(Anna Beliakova, Aaron D. Lauda) Categorification (B-Ok - Xyz)

684
Categorification in Geometry,
Topology, and Physics
Anna Beliakova
Aaron D. Lauda
Editors
American Mathematical Society

Anna Beliakova
Aaron D. Lauda
Editors
684
Anna Beliakova
Aaron D. Lauda
Editors
American Mathematical Society

Providence, Rhode Island
EDITORIAL COMMITTEE
Dennis DeTurck, Managing Editor
Michael Loss Kailash Misra Catherine Yan
2010 Mathematics Subject Classification. Primary 81R50, 57M25, 14F05, 18D10, 58J28,
17B81, 20C08, 17B55, 17B67.
Library of Congress Cataloging-in-Publication Data

Library of Congress Cataloging-in-Publication Data
Names: Beliakova, Anna, 1968– editor. — Lauda, Aaron, 1981– editor.
Title: Categorification in geometry, topology, and physics / Anna Beliakova, Aaron Lauda,
editors.
Description: Providence, Rhode Island : American Mathematical Society, [2017] | Series: Con-
temporary mathematics ; volume 684 | Includes bibliographical references.
Identifiers: LCCN 2016042028 | ISBN 9781470428211 (alk. paper)
Subjects: LCSH: Categories (Mathematics) | Mathematical analysis. | Topology. | Geometry.
| Physics. | AMS: Quantum theory – Groups and algebras in quantum theory – Quantum groups
and related algebraic methods. msc | Manifolds and cell complexes – Low-dimensional topology
– Knots and links in S 3 . msc | Algebraic geometry – (Co)homology theory – Sheaves, derived
categories of sheaves and related constructions. msc | Category theory; homological algebra –
Categories with structure – Monoidal categories (multiplicative categories), symmetric monoidal
categories, braided categories. msc | Global analysis, analysis on manifolds – Partial differential
equations on manifolds; differential operators – Eta-invariants, Chern-Simons invariants. msc |
Nonassociative rings and algebras – Lie algebras and Lie superalgebras – Applications to physics.
msc | Group theory and generalizations – Representation theory of groups – Hecke algebras and
their representations. msc | Nonassociative rings and algebras – Lie algebras and Lie superalge-
bras – Homological methods in Lie (super)algebras. msc | Nonassociative rings and algebras –
Lie algebras and Lie superalgebras – Kac-Moody (super)algebras; extended affine Lie algebras;
toroidal Lie algebras. msc
Classification: LCC QA169 .C3746 2017 | DDC 512/.62–dc23 LC record available at
https://lccn.loc.gov/2016042028
Contemporary Mathematics ISSN: 0271-4132 (print); ISSN: 1098-3627 (online)
DOI: http://dx.doi.org/10.1090/conm/684
Color graphic policy. Any graphics created in color will be rendered in grayscale for the
printed version unless color printing is authorized by the Publisher. In general, color graphics will
appear in color in the online version.
Copying and reprinting. Individual readers of this publication, and nonprofit libraries acting
for them, are permitted to make fair use of the material, such as to copy select pages for use
in teaching or research. Permission is granted to quote brief passages from this publication in
reviews, provided the customary acknowledgment of the source is given.
Republication, systematic copying, or multiple reproduction of any material in this publication
is permitted only under license from the American Mathematical Society. Permissions to reuse
portions of AMS publication content are handled by Copyright Clearance Center’s RightsLink
service. For more information, please visit: http://www.ams.org/rightslink.
Send requests for translation rights and licensed reprints to reprint-permission@ams.org.
Excluded from these provisions is material for which the author holds copyright. In such cases,
requests for permission to reuse or reprint material should be addressed directly to the author(s).
Copyright ownership is indicated on the copyright page, or on the lower right-hand corner of the
first page of each article within proceedings volumes.

c 2017 by the American Mathematical Society. All rights reserved.
The American Mathematical Society retains all rights
except those granted to the United States Government.
Printed in the United States of America.

∞ The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
Visit the AMS home page at http://www.ams.org/
10 9 8 7 6 5 4 3 2 1 22 21 20 19 18 17
Dedicated to Christian Blanchet
on the occasion of his sixtieth birthday
Contents
Preface ix
Geometry and categorification
Ben Webster 1
A geometric realization of modified quantum algebras
Yiqiang Li 23
The cube and the Burnside category
Tyler Lawson, Robert Lipshitz, and Sucharit Sarkar 63
Junctions of surface operators and categorification of quantum groups
Sungbong Chun, Sergei Gukov, and Daniel Roggenkamp 87
Khovanov-Rozansky homology and 2-braid groups
Raphaël Rouquier 147
DAHA approach to iterated torus links
Ivan Cherednik and Ivan Danilenko 159
vii
Preface
The emergent mathematical philosophy of categorification is reshaping our view

of modern mathematics by uncovering a hidden layer of structure in mathematics,
revealing richer and more robust structures capable of describing more complex
phenomena. This philosophy has led to a number of shocking new results and has
ushered in a new area of interaction between algebra, geometry, and theoretical
physics. Categorification is a powerful tool for relating various branches of math-
ematics and exploiting the commonalities between fields. It provides a language
emphasizing essential features and allowing precise relationships between vastly
different fields.
Categorification exemplifies the duality between pure mathematics and theo-
retical physics, with ideas from one area inspiring innovation in the other. Key
concepts are exchanged between fields and transported to other areas using the
powerful descriptive ability enabled by categorification. Indeed, the term “categori-
fication” originated in work of Crane and Frenkel who were studying techniques for
extending 3-dimensional topological quantum field theories to 4-dimensions. This
area has rapidly developed over the last twenty years far beyond these original con-
siderations, blossoming into an exciting and dynamic area of modern mathematics,
with deep relations to geometric representation theory, low-dimensional topology,
algebraic geometry, and mathematical physics.
Classical structures in representation theory, such as simple Lie algebras and
their representations, allow deformations into quantum groups and modules over
them, as discovered by Drinfeld and Jimbo. In turn, quantum groups lead to
a representation-theoretical interpretation of the Jones, Kauffman, HOMFLY-PT
and other link polynomials. These theories are closely connected to physics via the
Chern-Simons path integral, as shown by Witten in his seminal paper on the Jones
polynomial. More recently, it became clear that quantum deformation was only
a prelude into deeper structural beauty and meaning. In the work of Khovanov,
Ozsvath, Szabo, Rasmussen, Rozansky, and many others, quantum link invariants
were interpreted as Euler characteristics of various bigraded and multi-graded link
homology theories. These theories carry four-dimensional information, providing
invariants of link cobordisms. The resulting link homology theories appear to be a
nexus between some of the most sophisticated directions in modern mathematics
and theoretical physics.
The discovery of link homology gave rise to new advances in representation
theory, as quantum groups and related algebras were categorified. The resulting
higher representation theory created a new bridge between representation theory
and the link homology theories, which despite the increasing number of publications
in this area is still far from being fully explored.
ix
x PREFACE
This volume was inspired by the conference Categorification in Algebra, Geom-

etry, and Physics (a conference in honor of the 60th Birthday of Christian Blanchet),
which took place at IESC, Menasina, Cargese, France, from May 4th to May 8th,
2015. This conference made clear that categorification is a rapidly emerging area
of intense study. It also elucidated the need for a reference for newcomers to the
field to learn the types of tools used in categorification, the problems where these
tools have been successfully applied, and the future directions in which the field is
moving. Our aim with this volume is to address this need. To this end, we have
solicited articles from experts in categorification from around the world who were
invited to share their unique perspective.
This volume focuses on the role categorification plays in geometry, topology,
and physics. These articles illustrate many important trends for the field includ-
ing geometric representation theory, homotopical methods in link homology, inter-
actions between higher representation theory and gauge theory, and DAHA ap-
proaches to link homology.
The organizers wish to thank the John Templeton Foundation and the Univer-
sity of Zurich for their generous support in making this conference possible.
Anna Beliakova
Aaron Lauda
Contemporary Mathematics
Volume 684, 2017
http://dx.doi.org/10.1090/conm/684/13708
Geometry and categorification
Ben Webster
Abstract. We describe a number of geometric contexts where categorification
appears naturally: coherent sheaves, constructible sheaves and sheaves of modules
over quantizations. In each case, we discuss how “index formulas” allow us to
easily perform categorical calculations, and readily relate classical constructions of
geometric representation theory to categorical ones.
1. Introduction
“Categorification” is a very flexible concept. It simply refers to the idea that
it can be very interesting to take a set and add morphisms between its objects. Its
very flexibility means that it is an idea which must be employed carefully. It has not
proven very effective to start with a simple algebraic object and to hunt aimlessly
for categorifications of it. It is much more reliable to have a “machine” which
produces categories for you in a way that gives you some hope of understanding
how they decategorify.
Thus, geometry is a natural context for categorification because it is a natural
source of categories. The categories that appear in geometry also have a natural
geometric toolkit for producing functors (using push-pull along correspondences)
and calculation (using index formulas). Both of these can be more difficult to
understand in other approaches to categorification, such as algebraic or diagram-
matic. The focus of this paper will be on describing some of the basic ways of
applying geometry to construct categories, how the underlying geometry can help
with understanding these categories, and how to apply these principles in some
of the most illuminating special cases.
Categorification can also help us to better understand geometry. The categories
which appear naturally in this context shed light on the nature of the spaces they
are connected to. Often, the full structure of a category like coherent sheaves on
an algebraic variety is simply “too rich for our blood,” an amount of information
that exceeds our ability to take it in. Decategorification allows us cut out much of
the extraneous complication and understand some of the structure of this category,
and thus learn something about the underlying space.
This paper is structured around 3 different geometric contexts which naturally
lead to categorification:
The author was supported by the NSF under Grant DMS-1151473 and the Alfred P. Sloan
Foundation.
2017
c Ben Webster
1
2 BEN WEBSTER
• In Section 2, we consider categories of coherent sheaves and algebraic K-

theory. This is arguably the first place in the literature where the modern
philosophy of categorification appears, and the most likely to be some-
what familiar to the general reader.
• In Section 3, we consider categories of constructible sheaves and the
function-sheaf correspondence. While perhaps a more specialized taste,
this is actually an incredibly powerful theory, with connections to deep
number theory. In this author’s opinion, it is one any aspiring categorifier
should know a bit of.
• In Section 4, we consider categories of sheaves of modules over quantiza-
tions. This is the least familiar context, and one still under development.
Unlike the other two examples, we have not had the benefit of having
Grothendieck around to help us with it. However, progress on it has been
made, which we will briefly discuss here.
2. K-theory
The first appearance of categorification in its modern form was in topolog-
ical K-theory. Given a topological space X, we can consider the additive cate-
gory of vector bundles on X. To better understand this category we consider its
Grothendieck group K(X). This is the abelian group generated by symbols [A] for
A a vector bundle, subject to the relation
(1) [A] + [B] = [A ⊕ B].
Note, most authors consider vector bundles up to stable equivalence, which is the
same as considering the kernel of the map K(X) → Z sending a vector bundle to its
rank. Of course, this construction has many variations where we consider bundles
with additional structure. See the classic books of Atiyah [Ati67] or Karoubi
[Kar08] for more details on K-theory.
2.1. Coherent sheaves. This topological introduction is perhaps a little mis-
leading. We’ll instead be working with algebraic varieties. Both a curse and
a blessing of this geometric approach is that almost every construction that ap-
pears has several variations that make sense in different contexts. Throughout, I’ll
usually work in whatever context is most convenient for me. Thus, I could con-
sider holomorphic vector bundles on a complex manifold or locally free coherent
sheaves on a scheme or variety (over the complex numbers or some other field).
For a projective variety over C, these notions are the same by [Ser56], so the reader
will not lose much by thinking about whichever one they prefer.
Now, we will more seriously study the category Coh(X) of coherent sheaves
on a scheme X. Readers who are less happy with algebraic geometry might also
think about the category of modules over a commutative ring R, which is the same
as Coh(Spec R), the coherent sheaves on the spectrum Spec R of R.
Definition 2.1. Attached to this category, we have two natural Grothendieck groups:
• Let K0 (X) := K0 (Coh(X)) be the formal span of [F ] for all locally free1 coherent
sheaves F on X, modulo the relation
(2) [E] − [F ] + [G] = 0
1 Those of you thinking about R-modules should restrict to projective R-modules.
GEOMETRY AND CATEGORIFICATION 3
for any short exact sequence

(3) 0 −→ E −→ F −→ G −→ 0,
which is a slight modification of the relation (1), and
• Let G0 (X) := G0 (Coh(X)) be the span of [F ] for all coherent sheaves modulo the
same relation (2).
This modified relation is needed since neither of these categories is semi-
simple, while every short exact sequence of topological vector bundles splits.
Note that there’s an obvious homomorphism K0 (X) → G0 (X), but this need
not be an isomorphism. If X is quasi-projective and smooth, then it induces a
natural isomorphism K0 (X) G0 (X) since every sheaf has a finite length locally
free resolution2 . On the other hand, this will not hold in many other cases.
Remark 2.2. If R = C[t]/(t2 ), X = Spec R, then K0 (X) is spanned by the class of
the regular module R, and G0 (X) is spanned by the class of the 1-dimensional module
C R/tR tR. The short exact sequence
0 −→ tR −→ R −→ R/tR −→ 0,
shows that under the isomorphism G0 (X) Z, the subgroup K0 (X) is sent to 2Z. The
difference between this case and the smooth case is that the minimal projective resolution
of R/tR is the infinite complex
t t t
· · · −→ R −→ R −→ · · · −→ R −→ R/tR.
Note that K0 (X) is a ring, with multiplication given by [E][F ] = [E ⊗OX F ]. We
cannot endow G0 (X) with a compatible ring structure in general since ⊗OX is not
exact. As we’ll discuss below, we can sometimes fix non-exactness by considering
higher derived functors, but our example above shows that this can’t work here:
if X = Spec C[t]/(t2 ) then K0 (X) Z as a ring, so G0 (X) can only be a ring if 1/2Z is.
The problem is that ToriR (C, C) C for all i, so we would have to use the divergent
series 1/2 = 1 − 1 + 1 − 1 + · · · .
2.2. Functoriality. Our first task is to understand how maps between varieties
induce functors between categories and thus maps between K-groups. Of course,
the desired functors are pushforward and pullback of coherent sheaves, as defined
in [Har77, II.5]. Unfortunately, interpreted naively, neither of these functors is
exact. Pushforward f∗ is left exact, and thus has right derived functors Ri f∗ (see
[Har77, III.8]), and pullback f ∗ is right exact and thus has left derived functors Li f ∗ .
These have the usual long exact sequences
(4) · · · −→ Ri−1 f∗ G −→ Ri f∗ E −→ Ri f∗ F −→ Ri f∗ G −→ Ri f∗ E −→ Ri+1 f∗ E −→ · · ·
(5) · · · −→ Li+1 f ∗ G −→ Li f ∗ E −→ Li f ∗ F −→ Li f ∗ G −→ Li f ∗ E −→ Li−1 f ∗ E −→ · · ·

which show that the maps

∞
∞
(6) [F ] → (−1)i [Ri f∗ F ] [G] → (−1)i [Li f ∗ G]
i=0 i=0
2 In fact, by a theorem of Serre, a quasi-projective variety is smooth if and only if every coherent
sheaf has a finite length resolution.

4 BEN WEBSTER
are compatible with the relation (2) whenever these infinite sums make sense. As
often happens, the two different maps above make sense for the two different
versions of the Grothendieck group:
Proposition 2.3. For a projective morphism f : X → Y, the formulas of (6) define
maps
(7) G0 ( f∗ ) : G0 (X) → G0 (Y) K0 ( f ∗ ) : K0 (Y) → K0 (X)
Proof. For any coherent sheaf F on X, the sheaves Ri f∗ F are coherent by
[Har77, III.8.8(b)], and vanish for i > dim X by [Har77, III.2.7]. Thus, the first sum
of (6) is finite and well-defined. Note that we could not do this for K0 since Ri f∗ F
might not be locally free even if F is.
For a locally free sheaf G on Y, the pullback f ∗ G is locally free and the higher
pullbacks Li f ∗ G for i > 0 are 0, so the second sum of (6) is well-defined (and in fact
only has 1 non-zero term).
The map Spec C → Spec R induced by the unique ring homomorphism R → C
illustrates that it’s impossible to define these maps on the “wrong” Grothendieck
group:
• the regular module C is free, but its pushforward f∗ C is just the module
R/tR, which doesn’t have a corresponding class in K0 .
• the pullback Li f ∗ (R/tR) C for all i, so the second sum in (6) doesn’t
converge.
Of course, this map is somewhat pathological, and there are conditions one can
impose that will guarantee this maps make sense. In particular:
• If the source and target are smooth and quasi-projective, then we can
freely switch between K0 and G0 and so G0 ( f ∗ ) and K0 ( f∗ ) make sense for
a projective (or more generally, proper) morphism.
• If f is flat, then f ∗ is exact, and G0 ( f ∗ ) is well-defined.
2.3. Chern character and index formulas. Thus, we can supply ourselves with
a great number of categories and exciting functors between them. But as we said
in the introduction, this is particularly powerful because we can understand the
Grothendieck group and calculate the behavior of these maps in geometric terms.
Assume from now on that X is smooth and projective over C.
Theorem 2.4. There is a unique homomorphism ch : K0 (X) G0 (X) → H∗ (X; Q)
called Chern character which is:
(1) compatible with pullback
ch( f ∗ F ) = f ∗ ch(F )
(2) sends the class [L] of a line bundle L to
c1 (L)2
exp(c1 (L)) = 1 + c1 (L) + +···
2
where c1 (L) is the first Chern class, the cohomology class dual to the divisor
defined by a meromorphic section of L.
This definition is well-defined because of the splitting principle: given a vector
bundle E → X of rank n, we can consider the flag space Fl(E) given by pairs of a
point x ∈ X and a complete flag V1 ⊂ V2 ⊂ · · · ⊂ Vn = Ex . The cohomology ring is

given by
H∗ (Fl(E); Q) H∗ (X; Q)[α1 , . . . , αn ]/(ek (α) = ck (E));
here the classes αi are the first Chern classes of the induced vector bundles V
i /Vi−1 .
We can use ek (α) = ck (E) as a definition of the Chern classes, and ch(E) = ni=1 eαi .
Note that this implicitly gives a complicated but concrete formula for ch in terms of
Chern classes, since each homogeneous part is a symmetric polynomial in Chern
roots αi , and
thus a polynomial in the Chern classes. We can also define the Todd
class td(E) = ni=1 1−eα−α
i
i
.
Remark 2.5. Note that even if X is defined over a field other than C, we can still define
this homomorphism, with the target given by the Chow ring A∗ (X; Q), the ring spanned
by subvarieties of X modulo rational equivalence.
The Chern character is certainly not an isomorphism, but is not so far from
being one either:
Proposition 2.6. The kernel of ch ⊗ Q : K0 (X) ⊗ Q → H∗ (X; Q) is spanned by
elements [F ] − [G] where we have an isomorphism of underlying topological C-vector
bundles F C G.
The inelegance of this theorem comes from the fact that we are relating two “in-
compatible” structures. If we consider the topological K-theory of X as a manifold,
or replace H∗ (X; Q) by the Chow ring A∗ (X; Q), then this map will be an isomor-
phism. For many nice varieties, we have an isomorphism H∗ (X; Q) A∗ (X; Q), so
the map above becomes an isomorphism. For example, this is the case for any
variety which has an affine paving.
While there are many things to be said about K-theory, perhaps the most
important for us is the first example of an index formula: the Grothendieck-
Hirzebruch-Riemann-Roch theorem.
Theorem 2.7 (Grothendieck, Hirzebruch; [BS58, §7]). Let X, Y be smooth and
projective over C. Then for any map f : X → Y and any locally free sheaf F , we have that
ch( f∗ F ) td(TY ) = f∗ (ch(F ) td(TX )).
In particular, in the case where Y = Spec C is a point, we have that
∞
(−1)i dim Hi (X; F ) = ch(F ) td(TX ).
i=0 X
Since this is a particularly focus of categorifiers, let us note that using internal
Hom allows us to compute the Euler form on K0 (X), which is defined by

∞
[E], [F ] = Exti (E, F ).
i=0
This sum is well-defined because we have that

Exti (E, F ) = Hi (X; E∨ ⊗ F ),
which vanishes for i > dim X. This also allows us to see that

[E], [F ] = ch(F ) ch(E∨ ) td(TX ).
X
6 BEN WEBSTER
2.4. The Weyl character formula. One of the most important categorification
problems is understanding the characters of representations of groups. For com-
plex simple Lie groups, this problem is solved by the Weyl character formula.
While there are many proofs of this beautiful formula, one of the most remarkable
is obtained by combining GHRR as above with the Borel-Weil-Bott theorem, which
shows that for each highest weight λ, there is a line bundle Lλ on the flag variety
X := G/B of G such that H0 (X; Lλ ) Vλ and Hi (X; Lλ ) = 0 for i > 0. Thus, we have
that:
dim Vλ = ch(Lλ ) td(TX ).
X
Consider the symmetric algebra Sym• (t∗Q ) in the weights of the maximal torus T,
and let ( f ) be the constant term of f ∈ Sym• (t∗Q ). We can identify H∗ (X; Q) with
the symmetric algebra Sym• (t∗Q ) modulo Sym• (t∗Q )W + , the polynomials with ( f ) = 0
which are symmetric under the action of the Weyl group; in this realization, the
operation of integration is given by

w∈W (−1) w · p
w
(8) p= .
X α∈Δ+ α
+
rootsα α ∈ Δ are the Chern roots of the tangent bundle, so we λhave
The positive
td(TX ) = α∈Δ+ 1−e−α , and the Chern character of the line bundle is ch(Lλ ) = e , so

eλ+ρ α∈Δ+ α
ch(Lλ ) td(TX ) = α/2 − e−α/2 .
α∈Δ+ e
Thus, we have that

dim Vλ = ch(Lλ ) td(TX )
X

1 ew(λ+ρ) α∈Δ+ α
(9) = (−1) w
α∈Δ+ α w∈W
α/2 − e−α/2
α∈Δ+ e

(−1)w ew(λ+ρ)
= w∈W α/2 .
α∈Δ+ e − e−α/2
If we evaluate the RHS using L’Hôpital’s rule, we obtain the Weyl dimension
formula:
λ + ρ, α
dim Vλ = .
+
ρ, α
α∈Δ
We can also think T-equivariantly for the natural torus action on X. In this
case, the equation

w∈W (−1) w · p
w
p= ,
X α∈Δ+ α
gives the integral in equivariant cohomology, valued in HT (pt) Sym• (t∗Q ). We
can then interpret (9) in the completion of this T-equivariant cohomology:

α∈Δ+ α
w(λ+ρ) w w(λ+ρ)
1 w e w∈W (−1) e
ch(Lλ ) td(TX ) = (−1) = .
α∈Δ+ α w∈W
α/2 − e−α/2 α/2 − e−α/2
X α∈Δ+ e α∈Δ+ e
The Chern character of Vλ , considered as a T-equivariant coherent sheaf on a

point, is the sum over weight spaces μ dim(Vλ )μ eμ , so from equivariant GHRR,
we obtain the usual Weyl character formula:

(−1)w ew(λ+ρ)
dim(Vλ )μ eμ = w∈W α/2 .
μ α∈Δ+ e − e−α/2
3. The function-sheaf correspondence

3.1. Euler characteristic. Instead of coherent sheaves, we can also consider
constructible sheaves. Let k be a commutative ring, and for any topological space
X, we let kX denote the sheaf of locally constant3 k-valued functions on Ui .
Definition 3.1. We call a sheaf of k-modules F on a topological space X a local
⊕m
system if it is locally constant, i.e. if there is a finite open cover {Ui }, such that F |Ui kUi
for some integer m.
A great example of a local system is the flat sections of a vector bundle E with
a flat connection on a compact manifold. While globally there may be no sections,
there is always an open cover Ui where for any u ∈ Ui , each element of the fiber
Eu extends uniquely to a covariantly constant section of E over Ui . Thus a local
system is a sheaf where nearby fibers are isomorphic, but in order to make this
identification canonical, we have to shrink to a smaller neighborhood (and the
identification may depend on the neighborhood we choose).
Now, assume that the topological space we consider is a complex quasi-
projective variety. This space has two natural topologies both induced from the
embedding in projective space: the Zariski topology, whose generating open sets
are the locus where some meromorphic function on projective space has neither a
zero nor a pole, and the classical topology which is induced by the usual smooth
manifold structure on CPn . Constructible sheaves are sheaves of finitely generated
k-modules on algebraic varieties that are locally modeled on local systems in the
classical topology. We can define them inductively by saying that a sheaf F on
a variety of dimension n is constructible if there is a Zariski open subset U such
that F |U is a local system in the classical topology and on the complementary
Zariski closed subset V (of lower dimension), F |V is constructible4 . We let Sh(X)
be the category of constructible sheaves on X, and Db Sh(X) the subcategory of
the bounded derived category of all sheaves of vector spaces where all complexes
have constructible cohomology.
Certain aspects of constructible life are actually much simpler than coherent
sheaves. For example, instead of an index formula valued in homology we obtain
one valued in the space of constructible functions on X; as with sheaves, we in-
ductively define constructible functions on a variety of dimension n to be those
constant on a Zariski open subset U, with the restriction to the complement con-
structible. Let C(X) be the ring of constructible functions on X (with pointwise
3 The presheaf that assigns the constant k-valued functions to any open subset is not a sheaf, since
if a subset U = U1 ∪ U2 is the union of two open subsets with U1 ∩ U2 = ∅, then, a function that takes
value a on U1 and b on U2 must define a section of the sheafification by the gluing property. The locally
constant functions are, essentially by definition, the sheafification of this presheaf.
4 Note the odd mix of the classical and Zariski topologies here; this will be eased a bit when we
consider the ètale topology.

8 BEN WEBSTER
addition and multiplication). The map from sheaves to functions is one version of
the function-sheaf correspondence:
φF (x) = dim Fx .
More generally, for a complex of sheaves with constructible cohomology, we take

φF (x) = (−1)i dim H i (Fx ).
i∈Z
Here, we use H to emphasize we are just taking cohomology of a complex, not any
kind of sheaf cohomology. The long exact sequence (of sheaves) induced by a short
exact sequence (of complexes of sheaves) shows that this map factors through the
Grothendieck group G0 (Sh(X)), that this:
(10) φE − φ F + φ G = 0
whenever we have a short exact sequence of complexes like (3). Furthermore,
note that φ f −1 F = f ∗ φF , where f ∗ is the usual pullback of functions. However,
for constructible sheaves, there are two kinds of natural pushforward: the usual
pushforward f∗ F (U) = F ( f −1 (U)) and the pushforward with proper supports:
f! F (U) = {s ∈ F ( f −1 (U))| supp(s) → U is proper}.
While the former is probably more familiar, it is actually more convenient for us to
use the latter.
Just like coherent sheaves, constructible sheaves have an index formula.
Definition 3.2. For a constructible function φ on X and a map f : X → Y, we let f!
be the unique linear map such that:
• If X = U V with U open and V closed, we have f! φ = f! φ|U + f! φ|V .
• If φ = 1 is the constant function, then f! 1(y) = χc ( f −1 (y)) is the compactly-

supported Euler characteristic χc (X) = ∞ i i
i=0 (−1) dim Hc (X) of the fiber over
each point.
The pushforward
in the case where Y = ∗ is a point is sometimes called Euler integration
and denoted φ dχ. Pushforward is simply performing this integration over fibers.
Note, this is very close to the functor defined in [Mac74], but using compactly
supported Euler characteristics. This pushforward is a small modification of that
defined in [GM99, §1.4], by using the standard trick of factoring f! into the extension
by zero into a compactification, and then usual pushforward.
With this convention, we can now show:
Theorem 3.3. φ f! F = f! φF
i
Proof. First, let us prove this in the case of an open inclusion U → X. In this
case, the value at a point in u is obviously unchanged. At a point in X \ U, the stalk
of f! φF is 0, so the value of φ f! F at the point is 0, as is true of f! φF .
Let us prove this by induction on the dimension of the source variety X. Note
i j
that if we have a decomposition into open and closed subsets U → X ← V, then
we have a short exact sequence
0 −→ i! i∗ F −→ F −→ j! j∗ F → 0.
Thus, we have
φ f! F − f! φF = φ f! i! i∗ F + φ f! j! j∗ F − ( f i)! φi∗ F − ( f j)! φ j∗ F .
Since dim V < dim X, we have φ f! j! j∗ F = ( f j)! φ j∗ F . Thus, it suffices to check the
result after removing an arbitrary closed subvariety from X. Thus, we can assume
that F is a local system since this holds on an open subset. Taking an open cover
of X (in the classical topology) that trivializes F , and applying the Mayer-Vietoris
spectral sequence associated to this cover, we see that we get the same answer for
any local system, and thus can consider F = kX . In this case, the stalk of Ri f! kX at
a point y ∈ Y is given by Hci ( f −1 (y)) by base change5 [Del77, 4.5.4], so indeed, the
result follows from the equality f! 1(y) = χc ( f −1 (y)).
3.2. Étale cohomology. There is a more refined version of this theorem, which
is much more difficult to prove; it was the endpoint of 3 decades of remarkable work
in algebraic geometry.
Computing the Euler characteristic of a complex algebraic variety is a crude
analogue of counting the number of points in an algebraic variety over a finite
field (it behaves a lot like the number of points when p = 1). In particular, both
quantities are invariant under scissors congruence.
Like Euler characteristic, the number of points in an algebraic variety has a
cohomological interpretation, which is again an index formula. Understanding
this interpretation correctly requires a lot of difficult technical details, but these are
surprisingly easy to bypass to understand the general framework. Those looking
for more details should look first at the notes of Milne [Mil13], which cover all the
basic ideas while remaining relatively accessible. If still more details are sought, the
reader can turn to the earlier book of Milne [Mil80] or that of Kiehl and Weissauer
[KW01], both in English, or earlier French sources, such as [Del77].
The first scary-sounding thing is the étale topology on an algebraic variety X
(see [Mil13, §2-7] for general discussion). This is not a topology in the usual sense,
but a Grothendieck topology: its “open subsets” are given by maps ν : U → X (see,
for example, [Mil13, §4]). You can think of this as imposing a topology where
certain maps are formally locally invertible (even if there’s no underlying map
of spaces that really achieves this). The étale topology on a smooth manifold is
the topology where the open sets are manifolds U equipped with a smooth map
ν : U → X such that at each point u ∈ U, the differential Tu ν : Tu U → Tν(u) X is
an isomorphism (so this map is both an immersion and a submersion, a property
we call étale). You can easily work out that on a manifold, the étale topology
is equivalent to the usual one, by the inverse function theorem6 . Thus the étale
topology is essentially the topology where we declare a priori that the inverse
function theorem is true. Note that while this is not a “real” topology, we can
∗
still make sense of sheaves in this topology, and define cohomology Hét (X; Λ) in
5 Note the importance of using compactly supported pushforward here, since the usual pushfor-
ward does not have this property for non-proper maps.
6 Two Grothendieck topologies are equivalent if for any “neighborhood” ν : U → X in one topology,
there is a neighborhood η : V → X in the other topology such that ν factors through η (so η “contains” ν)
and another η : V → X such that η factors through ν (so ν “contains” η ). An open subset in the usual
topology includes via an étale map, and for any étale map, the inverse function theorem guarantees
there’s a neighborhood V of u that maps diffeomorphically to an open subset of X.
10 BEN WEBSTER
∗
any abelian group Λ, and compactly supported cohomology Hét,c (X; Λ) using Čech
7
cohomology .
For manifolds, this yields nothing new, since the inverse function theorem
really is true. However, we can apply the same trick in situations where it is
not, like the Zariski topology on an algebraic variety. This genuinely changes
the topological behavior of this variety. For example, it’s a well-known fact that
the cohomology of a complex algebraic variety in the Zariski topology is trivial,
whereas if we compute it in the étale topology with coefficients in a finite abelian
group Λ, then, by [Mil13, 21.1], it coincides with the usual Betti cohomology of the
complex points (in the classical topology):
∗
Hét (X; Λ) H∗ (X(C; Λ)).
The fact that a finite group is required here is a minor nuisance; it is mostly one
of the technicalities I suggest the reader ignore, but it does mean that typically, we
work with coefficient groups and rings which are built from finite ones. Thus, by
definition, we let
∗ ∗
Hét (X; Z ) := lim Hét (X; Z/n Z)
←−−
∗ ∗
Hét (X; Q ) := Hét (X; Z ) ⊗Z Q .
for any prime . Unfortunately, these do not give the same result as computing “di-
rectly” in the étale topology. The same comparison theorems to Betti cohomology
exist for these groups by the universal coefficient theorem.
There are other differences between the étale and Zariski topologies which are
relevant for local systems. For example, on C∗ , consider the local system defined
df
by solutions of x dx = 12 f . Solutions to this are provided by the branches of the
square root function. Thus, it is very easy to find an open subset in the classical
topology where this local system is trivialized by removing a single ray; however,
there is no Zariski open subset (that is, the complement of finitely many points)
where this local system is trivialized. On the other hand, in the étale topology, it is
x→x2
trivialized by the “neighborhood” C∗ −→ C∗ .
In the previous section, we defined the constructible sheaves using a funny
mix of the classical and Zariski topologies, essentially because there aren’t enough
interesting local systems in the Zariski topology; now knowing about the étale
topology, we might prefer to consider étale constructible sheaves8 .
The remarkable thing about the étale topology on a complex algebraic variety
is that it is a purely algebraic object: the tangent spaces and differential can be
rephrased in algebraic language, so we can speak of a map between schemes being
étale and thus define étale neighborhoods and the étale topology for an arbitrary
∗
scheme. Thus, we can define cohomology groups Hét (X; Λ) for an arbitrary scheme.
This results in a second remarkable comparison theorem: assume that we
have a scheme X defined over Z (for example, a projective variety defined by
7 It’s more “morally correct” to define this cohomology using derived functors; however, for
reasonable schemes, this is the same by [Mil13, 10.2].

8 Note that certain classical local systems cannot be trivialized in the étale topology. Solutions to
df
x dx = α f only will be if α is rational. This is yet another complication it will probably not greatly benefit
the reader to cogitate upon.
polynomials with integral coefficients)9 . We can consider the base-change of this

variety C or to the algebraic closure F̄p of any finite field of characteristic p.
Theorem 3.4 ([Mil13, 20.5 & 21.1]). If is any prime distinct from p, then we have
Hét∗ (X ⊗ F̄p ; Z ) Hét∗ (X ⊗ C; Z ) H∗ (X(C); Z ).
Since the former group is defined purely using the characteristic p geometry
of X ⊗ F̄p (that is, the solutions to our polynomials over finite fields) and the latter
purely using the topology of the complex solutions, this is a pretty remarkable
theorem.
3.3. The Grothendieck trace formula. However, one might wonder what pur-
pose it serves in relation to categorification. These results about cohomology are
in fact proven in a categorical context. We can consider étale local systems and
constructible sheaves not just on complex algebraic varieties, but on any scheme,
in particular one of characteristic p. Just like on complex algebraic varieties, these
sheaves are endowed with pushforward and pullback functors.
But rather than just considering Euler characteristic of stalks, we have a much
richer invariant, which incorporates the action of the Frobenius. The Frobenius
of interest to us is the relative Frobenius Fr : X ⊗ F̄p → X ⊗ F̄p which is induced
by raising functions on X ⊗ Fp to the pth power. For a projective variety, this is
the map of raising the projective coordinates to the pth power; note that this is an
automorphism of the variety since the polynomials have integer coefficients. For
any constructible sheaf F on X ⊗ Fp , we have a canonical isomorphism F Fr∗ F .
This means that if x ∈ X is an Fp -rational point (i.e. one whose coordinates lie in
Fp ), then, we have an induced Frobenius map Fr : Fx → Fx . If we have that F is a
complex of sheaves with Q -constructible cohomology defined on X ⊗ Fp , then we
get an action on the stalks of the cohomology on this point. This action respects
the differentials of the long exact sequence on cohomology, so we find that:
Proposition 3.5. The function

Φ(F )(x) = (−1)i tr(Fr | H i (Fx ))
i∈Z
defines a map from the Grothendieck group K(Sh(X ⊗ Fp )) to Q -valued functions on

X(Fp ).
More generally, we can consider the nth power of the Frobenius map, and
define
Φ(n) (F )(x) = (−1)i tr(Frn | H i (Fx ))
i∈Z
for x ∈ X(Fpn ). Compared to just taking Euler characteristic, this map is much more
powerful.

Theorem 3.6. The map from the Grothendieck group to ∞ n=1 Q [X(Fpn )] defined by
∞
n=1 Φ
(n)
is injective.
9 This is really a much stronger hypothesis than we need. With a bit more work, this theory can be
made to work for any variety over the complex numbers.

12 BEN WEBSTER
These functions also satisfy a trace formula, remarkably (or maybe not so
remarkably?) also due in large part to Grothendieck. Compatibly with our notation
before, if f : S → T is a map of finite sets, andΦ : S → k is a map to any ring k (or
more generally abelian group), then f! Φ(t) = s∈ f −1 (t) Φ(s).
Theorem 3.7. For any map between Fp -schemes f : X → Y, we have Φ(n) ( f! F ) =

f! Φ (F ), where on the RHS, we use f to denote the induced map X(Fpn ) → Y(Fpn ). In
(n)
particular, if f : X → Spec Fp , and F = Q , then we find that

∞
#X(F ) =
pn (−1)i tr(Frn | Hét,c
i
(X; Q )).
i=0
While it might seem strange, this theorem is thoroughly topological in nature:

it’s simply the Lefschetz fixed point theorem applied to the Frobenius.
Note, this means that the eigenvalues of Frobenius have a powerful effect on
the number of points in X(Fpn ) as we change n. For example, Wiles’s proof of
Fermat’s last theorem proceeded by showing that a counter-example would lead
to the existence of an elliptic curve whose Frobenius eigenvalues are too strange
to actually exist.
3.4. Grassmannians and sl2 . Now, let’s actually apply these theorems a bit.
One very interesting and relevant example is given by the system of Grassmannians
and partial flag varieties. Let Gr(r, n) be the Grassmannian defined over Z; you
can either think of this as the projective variety defined by the Plucker relations
(which have integer coefficients) or as the variety whose functor of points sends a
ring k to the set of module quotients kn → V such that V kn−r . If k is a field, this
is the collection of r-dimensional subspaces in kn .
Given r < r ≤ n, we have a partial flag variety Gr(r ⊂ r , n), given by pairs of
subspaces with one inside the other. These have their own Plücker relations, also
defined over Z, and natural maps
πr,r πr ,r
Gr(r, n) ←− Gr(r ⊂ r , n) −→ Gr(r , n).
There are functors
E = (πr,r+1 )∗ π∗r+1,r [r] F = (πr,r−1 )∗ π∗r−1,r [n − r]
relating the categories Cr := Sh(Gr(r, n)) for all 0 ≤ r ≤ n. The brackets indicate
homological shift in the derived category, but also require changing the action of
the Frobenius in order to keep the mixed structure pure of weight 0 by a factor of
p−r/2 ; the square of this operation is called “Tate twist.” The overall effect is that
Φ(m) (G[r]) = p−rm/2 Φ(m) (G).
These functors are biadjoint up to shift since πr ,r is smooth and proper. We can
understand the action of these functors using the index formulas we’ve defined
1
(Theorem 3.7). For two subspaces V, V in a larger vector space, we write V ⊂ V if
V ⊂ V and dim(V /V) = 1. Let q = pm .
Proposition 3.8. For any constructible function G on Gr(r, n) and V ∈ Gr(r, n), we
have
Φ(m) (EG)(V) = q−r/2 Φ(m) (G)(V )
1
V ⊂V

Φ(m) (F G)(V) = q(r−n)/2 Φ(m) (G)(V )
1
V ⊂V
Now, consider the difference

Φ(m) (EF G)(V) − Φ(m) (F EG)(V) = q(−n+1)/2 Φ(m) (G)(V) − q(−n+1)/2 Φ(m) (G)(V)
1 1
V ⊂V V ⊂V
(−n+1)/2
=q (1 + · · · + qn−r−1 − 1 − · · · − qr−1 )Φ(m) (G)(V)
(−n+1)/2 qn−r − qr (m)
=q Φ (G)(V)
q−1
q(n−2r)/2 − q(2r−n)/2 (m)
= Φ (G)(V)
q1/2 − q−1/2
By the joint injectivity of the maps Φ(m) , this implies that
(11) EF G F EG ⊕ G[n − 2r − 1] ⊕ · · · ⊕ G[2r − n + 1] (n − 2r ≥ 0)
(12) EF G ⊕ G[2r − n − 1] ⊕ · · · ⊕ G[n − 2r + 1] F EG (n − 2r ≤ 0)
This is a categorified version of the Uq (sl2 ) relation
K − K−1
EF − FE = .
q − q−1
In fact the functors E and F define a categorical action of sl2 , in the sense of Chuang
and Rouquier.
Definition 3.9. A collection of categories Cn and functors
E : Cn → Cn+2 F : Cn → Cn−2
form a categorical sl2 -action if
(1) the functors E and F are biadjoint, and
(2) satisfy the relations (11–12) up to isomorphism, and
(3) there is an action of the nilHecke algebra on the m-fold composition Em .
The first 2 of these properties have geometric interpretations we’ve discussed:
they arise from the properness and smoothness of the maps, and from the point
counting above. The third also has a geometric interpretation. The functor Em is a
push-pull functor for the correspondence
1 1 1
Gr(r ⊂ r + 1 ⊂ · · · ⊂ r + m, n) = {Vr ⊂ Vr+1 ⊂ · · · ⊂ Vr+m ⊂ kn }
over Gr(r, n) and Gr(r + m, n). This fits into the diagram:
Gr(r ⊂ r + 1 ⊂ · · · ⊂ r + m, n)
q
πr,r+m πr+m,r
Gr(r, n) Gr(r ⊂ r + m, n) Gr(r + m, n)
The vertical map q is a fiber bundle with fiber given by the complete flag variety
Fl(m) on an m-dimensional space; the functor Em can be rewritten as convolution
with the pushforward of the constant sheaf on Gr(r ⊂ r + 1 ⊂ · · · ⊂ r + m, n),
14 BEN WEBSTER
which is isomorphic to H∗ (Fl(m)) tensored with the constant sheaf on Gr(r ⊂ r +

m, n). Thus, the action of the nilHecke algebra on this functor is inherited from its
action on the cohomology of the flag variety. This is a more sheafy interpretation
of the results of [Lau10, Lau11] which phrases the same action in terms of the
cohomology of Grassmannians and these correspondence, which arise when we
take hypercohomology of the sheaves discussed above.
Remark 3.10. This action can be extended to other partial flag varieties, obtaining a
categorification of sln when we consider n-step flags. This corresponds to the calculations
of Khovanov and Lauda in [KL10]. This action is discussed in greater detail in [Webb].
3.5. Flag varieties. The structure of constructible sheaves on flag varieties is
a deep and beautiful subject. To stay within the bounds on this paper, we will
mainly concentrate on the relationship to the Hecke algebra. Consider a Coxeter
group W generated by the set S with the relations
s2 = e (st)m(s,t) = e
Definition 3.11. Let Hv (W) be the algebra over Z[v, v−1 ] generated by Ts for s ∈ S
with the relations:
(Ts + 1)(Ts − v) = 0 Ts Tt · · · = Tt Ts · · · .
where the latter two products both have m(s, t) terms.
On the other hand, we can consider the Fq -points of a split simple algebraic
group G over Fq (for example, PGLn (Fq ), Sp2n (Fq ), SOm (Fq ), · · · ), and let B be the
Fq -points of a Borel. This is just the group elements which preserve an appropriate
flag (which must be self-dual for symplectic or orthogonal groups). We have a
Bruhat decomposition which gives a bijection between the Weyl group W and the
double cosets BwB for B in G. The Weyl group of G is a Coxeter group (for example,
for PGLn (Fq ), it is the symmetric group Sn ).
The set of functions on the double cosets B\G/B has a natural multiplication:
let
f g(BwB) = f (Bw B)g(Bw B).
Bw ∈B\G
w B∈G/B
BwB=Bw w B
This arises naturally when we identify k[B\G/B] with EndG (k[G/B]), acting by the
same formula. We can also identify B\G/B with the set of G-orbits on G/B×G/B via
the map (g1 B, g2 B) → Bg−1
1
g2 B. In this realization, we can write this multiplication
as
f g = (π13 )! (π∗12 f · π∗23 g)
where πi j : (G/B)3 → (G/B)2 is the map forgetting the factor which is not listed.
Theorem 3.12 ([Iwa64]). The set of functions on B\G/B is isomorphic to the special-
ization Hq (W) at v = q.
Just to give a taste of how this map works, let’s consider the case of G =
PGLn (Fq ). In this case, G acts transitively on the set of complete flags in Fnq , with
the stabilizer of each flag being a Borel. Thus G/B is just the set of complete
flags in this vector space, with choosing a Borel B specifying a preferred flag
V1 ⊂ V2 ⊂ · · · ⊂ Vn = Fnq . The identity in the convolution multiplication is the
indicator function of this preferred flag V• ; as a function on G/B×G/B, this becomes
the the indicator of the diagonal. Our generators of the Hecke algebra are T(i,i+1)
for (i, i + 1) one of the adjacent transpositions in the symmetric group Sn . We send
T(i,i+1) to the indicator function of the Schubert cell
C(i,i+1) = {V• |Vi Vi , V j = V j for all j i}.
as a function on G/B × G/B, this becomes the the indicator of the pairs of flags that
differ only in i-dimensional subspace.
From our earlier discussion, we know that the natural way to categorify func-
tions on G/B is to consider constructible sheaves on G/B. In order to obtain
B-invariant functions, we need to consider B-equivariant sheaves, or equivalently
G-equivariant sheaves on G/B × G/B.
Definition 3.13. We call a sheaf on a G-scheme X weakly G-equivariant if there is
an isomorphism of sheaves a∗ F = p∗ F where a, p : G×X → X are the action and projection
maps (g, x) → gx, x respectively.10
We let Db (ShG (X)) denote the full subcategory of the derived category whose coho-
mology is weakly equivariant and constructible.
Now, let us consider the underlying algebraic group G, and G/B as an algebraic
variety over Fq . We’ll be interested in the derived category Db (ShG (G/B × G/B)) of
weakly G-equivariant sheaves. The G-orbits of G/B × G/B are all simply connected
and there are finitely many of them, so you can think of these as trivial vector
bundles on the different orbits, with some sort of topological glue holding them
together.
In particular, if we take the function Φ for one of these sheaves, we obtain a
function on G/B × G/B which is constant on G-orbits. Put differently, Φ defines
a natural map from the Grothendieck group Db (ShG (G/B × G/B)) to the Hecke
algebra. This map is surjective, since we can consider the extension i! (Q )Y from
an orbit Y, which hits the indicator function of the orbit.
This map is not injective, but this is only because we only considered a single
field. Thus, given one sheaf F on G/B, we have that Φ(F ⊕q ) = Φ(F (1)) = qΦ(F )
where F (1) is the Tate twist of F . However,
Φ(n) (F ⊕q ) = qΦ(n) (F ) Φ(n) (F (1)) = qn Φ(n) (F )
so considering larger n will fix this problem. If we do this carefully, we can
construct a category whose Grothendieck group is Hv (W): the subcategory of
Db (ShG (G/B × G/B)) where the Frobenius acts by elements of qZ on every stalk,
and v corresponds to Tate twist.
Thus we find that:
Corollary 3.14. The category Db ShG (G/B × G/B) categorifies the Hecke algebra.
In fact, the category Db ShG (G/B × G/B) is a geometric realization of the cate-
gorification of the Hecke algebra by Soergel bimodules, or the diagrammatics of
Elias and Williamson (see [Webb, Th. 6] and its proof).
10 The reader might rightly protest that it would be better to consider strong equivariant sheaves,
which is are sheaves together with a choice of isomorphism which satisfies a cocycle (associativity)
condition. This is fair, but not necessary at the present. Ultimately, if the reader wants to “do this
right,” they should use equivariant derived categories as well [BL94, WW], but in this context, this is
not necessary for understanding the basic point.
16 BEN WEBSTER
This is only a very small taste of a very large story; this is discussed in much
greater detail in the book of Hotta, Takeuchi and Tanisaki [HTT08]. The category
ShG (G/B × G/B) is related to very interesting categories of representations of the
Lie algebra g called Harish-Chandra bimodules and category O.
One topic we did not have the space to consider is that of perverse sheaves
and intersection cohomology. This is a huge topic, and provides an important lens
for all of the examples with constructible sheaves we’ve considered. One good
starting point is the introductory article [dCM10]. One of the key consequences
of this theory is that the category Db (ShG (G/B × G/B)) contains special objects
called intersection cohomology sheaves. These special complexes of sheaves give
a special basis of the Hecke algebra called the Kazhdan-Lusztig basis. This is
one of the prototypical examples of a special (or “canonical”) basis arising from
categorification.
3.6. Hall algebras. Hall algebras are discussed in much greater detail in the
survey article of Schiffmann [Sch]. Here, we will consider the rather narrow topic
of how they fit in with the function-sheaf correspondence; the general experience
of the author is that this connection is not nearly as well-known as it should be.
Hall algebras arise from our philosophy when the underlying space X is the
moduli space of objects ObA in some abelian category. Making careful mathemat-
ical sense of such a space can be quite tricky; usually it must be thought of as some
kind of stack. Thus, throughout, we’ll deal mostly with the most familiar example:
if Γ is a quiver, then there is an abelian category of representations of Γ, that is, of
representations of Γ’s path algebra.
There is a geometric space whose points are the isomorphism classes of repre-
sentations of Γ. For a dimension vector d : Γ → Z≥0 , let
Ed (k) = ⊕i→ j Hom(kdi , kd j )
for k any commutative ring; this defines the functor of points for an
algebraic variety
over Z. This space has an action of the affine algebraic group Gd = GLdi (k), where
GLdi (k) act by precomposition on arrows from i and by postcomposition on arrows
to i.
You can think of Ed as being a quiver representation on a free k-module with
a choice of basis in each of the spaces, and Gd as the group that acts by changing
bases. Thus, two elements of Ed (k) represent the same representation of Γ if and
only if they are in the same orbit under Gd (k).
Definition 3.15. We let the moduli space of representations of this quiver be the
union of the quotient spaces11 Ed /Gd for all d.
This moduli space has an additional structure, which arises from the fact that
there are short exact sequences, which form a related moduli space.
For the quiver Γ, you can break these sequences up into components where
the submodule has dimension d and the quotient d . You can think of this as the
space
d d d
Ed ,d (k) = ⊕i→ j Hom(kdi , k j ) ⊕ Hom(kdi , k j ) ⊕ Hom(kdi , k j )
11 In this case, it’s much harder to sweep the issue of equivariance under the rug. However, we are
nothing if not persistent, and thus will do our best to achieve said sweeping.
modulo the action of the group

Gd ,d := {g ∈ GL(kdi ⊕ kdi )|g(kdi ) = kdi }.
i
The moduli space of short exact sequences has 3 projection maps πs , πt , πq
considering the submodule, total module and quotient. In the case of quiver

representations, this is considering the action on kdi , kdi ⊕ kdi , and kdi respectively.
Definition 3.16. We let the Hall algebra be the constructible functions on ObA (k)
the points of the moduli space for k = C, Fq equipped the algebra structure
(13) f g = (πt )∗ (π∗s f · π∗q g); .
Similarly, we can define a monoidal structure on the category of constructible
sheaves on ObA via essentially the same formula:
(14) F G = (πt )∗ (π∗s F ⊗ π∗q G).
Remark 3.17. Actually all of these formulas are slightly wrong. The equation (13)
should have included a power of q (where q is the size of the field or −1 if k = C), depending
on the dimension of the fiber of πs × πq . Similarly, (14) should include a homological shift
by the same dimension. This shift assures that convolution commutes with Verdier duality.
Thus Theorem 3.7 implies that (13) categorifies (14):
Theorem 3.18. The map Φ(n) : K(Sh(ObA (Fq ))) → Q [ObA (Fqn )] is a homomor-
phism of rings and for every class in K(Sh(ObA (Fq ))), there is a n for which Φ(n) does not
kill this class.
In the case of representations of Γ, we have a second way of thinking about
this Hall algebra. Work of Ringel [Rin90] shows that we have a homomorphism
from the quantized universal enveloping algebra Uq (n) of the maximal unipotent
subalgebra of the associated Kac-Moody algebra of Γ to the Hall algebra. This map
sends the Chevalley generator Ei to the indicator function of a trivial representation
which is 1 dimensional and supported on i. Thus, Theorem 3.18 suggests we
should be able to categorify Uq (n) by replacing these indicator functions with the
corresponding constant sheaf on Eαi /Gαi ∗/Gm . This was, in fact, carried out by
Lusztig [Lus91,Lus93] and leads to his construction of canonical bases for universal
enveloping algebras.
This construction was given a different spin in the work of Rouquier [Rou12,
Rou] and Varagnolo-Vasserot [VV11], who showed that the resulting categories of
sheaves can be understood algebraically using KLR algebras.
4. Symplectic resolutions
There is one final context for categorification we want to discuss: that of conical
symplectic resolutions of singularities. These are closely allied to the constructible
sheaves we discussed, but also bear some similarities to coherent sheaves.
A conical symplectic resolution is an algebraic variety M over C which is
equipped with:
• a birational projective morphism M → N to an affine variety
• an algebraic 2-form Ω such that Ω is symplectic
• a conical S C∗ -action on M and N compatible with the map such that Ω
has weight n > 0 for this S-action.
18 BEN WEBSTER
These varieties have many remarkable properties. The most relevant for us is
that they can be quantized. The symplectic form Ω induces a Poisson bracket on
functions; this Poisson bracket is actually the leading order part of a quantization.
That is:
Theorem 4.1. There is a S-equivariant sheaf of algebras Q flat over C[[h]] on M
such that Q/hQ OM , and [ f, g] ≡ h{ f, g} (mod h)2 . In fact, up to isomorphism, such
algebras are in canonical bijection with H2 (M; C).
Since [−, −] has weight 0 under S, and {−, −} weight −n, we must have that h
has S-weight n for the desired equation to hold.
The corresponding cohomology class is called the period of a quantization. If
we take the global sections Γ(M; Q), then we can “set h = 1” by adjoining h−1/n , and
considering the invariant sections. Since h−1/n has weight −1, any section which
is a S-weight vector can be multiplied by an appropriate power of h to make it
invariant. Let D := Q[h−1/n ].
Theorem 4.2. The invariant sections A := Γ(M; D)S is a non-commutative algebra
filtered by its intersections A(m) := A ∩ Γ(M; h−m/n Q). There is a canonical isomorphism
C[M] A.
We call a S-equivariant module over D good if it is isomorphic to the base
extension of a coherent (i.e. locally finitely generated) Q module. The categories of
good D modules and finitely generated A-modules are related by an adjoint pair
of functors
ΓS : D -mod → A -mod Loc : A -mod → D -mod
which are often, but not always equivalences.
Remark 4.3. The category D -mod is much less sensitive to changing the parameter
that defines the quantization. Any line bundle L on M can be quantized to a right module
L over D; however, this quantization does not have a natural left D-module structure.
Instead EndD (L) is a new quantization, corresponding to a cohomology class differing
from that of D by the Euler class of L.
Since every element of H2 (M; Z) is the Euler class of a line bundle, this means that the
H (M; Z)-cosets of H2 (M; C) give Morita equivalent quantizations. Thus, without loss of
2
generality, we can add a large integer multiple of an ample class on M to the period of D.
For a sufficiently large multiple, we have that localization holds by [BPW].
The category D -mod also has the enormous advantage of allowing one to
use the methods of geometric categorification. Kashiwara and Schapira [KS12]
define a notion of a Euler class for a good D-module. While this has a general
definition using Hochschild homology, it can actually be thought of in a relatively
straightforward way in this special case.
The most interesting D-modules are holonomic; that is, their support is a half-
dimensional subvariety of M. If we fix a good D-module sheaf M, and x ∈ supp M
a point where the support is smooth, then the completion of M at this point will be
a free module over the functions on the completion of supp M at this point. The
rank of this module is constant on an open subset of the component containing x,
and thus defines an invariant rC of this component.

Definition 4.4. The Euler class of M is the sum e(M) = rC [C] ∈ Htop
BM
(supp M).
This is shown in [KS12, 7.3.5] to be equivalent to the more general definition.

Unfortunately, one has to be careful about what an index formula means in
this case. There is no definition of a pushforward in this context. Instead, one has
to rely on the interactions between modules. For example:
Theorem 4.5. Let M and N be good holonomic modules such that supp M∩supp N
is compact. Then
∞
(−1)i dim Exti (M, N) = e(M) ∩ e(N),
i=0 supp M∩supp N
where ∩ is the usual intersection product of homology classes.

More generally, one can replace pushforwards with convolutions. If we quan-
tize M ⊗ M̄ (as usual, M̄ is the space M with opposite symplectic form) with the
quantization DD ˆ op , then we can view modules over this quantization as bi-
modules. In fact, the localization and sections functors relate this category to the
category of A - A bimodules.
Definition 4.6. A DD ˆ op -module H is Harish-Chandra if it possesses a Q-lattice
H such that H /hH is killed by 1 ⊗ f − f ⊗ 1 for any global function f on M. This is
0 0 0
equivalent to H 0 /hH 0 having support on the scheme M ×N M.

Let HC be the category of Harish-Chandra bimodules on M, and Db (HC) the
subcategory of the bounded derived category with Harish-Chandra cohomology.
Let ν : M → N be the resolution map with Steinberg variety Z := M ×N M. Con-
sider the three different projections pi j : M × M × M → M × M as well as the two
projections pi : M × M → M. The cohomology HZ2 dim M (M × M; C) with supports in
Z is isomorphic to H2BM
dim M
(Z) and has a convolution product given by the formula
α β := (p13 )∗ (p∗12 α · p∗23 β),
making it into a semisimple algebra [CG97, 8.9.8]. For any closed subvariety
L ⊂ M with the property that L = ν−1 (ν(L)), there is a degree-preserving action of
this algebra on the cohomology HL∗ (M; C) given by the formula
α γ := (p2 )∗ (α · p∗1 γ).
Similarly, we have a convolution product on Harish-Chandra bimodules de-
fined by
L
(15) H1 H2 := (p13 )∗ (p−1 −1
12 H1 ⊗p−1 Dλ p23 H2 ),
2
making Db (HC) into a monoidal category. The same formula (15) defines an action
of Db (HC) on the derived category DbL (D -mod) of complexes with cohomology
supported on L
Theorem 4.7. Euler class defines an algebra homomorphism K(HC) → Htop
BM
(M×N M)
b BM
which interwines the representations on K(DL (D -mod)) and Htop (L).
A number of examples of these resolutions are discussed in [BLPW, §9]. These
include:
• the Springer resolutions of nilcones, and the induced resolutions of Slodowy
slices.
20 BEN WEBSTER
• Nakajima quiver varieties [Nak94, Nak98] (also discussed below), which

are geometric avatars of the representations of Lie algebras.
• Hypertoric varieties [Pro08], which are quaternionic analogues of toric
varieties, and give geometric versions of various notions in the theory of
hyperplane arrangements.
• Lusztig slices in affine Grassmannians [KWWY]. These also serve to ge-
ometrize representations of Lie algebras, but in a way “dual” to Nakajima
quiver varieties.
In [BLPW, §10], we propose a notion of duality for these symplectic resolutions.
While many complex structures appear in the conjectured duality, the one closest
to this paper is that there is a category O of special modules over D -mod, and a
dual category O! for the symplectic dual variety.
Conjecture 4.8 (Braden-Licata-Proudfoot-Webster). There is an isomorphism
K(O) K(O! ) such that the induced actions of K(HC) and K(HC! ) commute and form a
dual pair (in the sense of Schur-Weyl duality).
Schur-Weyl duality itself arises from considering this for flag varieties in type
A. Skew-Howe duality (for type A quiver varieties/Lusztig slices) and rank-level
duality (for affine type A quiver varieties/Lusztig slices) are special cases.
Now, let use turn to dicsussing some examples in more detail:
4.1. Flag varieties. The most familiar example is when M = T∗ G/B. In this
case, we have little to say beyond the results of Section 3.5, since D-modules on
T∗ G/B are just D-modules on G/B by [BPW, §4.1], and the categories Db (HC) is
essentially just Db ShG (G/B × G/B) actually coincide, since the Steinberg M ×N M is
just the union of conormal bundles to G-orbits in G/B × G/B. However, this gives
a slightly different perspective, using the Springer resolution. The cohomological
side of this picture is discussed in great detail in [CG97]. The most important
results are that:
Theorem 4.9. We have an algebra isomorphism Htop BM
(M×N M) C[W]. The induced
representations on the homology of the fibers of ν are the Springer representations.
Thus, any appropriate category of D-modules on G/B gives a categorified
representation of the Weyl group W.
4.2. Quiver varieties. Nakajima defined a remarkable set of varieties attached

to finite graphs, called quiver varieties. We’ll leave the details of this construction
to other papers [Nak94, Nak98]. Let I be the vertex set of an oriented Dynkin
12
diagram
. For each
pair of dimension vectors v, w, we consider the weights
λ = wi ωi , μ = λ − vi αi . The quiver variety M(v, w) (as defined in [Nak98, (3.5)])
has geometry which reflects the structure of the μ-weight space of the represen-
tation with highest weight λ. It is often useful to consider the union v M(v, w)
which controls the whole structure of the representation. Nakajima also defines a
Lagrangian subvariety Z ⊂ v,v M(v, w) × M(v , w) which plays a similar role to
the Steinberg. The principal results of [Nak98] show that:
12 We’ll only consider the Dynkin case for simplicity, but the results carry over the general case
with various minor caveats.

Theorem 4.10. We have a surjective algebra map U(g) → Htop BM

(Z). The induced
representations on the top homology of the fibers of ν are the different irreducible represen-
tations with highest weight ≤ λ.
In this case, the category of Harish-Chandra bimodules is a categorification
of a quotient of the universal enveloping algebra. While we can study it in its
own terms, it is also closely related to the categorification of these enveloping
algebras as developed by Khovanov, Lauda and Rouquier [KL10, Rou]. In fact, it
is a quotient of the categorified universal enveloping algebra, in a certain sense.
Theorem 4.11 ([Weba, 3.1, 3.3 & 3.9]). There is a 2-functor U → HC which is
“surjective” (i.e. essentially surjective and full) categorifying the map of Theorem 4.10.
Thus, when applied to quiver varieties, we obtain a geometric avatar of the
2-category U. As in previous cases, this gives a natural geometric avatar for
canonical bases, this time of the whole universal enveloping algebra [Weba, 3.13].
References
[Ati67] M. F. Atiyah, K-theory, Lecture notes by D. W. Anderson, W. A. Benjamin, Inc., New York-
Amsterdam, 1967. MR0224083
[BL94] J. Bernstein and V. Lunts, Equivariant sheaves and functors, Lecture Notes in Mathematics,
vol. 1578, Springer-Verlag, Berlin, 1994. MR1299527
[BLPW] T. Braden, A. Licata, N. Proudfoot, and B. Webster, Quantizations of conical symplectic resolutions
II: category O, arXiv:1407.0964.
[BPW] T. Braden, N. J. Proudfoot, and B. Webster, Quantizations of conical symplectic resolutions I: local
and global structure, arXiv:1208.3863.
[BS58] A. Borel and J.-P. Serre, Le théorème de Riemann-Roch (French), Bull. Soc. Math. France 86
(1958), 97–136. MR0116022
[CG97] N. Chriss and V. Ginzburg, Representation theory and complex geometry, Birkhäuser Boston,
Inc., Boston, MA, 1997. MR1433132
[dCM10] M. A. de Cataldo and L. Migliorini, What is. . . a perverse sheaf?, Notices Amer. Math. Soc. 57
(2010), no. 5, 632–634. MR2664042
[Del77] P. Deligne, Cohomologie étale, Lecture Notes in Mathematics, Vol. 569, Springer-Verlag, Berlin-
New York, 1977. Séminaire de Géométrie Algébrique du Bois-Marie SGA 4 12 ; Avec la collab-
oration de J. F. Boutot, A. Grothendieck, L. Illusie et J. L. Verdier. MR0463174
[GM99] M. Grinberg and R. MacPherson, Euler characteristics and Lagrangian intersections, Symplectic
geometry and topology (Park City, UT, 1997), IAS/Park City Math. Ser., vol. 7, Amer. Math.
Soc., Providence, RI, 1999, pp. 265–293. MR1702946
[Har77] R. Hartshorne, Algebraic geometry, Springer-Verlag, New York-Heidelberg, 1977. Graduate
Texts in Mathematics, No. 52. MR0463157
[HTT08] R. Hotta, K. Takeuchi, and T. Tanisaki, D-modules, perverse sheaves, and representation theory,
Progress in Mathematics, vol. 236, Birkhäuser Boston, Inc., Boston, MA, 2008. Translated from
the 1995 Japanese edition by Takeuchi. MR2357361
[Iwa64] N. Iwahori, On the structure of a Hecke ring of a Chevalley group over a finite field, J. Fac. Sci. Univ.
Tokyo Sect. I 10 (1964), 215–236 (1964). MR0165016
[Kar08] M. Karoubi, K-theory, Classics in Mathematics, Springer-Verlag, Berlin, 2008. An introduction;
Reprint of the 1978 edition; With a new postface by the author and a list of errata. MR2458205
[KL10] M. Khovanov and A. D. Lauda, A categorification of quantum sl(n), Quantum Topol. 1 (2010),
no. 1, 1–92, DOI 10.4171/QT/1. MR2628852
[KS12] M. Kashiwara and P. Schapira, Deformation quantization modules (English, with English and
French summaries), Astérisque 345 (2012), xii+147. MR3012169
[KW01] R. Kiehl and R. Weissauer, Weil conjectures, perverse sheaves and l’adic Fourier transform, Ergeb-
nisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in
Mathematics [Results in Mathematics and Related Areas. 3rd Series. A Series of Modern
Surveys in Mathematics], vol. 42, Springer-Verlag, Berlin, 2001. MR1855066
22 BEN WEBSTER
[KWWY] J. Kamnitzer, B. Webster, A. Weekes, and O. Yacobi, Yangians and quantizations of slices in the
affine Grassmannian, arXiv:1209.0349.
[Lau10] A. D. Lauda, A categorification of quantum sl(2), Adv. Math. 225 (2010), no. 6, 3327–3424, DOI
10.1016/j.aim.2010.06.003. MR2729010
[Lau11] A. D. Lauda, Categorified quantum sl(2) and equivariant cohomology of iterated flag varieties, Al-
gebr. Represent. Theory 14 (2011), no. 2, 253–282, DOI 10.1007/s10468-009-9188-8. MR2776785
[Lus91] G. Lusztig, Quivers, perverse sheaves, and quantized enveloping algebras, J. Amer. Math. Soc. 4
(1991), no. 2, 365–421, DOI 10.2307/2939279. MR1088333
[Lus93] G. Lusztig, Introduction to quantum groups, Progress in Mathematics, vol. 110, Birkhäuser
Boston, Inc., Boston, MA, 1993. MR1227098
[Mac74] R. D. MacPherson, Chern classes for singular algebraic varieties, Ann. of Math. (2) 100 (1974),
423–432. MR0361141
[Mil80] J. S. Milne, Étale cohomology, Princeton Mathematical Series, vol. 33, Princeton University
Press, Princeton, N.J., 1980. MR559531
[Mil13] J. S. Milne, Lectures on etale cohomology (v2.21), 2013, Available at www.jmilne.org/math/,
p. 202.
[Nak94] H. Nakajima, Instantons on ALE spaces, quiver varieties, and Kac-Moody algebras, Duke Math. J.
76 (1994), no. 2, 365–416, DOI 10.1215/S0012-7094-94-07613-8. MR1302318
[Nak98] H. Nakajima, Quiver varieties and Kac-Moody algebras, Duke Math. J. 91 (1998), no. 3, 515–560,
DOI 10.1215/S0012-7094-98-09120-7. MR1604167
[Pro08] N. J. Proudfoot, A survey of hypertoric geometry and topology, Toric topology, Contemp. Math.,
vol. 460, Amer. Math. Soc., Providence, RI, 2008, pp. 323–338, DOI 10.1090/conm/460/09027.
MR2428365
[Rin90] C. M. Ringel, Hall algebras and quantum groups, Invent. Math. 101 (1990), no. 3, 583–591, DOI
10.1007/BF01231516. MR1062796
[Rou] Raphael Rouquier, 2-Kac-Moody algebras, arXiv:0812.5023.
[Rou12] R. Rouquier, Quiver Hecke algebras and 2-Lie algebras, Algebra Colloq. 19 (2012), no. 2, 359–410,
DOI 10.1142/S1005386712000247. MR2908731
[Sch] O. Schiffmann, Lectures on Hall algebras, arXiv:math/0611617.
[Ser56] J.-P. Serre, Géométrie algébrique et géométrie analytique (French), Ann. Inst. Fourier, Grenoble 6
(1955), 1–42. MR0082175
[VV11] M. Varagnolo and E. Vasserot, Canonical bases and KLR-algebras, J. Reine Angew. Math. 659
(2011), 67–100, DOI 10.1515/CRELLE.2011.068. MR2837011
[Weba] B. Webster, A categorical action on quantized quiver varieties, arXiv:1208.5957.
[Webb] , Comparison of canonical bases for Schur and universal enveloping algebras.,
arXiv:1503.08734.
[WW] B. Webster and G. Williamson, The bounded below equivariant derived category, In preparation.
Preliminary version available at http://people.virginia.edu/˜btw4e/publications.
html.
Department of Mathematics, University of Virginia, Charlottesville, Virginia

E-mail address: bwebster@virginia.edu
Volume 684, 2017
A geometric realization of modified quantum algebras
Yiqiang Li
Abstract. A geometric construction of Lusztig’s modified quantum algebra

of symmetric type is presented by using certain localized equivariant derived
categories of double framed representation varieties of quivers.
1. Introduction
Let U̇ be Lusztig’s modified quantum algebra of symmetric type and Ḃ its
canonical basis. This paper is an attempt to geometrize the pair (U̇, Ḃ) by using
the geometry of double framed representation varieties EΩ of quivers. There are
several works in literature strongly suggesting the existence of such a construction.
Meanwhile, there are obstacles in getting this construction, largely because EΩ is
not really in complete analogy with partial flag varieties used in Beilinson-Lusztig-
MacPherson’s construction of the pair (U̇, Ḃ) of type A in [BLM90]. As a matter
of fact, partial flag varieties are geometric-invariant-theory (GIT) quotients of cer-
tain open subvarieties of EΩ of type A. Although the GIT quotients still make
sense in other types, they do not produce a construction of the pair (U̇, Ḃ). We
refer to [Re08] and [Li10a] for the progress in this direction. In another direction,
Zheng constructed in [Zh08] a set of endomorphism functors between certain local-
ized equivariant derived categories of framed representation varieties and showed
that it satisfies the defining relations of U̇. It is natural to see if similar localiza-
tion in EΩ helps us in obtaining the pair (U̇, Ḃ). Indeed, it does as we obtain the
following results in this paper.
Theorem 1.1. (1) Similar localization Md of equivariant derived cate-
gories of EΩ admits an associative convolution product.
(2) The convolution product is independent of the choices of orientations of
quivers.
(3) A set of bounded complexes in Md satisfies the defining relations of the
integral form of U̇.
(4) There is a surjective algebra homomorphism from the integral form of U̇
to the Grothedieck group of Md .
(5) Our construction is compatible with that of [BLM90] and [Zh08].
2010 Mathematics Subject Classification. Primary 17B37, 14L30, 14F05, 14F43.

Key words and phrases. Modified quantum algebra, canonical basis, equivariant derived cat-
egory, equivariant perverse sheaf, framed representation variety.
2017
c American Mathematical Society
23
24 YIQIANG LI
(6) Up to an assumption, the complexes involved in this construction are

semisimple.
(7) The canonical basis elements get sent to the simple perverse sheaves or
zero, up to shifts, arising in this construction in the special case corre-
sponding to the one studied in [BLM90].
(8) In the setting of item (7), the simple perverse sheaves are exactly those
microlocal perverse sheaves ([KS90], [GMV05], [W04]) defined by using
Nakajima’s quiver varieties ([N98]-[N00]).
The convolution product on the category Md is obtained by using left adjoints
of the localization functors, whose existence is shown in this paper, and the gen-
eral direct/inverse image functors with compact support introduced by Bernstein
and Lunts ([BL94], [LO08b]). The assumption mentioned in Theorem 1.1 (6)
assumes that the complexes corresponding to the idempotent generators in U̇ are
semisimple. It is proved in full generality in a closely related setting in [W12].
Based on Theorem 1.1 (7) and (8), we conjecture that the assumption holds in our
setting in full generality also and that the simple perverse sheaves arising from this
construction corresponds to elements in Ḃ.
It will be very interesting to relate the construction to the Harish-Chandra-
bimodule construction in [MS08, Theorem47] via localization, that in affine type
A cases in [GV93] and [L99], the one in type D cases in [Li10a], to the geometric
realization of quantum affine algebras in [N00] (in view of [T87] and [G94]), and
to the categorification of U̇ in [KL08a]-[KL08c], [R08] and [MSV10].
Acknowledgement. We thank Professor Z. Lin, H. Nakajima, M. Shimozono,
W. Wang and B. Webster for interesting discussions. We thank Professor C. Strop-
pel for pointing out the connection with [MS08]. A large part of the work was
done while the author was visiting Virginia Tech. This work is partially supported
by NSF grant: DMS-1160351.
Contents
1. Introduction
2. Preliminary, I
3. Preliminary, II
4. Convolution product
5. Defining relation
6. Algebra Kd
7. Relation with the work [Zh08]
8. BLM case
References
2. Preliminary, I
We shall recall the definition of a symmetric quantum modified algebra from
[L93].
2.1. Symmetric Cartan datum, root datum and graph. Let I be a finite
set, and
· : Z[I] × Z[I] → Z
A GEOMETRIC REALIZATION OF MODIFIED QUANTUM ALGEBRAS 25
be a symmetric bilinear form. The pair (I, ·) is called a symmetric Cartan datum
if
i · i = 2 and i · j ∈ {0, −1, −2, · · · }, ∀i = j ∈ I.
We call a triple (X, Y, (, )) a root datum of (I, ·) if the following properties are
satisfied:
• X and Y are finitely generated free abelian groups and (, ) : Y × X → Z
is a perfect pairing;
• there are two embeddings I → X, i → αi and I → Y, i → α̌i such that
(α̌i , αj ) = i · j for any i, j ∈ I.
Let X+ = {λ ∈ X|(α̌i , λ) ∈ N, ∀i ∈ I} be the set of all dominant integral
weights.
To a symmetric Cartan datum, associated a graph Γ consisting of the following
data:
Γ = (I, H; , : H → I; ¯ : H → H)
where I is the vertex set and H is the edge set, the maps and are the source
and target maps, respectively, and the map ¯ is the fixed-point-free involution such
that
h = h , h = (h̄) and #{h ∈ H|h = i, h = j} = −i · j, ∀h ∈ H, i = j ∈ I.
2.2. Modified quantum algebra. Let Q(v) be the rational field with the
indeterminate v. Let
v s − v −s
[[s]] = , [[s]]! = [[s]][[s − 1]] · · · [[1]], for any s ∈ N.
v − v −1
The modified quantum algebra U̇ of Lusztig attached to a root datum (X, Y, (, ))
of a symmetric Cartan datum (I, ·) is a Q(v)-algebra without unit determined by
the following generators and relations. The generators are
1λ , Eλ+αi ,λ and Fλ−αi ,λ , ∀i ∈ I, λ ∈ X.
The relations are
(U̇a)
1λ 1λ = δλ,λ 1λ , ∀λ, λ ∈ X;

(U̇b)
Eλ+αi ,λ 1λ = δλ,λ Eλ+αi ,λ , 1λ Eλ+αi ,λ = δλ ,λ+αi Eλ+αi ,λ , ∀i ∈ I, λ, λ ∈ X;
(U̇c)
Fλ−αi ,λ 1λ = δλ,λ Fλ−αi ,λ , 1λ Fλ−αi ,λ = δλ ,λ−αi Fλ−αi ,λ , ∀i ∈ I, λ, λ ∈ X;
(U̇d)
Eλ−αj +αi ,λ−αj Fλ−αj ,λ − Fλ+αi −αj ,λ+αi Eλ+αi ,λ = δij [[(α̌i , λ)]]1λ , ∀i, j ∈ I, λ ∈ X;
(U̇e)
m
(m−p) (p)
(−1)p Eλ+mαi +αj ,λ+pαi +αj Eλ+pαi +αj ,λ+pαi Eλ+pαi ,λ = 0, ∀i = j ∈ I, λ ∈ X;
p=0
(U̇f)
m
(m−p) (p)
(−1)p Fλ−mαi −αj ,λ−pαi −αj Fλ−pαi −αj ,λ−pαi Fλ−pαi ,λ = 0, ∀i = j ∈ I, λ ∈ X;
p=0
26 YIQIANG LI
where we set m = 1 − i · j in the q-Serre relations (U̇e) and (U̇f) and

(n) 1
Eλ+nαi ,λ = Eλ+nαi ,λ+(n−1)αi · · · Eλ+2αi ,λ+αi Eλ+αi ,λ ,
[[n]]!
(n) 1
Fλ−nαi ,λ = Fλ−nαi ,λ−(n−1)αi · · · Fλ−2αi ,λ−αi Fλ−αi ,λ , ∀n ∈ N, i ∈ I, and λ ∈ X.
[[n]]!
Let A = Z[v, v −1 ] be the subring of Laurent polynomials in Q(v). Let AU̇ be

(n) (n)
the A-subalgebra of U̇ generated by the elements 1λ , Eλ+nαi ,λ and Fλ−nαi ,λ for all
λ ∈ X, n ∈ N and i ∈ I. Let Ḃ be the canonical basis of U̇ defined in [L93, 25.2.4].
3. Preliminary, II
We shall recall the equivariant derived category from [BL94], [LMB00],
[LO08a]-[LO08b]. We will use the presentations in [S08] and [WW09].
3.1. Derived category D∗ (X). Let p be a prime number. Fix an algebraic
closure k of the finite field Fp of p elements. All algebraic varieties in this paper
will be over k.
Let l be a prime number different from p. Let Q̄l be an algebraic closure of the
field of l-adic numbers.
We write Db (X) for the bounded derived category of complexes of Q̄l -con-
structible sheaves on an algebraic variety X defined in [BBD82, 2.2.18]. See
also [FK88, I, §12] and [KW01, II 5]. We denote D ∗ (X), ∗ = {φ, +, −}, for
the similar derived category with the term “bounded” replaced by “unbounded”,
“bounded below”, “bounded above”, respectively.
The shift functor will be denoted by “[−]”. The functors Rf∗ , Rf! , Lf ∗ , and
Rf ! associated to a given morphism f : Y → X of varieties will be written as f∗ ,
f! , f ∗ and f ! , respectively, in this paper. We also write K → L → M → for any
distinguished triangle K → L → M → K[1] in the derived categories just defined.
∗
3.2. Equivariant derived category DG (X). Let us recall the equivariant
derived categories following [BL94], [S08], and [WW09].
A morphism f : Y → X of varieties over k is called n-acyclic if for any sheaf
F over X, considered as a complex of sheaves concentrated on the zeroth degree,
then
(1) F τ≤n f∗ f ∗ (F),
and the property (1) holds under base changes, where τ≤n : D(X) → D ≤n (X) is
the truncation functor.
Suppose a linear algebraic group G act on X.
e
A G-equivariant morphism E → X is called a resolution of X (with respect
to G) if the action of G on E is free and its quotient map E → G\E is a locally
trivial fibration with fiber G. We thus have the following diagram
e ē
X ←−−−− E −−−−→ Ē,
where Ē = G\E. Let [G\X] be the category whose objects are all resolutions of X
and morphisms between resolutions of X are defined to be G-equivariant morphisms
over X. By [LMB00, 18.7.5], [G\X] is an algebraic stack of Bernstein-Lunts.
e
A resolution E → X of X is called smooth if e is smooth. If, moreover, ē is
e
n-acyclic, we call the resolution E → X an n-acyclic smooth resolution of X.
A sequence, (En , fn )n∈N , of smooth resolutions of X:

f0 f1 fn
E0 −−−−→ E1 −−−−→ · · · −−−−→ En −−−−→ En+1 −−−−→ · · · ,
is called acyclic if en : En → X is n-acyclic and fn is a closed inclusion, ∀n ∈ N.
e
Given any smooth resolution E → X, the category D+ (Ē|e) of objects f rom
X is the full subcategory of D (Ē) consisting of all objects K such that there is a
+
complex L ∈ D + (X) satisfying e∗ (L) ē∗ (K).

If f : E → E is a morphism of smooth resolutions of X, then it induces a
map f¯ : Ē → Ē , which gives rise to the inverse image functor f¯∗ : D+ (Ē |e ) →
D + (Ē|e).
Given any acyclic sequence, (En , fn )n∈N , of smooth resolutions of X, we then
have a sequence of functors:
f¯∗ f¯∗ f¯∗
D+ (Ē0 |e0 ) ←−−0−−− D+ (Ē1 |e1 ) ←−−1−−− · · · D+ (Ēn |en ) ←−−n−−− D+ (Ēn+1 |en+1 ) ←−−−−− · · · .
−D (Ēn |en ) is defined to be the category whose

+
The inverse limit category ← lim
objects are pairs (An ; φn )n∈N , with An ∈ D+ (Ēn |en ) and φn : f¯n∗ (An+1 )→A
˜ n
isomorphisms; and whose morphisms α : (An ; φn )n∈N → (Bn ; ψn )n∈N are collections
of morphisms αn : An → Bn such that αn φn = ψn f¯n∗ (αn+1 ), ∀n ∈ N.
Definition 3.3. The bounded below G-equivariant derived category of X is
defined to be
DG
+
(X) = ← − D (Ēn |en ).
lim +
Similarly, we define the bounded (resp. bounded above; unbounded) G-equi-

−
variant derived category DG b
(X) (resp. DG (X); DG (X)), with lim←− D (Ēn |en )
+
−
←− D (Ēn |en ) (resp. ← − D (Ēn |en ); lim
←− D(Ēn |en )).
b
replaced by lim lim
We call the objects in the categories just defined equivariant complexes.
The following results are proved in [S08] and [WW09].
Proposition 3.4. (a) Acyclic sequences of smooth resolutions of X exist;
(b) The definitions of equivariant derived categories are independent of the
choices of the acyclic sequences of smooth resolutions of X;
−
(c) The category DG +
(X) (resp. DG b
(X), DG (X)) is a triangulated category.
−
Σ is a distinguished triangle in DG (X) (resp. DG
+ b
(X), DG (X)) if and only
if the projection of Σ to D (Ēn |en ) (resp. D (Ēn |en ), D− (Ēn |en )) is a
+ b
distinguished triangle for all n ∈ N.

For any object (An , φn ) ∈ DG
+
(X), there exists Bn ∈ D + (X) such that ē∗n An
e∗n Bn for any n ∈ N. So en Bn fn∗ ē∗n+1 (An+1 ) e∗n Bn+1 . Thus the stalks of Bn
∗
and Bn+1 at any given points are isomorphic to each other. Hence supp(Bn ) =
supp(Bn+1 ), for all n ∈ N. Therefore, we can define the support of the object
(An , φn ) as follows.
Definition 3.5. The support of the object (An , φn ) ∈ DG
+
(X) is defined to be
the support of the Bn ’s.
−
Similarly, we can define the supports of any objects in DGb
(X), DG (X), and
DG (X).
Finally, we should give another description of the bounded derived category
DGb
(X) in order to talk about the singular supports, which will be defined in
Section 8, of objects in DGb
(X) for varieties X defined over a characteristic zero
algebraically closed field.
28 YIQIANG LI
e ē
Let X ← E → Ē be a smooth resolution of X with respect to G. We define
the category DG b
(X, E) to be the category whose objects are
• triples K = (KX , K̄, θ), where KX ∈ D b (X), K̄ ∈ Db (Ē) and θ : e∗ KX →
ē∗ K̄ is an isomorphism;
and whose morphisms are
• pairs α = (αX , ᾱ) : K → L, with αX : KX → LX ∈ D b (X) and ᾱ : K̄ →
L̄ ∈ Db (Ē) such that θe∗ (αX ) = e∗ (ᾱ)θ.
Given any morphism ν : E → E1 of smooth resolutions of X, it induces a
morphism ν̄ : Ē → Ē1 , and a functor
ν ∗ : DG
b
(X, E1 ) → DG
b
(X, E), (KX , K̄, θ) → (KX , ν̄ ∗ K̄, ξ),
ν∗θ
where ξ is the composition e∗ (KX ) = ν ∗ e∗1 KX → ν ∗ ē∗1 (K̄) = e∗ ν̄ ∗ K̄.
An acyclic sequence (En , fn ) of smooth resolutions of X gives rise to a sequence
of functors
f∗ f∗
DG
b
(X, E0 ) ←−−
0
−− DG
b
(X, E1 ) ←−−
1
−− · · · ←−−−− DG
b
(X, En ) ←−−−− · · · .
The following proposition is proved in [S08, 5.3] and [BL94, Part I, 2].
Proposition 3.6. DG
b
(X) lim
←− D (X, En ).
b
Recall from [BL94, Part I, 3.2] that there exists a functor

(2) ⊗ : DG
b
(X) × DG
b
(X) → DG
b
(X)
which is exact on each variable. Equipped with the tensor functor ⊗, the derived
category DGb
(X) is a tensor triangulated category ([B05]).
We refer to [BL94, I, 5] for the definition of equivariant perverse sheaves.
Recall that to any irreducible, G-invariant, closed subvariety X1 in X, associated
the equivariant intersection cohomology, denoted by ICG (X1 ). Let
(3) G (X1 ) = ICG (X1 )[− dim X1 ].
IC
G (X) = Q̄l,X , the constant sheaf on X.
Note that if X is smooth, IC
∗
Remark 3.7. As in [LMB00, 18.7], the equivariant derived category DG (X)
∗
for ∗ = {φ, +, −, b} is equivalent to the derived category Dc ([G\X], Q̄l ) of complexes
of lisse-étale Q̄l -sheaves on the quotient stack [G\X] of cartesian constructible co-
homologies defined in [LMB00]. We shall identify these two categories throughout
the paper.
3.8. General inverse functor and general direct image functor. Through-
out this paper, we assume that for any homomorphism, say φ : H → G, of linear
algebraic groups,
• H = G × G1 for some linear algebraic group G1 and φ is the projection
to G.
Let Y be an H-space and X a G-space. A morphism f : Y → X is called a
φ-map if f (h.y) = φ(h).f (y) for any h ∈ H, y ∈ Y . Any φ-map f : Y → X gives
rise to a morphism
(4) Qf : [H\Y ] → [G\X]
e
of algebraic stacks ([LMB00]) defined as follows. For any given resolution Y ← E,
fe
the composition X ← E factors through the quotient map q : E → G1 \E due to
the assumption that G1 acts trivially on X. In other words, there exists a unique
e1
morphism X ← G1 \E such that e1 q = f e. Moreover, H\E = G\(G1 \E). For any
morphism ν : E → E in [H\Y ], it naturally induces a morphism ν̄ : Ē → Ē in
[G\X]. The morphism Qf is then defined to be
e e e
Qf : Y ← E → X ←
1
G1 \E, ∀H-resolution Y ← E ∈ [H\Y ];
Qf : E → E → Ē → Ē , ∀ morphism E → E ∈ [H\Y ].
ν ν̄ ν
By [LMB00], [LO08b], and Remark 3.7, the morphism Qf gives rise to the
following functors:
− −
Qf∗ : DH
+
(Y ) → DG
+
(X), Qf! : DH (Y ) → DG (X),
(5)
Qf ∗ : DG (X) → DH (Y ), Qf ! : DG (X) → DH (Y ).
When H = G, we simply write f for the morphism Qf , and f∗ , f! , f ∗ , f ! for the
functors Qf∗ , Qf! , Qf ∗ and Qf ! , respectively. These functors satisfy the follow-
ing properties. We refer to [LMB00] and [LO08a, LO08b] for proofs of these
properties.
Lemma 3.9. The pairs (Qf ∗ , Qf∗ ) and (Qf! , Qf ! ) are adjoint pairs.
Lemma 3.10. Qf ∗ (A ⊗ A ) = Qf ∗ A ⊗ Qf ∗ A , for any A, A ∈ DG
b
(X).
Lemma 3.11. If, moreover, ψ : I → H is a morphism of linear algebraic groups,
and g : Z → Y a ψ-map, we have
Qg ∗ Qf ∗ Q(f g)∗ and Qf∗ Qg∗ Q(f g)∗ .
Lemma 3.12 ([LO08b] (9.6.1)). Qf! (A ⊗ Qf ∗ (B)) Qf! (A) ⊗ B, for any A ∈
− −
DH (Y ) and B ∈ DG (X).
Lemma 3.13. If, moreover, ψ : I → H is a morphism of linear algebraic groups,
and g : Z → Y a ψ-map, we have
Qg ! Qf ! Q(f g)! and Qf! Qg! Q(f g)! .
Suppose that
I = G × G1 × G2 , H = G × G1 , and H1 = G × G2 .
We denote the projections of linear algebraic groups as follows.
ψ φ ψ1 φ1
I −−−−→ H −−−−→ G, and I −−−−→ H1 −−−−→ G.
Assume, further, that we have a cartesian diagram
g
Z −−−−→ Y
⏐ ⏐
⏐
f1
⏐
f
g1
Y1 −−−−→ X,
and the morphisms f , g, f1 and g1 are φ, ψ, ψ1 and φ1 maps, respectively. This
cartesian diagram then gives rise to the following cartesian diagram
Qg
[I\Z] −−−−→ [H\Y ]
⏐ ⏐
(6) ⏐
Qf1
⏐
Qf
Qg1
[H1 \Y1 ] −−−−→ [G\X].
30 YIQIANG LI
In fact, one can define a morphism,

ξ
X → [I\Z],
from the fiber product X := [H1 \Y1 ] ×[G\X] [H\Y ] to [I\Z] as follows. To a triple
T = (E1 → Y1 , E → Y, α : Qg1 (E1 → Y1 ) Qf (E → Y )) in the fiber product
X , we form the fiber product E1 ×(G1 \E) E, on which there is a free I-action
induced from the free H1 (resp. H, G) action on E1 (resp. E, G1 \E). It is clear
that I\E1 ×(G1 \E) E H\E. By the universal property of the cartesian diagram
(g, f ; f1 , g1 ), we see that there is a unique I-equivariant morphism E1 ×(G1 \E) E →
Z. The morphism ξ is defined by sending the triple T to E1 ×(G1 \E) E → Z. By the
universal property of X , there is a morphism ξ1 : [I\Z] → X . It is not difficult to
show that ξξ1 = id[I\Z] and ξ1 ξ = idX . So [I\Z] is isomorphic to the fiber product
X . Therefore, the diagram above is cartesian.
By Theorem 12.1 in [LO08b], the following lemma holds from the above carte-
sian diagram (6) of algebraic stacks.
Lemma 3.14. Qf ∗ Qg1! Qg! Qf1∗ .
3.15. Compatibility with localization. Let A, B and C be three triangu-
lated categories with exact functors
f1 f2
A → B → C.
Suppose that TA , TB and TC are thick subcategories of A, B, and C, respectively.
Then we have three “exact sequences” of categories ([BBD82, 1.4.4])
ι Q ι Q ι Q
0 → TA → A → A/TA → 0, 0 → TB → B → B/TB → 0, 0 → TC → C → C/TC → 0
Moreover, we assume that each localization functor Q admits a left adjoint Q! and
a right adjoint Q∗ . This implies that the functor ι admits a left adjoint ι∗ and a
right adjoint ι! ([V76, §2]) . We form the following functors
F1 = Q ◦ f1 ◦ Q! :A/TA → B/TB ; F2 = Q ◦ f2 ◦ Q! : B/TB → C/TC ;
F3 = Q ◦ f2 f1 ◦ Q! : A/TA → C/TC .
Lemma 3.16. If ι∗ f1 Q! = 0 or Q ◦ f2 ◦ ι = 0, then F2 ◦ F1 = F3 .
Proof. From [V76, 6.7], we have a distinguished triangle of functors
Q! Q → Id → ιι∗ → .
Thus we have a distinguished triangle of functors
Q ◦ f2 (Q! Q)f1 Q! → Q ◦ f2 (Id)f1 Q∗ → Q ◦ f2 (ιι∗ )f1 Q! → .
But the third term is zero by the assumption, and therefore the lemma follows.
Suppose that D is a fourth triangulated category with a thick subcategory

TD such that the localization functor Q : D → D/TD admits a right adjoint Q∗
and a left adjoint Q! . Assume, moreover, that we have a pair of exact functors
g1 : A → D, g2 : D → C such that f2 f1 g2 g1 . We can form the following functors:
G1 = Q ◦ g1 ◦ Q! and G2 = Q ◦ g2 ◦ Q! .
Lemma 3.17. If Q ◦ f2 ◦ ι = 0 and ι∗ g1 Q! = 0, then F2 F1 G2 G1 .
This lemma follows since by Lemma 3.16 we have F2 F1 = Q ◦ f2 f1 ◦ Q! =

Q ◦ g2 g1 ◦ Q! = G2 G1 .
Recall from [B05] that a tensor triangulated category is a triangulated category
equipped with a symmetric monoindal functor which is exact on each variable. A
thick subcategory T of a tensor triangulated category A is called a thick tensor
ideal if for any T ∈ T and A ∈ A, we have T ⊗ A ∈ T . If T is a thick tensor ideal
of A, then the quotient categories A/T is again a tensor triangulated category.
Moreover, the tensor structures are compatible with the localization functor in the
sense that Q(A ⊗ A ) Q(A) ⊗ Q(A ) for any A, A ∈ A.
Lemma 3.18. Assume that A and B are tensor triangulated categories and TA
and TB are thick tensor ideals. Suppose that f1 : A → B is an exact functor such
that f1 (A ⊗ A ) = f1 A ⊗ f1 A and Q ◦ f1 ◦ ι = 0, then
F1 (Q(A) ⊗ Q(A )) F1 (A) ⊗ F1 (A ).
Proof. We have F1 (Q(A) ⊗ Q(A )) F1 Q(A ⊗ A ) = Q ◦ f1 Q! Q(A ⊗ A ).
Consider the distinguished triangle
F1 (Q(A) ⊗ Q(A ) → Q ◦ f1 (A ⊗ A ) → Q ◦ f1 ιι∗ (A ⊗ A ) → .
By assumption, the third term of the above distinguished triangle is zero. So
F1 (Q(A) ⊗ Q(A )) Q ◦ f1 (A ⊗ A ) = F (QA) ⊗ F (QA ).

Lemma 3.19. Assume that A and B are tensor triangulated categories and TA
and TB are thick tensor ideals. Assume, further, that Q! (A⊗Q(A )) Q! (A)⊗A for
any A, A in A. Suppose that h1 : B → A is an exact functor. Let H1 = Q ◦ h1 ◦ Q! .
If Q ◦ h1 ι = 0, Q! ◦ F1 = f1 ◦ Q! , and f1 (A ⊗ h1 (B)) f1 (A) ⊗ B, for A ∈ A and
B ∈ B, then
F1 (Q(A) ⊗ H1 (Q(B))) F1 (Q(A)) ⊗ Q(B).
Proof. By assumption, H1 ◦ Q Q ◦ h1 . So
F1 (Q(A) ⊗ H1 (Q(B))) = F1 (Q(A) ⊗ Q ◦ h1 (B)) Q ◦ f1 ◦ Q! (Q(A) ⊗ Q ◦ h1 (B))
= Q ◦ f1 (Q! Q(A) ⊗ h1 (B)) = Q ◦ f1 (Q! Q(A) ⊗ h1 (B)
= Q(f1 Q! Q(A) ⊗ h1 (B) = Q(Q! ◦ F1 Q(A) ⊗ B) = F1 (Q(A)) ⊗ Q(B).
The lemma follows.
4. Convolution product
4.1. Framed representation variety. Recall from section 2.1 that Γ =
(I, H, , ,¯ ) is a graph attached to a Cartan datum (I, · ). Fix an orientation
Ω of Γ, i.e., Ω is a subset of H such that Ω Ω̄ = H. We call the pair (Γ, Ω) a
quiver. To an I-graded vector space V = ⊕i∈I Vi over k, we set

GV = GL(Vi ),
i∈I
the product of the general linear groups GL(Vi ). To a pair (D, V ) of finite dimen-
sional I-graded vector spaces over the field k, attach the f ramed representation
variety of the quiver (Γ, Ω):
EΩ (D, V ) = ⊕h∈Ω Hom(Vh , Vh ) ⊕ ⊕i∈I Hom(Vi , Di ).
32 YIQIANG LI
Elements in EΩ (D, V ) will be denoted by X = (x, q) where x (resp. q) is in the first

(resp. second) component. The group GD × GV acts on EΩ (D, V ) by conjugation:
(f, g).(x, q) = (x , q ), where xh = gh xh gh−1 −1
, qi = fi qi gi , ∀h ∈ Ω, i ∈ I,
for any (f, g) ∈ GD × GV , and (x, q) ∈ EΩ (D, V ). To each i ∈ I, we set

X(i) = qi + xh : Vi → Di ⊕ Vh .
h∈Ω:h =i h∈Ω:h =i

To a triple (D, V, V ) of I-graded vector spaces, we set
EΩ (D, V, V ) = EΩ (D, V ) ⊕ EΩ (D, V ).
We write EΩ for EΩ (D, V, V ) if it causes no ambiguity. The group
G = GD × GV × GV
acts on EΩ by (f, g, g ).(X, X ) = ((f, g).X, (f, g ).X ), for any (f, g, g ) ∈ G, (X, X )
∈ EΩ .
Similarly, to a quadruple (D, V, V , V ) of I-graded vector spaces, we set
EΩ (D, V, V , V ) = EΩ (D, V ) ⊕ EΩ (D, V ) ⊕ EΩ (D, V ).
The group H = GD × GV × GV × GV acts on EΩ (D, V, V , V ) by
(f, g, g , g ).(X, X , X ) = ((f, g).X, (f, g ).X , (f, g ).X ),
for any (f, g, g , g ) ∈ H, (X, X , X ) ∈ EΩ (D, V, V , V ).
4.2. Fourier-Deligne transform. Let Ω be another orientation of the graph
Γ. The various varieties defined in 4.1 can be defined with respect to Ω and

Ω ∪ Ω . Define a pairing us : EΩ∪Ω (D, V ) → k by us (X ) = h∈Ω\Ω tr(xsh xsh̄ )
s s
for any X i ∈ EΩ∪Ω (D, V s ) where tr(−) is the trace of the endomorphism in the
parenthesis. Fix a non-trivial character χ from the field Fp of p elements to Q̄∗l :=
Q̄l \{0}. Denote by Lχ the local system on k corresponding to χ. Let
(7) Ls = u∗s Lχ .
Let ust : EΩ∪Ω (D, V i , V j ) → k be the pairing defined by ust (X s , X t ) = −us (X s )+
ut (X t ) for any (X s , X t ) ∈ EΩ∪Ω (D, V s , V t ). Via this pairing, we may regard EΩ
as the dual bundle of the vector bundle EΩ over EΩ∩Ω . Note that this pairing is
G-invariant, i.e., ust (g.(X s , X t )) = ust (X s , X t ), for any g ∈ G. We set
(8) Lst = u∗st Lχ .
Consider the diagram
m m
EΩ (D, V s , V t ) ←−−st−− EΩ∪Ω (D, V s , V t ) −−−st
−→ EΩ (D, V s , V t ),
where the morphisms are obvious projections. It is clear that mst and mst are
G-equivariant morphisms. The F ourier-Deligne transform for the vector bundle
EΩ over EΩ∩Ω , associated with the character χ, is the triangulated functor

Ω : DG (EΩ (D, V , V )) → DG (EΩ (D, V , V ))
ΦΩ b s t b s t
(9)

∗
Ω (K) = mst! (mst (K) ⊗
defined to be ΦΩ Lst )[rst ], where rst is the rank of the vector
bundle EΩ → EΩ∩Ω . Note that rst = h∈Ω\Ω dim Vhs dim Vhs + dim Vht dim Vht .
Let a be the map of multiplication by −1 along the fiber of the vector bundle
EΩ over EΩ∩Ω .

Theorem 4.3. The transform ΦΩ
Ω is an equivalence of triangulated categories.
Ω Ω ∗
Moreover, ΦΩ ΦΩ a .
4.4. Localization. To each i ∈ I, we fix an orientation Ωi of the graph Γ such
that i is a source, i.e., all arrows h incident to i have h = i. Let Fi be the closed
subvariety of EΩi consisting of all elements (X, X ) such that either X(i) or X (i)
is not injective. Let Ui be its complement. Thus we have a decomposition
(10) γi : Fi → EΩi ← Ui : βi .
Notice that Fi and Ui are G-invariant. Following Zheng [Zh08], let Ni be the
thick subcategory of DG
b
(EΩ ) generated by the objects K ∈ DG b
(EΩ ) such that the
support of the complex ΦΩ (K) is contained in the subvariety Fi . Let N be the
Ωi
thick subcategory of DG
b
(EΩ ) generated by Ni for all i ∈ I. We define
DG
b
(EΩ ) = DG
b
(EΩ )/N
to be the localization of DG
b
(EΩ ) with respect to the thick subcategory N ([V76],
[KS90]). Let
Q : DGb
(EΩ ) → DGb
(EΩ )
denote the localization functor.

If dim Via > dim Di + h∈Ω:h =i dim Vha + h∈Ω:h =i dim Vha for a = 1 or 2
and some i ∈ I, we have Ni = DG b
(EΩ ). In this case, we have DG b
(EΩ ) = 0.
Recall that supp(K ⊗ L) = supp(K) ∩ supp(L) for any complexes K and L.
From these properties and the fact that ⊗ is exact on each variable, we have that
N is a thick tensor ideal of DG
b
(EΩ ) if DG
b
(EΩ ) is equipped with the derived tensor
functor ⊗. We see that DG (EΩ ) is a tensor triangulated category with the tensor
b
structure inherited from that of DG b

(EΩ ).
Lemma 4.5. The localization functor Q admits a right adjoint Q∗ and a left
adjoint Q! . Moreover, Q∗ and Q! are fully faithful.
Proof. Let Qi : DG b
(EΩ ) → DG b
(EΩ )/Ni be the localization functor with
respect to the thick subcategory Ni . It is well-known, for example, [BBD82,
1.4], [BL94, Theorem 3.4.3], that the functor βi∗ : DG b
(EΩi ) → DG
b
(Ui ) admits a
fully faithful right adjoint βi∗ and a fully faithful left adjoint βi! . Now that the
functor Qi can be identified with the functor βi∗ via the transform ΦΩ Ω , it then
i
admits a fully faithful right adjoint Qi∗ and a fully faithful left adjoint Qi! .
For simplicity, let us assume that the graph Γ consists of only two vertices
i and j. By the universal property of the localization functor Qi , the functor Q
factors through Qi , i.e., there exists a functor Q̂i : DG
b
(EΩ )/Ni → DG b
(EΩ )/N such
that Q = Q̂i ◦ Qi . Similarly, we have Q = Q̂j ◦ Qj for some Q̂j : DG (EΩ )/Nj →
b
DGb
(EΩ )/N . Let us consider the following diagram
Qj Q̂j
DG
b
(EΩ ) −−−−→ DG
b
(EΩ )/Nj −−−−→ DG
b
(EΩ )/N
⏐
⏐
Qi Qj∗
DG
b
(EΩ )/Ni .
Given any K ∈ N , we have Qi Qj∗ Qj (K) = 0. Indeed, if K ∈ Nj , then Qj (K) = 0.
So Qi Qj∗ Qj (K) = 0. If K ∈ Ni , we have supp(ΦΩΩ Qj∗ Qj (K)) = supp(ΦΩ K) ⊆
i Ωi
34 YIQIANG LI
Fi . So Qj∗ Qj (K) ∈ Ni . Thus, Qi Qj∗ Qj (K) = 0. If K1 → K → K2 → is a distin-

guished triangle such that K1 and K2 are in either Ni or Nj , then Qi Qj Qj ∗ (K) = 0
because the functor Qi Qj Qj ∗ is exact. Since any object in N is generated by objects
from Ni we see that Qi Qj∗ Qj (K) = 0 for any K ∈ N .
From this, we see that the functor Qi Qj∗ Qj factors through Q, i.e., there is a
unique exact functor Q̂i∗ : DGb
(EΩ )/N → DG b
(EΩ )/Ni such that
(11) Q̂i∗ Q = Qi Qj∗ Qj .
By applying Qj∗ to (11) and using the fact that Qj Qj∗ id, we have Q̂i∗ Q̂j =
Qi Qj∗ .
Let ηj : Id → Qj∗ Qj and j : Qj Qj∗ → Id be the associated natural transfor-
mations of the adjoint pair (Qj , Qj∗ ). We set η̂i : Id → Q̂i∗ Q̂i be the composition
Qi ηj
Qi → Qi Qj∗ Qj = Q̂i∗ Q̂j Qj = Q̂i∗ Q̂i Qi ,
and î : Q̂i Q̂i∗ → Id be the composition
Q̂j j
Q̂i Q̂i∗ Q̂j = Q̂i Qi Qj∗ = Q̂j Qj Qj∗ → Q̂j .
The fact that (Qj , Qj∗ ) is an adjoint pair implies that the compositions of natural
transformations
ηj Qj∗ Qj∗ j Qj ηj j Qj
(12) Qj∗ → Qj∗ Qj Qj∗ → Qj∗ , Qj → Qj Qj∗ Qj → Qj
are the identity transformation Id : Qj∗ → Qj∗ and id : Qj → Qj , respectively.
By applying Qi to the first term in (12) and using the definitions of î and η̂i ,
we get that the following composition of natural transformations is again identity
transformation
Qi ηj Qj∗ Qi Qj∗ j
Qi Qj∗ −−−−−−→ Qi Qj∗ Qj Qj∗ −−−−−→ Qi Qj∗

η̂i Q̂i∗ Q̂j Q̂i∗ î Q̂j
Q̂i∗ Q̂j −−−−−→ Q̂i∗ Q̂i Q̂i∗ Q̂j −−−−−→ Q̂i∗ Q̂j ,
where the vertical middle term comes from the following identity:
Qi Qj∗ Qj Qj∗ = Q̂i∗ Q̂j Qj Qj∗ = Q̂i∗ Q̂i Qi Qj∗ = Q̂i∗ Q̂i Q̂i∗ Q̂j .
In other words, we have
(13) (Q̂i∗ î )(η̂i Q̂i∗ ) = IdQ̂i∗ , i Q̂i )(Q̂i η̂i ) = IdQ̂i ,
(ˆ
where the second identity can be obtained by applying Q̂j to the second term in
(12). It is well known, for example [M71], that the pair (Q̂i , Q̂i∗ ) together with the
transformations (ˆ i , η̂i ) and the above two identities (13) make the pair (Q̂i , Q̂i∗ )
an adjoint pair. Moreover, the fact that j is a natural isomorphism implies that î
is a natural isomorphism, which is equivalent to say that Q̂i∗ is fully faithful. The
existence and the fully-faithfulness of the left adjoint Q̂i! of Q̂i can be proved in a
similar way.
Since Qi∗ and Q̂i∗ are fully faithful right adjoints of Qi and Q̂i , respectively, the
composition functor Q∗ = Qi∗ ◦ Q̂i∗ is the fully faithful right adjoint of Q = Q̂i Qi .
Similary, Q! = Qi! ◦ Q̂i! is the fully faithful left adjoint of Q. The lemma is proved
for the graph with only two vertices i and j.
In general, let us order the vertex set I as i1 , · · · , in . Define a sequence of thick

subcategories:
N1 ⊆ · · · ⊆ Nm ⊆ · · · ⊆ Nn = N ,
where Nm is the thick subcategory generated by the subcategories Ni1 , · · · , Nim .
Then the functor Q is the composition of the following functors
Qm
DG
b
(EΩ ) → DG
b
(EΩ )/N1 → · · · → DG
b
(EΩ )/Nm → DG
b
(EΩ )/Nm+1 → · · · → DG
b
(EΩ )/N .
Each Qm admits a fully faithful right adjoint Qm∗ and a fully faithful left adjoint
Qm! due to the fact that the functor Qim : DG b
(EΩ ) → DGb
(EΩ )/Nim admits fully
faithful right and left adjoint functors. This can be proved inductively in a similar
manner as the proof in the case when the graph has only two vertices, with the pair
(Ni , Nj ) replaced by (Nm , Nim+1 ). From this observation, we see that the functors
Q∗ = Q1∗ ◦ · · · ◦ Qn∗ and Q! = Q1! ◦ · · · ◦ Qn!
are the fully faithful right and left adjoint functors of Q, respectively.
A implicit property in the above proof of Lemma 4.5 for Γ = {i, j} is that Q̂j
admits a fully faithful right adjoint Q̂j∗ and the composition Qj∗ Q̂j∗ is the right
adjoint of Q. By the uniqueness of the right adjoint of Q, we have Qj∗ Q̂j∗ = Qi∗ Q̂i∗ .
Now we have
Qj∗ Q̂j∗ Q = Qj∗ Q̂j∗ Q̂i Qi = Qj∗ Qj Qi∗ Qi , Qi∗ Q̂i∗ Q = Qi∗ Q̂i∗ Q̂j Qj = Qi∗ Qi Qj∗ Qj .
So Q∗ Q(K) = Qj∗ Qj ◦ Qi∗ Qi (K) = Qi∗ Qi ◦ Qj∗ Qj (K). So it makes sense to state
the identities as follows:

∗ Ωi ∗ Ωi
Q∗ (Q(K)) = ΦΩ
Ωi βi∗ βi ΦΩ (K) and Q! (Q(K)) = ΦΩ
Ωi βi! βi ΦΩ (K)
i∈I i∈I
for any K ∈ DG
b
(EΩ ).
It is clear that such an expression works for an arbitrary
graph too. We also have
(14) Q! (A ⊗ Q(B)) Q! (A) ⊗ B, ∀A ∈ DG
b
(EΩ )/N , B ∈ DG
b
(EΩ ).
This is because each pair (Qa! , Qa ) in the proof of Lemma 4.5 has such a property.
Moreover,
Lemma 4.6. (a) The inclusion ι : N → DG b
(EΩ ) admits a left adjoint ι∗
!
and a right adjoint ι .
(b) One has Qι = 0, ι∗ Q! = 0, ι! Q∗ = 0, and, ∀A ∈ N , B ∈ DG b
(EΩ )/N ,
Hom(Q! B, ιA) = 0 and Hom(ιA, Q∗ B) = 0;
(c) For any K ∈ DG
b
(EΩ ), there are distinguished triangles
Q! Q(K) → K → ιι∗ (K) → and ιι! (K) → K → Q∗ Q(K) →;
(d) The functors ι, Q∗ , Q! are fully faithful, i.e., the following adjunctions are
isomorphic:
ι∗ ι → Id → ι! ι and QQ∗ → Id → QQ! .
Proof. These results follow from Lemma 4.5 and results from [V76, Prop.
6.5, 6.6, 6.7] and [KS06, Ch. 7, 10].
36 YIQIANG LI
Remark 4.7. (1). Comparing Lemma 4.5-4.6 with [BBD82, 1.4.3], [BL94],
we may regard DG b
(EΩ )/N as the equivariant derived category DG b
(U ) of an “imag-
inary” open subvariety U of EΩ . It is not clear to the author if the existence of
such a variety can be justified by the method in [B05].
∗
(2). One may define DG (EΩ (D, V, V )) for ∗ = {+, −} in a similar way and
Lemmas 4.5, 4.6 and (27) still hold, where the thick subcategory Ni can be defined
to be the one consisting of all objects K such that the support of the cohomology
sheaf H a (K) is in Fi for any a.
4.8. Convolution product. Let
pst : EΩ (D, V 1 , V 2 , V 3 ) → EΩ (D, V s , V t ), φ : H → G,
denote the projections to the (s, t)-components, where the spaces EΩ ’s, the groups
H and G are defined in Section 4.1. It is clear that pst is a φ-map. So, by Section 3.8
(4), we have a morphism of algebraic stacks:
Qpst : [H\EΩ (D, V 1 , V 2 , V 3 )] → [G\EΩ (D, V s , V t )].
From Section 3.8 (5), we have the following functors:
Qp∗st : DG
b
(EΩ (D, V s , V t )) → DH
b
(EΩ (D, V 1 , V 2 , V 3 ));
(15) − −
(Qpst )! : DH (EΩ (D, V 1 , V 2 , V 3 )) → DG (EΩ (D, V s , V t )).
Lemma 4.9. We have
(16) Q ◦ Qp∗st ◦ ι = 0 and ι∗ ◦ Qpst! ◦ Q! = 0.
Proof. For the first identity, it suffices to show that Qp∗st (Ni ) ⊆ Ni for
any i ∈ I. Fix an orientation Ωi such that i is a source in Ωi . The morphism
pst is a morphism of vector bundles over the base space EΩ∩Ωi (D, V s , V t ). Let
pst : EΩ (D, V s , V t ) → EΩ (D, V 1 , V 2 , V 3 ) be the transpose of pst . We identify
EΩ (D, V s , V t ) with EΩi (D, V s , V t ) under the pairing ust and EΩ (D, V 1 , V 2 , V 3 )
with EΩi (D, V s , V t ) × EΩ (D, V u ) where u is the number such that {s, t, u} =
{1, 2, 3}. Then the transpose pst is noting but the inclusion of EΩi (D, V s , V t ) into
EΩi (D, V s , V t ) × EΩ (D, V u ). Let ψ : G → H be the obvious imbedding. It is clear
that pst is a ψ-map. This induces a morphism of algebraic stacks:
Qpst : [G\EΩi (D, V s , V t )] → [H\EΩi (D, V s , V t ) × EΩ (D, V u )].
Let H act on EΩi (D, V s , V t ) by declaring that GV u acts trivially. Then Qpst
factors through [H\EΩi (D, V s , V t )]. So the functor Qpst! is a composition of the
integration functor
Ind! : DG (EΩi (D, V s , V t )) → DH (EΩi (D, V s , V t ))
from [BL94, 3.7] and the functor
pij! : DH (EΩi (D, V s , V t )) → DH (EΩi (D, V s , V t ) × EΩ (D, V u )).
Since GV u acts trivially on EΩi (D, V s , V t ), we see that the supports of the
complexes K and Ind! (K) are the same. Moreover, we have supp(pij! (K)) =
pij supp(K). So the decomposition pst! Ind! of Qpst! implies that
(17) Qpst! (Ni ) ⊆ Ni .
Note that the notion of Ni also makes sense on DH (EΩi (D, V s , V t ) × EΩ (D, V u )).
Now we have
(18) Qp∗st (ΦΩ
Ω K) = ΦΩ Qpst! (K)[f1 − f2 ],
Ω
where f1 and f2 are the ranks of EΩ (D, V s , V t ) and EΩ (D, V 1 , V 2 , V 3 ) over

EΩ∩Ωi (D, V s , V t ), respectively. This can be proved by a similar argument as the
proof of Theorem 1.2.2.4 in [L87], see also [KW01, Theorem 13.2]. By combining
(17) and (18), we get the statement Qp∗st (Ni ) ⊆ Ni for any i ∈ I.
The second identity can be shown similarly by replacing (17) by ι∗i (Qpst )∗ Qi! =
0 where ι∗i is the left adjoint of the inclusion functor ιi of Ni into the ambient
category and Qi! is the left adjoint to the quotient functor Qi with respect to
Ni .
We set
∗
Pst = Q ◦ Qp∗st ◦ Q! : DG
b
(EΩ (D, V s , V t )) → DH
b
(EΩ (D, V 1 , V 2 , V 3 ));
(19) − −
Pst! = Q ◦ (Qpst )! ◦ Q! : DH (EΩ (D, V 1 , V 2 , V 3 )) → DG (EΩ (D, V s , V t )).
By Lemma 4.6 (c) and (16) in the above Lemma, we have
∗
Lemma 4.10. Pst ◦ Q = Q ◦ Qp∗st and Q! Pst! = Qpst! Q! .
− −
To any objects K ∈ DG (EΩ (D, V 1 , V 2 )) and L ∈ DG (EΩ (D, V 2 , V 3 )), we
associate
∗ ∗ −
(20) K · L = P13! (P12 (K) ⊗ P23 (L)) ∈ DG (EΩ (D, V 1 , V 3 )).
−
If, in addition, M ∈ DG (EΩ (D, V 3 , V 4 )), we have
Proposition 4.11. (K · L) · M K · (L · M ).
Proof. Let Xs = EΩ (D, V s ) and Gs = GV s , ∀s = 1, 2, 3, 4. Let

qst : X1 × X3 × X4 → Xs × Xt , φst : GD × Ga → GD × Ga ;
a=1,3,4 a=s,t

rstu : X1 × X2 × X3 × X4 → Xs × Xt × Xu , φstu : GD × Ga → GD × Ga ;
a=1,2,3,4 a=s,t,u

sst : X1 × X2 × X3 × X4 → Xi × Xj , φst : GD × Ga → GD × Gs × Gt ;
a=1,2,3,4
be the self-explained projections. It is clear that qst , rstu and sst are φst -map, φstu -
map, and φst -map, respectively. Similar to the functors Pst! and Pst ∗
, we define the
∗ ∗ ∗
functors Qst! , Qst (resp. Rstu! , Rstu ; Sst! , Sst ) for the map qst (resp. rstu , sst ). By
definition,
(K·L)·M = Q14! (Q∗13 (K·L)⊗Q∗34 (M )) = Q14! (Q∗13 P13! (P12
∗ ∗
(K)⊗P23 (L))⊗Q∗34 (M )).
Since the square (r123 , p13 ; r134 , q13 ) is cartesian and by Lemma 3.14, we have
∗ ∗
(21) Qq13 Qp13! Qr134! Qr123 .
∗
By (16), we have Q ◦ Qr123 ◦ ι = 0 and ι∗ ◦ Qp13! ◦ Q! = 0. So by Lemma 3.17 and
(21),
(22) Q∗13 P13! = R134! R123
∗
.
Thus
∗ ∗ ∗
(K · L) · M Q14! (R134! R123 (P12 (K) ⊗ P23 (L)) ⊗ Q∗34 (M )).
38 YIQIANG LI
∗
By Lemma 3.12, Lemma 3.19, and the fact that Q ◦ Qr134 ι = 0,
∗ ∗ ∗ ∗
(23) (K · L) · M Q14! R134! (R123 (P12 (K) ⊗ P23 (L)) ⊗ R134 Q∗34 (M )).
By Lemma 3.13, Lemma 3.16, and the fact that ι∗ ◦ Qr134! Q! = 0, we have
(24) Q14! R134! = S14! .
∗
By Lemma 3.10, Lemma 3.18 and the fact that Q ◦ Qr123 ι = 0, we have
∗ ∗ ∗ ∗ ∗ ∗ ∗
(25) R123 (P12 K ⊗ P23 L) = R123 P12 K ⊗ R123 P23 L.
By Lemma 3.11 and Lemma 3.16, we have
∗ ∗ ∗ ∗ ∗ ∗ ∗
(26) R123 P12 = S12 , R123 P23 = S23 , and R134 Q∗34 = S34
∗
.
Combining (23)-(26), we get
∗ ∗ ∗
(K · L) · M S14! (S12 (K) ⊗ S23 (L) ⊗ S34 (M )).
∗ ∗ ∗
Similarly, we can show that K · (L · M ) S14! (S12 (K) ⊗ S23 (L) ⊗ S34 (M )). The
lemma follows.
Remark 4.12. The proofs of the equalities (22)-(26) can be generalized to
similar situations. In the sequel, we simply state the similar equalities without
mentioning how they are proved.
4.13. Independence of choice of orientation. Let Ω be another orienta-
tion of the graph Γ. Let NΩ be the thick subcategory of DG (EΩ ) defined in the

same way as N . One has, by definition, ΦΩ
Ω (N ) = NΩ . So we have an equivalence,

induced by ΦΩ Ω ,

Ω : DG (EΩ ) DG (EΩ ).
ΦΩ b b
(27)
Similarly, one can define the category DHb
(EΩ (D, V 1 , V 2 , V 3 )) and the equiva-
Ω
lence of categories ΦΩ : DH (EΩ (D, V , V , V )) DH
b 1 2 3 b
(EΩ (D, V 1 , V 2 , V 3 )).
−
We define a convolution product “·Ω ” on the categories DG (EΩ (D, V s , V t ))
similar to the one in Section 4.8. The following proposition shows that the convo-
lution products are compatible with the Fourier-Deligne transform.

Proposition 4.14. ΦΩ Ω (K · L) = ΦΩ (K) ·Ω ΦΩ (L), for any objects K ∈
Ω Ω
− −
DG (EΩ (D, V , V )) and L ∈ DG (EΩ (D, V , V )).
1 2 2 3
Proof. Due to the fact that m∗13 is a fully faithful functor and that the con-
dition to define the thick subcategory N on EΩ and EΩ∪Ω are the same, one can
−
deduce that ι∗ m∗13 Q! = 0. From this fact, the functor ΦΩ
Ω : DG (EΩ (D, V , V )) →
1 3
−
DG (EΩ (D, V , V )) can be rewritten as
1 3

∗
Ω (K) = M13! (M13 (K) ⊗ L13 )[r13 ].
ΦΩ
∗
where the notations L13 and r13 are defined in Section 4.2 and M13! and M13 are
obtained from m13! and m∗13 as in (19). By definition, we have

∗ ∗ ∗ ∗ ∗
Ω (K · L) = ΦΩ (P13! (P12 (K) ⊗ P23 (L))) = M13! (M13 P13! (P12 (K) ⊗ P23 (L)) ⊗ L13 )[r13 ].
ΦΩ Ω
Consider the following cartesian diagram

s
EΩ∪Ω (D, V 1 , V 3 ) × EΩ (D, V 2 ) −−−−→ EΩ∪Ω (D, V 1 , V 3 )
⏐ ⏐
⏐
r
⏐
m13
p13
EΩ (D, V 1 , V 2 , V 3 ) −−−−→ EΩ (D, V 1 , V 3 ).
∗
By an argument similar to the proof of (22), we have M13 P13! = S! R∗ . So

∗ ∗ ∗
Ω (K · L) = M13! (S! R (P12 (K) ⊗ P23 (L)) ⊗ L13 )[r13 ]
ΦΩ
(28)
= M13! S! (R∗ P12
∗
(K) ⊗ R∗ P23
∗
(L) ⊗ S ∗ L13 )[r13 ].
On the other hand, we have

∗ ∗
Ω (K) ·Ω ΦΩ (L) = P13! ((P12 ) (ΦΩ (K)) ⊗ (P23 ) (ΦΩ (L)))
ΦΩ Ω Ω Ω
∗ ∗ ∗ ∗
= P13! ((P12 ) M12! (M12 (K) ⊗ L12 )[r12 ] ⊗ (P23 ) M23! (M23 (L) ⊗ L23 )[r23 ]),
∗
where Pst! and (Pst ) are obtained from the projections pst from EΩ (D, V 1 , V 2 , V 3 )
∗
to EΩ (D, V , V ), Lst and rst are from Section 4.2, and the functors Mst!
s t
and Mst
∗
are obtained from the projections mst in Section 4.2 as Pst! and Pst from pst in
(19). Consider the following cartesian diagrams
s
EΩ∪Ω (D, V 1 , V 2 ) × EΩ (D, V 3 ) −−−1−→ EΩ (D, V 1 , V 2 , V 3 )
⏐ ⏐
⏐
r1
⏐
p12
m
EΩ∪Ω (D, V 1 , V 2 ) −−−12
−→ EΩ (D, V 1 , V 2 ),
and
s
EΩ (D, V 1 ) × EΩ∪Ω (D, V 2 , V 3 ) −−−2−→ EΩ (D, V 1 , V 2 , V 3 )
⏐ ⏐
⏐
r2
⏐
p23
m
EΩ∪Ω (D, V 2 , V 3 ) −−−23
−→ EΩ (D, V 2 , V 3 ).
∗
From these cartesian diagrams and similar to the proof of (22), we have (P12 ) M12!
∗ ∗ ∗
= S1! (R1 ) and (P23 ) M23! = S2! (R2 ) . So

∗ ∗ ∗ ∗
Ω (K) ·Ω ΦΩ (L) = P13! (S1! (R1 ) (M12 (K) ⊗ L12 ) ⊗ S2! (R2 ) (M23 (L) ⊗ L23 ))[r12 + r23 ]
ΦΩ Ω

= P13! (S1! ((R1 )∗ M12
∗
(K) ⊗ (R1 )∗ L12 ) ⊗ S2!

((R2 )∗ M23
∗
(L) ⊗ (R2 )∗ (L23 )))[r12 + r23 ]

= P13! S1! ((R1 )∗ M12
∗
(K) ⊗ (R1 )∗ L12 ⊗ (S1 )∗ S2!

((R2 )∗ M23
∗
(L) ⊗ (R2 )∗ (L23 )))[r12 + r23 ].
We form the following cartesian diagram
t
EΩ∪Ω (D, V 1 , V 2 , V 3 ) × E2Ω\Ω (D, V 2 ) −−−2−→ EΩ∪Ω (D, V 1 , V 2 ) × EΩ (D, V 3 )
⏐ ⏐
⏐
t1
⏐
s1
s
EΩ (D, V 1 ) × EΩ∪Ω (D, V 2 , V 3 ) −−−2−→ EΩ (D, V 1 , V 2 , V 3 ).
This cartesian diagram gives rise to the identity, (S1 )∗ S2!

= T2! (T1 )∗ . So

Ω (K) ·Ω ΦΩ (L) =
ΦΩ Ω

= P13! S1! ((R1 )∗ M12
∗
(K) ⊗ (R1 )∗ L12 ⊗ T2!

(T1 )∗ ((R2 )∗ M23
∗
(L) ⊗ (R2 )∗ (L23 )))[r123 ]

= P13! S1! T2! ((T2 )∗ (R1 )∗ M12
∗
K ⊗ (T2 )∗ (R1 )∗ L12 ⊗ (T1 )∗ (R2 )∗ M23
∗
L ⊗ (T1 )∗ (R2 )∗ L23 )[r123 ]

= P13! S1! T2! ((T2 )∗ (R1 )∗ M12
∗
K ⊗ (T1 )∗ (R2 )∗ M23
∗
L ⊗ (T2 )∗ (R1 )∗ L12 ⊗ (T1 )∗ (R2 )∗ L23 )[r123 ].
where r123 = r12 + r23 . Let F1 = EΩ∪Ω (D, V 1 , V 3 ) and

t3 : F ≡ EΩ∪Ω (D, V 1 , V 2 , V 3 )×E2Ω\Ω (D, V 2 ) → Z ≡ F1 ×EΩ (D, V 2 )×E2Ω\Ω (D, V 2 )
40 YIQIANG LI
be the obvious projection. Note that in the component EΩ∪Ω (D, V 2 ), there is a
copy of EΩ\Ω (D, V 2 ), denoted by E1Ω (D, V 2 ). Observe that
p13 s1 t2 = wt3 , m12 r1 t2 = y2 t3 , and m23 r2 t1 = y1 t3 ,
where w is the projection from Z to EΩ (D, V 1 , V 3 ), and y1 and y2 are the projec-
tions from Z to EΩ (D, V 2 , V 3 ) (for y1 ) and EΩ (D, V 1 , V 2 ), respectively. Note that
there are two choices for the projections for each yi , but we choose the unique one
such that the above identities hold. So

∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗
Ω (K) ·Ω ΦΩ (L) = W! T3! ((T3 ) Y2 K ⊗ (T3 ) Y1 L ⊗ (T2 ) (R1 ) L12 ⊗ (T1 ) (R2 ) L23 )[r123 ]
ΦΩ Ω
= W! (Y2∗ K ⊗ Y1∗ L ⊗ T3!

((T2 )∗ (R1 )∗ L12 ⊗ (T1 )∗ (R2 )∗ L23 ))[r123 ].
Let F2 = EΩ∪Ω (D, V 2 ) × EΩ\Ω (D, V 2 ). Thus, F = F1 × F2 . Each component

EiΩ (D, V 2 ) defines a projection, πs,s+1 : F2 → EΩ∪Ω (D, V 2 ), for any s = 1, 2. We
have
r1 t2 = w1 (1 × π23 ) and r2 t1 = w2 (1 × π12 ),
where 1 : F1 → F1 is the identity map and ws is the projection of EΩ∪Ω (D, V 1 , V 2 , V 3 )
to EΩ∪Ω (D, V s , V s+1 ) for any s = 1, 2. Hence,

∗ ∗ ∗ ∗ ∗ ∗
Ω (K)·Ω ΦΩ (L) = W! (Y2 K⊗Y1 L⊗T3! ((1×Π23 ) W1 L12 ⊗(1×Π12 ) W2 L23 ))[r123 ].
ΦΩ Ω
Observe that W1∗ L12 = P1∗ L∗1 ⊗ P2∗ L2 and W2∗ L23 = P3∗ L3 ⊗ P2∗ L∗2 where L∗s is the
dual of the local system Ls in 4.2, and ps are the projections from
EΩ∪Ω (D, V 1 , V 2 , V 3 ) to EΩ∪Ω (D, V s ).
So
T3! ((1 × Π23 )∗ W1∗ L12 ⊗ (1 × Π12 )∗ W2∗ L23 ) =
= T3! (((1 × Π23 )∗ (P1∗ L∗1 ⊗ P2∗ L2 ) ⊗ (1 × Π12 )∗ P2∗ (L∗2 ⊗ P3∗ L3 ))
= T3! ((T3 )∗ P ∗ L13 ⊗ (1 × Π23 )∗ P2∗ L2 ⊗ (1 × Π12 )∗ P2∗ (L∗2 ))
= P ∗ L13 ⊗ T3! (1 × Π23 )∗ P2∗ L2 ⊗ (1 × Π12 )∗ P2∗ (L∗2 )),
where P ∗ comes from the projection p : Z → EΩ∪Ω (V 1 , V 3 ). By a similar argument
as [KW01, p. 44], we have
T3! (1 × Π23 )∗ P2∗ L2 ⊗ (1 × Π12 )∗ P2∗ (L∗2 )) = Δ! Q̄l,Z1 [−2r ].
where Δ! is from the diagonal map δ : Z1 ≡ EΩ∪Ω (D, V 1 , V 3 ) × EΩ (D, V 2 ) → Z

and r is the rank of t3 , which is equal to h∈Ω\Ω dim Vh2 dim Vh2 . So
P ∗ L13 ⊗ T3! (1 × Π23 )∗ P2∗ L2 ⊗ (1 × Π12 )∗ P2∗ (L∗2 )) = P ∗ L13 ⊗ Δ! Q̄l,Z1 [−2r ]
= Δ! S ∗ L13 [−2r ].
Therefore,

∗ ∗ ∗
Ω (K) ·Ω ΦΩ (L) = W! (Y2 K ⊗ Y1 L ⊗ Δ! S L13 )[r123 − 2r ]
ΦΩ Ω
(29) = W! Δ! (Δ∗ Y2∗ K ⊗ Δ∗ Y1∗ L ⊗ S ∗ L13 )[r13 ]

= M13! S! (R∗ P12
∗
(K) ⊗ R∗ P23
∗
(L) ⊗ S ∗ L13 )[r13 ],
where the last identity comes from the observation that wδ = m13 s, y2 δ = p12 r
and y1 δ = p23 r. The proposition follows from (28) and (29).
5. Defining relation
5.1. Generator. Given any pair (X 1 , X 2 ) ∈ EΩ (D, V 1 , V 2 ), we write
“ X 1 → X 2 ”
if there exists an I-graded inclusion ρ : V 1 → V 2 such that ρh x1h = x2h ρh , qi1 =
qi2 ρi , for any h in Ω and i in I. We also write “ρ : X 1 → X 2 ” for such a ρ and
ρ
“X 1 → X 2 ” for the triple (X 1 , X 2 , ρ). Consider the smooth variety
(30)
ZΩ ≡ ZΩ (D, V 1 , V 2 ) = {(X 1 , X 2 , ρ)|(X 1 , X 2 ) ∈ EΩ (D, V 1 , V 2 ) and ρ : X 1 → X 2 }.
We have a diagram
π π
EΩ (D, V 1 ) ←−−1−− ZΩ (D, V 1 , V 2 ) −−−2−→ EΩ (D, V 2 ),
⏐
⏐
(31) π12
p1 p2
EΩ (D, V 1 ) ←−−−− EΩ (D, V 1 , V 2 ) −−−−→ EΩ (D, V 2 ),
where π1 and p1 are projections to the first components, π2 and p2 are the projec-
tions to the second components, and π12 is the projection to (1, 2) components.
Similar to ZΩ (D, V 1 , V 2 ), let ZtΩ (D, V 1 , V 2 ) = {(X 1 , X 2 , ρ)|ρ : X 2 → X 1 }.
Let π12 denote the projection ZtΩ (D, V 1 , V 2 ) → EΩ (D, V 1 , V 2 ). Note that
ZΩ (D, V 1 , V 2 ) ZtΩ (D, V 2 , V 1 ).
We set the following complexes in DG b
(EΩ ) with μ = λ − ν and n ∈ N:

Iμ = Q π12! (Q̄l,ZΩ ) , if dim V 1 = dim V 2 = ν;
(n)

(32) Eμ,μ−nαi = Q π12! (Q̄l,ZΩ )[eμ,nαi ] , if dim V = ν and dim V = ν + ni;
1 2

(n)
Fμ,μ+nαi = Q π12! (Q̄l,ZtΩ )[fμ,nαi ] , if dim V 1 = ν and dim V 2 = ν − ni;
where
(33)

eμ,nαi = n di + νh − (νi + n) and fμ,nαi = n (νi − n) − νh .
(n) (n)
We set Iμ , Eμ,μ−nαi and Fμ,μ+nαi to be zero if μ ∈ X can not be written as the
form μ = λ − ν for some ν ∈ N[I]. We have

Ω (n) (n) (n)
Lemma 5.2. ΦΩ Ω (Iμ ) = Iμ , ΦΩ Eμ,μ−nαi = Eμ,μ−nαi and ΦΩΩ Fμ,μ+nαi =
(n)
Fμ,μ+nαi where the elements on the right-hand sides are complexes in DG
b
(EΩ ) de-
fined in a similar way as the complexes on the left-hand sides.

(n) (n)
Proof. We shall show that ΦΩ Ω Eμ,μ−nαi = Eμ,μ−nαi . The others can be
shown in a similar way. Consider the following diagram
c
ZΩ ←−−−− Ẑ −−−−→ ẐΩ
⏐ ⏐ ⏐
⏐
π12
⏐
β
⏐

m m
EΩ ←−−−− EΩ∪Ω −−−12
12
−→ EΩ ,
42 YIQIANG LI
where the morphisms m12 and m12 are defined in Section 4.2, π12 is defined in (31),
Ẑ = ZΩ ×EΩ EΩ∪Ω , the morphism c is the map by forgetting the components x1h
and x2h for h ∈ Ω\Ω , and the rest of the morphisms are clearly defined.
By definition, we see that ZΩ is a locally closed subvariety in the variety EΩ ×
1 2
i∈I Hom(Vi , Vi ), which, in turn, can be embedded as an open subvariety into a
certain variety EΩ × P, where P is a certain projective variety, by [Z85]. Then the
morphism π12 is a compactifiable morphism defined in [FK88, p. 86]. From this
and the fact that the left square of the above diagram is cartesian, we may apply
the base change for compactifiable morphisms in [FK88, Theorem 8.7] to get the
following identity

∗
Ω (π12! (Q̄l,ZΩ )) = m12! (m12 π12! (Q̄l,ZΩ ) ⊗ L12 )[r12 ]
ΦΩ
= m12! (β! (Q̄l,Ẑ ) ⊗ L12 )[r12 ] = m12! β! (β ∗ L12 )[r12 ],
where L12 and r12 are defined in 4.2.
ρ
Let Ẑ1 be the subvariety of Ẑ defined by the condition x1h → x2h for any

h ∈ Ω \Ω. Observe that that the map c is a vector bundle with fiber dimension

fc = νh2 νh2 + νh1 νh1 ,
h∈Ω\Ω :h =i h∈Ω\Ω :h =i
and the restriction of the map u12 β, where u12 is defined in Section 4.2, to the
fiber c−1 (X, ρ) of c is 0 if c−1 (X, σ) ⊆ Ẑ1 and a non-constant affine linear function,
otherwise. By arguing in exactly the same way as the proof of Proposition 10.2.2
in [L93], we get

Ω (π12! (Q̄l,ZΩ )) π12! (Q̄l,ZΩ [r12 − 2fc ]),
ΦΩ Ω

Ω
where π12 is defined similar to π12 . It is clear that

r12 − 2fc = −n νh + n νh .
h∈Ω\Ω :h =i h∈Ω\Ω :h =i
By combining the above analysis, we have

(n) Ω (n)
ΦΩ
Ω Eμ,μ−nαi =Q π12! (Q̄l,ZΩ [r12 − 2fc ][n(di + νh − (νi + n))] = Eμ,μ−nαi ,
h∈Ω:h =i
where the second equality is due to the following computation.

r12 − 2fc + n(di + νh − (νi + n)) = n(di + νh − (νi + n)).
h∈Ω:h =i h∈Ω :h =i
The lemma follows.
(n)
5.3. Defining relation. We shall show that the complexes Iμ , Eμ,μ−nαi and
(n)
Fμ,μ+nαi satisfy the defining relations of AU̇.
Lemma 5.4. Iμ Iμ = δμ,μ Iμ where μ = λ − ν and ν ∈ N[I].
Proof. Assume that V 1 = V 2 = V 3 has dimension ν. Consider the following
cartesian diagram
s
ZΩ (D, V 1 , V 2 ) ←−−−− Z1 −−−1−→ Z
⏐ ⏐ ⏐
⏐
π12
⏐
r1
⏐
s2
p12 r
EΩ (D, V 1 , V 2 ) ←−−−− EΩ (D, V 1 , V 2 , V 3 ) ←−−
2
−− Z2 ,
where Z1 = ZΩ (D, V 1 , V 2 ) × EΩ (D, V 3 ), Z2 = EΩ (D, V 1 ) × ZΩ (D, V 2 , V 3 ),

ρ1 ρ2
Z = Z1 ×EΩ (D,V 1 ,V 2 ,V 3 ) Z2 = {(X 1 , X 2 , X 3 , ρ1 , ρ2 )|X 1 → X 2 → X 3 }.
and the morphisms are the obvious projections. The cartesian square on the
∗
left gives rise to the identity P12 Π12! (Q̄l,ZΩ (D,V 1 ,V 2 ) ) = R1! (Q̄l,Z1 ). Similarly,
∗
P23 Π12! (Q̄l,ZΩ (D,V 2 ,V 3 ) ) = R2! (Q̄l,Z2 ). So
∗ ∗
Iμ · Iμ = P13! (P12 (Iμ ) ⊗ P23 (Iμ )) = P13! (R1! (Q̄l,Z1 ) ⊗ R2! (Q̄l,Z2 ))
= P13! R1! (Q̄l,Z1 ⊗ R1∗ R2! (Q̄l,Z2 )) = P13! R1! (R1∗ R2! (Q̄l,Z2 )).
The right cartesian square in the above diagram implies that R1∗ R2! = S1! S2∗ . Thus
Iμ · Iμ = P13! R1! (R1∗ R2! (Q̄l,Z2 )) = P13! R1! S1! S2∗ (Q̄l,Z2 ) = P13! R1! S1! (Q̄l,Z ).
Consider the following commutative diagram
t
Z −−−−→ ZΩ (D, V 1 , V 3 )
⏐ ⏐
⏐
r1 s1
⏐
π12
p13
EΩ (D, V 1 , V 2 .V 3 ) −−−−→ EΩ (D, V 1 , V 3 ),
ρ1 ρ2 ρ2 ρ1
where t sends X 1 → X 2 → X 3 to X 1 → X 3 . Then we have
Iμ · Iμ = P13! R1! S1! (Q̄l,Z ) = Π12! T! (Q̄l,Z ).
Observe that t is the quotient map of Z by the group GV 2 , thus the induced
morphism Qt : [H\Z] → [G\ZΩ (D, V 1 , V 3 )] is an isomorphism. Hence T! (Q̄l,Z ) =
Q̄l,ZΩ (D,V 1 ,V 3 ) . Therefore,
Iμ · Iμ = Π12! (Q̄l,ZΩ (D,V 1 ,V 3 ) ) = π12! (Q̄l,ZΩ (D,V 1 ,V 3 ) ) = Iμ .
It is clear that Iμ · Iμ = 0 if μ = μ from the above argument. The lemma
follows.
Lemma 5.5. For any μ and μ , we have
(n) (n) (n) (n)
Eμ,μ−nαi Iμ = δμ−nαi ,μ Eμ,μ−nαi , Iμ Eμ,μ−nαi = δμ ,μ Eμ,μ−nαi ;
(n) (n) (n) (n)
Fμ,μ+nαi Iμ = δμ+nαi ,μ Fμ,μ+nαi , Iμ Fμ,μ+nαi = δμ ,μ Fμ,μ+nαi .
This lemma can be proved in exactly the same way as the proof of Lemma 5.4.
Lemma 5.6. Eμ−αj +αi ,μ−αj Fμ−αj ,μ = Fμ+αi −αj ,μ+αi Eμ+αi ,μ , for any i = j.
Proof. Fix four I-graded vector spaces V s for s = 1, 2, 3, 4 such that
(34) dim V 1 = ν + j − i, dim V 2 = ν + j, dim V 3 = ν and dim V 4 = ν − i.
Let
ρ1 ρ2
Z1 = {(X 1 , X 2 , X 3 , ρ1 , ρ2 )|(X 1 , X 2 , X 3 ) ∈ EΩ (D, V 1 , V 2 , V 3 ), X 1 → X 2 ← X 3 },
and π1 be the projection from Z1 to EΩ (D, V 1 , V 2 , V 3 ). Then an argument similar
to the proof of Lemma 5.4 yields that
Eμ−αj +αi ,μ−αj Fμ−αj ,μ = P13! Π1! (Q̄l,Z1 )[m],
where m = eμ−αj +αi ,αi + fμ−αj ,αj . Denote by V̌ the I\{j}-graded vector space
obtained from V by deleting the component Vj . Let Z be the variety of quadruples
44 YIQIANG LI
(X 1 , X 3 , ρ̌ : V̌ 1 → V̌ 3 , σj : Vj3 → Vj1 ), where (X 1 , X 3 ) ∈ EΩ (D, V 1 , V 3 ), such that

all the diagrams incurred in the quadruples are commutative, i.e.,
x3h ρ̌h = ρ̌h x1h , if {h , h } = j; x1h = σj x3h ρ̌h , if h = j; and σj x1h = x3h ρ̌h , if h = j.
Define a morphism of varieties
r13 : Z1 → Z
by r13 (X 1 , X 2 , X 3 , ρ1 , ρ2 ) = (X 1 , X 3 , ρ̌, σj ) where ρ̌k = ρ−12,k ρ1,k for any k ∈ I\{j}
and σj = ρ−1 ρ
1,j 2,j . Then, we have p π
13 2 = πr 13 , where π is the projection from Z
to EΩ (D, V 1 , V 3 ). Moreover, we observe that r13 is a quotient map of Z1 by GV 2 .
From these facts, we have
(35)
Eμ−αj +αi ,μ−αj Fμ−αj ,μ = P13! Π1! (Q̄l,Z1 )[m] = Π! R13! (Q̄l,Z1 )[m] = Π! (Q̄l,Z )[m].
On the other hand, let
ρ1 ρ2
Z2 = {(X 1 , X 4 , X 3 , ρ1 , ρ2 )|(X 1 , X 4 , X 3 ) ∈ EΩ (D, V 1 , V 4 , V 3 ), X 1 ← X 4 → X 3 }
and π2 be the projection from Z2 to EΩ (D, V 1 , V 4 , V 3 ). Define a morphism r̃13 :
Z2 → Z by r̃13 (X 1 , X 4 , X 3 , ρ1 , ρ2 ) = (X 1 , X 3 , ρ̌, σj ) where ρ̌k = ρ2,k ρ−1 1,k for any
k ∈ I\{j} and σj = ρ1,j ρ−1 2,j . Then we have p̃ π
13 2 = πr̃ 13 , where p̃ 13 is the projection
1 3
from Z2 to EΩ (D, V , V ), moreover r̃13 is quotient map of Z2 by GV 4 . From these
facts, we get
(36)
Fμ+αi −αj ,μ+αi Eμ+αi ,μ = P̃13! Π2! (Q̄l,Z2 )[m ] = Π! R̃13! (Q̄l,Z2 )[m ] = Π! (Q̄l,Z )[m ],
where m = fμ+αi −αj ,αj + eμ+αi ,αi . By (35), (36) and the fact that m = m , we
have the lemma.

Lemma 5.7. Let ν(i) = di + h∈H:h =i νh . For any vertex i ∈ I,
i −1
ν ν(i)−νi −1

Eμ,μ−αi Fμ−αi ,μ ⊕ Iμ [ν(i)−1−2p] = Fμ,μ+αi Eμ+αi ,μ ⊕ Iμ [ν(i)−1−2p].
p=0 p=0
Proof. Due to Lemma 5.2, we may assume that i is a source in Ω. Let us fix

four I-graded vector spaces, V a , for a = 1, 2, 3, 4, such that
(37) dim V 1 = dim V 3 = ν, dim V 2 = ν + i and dim V 4 = ν − i.
Then we have
Eμ,μ−αi Fμ−αi ,μ = P13! Π1! (Q̄l,Z1 )[m] and Fμ+αi −αj ,μ+αi Eμ+αi ,μ = P̃13! Π2! (Q̄l,Z2 )[m],
where m = ν(i) − 1 and the other notations on the right-hand sides are defined in
the proof of Lemma 5.6 with the condition (34) replaced by (37).
Let Zs1 be the open subvariety of Z1 defined by the condition that X 1 (i),
X (i) and X 3 (i) are injective. Similarly, we define the open subvariety EsΩ in
2
EΩ (D, V 1 , V 3 ).
Denote by X̌ the element obtained from X ∈ EΩ (D, V ) by deleting any com-
ponent xh such that h = i. Let Žs1 be the variety of tuples (X 1 , X̌ 2 , X 3 , ρ1 , ρ2 ).
Let Y1 be the variety of tuples (X̌ 1 , X̌ 2 , X̌ 3 , ρ̌1 , ρ̌2 , V1 , V2 , V3 ), where V1 , V2 , V3 ⊆
Di ⊕⊕h∈Ω:h =i Vh , such that V1 , V3 ⊆ V2 , dim V1 = dim V3 = νi and dim V2 = νi +1.
Similarly, we define the variety X1 of tuples (X̌ 1 , X̌ 2 , X̌ 3 , ρ̌1 , ρ̌2 , V1 , V3 ) and the va-
riety W of tuples (X̌ 1 , X̌ 3 , V1 , V3 ). Then we have the following cartesian diagram
r r
Zs1 −−−1−→ Žs1 −−−2−→ EsΩ
⏐ ⏐ ⏐
⏐
s1
⏐
s2
⏐
s3
r r
Y1 −−−3−→ X1 −−−4−→ W,
where the ra ’s are the obvious projections, and s2 and s3 are induced from s1 ,
which is defined by s1 (X 1 , X 2 , X 3 , ρ1 , ρ2 ) = (X̌ 1 , X̌ 2 , X̌ 3 , ρ̌1 , ρ̌2 , V1 , V2 , V3 ) with

V2 = im (qi2 + x2h ), and Va = im (qia + ρa,h xah ), ∀a = 1, 3.
Observe that s1 and s2 are the quotient maps of Zs1 and Žs1 by the group GVi1 ×
GVi2 × GVi3 , respectively, and s3 is the quotient map of EsΩ by GVi1 × GVi3 . Thus
we have
(38) R2! R1! (Q̄l,Zs1 ) = S3∗ R4! R3! (Q̄l,Y1 ).
Let Y1c be the closed subvariety of Y1 defined by the condition V1 = V3 and Y1o be
its complement. Let i1 : Y1c → Y1 and j1 : Y1o → Y1 be the inclusions. Sine r3 is
proper and Y1 is smooth, the complex R3! (Q̄l,Y1 ) is semisimple. So we have
r3! (Q̄l,Y1 ) = j1!∗
o o
r3! (Q̄l,Y1o ) ⊕ r3! i1! (Q̄l,Y1c ),
ro jo
where j1o and r3o are the morphisms Y1o → 3
Xo1 →
1
X1 with Xo1 the image of Y1o under
r3 . Observe that the morphism r3 i1 is a projective bundle of relative dimension
ν(i)−ν −1
ν(i) − νi − 1. Thus r3! i1! (Q̄l,Y1c ) = ⊕p=0 i Q̄l,Xc1 [−2p] where Xc1 is the closed
subvariety of X1 defined by the condition V1 = V3 . Then, we have
ν(i)−νi −1
R3! (Q̄l,Y1 ) = J!∗
o o
R3! (Q̄l,Y1o ) ⊕ ⊕p=0 Q̄l,Xc1 [−2p].
By combining the above analysis, we see that the restriction of the complex
Eμ,μ−αi Fμ−αi ,μ to EsΩ is equal to
(39)
ν(i)−ν −1
R2! R1! (Q̄l,Zs1 ) = S3∗ R4! R3! (Q̄l,Y1 ) = S3∗ R4! J1!∗
o o
R3! (Q̄l,Y1o ) ⊕ ⊕p=0 i S3∗ R4! Q̄l,Xc1 [m − 2p].
On the other hand, we may define the open subvarieties Zs2 of Zs2 similar to the
subvariety Zs1 of Z. Then the following varieties Ž2 , Y2 , Y2c , Y2o and X2 in the
diagram below are defined in a way similar to the varieties having subscript 1:
t t
Zs2 −−−1−→ Žs2 −−−2−→ EsΩ
⏐ ⏐ ⏐
⏐
w1
⏐
w2
⏐
s3
t t
Y2 −−−3−→ X2 −−−4−→ W,
where the morphism w1 is defined by
w1 (X 1 , X 4 , X 3 , σ1 , σ2 ) = (X̌ 1 , X̌ 4 , X̌ 3 , σ̌1 , σ̌2 , V1 , V4 , V3 )
−1
with V4 = im (qi4 + h∈Ω:h =i x4h ) and Va = im (qia + h∈Ω:h =i σa,h a
xh ) for any
a = 1, 3.
Let i2 : Y2c → Y2 ← Y2o : j2 be the inclusions. Then we see that t3 i2 is a
projective bundle of relative dimension νi − 1. An argument similar to the proof
46 YIQIANG LI
of (39) shows that the restriction of the complex Fμ+αi −αj ,μ+αi Eμ+αi ,μ to EsΩ is
equal to
(40)
i −1 ∗
T2! T1! (Q̄l,Zs2 ) = S3∗ T4! T3! (Q̄l,Y2 ) = S3∗ T4! J2!∗
o
T3!o (Q̄l,Y2o ) ⊕ ⊕νp=0 S3 T4! Q̄l,Xc2 [m − 2p],
to jo
where j2o and to3 are the morphisms Y2o →3
Xo2 →2
X2 with Xo2 . Finally, observe that
there are isomorphisms Y1 Y2 , X1 X2 and moreover the complex S3∗ T4! (Q̄l,Xc2 )
o o c c
is the restriction of Iμ to EsΩ . The lemma follows by comparing (39) and (40) and
using the observations.
Lemma 5.8. For any i = j ∈ I, let m = 1 − i · j. We have
(m−p) (p)
(m−p) (p)
Eμ3 ,μ2 Eμ2 ,μ1 Eμ1 ,μ = Eμ3 ,μ2 Eμ2 ,μ1 Eμ1 ,μ ;
0≤p≤m 0≤p≤m
p even p odd
(p) (m−p)
(p) (m−p)
Fμ,μ1 Fμ1 ,μ2 Fμ2 ,μ3 = Fμ,μ1 Fμ1 ,μ2 Fμ2 ,μ3 ;
0≤p≤m 0≤p≤m
p even p odd
where μ1 = μ + pαi , μ2 = μ + pαi + αj , and μ3 = μ + mαi + αj .

Proof. Without lost of generality, we assume that i is a source in Ω. For
a = 1, 2, 3, 4, let V a be the I-graded vector spaces such that
dim V 1 = ν − mi − j, dim V 2 = ν − pi − j, dim V 3 = ν − pi and dim V 4 = ν.
1 ρ1 2 ρ2 3 ρ3
Let Z be the variety of the data (X → X → X → X 4 ) where X a ∈ EΩ (D, V a )
for a = 1, 2, 3, 4. Let π : Z → EΩ (D, V 1 , V 4 ) be the obvious projection. Then
(m−p) (p)
Eμ3 ,μ2 Eμ2 ,μ1 Eμ1 ,μ = Π! (Q̄l,Z )[sm−p ],

where sm−p = m(ν(i)−νi )+(dj + h∈Ω:h =j νh −νj )+(m−p)(1−(m−p)). More-
over, π factors through ZΩ (D, V 1 , V 4 ), where the map r from Z to ZΩ (D, V 1 , V 4 )
ρ1 ρ2 ρ3 ρ3 ρ2 ρ1
is given by r(X 1 → X 2 → X 3 → X 4 ) = (X 1 → X 4 ). Let Bm−p = R! (Q̄l,Z ).
Thus,
(m−p) (p)
Eμ3 ,μ2 Eμ2 ,μ1 Eμ1 ,μ = Π! (Q̄l,Zp )[sm−p ] = Π12! Bm−p [sm−p ].
The identity for the E ’s is reduced to show that

(41) Bm−p [(m − p)(1 − (m − p))] = Bm−p [(m − p)(1 − (m − p))].
0≤p≤m 0≤p≤m
p even p odd
This is shown in [Zh08, 2.5.8]. For the sake of completeness, let us reproduce
here. Let Zs be the open subvariety of Z defined by the condition that X a (i)
are injective for a = 1, 2, 3, 4. The variety ZsΩ (D, V 1 , V 4 ) is defined similarly. Let
s : Zs → ZsΩ (D, V 1 , V 4 ) be the restriction of r to Zs and Cm−p := S! (Q̄l,Zs ). Then
the condition on the localization implies that we only need to show the identity
(41) when we restrict the complexes involved to the variety ZsΩ (D, V 1 , V 4 ), i.e., to
show that (41) holds with the complexes Bm−p replaced by the complexes Cm−p .
Observe that the group GV 2 × GV 3 × GL(Vi1 ) × GL(Vi4 ) acts freely on Zs
ρ
and the quotient variety Y is the variety of tuples (X̌ 1 → X̌ 4 , V1 , V2 , V3 ) where
V1 , V2 ⊆ V (i), V3 ⊆ V (i)) such that V1 ⊆ V2 and ρ(V2 ) ⊆ V3 ; and dim V1 = νi −m,
1 4
dim V2 = νi − p and V3 = νi . Moreover, the group GL(Vi1 ) × GL(Vi4 ) acts freely

on ZsΩ (D, V 1 , V 4 ) and its quotient variety X is the variety obtained from Y by
deleting V2 and replacing the condition ρ(V2 ) ⊆ V3 by ρ(V1 ) ⊆ V3 . Let t : Y → X

be the projection. Let Am−p = T! (Q̄l,Y ). To show (41), it reduces to show that the
identity holds with the complexes Bm−p replaced by the complexes Am−p .
Now define a partition (Xn )m n=0 of X such that elements in Xn satisfying the
condition that dim ρ(V 1 (i))∩V3 = νi −m+n. Let Yn = t−1 (Xn ), and tn : Yn → Xn
be the restriction of t to Yn . Then the restriction of r to Yn has fiber at any point
of Xn isomorphic to the Grassmannian Gr(m−p, n) of (m−p)-subspaces in n-space.
By the property of the cohomology of Gr(m − p, n), we have

m−p
Tn! (Q̄l,Yn ) = ⊕κ Q̄l,Xn [−2 (κa − a)],
a=1
where κ runs through the sequences (1 ≤ κ1 < κ2 < · · · < κm−p ≤ n). Since the
complexes Am−p are semisimple, it suffices to show that (41) holds when restricts
to the strata Xn for all n, which is left to show that

p
p
⊕κ Q̄l,Xn [−2 (κa −a)+p(1−p)] = ⊕κ Q̄l,Xn [−2 (κa −a)+p(1−p)].
0≤p≤n a=1 0≤p≤n a=1
p even p odd
To any sequence κ = (1 ≤ κ1 < · · · < κp ≤ n) of odd length, attached a sequence

κ of even length by κa = κa+1 for a = 1, · · · , p, if κ1 = 1; κ1 = 1 and κa+1 = κa
for a = 1, · · · , p if κ1 = 1. This defines a bijection between the set of sequences
κ of even length and the set of sequences κ of odd length and it is clear that the
shifts on both sides are the same under this bijection. Thus the identity holds. The
identity for the F ’s can be proved similarly.
Lemma 5.9. For any i ∈ I and m ∈ N, we have

(m)
(m+1)
Eμ+(m+1)αi ,μ+mαi Eμ+mαi ,μ = Eμ+(m+1)αi ,μ [m − 2p];
0≤p≤m
(m)
(m+1)
Fμ−(m+1)αi ,μ−mαi Fμ−mαi ,μ = Fμ−(m+1)αi ,μ [m − 2p].
0≤p≤m
Proof. Fix three I-graded vector spaces V a for a = 1, 2, 3 such that dim V 1 =
ν − (m + 1)i, dim V 2 = ν − mi and dim V 3 = ν. Let Z1 = ZΩ (D, V 1 , V 3 ) and Z be
the variety of tuples (X 1 , X 2 , X 3 , ρ1 , ρ2 ), where (X 1 , X 2 , X 3 ) ∈ EΩ (D, V 1 , V 2 , V 3 ),
ρ1 ρ2
such that X 1 → X 2 → X 3 . Let t : Z → Z1 be the map defined by
t(X 1 , X 2 , X 3 , ρ1 , ρ2 ) = (X 1 , X 3 , ρ2 ρ1 : X 1 → X 3 ).
As in the proof of Lemma 5.7, we have T! (Q̄l,Z ) = ⊕m
p=0 Q̄l,Z1 [−2p]. So
(m)
E μ+(m+1)αi ,μ+mαi Eμ+mαi ,μ = Π12! T! (Q̄l,Z )[eμ+(m+1)αi ,αi + eμ+mαi ,mαi ]
= ⊕m
p=0 Π12! (Q̄l,Z1 )[eμ+(m+1)αi ,αi + eμ+mαi ,mαi − 2p]
(m+1)
= ⊕m
p=0 Eμ+(m+1)αi ,μ [m − 2p].
The proof for the F ’s is similar.
By specializing the shift [z] to v z for any z ∈ Z, the identities in Lemmas 5.4-5.9
become the defining relations of the integral form AU̇. In short, we have
48 YIQIANG LI
(n) (n)
Proposition 5.10. The complexes Iμ , Eμ,μ−nαi and Fμ,μ+nαi satisfy the
defining relations of the integral form AU̇ defined in Section 2.2.
Remark 5.11. In many respects, if not all, the proof of Proposition 5.10 is very
similar to that of [Zh08, Theorem 2.5.2] (see also Proposition 7.7 in this paper).
6. Algebra Kd
6.1. Complex K• . Consider the complexes of the form
−
(42) K• = K1 · K2 · ... · Km ∈ DG (EΩ (D, V 1 , V 2 )),
(n) (n)
where the Ka ’s are either Eμ ,μ or Fμ ,μ in Section 5.1.
Proposition 6.2. The complexes K• in ( 42) are bounded.
Proof. For any pair (i, a) of sequences, where i = (im , · · · , i1 ) ∈ I m and
a = (am , · · · , a1 ) ∈ Nm , we write
(a ) (a ) (a ) (a ) (a ) (a )
E(i,a),μ = Eμ,μmm−1 · · · Eμ2 ,μ
2
1 Eμ1 ,μ0
1
and Fμ,(i,a) = Fμ0 ,μ
1
1 · · · Fμm−2 ,μm−1 Fμm−1 ,μ ,
m−1 m
such that μl − μl−1 = al αil for l = 1, · · · , m. By Lemmas 5.6 and 5.7, it suffices to
show the boundedness of the complex K• if K• is of the form Fμ,(j,b) E(i,a),μ for any
two pairs (i, a) and (j, b). An argument similar to the proof of Lemma 5.8 yields
that
(43) Fμ,(j,b) E(i,a),μ = Π! (Q̄l,Z )[m],
for some m, where π is the projection from Z to the variety EΩ (D, V 1 , V 3 ) with the
dimensions of V 1 and V 3 determined by the pairs of sequences, and Z is the variety
ρ1 ρ2
of the data (X 1 ← X 2 → X 3 ) together with a pair (U s , W t )1≤s≤m−1,1≤t≤n−1
of I-graded partial flags of V 1 and V 3 , where the dimensions of U s and W t are
determined by the μ’s in the pairs (i, a)) and (j, b), respectively, such that
ρ1 (X 2 ) ⊆ U n−1 ⊆ U n−2 ⊆ · · · ⊆ U 1 ⊆ V 1 , ρ1 (X 2 ) ⊆ W n−1 ⊆ · · · ⊆ W 1 ⊆ V 3 ,
and U s and W t are invariant under X 1 and X 3 , respectively, for any s and t. The
morphism π factors through the following varieties
π π π
Z →1 Z1 →2 Z2 →3 EΩ (D, V 1 , V 3 ),
where Z1 is the variety obtained from Z by forgetting the maps ρ2 , the variety Z2 is
the quotient variety of Z1 by the group GV 2 and the morphisms are clearly defined.
It is clear that the functors Π1! , Π2! and Π3! send bounded complexes to bounded
complexes and Π! = Π3! Π2! Π1! . The proposition follows.
Lemma 6.3. If Iμ is semisimple, then K• is semisimple.
Proof. Since Fμ,(j,b) E(i,a),μ = Fμ,(j,b) Iμ E(i,a),μ , we still have (43) with Z
replaced by
ρ1 ρ ρ2
Y = {(X 1 ← X 2 → X 4 → X 3 ; (U s , W t )s,t )},
where ρ is an isomorphism and (X 2 , X 4 ) ∈ EΩ (D, V 2 , V 4 ). Consider the following
cartesian diagram
Y −−−−→ ZΩ
⏐ ⏐
⏐
π1
⏐
π12
p
Y1 −−−−→ EΩ (D, V 2 , V 4 ),
ρ1 ρ2
where Y1 = {(X 1 ← X 2 , X 4 → X 3 ; (U s , W t )s,t )} and the horizontal maps are
ρ1 ρ2
projections. In particular, p(X 1 ← X 2 , X 4 → X 3 ; (U s , W t )s,t ) = (X 2 , X 4 ). By
base change theorem, we have

π1! (Q̄l,Y ) = p∗ (Iμ ).

Since p is smooth and, by assumption, Iμ is semisimple, we see that π1! (Q̄l,Y ) is
semisimple.
Let Y2 = Y1 /GV 2 ×GV 4 . Then π is the composition of the following morphisms
π π π
Y →1 Y1 →2 Y2 →3 EΩ (D, V 1 , V 3 ).
where π2 is a quotient map and π3 is a proper map. Again Π! is the composition
of Π1! , Π2! and Π3! . From this and that π1!

(Q̄l,Y ) is semisimple, we see that
Fμ,(j,b) E(i,a),μ is semisimple.
Conjecture 6.4. Iμ is a simple perverse sheaf, up to a shift, for any μ.

This conjecture is partially proved in Section 8. It is proved in full generality
in [W12] in a closely related setting.
6.5. Category Md . Let Md,ν 1 ,ν 2 be the full subcategory of DG

b
(EΩ (D, V 1 , V 2 ))
whose objects are finite direct sums of shifts of the complexes K• in Section 6.1.
We set
Md = ⊕ν 1 ,ν 2 ∈N[I] Md,ν 1 ,ν 2 .
It is clear that Md is closed under the convolution product “·” in (20). Let Ld be
the Grothendieck group of the category Md . By definition, this is free A-module
spanned by the isomorphism classes of objects in Md and subject to the relation
that [A ⊕ B] = [A] + [B] and [A[1]] = v[A] for any objects A, B ∈ Md . By
Lemma 6.9 and Proposition 4.11, the space Ld is an associative algebra, where the
multiplication on Ld is descended from the convolution product “·” in (20). By
Proposotion 5.10,
Theorem 6.6. There is a unique surjective A-algebra homomorphism
Ψd : AU̇ → Ld ,
(n) (n) (n) (n)
sending 1μ , Eμ+αi ,μ
and Fμ−αi ,μ , to Iμ , Eμ,μ−nαi and Fμ,μ+nαi , respectively, for
any i ∈ I and μ ∈ X.
If Ω is a second orientation of the graph Γ, we can define a similar category

MdΩ and its Grothendieck group LdΩ . We have an equivalence of categories:

Ω : Md → Md .
ΦΩ Ω
The following proposition follows from Lemma 4.14 and Lemma 5.2,

Proposition 6.7. We have an isomorphism of algebras Φd : Ld → LdΩ send-
(n) (n)
ing Iμ , Eμ,μ−nαi and Fμ,μ+nαi to the respective elements in LdΩ .
The algebra Ld should be a generalized q-Schur algebra ([D03]) when the

graph Γ is of finite type.
50 YIQIANG LI
6.8. Category Cd . Let Bd,ν 1 ,ν 2 be the set of isomorphism classes of sim-

ple perverse sheaves on DG b
(EΩ (D, V 1 , V 2 )) appeared as simple subquotients in
H (K• ) for any s ∈ Z, where K• is in (42) and pH i (−) is the perverse cohomology
p s
functor. We set
Bd = ν 1 ,ν 2 ∈N[I] Bd,ν 1 ,ν 2 .
Let Cd,ν 1 ,ν 2 be the full subcategory of DG
b
(EΩ (D, V 1 , V 2 )) consisting of semisimple
objects K such that the isomorphism classes of simple summands of pH s (K) are in
Bd,ν 1 ,ν 2 . We set
Cd = ⊕ν 1 ,ν 2 ∈N[I] Cd,ν 1 ,ν 2 .
By Proposition 6.3, we have
Lemma 6.9. Assume that Iμ is semisimple for any μ, then Md ⊆ Cd and,

moreover, the category Cd is closed under the convolution product “·” in ( 20).
Let Kd be the Grothendieck group of the category Cd . We see that Bd is a

basis of Kd . By Lemma 6.9,
Corollary 6.10. Assume that Iμ is semisimple for any μ, then Kd is an

associative algebra with multiplication induced from the convolution product “·” in
( 20). Moreover, the algebra Ld is a subalgebra of Kd .
Conjecture 6.11. We conjecture that Ld = Kd and the image of Ḃ under Ψd

is Bd with possible shifts, i.e., for any b ∈ Ḃ, ΨΩ
d (b) = v
s(b)
[K] for some s(b) ∈ Z
and [K] ∈ Bd .
Ω
We shall prove this conjecture partially in Section 8. In general, the identity

Ld = Kd is likely to be proved by the analysis of the geometry on the sink vertex
and in a way similar to the proof of Lemma 3.10 in [L03].
Let Vλ = Vλ1 ⊗ · · · ⊗ Vλn be the tensor product of the irreducible integrable
representations of U̇ with highest weights λ1 , · · · , λn in X+ . Denote by Dλ the
−
full subcategory of ⊕V DG (EΩ (D, V )) such that its Grothendieck group Vλ is iso-
morphic to the integral form of Vλ (see [Zh08]). Then the bifunctor (49) gives rise
to a bifunctor Cd × Dλ → Dλ by restriction, which descends to a bilinear map
◦ : Kd × Vλ → Vλ .
Let Bλ be the set of all isomorphism classes of simple perverse sheaves appearing
in Dλ . We then have

a◦b= sca,b c, where sca,b ∈ N[v, v −1 ],
c∈Bλ
for any a ∈ Bd and b ∈ Bλ . From this, we have
Corollary 6.12. If Conjectures 6.4 and 6.11 hold, then the action of the
canonical basis elements in U̇ on the canonical basis elements in Vλ has structure
constants in N[v, v −1 ] with respect to the canonical basis in Vλ .
It has been proved in [Zh08] that the generators of the quantum group attached
to Γ act positively on the canonical basis of the representation Vλ .
7. Relation with the work [Zh08]

−
7.1. Functor T . Let FΩ,G (D, V 1 , V 2 ) be the category of functors from the
− − ∗
category DG (EΩ (D, V )) to DG (EΩ (D, V 2 )). Similar to the functors Pst! and Pst
1
,
we define (∀s = 1, 2)
(44) Ps! = Q ◦ Qps! ◦ Q! , Ps∗ = Q ◦ Qp∗s ◦ Q! .
Define a functor
− −
ΘΩ : D G (EΩ (D, V 1 , V 2 )) → FΩ,G (D, V 1 , V 2 )
−
by ΘΩ (K) = P2! (K ⊗ P1∗ (−)) for any object K in DG (EΩ (D, V 1 , V 2 )) and the
∗
functors P2! and P1 are defined in (44).
Proposition 7.2.
ΘΩ (K · L) = ΘΩ (L)ΘΩ (K)
− −
for any objects K in DG (EΩ (D, V 1 , V 2 )) and L in DG (EΩ (D, V 2 , V 3 )).
Proof. By definition, we have
T (L)T (K)(M ) = P2! (L ⊗ (P1 )∗ T (K)(M )) = P2! (L ⊗ (P1 )∗ P2! (K ⊗ P1∗ (M )),
where P2! and (P1 )∗ are corresponding to the maps p2 and p1 in the following
cartesian diagram
p13 p
EΩ (D, V 1 , V 3 ) ←−−−− EΩ (D, V 1 , V 2 , V 3 ) −−−12−→ EΩ (D, V 1 , V 2 )
⏐ ⏐ ⏐
⏐
p̃2
⏐
p23
⏐
p2
p p
EΩ (D, V 3 ) ←−−2−− EΩ (D, V 2 , V 3 ) −−−1−→ EΩ (D, V 2 ).
By an argument similar to (22), we have (P1 )∗ P2! = P23! (P12
∗
) . So
T (L)T (K)(M ) = P2! (L ⊗ P23! (P12
∗
) (K ⊗ P1∗ (M )).
∗
By an argument similar to (23), we have P23! (A ⊗ P23 (B)) = P23! (A) ⊗ B. Thus,
T (L)T (K)(M ) = P2! P23! (P23
∗ ∗
(L) ⊗ (P12 ) (K ⊗ P1∗ (M )))
(45)
= P2! P23! (P23
∗ ∗
(L) ⊗ (P12 ∗ ∗
) (K) ⊗ (P12 ) P1 (M )).
Similarly, we have
∗ ∗ ∗
(46) T (K · L) = P̃2! P13! ((P12 ) (K) ⊗ P23 (L) ⊗ P13 P̃1∗ (M )),
where P̃1∗ comes from the projection EΩ (D, V 1 , V 3 ) → EΩ (D, V 1 ). The lemma
follows by comparing (45) with (46) and the following identity
P2! P23! = P̃2! P13! and ∗ ∗
(P12 ∗
) P1 = P13 P̃1∗ ,
which can be proved by a similar way as (24) and (26).
Define a functor of equivalence

− −
Ω : FG,Ω (D, V , V ) → FG,Ω (D, V , V )
ΨΩ 1 2 1 2

∗ Ω
by ΨΩΩ (F ) = ΦΩ F a ΦΩ , where a is the map of multiplication by −1 along the fiber
Ω
∗ Ω Ω
of the vector bundle EΩ over EΩ∩Ω . Its inverse is given by ΨΩ
Ω (−) = a ΦΩ (−)ΦΩ ,
Ω ∗
since ΦΩΩ ΦΩ = a . Moreover, we have
52 YIQIANG LI

Lemma 7.3. ΨΩ Ω commutes with the composition: ΨΩ (F2 ◦ F1 ) = ΨΩ (F2 ) ◦
Ω Ω
− −
Ω (F1 ) for any F1 ∈ FG,Ω (D, V , V ) and F2 ∈ FG,Ω (D, V , V ).
ΨΩ 1 2 2 3
− −
Let 1 DG (EΩ (D, V 1 , V 2 )) (resp. 1 DG (EΩ (D, V s )), s = 1, 2) be the full sub-
− −
category of the category DG (EΩ (D, V , V )) (resp. DG
1 2
(EΩ (D, V i ))) consisting of
∗ −
all objects such that a (K) K. Let FG,Ω (D, V , V 2 ) denote the category of
1 1
− −
functors from the category 1 DG (EΩ (D, V 1 )) to 1 DG (EΩ (D, V 2 )). We have
Lemma 7.4. The following diagram commutes
− Θ −
1
DG (EΩ (D, V 1 , V 2 )) −−−Ω−→ 1
FG,Ω (D, V 1 , V 2 )
⏐ ⏐
⏐ ⏐
Ω
ΦΩ Ω
ΨΩ
Θ
1
DG
b
(EΩ (D, V 1 , V 2 )) −−−Ω−→ 1 FG,Ω (D, V 1 , V 2 ).
− −
Proof. For any K ∈ 1
DG (EΩ (D, V 1 , V 2 )) and K1 ∈ 1
DG (EΩ (D, V 1 )), we
have

∗ Ω ∗ ∗ ∗
Ω T (K)(K1 ) = ΦΩ T (K)a ΦΩ (K1 ) = ΦΩ P2! (K ⊗ a P1 Π1! ((Π1 ) (K1 ) ⊗ L1 ))[d1 ],
ΨΩ Ω Ω
where P2! and P1∗ are from (44), Π1! and (Π1 )∗ come from the following projections
π π
EΩ (D, V 1 ) ←−−1−− EΩ∪Ω (D, V 1 ) −−−1−→ EΩ (D, V 1 );
d1 is the rank of π1 and L1 is defined in (7). Consider the following cartesian
diagram
π̃
EΩ∪Ω (D, V 1 ) × EΩ (D, V 2 ) −−−1−→ EΩ (D, V 1 , V 2 )
⏐ ⏐
⏐
p̃1
⏐
p1
π
EΩ∪Ω (D, V 1 ) −−−1−→ EΩ (D, V 1 ).
By an argument similar to (22), we have P1∗ Π1! = Π̃1! P̃1∗ . So

∗ ∗ ∗
Ω T (K)(K1 ) = ΦΩ P2! (K ⊗ a Π̃1! P̃1 ((Π1 ) (K1 ) ⊗ L1 ))[d1 ]
ΨΩ Ω

∗ ∗ ∗ ∗ ∗ ∗
Ω P2! Π̃1! (Π̃1 (K) ⊗ a P̃1 (Π1 ) (K1 ) ⊗ a P̃1 L1 ))[d1 ]
= ΦΩ
= R2! (R1∗ P2! Π̃1! (α) ⊗ L2 )[d2 ],
where R2! and R1∗ come from the following projections
r r
EΩ (D, V 2 ) ←−−
1
−− EΩ∪Ω (D, V 2 ) −−−2−→ EΩ (D, V 2 ),
d2 is the rank of r1 , α = Π̃∗1 (K) ⊗ a∗ P̃1∗ (π1 )∗ (K1 ) ⊗ a∗ P̃1∗ L1 )[d1 ], and L2 is defined
in (7). The following cartesian diagram
t
EΩ∪Ω (D, V 1 , V 2 ) −−−1−→ EΩ∪Ω (D, V 1 ) × EΩ (D, V 2 )
⏐ ⏐
⏐
s1
⏐
p2 π̃1
r
EΩ∪Ω (D, V 2 ) −−−1−→ EΩ (D, V 2 ),
gives rise to the identity R1∗ P2! Π̃1! = S1! T1∗ . So we have

∗ ∗ ∗
Ω T (K)(K1 ) = R2! (S1! T1 (α) ⊗ L2 )[d2 ] = R2! S1! (T1 (α) ⊗ S1 L2 )[d2 ]
ΨΩ
(47)
= R2! S1! (T1∗ Π̃∗1 (K) ⊗ T1∗ P̃1∗ (Π1 )∗ a∗ (K1 ) ⊗ a∗ T1∗ P̃1∗ L1 ⊗ S1∗ L2 )[d1 + d2 ].
On the other hand, we have

∗
Ω (K)(K1 ) = P2! (ΦΩ (K) ⊗ (P1 ) (K1 ))
ΘΩ ΦΩ Ω
(48) = P2! (M12!

∗
(M12 (K) ⊗ L12 )[r12 ] ⊗ (P1 )∗ (K1 ))
= P2! M12!
∗
(M12 ∗
(K) ⊗ L12 ⊗ (M12 ) (P1 )∗ (K1 ))[r12 ],
where P2! , (P1 )∗ come from the following projections
p p
EΩ (D, V 1 ) ←−−1−− EΩ (D, V 1 , V 2 ) −−−2−→ EΩ (D, V 2 ),
∗
and M12! , M12 , L12 and r12 are from 4.2. By comparing (47) with (48), the lemma
follows from the following observations: r2 s1 = p2 m12 , π̃1 t1 = m12 , π1 p̃1 t1 =
p1 m12 , p1 t1 = p1 , s1 = p2 and L12 = a∗ (P1 )∗ L1 ⊗ (P2 )∗ L2 . Note that the last
identity can be deduced from the following well-known fact. Let s, p1 , p2 : k ×k → k
be the addition, first and second projections, respectively. Then s∗ Lχ = p∗1 Lχ ⊗
p∗2 Lχ .
−
We define the following functors in FΩ,G (D, V 1 , V 2 ):
Iμ = Π2! Π∗1 , if dim V 1 = dim V 2 = ν;
Fμ,μ−nαi = Π2! Π∗1 [eμ,nαi ],
(n)
if dim V 1 = ν and dim V 2 = ν + ni;
Eμ,μ+nαi = Π1! Π∗2 [fμ,nαi ],
(n)
if dim V 1 = ν and dim V 2 = ν − ni;
where the functors Πi! and Π∗i are defined in (44) and eμ,nαi and fμ,nαi are defined
in (33). Note that Iμ = IdD − (EΩ (D,V 1 )) , the identity functor, since π1 and π2 are
G
principal GV 1 -bundles. We have
(n) (n) (n)
Proposition 7.5. ΘΩ (Iμ ) = Iμ , ΘΩ (Eμ,μ−nαi ) = Fμ,μ−nαi and ΘΩ (Fμ,μ+nαi )
(n)
= Eμ,μ+nαi .
(n) (n)
Proof. We shall show that ΘΩ (Eμ,μ−nαi ) = Fμ,μ−nαi . For any
−
K1 ∈ D G (EΩ (D, V 1 )),
we have
Fμ,μ−nαi (K1 ) = Π2! Π∗1 (K1 )[eμ,nαi ] = P2! Π12! Π∗12 P1∗ (K1 )[eμ,nαi ]
(n)
= P2! Π12! (Q̄l,ZΩ ⊗ Π∗12 P1∗ (K1 )[eμ,nαi ] = P2! (Π12! (Q̄l,ZΩ )[eμ,nαi ] ⊗ P1∗ (K1 ))
= P2! (Q(π12! (Q̄l,ZΩ )[eμ,nαi ]) ⊗ P1∗ (K1 )) = ΘΩ (Eμ,μ−nαi )(K1 ).
(n)
The rest can be proved similarly.

By Lemmas 5.2, 7.4, and Proposition 7.5, we have
(n) (n) (n)
Corollary 7.6. ΨΩ Ω Ω
Ω (Iμ ) = Iμ , ΨΩ (Fμ,μ−nαi ) = Fμ,μ−nαi and ΨΩ (Eμ,μ+nαi )
(n)
= Eμ,μ+nαi .
Actually, we need to show that the complexes in Corollary 7.6 are invariant
under the functor a∗ . This can be proved as in [L93, 10.2.4].
From Proposition 7.2, Corollary 7.6 and Proposition 5.10, we have
(n) (n)
Proposition 7.7. The functors Iμ , Eμ,μ−nαi and Fμ,μ+nαi satisfy the defining
relations of AU̇.
54 YIQIANG LI
(n) (n)
From Corollary 7.6, one sees that the functors Fμ,μ−nαi and Eμ,μ+nαi are the
(n) (n)
functors Fν,i and Eν,i in [Zh08], respectively. Proposition 7.7 was first proved
in [Zh08, 2.5.8].
Now that the functor ΘΩ induces a bifunctor
− − −
(49) ◦ : DG (EΩ (D, V 1 , V 2 )) × DG (EΩ (D, V 1 )) → DG (EΩ (D, V 2 ))
−
given by K ◦K1 = ΘΩ (K)(K1 ) = P2! (K ⊗P1∗ (K1 )) for any K ∈ DG (EΩ (D, V 1 , V 2 ))
−
and K1 ∈ DG (EΩ (D, V )).
1
Remark 7.8. (1). It should be true that the functor ΘΩ is fully faithful.
(2). We are not sure if the superscript 1 in the categories in Lemma 7.4 can be
dropped.
8. BLM case
In this section, we put the following extra assumptions on the graph Γ and the
I-graded vector space D in Section 4.1.
• The graph Γ is of type AN : 1 − 2 − · · · − N .
• The space D concentrates on the vertex N , i.e., Di = 0 for 1 ≤ i ≤ N − 1
and DN is a d-dimensional vector space over k.
8.1. Relation with [BLM90]. We fix an orientation Ω of Γ as follows: 1 →

2 → · · · → N . Let
(50) Ω2 : 1 → 2 → · · · → N → (N + 1) ← N ← · · · ← 2 ← 1 .
To a pair (V, V ) of I-graded vector spaces, the space EΩ (D, V, V ) defined in sec-
tion 4.1 is the representation variety of Ω2 with Va attached to the vertex a and Va
to the vertex a and D to the vertex N + 1.
Let us fix some notations. We will use xa+1,a to denote elements in
Hom(Va , Va+1 ). In particular, the element qN ∈ Hom(VN , DN ) is denoted by
xN +1,N in this section. For a pair (i, j) such that i ≤ j, 1 ≤ i ≤ N and
1 ≤ j ≤ N + 1, we fix a representative Si,j for the indecomposable representa-
tion of Ω2 of dimension equal to 1 at the vertices i, (i + 1), · · · , j, and equal to 0
otherwise. The notation Si ,j is defined similarly. For 1 ≤ i, j ≤ N + 1, let Ti,j
denote the indecomposable representation of the quiver Ω2 such that the dimension
of Ti,j equals 1 at the vertices i, i + 1, · · · , N, (N + 1), N , (N − 1) , · · · j , and zero
otherwise. When i = j, we simply write Si for Si,i , and Ti for Ti,i .
Let
U = {(X, X ) ∈ EΩ (D, V, V )|xa+1,a , xa+1,a , are injective, ∀a = 1, · · · , N }.
It is clear that U is nonempty only when dim V1 ≤ dim V2 ≤ · · · dim VN ≤ dim DN =
d and the same property for V . The set of isomorphism classes, [V ], of I-graded
vector spaces V of such a property is then in bijection with the set Sd of all non-
decreasing N + 1 step sequences, ν = (0 ≤ ν1 ≤ · · · ≤ νN ≤ d), of nonnegative
integers., via the map
[V ] → |V | = (dim V1 , dim V2 , · · · , dim VN ).
By abuse of notations, we write V ∈ Sd if dim Vi ≤ dim Vj for i ≤ j. To any ν ∈ Sd ,
we attach the partial flag variety Fν consisting of all flags F = (0 ≡ F0 ⊆ F1 ⊆
· · · ⊆ FN ⊆ FN +1 ≡ D) such that dim Fa = νa for a = 1, · · · , N .
For any pair V, V ∈ Sd such that |V | = ν and |V | = ν , define a morphism of

varieties
u : U → Fν × Fν ; (X, X ) → (F, F ),
where F = (0 ⊆ im(xN +1,N xN,N −1 · · · x2,1 ) ⊆ · · · ⊆ im(xN +1,N ) ⊆ DN ), and F is
defined similarly. It is well-known ([N94]) that the algebraic group G1 = GV ×GV
acts freely on U , and u can be identified with the quotient map q : U → G1 \U .
The following diagram of morphisms
β u
EΩ (D, V, V ) ←−−−− U −−−−→ Fν × Fν
induces a diagram of morphisms of algebraic stacks
β Qu
[G\EΩ (D, V, V )] ←−−−− [G\U ] −−−−→ [GD \Fν × Fν ],
which, in turn, gives rise to the following diagram of functors
β∗ Qu∗
DG
b
(EΩ (D, V, V )) −−−−→ DG
b
(U ) ←−−−− DG
b
D
(Fν × Fν ).
Lemma 8.2. Suppose that K is a G-equivariant complex on EΩ (D, V, V ). Then
K ∈ N in Section 4.4 if and only if K satisfies that supp(K) ∩ U = Ø.
Proof. To each a, we fix an orientation of the graph Γ:
Ωa : 1 → 2 → · · · → (a − 1) ← a → (a + 1) → · · · → N.
Let WaΩ(resp. WaΩ∪Ωa ) be the open subvariety of EΩ (resp. EΩ∪Ωa ) consisting
of all elements such that the component xa+1,a is injective. We have the following
diagram
τ τ
WaΩ ←−−−− WaΩ∪Ωa −−−−→ WaΩa
⏐ ⏐ ⏐
⏐
β
⏐
β
⏐
β
π π
EΩ ←−−−− EΩ∪Ωa −−−−→ EΩa ,
where the β’s are the open inclusions, and τ (resp. τ ) is the restriction of π
(resp. π ) to the variety WaΩ∪Ωa . Moreover, the squares in the above diagram are
cartesian. From this diagram, we have
β ∗ ΦΩ ∗ ∗ ∗
Ω (K) = τ! (τ β K ⊗ β L)[r].
a
This implies that
Ω (K)) ∩ Wa = Ø if and only if

supp(ΦΩ supp(K) ∩ WaΩ = Ø.
Ωa
(51) a
Assume that K ∈ Na , then by (51), supp(K) ∩ WaΩ = Ø. Since WaΩ ⊇ U , we have

supp(K) ∩ U = Ø. Since N is generated by the Na ’s, we see that any object K ∈ N
has the property that supp(K) ∩ U = Ø.
Now we shall show that if supp(K) ∩ U = Ø, then K ∈ N . Since any object
in DGb
(EΩ ) can be generated by the G-equivariant simple perverse sheaves, it is
enough to show this statement for K a simple perverse sheaf, which we shall assume
from now on.
Recall that EΩ (D, V, V ) can be regarded as a representation space of the
quiver Ω2 in (50). By Gabriel’s theorem, there is only finitely many G-orbits
in EΩ (D, V, V ) and, moreover, the stabilizers of the orbits in EΩ (D, V, V ) are
connected. So the G-equivariant simple perverse sheaves on EΩ (D, V, V ) are the
intersection cohomology complexes ICG (O(X,X ) ), attached to the G-orbit O(X,X )
56 YIQIANG LI
in EΩ (D, V, V ) containing the element (X, X ). Therefore, K ∈ N if the following

claim holds:
Claim. If X(j) + X (j ) is injective for 1 ≤ j ≤ i − 1, and X(i) + X (i ) is not
injective for some i ∈ [1, N ], then ICG (O(X,X ) ) ∈ Ni .
The claim can be shown as follows. Let i = (i1 , · · · , in ) be a sequence of
vertices in Ω2 and a = (a1 , · · · , an ) be a sequence of positive integers such that
Ω
il =i al = dim Vi for any vertex i. Let Fi,a be the variety of all triples (X, X , F),

where (X, X ) ∈ EΩ (D, V, V ) and F = (F = D ⊕ V ⊕ V ⊃ F ⊃ · · · ⊃ F = 0)
0 1 n
is a flag of graded vector subspaces in D ⊕ V ⊕ V , such that

dim Fl−1 /Fl = al il and (X, X )(Fl ) ⊂ Fl , ∀l = 1, · · · , n.
Consider the projection to the (1, 2)-components:
(52) Ω
πi,a : Fi,a
Ω
→ EΩ (D, V, V ), (X, X , F) → (X, X ).
By [R03, Theorem 2.2], one can choose a particular pair (i, a) such that the image of
Ω −1
Ω
πi,a is the closure O(X,X ) of the orbit O(X,X ) , and the restriction (πi,a ) (O(X,X ) )
→ O(X,X ) is an isomorphism. Thus, the complex ICG (O(X,X ) ) is a direct sum-
mand of the semisimple complex (πi,a Ω
)! (Q̄l,FΩ ), up to a shift.
i,a
So the complex ΦΩ Ω (ICG (O(X,X ) )) is a direct summand of the complex

i
ΦΩΩ ((πi,a )! (Q̄l,F

Ω )), up to a shift. By [L93, 10.2], the complex ΦΩ ((πi,a )! (Q̄l,F
Ω Ωi Ω
Ω ))
i
i,a i,a
is isomorphic to the complex (π )! (Q̄ Ωi ), up to a shift, where π and F are

Ωi Ωi Ωi
i,a l,Fi,a i,a i,a
defined in exactly the same manner and Fi,a

Ω
Ω
as πi,a with EΩ (D, V, V ) replaced by
Ωi
EΩi (D, V, V ). So the support of the complex ΦΩ (ICG (O(X,X ) )) is contained in
Ωi
the image of the morphism πi,a .
Observe that the conditions in the claim imply that
• either S1,i or S1 ,i is a direct summand of the representation M of Ω2
corresponding to the element (X, X ) in the claim;
• the representation M does not contain any direct summand of the form
S1,t and S1 ,t for 1 ≤ t < i.
From this observation, we see that the chosen sequence i = (i1 , · · · , in ) satisfies
that in = i or i , from the construction in [R03] and the Auslander-Reiten quiver
of Ω2 . This implies that for any (X, X , F) ∈ Fi,a
Ωi
, either ker(X(i)) = 0 if in = i or

ker(X (i )) = 0 if in = i . Therefore, the image of πi,aΩi
is contained in the subvariety
of EΩi (D, V, V ) consisting of all elements (X, X ) such that ker(X(i)) = 0 or
ker(X (i )) = 0. The claim follows.
By Lemma 8.2, there is an equivalence of triangulated categories

Q̄ : DG
b
(U ) → DG
b
(EΩ (D, V, V )) ≡ DG
b
(EΩ (D, V, V ))/N
such that Q = Q̄β ∗ and β! = Q! Q̄. For 1 ≤ i ≤ N , n ∈ N, we set
Δν = {(F, F ) ∈ Fν × Fν |F = F },
Yν,−nαi = {(F, F ) ∈ Fν × Fν |Fj ⊆ Fj , dim Fj /Fj = nδij , ∀1 ≤ j ≤ N },
Yν,+nαi = {(F, F ) ∈ Fν × Fν |Fj ⊇ Fj , dim Fj /Fj = nδij , ∀1 ≤ j ≤ N },
if ν exists. Let
(n) (n)
(53) 1ν = Q̄l,Δν , Eν,−nαi = Q̄l,Yν,−nαi [eν,−nαi ], Fν,+nαi = Q̄l,Yν,+nαi [fν,+nαi ],
where eν,−nαi = n(νi+1 − (νi + n)), fν,+nαi = n((νi − n) − νi−1 ), νN +1 = d and
ν0 = 0. By combining the above analysis, we have the following proposition.
Proposition 8.3. We have a sequence of functors of equivalence
Qu∗ Q̄
(54) DG
b
D
(Fν × Fν ) → DG
b
(U ) → DG
b
(EΩ (D, V, V )).
Moreover, for any μ = λ − ν,
(55)
Iμ = Q̄Qu∗ (1ν ), Eμ,μ−nαi = Q̄Qu∗ Eν,−nαi , Fμ,μ+nαi = Q̄Qu∗ Fν,+nαi .
(n) (n) (n) (n)
and
where Iμ , E• and F• are from ( 32) and 1ν , E• and F• ’s are from ( 53).
Define a convolution product “·” on the categories DG
b
(Ui ) as follows. We have
the following commutative diagram
uij
U −−−−→ U
⏐ ⏐
⏐
β
⏐
β
pij
EΩ (D, V, V , V ) −−−−→ EΩ (D, V i , V j ),
where pij is the projection to the (i, j)-component and uij is the restriction of pij
to U . The morphism uij then defines a morphism Quij : [H\U ] → [G\U ]. To any
objects K and L in DG b
(U ), associated an object
K · L = Qu13! (Qu∗12 (K) ⊗ Qu∗23 (L)) ∈ DG
b
(U ).
Similarly, we define a convolution product
(56) ◦ : DG
b
D
(Fν × Fν ) × DG
b
D
(Fν × Fν ) → DG
b
D
(Fν × Fν )
The following proposition shows that the construction given in [BLM90] is
compatible with the one given in this paper.
Proposition 8.4. The convolution products on DG b
D
(F × F), DG b
(U ) and
DG
b
(EΩ (D, V, V )) are compatible with the functors Qu∗ and Q̄ in the diagram ( 54).
Proof. First, we show that the convolution products are compatible with the
functors Q̄. Recall that Q = Q̄ ◦ β ∗ and β! = Q! Q̄. We have
(57) Q̄ ◦ Qu∗ij = Q̄ ◦ Qu∗ij β ∗ β! = Q̄β ∗ ◦ Qp∗ij β! = Q ◦ Qp∗ij ◦ Q! Q̄ = Pij∗ Q̄.
Similarly,
(58) Q̄ ◦ Quij! = Q̄β ∗ β! Quij! = Q ◦ Qpij! β! = Q ◦ Qpij! Q! Q̄ = Pij! Q̄.
From (57) and (58), we see that the convolution products on DG b
(Ui ) and
DGb
(EΩ (D, V, V )) are compatible with the functor Q̄.
Now, we show that the convolution products are compatible with the functors
Qu∗ . We have the following cartesian diagram:
Quij
[H\U ] −−−−→ [G\U ]
⏐ ⏐
⏐
Qu
⏐
Qu
qij
[GD \(Fν × Fν × Fν )] −−−−→ [GD \(Fν i × Fν j )].
58 YIQIANG LI
where qij is a projection and the first U is contained in EΩ (D, V, V , V ) such that
|V | = ν, |V | = ν and |V | = ν . This cartesian diagram gives rise to the following
identities:
Qu∗ qij
∗
= Qu∗ij Qu∗ , and Qu∗ qij! = Quij! Qu∗ .
So for any K ∈ DG
b
D
(Fν × Fν ) and L ∈ DG
b
D
(Fν × Fν ), we have
Qu∗ (K ◦ L) = Qu∗ q13! (q12
∗ ∗
(K) ⊗ q13 (L)) = Qu13! Qu∗ (q12
∗ ∗
(K) ⊗ q13 (L))
= Qu13! (Qu∗12 Qu∗ (K) ⊗ Qu∗23 Qu∗ (L)) = Qu∗ (K) · Qu∗ (L).
Therefore, the convolution products commute with the functor Qu∗ . The lemma
follows.
The following theorem follows from (55) in Proposition 8.3, Proposition 8.4,
and the results in [BLM90] and [SV00].
Theorem 8.5. Under the assumption in this section, the conjectures 6.4 and
6.11 hold.
8.6. Singular support. Let Ω be the quiver obtained from Ω by reversing
all arrows in Ω. Let EΩ (D, V ) = ⊕h∈Ω Hom(Vh , Vh ) ⊕ ⊕i∈I Hom(Di , Vi ) and
EΩ (D, V, V ) = EΩ (D, V ) × EΩ (D, V ). We shall identify the space
E := EΩ (D, V, V ) ⊕ EΩ (D, V, V )
with the cotangent bundles of EΩ (D, V, V ). (See [L91, 12] for more details.) Let
Λ be the closed subvariety of E defined by the following (ADHM or GP) relations:
xa,a+1 xa+1,a = xa,a−1 xa−1,a , xa ,(a+1) x(a+1) ,a = xa ,(a−1) x(a−1) ,a , ∀1 ≤ a ≤ N ;
xN +1,N xN,N +1 = xN +1,N xN ,N +1 .

This is Lusztig’s nilpotent quiver variety in [L91, 12]. Let Nil be the variety
of nilpotent elements in End(D). Let GD acts on Nil by conjugation. Define a
morphism
π : Λ → Nil
of varieties by sending elements, say X, in Λ to xN +1,N xN,N +1 . Note that the
morphism π is G1 -equivariant.
Let Λs be the open subvariety of Λ consisting of all elements such that xa+1,a +
xa−1,a and x(a+1) ,a + x(a−1) ,a are injective for 1 ≤ a ≤ N . By [N94], the G1 -
action on Λs is free and, moreover, admits a GIT quotient G1 \Λs , isomorphic to
the generalized Steinberg variety
Z := {(x, F, F ) ∈ Nil × Fν × Fν |x(Fa ) ⊆ Fa−1 , x(Fa ) ⊆ Fa−1

, 1 ≤ a ≤ N + 1}.
We shall identify G1 \Λs with Z. Due to the fact that π is G1 -equivariant, it factors
q π
through Z, i.e., π is the composition of the morphisms Λs → Z → Nil.
Given a simple perverse sheaf K in DG b
(X), the singular support, SS(K), of K
is defined to be the singular support of the complex KX in [KS90]. Here we abuse
the notion slightly, we mean the singular support of its counterpart on X(C) via
the principals in [BBD82, Ch. 6].
Proposition 8.7. K ∈ N if and only if SS(K) ∩ Λs = Ø.
Proof. Suppose that SS(K)∩Λs = Ø. From [KS90], SS(K)∩EΩ (D, V, V )

supp(K) where EΩ (D, V, V ) is identified with the subspace 0 ⊕ EΩ (D, V, V ) of E.
So we have
Ø = SS(K) ∩ Λs ⊇ SS(K) ∩ Λs ∩ EΩ (D, V, V ) = Supp(K) ∩ U
where U is the open subvariety of EΩ (D, V, V ) defined in the section 8.1. By
Lemma 8.2, we see that K ∈ N .
On the other hand, if SS(K) ∩ Λs = Ø, then it is a non empty closed subvariety
of Λs . Hence, π(SS(K) ∩ Λs ) is a closed subvariety of Nil due to the fact that
π can be decomposed as π q with π a proper map and q a quotient map. Note
that π(SS(K) ∩ Λs ) is also GD -invariant since K is a G-equivariant complex. So
0 ∈ π(SS(K) ∩ Λs ), which implies that SS(K) ∩ Λs ∩ π −1 (0) = Ø. Observe that
Λs ∩π −1 (0) = U . Thus, Supp(K)∩U = Ø. By Lemma 8.2, K ∈ N . The proposition
follows.
References
[B05] P. Balmer, The spectrum of prime ideals in tensor triangulated categories, J. Reine
Angew. Math. 588 (2005), 149–168, DOI 10.1515/crll.2005.2005.588.149. MR2196732
[BBD82] A. A. Beı̆linson, J. Bernstein, and P. Deligne, Faisceaux pervers (French), Analysis and
topology on singular spaces, I (Luminy, 1981), Astérisque, vol. 100, Soc. Math. France,
Paris, 1982, pp. 5–171. MR751966
[BL94] J. Bernstein and V. Lunts, Equivariant sheaves and functors, Lecture Notes in Mathe-
matics, vol. 1578, Springer-Verlag, Berlin, 1994. MR1299527
[BLM90] A. A. Beilinson, G. Lusztig, and R. MacPherson, A geometric setting for the quantum
deformation of GLn , Duke Math. J. 61 (1990), no. 2, 655–677, DOI 10.1215/S0012-
7094-90-06124-1. MR1074310
[CG97] N. Chriss and V. Ginzburg, Representation theory and complex geometry, Birkhäuser
[D03] S. Doty, Presenting generalized q-Schur algebras, Represent. Theory 7 (2003), 196–213
(electronic), DOI 10.1090/S1088-4165-03-00176-6. MR1990659
[DJ89] R. Dipper and G. James, The q-Schur algebra, Proc. London Math. Soc. (3) 59 (1989),
no. 1, 23–50, DOI 10.1112/plms/s3-59.1.23. MR997250
[FK88] E. Freitag and R. Kiehl, Étale cohomology and the Weil conjecture, Ergebnisse der
Mathematik und ihrer Grenzgebiete (3) [Results in Mathematics and Related Areas (3)],
vol. 13, Springer-Verlag, Berlin, 1988. Translated from the German by Betty S. Water-
house and William C. Waterhouse; With an historical introduction by J. A. Dieudonné.
MR926276
[GMV05] S. Gelfand, R. MacPherson, K. Vilonen, Microlocal perverse sheaves,
arXiv:math/0509440.
[GV93] V. Ginzburg and É. Vasserot, Langlands reciprocity for affine quantum groups of type
An , Internat. Math. Res. Notices 3 (1993), 67–85, DOI 10.1155/S1073792893000078.
MR1208827
[G01] T. L. Gómez, Algebraic stacks, Proc. Indian Acad. Sci. Math. Sci. 111 (2001), no. 1,
1–31, DOI 10.1007/BF02829538. MR1818418
[G94] I. Grojnowski, Affinizing quantum algebras: from D-modules to K-theory, preprint
(1994).
[J98] R. Joshua, Modules over convolution algebras from equivariant derived categories. I, J.
Algebra 203 (1998), no. 2, 385–446, DOI 10.1006/jabr.1997.7346. MR1622787
[KL08a] M. Khovanov and A. D. Lauda, A diagrammatic approach to categorification of quantum
groups. I, Represent. Theory 13 (2009), 309–347, DOI 10.1090/S1088-4165-09-00346-X.
MR2525917
[KL08b] M. Khovanov and A. D. Lauda, A diagrammatic approach to categorification of quantum
groups II, Trans. Amer. Math. Soc. 363 (2011), no. 5, 2685–2700, DOI 10.1090/S0002-
9947-2010-05210-9. MR2763732
60 YIQIANG LI
[KL08c] M. Khovanov and A. D. Lauda, A categorification of quantum sl(n), Quantum Topol.

1 (2010), no. 1, 1–92, DOI 10.4171/QT/1. MR2628852
[K91] M. Kashiwara, On crystal bases of the Q-analogue of universal enveloping alge-
bras, Duke Math. J. 63 (1991), no. 2, 465–516, DOI 10.1215/S0012-7094-91-06321-0.
MR1115118
[KS97] M. Kashiwara and Y. Saito, Geometric construction of crystal bases, Duke Math. J. 89
(1997), no. 1, 9–36, DOI 10.1215/S0012-7094-97-08902-X. MR1458969
[KS90] M. Kashiwara and P. Schapira, Sheaves on manifolds, Grundlehren der Mathematischen
Wissenschaften [Fundamental Principles of Mathematical Sciences], vol. 292, Springer-
Verlag, Berlin, 1994. With a chapter in French by Christian Houzel; Corrected reprint
of the 1990 original. MR1299726
[KS06] M. Kashiwara and P. Schapira, Categories and sheaves, Grundlehren der Mathema-
tischen Wissenschaften [Fundamental Principles of Mathematical Sciences], vol. 332,
Springer-Verlag, Berlin, 2006. MR2182076
[KW01] R. Kiehl and R. Weissauer, Weil conjectures, perverse sheaves and l’adic Fourier trans-
form, Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern
Surveys in Mathematics [Results in Mathematics and Related Areas. 3rd Series. A Series
of Modern Surveys in Mathematics], vol. 42, Springer-Verlag, Berlin, 2001. MR1855066
[LO08a] Y. Laszlo and M. Olsson, The six operations for sheaves on Artin stacks. I. Finite coeffi-
cients, Publ. Math. Inst. Hautes Études Sci. 107 (2008), 109–168, DOI 10.1007/s10240-
008-0011-6. MR2434692
[LO08b] Y. Laszlo and M. Olsson, The six operations for sheaves on Artin stacks. II. Adic coeffi-
cients, Publ. Math. Inst. Hautes Études Sci. 107 (2008), 169–210, DOI 10.1007/s10240-
008-0012-5. MR2434693
[LO09] Y. Laszlo and M. Olsson, Perverse t-structure on Artin stacks, Math. Z. 261 (2009),
no. 4, 737–748, DOI 10.1007/s00209-008-0348-z. MR2480756
[L87] G. Laumon, Transformation de Fourier, constantes d’équations fonctionnelles et con-
jecture de Weil (French), Inst. Hautes Études Sci. Publ. Math. 65 (1987), 131–210.
MR908218
[LMB00] G. Laumon and L. Moret-Bailly, Champs algébriques (French), Ergebnisse der Math-
ematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics
[Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in
Mathematics], vol. 39, Springer-Verlag, Berlin, 2000. MR1771927
[Li10a] Y. Li, A geometric realization of quantum groups of type D, Adv. Math. 224 (2010),
no. 3, 1071–1096, DOI 10.1016/j.aim.2009.12.022. MR2628804
[Li10b] Y. Li, On geometric realizations of quantum modified algebras and their canonical bases,
II, arXiv:1009.0838.
[L90] G. Lusztig, Canonical bases arising from quantized enveloping algebras, J. Amer. Math.
Soc. 3 (1990), no. 2, 447–498, DOI 10.2307/1990961. MR1035415
[L91] G. Lusztig, Quivers, perverse sheaves, and quantized enveloping algebras, J. Amer.
Math. Soc. 4 (1991), no. 2, 365–421, DOI 10.2307/2939279. MR1088333
[L93] G. Lusztig, Introduction to quantum groups, Progress in Mathematics, vol. 110,
Birkhäuser Boston, Inc., Boston, MA, 1993. MR1227098
[L99] G. Lusztig, Aperiodicity in quantum affine gln , Asian J. Math. 3 (1999), no. 1, 147–
177, DOI 10.4310/AJM.1999.v3.n1.a7. Sir Michael Atiyah: a great mathematician of
the twentieth century. MR1701926
[L03] G. Lusztig, Constructible functions on varieties attached to quivers, Studies in memory
of Issai Schur (Chevaleret/Rehovot, 2000), Progr. Math., vol. 210, Birkhäuser Boston,
Boston, MA, 2003, pp. 177–223. MR1985727
[MSV10] M. Mackaay, M. Stošić, P. Vaz, A diagrammatic categorification of the q-Schur algebra,
arXiv:1008.1348.
[MS08] V. Mazorchuk and C. Stroppel, Categorification of (induced) cell modules and the rough
structure of generalised Verma modules, Adv. Math. 219 (2008), no. 4, 1363–1426, DOI
10.1016/j.aim.2008.06.019. MR2450613
[M71] S. Mac Lane, Categories for the working mathematician, 2nd ed., Graduate Texts in
Mathematics, vol. 5, Springer-Verlag, New York, 1998. MR1712872
[N94] H. Nakajima, Instantons on ALE spaces, quiver varieties, and Kac-Moody alge-
bras, Duke Math. J. 76 (1994), no. 2, 365–416, DOI 10.1215/S0012-7094-94-07613-8.
MR1302318
[N98] H. Nakajima, Quiver varieties and Kac-Moody algebras, Duke Math. J. 91 (1998), no. 3,
515–560, DOI 10.1215/S0012-7094-98-09120-7. MR1604167
[N00] H. Nakajima, Quiver varieties and finite-dimensional representations of quantum affine
algebras, J. Amer. Math. Soc. 14 (2001), no. 1, 145–238, DOI 10.1090/S0894-0347-00-
00353-2. MR1808477
[R03] M. Reineke, Quivers, desingularizations and canonical bases, Studies in memory of Issai
Schur (Chevaleret/Rehovot, 2000), Progr. Math., vol. 210, Birkhäuser Boston, Boston,
MA, 2003, pp. 325–344. MR1985731
[Re08] M. Reineke, Framed quiver moduli, cohomology, and quantum groups, J. Algebra 320
(2008), no. 1, 94–115, DOI 10.1016/j.jalgebra.2008.01.025. MR2417980
[R08] R. Rouquier, 2-Kac-Moody algebras, arXiv:0812.5023.
[SV00] O. Schiffmann and E. Vasserot, Geometric construction of the global base of the
quantum modified algebra of gl , Transform. Groups 5 (2000), no. 4, 351–360, DOI
n
10.1007/BF01234797. MR1800532
[S08] O. M. Schnürer, Equivariant sheaves on flag varieties, Math. Z. 267 (2011), no. 1-2,
27–80, DOI 10.1007/s00209-009-0609-5. MR2772241
[T87] T. Tanisaki, Hodge modules, equivariant K-theory and Hecke algebras, Publ. Res. Inst.
Math. Sci. 23 (1987), no. 5, 841–879, DOI 10.2977/prims/1195176035. MR934674
[V76] J.L. Verdier, Categories derivees, etat 0. In SGA 4.5, LNM 569.
[V96] J.-L. Verdier, Des catégories dérivées des catégories abéliennes (French, with French
summary), Astérisque 239 (1996), xii+253 pp. (1997). With a preface by Luc Illusie;
Edited and with a note by Georges Maltsiniotis. MR1453167
[W04] I. Waschkies, The stack of microlocal perverse sheaves (English, with English and French
summaries), Bull. Soc. Math. France 132 (2004), no. 3, 397–462. MR2081221
[W12] B. Webster, A categorical action of quantized quiver varieties, arXiv:1208.5957.
[WW09] B. Webster, G. Williamson, The bounded below equivariant derived category, preprint.
[Z85] A. V. Zelevinskiı̆, Two remarks on graded nilpotent classes (Russian), Uspekhi Mat.
Nauk 40 (1985), no. 1(241), 199–200. MR783619
[Zh08] H. Zheng, Categorification of integrable representations of quantum groups, Acta
Math. Sin. (Engl. Ser.) 30 (2014), no. 6, 899–932, DOI 10.1007/s10114-014-3631-4.
MR3200442
Department of Mathematics, University at Buffalo, State University of New York,

244 Mathematics Building, Buffalo, New York 14260
E-mail address: yiqiang@buffalo.edu
Volume 684, 2017
The cube and the Burnside category
Tyler Lawson, Robert Lipshitz, and Sucharit Sarkar

Abstract. In this note we present a combinatorial link invariant that un-
derlies some recent stable homotopy refinements of Khovanov homology of
links. The invariant takes the form of a functor between two combinatorial
2-categories, modulo a notion of stable equivalence. We also develop some
general properties of such functors.
1. Introduction
In the last five years, two stable homotopy refinements of the Khovanov ho-
mology of links were introduced [LS14a, HKK16]. In a recent paper, we showed
that these two refinements agree, and in the process gave a simplified construction
of these invariants [LLS]. That paper still involved a certain amount of topology.
In this note, we present a combinatorial link invariant from which one can extract
the Khovanov homotopy types, with the goal of making aspects of our earlier work
more broadly accessible.
We will study functors from the cube category 2n , the small category associ-
ated to the partially ordered set {0, 1}n (Section 2), to the Burnside category B,
which is the 2-category whose objects are finite sets and morphisms are finite corre-
spondences (Section 3). The original construction of Khovanov homology [Kho00]
using Kauffman’s n-dimensional cube of resolutions [Kau87] of an n-crossing link
diagram K can be generalized to construct a 2-functor
FKh (K) : 2n → B.
See Section 6 for the exact definition, and for some additional grading shifts that
are involved. (Such functors also arise in other contexts, such as from a simplicial
complex with n vertices; see Example 4.2.)
To any such functor F : 2n → B, one can associate a chain complex Tot(F ) ∈
Kom, the totalization of F . Indeed, this construction can be thought of a functor
n
Tot : B 2 → Kom.
The totalization of the functor FKh recovers the dual of the Khovanov complex of
K:
(CKh (K))∗ = Tot(FKh ).
2010 Mathematics Subject Classification. 57M25, 55P42.

TL was supported by NSF Grant DMS-1206008.
RL was supported by NSF Grant DMS-1149800.
SS was supported by NSF Grant DMS-1350037.
2017
63
64 TYLER LAWSON, ROBERT LIPSHITZ, AND SUCHARIT SARKAR
(The dual only appears to maintain consistency with earlier conventions [LS14a].)
Given a functor F : 2n → B and a sufficiently large ∈ Z, one can construct a
based CW complex F ∈ CW. The dependence on is simple: F +1 is merely
the reduced suspension ΣF . The reduced cellular chain complex of F is the
earlier chain complex up to a shift:
•cell (F ) = Σ Tot(F ).
C
Similarly, to any natural transformation η : F → F between two such 2-functors
F, F : 2n → B, one can associate a pointed cellular map F → F that induces
the map Σ Tot(η) on the reduced cellular chain complexes. Unfortunately, the
definitions of these spaces and these maps depend on certain choices, and therefore
the construction is not strictly functorial. We may eliminate these choices if we are
willing to work with spectra. Indeed, if S is any reasonable category of spectra,
then there is a canonical functor
n
| · | : B2 → S
so that for any F : 2n → B, |F | is isomorphic to Σ− (Σ∞ F ), the th desuspension
of the suspension spectrum of F . Up to a grading shift, the Khovanov homotopy
type associated to the n-crossing link diagram K is
XKh (K) = |FKh |,
and therefore, its cohomology H • (XKh (K)) is isomorphic to the Khovanov homol-
ogy Kh(K) = H• (CKh (K)).
In this note, we only sketch the construction of F and only hint at the
construction of |F | (Section 7). We will focus on functors from cubes to the Burnside
category and define an equivalence relation on such functors, generated by the
following two relations:
(1) If η : F → F is a natural transformation between two functors F, F : 2n → B
and the induced map Tot(η) : Tot(F ) → Tot(F ) is a chain homotopy equiva-
lence, then F is stably equivalent to F .
(2) If ι : 2n → 2N is a face inclusion, and F : 2n → B, then there is an induced
functor Fι : 2N → B, induced by Fι ◦ ι = F and for any v ∈ 2N \ ι(2n ),
Fι (v) = ∅. The functor Fι is stably equivalent to F (up to a grading shift).
See Definition 5.9 for the precise version. This notion of stable equivalence ensures
that if F, F are stably equivalent functors, then |F | and |F | are homotopy equiv-
alent spectra; or equivalently, F and F are stably homotopy equivalent CW
complexes.
The main result of this paper is that the Khovanov functor FKh is a link
invariant. Namely, if K and K are isotopic link diagrams, then the Khovanov
functors FKh (K) and FKh (K ) are stably equivalent (Theorem 1, Section 6).
2. The cube
The one-dimensional cube is 2 = {0, 1}. It can be viewed as a partially ordered
set by declaring 1 > 0. It can also be viewed as a category with a single non-identity
morphism from 1 to 0. There is a notion of grading, where we declare the grading
of v ∈ 2 to be |v| = v.
THE CUBE AND THE BURNSIDE CATEGORY 65
The n-dimensional cube is the Cartesian product 2n = {0, 1}n . It has an

induced partial order, where
(u1 , . . . , un ) ≥ (v1 , . . . , vn ) if and only if ∀i(ui ≥ vi ),
and an induced categorical structure: the morphism set Hom2n (u, v) has a single
element if u ≥ v, and is empty otherwise. For u ≥ v, we will write ϕu,v to denote
the unique morphism in Hom2n (u, v). Finally, there is an induced grading, which
is simply the L1 -norm:
|v| = vi .
i
For convenience, we will write u ≥k v if u ≥ v and |u| − |v| = k; and we will
sometimes write u • v if u ≥1 v.
We will need the following sign assignment function.
Definition 2.1. Given u = (u1 , . . . , un ) ≥1 v = (v1 , . . . , vn ), let k be the
unique element in {1, . . . , n} satisfying uk > vk , and define

k−1
su,v = ui (mod 2).
i=1
3. The Burnside category

For us, the Burnside category B is the 2-category of finite sets, finite cor-
respondences, and bijections of correspondences. The objects Ob(B) are finite
sets; for any two objects A, B, the morphisms HomB (A, B) are the finite corre-
spondences (or spans) from A to B, that is, triples (X, s, t) where X is a finite
set, and s : X → A and t : X → B are set maps, called the source map and the
target map, respectively. We usually denote such correspondences by diagrams
smmm X QQQQ
vmmm
t
QQ( ; and we often drop s and t from the notation if they are
A B
irrelevant to the discussion. The identity morphism IdA ∈ HomB (A, A) is the cor-
Idnnn A PPPId
PPP . Composition is given by fiber product. That is,
respondence vnnn (
A A
for X in HomB (A, B) and Y in HomB (B, C), the composition Y ◦ X is defined to
be the fiber product Y ×B X = {(y, x) ∈ Y × X | s(y) = t(x)} in HomB (A, C):
Y ×B X T
jj TTTT
tjjjj T*
mm X U UUUU jjj Y QQQQQQ
mm UUU* tjjjjjj
vmm (
A B C.
For morphisms X, Y in HomB (A, B), define the 2-morphisms from X to Y to be
∼
=
the bijections X −→ Y so that the following diagram commutes:
X
<<<
<<
Y Q <<<
QQQ <
vmmmmmm QQ(
A B.
The Burnside category B is self-dual, in the sense of being isomorphic to
its own opposite B op . The isomorphism preserves objects, and sends (X, s, t) ∈
HomB (A, B) to (X, t, s) ∈ HomBop (B, A). The category of finite sets is essentially
a subcategory of B since we can view any set map f : A → B as the correspondence
Idnnn A QQQQ
f
vnnn QQ( . (This statement is not literally true because composition is
A B
only preserved up to 2-isomorphism.)
We will need the following functor from B to the category of Abelian groups
Ab.
Definition 3.1. Define the functor A : B → Ab as follows. For A ∈ Ob(B),
define A(A) = ZA, the free Abelian group with basis A. For a correspondence
(X, s, t) ∈ HomB (A, B), define the map A(X) : A(A) → A(B) by defining it on the
basis elements a ∈ A as

A(X)(a) = s−1 (a) ∩ t−1 (b) b.
b∈B
If, in B, we replace correspondences with isomorphism classes of correspon-

dences, we obtain an ordinary category. The functor A identifies this category
with a subcategory of Ab, whose objects are the finitely generated free Abelian
groups with a chosen basis and whose morphisms are represented by matrices of
nonnegative integers.
4. Functors from the cube to B

The central objects that we will study are pairs (F, r), where F is a functor
2n → B and r ∈ Z is an integer; we will use the notation Σr F to denote such a
pair, and call it a stable functor. When it is unlikely to cause confusion we will
suppress Σr from the notation, and refer to a stable functor F .
The cube category 2n is a strict category, while the Burnside category B is a
weak 2-category, so we owe some explanation regarding what we mean by a functor
F : 2n → B. We will treat 2n as a 2-category with no non-identity 2-morphisms,
and the functor F will be a strictly unitary lax 2-functor. (For general definitions,
see [Bén67, Definition 4.1 and Remark 4.2] which calls such functors strictly unitary
homomorphisms.)
Definition 4.1. A strictly unitary lax 2-functor F from the cube category 2n
to the Burnside category B consists of the following data:
(1) a finite set F (v) ∈ Ob(B) for every v ∈ {0, 1}n ,
(2) a finite correspondence F (ϕu,v ) ∈ HomB (F (u), F (v)) for every u > v, and
(3) a 2-isomorphism Fu,v,w : F (ϕv,w )×F (v) F (ϕu,v ) → F (ϕu,w ) for every u > v > w,
so that for all u > v > w > z, the following diagram commutes:
(4.1)
Id ×Fu,v,w
F (ϕw,z ) ×F (w) F (ϕv,w ) ×F (v) F (ϕu,v ) / F (ϕw,z ) ×F (w) F (ϕu,w )
Fv,w,z ×Id Fu,w,z

F (ϕv,z ) ×F (v) F (ϕu,v ) / F (ϕu,z ).
Fu,v,z

(Here, F (ϕw,z )×F (w) F (ϕv,w )×F
(v) F (ϕu,v ) denotes either F (ϕw,z )×F (w) F (ϕv,w )
×F (v) F (ϕu,v ) or F (ϕw,z ) ×F (w) F (ϕv,w ) ×F (v) F (ϕu,v ) , which are not the same
but are canonically identified.)
We will typically refer to strictly unitary lax 2-functors simply as 2-functors or

even just functors.
Example 4.2. Let X be a finite Δ-complex (as in [Hat02, Section 2.1]) such
that every k-simplex contains (k + 1) distinct vertices. (For example, X could be
a finite simplicial complex.) For each k, let X(k + 1) denote the set of k-simplices
of X, and let X(0) = ∅. Let {p1 , . . . , pn } = X(1) be the vertices of X. To this we
associate the stable functor Σ−1 F X where F X : 2n → B is given by

F X (v) = {Δ ∈ X(|v|) | ∀i (vi = 1) ⇐⇒ (pi ∈ Δ) } for every v ∈ {0, 1}n ,
F X (ϕu,v ) = {(Δv , Δu ) ∈ F X (v) × F X (u) | Δv ⊂ Δu } for every u > v in {0, 1}n .
The source and target maps for the correspondences are the two projection maps.
X
For every u > v > w, the 2-isomorphisms Fu,v,w are uniquely determined, and the
uniqueness forces Diagram (4.1) to commute.
Like the name, the data for a strictly unitary lax 2-functor 2n → B might seem
unwieldy, but fortunately there is a smaller formulation. Consider the following
three pieces of data:
(D-1) for every vertex v ∈ {0, 1}n of the cube, a finite set F (v) ∈ Ob(B),
(D-2) for every edge u • v of the cube, a finite correspondence F (ϕu,v ) ∈
HomB (F (u), F (v)), and
rr• v MM•
(D-3) for every two-dimensional face u I t• w of the cube, a 2-morphism
•
v
Fu,v,v ,w : F (ϕv,w ) ×F (v) F (ϕu,v ) → F (ϕv ,w ) ×F (v ) F (ϕu,v ),
satisfying the following two conditions:
rr• v MM• −1
(C-1) for every two-dimensional face u I t• w , Fu,v ,v,w = Fu,v,v ,w , and
•
v
v • w F
• HH
zzz vvHv•• FF•
(C-2) for every three-dimensional face u E • v H w •• z , the following com-
E• vHvH• yy
• y
v • w
mutes:
×Id
v,w ,w ,z
F
F (ϕw ,z ) ×F (w ) F (ϕv,w ) ×F (v) F (ϕu,v ) / F (ϕw ,z ) ×F (w ) F (ϕv,w ) ×F (v) F (ϕu,v )
Id ×F Id ×F
u,v,v ,w u,v,v ,w

F (ϕw ,z ) ×F (w ) F (ϕv ,w ) ×F (v ) F (ϕu,v ) F (ϕw ,z ) ×F (w ) F (ϕv ,w ) ×F (v ) F (ϕu,v )
F ×Id ×Id
v ,w ,w,z
F
v ,w ,w,z
Id ×F
u,v ,v ,w

F (ϕw,z ) ×F (w) F (ϕv ,w ) ×F (v ) F (ϕu,v ) / F (ϕw,z ) ×F (w) F (ϕv ,w ) ×F (v ) F (ϕu,v ).
A strictly unitary lax 2-functor F produces data (D-1)–(D-3) satisfying condi-

−1
tions (C-1)–(C-2), by simply declaring that Fu,v,v ,w = Fu,v ,w ◦ Fu,v,w . Conversely:
Proposition 4.3. Assume we are given data (D-1)–(D-3) satisfying condi-

tions (C-1)–(C-2). Then up to natural isomorphism, there is exactly one strictly
unitary 2-functor F : 2n → B that produces it.
Proof. This is [LLS, Lemma 2.12], but for completeness, we give a proof. For
both existence and uniqueness, we need the following facts about maximal chains
on the cube 2n . Fix u ≥k v, and consider maximal chains
u = z0 •· · · • zi •· · · • zk =v
. Then:
(m-1) Any two such maximal chains m1 and m2 can be connected by a sequence of
swaps across two-dimensional faces, that is, by a sequence of replacements

of chains · · · • zi−1 • zi • zi+1 • · · · by · · · • zi−1 • zi • zi+1 • · · · .
(m-2) Any two such sequences s1 and s2 connecting any two such maximal chains
m1 and m2 can be related by a sequence of moves of the following three
types:
(a) Replacing a sequence of the form {. . . , m , m , m , . . .} with the sequence
{. . . , m , . . .}.
(b) Exchanging a sequence of one of the following two forms for the other:
. .
. .
. .
· · · • zi−1 • zi • zi+1 • · · · • zj−1 • zj • zj+1 • · · • zi−1 • zi • zi+1 • · · · • zj−1 • zj • zj+1 • · · ·
· · · • zi−1 • zi • zi+1 • · · · • zj−1 • zj • zj+1 • · o / · • zi−1 • zi • zi+1 • · · · • zj−1 • zj • zj+1 • · · ·
· · · • zi−1 • zi • zi+1 • · · · • zj−1 • zj • zj+1 • · · • zi−1 • zi • zi+1 • · · · • zj−1 • zj • zj+1 • · · ·
. .
. .
. . ,
where j − i ≥ 2. (This corresponds to exchanging the order of swaps across

two faces which share no edges.)
(c) Exchanging a sequence of one of the following two forms for the other:
.. ..
. .

··· • zi • zi+1 • zi+2 • zi+3 •· · · ··· • zi • zi+1 • zi+2 • zi+3 •· · ·
··· • zi
• zi+1
• zi+2 • zi+3 •· · · o / ··· • zi • zi+1
• zi+2 • zi+3 •· · ·


.. ..
. . .
(This corresponds to the six faces of the cube

• zi+2
zi+1
JttJ• JJJ
yyy
•
• •

· · · • zi EE• zi+1 zi+2 • zi+3 •· · · .)
E• J
tJ
t J• tt
• t
•
• zi+2
zi+1
These facts are easy to check directly. Alternatively, they are obvious from a
geometric reformulation. The maximal chains in the cube correspond to the vertices
of the permutohedron Πk−1 . The edges of Πk−1 correspond to swaps across two-
dimensional faces, as described in (m-1), and the two-dimensional faces of Πk−1 are
either squares or hexagons, corresponding to the second and third moves of (m-2).
(For more details about the permutohedron, see, for instance, [Zie95].) Therefore,
(m-1) can be restated by saying that any two vertices of Πk−1 can be connected
by a path along the edges, and (m-2) can be restated by saying that any two such
paths can be connected by homotoping across two-dimensional faces.
Now consider the given data (D-1)–(D-3) that satisfies conditions (C-1)–(C-2).
For each u ≥k v, choose a maximal chain
mu,v = { u = z0u,v •· · · •z
i
u,v
•· · · •z
k
u,v
=v}
and define
F (ϕu,v ) = F (ϕzk−1
u,v
,zku,v ) ×F (zk−1
u,v
) · · · ×F (z1u,v ) F (ϕz0u,v ,z1u,v ).
For u ≥k v ≥ w, define the 2-isomorphism Fu,v,w : F (ϕv,w ) ×F (v) F (ϕu,v ) →

F (ϕu,w ) by choosing a sequence of maximal chains connecting mv,w ∪ mu,v to mu,w
of the type described in (m-1), and using the maps provided by the data (D-3). Since
the data satisfies conditions (C-1)–(C-2), (m-2) implies that the 2-isomorphism is
independent of the choice of the sequence of maximal chains. The same argument
shows that these 2-isomorphisms satisfy Diagram (4.1). Therefore, this defines a
strictly unitary lax 2-functor F : 2n → B. The construction is clearly unique up to
natural isomorphism.
Notice that the existence of the Fu,v,v ,w implies the following:

rr• v MM•
(C-0) for every two-dimensional face u I t• w , F (ϕv,w ) ×F (v) F (ϕu,v ) =
•
v
F (ϕv ,w ) ×F (v ) F (ϕu,v ) .
The converse is not true: given data (D-1)–(D-2) satisfying condition (C-0), there
might be no way, one way, or more than one way of constructing data (D-3) satis-
fying conditions (C-1)–(C-2), as the following two examples illustrate.
Example 4.4. Assume we have the following data (D-1)–(D-2) on the 2-
s• 10 K•
dimensional cube 11 K 00 :
• s•
01
◦
a1 b1
a2 b2
◦ ◦
c1 d1
c2 d2
◦
(We have not labeled the elements of the one-element sets F (v), only the ele-
ments of the two-element correspondences F (ϕu,v ).) This does not uniquely specify
data (D-3). Up to natural isomorphism, we may assume
F11,10,01,00 : {a1 b1 , a1 b2 , a2 b1 , a2 b2 } → {c1 d1 , c1 d2 , c2 d1 , c2 d2 }
sends a1 b1 to c1 d1 ; but there are still six ways to define the bijection F11,10,01,00
that are not naturally isomorphic. (Indeed, some of these six functors are not even
equivalent in the sense of Definition 5.9; see Remark 5.13.)
Example 4.5. Assume we have the following data (D-1)–(D-2) on the 3-

111 • 110
• •
101 • • 100
dimensional cube • :
011 • • 010 •
• •
001 • 000
a
p]1=D aaaaaaaaaaaaaaaa2 1 aaaaa◦@◦ 999
p2" ;] AA ]=D]=D=]D]]]]]]]]aa4 3]]]]] ◦## >> @@9b19
"" ; ;AA =D=D ◦## EE# E> >>b2@ 9
"" ;;A;AA =D=D=D ## ## E EEb3> @@9@99
" ;;AA cc1DD
; c3A2===DD ## ## b4 E>> @9@9@9
"" # # EE>> @9
"" c
;; A ◦ ◦ ]]]]]dd2 1 ]]## ]## ]]]]]]E]E>]E>]]@9◦
A
4; A =
" ; ◦! ` `` d3`````# ` # ````````◦"
"" ◦%%`!!` ` d4 ## ""
" %% ! ## ## ""
" !! ## # #
" " % %%! # # ""
◦" ?E ` ` %!!` ```è ```` ` `# ◦; ""
`
È````` !% ;;
◦ A ""
e2 1
] ] ]?] ] ] ]
◦:@:@ ?E?E ]]]] ]! % ]e4 3]]]]]◦ C◦== A e A
:@:@ ?E?E ! !%% =
CC f2A ; ; "
""
f
:@:@@ ?g?1E ! %
1
f3= AA;
::@@ g2?EEE !! %% f 4 ;
CC== AA; "
::@ ?? E E! % CC== A;A; ""
g3@ ?? ! CC== A; ;"
g4: @@ ◦]]]]]h1 ]]]]]]]]]] C=C=]] A
C]
:: @ ◦
: ◦ aaa h3aaaaaaaaaaaaaaq2 q1
h2
◦aaa h4
(We have only labeled the elements of the two-element sets F (111) and F (000),
and the elements of some of the correspondences F (ϕu,v ). In case the picture is not
clear, the correspondence F (ϕ110,010 ) has eight elements with the following (source,
target) pairs: (t(a1 ), t(e1 )), (t(a1 ), t(e3 )), (t(a2 ), t(e2 )), (t(a2 ), t(e4 )), (t(a3 ), t(e1 )),
(t(a3 ), t(e3 )), (t(a4 ), t(e2 )), and (t(a4 ), t(e4 )); and the correspondence F (ϕ101,001 )
also has eight elements with the following (source, target) pairs: (t(c1 ), t(g1 )),
(t(c1 ), t(g3 )), (t(c2 ), t(g2 )), (t(c2 ), t(g4 )), (t(c3 ), t(g2 )), (t(c3 ), t(g3 )), (t(c4 ), t(g1 )),
and (t(c4 ), t(g4 )).) Let us attempt to construct data (D-3). Up to natural isomor-
phism, we may assume the bijection F111,110,101,100 sends a1 b1 to c1 d1 . Looking at
the correspondence from p1 to q1 , conditions (C-1)–(C-2) force F011,010,001,000 (e1 f1 )
= g1 h1 . Similarly, using the correspondence from p1 to q2 , we get F011,010,001,000 (e3 f3 )
= g3 h3 . Using the correspondence from p2 to q1 , the former implies
F111,110,101,100 (a3 b3 ) = c4 d4 ;
but using the correspondence from p2 to q2 , the latter implies F111,110,101,100 (a3 b3 )
= c3 d3 , which is a contradiction. So even though the given data satisfies condi-
tion (C-0), it is impossible to construct data (D-3) satisfying conditions (C-1)–(C-2).
5. Properties of such functors

In this section, we will discuss a few constructions that can be carried out with
stable functors from the cube to the Burnside category.
Definition 5.1. If F : 2n → B, then Tot(F ) ∈ Kom is defined to be the
following chain complex. The chain group is defined to be

Tot(F ) = A(F (v)),
v∈2n
the homological grading of the summand A(F (v)) is defined to be |v|, and the
differential ∂ : Tot(F ) → Tot(F ) is defined by declaring the component ∂u,v of ∂
that maps from A(F (u)) to A(F (v)) to be

(−1)su,v A(F (ϕu,v )) if u ≥1 v,
∂u,v =
0 otherwise.
(Here su,v is the sign assignment from Definition 2.1 and A : B → Ab is the functor
from Definition 3.1.) For all r ∈ Z, define Tot(Σr F ) = Σr Tot(F ), the chain complex
with gradings shifted up by r.
Remark 5.2. In terms of the reformulation from Proposition 4.3, the functor
F : 2n → B is equivalent to data (D-1)–(D-3) satisfying conditions (C-1)–(C-2). In
order to define the chain complex Tot(F ), it is enough to have data (D-1)–(D-2)
satisfying condition (C-0).
Definition 5.3. If F, F : 2n → B are two 2-functors, then the coproduct
F F : 2n → B is defined as follows.
(1) For all v ∈ 2n , (F F )(v) = F (v) F (v).
(2) For all u > v, (F F )(ϕu,v ) = F (ϕu,v ) F (ϕu,v ) with the source and target
maps defined component-wise.
(3) For all u > v >
w, the map (F F )u,v,w
from (F F )(ϕv,w ) ×(F F )(v)
(F F )(ϕu,v ) ∼

= F (ϕv,w ) ×F (v) F (ϕu,v ) F (ϕv,w ) ×F (v) F (ϕu,v ) to (F

F )(ϕu,w ) = F (ϕu,w ) F (ϕu,w ) is defined to be Fu,v,w on the first component

and Fu,v,w on the second component.
It is straightforward to check that this defines a strictly unitary lax 2-functor 2n →
B, and Tot(F F ) = Tot(F ) ⊕ Tot(F ).
Definition 5.4. If F1 : 2n1 → B, F2 : 2n2 → B are two 2-functors, then the
product F1 × F2 : 2n1 +n2 → B is defined as follows.
(1) For all (v1 , v2 ) ∈ 2n1 × 2n2 , (F1 × F2 )((v1 , v2 )) = F1 (v1 ) × F2 (v2 ).
(2) For all (u1 , u2 ) > (v1 , v2 ), (F1 × F2 )(ϕ(u1 ,u2 ),(v1 ,v2 ) ) = F1 (ϕu1 ,v1 ) × F2 (ϕu2 ,v2 )
with the source and target maps defined component-wise, with the understand-
ing that if ui = vi the correspondence Fi (ϕui ,vi ) is the identity.
(3) For all (u1 , u2 ) > (v1 , v2 ) > (w1 , w2 ), the map (F1 × F2 )(u1 ,u2 ),(v1 ,v2 ),(w1 ,w2 ) is
defined as

(F1 × F2 )(u1 ,u2 ),(v1 ,v2 ),(w1 ,w2 ) (x1 , x2 ) = (F1 )u1 ,v1 ,w1 (x1 ), (F2 )u2 ,v2 ,w2 (x2 ) ,
with the understanding that if ui = vi or vi = wi then (Fi )ui ,vi ,wi is the identity
map.
Once again, it is straightforward to check that this defines a strictly unitary lax
2-functor 2n1 +n2 → B. This time, Tot(F1 × F2 ) = Tot(F1 ) ⊗ Tot(F2 ): the sign
assignment from Definition 2.1 translates into the Koszul sign convention on the
tensor product.
Definition 5.5. A face inclusion ι is a functor 2n → 2N that is injective on
objects, and preserves the relative gradings. Face inclusions can be described as
functors of the following form: Fix U ≥n W in 2N , and let {V1 , . . . , Vn } = {V ∈
2N | U ≥n−1 V ≥1 W }. The functor that sends the object (v1 , . . . , vn ) ∈ 2n to the
object (W + i vi (Vi − W )) ∈ 2N is a face inclusion.
Let |ι| = |ι(v)| − |v| for any v ∈ 2n be the grading of W .
Remark 5.6. The autoequivalences ι : 2n → 2n are face inclusions. Note that

the group of autoequivalences of 2n is the permutation group Sn , where σ ∈ Sn cor-
responds to the autoequivalence that sends (v1 , . . . , vn ) ∈ 2n to (vσ−1 (1) , . . . , vσ−1 (n) ).
Definition 5.7. If ι : 2n → 2N is a face inclusion, and F : 2n → B is a functor,
then the induced functor Fι : 2N → B is uniquely defined by imposing F = Fι ◦ ι,
and for all objects v ∈ 2N \ ι(2n ), Fι (v) = ∅.
Note that the chain complexes Tot(Fι ) and Σ|ι| Tot(F ) are canonically isomor-
phic except for signs. For all u ≥1 v in 2n , the component of the differential
from A(F (u)) to A(F (v)) has sign (−1)sι(u),ι(v) in Tot(Fι ) and sign (−1)|ι|+su,v in
Σ|ι| Tot(F ). To every v ∈ 2n , assign tv ∈ Z/2, so that for all u ≥1 v, tu + tv =
|ι| + su,v + sι(u),ι(v) . (Such an assignment exists, and is unique up to adding a
constant to all the assignments.) Then the map Tot(Fι ) → Σ|ι| Tot(F ), defined to
be (−1)tv times the identity on the summand A(F (v)) for all v is an isomorphism
of chain complexes, and is canonical up to an overall sign.
Definition 5.8. A natural transformation η : F → F between two 2-functors
F, F : 2n → B is a strictly unitary lax 2-functor η : 2n+1 → B so that η|{1}×2n = F
and η|{0}×2n = F (with respect to the obvious identifications of {i} × 2n with 2n ).
Note that a natural transformation η induces a chain map Tot(η) : Tot(F ) →
Tot(F ), and this is functorial in the following sense: If η : F → F and θ : F → F
are two natural transformations, then Tot(θ ◦ η) = Tot(θ) ◦ Tot(η). Moreover, Tot
of the functor 2n+1 → B determined by η is the mapping cone of the chain map
Tot(η).
Definition 5.9. Two stable functors (E1 : 2m1 → B, q1 ) and (E2 : 2m2 →
B, q2 ) are defined to be stably equivalent if there is a sequence of stable functors
{(Fi : 2ni → B, ri )} for 0 ≤ i ≤ , with Σq1 E1 = Σr0 F0 and Σq2 E2 = Σr F , such
that for every adjacent pair {Σri Fi , Σri+1 Fi+1 }, one of the following holds:
(1) (ni , ri ) = (ni+1 , ri+1 ), and there is a natural transformation η, either from Fi
to Fi+1 or from Fi+1 to Fi , so that the induced map Tot(η) is a chain homotopy
equivalence.
(2) ri+1 = ri − |ι| and Fi+1 = (Fi )ι for some face inclusion ι : 2ni → 2ni+1 ; or
ri+1 = ri + |ι| and Fi = (Fi+1 )ι for some face inclusion ι : 2ni+1 → 2ni .
We call such a sequence {Σri Fi } a stable equivalence between Σq1 E1 and Σq2 E2 .
Note that a stable equivalence induces a chain homotopy equivalence Tot(Σq1 E1 ) →
Tot(Σq2 E2 ), well-defined up to choices of inverses of chain homotopy equivalences
and an overall sign.
As is standard, instead of having to consider an arbitrary sequence of zig-zags
of natural transformations, it is enough to consider a single zig-zag.
Proposition 5.10. If stable functors (E1 : 2m1 → B, q1 ) and (E2 : 2m2 →
B, q2 ) are stably equivalent, then there exist stable functors (F1 : 2n → B, r),
(F2 : 2n → B, r), and (G : 2n → B, r), satisfying the following for all i ∈ {1, 2}:
(1) Fi = (Ei )ιi for some face inclusion ιi : 2mi → 2n , and qi = r + |ιi |.
(2) There is a natural transformation ηi : Fi → G, so that Tot(ηi ) is a chain ho-
motopy equivalence.
Proof. We can compose natural transformations η : F → F and η : F → F
to get a natural transformation η ◦ η : F → F ; and since Tot(η ◦ η) = Tot(η ) ◦
Tot(η), if Tot(η) and Tot(η ) are chain homotopy equivalences, so is Tot(η ◦ η).

Similarly, for any face inclusions ι : 2n → 2n and ι : 2n → 2n , the composition

ι ◦ ι : 2n → 2n is a face inclusion with |ι ◦ ι| = |ι| + |ι |; and for any F : 2n → B,
(Fι )ι = Fι ◦ι . Finally, if η : F → F is a natural transformation between functors
F, F : 2n → B, and ι : 2n → 2N is a face inclusion, then there is an induced natural
transformation ηι : Fι → Fι ; and if Tot(η) is a chain homotopy equivalence, so is
Tot(ηι ).
Using these moves, we can convert any stable equivalence {Σri Fi } from Σq1 E1
to Σq2 E2 to one of the form {Σq1 E1 , Σr F1 = Σr G0 , Σr G1 , . . . , Σr G−1 , Σr G =
Σr F2 , Σq2 E2 }, where G0 , . . . , G are all functors 2n → B; Fi = (Ei )ιi for some
face inclusion ιi : 2mi → 2n and qi = r + |ιi |; and there is a zig-zag of natural
transformations connecting {G0 , . . . , G }, inducing chain homotopy equivalences
among Tot(Gi ).
So, in order to prove the proposition, it is enough to show that we can we can
η η
convert a zig-zag of the form F ← G → F with Tot(η) and Tot(η ) chain homotopy
θ θ
equivalences to a zig-zag of the form F → H ← F with Tot(θ) and Tot(θ ) chain
homotopy equivalences. We can achieve this by working on the cube 2n+1 instead
of 2n . That is, we will construct H : 2n+1 → B, and θ : Fι0 → H, θ : Fι0 → H so
that Tot(θ) and Tot(θ ) are chain homotopy equivalences (with ι0 denoting the face
inclusion 2n ∼
= {0} × 2n → 2n+1 ).
To define H, note that the natural transformations η and η are thought of
as functors 2n+1 → B satisfying η|{1}×2n = η |{1}×2n = Gι1 |{1}×2n (with ι1 de-
noting the face inclusion 2n ∼ = {1} × 2n → 2n+1 ), and η|{0}×2n = Fι0 |{0}×2n , and
η |{0}×2n = Fι0 |{0}×2n . Define H to be the quotient of η η by identifying η|{1}×2n

θ θ
with η |{1}×2n . The natural transformations Fι0 → H and Fι0 → H come from
inclusions. Since Tot(η) is a chain homotopy equivalence, so is Tot(θ ); and since
Tot(η ) is a chain homotopy equivalence, so is Tot(θ). This concludes the proof.
We conclude this section with some illustrative examples.
Example 5.11. Let P : 2 → B denote the functor that assigns one-element

sets to 1 and 0, and a two-element correspondence to ϕ1,0 . Extend the data from
(1)
Example 4.4 to a functor F (1) : 22 → B by declaring the matching F11,10,01,00 to
be the map
a1 b1 → c1 d1 a1 b2 → c2 d1 a2 b1 → c1 d2 a2 b2 → c2 d2 .
Then F (1) is naturally isomorphic to P × P.
Example 5.12. Consider once again the data from Example 4.4; this time
(2)
define the functor F (2) : 22 → B by declaring the matching F11,10,01,00 to be the
map
a1 b1 → c1 d1 a1 b2 → c1 d2 a2 b1 → c2 d1 a2 b2 → c2 d2 .
Then F (2) is stably equivalent to Pι0 Pι1 , where ιi is the face inclusion 2 = ∼ {i} ×
111 • 110
• •
2 101 • 100
2 → 2 . To see this, let G be the following diagram on the cube • • :
011 • • 010 •
• •
001 • 000
◦$$ III p3 GG
$$ I2I G
III 2GG
$$ GG
$ ◦$$ p4
2$$ $$ 2
$ $$
p5 I◦IIIII p1 GG
I II $$
2
G
2IIII $ 2GG
IIII $ GG
p6 ◦ 2 p2 .
(We have only labeled some of the elements of some of the sets G(v). We have not
labeled the elements of the correspondences G(ϕu,v ), but merely indicated their
cardinalities if bigger than one.) Define the matching G110,100,010,000 on the right-
(2)
most face to be isomorphic to the matching F11,10,01,00 above. It is straightforward
to verify that we can construct matchings on the other two-dimensional faces in a
unique fashion so that condition (C-2) from Section 4 is satisfied. Therefore, this
defines a functor G : 23 → B.
Let F be the diagram restricted to the objects {p1 , p2 , p3 , p4 } and the cor-
respondences between them, and let F be the diagram restricted to the objects
{p1 , p2 , p5 , p6 } and the correspondences between them. It is easy to verify that both
the inclusions F → G and F → G are natural transformations that induce chain
homotopy equivalences Tot(F ) → Tot(G) and Tot(F ) → Tot(G). Furthermore, F
is naturally isomorphic to Fι where ι is the face inclusion 22 ∼
(2)
= 22 × {0} → 23 ;
and F is naturally isomorphic to (Pι0 Pι1 )ι where ι is the face inclusion 22 ∼

=
{0} × 22 → 23 . Therefore, F (2) is stably equivalent to Pι0 Pι1 .
Remark 5.13. Properties of the topological realizations from Section 7 im-

ply that the two functors F (1) and F (2) from Examples 5.11–5.12 are not stably
equivalent. The cohomology of |P| is F2 , supported in grading zero (by prop-
erty (Sp-1)(c)), and therefore, the spectrum |P| is homotopy equivalent to Σ−1 RP2 .
Since F (1) is naturally isomorphic to P × P, the spectrum |F (1) | is homotopy equiv-
alent to Σ−2 (RP2 ∧ RP2 ) (by property (Sp-4)). On the other hand, since F (2)
is stably equivalent to Pι0 Pι1 , the spectrum |F (2) | is homotopy equivalent to
(Σ−1 RP2 ) ∨ RP2 (by properties (Sp-3), (Sp-5), and (Sp-6)). However, the spectra
Σ−2 (RP2 ∧ RP2 ) and (Σ−1 RP2 ) ∨ RP2 are not homotopy equivalent (being distin-
guished by the Steenrod square Sq2 for instance), and therefore, (by property (Sp-6)
once again) the diagrams F (1) and F (2) are not stably equivalent.
6. The Khovanov functor

We turn now to the refinement of Khovanov homology. Fix an oriented link
diagram K with n crossings and an ordering of the crossings of K. Khovanov associ-
ated an n-dimensional cube of free Abelian groups to this data, as follows [Kho00].
Each crossing of K has a 0-resolution and a 1-resolution:

??
?
←− ?? −→
0 1
.
?
So, given a vertex v = (v1 , . . . , vn ) ∈ {0, 1}n of the cube, performing the vi -
resolution at the ith crossing gives a collection Kv of disjoint, embedded circles
in S 2 . Further, for each edge u • v of the cube, Ku is obtained from Kv by either
merging two circles into one or splitting one circle into two. Let V = Zx+ , x− be a
free, rank-2 Z-module, which we endow with a multiplication and comultiplication
by
m(x+ ⊗x+ ) = x+ m(x+ ⊗x− ) = x− m(x− ⊗x+ ) = x− m(x− ⊗x− ) = 0
Δ(x+ ) = x+ ⊗ x− + x− ⊗ x+ Δ(x− ) = x− ⊗ x− .
Define a functor FKh,Ab : (2 ) n op
→ Ab on objects by setting

FKh,Ab (v) = V.
S∈π0 (Kv )
On morphisms, if u • v is an edge of the cube so that Ku is obtained from Kv

by merging two circles then FKh,Ab (ϕop u,v ) applies the multiplication map to the
corresponding factors of FKh,Ab (v) and the identity map to the remaining factors;
if instead Ku is obtained from Kv by splitting one circle then FKh,Ab (ϕop
u,v ) applies
the comultiplication map to the corresponding factor of FKh,Ab (v) and the identity
map to the remaining factors. It is straightforward to verify that the resulting
diagram commutes. The total complex of this cube (i.e., multiplying the map on
the edge u • v by (−1)su,v and summing over vertices of each grading, in a fashion
similar to Definition 5.1) is the Khovanov complex CKh (K).
The Khovanov cube decomposes as a direct sum over quantum gradings. To
define the quantum grading, notice that for v ∈ {0, 1}n , FKh,Ab (v) comes with a
preferred basis: the labelings of the circles in Kv by elements of {x− , x+ }, i.e.,
functions π0 (Kv ) → {x+ , x− }. The quantum grading of a basis element is
grq (v, x : π0 (Kv ) → {x+ , x− })
= n+ − 2n− + |v| + |{Z | x(Z) = x+ }| − |{Z | x(Z) = x− }| ∈ Z,
where n+ and n− are the number of positive and negative crossings of K, respec-
tively. There is also a homological grading shift of −n− .
Our main goal is to refine FKh,Ab to a functor FKh : 2n → B, satisfying the
following:
(a) For all v, FKh (v) = {x : π0 (Kv ) → {x+ , x− }} is the preferred basis of Kho-
vanov generators.
(b) The above identification induces an isomorphism Σ−n− Tot(FKh ) ∼ = (CKh )∗ .
By Proposition 4.3 it suffices to define FKh on vertices, edges, and 2-dimensional
faces. For all vertices v, FKh (v) is already defined. Next, notice
that for each edge
u • v and each element y ∈ FKh (v), FKh,Ab (ϕop u,v )(y) = x∈FKh (u) x,y x where
x,y ∈ {0, 1}. (In other words, all of the entries of the matrix FKh,Ab (ϕop u,v ) are 0
or 1.) Define FKh (ϕu,v ) = {(y, x) ∈ FKh (v) × FKh (u) | x,y = 1}, with the obvious
source and target maps to FKh (u) and FKh (v).
2 2 2
1 1 1
(a) (b) (c) (d) (e) (f)
Figure 6.1. The ladybug matching. (a) A piece of a (perhaps)

partially-resolved link diagram. (b) The 00-resolution, along with the arcs
av and av , drawn with thick lines. This is a ladybug configuration. (c)
The same configuration, up to isotopy in S 2 , looking more like a ladybug.
The right pair of arcs is labeled 1 and 2. (d)–(e) The 10 and 01 resolutions.
The circles in the 10 resolution and the circles in the 01 resolution can be
identified via the induced numbering by 1, 2. (f) The 11 resolution.
So far, there is no information in FKh beyond that in the Khovanov complex.

The new information is in the definition of Fu,v,v ,w : FKh (ϕv,w )×FKh (v) FKh (ϕu,v ) →
rr• v MM•
FKh (ϕv ,w ) ×F (v ) FKh (ϕu,v ) for the 2-dimensional faces u I t• w . The fact that
•
v
FKh,Ab is a commutative diagram implies that there is a 2-isomorphism between
FKh (ϕv,w ) ×FKh (v) FKh (ϕu,v ) and FKh (ϕv ,w ) ×FKh (v ) FKh (ϕu,v ). Namely, for x ∈
FKh (u) and z ∈ FKh (w), the cardinalities of
Ax,z := s−1 (x) ∩ t−1 (z) ⊆ FKh (ϕv,w ) ×FKh (v) FKh (ϕu,v ) and
Ax,z := s−1 (x) ∩ t−1 (z) ⊆ FKh (ϕv ,w ) ×FKh (v) FKh (ϕu,v )
op
are the (x, z) entries in the matrices FKh,Ab (ϕop op
u,v )◦FKh,Ab (ϕv,w ) and FKh,Ab (ϕu,v )◦
op
FKh,Ab (ϕv ,w ), respectively, and these two matrices are the same. Further, in
most instances, these sets have 0 or 1 elements, so there is a unique isomorphism
Fu,v,v ,w |Ax,z : Ax,z → Ax,z .
The exceptional case is when there is a circle Cw in Kw which splits to form
two circles in each of Kv and Kv , and these same two circles merge to form a
single circle Cu in Ku ; x labels Cu by x− ; and z labels Cw by x+ . We call this
configuration a ladybug configuration because of the following depiction. First, draw
the circle Cw . Then, draw the arc av with endpoints on Cw which pinches along the
edge v • w (i.e., performing embedded 1-surgery on Kw along av produces Kv ).
Similarly, draw another arc av with endpoints on Cw which pinches along the edge
v • w . Up to isotopy in S 2 , the result looks like the ladybug in Figure 6.1. Now,
distinguish a pair of arcs in (Cw , ∂av ∪ ∂av ), the right pair, as the two arcs you get
to by walking along av or av to Cw and then turning right. (Again, see Figure 6.1.)
Number the two arcs in the right pair as 1, 2—the numbering will not matter. Label
the two circles in Kv which come from Cw as Cv1 and Cv2 , according to whether
they contain the right arc numbered 1 or 2. Similarly, label the two circles in Kv
which come from Cw as Cv1 and Cv2 , according to whether they contain the right
arc numbered 1 or 2. The two elements of Ax,z are

α = ((Cw → x+ ), (Cv1 , Cv2 ) → (x− , x+ ), (Cu → x− )),
β = ((Cw → x+ ), (Cv1 , Cv2 ) → (x+ , x− ), (Cu → x− ))
while the two elements of Ax,z are
α = ((Cw → x+ ), (Cv1 , Cv2 ) → (x− , x+ ), (Cu → x− )),
β = ((Cw → x+ ), (Cv1 , Cv2 ) → (x+ , x− ), (Cu → x− ))
The bijection Fu,v,v ,w sends α to α and β to β . (See also [LS14a, Section 5.4].)
Proposition 6.1. The definitions above satisfy conditions (C-1)–(C-2) from
Section 4, and so induce a strictly unitary 2-functor FKh : 2n → B by Proposi-
tion 4.3.
Sketch of Proof. The proof is essentially the same as the proof of [LS14a,
Proposition 5.19] and, indeed, the result follows from [LS14a, Proposition 5.20]
and [LLS, Lemma 4.2], so we will only sketch the argument here.
Condition (C-1) is immediate, so we only need to check condition (C-2). Fix
v1 D• v2 B
}•
zzD•• BB
}} •
a three-dimensional face u A• v1 D v2 ••w . We have six correspondences from
A zDz•D ||
• • |
v1 • v2
FKh (u) to FKh (w) coming from the six maximal chains from u to w. Fix Khovanov
generators x ∈ FKh (u) and z ∈ FKh (w), and consider the subsets s−1 (x) ∩ t−1 (z) of
these six correspondences. The various coherence maps Fu,v(i) ,v(j) and Fv(i) ,v(j) ,w
1 2 1 2
produce six bijections connecting these six sets, and we need to check that these
bijections agree.
Unless one of the two-dimensional faces of the above cube is a ladybug con-
figuration, each of the above six sets will contain 0 or 1 element, and the check is
vacuous or trivial, respectively. So we concentrate on configurations that contain
ladybugs, and check them by hand. Using some underlying symmetries, we can
reduce to just four configurations [LS14a, Figure 5.3. a–c, e]. The first three are
similar, so there are essentially two case checks that need to be performed; and the
proofs of [LS14a, Lemmas 5.14, 5.17] imply that the checks succeed.
Hu-Kriz-Kriz gave an intrinsic description of the functor FKh [HKK16], that

does not require Proposition 4.3, as follows. On objects, FHKK agrees with FKh .
To define FHKK on morphisms, first fix a checkerboard coloring for K. There is an
induced checkerboard coloring of each resolution Kv . Further, each morphism ϕu,v
corresponds to a cobordism Σu,v in [0, 1]×S 2 from Ku to Kv , and the checkerboard
coloring of K induces a coloring of the components of ([0, 1] × S 2 ) \ Σu,v . Let
Bu,v ⊂ ([0, 1] × S 2 ) \ Σu,v denote the closure of the black region. Then

(6.1) H1 (Bu,v )/H1 (Bu,v ∩ {0, 1} × S 2 ) ∼
= Zg(Σ0 ) .
Σ0 ∈π0 (Σu,v )
If each connected component of Σu,v has genus 0 or 1 then we call a vector

(1 , . . . , k ) in the group (6.1) such that each i = ±1 a sign choice for Σu,v . By a
valid boundary labeling of Σu,v we mean a labeling of the circles in ∂Σu,v by x+ or
x− satisfying the condition that every component Σ0 of the cobordism has genus
equal to
1 − |{C ∈ π0 (Kv ∩ ∂Σ0 ) | x(C) = x− }| − |{C ∈ π0 (Ku ∩ ∂Σ0 ) | x(C) = x+ }| .
(In particular, if some component Σ0 has genus bigger than 1 then the set of
valid boundary labelings is empty.) Then FHKK (ϕu,v ) is the product of the set of
valid boundary labelings for Σu,v and the set of sign choices for Σu,v . The source
and target maps for FHKK (ϕu,v ) are given by restricting the boundary labelings
of Σu,v to Ku and Kv , respectively. The coherence map FHKK (ϕv,w ) ×FHKK (v)
FHKK (ϕu,v ) → FHKK (ϕu,w ) is obvious except when a genus 0 component Σ0 of
FHKK (ϕu,v ) and a genus 0 component Σ1 of FHKK (ϕv,w ) glue to give a genus 1
component Σ of FHKK (ϕu,w ). In this last case, there is a single circle C of Σ0 ∩ Σ1 ,
that is non-separating in Σ and is labeled x+ . Orient C and Σ as the boundaries of
the black regions in S 2 and [0, 1] × S 2 , respectively, and choose a circle D in Σu,w
with intersection number D ·C = 1. Either the pushoff of C or the pushoff of D into
Bu,w is a generator of the new Z-summand of (6.1); use this generator to extend the
sign choice to Σu,w . This finishes the construction of FHKK , and verifying that this
does, in fact, define a 2-functor is straightforward. The functor FHKK is naturally
isomorphic to the functor FKh defined via the ladybug matching [LLS, Lemma 8.1].
Definition 6.2. Associated to an oriented n-crossing link diagram K, let
FKh (K) : 2n → B be the functor constructed in Proposition 6.1. Define the Kho-
vanov functor to be the stable functor Σ−n− FKh (K), where n− is the number of
negative crossings in K.
There is also a reduced Khovanov functor associated to a pointed link (K, p):
the basepoint p chooses a preferred circle Cv,p in each resolution Kv , and we declare
that
(v) = {x ∈ FKh (v) | x(Cv,p ) = x− },
FKh
−1 −1
(ϕu,v ) = s
FKh (u)) ∩ t
(FKh (v)) ⊆ FKh (ϕu,v ),
(FKh
and similarly the coherence maps (FKh )u,v,w are restrictions of the coherence maps
(FKh )u,v,w for FKh . It is straightforward to see that FKh does define a strictly
unitary 2-functor. (Replacing x− with x+ in the definition would give a naturally
isomorphic functor F .)
Kh
(K, p) : 2 → B be the above functor associated to a

n
Definition 6.3. Let FKh
pointed, oriented n-crossing link diagram K. Define the reduced Khovanov functor
to be the stable functor Σ−n− FKh
(K), where n− is the number of negative crossings
in K.
Since the chain complexes CKh and CKh decompose according to quantum grad-
ings, so do the functors FKh and FKh
:
j j
FKh = FKh FKh
= F ,
Kh
j∈Z j∈Z
j
where FKh (v) = {x ∈ FKh (v) | grq (v, x) = j}, and F j (v) = FKh j−1
(v) ∩ FKh (v).
Kh
Theorem 1. If K1 and K2 are oriented link diagrams, with n1− and n2− nega-
tive crossings, respectively, representing isotopic oriented links then the Khovanov
functors Σ−n− FKh (K1 ) and Σ−n− FKh
1 j 2 j
(K2 ) are stably equivalent stable functors.
Similarly, if (K1 , p1 ) and (K2 , p2 ) are pointed, oriented link diagrams, with n1− and
n2− negative crossings, respectively, representing isotopic pointed, oriented links then
the reduced Khovanov functors Σ−n− F j (K1 , p1 ) and Σ−n− F j (K2 , p2 ) are stably
1 2
Kh Kh
equivalent stable functors.
Sketch of Proof. The proof is essentially the same as the proof of invariance
of the Khovanov spectrum [LS14a, Theorem 1.1], so we only give a sketch. We
start with a general principle. Since CKh (K) comes with a preferred basis, so does
the dual complex (CKh (K))∗ . Suppose that S is a subset of the preferred basis for
(CKh (K))∗ so that the span of S is a subcomplex of (CKh (K))∗ . Then there is a
S
functor FKh : 2n → B defined by
S
FKh (v) = S ∩ FKh (v),
(6.2) S
FKh (ϕu,v ) = s−1 (FKh
S
(u)) ∩ t−1 (FKh
S
(v)) ⊆ FKh (ϕu,v ),
with coherence maps induced by the coherence maps for FKh (K). Further, inclusion
induces a natural transformation η : FKhS
→ FKh (K), and if span(S) → (CKh (K))∗
is a quasi-isomorphism then, for any m ∈ Z, η is a stable equivalence between
the stable functors Σm FKh S
and Σm FKh (K). Similarly, if S is a subset of the
∗
preferred basis for (CKh (K)) so that the complement of S spans a subcomplex of
(CKh (K))∗ then Formula (6.2) still defines a functor FKh S
: 2n → B, and there is a
natural transformation η : FKh (K) → FKh defined as follows. Recall that a natural
S
transformation is a functor η : 21 × 2n → B. Define η(ϕ1,0 × Idv ) = FKh S

(v) =
S ∩ FKh (v), with source map given by the inclusion FKh (v) → FKh (v) and target
S
map the identity map FKh S

(v) → FKh
S
(v); the definition extends in an obvious way
to all of 2 × 2 . If the quotient map (CKh (K))∗ → span(S) is a quasi-isomorphism
1 n
then η is a stable equivalence between the stable functors Σm FKh (K) and Σm FKh S
.
To prove the theorem, it suffices to check invariance under the three Reidemeis-
ter moves. (For pointed links, we need to check invariance under the Reidemeister
moves in the complement of the basepoint.) By the previous paragraph, it suffices
to show that the (duals of the) isomorphisms on Khovanov homology induced by
Reidemeister moves come from sequences of:
• Inclusions of subcomplexes spanned by Khovanov generators, inducing isomor-
phisms on homology.
• Projections to quotient complexes spanned by Khovanov generators, inducing
isomorphisms on homology.
• Face inclusions of cubes.
For Reidemeister I and II moves, Bar-Natan’s formulation [Bar02] of Khovanov’s
invariance proof [Kho00] has this form, so proves invariance of the stable func-
tor as well (see also [LS14a, Propositions 6.2 and 6.3]). For Reidemeister III
moves, it suffices to prove invariance under the braid-like Reidemeister III move,
which locally has the form σ1 σ2 σ1 σ2−1 σ1−1 σ2−1 ∼ Id (where the σi are braid gen-
erators) [Bal11, Section 7.3]. For this braid-like Reidemeister III, one can again
reduce to the identity braid by a sequence of subcomplex inclusions and quotient
complex projections, though the sequence is somewhat tedious [LS14a, Proposition
6.4]. The result follows.
The following properties are strightforward from the definitions. (See [LLS,
Proposition 9.1].)
(X-1) If K1 and K2 are oriented links then

j
FKh (K1 K2 ) ∼
=
j1
FKh j2
(K1 ) × FKh (K2 ).
j1 +j2 =j
(X-2) If (K1 , p1 ) is a pointed, oriented link, and K2 is an oriented link then

F j (K1 K2 , p1 ) ∼= F j1 (K1 , p1 ) × FKh
j2
(K2 ).
Kh Kh
j1 +j2 =j
(X-3) If (K1 , p1 ) and (K2 , p2 ) are pointed, oriented links and (K1 #K2 , p) is the
connected sum of K1 and K2 at the basepoints and the new basepoint p
is chosen on one of the connected sum strands, then

F j (K1 #K2 , p) ∼
= F j1 (K1 , p1 ) × F j2 (K2 , p2 ).

Kh Kh Kh
j1 +j2 =j
Section 7 describes a recipe for turning a stable functor Σr F into a spectrum

|Σ F |. Applying that recipe to the Khovanov functor associated to a link K (re-
r
spectively, reduced Khovanov functor associated to a pointed link (K, p)) gives the
Khovanov stable homotopy type XKh (K) (respectively, XKh (K, p)).
7. Spaces
Finally, we return to the connection with topological spaces. Given a diagram
F : 2n → B, one can associate an essentially well-defined spectrum |F |. To give a
concrete construction of |F |, following [LLS, Section 5], start by fixing an integer
≥ n. We build a diagram G : 2n → CW of based CW complexes, as follows. Given
a vertex v, let

G(v) = (F (v) × [0, 1] )/(F (v) × ∂([0, 1] )) S .
x∈F (v)
For each v > w, choose an embedding

Φ(ϕv,w ) : F (ϕv,w ) × [0, 1] → F (v) × [0, 1]
satisfying the following conditions:
(1) For each a ∈ F (ϕv,w ), Φ({a} × [0, 1] ) ⊆ {s(a)} × [0, 1] ; that is, the following
diagram commutes:
Φ(ϕv,w ) /
F (ϕv,w ) × [0, 1] F (v) × [0, 1]

F (ϕv,w )
s / F (v).
(2) For each a ∈ F (ϕv,w ), the embedding ({a} × [0, 1] ) → ({s(a)} × [0, 1] ) is an
inclusion as a sub-box; that is, the lower arrow in the following diagram

{a} × [0, 1] / {s(a)} × [0, 1]
∼
= ∼
=

[0, 1] / [0, 1]
is a map of the form

(x1 , . . . , x ) → (c1 + d1 x1 , . . . , c + d x )
for some non-negative constants c1 , d1 , . . . , c , d .

Define G(ϕv,w ) to be the composition
F (v) × [0, 1] /∂ → F (ϕv,w ) × [0, 1] /∂ → F (w) × [0, 1] /∂,
where the ∂ symbols denote the unions of the boundaries of the cubes, the first
map sends a point of the form Φ(a, x) to (a, x) ∈ F (ϕv,w ) × [0, 1] and all other
points to the collapsed boundary, and the second map sends (a, x) to (t(a), x).
The resulting map G : 2n → CW does not commute on the nose: the maps
G(ϕv,w ) ◦ G(ϕu,v ) and G(ϕu,w ) are (probably) defined using different choices of
embeddings. However, the spaces of embeddings Φ are ( − 2)-connected, so since
any sequence of composable morphisms in 2n has length at most n ≤ , the cube
commutes up to coherent homotopies, i.e., is a homotopy coherent diagram in the
sense of Vogt [Vog73]. Homotopy coherent diagrams are essentially as good as
commutative diagrams: they appear naturally whenever one takes a commutative
diagram of spaces and replaces all of the objects by homotopy equivalent ones,
and conversely every homotopy coherent diagram can be replaced by a homotopy
equivalent honestly commutative diagram essentially canonically.
Next, we add a single object ∗ to the category 2n , and a single morphism
v → ∗ for each non-terminal object v of 2n , to get a bigger category 2+n
. Extend
G to a functor G+ : 2+ → CW by declaring that G+ (∗) is a single point. Finally,
n
take the homotopy colimit of G+ (as defined by Vogt for homotopy coherent dia-
grams [Vog73]) to define
F = hocolim(G+ ).
Adding the object ∗ and taking the homotopy colimit is an iterated version of
the mapping cone construction. In particular, if G : 21 → CW were a diagram
on the one-dimensional cube, then G would consist of the data of a cellular map
between two CW complexes, G(ϕ1,0 ) : G(1) → G(0), and hocolim(G+ ) would be the
mapping cone of G(ϕ1,0 ). This iterated mapping cone of G is a space-level version
of the totalization construction from Definition 5.1. So, unsurprisingly, F carries
a CW complex structure so that its reduced cellular chain complex C •cell (F ) can
be identified with Σ Tot(F ), implying H • (F ) ∼
= Σ H• (Tot(F )).
Similarly, given any natural transformation η : F → F between diagrams
F, F : 2n → B, and an ≥ n + 1, one can construct a based cellular map
η : F → F so that the induced map on the reduced cellular chain com-
plexes is the map Σ Tot(η) : Σ Tot(F ) → Σ Tot(F ). This is not functorial on the
nose, since the construction depends on the choices of embeddings Φ. However, the
space of choices of the coherent homotopies in the construction of the homotopy
coherent diagrams 2n → CW is an ( − n − 2)-connected space, and the space of
choices in the construction of the map is an ( − n − 1)-connected space. By allow-
ing to be arbitrarily large, we can make these constructions essentially canonical,
namely, parametrized by a contractible space.
Therefore, to make the space independent of and the choices of embeddings,
and to make the construction functorial, we replace the CW complex F by its
suspension spectrum, and then formally desuspend times:
|F | = Σ− (Σ∞ F ).
Alternately, one could replace the diagram G by its suspension spectrum before
taking the homotopy colimit. Finally for any stable functor Σr F , define |Σr F | to
be the formal suspension Σr |F |.
There are several other ways of turning a diagram F : 2n → S into a space:
(1) Produce a diagram G : 2n → S by viewing F as a functor F to the category
of permutative categories, by letting F(v) = Sets/F (v) be the category of sets
over F (v), and then applying K-theory. Then take the iterated mapping cone,
as above. This is the procedure used by Hu-Kriz-Kriz [HKK16].
(2) Turn F into a (rather special) flow category, in the sense of Cohen-Jones-Segal,
and then apply the Cohen-Jones-Segal realization [CJS95]. This is essentially
the approach taken in our previous work [LS14a].
(3) Without making any choices, produce a diagram F : 2 n
→ S from a slightly
larger category 2, and then take an appropriate iterated mapping cone of
n
F [LLS, Section 4].

These constructions all give homotopy equivalent spectra [LLS, Theorem 3].
For convenience, we summarize some of the properties of this functor
n
| · | : B2 → S .
(Sp-1) For any stable functor (F : 2n → B, r),
(a) |Σr F | is the formal suspension Σr |F |.
(b) The spectrum |F | is homotopy equivalent to a formal desuspension of the
suspension spectrum of some finite CW complex F .
(c) There is an identification of Tot(F ) with the cellular chain complex of
|F | from (Sp-1)(b). In particular, H• (|F |) ∼
= H• (Tot(F )) and H • (|F |) ∼
=
•
H (Tot(F )).
(Sp-2) For any natural transformation η : F → F between diagrams F, F : 2n →
B, and any integer r,
(a) The map |Σr η| is the formal suspension Σr |η|.
(b) The map |η| is homotopic to a formal desuspension of some cellular map
η : F → F .
(c) With respect to the identification from (Sp-1)(c) and (Sp-2)(b), the map
Tot(η) can be identified with the map induced by |η| on the reduced
cellular chain complex.
(Sp-3) For any two functors F, F : 2n → B, there is a canonical homotopy equiv-
alence |F F | |F | ∨ |F |.
(Sp-4) For any two functors F1 : 2n1 → B and F2 : 2n2 → B, there is a canonical
homotopy equivalence |F1 × F2 | |F1 | ∧ |F2 |.
(Sp-5) If ι : 2n → 2N is a face inclusion, then there is a canonical homotopy
equivalence |F | |Σ−|ι| Fι |.
(Sp-6) In particular, from (Sp-5) and (Sp-2)(c), stably equivalent stable functors
give homotopy equivalent spectra.
See [LLS], particularly Section 4, for more details about the construction of |F |,
and for proofs of these properties.
8. Some questions
j
We end with a few questions. First, the Khovanov homotopy type XKh (K)
i,j
contains more information than the Khovanov homology Kh (K) = H (X (K)).
i j
Kh
j
Specifically, the Khovanov spaces XKh (K) induce Steenrod operations on Kh i,j (K).
One can give a combinatorial formula for Sq2 : Kh i,j (K; F2 ) → Kh i+2,j (K; F2 )
[LS14c]; using this, Seed showed that there are knots with isomorphic Khovanov
homologies but distinct Steenrod squares [See]. Further, using Sq2 (K), one can
2
give refinements sSq
± [LS14b] of Rasmussen’s s-invariant [Ras10] and, using direct
computations and the connected sum formula from Properties (X-1) and (Sp-4),
one obtains (modest) new concordance results [LLS, Corollary 1.5].
The formula for Sq2 is combinatorial and not hard to formulate in terms of
FKh (K), but does not seem particularly obvious in this language. We do not know
how to generalize the formula to higher Steenrod squares, or (reduced) pth powers.
So:
Question 8.1. Are there nice formulations of the action of the Steenrod algebra
on Kh(K), purely in terms of the stable Khovanov functor from Definition 6.2? Of
other algebro-topological invariants of XKh (K) (such as the K-theory or bordism
groups)? Do these give additional concordance invariants?
As described in Section 7, one can turn stable functors 2n → B into spectra.
Perhaps this operation loses useful information:
Question 8.2. Are there stable functors Σk F and Σ G so that the spectra
|Σ F | and |Σ G| are homotopy equivalent but Σk F and Σ G are not stably equiva-
k
lent? If so, are there useful knot invariants that can be obtained from the Khovanov
functor which are lost on passing to XKh (K)?
There are other formulations of Khovanov homology, including Cautis-
Kamnitzer’s formulation via algebraic geometry [CK08], formulations by Web-
ster [Weba, Webb], Lauda-Queffelec-Rose [LQR15], and others via categorified
quantum groups, and Seidel-Smith’s formulation via Floer homology [SS06] (see
also [AS]). In the Floer homology case, the Cohen-Jones-Segal program [CJS95]
is expected to produce a Floer spectrum.
Question 8.3. Does the Cautis-Kamniter or Webster formulation of Khovanov
homology have a natural extension along the lines of stable functors to the Burnside
category?
Question 8.4. Is the Floer spectrum (conjecturally) given by applying the
Cohen-Jones-Segal construction to the Seidel-Smith formulation of Khovanov ho-
mology homotopy equivalent to XKh (K)?
Finally, it would be nice to have an explicit description of the stable Khovanov
functor or the Khovanov homotopy type in more cases. One possible direction
would be to understand the stable invariant in the sense of [Sto07, GOR13]:
Question 8.5. Do the Khovanov functors for (m, N ) torus knots admit a limit
as N → ∞, and if so, is there a simple description of the limit functor or the
associated spectrum?
Remark 8.6. Since the first version of this paper was written, Willis and
Lobb-Orson-Schuetz have shown that the Khovanov homotopy types of torus knots
stabilize (cf. Question 8.5) [Wilb, LOS]. In fact, they have deduced a more general
stabilization phenomenon and deduced the existence of colored Khovanov stable
homotopy types [LOS, Wila].
Acknowledgments
The authors thank the referee for comments on a draft of this paper. The
second author also thanks the organizers of the conference “Categorification in
Algebra, Geometry and Physics” for creating a stimulating environment in which
to discuss these results.
References
[AS] M. Abouzaid and I. Smith, Khovanov homology from Floer cohomology,
arXiv:1504.01230.
[Bal11] J. A. Baldwin, On the spectral sequence from Khovanov homology to Heegaard Floer
homology, Int. Math. Res. Not. IMRN (2011), no. 15, 3426–3470. MR2822178
(2012g:57021)
[Bar02] D. Bar-Natan, On Khovanov’s categorification of the Jones polynomial, Algebr. Geom.
Topol. 2 (2002), 337–370 (electronic), DOI 10.2140/agt.2002.2.337. MR1917056
[Bén67] J. Bénabou, Introduction to bicategories, Reports of the Midwest Category Seminar,
Springer, Berlin, 1967, pp. 1–77. MR0220789
[CJS95] R. L. Cohen, J. D. S. Jones, and G. B. Segal, Floer’s infinite-dimensional Morse theory
and homotopy theory, The Floer memorial volume, Progr. Math., vol. 133, Birkhäuser,
Basel, 1995, pp. 297–325. MR1362832 (96i:55012)
[CK08] S. Cautis and J. Kamnitzer, Knot homology via derived categories of coherent sheaves.
I. The sl(2)-case, Duke Math. J. 142 (2008), no. 3, 511–588, DOI 10.1215/00127094-
2008-012. MR2411561
[GOR13] E. Gorsky, A. Oblomkov, and J. Rasmussen, On stable Khovanov homology of torus
knots, Exp. Math. 22 (2013), no. 3, 265–281, DOI 10.1080/10586458.2013.798553.
MR3171092
[Hat02] A. Hatcher, Algebraic topology, Cambridge University Press, Cambridge, 2002.
MR1867354
[HKK16] P. Hu, D. Kriz, and I. Kriz, Field theories, stable homotopy theory, and Khovanov
homology, Topology Proc. 48 (2016), 327–360. MR3465966
[Kau87] L. H. Kauffman, State models and the Jones polynomial, Topology 26 (1987), no. 3,
395–407. MR899057 (88f:57006)
[Kho00] M. Khovanov, A categorification of the Jones polynomial, Duke Math. J. 101 (2000),
no. 3, 359–426, DOI 10.1215/S0012-7094-00-10131-7. MR1740682
[LLS] T. Lawson, R. Lipshitz, and S. Sarkar, Khovanov homotopy type, Burnside category,
and products, arXiv:1505.00213.
[LOS] A. Lobb, P. Orson, and D. Schuetz, A Khovanov stable homotopy type for colored links,
arXiv:1602.01386.
[LQR15] A. D. Lauda, H. Queffelec, and D. E. V. Rose, Khovanov homology is a skew Howe
2-representation of categorified quantum slm , Algebr. Geom. Topol. 15 (2015), no. 5,
2517–2608, DOI 10.2140/agt.2015.15.2517. MR3426687
[LS14a] R. Lipshitz and S. Sarkar, A Khovanov stable homotopy type, J. Amer. Math. Soc. 27
(2014), no. 4, 983–1042, DOI 10.1090/S0894-0347-2014-00785-2. MR3230817
[LS14b] R. Lipshitz and S. Sarkar, A refinement of Rasmussen’s S-invariant, Duke Math. J.
163 (2014), no. 5, 923–952, DOI 10.1215/00127094-2644466. MR3189434
[LS14c] R. Lipshitz and S. Sarkar, A Steenrod square on Khovanov homology, J. Topol. 7 (2014),
no. 3, 817–848, DOI 10.1112/jtopol/jtu005. MR3252965
[Ras10] J. Rasmussen, Khovanov homology and the slice genus, Invent. Math. 182 (2010), no. 2,
419–447, DOI 10.1007/s00222-010-0275-6. MR2729272
[See] C. Seed, Computations of the Lipshitz-Sarkar Steenrod square on Khovanov homology,
arXiv:1210.1882.
[SS06] P. Seidel and I. Smith, A link invariant from the symplectic geometry of nilpotent
slices, Duke Math. J. 134 (2006), no. 3, 453–514, DOI 10.1215/S0012-7094-06-13432-4.
MR2254624
[Sto07] M. Stošić, Homological thickness and stability of torus knots, Algebr. Geom. Topol. 7
(2007), 261–284, DOI 10.2140/agt.2007.7.261. MR2308944
[Vog73] R. M. Vogt, Homotopy limits and colimits, Math. Z. 134 (1973), 11–52. MR0331376
[Weba] B. Webster, Knot invariants and higher representation theory, arXiv:1005.4559.

[Webb] , Tensor product algebras, Grassmannians and Khovanov homology,
arXiv:1312.7357.
[Wila] M. Willis, A colored Khovanov homotopy type for links, and its tail for the unknot,
arXiv:1602.03856.
[Wilb] , Stabilization of the Khovanov homotopy type of torus links, arXiv:1511.02742.
[Zie95] G. M. Ziegler, Lectures on polytopes, Graduate Texts in Mathematics, vol. 152, Springer-
Verlag, New York, 1995. MR1311028
E-mail address: tlawson@math.umn.edu
Department of Mathematics, University of Minnesota, Minneapolis, Minnesota 55455
E-mail address: lipshitz@uoregon.edu
Department of Mathematics, University of Oregon, Eugene, Oregon 97403
E-mail address: sucharit@math.princeton.edu
Department of Mathematics, Princeton University, Princeton, New Jersey 08544

Volume 684, 2017
Junctions of surface operators and

categorification of quantum groups
Sungbong Chun, Sergei Gukov, and Daniel Roggenkamp

Abstract. We show how networks of Wilson lines realize quantum groups
Uq (slm ), for arbitrary m, in 3d SU (N ) Chern-Simons theory. Lifting this con-
struction to foams of surface operators in 4d theory we find that rich structure
of junctions is encoded in combinatorics of planar diagrams. For a particu-
lar choice of surface operators we make a connection to known mathematical
constructions of categorical representations and categorified quantum groups.
Contents
1. Introduction
2. Junctions of Wilson lines and quantum groups
2.1. Junctions of line operators
2.2. Web relations
2.3. Skew Howe duality and the quantum group
2.4. Why “categorification = surface operators”
3. Junctions of surface operators
3.1. Junctions in 4d N = 4 theory
3.2. Line-changing operators in class S and network cobordisms
3.5. OPE of surface operators and the Horn problem
3.6. OPE and Schubert calculus
Domain walls in 4d N = 2 SQCD
4. Categorification and the Landau-Ginzburg perspective
4.1. Physics perspectives on categorification
4.2. LG theory on “time × knot”
4.3. Junctions and LG interfaces
4.4. Junctions and matrix factorizations
4.5. Junctions and categorification of quantum groups
5. What’s next?
Acknowledgments
Appendix A. Wilson lines and categories N Web
Appendix B. Domain walls, junctions and Grassmannians
Appendix C. LG Interfaces and the cohomology of Grassmannians
The work of the second author was funded in part by the DOE Grant DE-SC0011632 and
the Walter Burke Institute for Theoretical Physics.
2017
87
88 SUNGBONG CHUN, SERGEI GUKOV, AND DANIEL ROGGENKAMP
Appendix D. LG Interfaces and 2-categories N Foam

D.1. Derivation of the bubble relation
D.2. Quantum group relations
References
1. Introduction
In this paper we develop a physical framework for categorification of quantum
groups, which is consistent with (and extends) the physical realization of homolog-
ical knot invariants as Q-cohomology of a certain brane system in M-theory (that
we review in section 4).
Our starting point will be the celebrated relation between quantum groups and
Chern-Simons TQFT in three dimensions. Even though it has a long history, many
aspects of this relation remain mysterious, even with respect to some of the most
basic questions. For instance, quantum groups are usually defined via generators
and relations, which are not easy to “see” directly in Chern-Simons gauge theory.
Instead, one can see certain combinations of the generators that implement braiding
of Wilson lines [1].
The realization of quantum groups we discuss in this paper is also based on
Chern-Simons gauge theory, but in constrast to the conventional one, it allows for
gauge group and quantum group to be of completely different rank! Moreover, the
quantum group generators have an immediate and very concrete interpretation.
This approach is based on networks of Wilson lines and gives a physical realization
of the skew Howe duality, which was introduced in the context of knot homology
in [2]: As we will show in section 2, upon concatenation, networks of Wilson lines
exhibit quantum group relations, which for the Lie algebra g = sl2 take a very
simple form:
KK −1 = 1 = K −1 K ,
(1.1) KE = q 2 EK , KF = q −2 F K
K − K −1
[E, F ] = .
q − q −1
Here, the quantum variable q is related to the Chern-Simons coupling constant in
the usual way, c.f. (2.8). What is not usual is that configurations of m Wilson lines
in totally antisymmetric representations
(1.2) R1 ⊗ . . . ⊗ Rm = Λk1 CN ⊗ . . . ⊗ Λkm CN
are interpreted as weight spaces of quantum slm , on which the Chevalley generators
act by adding extra Wilson line segments. This idea is due to [3]. So, generators
and relations of the quantum group have a very concrete realization (shown in
Figures 14 and 10, respectively); and the rank m − 1 of the quantum group has
nothing to do with the rank N − 1 of Chern-Simons gauge theory. Instead it is
determined by the number of incoming (equally, outgoing) Wilson lines.
By categorifying representation theory, one usually means replacing weight
spaces, such as (1.2), by graded categories Rλ on which raising and lowering op-
erators act as functors, c.f. [4–8]. The quantum group itself is promoted to a
JUNCTIONS OF SURFACE OPERATORS 89
2-category U̇, for which Rλ is a 2-representation, as illustrated in Figure 1. Sounds

a bit scary, doesn’t it?
N Foam categorical skew Howe duality U̇ 2-representation R

(2-category) (categorified (category)
(2-functor) quantum group)
categorified
representation
decategorification K0 K0
N Web skew Howe duality U̇ R

(category) (functor) (quantum group) representation (vector space)
Figure 1. Categorified representation theory and categorified

skew Howe duality.
Luckily, the rich and abstract structure of categorical representations and cat-
egorified quantum groups can be made very “user-friendly” and intuitive in the
diagrammatic approach recently developed by Khovanov and Lauda [9–12] (see
[13–15] for excellent expositions and [16] for a related algebraic construction).
One of our main goals is to provide a physical realization of this diagrammatic
approach.
In the context of quantum field theories, categorification can be achieved by
adding a dimension, and networks of Wilson line operators used to realize quantum
groups in Chern-Simons theory become foams of surface operators in 4d. Indeed,
foams have been related to Khovanov-Lauda diagrams in the math literature in
[17, 18]. The physical realizations of foams discussed here are built by sewing
surface operators along 1d junctions, which are interesting in their own right and
so far did not even appear in the physics literature. In section 3, we will not only
present evidence that BPS junctions of surface operators exist, but we will also
discuss various applications, ranging from math to physics.
In fact, the relevant surface operators [19, 20] have already been used to con-
struct group actions on categories (notably, in the context of the geometric Lang-
lands program [21]) and to realize many elements of geometric representation the-
ory [22]. In these realizations, as well as in many other similar problems, groups
acting on categories are generated by codimension-1 walls (interfaces) acting on
categories of boundary conditions. The group law comes from the “fusion” product
of interfaces. This is an instance of the well known fact that the algebraic structure
of various operators and defects in (n + 1)-dimensional topological quantum field
theories is governed by n-categories (see e.g. [20] for a review).
Our physical realization of categorified quantum groups is conceptually similar,
but with several twists. One way to describe it is to consider the world-volume
theory on the foam of surface operators. By moving the m parallel surface defects
representing the slm weight spaces close together we can describe the world-volume
theory of the combined system as tensor product of the theories on the individual
surface operators. Junctions of surface operators introduce interactions between
tensor factors along 1-dimensional loci, and are incorporated by interfaces between
generally different 2d theories. For instance, collapsing the x2 -direction of the
configuration of surface operators on the left of Figure 2, the combined world-

volume theory can be described by a 2d theory in the (x0 , x1 )-plane. The surface
operator supsended along the x2 -direction introduces a 1d interface.
Now, while in the spririt of the remark above the structure of a 2d TQFT
is encoded in a 1-category, different 2d TQFTs together with interfaces between
them form a 2-category, whose objects are not boundary conditions, but rather the
different 2d theories. 1-morphisms are interfaces, and 2-morphisms are interface
changing fields. As a bonus, these 2-categories come with 2-representations on the
categories of boundary conditions of the 2d TQFTs.
It is in this way that we extract the building blocks of categorified quantum
groups U̇ out of foams of surface operators: slm weight spaces correspond to tensor
products of the world-volume theories of m surface operators, and the generators
of the quantum group are realized as interfaces between them.
In fact, this 2d perspective on surface operators and their junctions provides
a physical realization of the planar diagrams of Khovanov-Lauda [9–12] and vast
generalizations, for more general types of surface operators.1
There are various ways to describe relevant 2d TQFTs, and in this paper we
mainly consider two variants: one is based on topological Landau-Ginzburg (LG)
models, while the second involves the UV topological twist of sigma-models whose
targets are certain flag-like subspaces of affine Grassmannians which play an impor-
tant role in the geometric Satake correspondence. We generally favor the former,
where interfaces and their compositions are easier to analyze using matrix factor-
izations [23].
Figure 2. Projecting Σ onto (x0 , x1 ) plane gives a product of

Landau-Ginzburg theories, in which junctions (singular edges of
Σ) are represented as interfaces.
In the Landau-Ginzburg approach (described in detail in section 4) the 2d

space-time is precisely the plane on which the planar diagrams (with dots on the
lines) of Khovanov-Lauda are drawn. Each 2d region of the plane colored by the
1 After embedding in eleven-dimensional M-theory, this will also provide an answer to the
following question: Which two dimensions of space-time, relative to the fivebranes, compose the
plane where the diagrams of [9–12] are usually drawn?
highest weigh λ defines a LG model or, to be more accurate, a product of LG

models. It is basically a projection of Σ from the three-dimensional space — that
later in the text we parameterize with (x0 , x1 , x2 ) — to a two-dimensional plane
(x0 , x1 ). Then, LG interfaces describe transitions between different sheets of Σ,
regarded as a multi-cover of the (x0 , x1 )-plane, c.f. Figure 2. These 1-dimensional
interfaces (defects) are precisely the arcs in the planar diagrams of [9–12]:
region on a plane ↔ weight space (1.2) ↔ LG model LGk1 ⊗ . . . ⊗ LGkm
lines ↔ 1-morphisms ↔ interfaces between LG models
dots ↔ 2-morphisms ↔ interface changing fields
The reader is invited to use this dictionary to translate virtually any question in one
subject to a question in another. For example, it would be interesting to study 2-
categories that one finds for more general types of surface operators in various gauge
theories, and interfaces in Landau-Ginzburg models, other than examples studied
in this paper. Mathematically, they should lead to interesting generalizations of
the Khovanov-Lauda-Rouquier (KLR) algebras [10, 11, 16]. Conversely, it would
be interesting to identify surface operators and LG interfaces for mathematical
constructions generalizing U̇ .
In summary, if you are studying junctions of surface operators or interfaces in
Landau-Ginzburg models, most likely you are secretly using the same mathemat-
ical structure — and, possibly, even the same diagrams — as those categorifying
representations and quantum groups, or interesting generalizations thereof.
2. Junctions of Wilson lines and quantum groups

In this section, we will discuss Wilson lines in SU (N ) Chern-Simons theory
and their junctions. We will show that (upon concatenation) certain networks of
Wilson lines satisfy the defining relations of (Lusztig’s idempotent form [24] of) the
quantum groups U̇q (slm ).
We start by reviewing the necessary techniques from [25–27] in section 2.1.
In section 2.2, we explore various relations satisfied by networks of Wilson lines
and their junctions, which surprisingly include the quantum group relations. We
give an explicit derivation of one such relation, relegating the rest to Appendix A.
Then, by playing with m strands, in section 2.3 we (re)discover the representation
category of the quantum group of slm . This gives a physical realization of the
diagrammatic approach of Cautis-Kamnitzer-Morrison [3] to quantum skew Howe
duality. Finally, in section 2.4, we give an a priori explanation why categorification
of these beautiful facts should involve surface operators and their junctions, leading
us into sections 3 and 4.
2.1. Junctions of line operators. Our starting point is Chern-Simons the-

ory in three dimensions with gauge group G = SU (N ). The action on a closed
3-manifold M3 is given by

κ 2
(2.1) SCS = (A ∧ dA + A ∧ A ∧ A).
4π M3 3
where A is an SU (N )-gauge connection on M3 , and the coupling constant κ ∈ N is
usually called the “level.”
The most important (topological) observables of this theory are derived from
the parallel transport with respect to A: To a path Γ ⊂ M3 and a representation
a Γa
b Γb
˜
c Γc
=

jj
ii i ˜ j
Ua (Γa )ij Ub (Γb )ij Uc (Γc )ij
(a)
k1 + k2
k1 k2
or
k1 + k2 k1 k2
(b)
Figure 3. (a) A trivalent network of Wilson lines and the cor-

responding gauge invariant observable. (b) Trivalent junctions of
Wilson lines in antisymmetric representations: Labels k refer to
representations Λk .
R of the gauge group one associates the Wilson line

(2.2) UR (Γ) = P exp ρR (A) : R −→ R .
Γ
Here P denotes path-ordering, and ρR is the associated representation on the Lie
algebra. For an open path Γ such UR (Γ) is not gauge invariant, but one can form
gauge invariant combinations, such as the familiar Wilson loops
(2.3) WR (Γ) = trR UR (Γ)
by closing the path Γ. More general gauge invariant combinations can be associated
to networks of Wilson lines [26] by forming junctions of Wilson lines in representa-
tions R1 , . . . , Rn and contracting the corresponding observables URi with invariant
tensors
(2.4) ∈ Hom(R1 ⊗ . . . ⊗ Rn , C) .
The latter correspond to additional junction fields, which need to be specified along
with representations of Wilson lines in order define the observable.
For instance, to the graph in Figure 3(a), one can associate an observable
by contracting the Wilson lines Ua (Γa ), Ub (Γb ) and Uc (Γc ) with invariant tensors
∈ a ⊗ b ⊗ c in the “incoming” junction and ∈ a ⊗ b ⊗ c in the “outgoing”
junction of the graph. Here R denotes the dual representation. (The Wilson lines
carry gauge indices i, j of representation a, i , j of b, and i , j of c, respectively.
After contracting the gauge indices of Wilson lines via and , we obtain a gauge
invariant observable.)
In the following we will mostly consider Wilson lines in totally antisymmetric

representations ∧k , of the fundamental representation of SU (N ), which for
ease of notation we will just label by k. These Wilson lines admit two possible
trivalent junctions, depicted in Figure 3(b), which are dual to each other. Since
Hom(Λk1 ⊗Λk2 , Λk1 +k2 ) is one-dimensional, there is just one possible junction
field, whose normalization we will fix shortly. We can then omit the labels of the
junctions. As it turns out, junctions of more than three such Wilson lines can all
be factorized into trivalent junctions, c.f. Figure 10(a).
In order to unambiguously define Wilson line invariants in the quantum theory,
a framing, i.e. the choice of orthogonal vector fields along the Wilson lines is re-
quired, which fit together in the junctions. This is in particular needed to regularize
the self-linking number of Wilson loops. In our discussion, we will always consider
two-dimensional projections of graphs of Wilson lines, and choose the corresponding
vertical framing.
Moreover, Wilson lines in the quantum theory are labeled only by those repre-
sentations of SU (N ) which correspond to integrable highest weight representations
of su(N )κ ; all other Wilson lines decouple in the quantum theory. The totally
antisymmetric representations considered here are integrable for all κ ≥ 1.
Now that we have defined the gauge invariant observables we are interested
in, let us proceed to summarize some relevant machinery from [25–27] to compute
their expectation values.
Hilbert space and the connected sum formula. Quantization of Chern-Simons
theory on a product Σ × R associates to any surface Σ a Hilbert space HΣ . The
Chern-Simons path integral on an open 3-manifold M3 with boundary ∂M3 = Σ
gives rise to a vector Z(M3 ) = |M3 Σ ∈ HΣ . Moreover, if a closed 3-manifold M3
(1) (2)
can be cut along a surface Σ into two disconnected components M3 = M3 #Σ M3 ,
then the path integral on M3 evaluates to the scalar product
(1) (2)
Z(M3 ) = Σ M3 |M3 Σ
(1) (2)
of the vectors associated to M3 and M3 . This also holds in the presence of Wilson
lines, as long as they intersect the surface Σ transversely. The corresponding Hilbert
space then also depends on the punctures, at which the surface is pierced by the
Wilson lines and the associated representations.
It turns out that the Hilbert spaces HΣ are isomorphic to the spaces of confor-
mal blocks of the s u(N )κ -WZW models on Σ, where, at the intersection points, pri-
mary fields with the respective integrable highest weight representations of s u(N )κ
are inserted [25]. These spaces are well studied. For instance, the Hilbert spaces
associated to the 2-sphere with two punctures labeled by irreducible representations
Ri and Rj is one-dimensional if Ri ∼ = Rj and zero-dimensional otherwise:
(2.5) dimHΣ{Ri ,Rj } = δRi ,Rj .
As before R is the dual of the representation R. This in particular implies the
following. Consider a configuration of Wilson lines on a 3-sphere M3 = S 3 , which
intersects a given great sphere Σ = S 2 at two points colored by representations
R and R , c.f. Figure 4. Denote the hemispheres (with Wilson lines) obtained by
(1) (2) (1)
cutting M3 along Σ by M3 and M3 . Now, according to (2.5), we have Z(M3 ) =

|M3 S 2 {R,R } = 0 if R R . If, on the other hand, R ∼
(1) (1)
= R , then |M3 S 2 {R,R} is
proportional to the vector |N3 S 2 {R,R} that corresponds to a hemisphere N3 with
R R
R
= δR,R̄
R
R
Figure 4. An illustration of the connected sum formula, applied

to a network of Wilson lines in S 3 and a separating 2-sphere S 2 that
cuts the network at two Wilson lines in representations R and R .

When R ∼ = R , its partition function factorizes into contributions
from the Wilson line networks in the left and right hemisphere,
respectively, where the R and R = R Wilson lines are joined.
a single strand of R Wilson line connecting two points in the boundary S 2 . The
proportionality constant is given by
(1)
(1) S 2 {R,R} N3 |M3 S 2 {R,R}
(2.6) |M3 S 2 {R,R} = |N3 S 2 {R,R} ,
S 2 {R,R} N3 |N3 S 2 {R,R}
which in particular implies the connected sum formula

(2.7)
(2) (1)
(2) (1) S 2 {R,R} M3 |N3 S 2 {R,R} S 2 {R,R} N3 |M3 S 2 {R,R}
S 2 {R,R} M3 |M3 S 2 {R,R} = .
S 2 {R,R} N3 |N3 S 2 {R,R}
Here, the inner products on the right are the S 3 partition function with three
configurations of Wilson lines: the two in the numerator are obtained by joining the
(2) (1)
R and R ends of the Wilson line configurations in M3 and M3 , respectively, while
the one in the denominator is a single Wilson loop (the unknot) in the representation
R. A pictorial representation of this formula is given in Figure 4.
Framing and skein relations. Surgeries other than a simple connected sum en-
able us to study braiding of Wilson lines, skein relations, and to compute the expec-
tation values of Wilson loops colored by various representations. In general, given
a surgery presentation of M3 , we can compute the expectation values of Wilson line
operators in it by the surgery formula [25].
R R R R
= = e2πihR R and = = e−2πihR R
Figure 5. Wilson lines with extra twists.
The simplest example is the relation among twisted Wilson lines. The path
integral on a 3-ball containing a straight Wilson line in some representation R
ending on the boundary 2-sphere determines a vector in the one-dimensional Hilbert
space HS 2 {R,R} . This is also true when the Wilson line has extra twists, as in
q −1/N − q 1/N + (q−q −1 ) = 0.
Figure 6. Skein relation of vertically framed Wilson lines in the

fundamental representation .
Figure 5. In particular, the vectors associated to Wilson lines with different number
of twists are proportional. Assuming vertical framing, the Wilson lines in Figure 5
are related by (±1)-Dehn twists on the boundary 2-sphere. The proportionality
constant in this case is e2πihR where hR is the conformal weight of the primary field
of the WZW model transforming in representation R.
In a similar fashion one obtains skein relations. Consider a 3-ball with two
Wilson lines labeled by the fundamental representation , ending on the boundary
2-sphere. The associated Hilbert space HS 2 ,{,,,} is two-dimensional, so the
three configurations of Wilson lines in Figure 6 have to satisfy a linear relation. The
coefficients of this relation can be determined from the fact that the configurations
are related by half-twists on the boundary 2-sphere, c.f. [25, 26]. In Figure 6 we
expressed them in terms of the variable
(2.8) q = eπi/(N +κ) .
Note that the exact form of the relation depends on the choice of framing. We
use the vertical framing throughout this paper, which is why the skein relation
here looks different from its more familiar form in the canonical framing2 , in which
the coefficients of the first two terms are q −N and −q N , respectively. A thorough
discussion of this can be found in [26].
Next, let us show how to obtain the expectation value of an unknotted Wilson
loop in S 3 labeled by an irreducible representation Ri . The idea is that such a
configuration can be obtained by Dehn surgery on S 2 × S 1 , with the Wilson line
running along the S 1 . More precisely, let M3 be a tubular neighborhood of the
Wilson line in S 2 × S 1 , and M3 its complement. Then S 3 can be obtained by
gluing M3 and M3 along the boundary torus T 2 , with a non-trivial identification
by the global diffeomorphism S ∈ SL(2, Z):
(2.9) Z(S 3 , Ri ) = T 2 M3 |S|M3 , Ri T 2 .
Here we use the crucial fact that the mapping class group of T 2 acts on HT 2 by the
modular transformation on the characters of the respective Kac-Moody algebra. In
particular,

(2.10) S|M3 , Ri T 2 = Sij |M3 , Rj T 2 ,
j
where Sij is the modular S-matrix. Hence,

(2.11) Z(S 3 , Ri ) = Sij T 2 M3 |M3 , Rj T 2 = Sij Z(S 2 × S 1 , Rj ) ,
j j
but Z(S × S , Rj ) = trHS2 {R } (1) = dim(HS 2 {Rj } ), which is 1 if Rj is the trivial

2 1
j
representation and 0 otherwise. It immediately follows that the expectation value
2 in which self-linking numbers of knots are zero

k1
k2 N
k1 +k2
˜ = k1 +k2 k1
k1 + k2
Figure 7. The evaluation of a “θ-web” that determines the nor-

malization of the vertices and
. Here k stands for a totally
antisymmetric representation ∧k .
of a Wilson loop in S 3 is given by the quantum dimension

Z(S 3 , Ri ) Si0
(2.12) WRj S 3 = = .
Z(S 3 , R0 = triv) S00
For Wilson lines in the totally antisymmetric representations in SU (N ) Chern-
Simons theory at level κ, this gives

S0,∧m N
(2.13) W∧m = = ,
S0,0 m
where

N [N ] · · · [N − m + 1] q N − q −N
(2.14) = , [N ] =
m [m] · · · [1] q − q −1
is the quantum binomial coefficient.
Tetrahedral network and normalization of trivalent vertices. As we explained
earlier, there are only two possible trivalent junctions between Wilson lines in totally
antisymmetric representations, c.f. Figure 3(b). Moreover, the spaces of junction
fields
(2.15) ∈ Hom(Λk1 +k2 , Λk1 ⊗ Λk2 ) ,

∈ Hom(Λk1 ⊗ Λk2 , Λk1 +k2 )
are each one-dimensional. We choose and as positive multiples of the respective
antisymmetrizations of the identity maps k1 ⊗ k2 ←→ k1 +k2 . Note that this
choice depends on the ordering of (k1 , k2 ), where a change of the ordering leads
to a sign factor (−1)k1 k2 . The normalization is fixed by requiring the relation in
Figure 7 to hold. Note that this differs from the normalization used in [26], where
N N N
the “θ-web” would evaluate to k1 k2 k1 +k2 . Also, our choice of signs leads
a a a a
b = eiπ(ha +hb −hc ) b b = e−iπ(ha +hb −hc ) b
c c , c c
a a a a
b = eiπ(hb +hc −ha ) b b = e−iπ(hb +hc −ha ) b
c c , c c
Figure 8. Relations between vertices in the conventions of [26].

Our choice for the vertices between totally antisymmetric repre-
sentations Λk1 , Λk2 and Λk1 +k2 involves an extra sign factor
(−1)k1 k2 .
k k
k+1k
= −q −(1+ N )
1 N
−q −(1+ N )
1
k+1 k−1 = k+1 k−1 k+1 1 1
k k
Figure 9. Evaluation of the tetrahedral web.
to an additional factor of (−1)k1 k2 in the vertex relations depicted in Figure 8,

which are valid in the conventions of [26], when a, b, c are totally antisymmetric
representations Λk1 , Λk2 and Λk1 +k2 .
We conclude this review by computing the expectation value of the tetrahedral
network in Figure 9. It has four Wilson lines in antisymmetric representations
∧k , ∧k , ∧k+1 , ∧k−1 and two diagonal Wilson lines in the fundamental ,
which are positively crossed. The braiding relations for vertices (c.f. Figure 8) on
the two junctions at the end of a diagonal line, imply that the expectation value of
this tetrahedron is proportional to the expectation value of the one, in which the
fundamental Wilson lines are negatively crossed. The constant of proportionality
can be easily calculated to be q −2− N . (The relevant conformal weights satisfy
2
exp(2πi h(Λk )) = q (1+ N )(N −k)k .) Using the skein relation of Figure 6 one arrives
1
at the first equation in Figure 9. The second equality follows from the connected
sum formula (Figure 4).
2.2. Web relations. Networks of Wilson lines in totally antisymmetric rep-
resentations satisfy the linear relations depicted in Figure 10. These relations have
interesting implications. For instance, identities (e) and (f) are part of the defining
relations of the quantum group U̇q (slm ). The connection between Wilson lines and
quantum groups is the subject of section 2.3 below. Here, we will demonstrate how
to derive such relations using the one in Figure 10(e) as an example. The proofs of
the other relations are deferred to Appendix A.
Consider the three configurations of Wilson lines in Figure 10(e) (all lying in 3-
balls with ends on the boundary 2-sphere). The path integral in these systems gives
three vectors in HS 2 ,{m,j,m,j} . Since the dimensions of this space is greater than 2
for general m and j, it is not a priori clear that a relation we are seeking exists.
In order to obtain it, we start with the relations among networks in Figure 11(a).
That such relations (with some coefficients) have to hold follows from the fact that
HS 2 ,{1,m,1,m} is 2-dimensional. Here, G stands for the adjoint representation, and
we have chosen non-trivial vertices at the end-points of the Wilson lines colored by
G. (In fact, HS 2 ,{G,m,m} is one-dimensional, so the gauge invariant tensor lying at
the junction of G, m, m-colored Wilson lines is proportional to the multiplication
G ⊗ ∧m → ∧m . Nevertheless, the precise normalization need not be specified,
since as long as xm and x m in Figure 11(a) are nonzero, we will eventually get
Figure 12. That xm and x m are nonzero can be easily shown by closing off the
Wilson lines of Figure 11(a) in two inequivalent ways.) Inserting these relations
into larger networks of Wilson lines leads to relations of Figure 11(b).
Next, from 1-dimensionality of HS 2 ,{G,j,j} one derives the relations in Figure
11(c). (Again, we have chosen non-trivial junction fields.) This allows us to relate
the first two terms in the identities of Figure 11(b), and hence to eliminate them
i j k i j k
(a) j +k = i+j
i+j +k i+j +k
j j j j
N −j j
(b) m j +m = and m j −m =
m m
j j
m m
m+1
(c) m = m + [N − m − 1] m−1
m+1
m m
l m l m l m
n
m−1 l−1 m−1
(d) l+n m−n = + m+l
n n−1
l+n−1
m+l−1 m+l−1 m+l−1
m j m j m j
(e) m+1 j −1 − m−1 j +1 = [j −m]
m j m j
m−2 j +1 k+1 m−2 j +1 k+1 m−2 j +1 k+1
j j +2 m−1 j
(f) [2] = +
j +1 m−1 j +1
m−1 j −1
m j k m j k m j k
Figure 10. Relations among networks of Wilson lines.

m m
m
G
xm + ym + zm m−1 = 0,
m m
(a)
m m
m
G
x̃m + ỹm + z̃m m+1 = 0.
m m
m j m j m j
G
xm j +1 + ym j +1 + zm m−1 j +1 = 0,
m j j m j
(b)
m j m j m j
G
x̃m j −1 + ỹm j −1 + z̃m m+1 j −1 = 0.
m j j m j
j j j j
1 1
(c)
G
j −1 = ηj G
and
G
j +1 = ηj G
1 1
j j j j
Figure 11. (a) Two linear relations among three Wilson lines in
HS 2 ,{1,m,1,m} , (b) two linear relations among three Wilson lines
containing those of (a) in the red dashed box, and (c) proportion-
ality relations between two vectors in HS 2 ,{G,j,j} .
from the relations. One arrives at the new relation depicted in Figure 12, which is
a linear relation of the type we are after. To deduce (e) of Figure 10, it remains to
determine the coefficients α and β.
We will do this in two steps. First we close off the m- and j-colored Wilson
lines in Figure 12. Using the expectation values of all the resulting networks of
Wilson lines that have been determined earlier we obtain the following relation
(2.16) [N ] = α [N − j] [m] − β [N − m] [j] .
Another relation can be found by connecting the incoming m- and j-colored
Wilson lines in Figure 12 with a junction to an incoming j + m Wilson line. Since
HS 2 ,{j,m,j+m} is one-dimensional, the vectors associated to all three configurations
of Wilson lines are proportional to one another. The constants of proportionality
can be easily found: close off Wilson lines in relation of Figure 10(a) in a way
m j m j m j
= α m+1 j −1 − β m−1 j +1
m j m j
Figure 12. A linear relation among three vectors in HS 2 ,{j,m,j,m} .

The coefficients α and β are functions of
m , ym , zm , ηj , ηj .
xm , ym , zm , x
shown in Figure 13(a), then insert the identity of Figure 10(b), and finally apply
the resulting identity twice. The result is depicted in Figure 13(b), from which we
obtain another relation on the coefficients α and β:
(2.17) 1 = α [m] [j + 1] − β [m + 1] [j] .
Together with (2.16) this fixes the sought-after coefficients to be
1 1
(2.18) = [m − j] = ,
α β
finally proving the relation of Figure 10(e).
Let us conclude this subsection by adding a remark on the expectation values
of closed and planar trivalent graphs Γ of Wilson lines. The relations in Figure
10, together with the expectation value of the Wilson loops (equation 2.13), form
the complete set of Murakami-Ohtsuki-Yamada (MOY) graph polynomial relations
which uniquely determines the MOY graph polynomials PN (Γ; q) ∈ Z[q, q −1 ], see
[28]. As the MOY graphs are closed, oriented graphs generated by the junctions
of Figure 3(b), the uniqueness theorem implies that the expectation value of any
planar Wilson lines in antisymmetric representations (and their junctions) can be
computed from the definition of MOY graph polynomials. This not only provides
us a consistency check for our methods, but also a combinatorial way to compute
the expectation value of networks of Wilson lines, as the MOY graph polynomials
are combinatorially defined.
2.3. Skew Howe duality and the quantum group. The quantum group
Uq (sl2 ) is usually defined by means of generators E, F, K, K −1 and relations (1.1).
In finite dimensional representations of Uq (sl2 ), K can be diagonalized with eigen-
values q n , here n ∈ Z is the respective sl2 -weight. E and F raise, respectively lower
the weights by 2. For such representations, one can trade the generators K, K −1
for idempotents 1n , n ∈ Z, projecting on weight spaces of weight n
(2.19) 1n 1m = δn,m 1n .
This leads [4] to a modified quantum group U̇q (sl2 ) generated by E, F and the 1n .
Since K1n = q n 1n , the algebra relations become
(2.20) E1n = 1n+2 E = 1n+2 E1n , F 1n = 1n−2 F = 1n−2 F 1n
[E, F ] 1n = [n] 1n .
i+j k i+j j +k i j +k
i k
(a) j = j and j = j
i k i k
j +k i+j j +k i+j
i+j +k i+j +k i+j +k i+j +k
i+j k i+j k i j +k i j +k
j j
i+j j+k i j j +k i+j j+k i+j j k
(b) i j
= and i j
=
i+j k i j +k
i+j +k i+j +k i+j +k i+j +k
Figure 13. (a) Capping off the “associativity identity”. (b) Re-

lations in HS 2 ,{j,m,j+m} .
It is now very easy to see that if we identify
6 6 6
-1 6 , 6
1 6
(2.21) 1k2 −k1 → , E → F → ,
6 6 6 6
k1 k2
as in [3], then, upon concatenation, these configurations of Wilson lines satisfy

the quantum group relations, (2.19) and (2.20). In particular, the commutation
relation of E and F is nothing but the identity of Figure 10(e), which was explained
in the previous subsection. Here, composition of webs (networks) is drawn from
bottom to top, and Wilson lines with different labels cannot be joined, i.e. their
concatenation vanishes. Note, that E and F do not change k = k1 + k2 , which
characterizes the quantum group representation. For fixed k the identification of
strands and idempotents is unambiguous.
Interestingly, while it is well known that the quantum group Uq (slN ) appears
in the description of SU (N ) Chern-Simons theories, here we realize the quantum
group sl2 in Chern-Simons theory with gauge group SU (N ) for any N . The choice
of gauge group merely restricts the possible sl2 weights which can appear. After
all, the labels ki stand for totally antisymmetric representations Λki in SU (N )
Chern-Simons theory. Hence, the sl2 weights can only lie between −N and N .
The networks of Wilson lines in SU (N ) Chern-Simons theory therefore only realize
quantum group representations with highest weights ≤ N .
k1 k2 km k1 ki −1 ki+1 +1 km k1 ki +1 ki+1 −1 km
1λ → ··· , Ei 1λ → ··· ··· , Fi 1λ → ··· ···
ki ki+1 ki ki+1
Figure 14. U̇q (slm ) idempotents and generators in terms of Wil-

son lines and their junctions.
This construction can easily be generalized by increasing the number of strands

of Wilson lines to m > 2. One defines idempotents 1λ , λ = (k2−k1 , · · · , km−km−1 ) ∈
Zm−1 by m parallel strands of Wilson line with labels k1 , . . . , km . For 1 ≤ i ≤ m, the
generators Ei (resp. Fi ) are defined by suspending a Wilson line in the fundamental
representation between the ith and (i + 1)st strand (resp. between the (i + 1)st
and the ith strand). These definitions are illustrated in Figure 14. Then, using in
particular identities (e) and (f) of Figure 10, one finds that these webs satisfy the
defining relations of the higher rank quantum group U̇q (slm ):
1λ 1λ = δλ,λ 1λ , Ei 1λ = 1λ+li Ei , Fi 1λ = 1λ−l+i Fi ,
(2.22) [Ei , Fj ]1λ = δi,j [λi ]1λ , [Ei , Ej ]1λ = 0 for |i − j| > 1,
(2) (2)
and Ei Ej Ei 1λ = Ei Ej 1λ + Ej Ei 1λ for |i − j| = 1.
Here λ denotes an slm weight. The generators Ei (resp. Fi ) associated to the
simple roots of slm raise (resp. lower) λ by li = (0, . . . , 0, −1, 2, −1, 0, . . . , 0) where
2 appears in position i.
To summarize, in SU (N ) Chern-Simons theory with arbitrary N , we obtain
a realization of U̇q (slm ) on a configurations of Wilson lines with m strands. The
rank of the Chern-Simons gauge group only restricts the possible representations
of the quantum group which can be obtained in this way. What we described is a
physical realization of the skew Howe duality of [2]

(2.23) Λk (Cm ⊗ CN ) ∼ = Λk1 CN ⊗ . . . ⊗ Λkm CN .
k1 +...+km =k
The actions of slm and slN , respectively, on Cm and CN commute, and the direct
sum on the RHS of this equation is indeed the weight decomposition with respect to
the slm action. This duality has a direct generalization to quantum groups U̇q (slm ).
In Chern-Simons theory we interpret each summand on the RHS of (2.23) as
the Hilbert space of m Wilson lines in representations, R1 = Λk1 CN , etc.,
(2.24) R1 ⊗ . . . ⊗ Rm = Λk1 CN ⊗ . . . ⊗ Λkm CN ,
so that raising and lowering operators, Ei and Fi , which relate different weight
spaces are realized as configurations of SU (N ) Wilson lines in totally antisymmetric
representations with m incoming and m outgoing strands.
N −k N −k
= (−1)k(N−k)
k k
Figure 15. “Tag” morphism relation of the N Web category.
In fact, the identification of Figure 14 exactly corresponds to the skew Howe

duality functor from the quantum group U̇q (slm ) to the category N Webm [3],
whose objects are tuples (k1 , . . . , km ), 0 ≤ ki ≤ N , and whose morphisms are “slN -
webs”. The latter are generated by the morphisms depicted in Figure 14 modulo
the relations of Figure 10.
Note that the category N Webm contains a “tag” morphism satisfying the rela-
tion in Figure 15. This tag corresponds to a junction with the N th antisymmetric
representation. Since this is a trivial representation, the respective Wilson line is
trivial and does not have to be drawn. The junction however still requires a spec-
ification of the junction field, i.e. an invariant tensor Λk ⊗ ΛN −k −→ ΛN ,
which was determined by the ordering of incoming Wilson lines. This ordering is
specified by the direction of the outgoing Wilson line, hence the tag, and the change
of ordering leads to a sign factor (−1)k(N −k) as in the tag relation.
2.4. Why “categorification = surface operators”. In general, a d-dimen-

sional TQFT assigns a number (= partition function) to every closed d-manifold
Md , a vector space H(Md−1 ) to every (d − 1)-manifold, a category C(Md−2 ) to
every (d − 2)-manifold, and so on. Therefore, the process of “categorification” that
promotes each of these gadgets to the higher categorical level has an elegant in-
terpretation as “dimensional oxidation” in TQFT, which promotes a d-dimensional
TQFT to a (d + 1)-dimensional one [29] (see also [20]).
categorification
Figure 16. Categorification means lifting a given theory to one

dimension higher. This operation of adding an extra dimension
turns line operators into surface operators [20] and graphs (net-
works) into foams (surfaces with singular edges).
Many interesting theories come equipped with non-local operators (often called
“defects”) that preserve and frequently ameliorate the essential structure of the
theory. Categorification has a natural extension to theories with such non-local
operators which also gain an extra dimension, much like the theory itself. Prominent
examples of such operators are codimension-2 line operators in 3d Chern-Simons
TQFT that we encountered earlier and that do not spoil the topological invariance
of the theory. One dimension lower, such codimension-2 operators would actually
be local operators supported at points pi on a 2-manifold M2 . And, one dimension
higher, in a 4d TQFT such codimension-2 operators would be supported on surfaces
Σ ⊂ M4 . Therefore, on general grounds a categorification of 3d Chern-Simons

invariants should be a functor (4d TQFT) that assigns
surface Σ ⊂ M4 number Z(M4 ; Σ)
(2.25) knot/web K ⊂ M3 vector space H(M3 ; K)
points p1 , . . . , pn ∈ M2 category C(M2 ; p1 , . . . , pn )
..
.
In particular, as illustrated in Figure 16, categorification of graphs and networks
is achieved by studying surface operators with singular edges (junctions) in four
dimensions. This naturally leads us to the study of junctions of surface operators
and the effective 2d theory on their world-sheet Σ which, respectively, will be the
subjects of sections 3 and 4.
Figure 17. A link cobordism Σ defines a map between the cor-

responding homology groups.
Moreover, functoriality implies that the structure at each level should be com-
patible with cobordisms as well as operations of cutting and gluing. It turns out to
be very rich even before one tackles the very interesting question of studying tangle
cobordisms in non-trivial ambient manifolds. In other words, even when M2 = R2 ,
M3 = R3 and M4 = R4 in (2.25) functoriality can be highly non-trivial due to
interesting topologies of K and Σ. In particular, a surface cobordism Σ between
links K1 and K2 gives rise to a linear map (see Figure 17):
(2.26) Z(Σ) : H(K1 ) −→ H(K2 )
where, to avoid clutter, we tacitly assumed that all components of K1 and K2 have
the same color. (Otherwise, Σ can not be a smooth surface and must be a foam,
i.e. have singular edges.) In fact, all our defects — line operators in 3d, surface
operators in 4d, as well as points in 2d — carry certain labels or colors which we
denote by Ri . For example, making this part of our notation explicit and, on the
other hand, suppressing M2 = R2 , the third line in (2.25) should read:
(2.27) (p1 , R1 ), . . . , (pn , Rn ) category CR1 ,...,Rn
The Hochschild homology of this category must be the cohomology of n copies of
the unknot, colored by R1 , . . . , Rn , respectively:

(2.28) HH ∗ (CR1 ,...,Rn ) = HR1 ⊗ . . . ⊗ H Rn
This is a part of the reverse process, called “decategorification,” which in TQFT

corresponds to dimensional reduction (of the “time” direction). It will prove very
useful later, in section 4.2, where it will help us to better understand the cohomology
of the colored unknot as well as the structure of the category (2.27). Notably,
applying (2.26) to the cobordism between on the one hand the disjoint union K1 =

K of a knot K with the unknot and on the other hand their connected sum
K2 = K# ∼
= K which gives back K, yields a map
(2.29) A ⊗ HR (K) → HR (K) ,
where
(2.30) A := HR ( ).
In fact, for the special case K = this defines an algebra structure
(2.31) : A⊗A → A,
on the homology of the unknot, and (2.29) promotes the homology HR (K) of any
knot K to an A-module.
It turns out that the algebra structure of the unknot homology (2.31) contains
a lot of useful information about how the theory behaves under cobordisms [30–32]
and has been used to construct various deformations of Khovanov-Rozansky homol-
ogy [33–38]. Much of this algebraic structure extends even to colored HOMFLY-PT
homology [39].
A 2d TQFT is basically determined by a Frobenius algebra, defined by a “pair-
of-pants” product as in (2.31). In case the theory is obtained from a supersymmetric
one by topological twisting, than this algebra is the chiral ring of the untwisted
theory. In our present context this implies that the chiral ring of the world-volume
theory of a surface operator labeled by representation R has to agree with the
algebra (2.31) associated to the unknot colored by R,
(2.32) A = HR ( ) = chiral ring of 2d TQFT on the surface operator .
This provides a useful clue for identifying the surface operators (and their world-
volume theories) which are relevant for the categorification of quantum groups.
For example, the original Khovanov homology [40] is based on the nilpotent
Frobenius algebra
(2.33) A = C[X]/X 2
A semi-simple deformation of this algebra studied by Lee [33] led to a number
of important developments, including Rasmussen’s proof of the Milnor conjecture
[41].
3. Junctions of surface operators

Surface operators are non-local operators which, in four-dimensional QFT, are
supported on two-dimensional surfaces embedded in four-dimensional space-time
M4 ,
(3.1) Σ ⊂ M4
Intuitively, surface operators can be thought of as non-dynamical flux tubes (or
vortices) much like Wilson and ’t Hooft line operators can be thought of as static
electric and magnetic sources, respectively. Compared to line operators, however,
surface operators have a number of peculiar features. See [42] for a recent review
on surface operators.
Thus, in gauge theory with gauge group G, line operators are labeled by discrete
parameters, namely electric and magnetic charges of the static source, while surface
operators in general are labeled by both discrete and continuous parameters. The
former are somewhat analogous to discrete labels of line operators, but the latter
are a novel feature of surface operators.
In practice, there are two basic ways of constructing surface operators:
• as a singularity along Σ, e.g.
(3.2) F = 2παδΣ + . . .
• as a coupled 2d-4d system, e.g. described by the action

(3.3) Stot = d x (L4d + δΣ · L2d ) =
4
d x L4d +
4
d2 x L2d
M4 M4 Σ
where the global symmetry G of the 2d Lagrangian L2d is gauged when
coupled to the 4d Lagrangian L4d .
Both descriptions can be very useful for establishing the existence and analyzing
junctions of surface operators, which locally look like a product of “time” direction
Rt with a Y-shaped graph where three (or more) surface operators meet along a
singular edge of the surface Σ, thus providing a physics home and a microscopic
realization of the ideas advocated in [43–45]. A surface Σ with such singular edges
is often called a “foam” or a “seamed surface”.
Before one can start exploring properties of junctions and their applications, it
is important to establish their existence at two basic levels:
a) at the level of “kinematics” as well as
b) at the level of field equations (or BPS equations, if one wants to preserve
some supersymmetry).
The former means that junctions of surface operators must obey certain charge
conservation conditions analogous to (2.4), while the latter means they must be a
solution to field equations or BPS equations, at least classically. Addressing both
of these questions will be the main goal of the present section, where we treat cases
with different amount of supersymmetry in parallel. In particular, sections 3.1–3.4
will be devoted to the latter question, whereas kinematics will be largely the subject
of section 3.5. Before we proceed to technicalities, however, let us give a general
idea of what the answer to each of these questions looks like.
When a configuration of surface operator junctions is static, we have Σ = Rt ×Γ,
where Rt is the “time” direction and Γ is a planar trivalent graph or, more generally,
a trivalent oriented graph in a 3-manifold M3 which, for most of what we need in
this paper, will be simply R3 (or S 3 ). Then, as we shall see, in many cases the
BPS equations on M3 will reduce to (or, at least, contain solutions of) the simple
flatness equations for the gauge field A:
(3.4) FA = 0
For example, the famous instanton equations FA+ = 0 in Donaldson-Witten theory
reduce to (3.4) on a 4-manifold of the form Rt × M3 when one requires solutions to
be invariant under translations along Rt . Since (3.4) is a “universal” part of BPS
equations for many different systems, with different amounts of supersymmetry, it
makes sense to illustrate how the questions of “kinematics” and the “existence of
Figure 18. A small ball around a trivalent junction.
solutions” look in the case of (3.4). By doing so, we also focus our attention on
the most interesting ingredient, namely the gauge dynamics. Then, adding more
fields and interactions to the system is a relatively straightforward exercise and we
comment on it in each case.
As explained in [19], a surface operator with non-zero α in (3.2) can be thought
of as a Dirac string of a magnetic monopole with improperly quantized magnetic
charge α. Magnetic charges of monopoles that obey Dirac quantization condition
take values in the root lattice Λrt of the gauge group G. When this condition is
not obeyed, the world-sheet of a Dirac string becomes visible to physics and this is
precisely what a surface operator is. In this way of describing surface operator —
as a singularity for the gauge field (and, possibly, other fields) — it should be clear
that α in (3.2) takes values in the Lie algebra of the maximal torus of the gauge
group, t = Lie(T), modulo the lattice of magnetic charges Λrt . For example, when
G = U (1) the parameter α is a circle-valued variable.
In this description of a surface operator — as a singularity (3.2) or, equivalently,
as a Dirac string of an improperly quantized magnetic charge — the orientation of
the edges of the graph Γ has a natural interpretation: it represents the magnetic
flux. Of course, one can change the orientation of each edge and simultaneously
invert the holonomy U = exp(2πiα) ∈ G of the gauge field around it, without
affecting the physics:
U U −1
(3.5) ←−− = −−−→
Moreover, if the gauge field satisfies (3.4) away from the singularity locus Σ = Rt ×Γ,
then the magnetic flux must be conserved at every junction. In the abelian theory
with G = U (1) it simply means that the signed sum (signs determined by the
orientation) of the parameters α for all incoming and outgoing edges is equal to
zero at every vertex of Γ, e.g.
(3.6) α = α + α (mod 1)
for a basic trivalent junction as depicted in Figure 18. A non-abelian version of flux
conservation is more involved and will be discussed in section 3.5. This preliminary
discussion, however, should give a reasonably good idea of what a junctions looks
like in the description of a surface operator as a singularity (or, ramification) along
Σ.
N)
k1 ,
r(
G
Gr(k,N) Flag
Gr
(k 2
,N
)
Figure 19. The core of a junction between surface operators
(3.8) carries a partial flag variety F l(k1 , N ), k = k1 + k2 . Shown
here is the (x1 , x2 ) plane.
Now, let us have a similar preliminary look at the same question from the
viewpoint of a 2d-4d coupled system à la (3.3). In this approach, each face of
the seamed surface Σ is decorated with a 2d theory that lives on it, in many
examples simply a 2d sigma-model with a target space C that enjoys an action
of the symmetry group G, e.g. a conjugacy class3 or a flag variety C = G/L.
In the approach based on 2d-4d coupled system, a junction of surface operators
(or, a more general codimension-1 defect along Σ) determines a “correspondence”
M ⊂ C × C × C :
M
(3.7) " #
C C × C
where C and C are target spaces of 2d sigma-models on “incoming” edges of Γ that
join into an edge that carries sigma-model with target space C. The description of
such interfaces (or line defects) in the dual4 Landau-Ginzburg model will be the
main subject of section 4.
Surface operators that will be relevant to skew Howe duality and categorifica-
tion of quantum groups are labeled by “Levi types” L = S(U (k) × U (N − k)) in
a theory with gauge group G = SU (N ), or simply by k ∈ {1, . . . , N − 1}. In the
2d-4d description of such “k-colored” surface operators the target space C = G/L
is the Grassmannian of k-planes in CN ,
(3.8) C = Gr(k, N ) = SU (N )/S(U (k) × U (N − k))
As we proceed, we will encounter Grassmannian varieties5 more and more often, as
moduli spaces of solutions in gauge theory (later in this section) as well as in the
description of interfaces and chiral rings in Landau-Ginzburg models (in section 4).
3 Notice, this is similar to labeling of surface operators defined as singularities, where conju-
gacy classes characterize possible values of the holonomies U ∈ C.

4 in the sense of LG / sigma-model duality [46, 47]
5 In case of a more general irreducible representation of highest weight λ, the space Gr(k, N )
λ
would be replaced by a finite-dimensional subspace Gr of the affine Grassmannian that plays an
important role in the geometric Satake correspondence. Since such spaces are generally singular,
we leave detailed study of the corresponding surface operators to future work and focus here on
the ones relevant to skew Howe duality and categorification of quantum groups.
What about junctions of surface operators that carry Grassmannian sigma-

model on their world-sheet? They can be conveniently described as correspondences
(3.7), where M is a partial flag variety F l(k1 , k, N ) of k1 -planes in k = (k1 + k2 )-
planes in CN , as illustrated in Figure 19:
Gr(k, N ) ←− F l(k1 , k, N ) −→ Gr(k1 , N ) × Gr(k2 , N )
(3.9)
V →− (V1 ⊂ V ) −→ (V1 , V1⊥ ⊂ V )
Now, once we had a preliminary discussion of how the decorations of surface
operators flow across junctions, both from the singularity perspective and in the
approach of 2d-4d coupled system, it is natural to tackle the question of their
existence at the level of field equations. As we already noted earlier, it involves
solving PDEs in three dimensions with prescribed boundary conditions around
Γ ⊂ M3 , which in general is not an easy task, even for the basic version of the
equations (3.4). While it would be extremely interesting to pursue the construction
of explicit solutions, we will only need to know whether they exist and what their
moduli space looks like. Luckily, this latter question can be addressed without
constructing explicit solutions and an illuminating way to do that is via embedding
the gauge theory into string / M-theory.
Many interesting SUSY field theories on M4 can be realized on stacks of M-
theory fivebranes, such as configurations of
(3.10) N M5-branes on M4 × C
Indeed, for different choices of 2-manifolds C and their embeddings in the eleven-
dimensional space-time one can realize 4d field theories on M4 with N = 1, N = 2,
or N = 4 supersymmetry. Below we consider each of these choices in turn.
In such constructions, one can “engineer” surface operators by introducing
additional M5-branes or M2-branes which share only the two-dimensional part of
the 4d geometry, Σ ⊂ M4 , with the original set of fivebranes (3.10). In particular,
in such brane constructions, the junctions of surface operators relevant to skew
Howe duality and categorification of quantum groups have a simple interpretation
where one stack of k branes splits up into two stacks of k1 and k2 = k − k1 branes,
as illustrated in Figure 20.
Figure 20. Brane realization of the junction of surface operators

in Figure 19.
3.1. Junctions in 4d N = 4 theory. When C = T 2 or R2 , the fivebrane

configuration (3.10) preserves maximal (N = 4) supersymmetry in 4d space-time
M4 . Then, depending on the geometry of Σ and M4 in (3.1), one finds different
fractions of unbroken supersymmetry.
If both M4 and Σ are flat, i.e. M4 = R4 and Σ = R2 , then surface operators
can be half-BPS and admit the following brane construction
0 1 2 3 4 5 6 7 8 9 10
(3.11) M5 × × × × × ×
M5 × × × × × ×
where, following conventions of [48], we assume that M4 is parametrized by (x0 , x1 ,
x2 , x3 ) and C is parametrized by x6 and x10 .
The half-BPS surface operators in 4d N = 4 theory can form 14 -BPS junctions
of the form Σ = Rt × Γ, where Rt is the x0 “time” direction and Γ is an arbitrary
trivalent graph in the (x1 , x2 )-plane. In fact, without breaking supersymmetry
further, we can take the 4d space-time to be M4 = Rt × M3 , with an arbitrary
3-manifold M3 locally parametrized by (x1 , x2 , x3 ), and a knotted trivalent graph
Γ ⊂ M3 . This requires partial topological twist of the 4d N = 4 theory along
M3 , which in the brane construction (3.11) is realized by replacing the space R3
parametrized by (x7 , x8 , x9 ) with the cotangent bundle to M3 .
Then, M5-branes are supported on R3 × M3 , while M5 -branes are supported
on R3 × LΓ , where M3 and LΓ are special Lagrangian submanifolds in a local
Calabi-Yau space T ∗ M3 , such that
(3.12) LΓ ∩ M3 = Γ
This configuration of fivebranes preserves 4 real supercharges, i.e. a quarter of the
original N = 4 SUSY.
When C = T 2 or R2 , we can consider a further generalization in which junctions
of surface operators are not necessarily static. Such configurations preserve only 2
real supercharges — namely, N = (0, 2) supersymmetry along C — and recently
have been studied [39, 49] in a closely related context. In particular, one can
combine the time direction with M3 into a general 4-manifold M4 and take Σ to
be an arbitrary surface (3.1) with singular trivalent edges, i.e. a foam or a seamed
surface [44, 45].
For generic Σ, such foams or non-static junctions of surface operators will break
supersymmetry unless we extend the partial topological twist along M3 to all of
M4 . In the fivebrane system (3.11) this twist is realized by replacing (x7 , x8 , x9 )-
directions with a non-trivial bundle over M4 , namely the bundle of self-dual 2-forms:
space-time: Λ2+ (M4 ) × R4
(3.13) N M5-branes: M4 × R2
k M5 -branes: LΣ × R 2
where, much as in (3.12), LΣ and M4 are coassociative submanifolds in a local
G2 -holonomy manifold Λ2+ (M4 ), such that
(3.14) LΣ ∩ M4 = Σ
As we already stated earlier, this generic configuration of surface operators preserves
only 2 real supercharges (which, moreover, are chiral from the 2d perspective of x6
and x10 ).
line−changing
operator
x0
x1
x2
Figure 21. 6d fivebrane theory on a 2-manifold M2 parametrized

by x1 and x2 yields 4d N = 2 theory T [M2 ] in the remaining
four dimensions. In this theory, a network of line defects on M2
determines a BPS line operator, whereas a “time evolution” of such
network corresponds to segments of different line operators joined
together at those values of “time” x0 where the topology of network
changes.
3.2. Line-changing operators in class S and network cobordisms. The

intersection of M5-branes with M5 -branes produces the so-called codimension-2
defect in the 6d (0, 2) theory on the fivebrane world-volume (3.10). The latter also
admits codimension-4 defects which too can be used to produce surface operators
in 4d theory on M4 and which in M-theory are realized by M2-branes ending on
M5-branes:
0 1 2 3 4 5 6 7 8 9 10
(3.15) M5 × × × × × ×
M2 × × ×
where the M2-brane world-volume is Σ × R+ , with R+ = {x7 ≥ 0}. In this

realization of surface operators, BPS junctions can be obtained by considering
codimension-4 defect supported on Σ = Rt × Γ. In fact, let M4 = Rt × M2 × R,
where the 2-manifold M2 is a Riemann surface (possibly with punctures); this re-
quires a partial topological twist of the fivebrane theory which, similar to (3.13), is
realized by embedding M2 in the space-time T ∗ M2 . Also, let Γ ⊂ M2 be a colored
trivalent graph on M2 without self-intersections:
6d (0,2) theory: Rt × M2 × R × C
(3.16) ∪
surface operator: Rt × Γ
If we apply this to C = R2 (or C = T 2 ), as in section 3.1, it is natural to ask

what this configuration looks like from the viewpoint of Rt × R × C ∼ = Rt × R3 ,
after compactification on the Riemann surface M2 . Due to partial topological twist
along M2 , the resulting 4d theory T [M2 ] has only N = 2 supersymmetry and the
codimension-4 defect (3.16) produces a line operator in this theory labeled by a
colored trivalent network Γ ⊂ M2 .
This is precisely the configuration recently used in [50–54] to study line op-
erators in 4d N = 2 theories T [M2 ]. In our application here, we simply wish to
interpret the Riemann surface M2 as part of the 4d space-time. Then, the same
configuration (3.16) describes “static” surface operators in 4d N = 4 theory on
M4 = Rt × M2 × R. Non-static surface operators correspond to networks Γ(t)
which vary with t ∈ Rt ; from the viewpoint of 4d N = 2 theory T [M2 ] they cor-
respond to segments of different line operators glued together in a single line via
local operators supported at those points ti ∈ Rt where the network Γ(t) changes
topology. (Recall, that due to partial topological twist along M2 only topology of
the network Γ matters.)
In other words, basic topological moves on Γ realized via cobordisms correspond
to line-changing operators in 4d N = 2 theory T [M2 ]:
(3.17) network cobordism ⇔ line-changing operator
Turning on Omega-background in 4d N = 2 theory T [M2 ] corresponds to replacing
C = R2 with C = R2 . This is precisely the setup we will use in application to knot
homologies and categorification of quantum groups, cf. (4.2).
3.3. Junctions in 4d N = 2 theory. When C is an arbitrary Riemann sur-

face (possibly with boundaries and punctures) of genus g = 1, in order for the
fivebrane configuration (3.10) to preserve supersymmetry its world-volume theory
must be (partially) twisted along C. For the embedding of the fivebrane world-
volume in the ambient space-time the partial topological twist means that C must
be either a holomorphic Lagrangian submanifold in a Calabi-Yau 2-fold (that lo-
cally, near C always looks like T ∗ C) or a holomorphic curve in a Calabi-Yau 3-fold.
The first choice preserves N = 2 supersymmetry in the 4d theory on M4 , while the
second option preserves only N = 1 SUSY and will be considered next.
Consistent with (3.11), our choice of coordinates is
M4 T ∗C R3
(3.18)
x , x , x2 , x3
0 1
x , x , x6 , x10
4 5
x , x8 , x9
7
In particular, the U (1)r × SU (2)R R-symmetry group in this setup is identified

with the U (1)45 × SU (2)789 symmetry of the eleven-dimensional geometry. When
both M4 = R4 and Σ = R2 are flat, we can still use brane configuration (3.11) to
describe half-BPS surface operators, this time in 4d N = 2 theory on M4 .
Also, as in the N = 4 case, we can describe junctions by taking M4 = Rt × M3
and replacing (x7 , x8 , x9 ) with the cotangent bundle to M3 :
space-time: Rt × T ∗ M3 × T ∗ C
(3.19) N M5-branes: Rt × M3 × C
k M5 -branes: R t × LΓ × C
where, as in (3.12), LΓ and M3 are special Lagrangian submanifolds in T ∗ M3
which intersect over a knotted trivalent graph Γ ⊂ M3 . Then, the resulting brane
configuration (3.19) describes junctions of half-BPS surface operators supported on

a surface Σ = Rt × Γ.
Due to the partial topological twist along M3 , the latter can be an arbitrary
3-manifold and Γ can be an arbitrary knotted graph. The price to pay for it is
that junctions (3.19) preserve only 2 real supercharges, i.e. again are 14 -BPS in the
original 4d theory on M4 .
3.4. Junctions in 4d N = 1 theory. Even though 4d N = 1 theories are
not directly related to the main subject of our paper, here for completeness we
consider possible junctions of surface operators in such theories.
First, in a typical brane construction [55] of 4d N = 1 theories, C is embedded
as a holomorphic curve in a local Calabi-Yau 3-fold:
(3.20) M4 × CY3 × R
This leaves little room for surface operators. In fact, even when M4 = R4 this
system admits a half-BPS surface operator that breaks supersymmetry down to
N = (0, 2) along Σ = R2 .
In the fivebrane construction (3.10) of the 4d N = 1 theory, this half-BPS
surface operator can be realized by an additional system of M5 -branes supported
on a holomorphic 4-cycle D ⊂ CY3 :
space-time: M4 × CY3 × R
(3.21) N M5-branes: M4 × C
M5 -branes: Σ × D
Equivalently, via a “brane creation” phase transition [56], one can represent half-
BPS surface operators in 4d N = 1 theory by M2-branes with a semi-infinite extent
in the x7 -direction (x7 ≥ 0):
space-time: M4 × CY3 × R
(3.22) N M5-branes: M4 × C
M2-branes: Σ × {pt} × R+
It is easy to verify that N = 1 supersymmetry on M4 is not sufficient to allow for
non-trivial BPS junctions of surface operators.
3.5. OPE of surface operators and the Horn problem. In non-abelian
theory, the analogue of the flux conservation (3.6) is known as the multiplicative
Horn problem:
(3.23) What conjugacy classes are contained in C · C ?
Recall, that when a surface operator is described as a singularity (ramification) for
the gauge field, it is naturally labeled by the conjugacy class C ⊂ G where the holo-
nomy U — defined only modulo gauge transformations, i.e. modulo conjugaction
— takes values.
If away from Σ = Rt × Γ, that is away from Γ ⊂ M3 , the gauge field satisfies
(3.4), then the product of holonomies around edges of each vertex in the graph Γ
should be trivial. This is a non-abelian version of the flux conservation (3.6) carried
by surface operators, which for a basic trivalent junction as in Figure 18 takes the
form
(3.24) U = U · U .
The same condition describes junctions of non-abelian vortex strings. When written
in terms of the respective conjugacy classes
(3.25) C ⊂ C · C
it indeed turns into the multiplicative Horn problem (3.23).
By reversing the orientation of the surface operator characterized by holonomy
U using (3.5), we can equivalently describe the more symmetric situation, in which
all surface operators are outgoing. This replaces U → U −1 in (3.24) and leads to
(3.26) 1 ∈ C · C · C
where 1 ∈ G is the identity.
In gauge theory, (3.23) and (3.26) can be interpreted as operator product ex-
pansion (OPE) of surface operators. Then, depending on the context, the “OPE

coefficient” is either the moduli space MG 0,3 (C, C , C ) of flat G-bundles on the punc-
tures 2-sphere S 2 \ {p, p , p } with holonomies around the punctures lying in C, C ,
and C , or its suitable cohomology or K-theory. Indeed, this moduli space describes
the space of gauge fields which satisfy (3.4) in the vicinity of a trivalent junction,
which we can surround with a small ball, as illustrated in Figure 18. The bound-
ary of such small ball is a trinion S 2 \ {p, p , p }. Since its fundamental group is
generated by loops around the punctures, with a single relation abc = 1,
(3.27) π1 (S 2 \ {p, p , p }) = a, b, c/abc = 1 ,
the moduli space of its representations into G is precisely the set of triples (3.24),
and each such triple determines a flat G-bundle on the trinion:

(3.28) 0,3 (C, C , C ) = {(U, U , U ) ∈ C × C × C | U · U · U = 1}/G
MG
In particular, this moduli space is non-empty if and only if (3.26) holds,

(3.29) 0,3 (C, C , C ) = ∅
MG ⇔ C · C · C & 1
This moduli space is a complex projective variety and is also symplectic. For
instance, when G = SU (2) and the conjugacy classes satisfy (3.26) we have6
(C, C , C ) = CP3 //U (1)3 = point .
SU(2)
(3.30) M0,3
Note, that the problem considered here has an obvious analogue for a 2-sphere
with an arbitrary number of punctures; it characterizes junctions where more than
three surface operators meet along a singular edge of Σ. Since this generalization
is rather straightforward, we shall focus on trivalent junctions, which are relevant
in the context of the skew Howe duality.
Our next goal is to describe the selection rules of the “OPE” of surface oper-
ators, i.e. to find solutions to the multiplicative Horn problem (3.23) or its more
symmetric form (3.29).
To do so, it is useful to parametrize conjugacy classes by the logarithm α of
the eigenvalues of U = exp(2πiα), that take values in the fundamental alcove of G,
(3.31) α ∈ U.
For example, for G = SU (N ) parameters α take values in the simplex

(3.32) U = {α1 ≥ α2 ≥ . . . ≥ αN ≥ α1 − 1| αi = 0} .
i
6 The “complex case” of G = SL(2, C) is even easier since eigenvalues of the holonomies
C
obey algebraic relations, analogous to the familiar A-polynomials for knots, see e.g. [20].
Since in a theory with gauge group G surface operators are parametrized by

points in the Weyl alcove (3.31) and, possibly, other data, it is natural to study the
image of (3.29) in U3 , which defines a convex polytope of maximal dimension:
(3.33) ΔG := {(α, α , α ) ∈ U3 : C · C · C & 1}
In other words, the polyhedron ΔG ⊂ U3 describes the set of triples of conjugacy
classes such that the moduli space (3.29) is non-empty. For instance, for G = SU (2),
U3 = [0, 12 ]3 is a cube and ΔG is a regular tetrahedron inscribed in it [57], cf. (3.6):
(3.34) |α − α | ≤ α ≤ min{α + α , 1 − α − α }
This tetrahedron is precisely the image of the moment map under U (1)3 action on
the “master space” CP3 in (3.30).
More generally, for G = SU (N ) the facets of ΔG are defined by linear inequal-
ities [58, 59]:

(3.35) αi + αj + αk ≤ d
i∈I j∈J k∈K
for each d ≥ 0, 1 ≤ r ≤ N and all triples I, J, K of r-element subsets of {1, . . . , N },

such that the degree-d Gromov-Witten invariant of the Grassmannian Gr(r, N )
satisfies
(3.36) GWd (σI , σJ , σK ) = 1 ,
where the Schubert cycles σI are the cohomology classes associated to the Schubert
subvarieties
(3.37) XJ = {W ∈ Gr(r, N ) | dim(W ∩ Fij ) ≥ j, j = 1, . . . , r}
for I = {i1 , . . . , ir }, which are defined with respect to a complete flag
(3.38) F0 = {0} ⊂ F1 ⊂ . . . ⊂ FN −1 ⊂ FN = CN .
The σI form a basis of H ∗ (Gr(r, N ), Z) and will be the subject of section 3.6.
A similar list of inequalities that defines ΔG for any compact group G was
explicitly described by Teleman and Woodward [60] and then further optimized in
[61, 62]. The inequalities of Teleman and Woodward also have an elegant inter-
pretation in terms of the small quantum cohomology, except that instead of the
Grassmannian they involve the flag variety F l = GC /P, where GC is the complex-
ification of G and P is a maximal parabolic subgroup.
For example, conjugacy classes of order-2 elements U 2 = (U )2 = (U )2 = 1 ∈
SU (N ) such that U · U · U = 1 have the moduli space
(α, α , α ) = point
SU(N )
(3.39) M0,3
and therefore belong to the boundary of the polyhedron ΔG , (α, α , α ) ∈ ∂ΔG .
The dimensions i, j, k of their −1 eigenspaces satisfy the Clebsch-Gordan rules:
|i − j| ≤ k ≤ i + j, etc.
Multiplying such order-2 conjugacy classes by ζ k , where k is the dimension
of (−1)-eigenspace and ζ = exp(iπ/N ) is a primitive N th root of −1, we obtain
conjugacy classes that correspond to R = Λk (CN ) and play an important role in
MOY invariants of colored trivalent graphs and in the skew Howe duality:
(3.40) Uk = diag( ζ k , . . . , ζ k , −ζ k , . . . , −ζ k ) .
!" # !" #
N −k k
Namely, as in section 2, let Γ be a planar oriented trivalent graph, whose edges are
colored by elements in {1, 2, . . . , N − 1} with the signed sum (signs given by the
orientation) of colorings around each vertex equal to 0.
With every such graph we can associate a configuration of surface operators
supported on Σ = Rt × Γ, such that an edge colored by k is represented by a surface
operator with holonomy in the conjugacy class of (3.40):
(3.41) color k ⇔ representation Λk (CN ) ⇔ holonomy Uk .

Note, this way of associating a particular type of surface operator to a colored edge
of Γ is consistent with the BPS equation (3.4) which, in turn, leads to flux conser-
vation condition (3.26). In particular, one can verify (3.35) using the parameters
(3.31) for the conjugacy class of (3.40):
k k k−N k−N
(3.42) α = ,..., , ,..., .
2N !" 2N# 2N !" 2N #
N −k k
In this case, the moduli space of solutions to the BPS equations with a “foam” of
surface operators on Σ = Rt × Γ is the representation variety of the fundamental
group of the graph complement into SU (N ) with meridional conditions on the edges
of the graph

(3.43) M(Γ) = Rep π1 (S 3 \ Γ); SU (N ) .
Lobb and Zentner [63] (see also [64]) point out that the Poincaré polynomial of this
space is closely related to the MOY polynomial PN (Γ) ∈ Z[q, q −1 ] of the colored
graph Γ. In fact,
(3.44) χ (M(Γ)) = PN (Γ)|q=1 .
Moreover, the authors of [63, 64] provide a simple model for M(Γ) by decorating
each edge colored with k by a point in the Grassmannian Gr(k, N ), with the condi-
tion that at every vertex with the three edges colored by k1 , k2 , and k = k1 + k2 the
corresponding decorations consist of two orthogonal k1 - and k2 -planes in CN and
the k-plane that they span. Such decorations are called admissible and the space
of all admissible decorations of the colored graph Γ is homeomorphic to M(Γ). For
example, the moduli space for the k-colored unknot is

(3.45) M k ∼ = Gr(k, N ) ,
in agreement with the fact that its cohomology indeed gives the colored homology
of the unknot (2.30) for R = Λk (CN ):
(3.46) A = HR ( ) ∼
= H ∗ (Gr(k, N )) .
Also note that the moduli space associated to the trivalent junction with the edges
colored by k1 , k2 , and k = k1 + k2 is precisely the partial flag variety that appears
in the correspondence (3.9). In section 4 we describe the corresponding interface
in the product of Landau-Ginzburg models with chiral rings A = H ∗ (Gr(k, N )),
which provide an equivalent (and more user-friendly) description of the topologically
twisted theory. But before that, in the next section, we will briefly comment on how
Schubert calculus can be employed to approach the “OPE of surface operators”.
3.6. OPE and Schubert calculus. Note, that in degree d = 0, the condition
(3.36) in the description (3.35) of the OPE selection rules corresponds to the cup
product in the classical cohomology ring of the Grassmannian Gr(r, N )

(3.47) σ λ · σμ = cνλ,μ σν ,
ν
where the structure constants cνλ,μ

are the Littlewood-Richardson coefficients. Here,
our goal will be to explain (3.47) and its “quantum deformation”
ν,d
(3.48) σλ σμ = cλ,μ q d · σν
d≥0 ν
that conveniently packages all Gromov-Witten invariants that appear in (3.35), see
e.g. [65]. A careful reader will notice that, compared to (3.35), we have labeled the
cohomology classes differently here. Instead of Schubert symbols J = {j1 , . . . , jr }
(which are sequences 1 ≤ j1 < j2 < . . . jr ≤ N ), we used partitions λ: Indeed, to
each Schubert symbol J, we can associate a partition
(3.49) λJ = (λ1 ≥ . . . ≥ λr ≥ 0) , λi = N − r − ji + i ,
whose associated Young diagram fits into a rectangle of size r × (N − r), i.e.
λ1 ≤ N − r. Note, that such Young diagrams are in one-to-one correspondence

with integrable highest weight representations of su(N − r + 1)r (equivalently, of

u(r)N −r+1 ). Moreover, there is a partial ordering of the labels: I ≤ J if in ≤ jn
for all n, which expressed in partitions becomes λI ⊆ λJ .
To explain this more slowly, let us recall a few basic facts about the geometry
of Gr(r, N ), the space of r-dimensional linear subspaces V of a fixed N -dimensional
complex vector space E. Thinking of these subspaces as spans of row vectors, to
each V ∈ Gr(r, N ) we can associate a matrix M ∈ GLr \Mat∗r,N , which leads to the
Plücker embedding
N
(3.50) Gr(r, N ) ⊂ CP( r )−1
and defines Gr(r, N ) as an algebraic variety, cut out by homogeneous polynomial
equations. For example, the Grassmannian Gr(2, 4), which we will use for illustra-
tions, is defined by a single equation in CP5 :
(3.51) X12 X34 − X13 X24 + X14 X23 = 0
We are interested in the cohomology ring of Gr(r, N ) and its quantum

defor-
mation (3.48). The classical cohomology H ∗ (Gr(r, N )) has total rank Nr and is
non-trivial only in even degrees ranging from zero to the dimension of the Grass-
mannian,
(3.52) dimR Gr(r, N ) = 2r(N − r)
In fact, the Grassmannian has a decomposition into a disjoint union
$
(3.53) Gr(r, N ) = SJ
J
of Schubert cells, which gives it the structure of a CW complex. The Schubert cells
can be defined with respect to a any fixed complete flag E1 ⊂ E2 ⊂ . . . ⊂ EN −1 ⊂
EN in E as follows:
(3.54) SJ = {V ∈ Gr(r, N ) | dim(V ∩ Ei ) = #{a|ja ≤ i} for all 1 ≤ i ≤ N }.
Their complex dimensions can be expressed as

(3.55) dim SJ = jn − n = |λJ |
n
where |λ| = λ1 + . . . + λr . Thus, our favorite example Gr(2, 4) has six Schubert
cells labeled by Schubert symbols J = (1, 2), (1, 3), (1, 4), (2, 3), (2, 4), and (3, 4),
of complex dimensions 0, 1, 2, 2, 3, and 4, respectively.
The closure S J of a Schubert cell is a union of Schubert cells, e.g. S 13 =
S13 ∪ S12 in Gr(2, 4). This inclusion of Schubert cells defines the so-called Bruhat
order which indeed is compatible with the partial order on the Schubert symbols
mentioned above:
(3.56) I≤J ⇔ SI ⊆ S J .
For Gr(2, 4) this partial order can be conveniently described by the following dia-
gram
(3, 4)
(2, 4)
(3.57) (2, 3) (1, 4)
(1, 3)
(1, 2)
A similar partial order on the set of conjugacy classes defined by closure plays an
important role in the gauge theory approach to the ramified case of the geometric
Langlands correspondence [21] and the geometric construction of Harish-Chandra
modules [22].
The Schubert variety XJ is the Zariski closure of SJ :
$
(3.58) XJ := SI
I≤J
Clearly, dim XJ = |λJ |; and the Schubert cycles σλ ∈ H 2|λ| (Gr(r, N )) form an
integral basis for the cohomology ring of the Grassmannian,

(3.59) H ∗ (Gr(r, N ); Z) = Z · σλ
λ⊆(N −r)t
For example, the Poincaré polynomial of Gr(2, 4) is (cf. (3.57))

(3.60) P (Gr(2, 4)) = 1 + t2 + 2t4 + t6 + t8
As alluded to above, in the “classical” Schubert calculus (3.47), the non-
negative integers cνλ,μ are the Littlewood-Richardson coefficients; they vanish unless
|λ| + |μ| = |ν|. Note that, σ(0) ≡ σ(0,...,0) associated to the partition λ = (0, . . . , 0)
is the unit in the cohomology ring of the Grassmannian. The classical cohomology
ring has two important generalizations (deformations), one of which we already
mentioned earlier. Namely, much like its classical counterpart, the (small) quan-
tum cohomology of the Grassmannian Gr(r, N ) has a Z[q]-basis formed by Schubert
classes {σλ } labeled by partitions λ = (λ1 , . . . , λr ) that fit into a rectangle of size
r × (N − r), i.e. N − r ≥ λ1 ≥ λ2 ≥ . . . ≥ λr ≥ 0. The structure constants cν,d λ,μ of
the quantum multiplication (3.48) are often called quantum Littlewood-Richardson
coefficients. They vanish, cν,d λ,μ = 0, unless |λ| + |μ| = |ν| + N d, which means that q
carries homological degree 2N .
For example, the quantum cohomology ring QH ∗ (Gr(2, 4)) can be described
by the following relations:
σ(1,0) σ(1,0) = σ(2,0) + σ(1,1) σ(1,0) σ(2,0) = σ(2,1)
σ(1,0) σ(1,1) = σ(2,1) σ(1,0) σ(2,1) = σ(2,2) + q
σ(1,0) σ(2,2) = qσ(1,0) σ(2,0) σ(2,0) = σ(2,2)
σ(2,0) σ(1,1) =q σ(2,0) σ(2,1) = qσ(1,0)
(3.61)
σ(2,0) σ(2,2) = qσ(1,1) σ(1,1) σ(1,1) = σ(2,2)
σ(1,1) σ(2,1) = qσ(1,0) σ(1,1) σ(2,2) = qσ(2,0)
σ(2,1) σ(2,1) = qσ(2,0) + qσ(1,1) σ(2,1) σ(2,2) = qσ(2,1)
σ(2,2) σ(2,2) = q2
Setting q = 0 we obtain the ordinary cohomology ring (3.47), while specializing to
q = 1 we get the Verlinde algebra [66] (namely, the algebra of Wilson loops in U (r)
Chern-Simons theory at level N − r in our conventions). A categorification of this
algebra was recently constructed in [67].
The second important deformation of the classical Schubert calculus (3.47) is
based on the fact that Gr(r, N ) admits a torus action of T = U (1)N . (Note that
since the diagonal U (1) acts trivially, we effectively have an action of U (1)N −1 .)
There is a fixed point for each subset J = {j1 < . . . < jr } ⊂ {1, . . . , N }, so that

∗ T ∗ N
(3.62) dim H (Gr(r, N ) ) = dim H (Gr(r, N )) = .
r
This is the general property of spaces with vanishing odd cohomology: all such
spaces are equivariantly formal. In general, the T -equivariant cohomology of such a
space X is conveniently described by the moment graph (a.k.a. the Johnson graph),
whose vertices are in bijection with X T , edges correspond to one-dimensional orbits,
etc. For example, the moment graph of the Grassmannian Gr(2, 4) is illustrated in
Figure 22, whereas for r = 1 and N = 4 we get precisely the regular tetrahedron
ΔSU(2) obtained in (3.34) as the image of the moment map under the U (1)3 action
on CP3 . In terms of the moment graph, the T -equivariant cohomology HT∗ (X) has
a simple explicit description as the ring
χE divides fx − fy whenever
(3.63) (fx ) ∈ HT∗ (pt) s.t.
x∈vertices
x and y lie in a common edge E
where

N
(3.64) HT∗ (pt) = C[mi ] = C[m1 , . . . , mN ] ,
i=1
and χE = mi − mj if the vertices of the edge E correspond to Schubert symbols
I = (I ∩ J) ∪ {i} and J = (I ∩ J) ∪ {j}.
The quantum and equivariant deformations of the cohomology ring (3.47)
can be combined into the T -equivariant quantum cohomology of Gr(r, N ) which
has multiplication of the form (3.48), where the equivariant quantum Littlewood-
Richardson coefficients cν,dλ,μ are homogeneous polynomials in the ring (3.64) of de-
gree |λ| + |μ| − |ν| − N d. The explicit form of these polynomials for Gr(2, 4) can
be found in [68, sec. 8]. In particular, cν,d
λ,μ is equal to the quantum Littlewood-
Richardson coefficient when |λ| + |μ| = |ν| + N d, while for d = 0 it is equal to the
ordinary equivariant Littlewood-Richardson coefficient cνλ,μ . The remarkable fact
about the equivariant quantum Littlewood-Richardson coefficients is that they are
<12> = 1100 =
13
23
<23> = 0110 = <13> = 1010 =
14
24
12
34 34
12
<24> = 0101 = <14> = 1001 =
14
24
13
23
<34> = 0011 =
Figure 22. The Schubert cells of the Grassmannian (as well as

vacua of 4d N = 2 SQCD) can be labeled by Young diagrams that
fit into r × (N − r) rectangle, or, equivalently, by a string that
contains r ones and N − r zeros, or, yet another way, by length-r
subsets I ⊂ {1, . . . , N }. Shown here is the example with r = 2 and
N = 4.
homogeneous polynomials in variables m1 − m2 , . . ., mN −1 − mN with positive

coefficients. Hence, they can be categorified!
Domain walls in 4d N = 2 SQCD. The geometry of Schubert varieties
described here has a simple interpretation in terms of moduli spaces of domain
walls in 4d N = 2 super-QCD with U (r) gauge group and Nf = N flavors in
the fundamental representation of the gauge group, see e.g. [69–71]. Since this
physical manifestation is somewhat tangential to the subject of the present paper
we relegate it to Appendix B.
4. Categorification and the Landau-Ginzburg perspective

As we already explained in section 2.4, surface operators supported on knot
and graph cobordisms provide a categorification of the corresponding knot/graph
invariants, thus making the results of section 3 directly relevant. Moreover, the
choice of the Riemann surface C in section 3 determines the type of knot homology
that one finds. Thus, a compact Riemann surface C typically leads to a singly-
graded homology theory (see [20] for concrete examples). In this paper, we are
mostly interested in applications to doubly-graded knot homologies categorifying
quantum group invariants; in the language of section 3, they correspond to a some-
what peculiar choice C ∼= R2 (= “cigar” in the Taub-NUT space) that we already
encountered in section 3.2 and will describe in more detail below.
In particular, we describe junctions of surface operators from the perspective

of the 2d TQFT on Σ. We restrict the discussion to surface operators associated
with antisymmetric tensor products of the fundamental representation of slN . Gen-
eralization of the discussion below to slN |M and to symmetric representations of
slN should provide categorification of the super q-Howe duality [72] and symmetric
q-Howe duality [73], respectively. We leave these generalizations to future work.
4.1. Physics perspectives on categorification. Soon after the advent of

the first homological knot invariants [40, 74, 75] it was proposed [20, 76] that, in
general, knot homology should admit a physical interpretation as a Q-cohomology
of a suitable physical system,
(4.1) knot homology = Q-cohomology ≡ HBPS
where Q is a supercharge. Since then, many concrete realizations of this general
scenario have been proposed for different knot homologies.
In most of the physics approaches to knot homologies, the starting point is one
of the two closely related fivebrane systems:
(4.2)
doubly-graded
triply-graded
space-time: Rt × T ∗ S 3 × T N4 ← −−
phase
−− − −−
→ Rt × CY3 × T N4
N M5-branes: Rt × S 3 × R2 transition
Rt × LΓ × R2
k M5 -branes: Rt × LΓ × R 2
which is precisely a variant of (3.19), with M3 = S 3 and C = R2 . (Other variants,

with C = S 2 and C = T 2 , have been studied in [20].) Keeping track of the
U (1) × U (1) quantum numbers associated with the rotation symmetry of C = R2
and its normal bundle in the Taub-NUT space where it is naturally embedded,
T N4 ∼= T ∗ C, leads to two gradings, namely the q-grading and the homological
t-grading.
Much like the familiar phases of water, the two fivebrane configurations in
(4.2) are conjecturally related by a phase transition [77, 78] and describe the same
physical system in different regimes of parameters, one of which has the fixed rank
N and hence is more suited for doubly-graded knot homologies, while the other has
an additional, third grading on the space of refined BPS states (= Q-cohomology)
and provides a physical realization of colored HOMFLY-PT homologies. Studying
the two fivebrane systems (4.2) from different vantage points led to different physical
perspectives on doubly-graded and triply-graded knot homologies:
• Enumerative invariants: Looking at the system (4.2) from the vantage
point of the Calabi-Yau space CY3 leads to the original physical descrip-
tion [76] in terms of enumerative invariants.
• Refined Chern-Simons: Looking from the vantage point of M3 = S 3
on the doubly-graded side of (4.2) leads to slN double affine Hecke algebra
(DAHA) and refined Chern-Simons theory [79–83].
• 5d gauge theory: Also on the doubly-graded side, analyzing the system
(4.2) from the vantage point of the world-volume theory on N M5-branes
leads to a formulation of knot homology via counting solutions to the
Haydys-Witten equations in five dimensions [84, 85].
• Landau-Ginzburg model is a two-dimensional theory that “lives” on
(time) × (knot) or, to be more precise, on the cylinder Rt × Sσ1 , where
Sσ1 ⊂ S 3 is the great circle and K → Sσ1 is a multiple cover [44, 45, 86,
87]. In this description of doubly-graded slN homology, each braiding
(crossing) in a two-dimensional projection of a link K is represented by
the corresponding line defect Di in the two-dimensional Landau-Ginzburg
model, as illustrated in Figure 23.
• 3d N = 2 theory labeled by the knot K that “lives” on a 3-dimensional
part, Rt × R2 , of the fivebrane world-volume provides yet another way to
compute the Q-cohomology on either side of (4.2). This approach is a
certain variant of the so-called “3d-3d correspondence” [88, 89].
• Vortex counting: Compactification of M-theory on the Calabi-Yau 3-
fold CY3 is known to engineer 5d N = 2 gauge theory [90], in which
fivebranes (4.2) produce a codimension-2 defect, a “surface operator”.
Thus, from the vantage point of 5d gauge theory on Rt × T N4 with a
ramification (determined by the knot K) along Rt × R2 , the problem of
counting BPS states can be formulated in terms of K-theoretic instanton-
vortex counting [39, 91] that involves Hilbert schemes of points, etc.
These physical perspectives stimulated development of many new structures in knot
homologies, including the formulation of triply-graded homology categorifying the
HOMFLY-PT polynomial [92], which came as a bit of surprise [93], led to many
new differentials (canceling, universal, colored, exceptional), to new connections with
knot contact homology [94], recursion relations with respect to color-dependence
[95], etc.
The setup (4.2) has a natural extension to knot cobordisms (on both sides,
although the two extensions are not obviously related by a phase transition), cf.
(3.13):
(4.3)
doubly-graded
triply-graded
space-time: Λ2+ (M4 ) × T N4
space-time: XG2 × T N4
N M5-branes: M4 × R2
M5 -branes: LΣ × R2
M5 -branes: LΣ × R2
where the cobordism Σ combines the “time” direction and the direction along the
knot into a non-trivial 2-manifold, XG2 is a 7-manifold with G2 holonomy (= Rt ×
CY3 in simple cases), and LΣ is a coassociative submanifold described in (3.14).
A variant of the transition [56] relates the doubly-graded side of (4.2) and
(4.3) with surface operators produced from codimension-2 defects on the fivebrane
world-volume to similar brane configurations with M2-branes (codimension-4 sur-
face operators). For instance, a suitable analogue of (4.3) is
space-time: Λ2+ (M4 ) × T N4
(4.4) N M5-branes: M4 × R2
M2-branes: Σ × R+ × {0}
where the twisted product Σ × R+ is an associative submanifold in Λ2+ (M4 ). In
the special case M4 = Rt × S and Σ = Rt × Γ we find Λ2+ (M4 ) = Rt × T ∗ S 3 and
3
Σ × R+ ∼ = Rt × Γ × R+ .
Of particular interest is the cobordism representing the connected sum with
the unknot (2.29) which can fully characterize the 2d “effective theory” that lives
on patches of a general foam Σ.
D1 D2 ... Di ...
... ...
... ... σ
S3
Figure 23. A knot or a link K can be represented as a closure

of a braid that runs along one of the directions in M3 (= S 3 here).
Then, crossings or braidings (denoted by “◦”) of a surface operator
in extra dimensions of M3 “project” to interfaces in 2d Landau-
Ginzburg model on Rt × Sσ1 .
4.2. LG theory on “time × knot”. Our next goal is to describe the effective
2d theory that “lives” on the world-sheet Σ of surface operators which appear in
categorification of quantum group invariants.
The first clue comes from the early construction of knot homology groups
[92, 96] based on matrix factorizations. Since matrix factorizations are known
to describe boundary conditions and interfaces in 2d Landau-Ginzburg models
[23, 97–101] it was proposed in [86, 87, 102] to interpret the matrix factoriza-
tions used by Khovanov and Rozansky as interfaces in 2d Landau-Ginzburg model
on what we call Σ, i.e. on the product of the time direction Rt and the strands of
a knot, or link, or tangle, as illustrated in Figure 23.
There are several ways to show this, including dualities that map intersecting
branes in M-theory to intersecting D-branes which allow to address the question
of what degrees of freedom live on Σ via the standard tools of perturbative string
theory. The upshot of this exercise is that for each strand in the braid or link K
colored by R one finds a 2d theory whose chiral ring agrees with the homology of
the unknot (2.30) colored by a representation R of Lie algebra g.
Using this physical interpretation, not only can one reproduce the superpo-
tential Wg,R (x) for g = slN and R = used by Khovanov and Rozansky but
also produce new Landau-Ginzburg potentials for more general representations.
This includes all symmetric and anti-symmetric representations of g = slN , cer-
tain representations of exceptional Lie algebras and g = soN [86], which later were
used in mathematical constructions of the corresponding knot homologies, see e.g.
[28, 103–105].
In all of these constructions, the category (2.27) is a category of matrix factor-

izations MF(W ) of a polynomial
(4.5) W = W R1 + . . . + W Rn
that is a sum of potentials WRi , one for each marked point (or strand). Corre-
spondingly, the category
(4.6) CR1 ,...,Rn = MF(W ) = MF(WR1 ) ⊗ . . . ⊗ MF(WRn )
is a product of categories MF(WRi ) associated to individual strands. As discussed
in section 2.4, the Hochschild homology of the MF(WR ) is supposed to agree with
the homology of the R-colored unknot,

(4.7) A = HH ∗ MF(WR ) = HR
if the knot homology at hand extends to a functor (2.25), c.f. (2.28). The Hochschild
homology can be computed as the the homology of the Koszul complex associated
with the sequence of partial derivatives of WR . It contains the Jacobi ring J (WR )
and is in fact equal to it if and only if WR has only isolated singularities. Indeed, it
is this case, which will be relevant for us. Hence, we get a non-trivial constraint on
the superpotential WR , namely that its Jacobi ring, which is also the chiral ring of
the Landau-Ginzburg model associated to an R-colored point, has to agree with the
respective knot homology. This identification has to hold on the level of algebras.
(Recall that the knot homology carries an algebra structure containing information
about knot cobordisms, c.f. the discussion around (2.31).)
Beyond this, also deformations of Khovanov-Rozansky homology groups cate-
gorifying quantum slN invariants [30, 32, 33, 35, 38, 106] must be correctly incor-
porated in the structure of the category (2.27) associated to marked points. Indeed,
this is not entirely unrelated from the algebra structure: differentials in spectral
sequences that relate different variants of knot homology are often represented by
generators of the algebra (2.30).
In the Landau-Ginzburg setup, such deformations should be realized by relevant
deformations of the Landau-Ginzburg potential WR , providing further consistency
checks for the proposed Landau-Ginzburg description. Moreover, also the physically
motivated generalizations of the original Khovanov-Rozansky construction to other
representations mentioned above give further credence to the Landau-Ginzburg
approach.
In addition, parallel developments in the study of surface operators led to a
number of alternative descriptions of the chiral ring (4.7) which, in all cases of our
interest, agrees with the chiral ring of the Landau-Ginzburg model with the super-
potential WR . Indeed, as we reviewed in section 3, the same brane configurations
(4.3)–(4.4) that we use for categorification of quantum group invariants describe
codimension-2 and codimension-4 defects in 6d theory on M5-branes. Thus, one
can use the description of surface operators as coupled 2d-4d systems to describe
the effective 2d theory on Σ.
Because the question about the chiral ring A of the 2d theory on Σ is completely
local (in a sense that it does not depend on the geometry away from Σ as well as
position along Σ), we can take M4 = R4 and Σ = R2 . Then, the question basically
reduces to the study of the chiral ring in 2d theory on the surface operator labeled
by R in 4d N = 4 gauge theory on M4 that we already analyzed in section 3. In
this paper, we are mainly interested in the case of the gauge group G = SU (N )
and its k-th anti-symmetric representation R = Λk CN . The corresponding surface

operators have Levy type L = SU (U (k) × U (N − k)) and in the description as
2d-4d coupled systems à la (3.3), the 2d theory on Σ is a N = (4, 4) sigma-model
with hyper-Kähler target space T ∗ (G/L) = T ∗ Gr(k, N ). Topological twist of the
ambient theory on M4 induces a topological twist of the 2d theory on Σ, which
then becomes a 2d TQFT with the chiral ring
(4.8) A = H ∗ (Gr(k, N ))
given by the classical cohomology of the Grassmannian, in agreement with (2.32)
and (3.46). A 2d TQFT with the same chiral ring can be obtained by a topolog-
ical B-twist of the N = (2, 2) Landau-Ginzburg model with k chiral superfields
x1 , . . . , xk of U (1)R -charge q = N2+1 and the superpotential
(4.9) W0 (x1 , . . . , xk ) = xN
1
+1
+ . . . + xN
k
+1
.
More precisely, a change of variables from the xi to the elementary symmetric
polynomials7 Xi = σi (x1 , . . . , xk ) (see e.g. section 8.3 of [107]):
(4.10) (x1 , . . . , xk ) −→ (X1 = σ1 (x1 , . . . , xk ), . . . , Xk = σk (x1 , . . . , xk ))
gives rise to a new Landau-Ginzburg model denoted by LGk . Its chiral superfields
Xi have U (1)R -charge qi = N2i+1 , and its superpotential W = W (X1 , . . . , Xk ) is
just W0 expressed in terms of the Xi , i.e. W (σ1 (x1 , . . . , xk ), . . . , σk (x1 , . . . , xk )) =
W0 (x1 , . . . , xk ). It is still quasi-homogeneous. In the IR, LGk flows to a supercon-
formal field theory of central charge8
3k(N − k)
(4.11) ,
c=
N +1
which is believed to be the level-1 Kazama-Suzuki model associated to the Grass-
mannian Gr(k, N ). The chiral ring of LGk is the Jacobi ring of W (X1 , . . . , Xk )
and indeed agrees with the classical cohomology ring H ∗ (Gr(k, N )) of the Grass-
mannian[66]. The chiral superfields Xi correspond to the Chern classes ci of the
tautological bundle over Gr(k, N ).
Note that the M-theory setup features two U (1)-symmetries induced by rota-
tions in T N4 . One of them, namely the q-grading, descends to the U (1)R -symmetry
of the Landau-Ginzburg model. The other one will not play a role in our discussion.
4.3. Junctions and LG interfaces. Let us now turn to the Landau-Ginzburg

description of junctions of surface operators. We are interested in junctions of sur-
face operators which are created when the stack of k M5 -branes is split up into
two stacks of k1 and k2 = k − k1 M5 -branes, c.f. Figure 20 and Figure 24.
As we already discussed below (3.44), from the point of view of the Grassman-
nian sigma-model this splitting corresponds to the following boundary condition
at the junction: the k1 - and k2 -dimensional subspaces defining the theory after
the splitting are orthogonal at the point of splitting and span the k = (k1 + k2 )-
dimensional subspace defining the theory before the splitting. This condition can
be described by means of the correspondence (3.9) between (products of) Grass-
(k) (k)
mannians. It identifies the Chern classes c1 , . . . , ck of the tautological bundle
7σ

j (x1 , . . . , xk )
= 1≤i1 <...<ik ≤k xi1 . . . xij

i (1 − qi ).
8 The central charge associated to LG-models can be obtained as c = 3
Figure 24. Junction between LGk , LGk1 and LGk2 =k−k1 . Upon
folding it can be described by an interface Ikk1 ,k2 between LGk and
LGk1 ⊗ LGk2 .
over Gr(k, N ) with the symmetrizations
(k)

i
(k1 ) (k2 )
(4.12) ci = cj ci−j
j=0
(k )
of the Chern classes cj i of the tautological bundles over Gr(ki , N ), see Appendix
C.
This translates to a rather simple identification of chiral fields between the
respective Landau-Ginzburg models: We regard the junction via the folding trick as
an interface Ikk1 ,k2 between Landau-Ginzburg models LGk on one side and LGk1 ⊗
LGk2 on the other. Both these theories can be realized by means of a change
of variables from one and the same model, LG⊗k 1 , the Landau-Ginzburg model
with k chiral superfields x1 , . . . , xk and superpotential (4.9). On the left side of
the interface, in the model LGk , the chiral superfields Xi are obtained as the
symmetrization of all the variables xi , c.f. (4.10), and correspond to the Chern
(k)
classes ci of Gr(k, N ). On the right side of the interface, the superfields of the
model LGk1 ⊗LGk2 are obtained by symmetrizing the first k1 variables (x1 , . . . , xk1 )
and the last k2 = k − k1 variables (xk1 +1 , . . . , xk ) separately:
(4.13)(x1 , . . . , xk ) −→ (Y1 , . . . , Yk ) =
(σ1 (x1 , . . . , xk1 ), . . . , σk1 (x1 , . . . , xk1 ), σ1 (xk1 +1 , . . . , xk ), . . . , σk2 (xk1 +1 , . . . , xk )) .
The Zi = Yi , i = 1, . . . , k1 and Zj = Yk1 +j , j = 1, . . . , k2 are the superfields of the

factor models LGk1 and LGk2 , respectively, and can be identified with the Chern
(k ) (k )
classes ci 1 and cj 2 . In terms of these fields, the junction condition (4.12) reads9

i

(4.14) Xi = σi (x1 , . . . , xk ) = fi (Y1 , . . . , Yk ) = Zj Zi−j .
j=0
9 Here one sets Z = 1 = Z and Z = 0 if i is out of bounds (i < 0 or i > k ), Z = 0 if j is

0 0 i 1 j
out of bounds (j < 0 or j > k2 ).
It trivially identifies the underlying models LG⊗k

1 on both sides of the interface,
and identifies the more symmetric variables Xi on its left with the respective sym-
metrization of the less symmetric variables Yi on its right:
X1 = σ1 (x1 , . . . , xk ) = f1 (Y1 , . . . , Yk ) = Y1 + Yk1 +1 ,
(4.15) X2 = σ2 (x1 , . . . , xk ) = f2 (Y1 , . . . , Yk ) = Y2 + Yk1 +2 + Y1 Yk1 +1 ,
...
This discussion immediately generalizes to more complicated configurations.
The junction created by splitting up the stack of M5 -branes into r > 2 stacks
of multiplicities k1 , . . . , kr , respectively, on the level of the Grassmannian sigma-
models is described by the correspondence via the partial flag manifold
F l(k1, k1 +
k2 , . . . , k1 + . . . + kr = k, N ) = G/L with the Levi subgroup L = S i U (ki ) , i.e.
the space of flags10
(4.16) Ck1 ⊂ Ck1 +k2 ⊂ · · · ⊂ CN
Figure 25. Interface Ikl11,...,l

,...,kr between LGk1 ⊗ . . . ⊗ LGkr and
s
LGl1 ⊗ . . . ⊗ LGls .
As before, this translates to a simple condition at the interface Ikk1 ,...,kr between
the respective Landau-Ginzburg models, LGk on one side and LGk1 ⊗ . . . ⊗ LGkr
on the other. Again, both models can be obtained by different changes of variables
from one and the same underlying model, LG⊗k 1 , which is trivially identified by
the interface. The superfields X1 , . . . , Xk of LGk are the symmetrization of all the
fields x1 , . . . , xk , whereas the fields Y1 , . . . , Yk of LGk1 ⊗ . . . ⊗ LGkr are obtained by
separately symmetrizing the sets of variables (x1 , . . . , xk1 ), (xk1 +1 , . . . , xk1 +k2 ), . . .,
(xk1 +...+kr−1 +1 , . . . , xk ). The interface then expresses the more symmetric polyno-
mials Xi in terms of the less symmetric ones Yj :
(4.17) Xi = σi (x1 , . . . , xk ) = fi (Y1 , . . . , Yk ) .
From this it is also evident how to describe a junction at which r stacks of
M5 -branes of multiplicities k1 , . . . , kr join and immediately split up into s stacks or
multiplicities l1 , . . . , ls , c.f. Figure 25. (In this and the following figures we will sup-
press the time direction. The surface operators will be represented by lines, which
will be labeled by the multiplicity of the respective stack of M5 -branes. Surface
operators also have an orientation, which, in case of ambiguity will be indicated by
arrows. The junctions discussed here, which arise by splitting and joining of stacks
10 See e.g. [108] for further details on topology of flag varieties and their coupling to 4d
N = 4 and N = 2 gauge theories.

Figure 26. Junction described by Ikl11,...,l s

,...,kr (left) factorizes as
Ikk1 ,...,rr ∗ Ikl1 ,...,ls over LGk (right).
of M5 -branes have the property that the sum of labels of incoming lines equals the
sum of labels of outgoing lines: M5 -branes cannot end in junctions.) The junction
of Figure 25 can be realized by an interface Ikl11,...,l s
,...,kr between LGk1 ⊗ . . . ⊗ LGkr

on the left side and LGl1 ⊗ . . . ⊗ LGls on the right. Since i ki = k = j lj , the
models on either side of the interface arise from one and the same model LG⊗k 1 by
means of different symmetrizations of variables. On the left, the sets (x1 , . . . , xk1 ),
(xk1 +1 , . . . , xk1 +k2 ), . . ., (xk1 +...+kr−1 +1 , . . . , xk ) are symmetrized separately giv-
ing rise to the chiral superfields (X1 , . . . , Xk ), whereas on the right the superfields
(Y1 , . . . , Yk ) are obtained by symmetrizing the (x1 , . . . , xl1 ), (xl1 +1 , . . . , xl1 +l2 ), . . .,
(xl1 +...+ls−1 +1 , . . . , xk ). The interface now relates the fields on the two sides of
the defect by expressing the symmetric polynomials σi (x1 , . . . , xk ) in terms of the
partially symmetrized ones on either side:
f1 (X1 , . . . , Xk ) = σ1 (x1 , . . . , xk ) = f1 (Y1 , . . . , Yk )
(4.18) f2 (X1 , . . . , Xk ) = σ2 (x1 , . . . , xk ) = f2 (Y1 , . . . , Yk )
... ,
In particular, this condition is the composition of the identifications imposed by
the interfaces Ilk1 ,...,ls describing the splitting of one stack into s stacks and the one
imposed by the interface Ikk1 ,...,kr describing the joining of the r stacks into one
stack. Thus, the respective interface can be obtained by fusion
(4.19) Ikl11,...,l s k l1 ,...,ls

,...,kr = Ik1 ,...,rr ∗ Ik
of these two interfaces over LGk , c.f. Figure 26.

In fact, more generally, the interfaces Ikl11,...,l s
,...,kr factorize into those interfaces
describing trivalent junctions (where either two stacks of M5 -branes join into one
or one stack splits up into two stacks), see Figure 27.
4.4. Junctions and matrix factorizations. In the following, we will study

the junctions of surface operators introduced in the last section in more detail using
an elegant representation of interfaces between Landau-Ginzburg models in terms
of matrix factorizations [23]. This representation in particular lends itself easily
to the description of fusion. (Of course one could equivalently use the language of
correspondences in Grassmannian sigma model to do the analysis.)
A matrix factorization over a polynomial ring R of a polynomial W ∈ R is a
pair of square matrices p0 and p1 with entries in R, which multiply to W times the
Figure 27. Junction (left) can be factorized into trivalent junc-

tions (right).
identity matrix:
(4.20)
p1
P: P1 ∼
= Rr −
←−
−−
−−
−−
−−
−−
−−
−→ r ∼
− R = P0 , p1 p0 = W idP0 , p0 p1 = W idP1 .
p0
A B-type supersymmetric interface connecting a Landau-Ginzburg model with chi-

ral superfields X1 , . . . , Xk and superpotential W (X1 , . . . , Xk ) to one with super-
fields Y1 , . . . , Yl and superpotential W (Y1 , . . . , Yl ) is determined by a matrix fac-
torization of the difference W (X1 , . . . , Xk ) − W (Y1 , . . . , Yl ) of superpotentials over
the ring R = C[X1 , . . . , Xk , Y1 , . . . , Yl ] of polynomials in the chiral superfields on
both sides.
From this description it is easy to read off certain properties of the interface,
such as the chiral ring of operators on the interface, which is given by the BRST-
cohomology on EndR (P0 ⊕P1 ), where the BRST-operator Q acts on Φ ∈ EndR (P0 ⊕
P1 ) by graded commutator
(4.21) QΦ = p Φ − σΦσ p
with the operator

0 p1
(4.22) p= .
p0 0
The Z2 -grading (fermion number) is given by

idP0 0
(4.23) σ= .
0 −idP1
Correlation functions of chiral primary fields in the presence of such interfaces can
be calculated by means of a residue formula [98].
Moreover, fusion of interfaces P and Q separating respectively a Landau-
Ginzburg model with superpotential W (X1 , . . . , Xr ) from one with superpotential
W (Y1 , . . . , Ys ), and the model with superpotential W (Y1 , . . . , Ys ) from one with
Figure 28. Fusion of interface P between LG models with su-

perpotentials W (Xi ) and W (Yi ), and Q between LG models with
superpotentials W (Yi ) and W (Zi ) (left) gives rise to interface
P ∗Q between LG models with superpotentials W (Xi ) and W (Zi )
(right).
superpotential W (Z1 , . . . , Zt ), c.f. Figure 28, has a very simple description in terms
of matrix factorizations. The interfaces are described by two matrix factorizations
p1 q1
(4.24) P : P1 −
←−
−−
−−
−−
−−
−−
−−
−→
− P0 and Q : Q1 −
←−
−−
−−
−−
−−
−−
−−
−→
− Q0
p0 q0
of W (X1 , . . . , Xr ) − W (Y1 , . . . , Ys ) and W (Y1 , . . . , Ys ) − W (Z1 , . . . , Zt ), respec-

tively. The fused interface P ∗ Q separates the Landau-Ginzburg models with

superpotential W (X1 , . . . , Xr ) from the one with superpotential W (Z1 , . . . , Zt ).
It is described by the tensor product matrix factorization
(4.25) ⎛ ⎞
⎛ ⎞ ⎝ p1 ⊗ idQ0 −idP0 ⊗ q1 ⎠ ⎛ ⎞
P1 ⊗ Q0 idP1 ⊗ q0 p0 ⊗ idQ1 P0 ⊗ Q0
(P ⊗ Q)1 := ⎝ ⊕ ⎠← −−−−−−−−−−−−−−−−−−−−→ ⎝
⎛ −−−−−−−−−−−−−−−−−−−− ⎞
⊕ ⎠ =: (P ⊗ Q) .
0
P0 ⊗ Q1 p 0 ⊗ idQ idP ⊗ q 1 P 1 ⊗ Q 1
⎝ 0 1 ⎠
−idP0 ⊗ q0 p1 ⊗ idQ1
This is a matrix factorization of the sum

(4.26) W (X1 , . . . , Xr ) − W (Y1 , . . . , Ys ) + W (Y1 , . . . , Ys ) − W (Z1 , . . . , Zt )
= W (X1 , . . . , Xr ) − W (Z1 , . . . , Zt ) ,
as it should be. It still involves the fields Yi of the Landau-Ginzburg model
squeezed in between the interfaces, which are promoted to new degrees of freedom
on the interface. Therefore, a priori it is a matrix factorization of infinite rank over
C[X1 , . . . , Xr , Z1 , . . . , Zt ]. However, it can be shown that by splitting off “trivial”
matrix factorizations11 these matrix factorizations always reduce to finite rank. See
[23, 87, 109] for a more detailed discussion of fusion in this context.
The Landau-Ginzburg models we are interested in here exhibit a U (1)R -sym-
metry, and the interfaces also preserve this symmetry. This means that the modules
P0 and P1 carry representations ρ0 and ρ1 of U (1)R , respectively, which are compat-
ible with the R-module structure, such that p0 and p1 are homogeneous of charge
112 . The respective matrix factorization is called “graded”. For ease of notation
11 matrix factorizations of the form 1 · W , are physically trivial
12 The superpotentials which are factorized by p0 and p1 have U (1)R -charge 2.
we will rescale the U (1)R -charges by the degree N + 1 of the potential. In these
units the superpotential has charge 2N + 2, the maps pi have charge N + 1, and the
fields Xi defined in (4.10) have charge 2i. All charges appearing will be integral.
Now, the interface Ikk1 ,k2 associated to the trivalent junction in Figure 24 sep-
arates the Landau-Ginzburg model LGk with superpotential W (X1 , . . . , Xk ) (the
superpotential of (4.9) written in the symmetrized fields (4.10)), from the Landau-
Ginzburg model LGk1 ⊗ LGk2 , which has superpotential W (f1 , . . . , fk ) (the same
superpotential, but written in the only partially symmetrized fields (4.13)). Here
fi = fi (Y1 , . . . , Yk ) is the map from (4.15) symmetrizing the partially symmetrized
Yi . The matrix factorization of W (X1 , . . . , Xk ) − W (f1 , . . . , fk ) which imposes the
identification (4.15) is now easy to construct. It is given by a Koszul type matrix
factorization, c.f. [23].
One can find homogeneous polynomials Ui (X1 , . . . , Xk , Y1 , . . . , Yk ), i = 1, . . . , k,
such that
(4.27) W (X1 , . . . , Xk ) − W (f1 , . . . , fk )

k
= (Xi − fi (Y1 , . . . , Yk )) Ui (X1 , . . . , Xk , Y1 , . . . , Yk ) .
i=1
One particular choice13 of such polynomials is given by

(4.28)
1
Ui = (W (f1 , . . . , fi−1 , Xi , . . . , Xk ) − W (f1 , . . . , fi , Xi+1 , . . . , Xk )) .
Xi − fi
The matrix factorization describing the interface is then given by the tensor product
) k *
k1 ,k2

(4.29) Ik = P {−k1 k2 }
i
i=1
of the rank-1 matrix factorizations
pi1 =(Xi −fi )
(4.30) Pi : P1i ∼
= R{2i − N − 1} −
←−
−−
−−
−−
−−
−−
−−
−→ ∼ i
− R{0} = P0
pi0 =Ui
over R = C[X1 , . . . , Xk , Y1 , . . . , Yk ]. Since, under the tensor product of matrix fac-

torizations, the factorized polynomials behave additively, (4.29) is indeed a matrix
factorization of (4.27) and has rank r = 2k−1 . Here, with the notation R{a} we
specify the U (1)R -representations on R by indicating the charge a of 1 ∈ R; and a
matrix factorization Q{a}, as in (4.29), denotes the matrix factorization Q where
all U (1)R -charges are shifted by a14 .
This construction immediately generalizes to the more general junctions Ikl11,...,l s
,...,kr
of Figure 25. The respective interface separates the Landau-Ginzburg model LGk1 ⊗
. . . ⊗ LGkr on the left from LGl1 ⊗ . . . ⊗ LGls on the right. Both models are ob-
tained from the Landau-Ginzburg model of (4.9) by means of different symmetriza-
tions of the variables x1 , . . . , xk . On the left, the sets of variables (x1 , . . . , xk1 ),
(xk1 +1 , . . . , xk1 +k2 ), . . ., (xk1 +...+kr−1 +1 , . . . , xk ) are symmetrized separately, giving
rise to the superfields (X1 , . . . , Xk ), whereas on the right the superfields (Y1 , . . . , Yk )
are obtained by symmetrizing the sets of variables (x1 , . . . , xl1 ), (xl1 +1 , . . . , xl1 +l2 ),
. . ., (xl1 +...+ls−1 +1 , . . . , xk ). The superpotentials on the two sides are just given by
13 Other choices lead to equivalent matrix factorizations.
14 Such a shift only changes the U (1)R -charges of interface changing fields.
(4.9), expressed in terms of the respective superfields on the two sides. On the
left it is W (f1 , . . . , fk ), and on the right W (f1 , . . . , fk ), where fi and fi are the
maps (4.18) symmetrizing the partially symmetric (X1 , . . . , Xk ) and (Y1 , . . . , Yk ),
respectively.
As for the trivalent case, the junction condition (4.18) can now be implemented
by a Koszul matrix factorization. The polynomials
1

(4.31) Ui = W (f1 , . . . , fi−1

, fi , . . . , fk ) − W (f1 , . . . , fi , fi+1

, . . . , fk ) ,
fi − fi
satisfy

k
(4.32) W (f1 , . . . , fk ) − W (f1 , . . . , fk ) = (fi − fi ) Ui ,
i=1
and the tensor product

) k *

(4.33) Ikl11,...,l s
,...,kr = Pi {− 1≤a<b≤k la lb }
i=1
of the rank-one matrix factorizations

pi1 =(fi −fi )
∼
= R{2i − N − 1} −− ∼ i
P : −−
−−
−−
−−
−−
−−→
− R{0} = P0
i
(4.34) P1i ← −
pi0 =Ui
implements the identification (4.18) across the interface.15
Figure 29. Fusion of Ikk1 ,k2 and Ikk1 ,k2 in LGk1 ⊗ LGk2 produces
copies of the identity defect in LGk .
4.5. Junctions and categorification of quantum groups. The interfaces

described in the last section satisfy many interesting properties. For instance, from
the construction it is evident, that they factorize through a surface operator LGk ,
i.e.
(4.35) Ikl11,...,l s l1 ,...,ls
,...,kr = Ik ∗ Ikk1 ,...,kr ,
c.f. Figure 26, and that more generally, they factorize into interfaces corresponding
to trivalent junctions, as displayed in Figure 27.
the Landau-Ginzburg defect Ikl11,...,l

15 Indeed, s
,...,kr for the case s = 1, ki = 1 was already
mentioned in [110].
Moreover, fusing the interfaces Ikk1 ,k2 and Ikk1 ,k2 (k = k1 + k2 ) in LGk1 ⊗ LGk2
produces a number of copies of the identity defect Idk = Ikk in LGk :
+ ,
k1 ,k2 k
(4.36) Ik1 ,k2 ∗ Ik
k
= Idk ,
k1
see Figure 29. Here, we use the notation ¶{q a1 + . . . + q ar } := ¶{a1 } ⊕ . . . ⊕ ¶{ar }
[k]!
for a matrix factorization ¶, and, as before, kk1 = [k1 ]![k−k 1 ]!
denotes the quantum
−k
−q k
binomial coefficient with [k]! = [k][k − 1] · · · [1], [k] = qq−q −1 = q
k−1
+ q k−3 + . . . +
q −k+1 . A simple proof of this relation can be found in Appendix D.1.
Let

k1 −1,1
(4.37) Ek1 ,k2 := Idk1 −1 ⊗ I1,k
k2 +1
2
∗ Ik 1
⊗ Id k 2

2 −1
Fk1 ,k2 := Ik1 ,1 ⊗ Idk2 −1 ∗ Idk1 ⊗ Ik1,k
k1 +1
2
,
be the combination of interfaces represented in Figure 30. Then, it is not difficult
to see (c.f. Appendix D.2) that
(4.38) ∼ Fk −1,k +1 ∗ Ek ,k ⊕ Idk ,k {[k2 − k1 ]}
Ek +1,k −1 ∗ Fk ,k =
1 2 1 2 1 2 1 2 1 2
for k1 ≤ k1 and
(4.39) Fk1 −1,k2 +1 ∗ Ek1 ,k2 ∼
= Ek1 +1,k2 −1 ∗ Fk1 ,k2 ⊕ Idk1 ,k2 {[k1 − k2 ]}
for k1 ≥ k2 .
Figure 30. Configurations of surface operators categorifying the

quantum group generators: Ek1 ,k2 (left) and Fk1 ,k2 (right).
Here Idk1 ,k2 = Idk1 ⊗ Idk2 denotes the identity defect in the tensor product
LGk1 ⊗ LGk2 . But this is nothing but a categorification of the relation [E, F ]1n =
[n]1n of the quantum group U̇q (sl2 ), c.f. (2.20). Here, the difference n = k2 − k1
plays the role of a sl2 weight.
In fact, the matrix factorizations Ikl11,...,l s
,...,kr have already appeared in the con-
struction of slN -link homology in the math literature [28]. There it was shown that
they satisfy relations categorifying the MOY relations depicted in Figure 10. This
means in particular, that the junctions of surface operators defined in section 4.3
provide a categorification of the U̇q (slm ) representation obtained by ladders of Wil-
son lines in Chern-Simons theory, c.f. section 2. This 2-representation was analysed
from a mathematical point of view in [111]. Building blocks are configurations of
m parallel stacks of M5 -branes, on which the quantum group generators Ei act
by peeling off one M5 -brane from the i-th stack and migrating it to the (i + 1)-st
stack, whereas the Fi are realized by peeling off one M5 -brane from the (i + 1)-st
and moving it to the i-th stack.
The 2-category structure is given as follows: Objects are m-tuples (k1 , . . . , km ),
0 ≤ ki ≤ N , corresponding to m parallel stacks of M5 -branes of multiplicities ki .
From a 2d perspective they are realized by tensor products LGk1 ⊗ . . . ⊗ LGkm
of the Landau-Ginzburg theories on the respective world-volumes. 1-Morphisms
correspond to configurations joining such stacks. In 2d they are interfaces between
the respective Landau-Ginzburg models. The 2-morphisms correspond to interface
changing fields.
5. What’s next?
We conclude with a list of questions and possible generalizations, some of which
have already been mentioned in the text:
• Even though the current work shows how the BPS junctions of surface
operators realize the generators and elementary relations of the 2-category
U̇, it still remains unproven whether these junctions reproduce the entire
set of KLR algebra relations, or equivalently, the foam relations given in
[18].
• It would be interesting to verify the existence of BPS junctions of surface
operators by alternative methods (in particular, by constructing explicit
solutions to the field equations) and extend the classification presented
here to more general 4d theories and more general surface operators.
• For example, it would be interesting to construct holographic duals of
surface operator junctions proposed here.
• Also it would be interesting to investigate, whether the categories of junc-
tions of surface operators can be reconstructed out of the representation
varieties appearing in the discussion of the charge conservation condition
at junctions in section 3.5.
• Mathematically, other types of junctions (for other types of surface op-
erators and LG models) also lead to diagrammatic algebras and may be
worth exploring.
• As a generalization in a different direction, one may keep the same types
of surface operators and LG interfaces, but consider changing space-time
geometry, in particular, the geometry of the fivebrane world-volume. In-
deed, in most of our discussion we assumed M3 = R3 , and one immediate
generalization would be to consider M3 = S 1 × R2 which would result in
planar diagrammatics on a cylinder. More generally, one might consider
junctions of surface operators in M4 = Σ × R2 which, via the arguments
of this paper, would lead to LG models and planar diagrammatics on a
Riemann surface Σ.
• The analysis of section 3 suggests to carry out the analogue of [63, 64] for
the complexified gauge group GC and to compare Poincaré polynomials of
the resulting moduli spaces M(GC , Γ) to MOY polynomials of the planar
graphs Γ.
• In section 4 we already mentioned two important generalizations to su-
pergroups and symmetric representations that should categorify the super
q-Howe duality [72] and symmetric q-Howe duality [73], respectively.
• It would be interesting to work out a precise dictionary between “fake bub-

bles” of the diagrammatic approach [14] and relevant deformations of the
Landau-Ginzburg potential [86], both of which are related to differentials
on Khovanov-Rozansky homology.
We plan to tackle some of these questions in the future work.
Acknowledgments
The authors thank A. Lauda and D. Rose for many patient and very help-
ful explanations, as well as J. Kamnitzer, M. Khovanov, A. Lobb, M. Mackaay,
L. Rozansky, M. Stosic, C. Stroppel, C. Teleman, B. Webster, and P. Wedrich for
helpful comments and advice. We also thank participants of the “Knot homologies,
BPS states, and SUSY gauge theories” program at the Simons Center for Geometry
and Physics for useful discussions.
Appendix A. Wilson lines and categories N Web

In this section, we derive the remaining relations of Figure 10. Before pro-
ceeding to the derivations, however, it is necessary to resolve an ambiguity in our
normalization. Let us denote the invariant tensors at the junctions of Figure 7 as
k1 ,k2 and k1 ,k2 , respectively. Since the RHS of Figure 7 only determines the nor-
malization of their product, it remains the same if one invariant tensor is multiplied
by a complex number and the other by the inverse. Thus, there is an ambiguity
in the “relative” normalization among the invariant tensors from different types of
junctions. Such ambiguity would not be observed if ’s and ’s always appear in
pairs, but they do not in Figure 10(a), 10(d) and 10(f).
We claim that the junctions can be normalized so as to satisfy the relations of
Figure 10. Let us first normalize those which involve Wilson lines in the defining
representations. The open Wilson lines of Figure 31(a) are proportional to each
other, since the associated Hilbert space is 1-dimensional. Renormalize the invariant
tensors i,1 and 1,i via multiplication by ηi (c.f. Figure 31(b)), and renormalize

i,1 and 1,i accordingly via multiplication by 1/ηi . This proves Figure 10(a) for
j = k = 1.
i i i i
2 2 2
(a) i+1 = ηi and i+1 i+1 = ηi2
i+2 i+2 i+2 i+2

i i 1 1
(b) 1 → ηi 1 and i → ηi i
i+1 i+1 i+1 i+1
Figure 31. (a) Definition of ηi and the closed Wilson lines which
determine the value of ηi2 . (b) Renormalization of junctions which
involve the Wilson lines in .
j
j
j −1 i
i j k i
j k j −1 k
i+1
[j] i+j = i+j =
i+j
i+j +k i+j +k i+j +k
j j
i k i j −1
j −1 k
j −1 +k
= j −1+k =
i+1 j +k
i+j +k i+j +k
j
j −1 j
i i k
k
j
= j +k = [j] j +k
i+j +k i+j +k
Figure 32. Induction on j. Apply the base case j = 1 and the

induction hypothesis for j − 1 in the red dashed cirlces.
Inductively renormalize the junctions of higher rank representations. Consider

Wilson lines of Figure 10(a) with j = 1, and assume WLOG that i ≥ k. Since
the associated Hilbert space is 1-dimensional, the Wilson lines on each side are
proportional to each other. Renormalizing i,k+1 by absorbing the proportionality
constant, we have turned the relation into an identity and determined the relative
normalization between i,k+1 and i+1,k . Proceed recursively until we reach i+k,1 ,
whose normalization is fixed by Figure 31(b). Applying this procedure for all i + k,
the relative normalizations among the invariant tensors can be determined such
that Figure 10(a) holds for j = 1. Finally, induction on j shows that Figure 10(a)
holds for any i, j and k (c.f. Figure 32.)
Now that we have fixed the normalization of junctions, let us proceed to derive
the remaining relations. Figure 10(b) is rather straightforward, once we close the
open ends by connecting them.
The three vectors of Figure 10(c) satisfy a linear relation, as the associated
Hilbert space is 2-dimensional. Close the open ends in two inequivalent ways
(c.f. Figure 12.) The resulting closed Wilson lines have computable expectation
values via the direct sum formula, Figure 10(a) and Figure 10(b).
Figure 10(d) follows from Figures 10(a), (b) and (e). First, apply Figure 10(a)
and 10(b) to replace the right-moving Wilson line labeled by ∧n (the left-moving
Wilson line labeled by ∧l+n−1 ) by n right-moving Wilson lines labeled by (l+n−1
left-moving Wilson lines labeled by ). Recursively apply Figure 10(e) until the
resulting Wilson lines are those of the RHS, and the coefficients follow.
Figure 10(f) also follows from Figures 10(a), (b) and (e). First, apply Figure
10(a) on LHS with (i, j, k) = (m−2, 1, 1). Use Figure 10(a) and (b) to replace the
upper Wilson line labeled by ∧j by j Wilson lines labeled by . Then, recursively
apply Figure 10(e) until the resulting Wilson lines differ from those of the RHS only
by a single right-moving Wilson line labeled by ∧2 . Use Figure 10(a) and (b) to
replace the latter by two Wilson lines labeled by , and the coefficients follow. This
concludes our proof of Figure 10.
Appendix B. Domain walls, junctions and Grassmannians

For N > r, and for generic values of mass parameters mi , 4d N = 2 SQCD
theory has massive isolated vacua (with in < in+1 ):
(B.1) φ = diag(mi1 , mi2 , . . . , mir )
in which the gauge group U (r) is completely broken and SU (Nf ) → U (1)Nf −1 .
When all masses are real — as e.g. in 5d lift of this theory — one can also as-
sume mi < mi+1 . Clearly, the set of such vacua is in one-to-one correspondence
with Schubert cells SI in Gr(r, N ) or, equivalently, vertices of the moment graph
associated to the T = U (1)N action on Gr(r, N ) discussed around (3.62).
Note, that the mass parameters mi can be identified with the equivariant pa-
rameters for the torus action. For instance, U (2) theory with Nf = 4 has six vacua
(3.57), also shown in Figure 22.
While vacua of 4d N = 2 SQCD correspond to vertices of the moment graph,
elementary (not composite) domain walls that interpolate between vacua labeled
by I and J correspond to edges of the moment graph connecting vertices I and J.
Two vertices I = J are joined by an edge if and only if
(B.2) |I ∪ J| = r − 1
in which case this edge is labeled by mi − mj , where i and j are the two elements
in the symmetric difference of I and J. For instance, U (2) theory with Nf = 4
has 12 elementary walls that correspond to edges of the octahedron in Figure 22.
In general, each elementary wall carries U (1) ⊂ U (r) symmetry left unbroken by
r − 1 components of the Higgs field which remain non-zero in the core of the wall in
the weak coupling limit. For this reason, such elementary walls are called abelian
[112]. For special values of mass parameters, one finds walls that support non-
abelian symmetries [71, 113].
The BPS condition relates the domain wall tension T and its orientation, say,
in the (x1 , x2 ) plane to the absolute value and phase of mi − mj , respectively:
(B.3) mi − mj = T eiθ
where θ is the angle in the (x1 , x2 ) plane. In particular, if all mass parameters mi
are real, the BPS condition means that walls must be parallel in four-dimensional
space-time and their ordering in x2 direction precisely follows the closure ordering
of Schubert cells (3.56). When mass parameters are complex, on the other hand,
in general one finds a web (or network) of walls at angles determined by (B.3), see
e.g. [114] for a nice “adiabatic” argument (for small values of Im(mi )).
Webs or networks of domain walls involve junctions, which can be either abelian
or non-abelian, depending on the unbroken gauge symmetry in the core of the
junction at weak coupling [112]. Specifically, a trivalent junction of walls separating
vacua labeled by Schubert symbols I1 = a · · ·, I2 = b · · ·, and I3 = c · · · has
vanishing components of the Higgs field φa , φb , and φc in the core of the junction
and, therefore, carries U (1) gauge symmetry left unbroken by the remaining r − 1
non-zero components of the Higgs field. On the other hand, a junction of walls
between vacua labeled by I1 = ab · · ·, I2 = bc · · ·, and I3 = ac · · · has only
r − 2 non-zero components of the Higgs field at the core and, therefore, carries U (2)
unbroken symmetry. This discussion easily generalizes to other types of walls and
junctions.
Appendix C. LG Interfaces and the cohomology of Grassmannians

The cohomology of the Grassmannian Gr(k, N ) can be described as the poly-
(k) (k)
nomial ring in the Chern classes c1 , . . . , ck of the tautological bundle and the
(k) (k)
Chern classes c1 , . . . , cN −k of the complementary bundle modulo the ideal Ik,N
generated by the relations
(1 + tc1 + . . . + tk ck )(1 + tc1 + . . . + tN −k cN −k ) = 1 .

(k) (k) (k) (k)
(C.1)
Similarly, the cohomology of the partial flag variety F l(k1 , k = k1 + k2 , N ) can be
realized as polynomial ring in c1 , . . . , ck1 , c1 , . . . , ck2 , c1 , . . . , cN −k modulo the ideal
Ik1 ,k,N generated by the relations
(C.2) (1 + tc1 + . . . + tk1 ck1 )(1 + tc1 + . . . + tk2 ck2 )(1 + tc + . . . + tN −k cN −k ) = 1 .

Now (3.9) maps
(C.3) H ∗ (Gr(k1 , N ) × Gr(k2 , N )) −→ H ∗ (F l(k1 , k, N ))
(k1 )
ci −→ ci

i
cj ci−j
(k1 )
ci −→
j=0
−→ ci
(k )
ci 2
(k )

i
ci 2 −→ cj ci−j
j=0
and
(C.4) H ∗ (Gr(k, N )) −→ H ∗ (F l(k1 , k, N ))

i
cj ci−j
(k)
ci −→
j=0
(k)
ci −→ ci .
(k) i (k1 ) (k2 )
Hence, the correspondence (3.9) identifies ci with j=0 cj ci−j .
Appendix D. LG Interfaces and 2-categories N Foam

The category N Foam can be described in terms of generators and relations
[18]. In this appendix we exemplary check that the realization we propose in terms
of surface operators satisfies two such relations.
D.1. Derivation of the bubble relation. To derive the bubble formula

(4.36), we make use of a correspondence between matrix factorizations of W over a
polynomial ring R and finitely generated modules over the respective hypersurface
ring R := R/(W ). Namely, free resolutions of the latter always turn 2-periodic after
finitely many steps, with the 2-periodic part given by a matrix factorization of W ,
[115]. Thus, questions about matrix factorizations can be turned into questions

about R-modules. This comes in handy in particular in the calculation of fusion of
interfaces in Landau-Ginzburg models [23].
The matrix factorization Ikk1 ,k2 for instance is related in this way to the module
(D.1) R{−k
M = R/J 1 k2 } ,
over the ring R = C[X1 , . . . , Xk , Y1 , . . . , Yk ]/(W (X1 , . . . , Xk ) − W (f1 , . . . , fk )),

where J is the ideal generated by (X1 − f1 (Y1 , . . . , Yk )), . . . , (Xk − fk (Y1 , . . . , Yk ).
Similarly Ikk1 ,k2 is related to the module
(D.2) M = R ,
/J R
over R = C[X , . . . , X , Y1 , . . . , Yk ]/(W (X , . . . , X ) − W (f1 , . . . , fk )), where the

1 k 1 k
ideal J is generated by (X1 − f1 (Y1 , . . . , Yk )), . . . , (Xk − fk (Y1 , . . . , Yk ). Here, the
Xi and Xi correspond to the fields of the incoming, respectively outgoing models
LGk , whereas the Yi are the fields of the intermediate model LGk1 ⊗ LGk2 .
The matrix factorization of the fusion product Ikk1 ,k2 ∗Ikk1 ,k2 is now given by the
2-periodic part of the free resolution of the module M := M ⊗ M considered as a
module over R := C[X1 , . . . , Xk , X1 , . . . , X ]/(W (X1 , . . . , Xk ) − W (X1 , . . . , X )).
k k
But
(D.3)
M ∼
=R ⊗ (C[Y1 , . . . , Yk ]/((Xi − fi (Y1 , . . . , Yk )), (Xi − fi (Y1 , . . . , Yk ))) {−k1 k2 }

∼
= R /(Xi − Xi ) ⊗ (C[Y1 , . . . , Yk ]/(fi (Y1 , . . . , Yk ))) {−k1 k2 } .
Now, the Yi are obtained from (x1 , . . . , xk ) by partial symmetrization with respect
to permutations of respectively the first k1 and the remaining k2 variables, and the
fi = σi (x1 , . . . , xk ) are a basis of the completely symmetrized (x1 , . . . , xk ). Hence,
(D.4) (C[Y1 , . . . , Yk ]/(fi (Y1 , . . . , Yk ))) {−k1 k2 }

∼
= C[x1 , . . . , xk ]Sk1 ×Sk2 /C[x1 , . . . , xk ]Sk {−k1 k2 } .
∼
= H ∗ (Gr(k1 , k)){−k1 k2 }
+ ,
∼ k
=C .
k1
The relation (4.36) follows from the fact that the 2-periodic part of the Koszul
resolution of

(D.5) R /(Xi − Xi )
is the matrix factorization corresponding to the identity defect Ikk of LGk .

Figure 33. Interface configurations associated to FE (left) and

EF (right).
D.2. Quantum group relations. Here, we briefly explain how to obtain

(4.39). Pictorially, the left hand side of this equation is given by the configuration of
interfaces depicted on the left of Figure 33. The blue symbols specify the Landau-
Ginzburg fields of the respective segments of surface operators. Let R be the
polynomial ring in the external variables (Xi , Xi , Yi , Yi ) and
(D.6) = R/(W (X1 , . . . , Xk ) + W (Y1 , . . . , Yk )

R 1 2
−W (X1 , . . . , Xk1 ) − W (Y1 , . . . , Yk2 ))

the respective hypersurface ring. The R-module corresponding to the matrix fac-
torization on the left hand side of (4.39) is then given by
(D.7) 1 /J R
M =R 1 {−k1 − k2 + 1} ,
where
(D.8) 1 = R
R [A, B, U1 , . . . , Uk −1 , V1 , . . . , Vk +1 ]
1 2
and J is the
is obtained by associating the internal variables (A, B, Ui , Vi ) to R,

ideal in R1 generated by the relations
(D.9) Xi = σi (A, U1 , . . . , Uk1 −1 )

(D.10) Xi = σi (B, U1 , . . . , Uk1 −1 )
(D.11) Vi = σi (A, Y1 , . . . , Yk2 ) = σi (B, Y1 , . . . , Yk2 ) .
The module (D.7) can be recast in the following way. The variables Vi can be
eliminated by (D.11). Introducing the new variables
(D.12) V i := Yi − BV i−1 , V0 = 1 ,
the second part of the relations (D.11) can be rephrased as
(D.13) Yi = σi (B, V 1 , . . . , V k2 ) ,
(D.14) Yi = σi (A, V 1 , . . . , V k2 ) ,
(D.15) (A − B)V k2 = 0.
Thus, we have traded the variables V1 , . . . , Vk2 +1 for variables V 1 , . . . , V k2 and

relations (D.11) for (D.13), (D.14) and (D.15). Defining

(D.16) 2 = R
R A, B, U1 , . . . , Uk −1 , V 1 , . . . , V k −1
1 2
one obtains
(D.17) M∼
=R 2
2 [Vk ] {−k1 − k2 + 1} ,
2 [Vk ]/J R
2
where J is defined by the relations (D.9), (D.10), (D.13), (D.14) and (D.15). Next,
one can use (D.15) to eleminate V k2 :
(D.18) M ∼
=R 2 /J R
2 {−k1 − k2 + 1} ⊕ R
2 /((A − B), J )R
2 {−k1 + k2 + 1} .
!" # !" #
=:M1 =:M2
2 generated by (D.9), (D.10), (D.13) and (D.14), where

Here J is the ideal in R
V k2 is set to zero. With A − B = 0, these equations can be solved on M2 . One
obtains Xi = Xi and Yi = Yi , and (D.9) can be used to eliminate the Ui . Only
the identity with i = k1 in (D.9) remains and gives rise to the degree k1 relation
k1 i
i=0 Xk1 −i (−A) = 0, which truncates the possible exponents of A (X0 := 1).
Thus,
- .
(D.19) M2 ∼ 2R
= R/J q −k1 −k2 +1 + q −k1 −k2 +3 + . . . + q k1 −k2 −1 ,
generated by the relations

where J2 is the ideal in R
(D.20) 0 = (Xi − Xi ) = (Yi − Yi ) .
Now, M1 is symmetric in the roles the variables (Xi , Xi ) and (Yi , Yi ) play. In par-
ticular, it also appears in the module M corresponding to the matrix factorization
on the right hand side of equation (4.39), represented on the right of Figure 33.
Indeed,
(D.21) M ∼
= M1 ⊕ M2 ,
with
- .
(D.22) M2 ∼ 2R
= R/J q −k1 −k2 +1 + q −k1 −k2 +3 + . . . + q −k1 +k2 −1 .
Therefore for k1 ≥ k2
/ 0
(D.23) M∼ 2R
= M ⊕ R/J q −k1 −+k2 +1 + q −k1 −k2 +3 + . . . + q k1 −k2 −1 .
!" #
[k1 −k2 ]
The module R/J 2R now corresponds to the identity defect in LGk ⊗ LGk , which
1 2
proves that upon fusion the configurations of interfaces E and F satisfy the quantum
sl2 -relations (4.39).
References
[1] G. Moore and N. Reshetikhin, A comment on quantum group symmetry in conformal field
theory, Nuclear Phys. B 328 (1989), no. 3, 557–574, DOI 10.1016/0550-3213(89)90219-8.
MR1030450
[2] S. Cautis, J. Kamnitzer, and A. Licata, Categorical geometric skew Howe duality, Invent.
Math. 180 (2010), no. 1, 111–159, DOI 10.1007/s00222-009-0227-1. MR2593278
[3] S. Cautis, J. Kamnitzer, and S. Morrison, Webs and quantum skew Howe duality, Math.
Ann. 360 (2014), no. 1-2, 351–390, DOI 10.1007/s00208-013-0984-4. MR3263166
[4] A. A. Beilinson, G. Lusztig, and R. MacPherson, A geometric setting for the quantum
deformation of GLn , Duke Math. J. 61 (1990), no. 2, 655–677, DOI
10.1215/S0012-7094-90-06124-1. MR1074310
[5] I. Grojnowski and G. Lusztig, On bases of irreducible representations of quantum GLn ,
Kazhdan-Lusztig theory and related topics (Chicago, IL, 1989), Contemp. Math., vol. 139,
Amer. Math. Soc., Providence, RI, 1992, pp. 167–174, DOI 10.1090/conm/139/1197834.
MR1197834
[6] J. Bernstein, I. Frenkel, and M. Khovanov, A categorification of the Temperley-Lieb
algebra and Schur quotients of U (sl2 ) via projective and Zuckerman functors, Selecta
Math. (N.S.) 5 (1999), no. 2, 199–241, DOI 10.1007/s000290050047. MR1714141
[7] I. Frenkel, M. Khovanov, and C. Stroppel, A categorification of finite-dimensional
irreducible representations of quantum sl2 and their tensor products, Selecta Math. (N.S.)
12 (2006), no. 3-4, 379–431, DOI 10.1007/s00029-007-0031-y. MR2305608
[8] H. Zheng, A geometric categorification of representations of Uq (sl2 ), Topology and
physics, Nankai Tracts Math., vol. 12, World Sci. Publ., Hackensack, NJ, 2008,
pp. 348–356, DOI 10.1142/9789812819116 0016. MR2503405
[9] A. D. Lauda, A categorification of quantum sl(2), Adv. Math. 225 (2010), no. 6,
3327–3424, DOI 10.1016/j.aim.2010.06.003. MR2729010
[10] M. Khovanov and A. D. Lauda, A diagrammatic approach to categorification of quantum
groups. I, Represent. Theory 13 (2009), 309–347, DOI 10.1090/S1088-4165-09-00346-X.
MR2525917
groups II, Trans. Amer. Math. Soc. 363 (2011), no. 5, 2685–2700, DOI
10.1090/S0002-9947-2010-05210-9. MR2763732
groups. III, Quantum Topol. 1 (2010), no. 1 1–92.
[13] M. Khovanov, Categorifications from planar diagrammatics, Jpn. J. Math. 5 (2010), no. 2,
153–181, DOI 10.1007/s11537-010-0925-x. MR2747932
[14] A. D. Lauda, An introduction to diagrammatic algebra and categorified quantum sl2 , Bull.
Inst. Math. Acad. Sin. (N.S.) 7 (2012), no. 2, 165–270. MR3024893
[15] J. Kamnitzer, Categorification of Lie algebras [d’apres Rouquier, Khovanov-Lauda],
Astérisque (2013) Exp. No. 1072. Séminaire Bourbaki. Volume 2012/2013. Exposés.
[16] R. Rouquier, 2-Kac-Moody algebras, arXiv:0812.5023.
[17] A. D. Lauda, H. Queffelec, and D. E. V. Rose, Khovanov homology is a skew Howe
2-representation of categorified quantum slm , Algebr. Geom. Topol. 15 (2015), no. 5,
2517–2608, DOI 10.2140/agt.2015.15.2517. MR3426687
[18] H. Queffelec and D. E. V. Rose, The sln foam 2-category: a combinatorial formulation of
Khovanov-Rozansky homology via categorical skew Howe duality, arXiv:1405.5920.
[19] S. Gukov and E. Witten, Gauge Theory, Ramification, And The Geometric Langlands
Program, hep-th/0612073.
[20] S. Gukov, Gauge theory and knot homologies, Fortschr. Phys. 55 (2007), no. 5-7, 473–490,
DOI 10.1002/prop.200610385. MR2331918
[21] S. Gukov and E. Witten, Rigid surface operators, Adv. Theor. Math. Phys. 14 (2010),
no. 1, 87–177. MR2684979
[22] S. Gukov and E. Witten, Branes and quantization, Adv. Theor. Math. Phys. 13 (2009),
no. 5, 1445–1518. MR2672467
[23] I. Brunner and D. Roggenkamp, B-type defects in Landau-Ginzburg models, J. High Energy
Phys. 8 (2007), 093, 46 pp. (electronic), DOI 10.1088/1126-6708/2007/08/093. MR2342020
[24] G. Lusztig, Introduction to quantum groups, Progress in Mathematics, vol. 110, Birkhäuser
[25] E. Witten, Quantum field theory and the Jones polynomial, Comm. Math. Phys. 121
(1989), no. 3, 351–399. MR990772
[26] E. Witten, Gauge theories and integrable lattice models, Nuclear Phys. B 322 (1989),
no. 3, 629–697, DOI 10.1016/0550-3213(89)90232-0. MR1010205
[27] E. Witten, Gauge theories, vertex models, and quantum groups, Nuclear Phys. B 330
(1990), no. 2-3, 285–346, DOI 10.1016/0550-3213(90)90115-T. MR1043385
[28] H. Wu, A colored sl(N ) homology for links in S 3 , Dissertationes Math. (Rozprawy Mat.)
499 (2014), 217, DOI 10.4064/dm499-0-1. MR3234803
[29] L. Crane and I. B. Frenkel, Four-dimensional topological quantum field theory, Hopf
categories, and the canonical bases, J. Math. Phys. 35 (1994), no. 10, 5136–5154, DOI
10.1063/1.530746. Topology and physics. MR1295461
[30] D. Bar-Natan, Khovanov’s homology for tangles and cobordisms, Geom. Topol. 9 (2005),
1443–1499, DOI 10.2140/gt.2005.9.1443. MR2174270
[31] D. Bar-Natan and S. Morrison, The Karoubi envelope and Lee’s degeneration of Khovanov
homology, Algebr. Geom. Topol. 6 (2006), 1459–1469, DOI 10.2140/agt.2006.6.1459.
MR2253455
[32] M. Mackaay and P. Vaz, The universal sl3 -link homology, Algebr. Geom. Topol. 7 (2007),
1135–1169, DOI 10.2140/agt.2007.7.1135. MR2336253
[33] E. S. Lee, An endomorphism of the Khovanov invariant, Adv. Math. 197 (2005), no. 2,
554–586, DOI 10.1016/j.aim.2004.10.015. MR2173845
[34] M. Khovanov, Link homology and Frobenius extensions, Fund. Math. 190 (2006), 179–190,
DOI 10.4064/fm190-0-6. MR2232858
[35] B. Gornik, Note on Khovanov link cohomology, math/0402266.
[36] H. Wu, Generic deformations of the colored sl(N )-homology for links, Algebr. Geom.
Topol. 11 (2011), no. 4, 2037–2106, DOI 10.2140/agt.2011.11.2037. MR2826932
[37] A. Lobb, A note on Gornik’s perturbation of Khovanov-Rozansky homology, Algebr.
Geom. Topol. 12 (2012), no. 1, 293–305, DOI 10.2140/agt.2012.12.293. MR2916277
[38] D. E. V. Rose and P. Wedrich, Deformations of colored sl(N) link homologies via foams,
arXiv:1501.0256.
[39] E. Gorsky, S. Gukov, and M. Stosic, Quadruply-graded colored homology of knots,
arXiv:1304.3481.
[40] M. Khovanov, A categorification of the Jones polynomial, Duke Math. J. 101 (2000),
no. 3, 359–426, DOI 10.1215/S0012-7094-00-10131-7. MR1740682
[41] J. Rasmussen, Khovanov homology and the slice genus, Invent. Math. 182 (2010), no. 2,
419–447, DOI 10.1007/s00222-010-0275-6. MR2729272
[42] S. Gukov, Surface operators, New dualities of sypersymmetric gauge theories, Math. Phys.
Stud., Springer, Cham, 2016, pp. 223–259. MR3467294
[43] M. G. Alford, K.-M. Lee, J. March-Russell, and J. Preskill, Quantum field theory of
nonabelian strings and vortices, Nuclear Phys. B 384 (1992), no. 1-2, 251–317, DOI
10.1016/0550-3213(92)90468-Q. MR1186591
[44] L. Rozansky, Topological A-models on seamed Riemann surfaces, Adv. Theor. Math.
Phys. 11 (2007), no. 4, 517–529. MR2354074
[45] M. Khovanov and L. Rozansky, Topological Landau-Ginzburg models on the world-sheet
foam, Adv. Theor. Math. Phys. 11 (2007), no. 2, 233–259. MR2322554
[46] W. Lerche, C. Vafa, and N. P. Warner, Chiral rings in N = 2 superconformal theories,
Nuclear Phys. B 324 (1989), no. 2, 427–474, DOI 10.1016/0550-3213(89)90474-4.
MR1025424
[47] E. Witten, Phases of N = 2 theories in two dimensions, Nuclear Phys. B 403 (1993),
no. 1-2, 159–222, DOI 10.1016/0550-3213(93)90033-L. MR1232617
[48] E. Witten, Solutions of four-dimensional field theories via M -theory, Nuclear Phys. B 500
(1997), no. 1-3, 3–42, DOI 10.1016/S0550-3213(97)00416-1. MR1471647
[49] A. Gadde, S. Gukov, and P. Putrov, Fivebranes and 4-manifolds, arXiv:1306.4320.
[50] D. Xie, Higher laminations, webs and N=2 line operators, arXiv:1304.2390.
[51] D. Xie, Aspects of line operators of class S theories, arXiv:1312.3371.
[52] M. Bullimore, Defect Networks and Supersymmetric Loop Operators, JHEP 1502 (2015)
066, [arXiv:1312.5001].
[53] I. Coman, M. Gabella, and J. Teschner, Line operators in theories of class S, quantized
moduli space of flat connections, and Toda field theory, J. High Energy Phys. 10 (2015),
143, front matter+88. MR3435496
[54] Y. Tachikawa and N. Watanabe, On skein relations in class S theories, J. High Energy
Phys. 6 (2015), 186, front matter+41. MR3370149
[55] E. Witten, Branes and the dynamics of QCD, Nuclear Phys. B 507 (1997), no. 3, 658–690,
DOI 10.1016/S0550-3213(97)00648-2. MR1491032
[56] E. Frenkel, S. Gukov, and J. Teschner, Surface operators and separation of variables, J.
High Energy Phys. 1 (2016), 179, front matter+52. MR3471401
[57] L. C. Jeffrey and J. Weitsman, Bohr-Sommerfeld orbits in the moduli space of flat
connections and the Verlinde dimension formula, Comm. Math. Phys. 150 (1992), no. 3,
593–630. MR1204322
[58] S. Agnihotri and C. Woodward, Eigenvalues of products of unitary matrices and quantum
Schubert calculus, Math. Res. Lett. 5 (1998), no. 6, 817–836, DOI
10.4310/MRL.1998.v5.n6.a10. MR1671192
[59] P. Belkale, Local systems on P1 − S for S a finite set, Compositio Math. 129 (2001), no. 1,
67–86, DOI 10.1023/A:1013195625868. MR1856023
[60] C. Teleman and C. Woodward, Parabolic bundles, products of conjugacy classes and
Gromov-Witten invariants (English, with English and French summaries), Ann. Inst.
Fourier (Grenoble) 53 (2003), no. 3, 713–748. MR2008438
[61] N. Ressayre, On the quantum Horn problem, arXiv:1310.7331.
[62] P. Belkale and S. Kumar, The multiplicative eigenvalue problem and deformed quantum
cohomology, Adv. Math. 288 (2016), 1309–1359, DOI 10.1016/j.aim.2015.09.034.
MR3436409
[63] A. Lobb and R. Zentner, The quantum sl(N ) graph invariant and a moduli space, Int.
Math. Res. Not. IMRN 7 (2014), 1956–1972. MR3190356
[64] J. Grant, The moduli problem of Lobb and Zentner and the colored sl(N ) graph invariant,
J. Knot Theory Ramifications 22 (2013), no. 10, 1350060, 16, DOI
10.1142/S0218216513500600. MR3125899
[65] A. Bertram, Quantum Schubert calculus, Adv. Math. 128 (1997), no. 2, 289–305, DOI
10.1006/aima.1997.1627. MR1454400
[66] E. Witten, The Verlinde algebra and the cohomology of the Grassmannian,
hep-th/9312104.
[67] S. Gukov and D. Pei, Equivariant Verlinde formula from fivebranes and vortices,
arXiv:1501.0131.
[68] L. Mihalcea, Equivariant quantum Schubert calculus, Adv. Math. 203 (2006), no. 1, 1–33,
DOI 10.1016/j.aim.2005.04.002. MR2231042
[69] M. Eto, Y. Isozumi, M. Nitta, K. Ohashi, K. Ohta, and N. Sakai, D-brane configurations
for domain walls and their webs, Particles, strings and cosmology, AIP Conf. Proc.,
vol. 805, Amer. Inst. Phys., Melville, NY, 2005, pp. 354–357, DOI 10.1063/1.2149729.
MR2222621
[70] M. Eto, T. Fujimori, Y. Isozumi, M. Nitta, K. Ohashi, et. al., Non-Abelian vortices on
cylinder: Duality between vortices and walls, Phys.Rev. D73 (2006) 085008,
[hep-th/0601181].
[71] M. Eto, T. Fujimori, M. Nitta, K. Ohashi, and N. Sakai, Domain walls with non-Abelian
clouds, Phys.Rev. D77 (2008) 125008, [arXiv:0802.3135].
[72] H. Queffelec and A. Sartori, Mixed quantum skew Howe duality and link invariants of type
A, arXiv:1504.0122.
[73] D. E. V. Rose and D. Tubbenhauer, Symmetric webs, Jones-Wenzl recursions and q-Howe
duality, arXiv:1501.0091.
[74] P. Ozsváth and Z. Szabó, Holomorphic disks and knot invariants, Adv. Math. 186 (2004),
no. 1, 58–116, DOI 10.1016/j.aim.2003.05.001. MR2065507
[75] J. Rasmussen, Floer homology and knot complements, math/0306378.
[76] S. Gukov, A. Schwarz, and C. Vafa, Khovanov-Rozansky homology and topological strings,
Lett. Math. Phys. 74 (2005), no. 1, 53–74, DOI 10.1007/s11005-005-0008-8. MR2193547
[77] R. Gopakumar and C. Vafa, M-Theory and Topological Strings-I, hep-th/9809187v1.
[78] R. Gopakumar and C. Vafa, M-Theory and Topological Strings-II, hep-th/9812127v1.
[79] M. Aganagic and S. Shakirov, Knot Homology from Refined Chern-Simons Theory,
arXiv:1105.5117.
[80] A. Iqbal and C. Kozçaz, Refined Hopf link revisited, J. High Energy Phys. 4 (2012), 046,
front matter+18. MR2967625
[81] I. Cherednik, Jones polynomials of torus knots via DAHA, Int. Math. Res. Not. IMRN 23
(2013), 5366–5425. MR3142259
[82] H. Nakajima, Refined Chern-Simons theory and Hilbert schemes of points on the plane,
Perspectives in representation theory, Contemp. Math., vol. 610, Amer. Math. Soc.,
Providence, RI, 2014, pp. 305–331, DOI 10.1090/conm/610/12157. MR3220632
[83] P. Samuelson, A topological construction of Cherednik’s sl 2 torus knot polynomials,
arXiv:1408.0483.
[84] E. Witten, Fivebranes and knots, Quantum Topol. 3 (2012), no. 1, 1–137, DOI
10.4171/QT/26. MR2852941
[85] A. Haydys, Fukaya-Seidel category and gauge theory, J. Symplectic Geom. 13 (2015),
no. 1, 151–207, DOI 10.4310/JSG.2015.v13.n1.a5. MR3338833
[86] S. Gukov and J. Walcher, Matrix factorizations and Kauffman homology, hep-th/0512298.
[87] N. Carqueville and D. Murfet, Computing Khovanov-Rozansky homology and defect
fusion, Algebr. Geom. Topol. 14 (2014), no. 1, 489–537, DOI 10.2140/agt.2014.14.489.
MR3183384
[88] H. Fuji, S. Gukov, M. Stosic, and P. Sulkowski, 3d analogs of Argyres-Douglas theories and
knot homologies, JHEP 1301 (2013) 175, [arXiv:1209.1416].
[89] H.-J. Chung, T. Dimofte, S. Gukov, and P. Sulkowski, 3d-3d Correspondence Revisited,
arXiv:1405.3663.
[90] S. Katz, A. Klemm, and C. Vafa, Geometric engineering of quantum field theories, Nuclear
Phys. B 497 (1997), no. 1-2, 173–195, DOI 10.1016/S0550-3213(97)00282-4. MR1467889
[91] T. Dimofte, S. Gukov, and L. Hollands, Vortex counting and Lagrangian 3-manifolds,
Lett. Math. Phys. 98 (2011), no. 3, 225–287, DOI 10.1007/s11005-011-0531-8. MR2852983
[92] M. Khovanov and L. Rozansky, Matrix factorizations and link homology. II, Geom. Topol.
12 (2008), no. 3, 1387–1425, DOI 10.2140/gt.2008.12.1387. MR2421131
[93] M. Khovanov, sl(3) link homology, Algebr. Geom. Topol. 4 (2004), 1045–1081, DOI
10.2140/agt.2004.4.1045. MR2100691
[94] M. Aganagic, T. Ekholm, L. Ng, and C. Vafa, Topological strings, D-model, and knot
contact homology, Adv. Theor. Math. Phys. 18 (2014), no. 4, 827–956. MR3277674
[95] H. Fuji and P. Sulkowski, Super-A-polynomial, String-Math 2012, Proc. Sympos. Pure
Math., vol. 90, Amer. Math. Soc., Providence, RI, 2015, pp. 277–303. MR3409800
[96] M. Khovanov and L. Rozansky, Matrix factorizations and link homology, Fund. Math. 199
(2008), no. 1, 1–91, DOI 10.4064/fm199-1-1. MR2391017
[97] I. Brunner, M. Herbst, W. Lerche, and B. Scheuner, Landau-Ginzburg realization of open
string TFT, J. High Energy Phys. 11 (2006), 043, 28 pp. (electronic), DOI
10.1088/1126-6708/2006/11/043. MR2270412
[98] A. Kapustin and Y. Li, Topological correlators in Landau-Ginzburg models with
boundaries, Adv. Theor. Math. Phys. 7 (2003), no. 4, 727–749. MR2039036
[99] K. Hori and J. Walcher, F -term equations near Gepner points, J. High Energy Phys. 1
(2005), 008, 23 pp. (electronic), DOI 10.1088/1126-6708/2005/01/008. MR2134919
[100] I. Brunner and D. Roggenkamp, Defects and bulk perturbations of boundary
Landau-Ginzburg orbifolds, J. High Energy Phys. 4 (2008), 001, 35, DOI
10.1088/1126-6708/2008/04/001. MR2425302
[101] I. Brunner, H. Jockers, and D. Roggenkamp, Defects and D-brane monodromies, Adv.
Theor. Math. Phys. 13 (2009), no. 4, 1077–1135. MR2661203
[102] S. Gukov and M. Stosić, Homological algebra of knots and BPS states, String-Math 2011,
Proc. Sympos. Pure Math., vol. 85, Amer. Math. Soc., Providence, RI, 2012, pp. 125–171,
DOI 10.1090/pspum/085/1377. MR2985329
[103] M. Khovanov and L. Rozansky, Virtual crossings, convolutions and a categorification of
the SO(2N ) Kauffman polynomial, J. Gökova Geom. Topol. GGT 1 (2007), 116–214.
MR2386537
[104] Y. Yonezawa, Quantum (sln , ∧Vn ) link invariant and matrix factorizations, Nagoya Math.
J. 204 (2011), 69–123. MR2863366
[105] R. Elliot and S. Gukov, Exceptional knot homology, J. Knot Theory Ramifications 25
(2016), no. 3, 1640003, 49, DOI 10.1142/S0218216516400034. MR3475070
[106] N. M. Dunfield, S. Gukov, and J. Rasmussen, The Superpolynomial for knot homologies,
math/0505662.
[107] S. Cecotti and C. Vafa, On classification of N = 2 supersymmetric theories, Comm. Math.
Phys. 158 (1993), no. 3, 569–644. MR1255428
[108] A. Gadde and S. Gukov, 2d index and surface operators, J. High Energy Phys. 3 (2014),
080, front matter+40. MR3190966
[109] T. Dyckerhoff and D. Murfet, Pushing forward matrix factorisations, arXiv:1102.2957.
[110] N. Behr and S. Fredenhagen, Variable transformation defects, String-Math 2011, Proc.
Sympos. Pure Math., vol. 85, Amer. Math. Soc., Providence, RI, 2012, pp. 303–312, DOI
10.1090/pspum/085/1386. MR2985337
[111] M. Mackaay and Y. Yonezawa, sl(N)-Web categories, arXiv:1306.6242.
[112] M. Eto, Y. Isozumi, M. Nitta, K. Ohashi, and N. Sakai, Non-abelian webs of walls, Phys.
Lett. B 632 (2006), no. 2-3, 384–392, DOI 10.1016/j.physletb.2005.10.017. MR2189503
[113] M. Shifman and A. Yung, Localization of non-abelian gauge fields on domain walls at
weak coupling: D-brane prototypes, Phys. Rev. D (3) 70 (2004), no. 2, 025013, 28, DOI
10.1103/PhysRevD.70.025013. MR2103282
[114] M. Eto, Y. Isozumi, M. Nitta, K. Ohashi, and N. Sakai, Webs of walls, Phys.Rev. D72
(2005) 085004, [hep-th/0506135].
[115] D. Eisenbud, Homological algebra on a complete intersection, with an application to group
representations, Trans. Amer. Math. Soc. 260 (1980), no. 1, 35–64, DOI 10.2307/1999875.
MR570778
Walter Burke Institute for Theoretical Physics, California Institute of Technol-

ogy, Pasadena, California 91125
Walter Burke Institute for Theoretical Physics, California Institute of Technol-

ogy, Pasadena, California 91125
Institut für Theoretische Physik, Universität Heidelberg, Philosophenweg 19,

69120 Heidelberg, Germany
Volume 684, 2017
Khovanov-Rozansky homology and 2-braid groups
Raphaël Rouquier
1. Introduction
Khovanov [Kh] has given a construction of the Khovanov-Rozansky link invari-
ants (categorifying the HOMFLYPT polynomials) using Hochschild cohomology of
2-braid groups. We give a direct proof that his construction does give link invari-
ants. We show more generally that, for any finite Coxeter group, his construction
provides a Markov “2-trace”, and we actually show that the invariant takes value
in suitable derived categories. This makes more precise a result of Trafim Lasy
who has shown that, after taking the class in K0 , this provides a Markov trace
[La1, La2]. It coincides with Gomi’s trace [Go] for Weyl groups (Webster and
Williamson [WeWi]) as well as for dihedral groups [La1].
In the first section, we recall the construction of 2-braid groups [Rou1], based
on complexes of Soergel bimodules. The second section is devoted to Markov traces,
and a category-valued version, 2-Markov traces. We provide a construction using
Hochschild cohomology. The third section is devoted to the proof of the Markov
property for Hochschild cohomology.
2. Notations
Let k be a commutative ring. We write ⊗ for ⊗k . Let A be a k-algebra. We
denote by Aopp the opposite algebra to A and we put Aen = A ⊗ Aopp .
We denote by A-Mod the category of A-modules, by A-mod the category of
finitely generated A-modules, by A-Proj the category of projective A-modules and
by A-proj the category of finitely generated projective A-modules. Assume A is
graded. We denote by A-modgr (resp. A-projgr) the category of finitely generated
(resp. and projective) graded A-modules.
Given M a graded k-module and n ∈ Z, we denote by M n the graded k-
module given by M ni = Mn+i .
Given A an additive category, we denote by Comp(A) (resp. Ho(A)) the cate-
gory (resp. the homotopy category) of complexes of objects of A. If A is an abelian
category, we denote by D(A) its derived category.
Given C a category, we denote by {1} the self equivalence of C Z given by
(M {1})i = Mi+1 .
2010 Mathematics Subject Classification. Primary 57M27, 17B10, 18D10, 20F36.

The author was partly supported by the NSF (grant DMS-1161999).
2017
147
148 RAPHAËL ROUQUIER
Let T be a triangulated category equipped with an automorphism M → M 1.

We denote by q the automorphism of K0 (T ) given by [M ] → [M 1]. This endows
K0 (T ) with a structure of Z[q, q −1 ]-module.
3. 2-braid groups
3.1. Braid groups. Let (W, S) be a finite Coxeter group. Let V be the
geometric representation of W over k = C: it comes with a basis {es }s∈S . Given
s ∈ S, we denote by αs the linear form on V such that s(x) − x = αs (x)es for all
x ∈ V . The set {αs }s∈S is a basis of V ∗ . Let P = PS = P(W,S) = k[V ] (we will
denote by XS or X(W,S) a given object constructed from (W, S)).
The braid group BS = B(W,S) associated to (W, S) is the group generated by
{σs }s∈S with relations
σs σt σs · · · = σt σs σt · · ·
!" # !" #
mst terms mst terms
for any s, t ∈ S such that the order mst of st is finite.

We denote by l : BS → Z the length function. It is the morphism of groups
defined by l(σs ) = 1 for s ∈ S.
3.2. Lift. Let us recall, following [Rou1], how to lift in a non-trivial way the
action of W on the derived category D(P ) to an action of BS on the homotopy
category Ho(P ).
Let s ∈ S. We put
m
θs = P ⊗P s P and Fs = 0 → θs 1 −→ P 1 → 0.
The latter is a complex of graded P en -modules, where P 1 is in cohomological

degree 1 and m denotes the multiplication map. We put
a→aα ⊗1+a⊗α
Fs−1 = 0 → P −1 −−−−−s−−−−−−→
s
θs → 0.
This is a complex of graded P en -modules, where P −1 is in cohomological degree

−1.
Let us recall a result of [Rou1, §9]. Given i1 , . . . , ir , j1 , . . . , jr ∈ S and
ε
δ1 , . . . , δr , ε1 , . . . , εr ∈ {±1} such that σiδ11 · · · σiδrr = σjε11 · · · σjrr , there is a canonical
isomorphism in Ho(P en -modgr)
∼ ε
Fiδ11 ⊗Pn · · · ⊗Pn Fiδrr → Fjε11 ⊗Pn · · · ⊗Pn Fjrr
and these isomorphisms form a transitive system of isomorphisms.

Given b ∈ BS , we put
Fb = lim Fiε11 ⊗P · · · ⊗P Fiεrr ∈ Ho(P en -modgr).

i1 ,...,ir
ε1 ,...,εr
ε
b=σi 1 ···σiεrr
1
The 2-braid group B(W,S) is the full monoidal subcategory of Ho(P en -modgr)
with objects the Fb ’s, with b ∈ BS .
KHOVANOV-ROZANSKY HOMOLOGY AND 2-BRAID GROUPS 149
3.3. Parabolic subgroups. 1 Let I ⊂ S and let WI be the subgroup of W

s and PI = k[VI ]. We have V = VI ⊕ V ,
I
generated by I. Let VI = s∈I ke
hence P = PI ⊗ k[V I ]. We deduce also that V ∗ = (V I )⊥ ⊕ (VI )⊥ , hence the
composition of canonical
1 maps (V I )⊥ → V ∗ (VI )∗ is an isomorphism. We
identify (V ) = s∈I kαs and (VI )∗ via this isomorphism.
I ⊥
2
The compositions of canonical maps s ∈I ker αs → V → V /V I and VI →
∼ 2
V → V /V I are isomorphisms: this provides an isomorphism VI → s ∈I ker αs . We
∼ 2
denote by ρI : P PI the morphism given by the composition VI → s ∈I ker αs →
V.
We have a functor γI : PIen -Mod → P en -Mod sending M to k[V I ] ⊗ M , where
k[V ] is the regular k[V I ]en -module and P en is decomposed as P en = k[V I ]en ⊗PIen .
I
We obtain a fully faithful monoidal functor

BWI → BW , F → γI (F ) = k[V I ] ⊗ F.
4. Hochschild cohomology and traces

4.1. Markov traces and 2-traces.
4.1.1. Markov traces. Let Cox be the poset of finite Coxeter groups, viewed as
a category. The objects are Coxeter groups (W, S) and Hom((W, S), (W , S )) is
the set of injective maps f : S → S such that mf (s),f (t) = mst for all s, t ∈ S.
Given s ∈ S, we denote by is : (WS\s , S \ s) → (W, S) the inclusion.
Let F be a full subposet of Cox closed below.
Let H(W,S) = Z[q ±1 ]B(W,S) /((Ts −1)(Ts +q))s∈S be the Hecke algebra of (W, S).
Definition 4.1. Let R be a Z[t− , t+ , q ±1 ]-module. A Markov trace on F is the
data of a family of Z[q ±1 ]-linear maps τ(W,S) : H(W,S) → R for (W, S) ∈ F such
that
• τS (hh ) = τS (h h) for h, h ∈ HS
• τS (hTs±1 ) = t± τS\s (h) for all s ∈ S and h ∈ HS\s .
Markov traces, with a possibly more general definition, have been studied by
Jones and Ocneanu in type A [Jo], Geck-Lambropoulou in type B [GeLa], Geck
in type D [Ge], and Kihara in type I2 (n) [Ki]. Gomi has provided a general con-
struction of Markov traces for Weyl groups, using Lusztig’s Fourier transform [Go].
More recently, Lasy has studied Markov traces in relation with Gomi’s definition
and Soergel bimodules [La1].
4.1.2. Markov 2-traces.
Definition 4.2. Let C : F → Cat be a functor.
A Markov 2-trace on F (relative to C) is the data of functors M(W,S) : B(W,S) →
C(W,S) such that the following holds
• MS (?1 ·?2 ) MS (?2 ·?1 ) as functors BS × BS → CS
• MS (γS\s (?) · Fs±1 ) TS,s,± C(is )MS\s (?) as functors BS\s → CS , for some
endofunctors TS,s,± of CS , for all s ∈ S.
One can ask in addition that the functors TS,s,± are invertible. On the other
hand, one can get a more general definition by dropping the functoriality of C and by
requiring the existence of functors DS,s,± : CS\s → CS such that MS (γS\s (?)·Fs±1 )
DS,s,± MS\s (?).
Remark 4.3. Let C¯ = colim C and assume there are endofunctors T± of C¯ which
¯ one can construct
restrict to TS,s,± for any S and s ∈ S. Replacing C(W,S) by C,
from a Markov 2-trace another one taking value in the constant category C, ¯ and
with fixed endofunctors T± .
The first “trace” condition, once formulated in the appropriate homotopical
setting, leads to a universal solution (“abelianization”) that can be described ex-
plicitly [BeNa]. It would be interesting to find a suitable formulation of the second
condition in this setting leading to a universal solution. It would also be interest-
ing to study functoriality with respect to cobordism in type A, along the lines of
[KhTh] and [ElKr].
4.1.3. From Markov 2-traces to Markov traces. Let Soe be the category of
Soergel bimodules: this is the full subcategory of P en -modgr whose objects are
direct summands of direct sums of objects of the form θs1 · · · θsn r, for some
s1 , . . . , sn ∈ S and r ∈ Z.
There is a Z[q ±1 ]-algebra morphism H → K0 (Soe) given by Ts → [Fs ], and
that morphism is actually an isomorphism (cf [Soe, Theorem 1] and [Li, Théorème
2.4]).
We consider now a Markov 2-trace in the following setting. Assume the functor
C takes values in graded triangulated categories and M(W,S) is the restriction of a
graded triangulated functor M(W,S) : Hob (Soe(W,S) ) → C(W,S) . In particular, it in-
duces a Z[q ±1 ]-linear map H(W,S) → K0 (C(W,S) ). Let R = colim(W,S)∈F K0 (C(W,S) ),
a Z[q ±1 ]-module. Assume there are commuting endomorphisms t± of R compatible
with the action of [TS,s,± ] on K0 (C(W,S) ), for all (W, S) ∈ F and s ∈ S, via the
canonical maps ι(W,S) : K0 (C(W,S) ) → R.
Define τ(W,S) : B(W,S) → R by τ (b) = ιS ([MS (Fb )]). We have the following
immediate proposition.
Proposition 4.4. The maps τ(W,S) come uniquely from Z[q ±1 ]-linear maps
H(W,S) → R. They define a Markov trace on F.
4.2. Hochschild homology.
(W,S) en
4.2.1. Main Theorem. We put HHi = HHi = TorP
i (P, −) : P en -modgr →
P -modgr. This gives rise to functors HHi : Ho (P -modgr) → Hob (P -modgr) and
b en
to a functor HH∗ : Hob (P en -modgr) → Hob (P -modgr)Z .

Given I ⊂ S, we have a functor ρ∗I : Db (PI -modgr)Z → Db (P -modgr)Z . This
defines a functor from Cox to graded triangulated categories
(W, S) → Db (PS -modgr)Z
. Our grading here is the one coming from PS -modgr.
The following theorem is a consequence of Theorem 5.1 below.
Theorem 4.5. The functors HHS∗ define a Markov 2-trace on finite Cox-
eter groups with value in CS = Db (PS -modgr)Z and with TS,s,+ = [−1]{−1} and
TS,s,− = Id.
Passing to homology, we obtain the following result.
Corollary 4.6. The functors H ∗ HHS∗ define a Markov 2-trace on finite Cox-

Z
eter groups with value in CS = (PS -modgr)Z and with TS,s,+ = [−1]{−1} and
TS,s,− = Id.
The construction of §4.1.3 provides a Markov trace, recovering a result of Lasy

[La1, La2]. Note that Webster and Williamson [WeWi] have shown that for finite
Weyl groups, this is actually Gomi’s trace, as conjectured by J. Michel. That has
been shown to hold also in type I2 (n) by Lasy [La1].
Corollary 4.7 (Lasy). The following defines a Markov trace on finite Coxeter
groups:

B(W,S) & b → (−1)j dim H j (HHSi (Fb ))d q −d t−i ∈ Z[q ±1 , t±1 ]
d,i,j
corresponding to t+ = −t and t− = 1.
4.2.2. Shift adjustment. By shifting suitably the invariants, we can get rid of
the automorphisms TS,s,± , but we lose functoriality (it would be interesting to
see if functoriality with respect to an appropriate notion of cobordisms can be
implemented). In order to do this, we need to use 12 Z-complexes.
Given A an additive category, the category of 12 -complexes in A has objects
(C , d )i∈ 12 Z where the differential has degree 1, and morphisms are 12 Z-graded maps
i i
commuting with the differential. Its homotopy category is denoted by Ho 12 (A) and,
when A is an abelian category, its derived category by D 12 (A).
Corollary 4.8. Given (W, S) a finite Coxeter group and b ∈ BS , let

|S| + l(b) 1
S
NS (Fb ) = HH∗+ |S|+l(b) (Fb ) ∈ Db1 (PS -modgr) 2 Z .
2 2 2
• We have NS (Fb Fb ) NS (Fb Fb ) for all b, b ∈ BS .

• Given s ∈ S and b ∈ BS\s , we have NS (γS\s (Fb )Fs±1 ) ρ∗S\s NS\s (Fb ).
4.3. Khovanov-Rozansky homology of links. We specialize now to the
case of the classical Artin braid groups considered by Khovanov in [Kh]. Note that
Khovanov conjectured fifteen years ago that the Fb ’s should give rise to interesting
link invariants. 1n
We take here V = ( i=1 kei )/k(e1 + · · · en ), the reflection representation of
W = Sn , with S = {(1, 2), . . . , (n − 1, n)}. Let Pn = k[V ] = k[α1 , . . . , αn−1 ], where
αi = Xi+1 − Xi . We put Bn = B(W,S) .
Let P∞ = limn Pn , where the limit is taken over the morphisms of Pn -algebras
ρn : Pn+1 → Pn , αn → 0. This provides functors between derived categories
· · · → Db (Pn -modgr) → Db (Pn+1 -modgr) → · · · → Db (P∞ -modgr).
Theorem 4.9. The assignment to b ∈ Bn+1 of the isomorphism class of
1
HH∗+ n+l(b) (Fb )[ n+l(b)
2 ] in Db1 (P∞ -modgr) 2 Z defines an invariant of oriented links.
2 2
Remark 4.10. Note that Webster gives in [We, §3] a related treatment of
Khovanov-Rozansky homology.
Passing to homology, we recover the following result of Khovanov [Kh]. Kho-
vanov identifies the invariant as the Khovanov-Rozansky homology, a categorifica-
tion of the HOMFLYPT polynomial.
Theorem 4.11 (Khovanov). The assignment to b ∈ Bn+1 of
±1/2 ±1/2
Xb = (t2 t3 )−(n+l(b))/2 dim H j (HHi (Fb ))d td1 ti2 tj3 ∈ N[t2 , t3 ][[t1 ]][t−1
1 ]
d,i,j
defines an invariant of oriented links.

Note that X1 = 1, where 1 ∈ B1 corresponds to the trivial knot.

We define now a two variables invariant Yb = (Xb )|t1/2 =√−1 . The following
3
corollary shows that Yb is the HOMFLYPT polynomial, as expected.
Corollary 4.12 (Khovanov). Given b, b ∈ Bn and r ∈ {1, . . . , n − 1}, we
have √ −1/2
(t1 t2 )−1/2 Ybσr−1 b + (t1 t2 )1/2 Ybσr b = −1(t1
1/2
− t1 )Ybb .
5. Proofs
5.1. Multiple complexes.
5.1.1. Total objects. For a more intrinsic approach to this section, cf [De, §1.1].
Let A be an additive category and n ≥ 0. We denote by Comp(A) the category
of complexes of objects of A. The category n-Comp(A) of n-fold complexes is
defined inductively by n-Comp(A) = Comp((n − 1)-Comp(A)) and 0-Comp(A) =
A. Its1objects are families (X, d1 , . . . , dn ) where X is an object of A graded by
Zn = i=1 Ze∗i , di is a graded map of degree ei and d2i = [di , dj ] = 0 for all i, j.
n
Given X an n-complex and i ∈ {1, . . . , n}, we define Y = X[ei ] as the n-complex

given by Y b = X ei +b and differentials ∂ib = (−1)δij djei +b .
Let f : {1, . . . , n} → {1, . . . , m} be a map. It induces a map σ : Zn → Zm and
gives by duality a map Zm → Zn . This provides a functor from Zn -graded objects
to Zm -graded objects of A. Let X be an n-complex. We have a corresponding
Zm -graded object X . We define a structure of m-complex by

da
i = (−1) k∈f −1 (i),k<j bk dbj
b∈σ −1 (a)
j∈f −1 (i)

where b = i bi ei .
Note that when f is an injection, the sum above has only positive signs. When
f is a bijection, then Totf is a self-equivalence of n-Comp(A). We write Tot = Totf
when m = 1.
We have defined an additive functor
Totf : n-Comp(A) → m-Comp(A).
Let g : {1, . . . , m} → {1, . . . , p} be a map and τ : Zm → Zp the associated
morphism. Let X ∈ n-Comp(A). The Zp -graded objects underlying Totgf (X) and
1
Totg (Totf (X)) have their component of degree a equal to c
c∈(τ σ)−1 (a) X . We
define an isomorphism between these p-complexes by multiplication by (−1)ε(c) on
X c , where
ε(c) = cl cl .
l<l
f (l)>f (l )
gf (l)=gf (l )
This gives an isomorphism of functors
∼
Totgf → Totg ◦ Totf .
Let k be a commutative ring and A, B and C be three k-algebras. Let
X ∈ n-Comp((A ⊗ B)-Mod) and Y ∈ m-Comp((B ⊗ C)-Mod). Then X ⊗B Y
defines an object of (n + m)-Comp((A ⊗ C)-Mod). We have (X ⊗B Y )(a1 ,...,an+m ) =
X (a1 ,...,an ) ⊗B Y (an+1 ,...,an+m ) .
5.1.2. Cohomology. Assume A is an abelian category. Let X ∈ n-Comp(A).

Let r ∈ {1, . . . , n} and Y = ker dr / im dr . This is an n-complex with dr,Y = 0. Let
i ∈ Z. Consider the map f1 : Zn−1 → Zn , (a1 , . . . , an−1 ) → (a1 , . . . , ar−1 , i, ar , . . . ,
an−1 ). We put Hdr (X) = a∈Zn−1 Y f (a) . This defines a functor
i
Hri : n-Comp(A) → (n − 1)-Comp(A).
Let g : {1, . . . , n} → {1, . . . , m} be a map and let r ∈ {1, . . . , m} such that

f −1 (r) = {s} for some s ∈ {1, . . . , n}. Define maps

i if i < s
α : {1, . . . , n − 1} → {1, . . . , n}, i →
i + 1 if i ≥ s
and
i if i < r
β : {1, . . . , m} → {1, . . . , m − 1}, i →
i−1 if i ≥ r
Then, we have a canonical isomorphism
∼
(1) Hdi r (Totf (M )) → Totβf α (Hdi s (M )).
5.1.3. Resolutions. Let k be a commutative ring and A a k-algebra. The k-
linear functor H 0 : Ho− (A-Proj) → A-Mod restricted to the full subcategory of
complexes M with H i (M ) = 0 for i = 0 is an equivalence. Let C = CA : A-Mod →
Ho− (A-Proj) be an inverse, composed with the inclusion functor. By construction
the resolution functor C is fully faithful. It induces a functor, still denoted by C,

C : Hob (A-Mod) → Hob Ho− (A-Proj) .

We view Hob Ho− (A-Proj) as a triangulated category with the canonical structure
on Hob (C), where C is the additive category Ho− (A-Proj). The functor C is a fully
faithful triangulated functor.
Assume A is projective as a k-module. Let X = CAen (A) be a projective
resolution of A as an Aen -module. The functor
− ⊗Aopp X : Compb (A-Mod) → 2-Comp(A-Proj)

composed with the canonical functor 2-Comp(A-Proj) → Ho Ho(A-Proj) is iso-
∼
morphic to C: we have M ⊗Aopp X → CA (M ) for M ∈ Hob (A-Mod).
−
Let i ∈ Z. The functor H i : Ho (A-Proj) → A-Mod induces a functor

Hdi 2 : Hob Ho− (A-Proj) → Hob (A-Mod).
It extends the functor Hdi 2 : 2-Comp(A-Mod) → Comp(A-Mod).

Let B, B be two k-algebras, projective as k-modules. Let L ∈ Hob (A ⊗

B opp )-Mod and M ∈ Hob ((B ⊗ B opp )-Mod). Assume the components of M are
projective right B -modules. We deduce from (1) an isomorphism

∼

(2) Tot13→1,2→2 CA⊗B opp (L) ⊗B M → CA⊗B opp Tot(L ⊗B M ) .
Note that when A is coherent, then A-Mod can be replaced by the abelian cat-
egory A-mod and A-Proj by A-proj. If A is graded, we can replace these categories
by the corresponding categories of graded modules.
5.2. Markov moves. Let (W, S) be a finite Coxeter group. Let P -exact be
the category of finitely generated graded P en -modules whose restrictions to P and
P opp are projective. Let M ∈ Hob (P -exact) and i ∈ Z. We put

K i (M ) = KSi (M ) = Hdi 2 P ⊗P en CP en (M ) ∈ Hob (P -modgr).
This defines a triangulated functor
K i : Hob (P -exact) → Hob (P -modgr).
Theorem 5.1. Given N, N ∈ Hob (P -exact), we have functorial isomorphisms
K (N ⊗P N ) K i (N ⊗P N ) in Ho(P -modgr).
i
Let s ∈ S and let zs be a non-zero element of (V ∗ )S\s . Let M ∈ Hob (PS\s -exact).
We have functorial isomorphisms

• KSi γS\s (M ) ⊗P Fs ρ∗S\s KS\s i+1
(M )[−1] in D(P -modgr)

• KSi γS\s (M ) ⊗P Fs−1 ρ∗S\s KS\s i
(M ) in D(P -modgr)

• KS γS\s (M )) P ⊗PS\s KS\s (M )−1⊕P ⊗PS\s KS\s

i i+1 i
(M ) in Ho(P -modgr).
Proof. Thanks to (2), we have

CP en Tot(N ⊗P N ) Tot12→1,3→2 (CP en (N ) ⊗P N )
hence

P ⊗P en CP en Tot(N ⊗P N ) Tot12→1,3→2 CP en (N ) ⊗P en N

Tot12→1,3→2 N ⊗P en CP en (N )

P ⊗P en CP en Tot(N ⊗P N )
and the first assertion follows.
We have

CP en (M ⊗ k[zs ]) Tot1→1,23→2 CPS\s
en (M ) ⊗ X
z ⊗1−1⊗z
where X = 0 → k[zs ]en −1 −−
s
−−−−−→
s
k[zs ]en → 0, the non-zero terms being in
degrees −1 and 0. Let

L = P ⊗P en CP en Tot (M ⊗ k[zs ]) ⊗P Fs .
By (2), we have

L P ⊗P en Tot13→1,2→2 CP en (M ⊗k[zs ])⊗P Fs Tot13→1,24→2 (Fs ⊗k[zs ]en X)⊗P en CP en (M )
S\s S\s
We have
θs
m /P θs
m /P
Fs ⊗k[zs ]en X zs ⊗1−1⊗zs 0 αs ⊗1−1⊗αs 0

θs 1 / P 1 θs 1 / P 1
m m
Indeed, there is c ∈ k such that s(zs ) − zs = −2cαs . Then zs − cαs ∈ (V ∗ )s , hence

∗
zs ⊗ 1 − 1 ⊗ zs and c(αs ⊗ 1 − 1 ⊗ αs ) are equal in θs .

Lemma 5.2 below shows that, when s∈Z(W ), then the exact sequence of P en -
modules
α ⊗1+1⊗αs a⊗b→as(b)
0 → P −−s−−−−−−→ θs −−−−−−−→ P s → 0
en
splits by restriction to PS\s . Here, P s = P as a left P -module, and the right action
of a ∈ P is given by multiplication by s(a). Note that when s ∈ Z(W ), then
P s = PS\s ⊗ k[αs2 ], and the splitting of the sequence holds trivially.
en
We deduce that there is an isomorphism of complexes of graded PS\s -modules
α ⊗1−1⊗αs
a→aα ⊗1−a⊗αs
0 → θs −−s−−−−−−→ θs 1 → 0 P [1]−1⊕ 0 → P s −−−−−s−−−−−−→ θs → 0 1.
P −1
2αs
/P Ps /0
Let Y1 = and Y2 = a→aαs ⊗1−a⊗αs

0 /0 θs 1 / P 1
We have an exact sequence of bicomplexes of graded P en -modules

0 → Y1 → Fs ⊗k[αs ]en X → Y2 → 0.
en
It splits after restricting to PS\s and applying ?(i,∗) . It follows that we have an
exact sequence of complexes of graded P en -modules

0 → Hdi 2 Tot13→1,24→2 Y1 ⊗PS\s en CP en (M )
S\s
→ Hdi 2 (L) →

→ Hdi 2 Tot13→1,24→2 Y2 ⊗PS\s en CP en (M )
S\s
→ 0.
We have an exact sequence of P en -modules

a→aα ⊗1−a⊗α m
0 → P s −−−−−s−−−−−−→
s
θs −→ P → 0.
So, the morphism of bicomplexes Y2 → Y2 :
Ps /0 0 /0
a→aαs ⊗1−a⊗αs

θs 1 / P 1 P 1
id / P 1
m 9 9
m id
induces an isomorphism of complexes

∼ i

Hdi 2 Tot13→1,24→2 Y2 ⊗PS\s
en CP en (M )
S\s
→ H d2
Tot13→1,24→2
Y2 ⊗P en CP en (M )
S\s S\s
and these complexes vanish in Hob (P en -modgr).

We deduce that

Hdi 2 (L) Hdi 2 Tot13→1,24→2 Y1 ⊗PS\s en CP en (M )
S\s

ρ∗S\s Hdi 2 Tot13→1,24→2 PS\s [(−1, 1)] ⊗PS\s
en CP en (M )
S\s

ρ∗S\s Hdi+1
2
P ⊗ en C en
S\s PS\s PS\s (M ) [−1]
in Db (P en -modgr). Note that the multiplication map P en → P is a split surjection
of algebras. We deduce the first and last terms of the sequence of isomorphisms
above are actually isomorphic in Db (P -modgr). This shows the second assertion.
The proof of the assertion involving Fs−1 is similar.
We have k[zs ] ⊗k[zs ]en X k[zs ]−1[1] ⊕ k[zs ] and the last assertion follows
immediately.
Lemma 5.2. Let s ∈ S. Assume s∈Z(W ). Let L = (V S\s )⊥ ∩ (V ∗ )s , a
hyperplane of (V S\s )⊥ = (VS\s )∗ . We have a commutative diagram of PS\s
en
-modules
where the diagonal map is an isomorphism
a⊗b→as(b)
θOs / Ps
qqq8
∼ qqq
a⊗b→a⊗b
qqqq →as(b)
qq a⊗b
PS\s ⊗S(L) PS\s
Here, P s denotes the left PS\s -module P endowed with a right action of a ∈ PS\s
by multiplication by s(a).
Proof. We will show that φ : PS\s ⊗S(L) PS\s → P, a ⊗ b → as(b) is an
isomorphism. It is a morphism of algebras, and a morphism of graded left PS\s -
modules.
By assumption, there is t ∈ S \ s such that mst = 2, so that s(αt ) − αt is
a non-zero multiple of αs . It follows that φ(αt ⊗ 1 − 1 ⊗ αt ) ∈ k× αs . Since
V ∗ = (V S\s )⊥ ⊕ kαs , we have P = PS\s ⊗ k[αs ] and we deduce that φ is surjective.
Since φ is a morphism of graded free left PS\s -modules with the same graded
ranks 1 + t + t2 + · · · , we deduce that φ is an isomorphism.
Remark 5.3. While we do not expect the category of Soergel bimodules to
exist for complex reflection groups, we hope that its homotopy category does exist,
as well as the 2-braid group. Its center would be a starting point for a structural
approach to the construction of unipotent data in Broué–Malle–Michel’s theory of
spets [BroMaMi].
References
[BeNa] D. Ben-Zvi and D. Nadler, The Character Theory of a Complex Group, preprint
arXiv:0904.1247v2.
[BroMaMi] M. Broué, G. Malle, and J. Michel, Towards spetses. I, Transform. Groups 4 (1999),
no. 2-3, 157–218, DOI 10.1007/BF01237357. Dedicated to the memory of Claude
Chevalley. MR1712862
[De] Théorie des topos et cohomologie étale des schémas. Tome 3 (French), Lecture Notes
in Mathematics, Vol. 305, Springer-Verlag, Berlin-New York, 1973. Séminaire de
Géométrie Algébrique du Bois-Marie 1963–1964 (SGA 4); Dirigé par M. Artin, A.
Grothendieck et J. L. Verdier. Avec la collaboration de P. Deligne et B. Saint-Donat.
MR0354654
[ElKr] B. Elias and D. Krasner, Rouquier complexes are functorial over braid cobordisms,
Homology, Homotopy Appl. 12 (2010), no. 2, 109–146. MR2721032
[Ge] M. Geck, Trace functions on Iwahori-Hecke algebras, Knot theory (Warsaw, 1995),
Banach Center Publ., vol. 42, Polish Acad. Sci., Warsaw, 1998, pp. 87–109.
MR1634450
[GeLa] M. Geck and S. Lambropoulou, Markov traces and knot invariants related to Iwahori-
Hecke algebras of type B, J. Reine Angew. Math. 482 (1997), 191–213. MR1427662
[Go] Y. Gomi, The Markov traces and the Fourier transforms, J. Algebra 303 (2006),
no. 2, 566–591, DOI 10.1016/j.jalgebra.2005.09.034. MR2255123
[Jo] V. F. R. Jones, Hecke algebra representations of braid groups and link polynomials,
Ann. of Math. (2) 126 (1987), no. 2, 335–388, DOI 10.2307/1971403. MR908150
[Ki] Y. Kihara, Weights of a Markov trace of dihedral type, Master Thesis, Sophia Uni-
versity, 2004.
[Kh] M. Khovanov, Triply-graded link homology and Hochschild homology of Soergel bimod-
ules, Internat. J. Math. 18 (2007), no. 8, 869–885, DOI 10.1142/S0129167X07004400.
MR2339573
[KhTh] M. Khovanov and R. Thomas, Braid cobordisms, triangulated categories, and flag
varieties, Homology Homotopy Appl. 9 (2007), no. 2, 19–94. MR2366943
[La1] T. Lasy, Special Markov traces and Gomi’s formula for reflection groups, PhD thesis,
University Denis Diderot, 2012.
[La2] T. Lasy, Markov property and Khovanov-Rozansky homology: Coxeter Case, preprint
arXiv:1202.5547.
[Li] N. Libedinsky, Équivalences entre conjectures de Soergel (French, with French sum-
mary), J. Algebra 320 (2008), no. 7, 2695–2705, DOI 10.1016/j.jalgebra.2008.05.030.
MR2441995
[Rou1] R. Rouquier, Categorification of sl2 and braid groups, Trends in representation theory
of algebras and related topics, Contemp. Math., vol. 406, Amer. Math. Soc., Provi-
dence, RI, 2006, pp. 137–167, DOI 10.1090/conm/406/07657. MR2258045
[Soe] W. Soergel, The combinatorics of Harish-Chandra bimodules, J. Reine Angew. Math.
429 (1992), 49–74, DOI 10.1515/crll.1992.429.49. MR1173115
[We] B. Webster, Khovanov-Rozansky homology via a canopolis formalism, Algebr. Geom.
Topol. 7 (2007), 673–699, DOI 10.2140/agt.2007.7.673. MR2308960
[WeWi] B. Webster and G. Williamson, The geometry of Markov traces, Duke Math. J. 160
(2011), no. 2, 401–419, DOI 10.1215/00127094-1444268. MR2852120
Department of Mathematics, UCLA, Box 951555, Los Angeles, California 90095-1555

E-mail address: rouquier@math.ucla.edu
Volume 684, 2017
DAHA approach to iterated torus links
Ivan Cherednik and Ivan Danilenko

Abstract. We extend the construction of DAHA-Jones polynomials for any
reduced root systems and DAHA-superpolynomials in type A from iterated
torus knots (our previous paper) to links, including arbitrary algebraic links.
Such a passage essentially corresponds to the usage of the products of Macdo-
nald polynomials and is directly connected to the splice diagrams (Neumann et
al.). The specialization t = q of our superpolynomials results in the HOMFLY-
PT polynomials. The relation of our construction to the stable Khovanov-
Rozansky polynomials and the so-called ORS-polynomials of the correspond-
ing plane curve singularities is expected for algebraic links in the uncolored
case. These 2 connections are less certain, since the Khovanov-Rozansky the-
ory for links is not sufficiently developed and the ORS polynomials are quite
involved. However we provide some confirmations. For Hopf links, our con-
struction produces the DAHA-vertex, similar to the refined topological vertex,
which is an important part of our work.
Contents
0. Introduction
0.1. Overview
0.2. Brief history
1. Double Hecke algebras
1.1. Affine root systems
1.2. Definition of DAHA
1.3. The automorphisms
1.4. Macdonald polynomials
1.5. Evaluation formula
2. Integral forms
2.1. A preliminary version
2.2. The integrality theorem
2.3. Using the duality
2.4. Concluding the proof
2.5. Affine exponents
2.6. J -polynomials in type A
2010 Mathematics Subject Classification. Primary 14H50, 17B22, 17B45, 20C08, 20F36,
33D52, 30F10, 55N10, 57M25.
Key words and phrases. Double affine Hecke algebra, Jones polynomial, HOMFLY-PT poly-
nomial, Khovanov-Rozansky homology, iterated torus link, cabling, Macdonald polynomial, topo-
logical vertex.
The first author was partially supported by NSF grant DMS–1363138.
2017
159
160 IVAN CHEREDNIK AND IVAN DANILENKO
3. Topological vertex
3.1. Theta-functions
3.2. Mehta-Macdonald identities
3.3. Norm-formulas
3.4. The case of An
3.5. High–level 3j–symbols
3.6. The coinvariant approach
4. DAHA-Jones theory
4.1. Iterated torus knots
4.2. From knots to links
4.3. Splice diagrams
4.4. DAHA-Jones polynomials
4.5. The polynomiality
4.6. Major symmetries
5. DAHA-superpolynomials
5.1. Main theorem
5.2. Sketch of the proof
5.3. Super-vertex
5.4. HOMFLY-PT polynomials
5.5. Connection Conjecture
6. Multiple torus knots
6.1. Preliminary remarks
6.2. Uncolored 2-fold trefoil
6.3. Similar links
6.4. Uncolored 2-fold T (4, 3)
6.5. Uncolored 2-fold T (5, 3)
6.6. Uncolored 3-links
7. Hopf links
7.1. Basic constructions
7.2. Colored Hopf 3-links
7.3. Hopf 2-links
7.4. Algebraic Hopf links
7.5. Twisted T (2κ, κ)
8. Further examples
8.1. On cable notations
8.2. Colored 2-links
8.3. 3-hook for trefoil
8.4. Alexander polynomials
8.5. Two different paths
9. Generalized twisted union
9.1. Uncolored trefoil-prime
9.2. Colored/iterated examples
9.3. Generalized twisting
9.4. Some examples
9.5. Toward the Skein
Appendix A. Links and splice diagrams
A.1. Links, cables and splices
A.2. Splice diagrams
DAHA APPROACH TO ITERATED TORUS LINKS 161
A.3. Operations on links

A.4. Equivalent diagrams
A.5. Connection with DAHA
References
0. Introduction
We extend the construction from [ChD] of the DAHA-Jones polynomials (any
reduced root systems) and the DAHA superpolynomials in type A from knots to
iterated torus links , including all algebraic links. There is solid evidence that un-
der t = q our polynomials become correspondingly the Quantum Group (WRT)
invariants of such links. This was checked in particular cases in [Ch2, ChD] using
[Ste1] and proper variants of the Rosso-Jones formula ; see [RJ,Mo,ChE]. For the
HOMFLY-PT polynomials , this coincidence was recently verified via the approach
from [MS] (they proved it for iterated torus knots). This paper is devoted to the
DAHA-Jones theory (for generic q, t).
Our main technical tool is the switch from the Macdonald polynomials in the
DAHA-Jones theory from [Ch2] and further works to their products, considered in
terms of X and in terms of Y . The projective action of GL2 (Z) in DAHA is the
key here. The Hopf links play an important role in our work; they directly lead to
the DAHA-vertex , similar to the refined topological vertex . See [GIKV, AKMV,
AFS], but no exact relation to these papers is expected because we do a different
version of the Macdonald theory (fixed root systems). Though the topological
2–vertex from [AKMV] is connected to our one.
Due to the novelty of the direction we present in this work, explicit examples are
its very important part. We note that the products of Macdonald polynomials in
the context of superpolynomials of links appeared in physics works. See for instance
[GIKV] (the uncolored Hopf 2–link) and especially Section 4.3 from [DMS]; our
superpolynomials match those suggested there (for simple multiple torus knots).
0.1. Overview. Let try to put our work into perspective.

0.1.1. Topological connections. The relation of our construction to the HOMFLY-
PT homology [KhR1, KhR2, Kh, Ras] is not too certain, since the theory of the
latter for links and with colors is not sufficiently developed. Nevertheless, we con-
jecture the coincidence of our superpolynomials (type A) with the stable Khovanov-
Rozansky polynomials for algebraic uncolored links in the unreduced setting.
We also establish some relation to the reduced Khovanov polynomials (for sl2 )
for small iterated links, not only algebraic. The reduction to sl2 is actually a
special case of the approach from [DGR] and [Ch2, ChD], which is based on the
theory of differentials from [Kh, Ras]. We can recover certain (reduced) Khovanov-
Rozansky polynomials for small links directly from the corresponding DAHA-Jones
polynomials.
There is a fundamental connection of our construction with the splice diagrams
from [EN]; we mainly use them in the form of pairs of incidence graphs , which
are collections of labeled trees. The trees with strict positivity of r, s (and under
certain further conditions in the case of twisted unions ) describe the links of germs
of any plane curve singularities. As for non-algebraic links, the simplest (and
important) example is the so-called Hopf κ–link, which corresponds to the tree
with one vertex labeled by [1, −1] and κ arrowheads (with arbitrary colors). The
topological symmetries of links and the corresponding splice diagrams appeared
equivalent to the invariance of the polynomial representation of DAHA under the
automorphisms τ− , η, σ 2 , ι and the ϕ–invariance of the DAHA-coinvariant . See
Section 1.3 for the definitions.
For the Hopf 3–links (3–fibers of the Hopf fibration), the corresponding DAHA-
Jones polynomials and superpolynomials constitute the topological DAHA-vertex .
They satisfy important associativity identities, generalizing those from paper [ChF]
on the nil-DAHA approach to the Rogers-Ramanujan sums (upon the limit t → 0).
This associativity we establish is an interesting q, t–generalization (with an
additional parameter a for superpolynomials) of the relations for the so-called 3j–
symbols. The Macdonald polynomials times the powers of Gaussians are the key
here. This power is the level of the theory; it is 0 for the q, t–deformation of
the 3j–symbols. The construction possesses the permutation invariance, which is
important in TQFT , and the associativity (3.31). To be exact, the latter holds for
the Hopf 3–links with the linking numbers 1, 1, −1 . The standard Hopf 3–links
with the linking numbers −1, −1, −1 result in the S3 –invariance; and then the
associativity must be properly adjusted.
Not all solvable links in S3 can be obtained by our construction, but all algebraic
links are reached [EN]. The basic algebraic example is T (κ r, κ s), described by the
tree with one vertex labeled by [r, s] and κ arrowheads colored by dominant weights.
Our labels are Newton pairs [r, s], which however allowed to be negative. They
naturally emerge in our construction vs. the “topological” a, r–pairs.
0.1.2. Algebraic links. For algebraic links, we conjecture in the uncolored case
(i.e. for the fundamental representation) that unreduced DAHA-superpolynomials,
defined as the series H min /(1 − t)κ−1 for reduced ones H min are proportional to
P alg from [ORS] defined in terms of the nested Hilbert schemes of (the germs of)
plane curve singularities C. This is not simple to verify; the weight filtration , which
is the key ingredient of the ORS polynomials, is generally quite involved.
Generalizing Conjecture 2 in [ObS] (extended to arbitrary colors and proved
in [Ma]), the ORS conjecture from [ORS] claims the connection of their polyno-
mial to the Poincarè polynomial P KhR (Link of C) of the unreduced triply-graded
HOMFLY-PT homology of Khovanov and Rozansky. This was stated there for any
algebraic uncolored links.
Thus we have three independent mathematical constructions of link superpoly-
nomials, presumably coinciding (up to normalization) with each other in the case
of uncolored algebraic links. The DAHA construction is certainly the simplest the-
oretically and practically and it works for any root systems and with arbitrary
colors. Finding P KhR is an involved task, even for simple knots (and quite a chal-
lenge for links and with non-trivial colors). The polynomials P alg are sophisticated
too, though the super-duality and some other important symmetries were checked
for them [ORS], which match those in the DAHA theory. There is also the 4th
important approach based on the rational DAHA [Gor, GORS], but it is restricted
to torus knots.
We mainly omit in this work the Jacobian factors of plane curve singularities.
The conjectural relation of the reduced DAHA superpolynomials under a = 0, q = 1
to the Betti numbers of Jacobian factors of unibranch C played an important role

in [ChD]. The passage from algebraic knots to links does not add too much new
here. The DAHA-superpolynomials at q = 1 (the case of the trivial center charge)
are products of those over the branches; the corresponding product formula for the
Betti numbers is geometrically obvious.
0.1.3. Our construction. The data are the pairs of incidence graphs {L, L} ,
a unions of subtrees, which are labeled, colored and with arrowheads; see Section
4.3 for the splice interpretation. The labels are Newton pairs, lifted to γ ∈ GL2 (Z)
and interpreted as DAHA-automorphisms γ . We check that the final formulas do
not depend on the flexibility here and, moreover, depend only on the corresponding
link, which is quite a confirmation of the connections with topology.
The tree structure and labels determine which products of the corresponding
polynomials must be considered and which γ must be applied; the resulting DAHA
operators (γ act only in HH) are always projected onto its polynomial representation
before the next step [ChD]. In the presence of L , we replace X by Y ∓1 in the
pre-polynomial for L, L∨ and apply this operator to the pre-polynomial for L.
The DAHA-coinvariant is applied to the last pre-polynomial.
In the simplest case of the κ–fold torus knot T (κr, κs), there is no L and the
matrix γ is an arbitrary lift of (r, s)tr considered as its first column. It is then
lifted to the corresponding DAHA automorphism γ and is applied to the product
of Macdonald polynomials for the dominant weights assigned to the κ arrowheads.
The usage of Y ∓1 for L, L∨ is connected with adding the meridian in topology.
The sign corresponds to the orientation. Our positivity claim from Part (ii) of
the Connection Conjecture 5.3 and formula (5.40) for dega from Theorem 5.1 are
expected to hold only for Y +1 there (i.e. with ∨). The classical Hopf links are with
the pairwise linking numbers −1, i.e. for Y −1 ; the positivity of the corresponding
min /((1−t)(1−q))M does not hold for them.
series H
0.1.4. DAHA theory. The algebraic properties of the DAHA-Jones polynomials

and DAHA-superpolynomials from [Ch2, GN, Ch3, ChD] can be fully extended
to arbitrary iterated torus links , which includes the polynomiality, duality, special
values at q = 1, the color exchange and further symmetries. The main new feature
(vs. knots) is that the construction now depends on the choice of the q, t–integral
form of the Macdonald polynomials. The normalization is the division by the LCM
of all evaluations of the Macdonald polynomials involved. Upon this division, the
notation is JDmin and H min (for the superpolynomials). The integral form does
not matter if only one color is involved, say in the case of knots.
min is very reasonable because all colors are on equal footing in
Algebraically, H
its definition and this is a natural setting for our polynomiality theorem. However
H min seems generally too “small” to be interpreted topologically. Namely, the
reduced HOMFLY-PT polynomial (t = q) requires picking one of the branches and
then the division by the evaluation of the corresponding Macdonald polynomial,
not by the total LCM . The unreduced HOMFLY-PT polynomials are without
any divisions at all. They are topological, but have poles for links with respect
to q; so must be generally considered as q–series. Let us comment a little on the
HOMFLY-PT polynomials in the case t = q.
The Macdonald P –polynomials becomes Schur functions; they do not depend

on q as t = q. So no integral form is needed in the DAHA approach to QG/WRT-
invariants for t = q. The unreduced HOMFLY-PT polynomials are then uniquely
recovered from the corresponding QG–invariants for (all) An upon the substitution
a = q n+1 for any n (or any infinite sequence of them). This can be considered
as a definition of HOMFLY-PT polynomials; another approach is based on the
theory of the Skein. The corresponding QG–invariants are q–polynomials (without
denominators). However their a–stabilizations are generally rational in terms of
q. Thus we need to switch to the (infinite) q–expansion to make the coefficients
of HOMFLY-PT polynomials meaningful geometrically and topologically simply
because of the a–stabilization. Such unreduced setting is not necessary for knots.
Generally we use the J–polynomials in this paper instead of P –polynomials,
which provides the absence of the denominators in the DAHA superpolynomials.
We note that the corresponding H min are not always irreducible as polynomials of
±1
a, q, t ; they are really minimal possible (irreducible) only for sufficiently small
dominant weights.
The J–polynomials have no exact counterparts for arbitrary root systems.
However the QG–invariants are well defined for any root systems. Therefore
the refined theory of such invariants can be expected to involve certain theory
of J–polynomials. Unfortunately the q, t–integral forms of the Macdonald P –
polynomials are not generally settled for arbitrary root systems. Following [Ch5],
we prove that the spherical P –polynomials become q, t–integral upon multiplication
by certain products of q, t–binomials of multiplicity one (with a minor reservation
for the root systems D2m ). This is for nonsymmetric and symmetric Macdonald
polynomials, which provides an important technical tool in the theory of the decom-
position of the polynomial DAHA representation and is of independent algebraic
interest. For instance, this can be used to simplify and generalize the theory from
[Ch5] among other applications. Some relations to [HHL],[RY],[OS] can be ex-
pected.
The corresponding q, t–integral polynomials seem as close as possible to the
Macdonald J–polynomials. The latter are actually not minimal as q, t–integral
ones; the focus of their theory is the stabilization and positivity (not only the q, t–
integrality). Say for t = q, they are greater than the Macdonald P –polynomials,
which are Schur functions. The main deviation of our modified J–polynomials,
denoted by P in this work, from the classical J–theory in type A is that our
construction is naturally invariant with respect to the automorphism ι = −w0 for
the element of maximal length w0 ∈ W . This is not the case for the J–polynomials;
their stabilization property is incompatible with ι.
0.2. Brief history. Due to the novelty of the DAHA-Jones theory, a brief
account of the history of this (very recent) direction can be useful to the readers.
We will not discuss other instances of using Macdonald polynomials and DAHA in
algebra, geometry, topology and physics; there are quite a few now (including String
Theory, rational DAHA, Hilbert schemes and so on). The focus of this section will
be on using q, t–DAHA in Knot Theory and related theory of plain curve singular-
ities, especially on the applications to the refined WRT invariants, HOMFLY-PT
polynomials and HOMFLY-PT homology (stable Khovanov-Rozansky polynomials)
and on using the splice diagrams.
0.2.1. Torus knots via DAHA. Using the Macdonald polynomials instead of
Schur functions in the so-called knot opertators was suggested in [AS], which
directly influenced [Ch2]. These operators naturally appear in the approach to
the Jones and WRT invariants of torus knots via Conformal Field Theory. See
[Ste1, Ch2, GN, Ch3]. The main difficulty of the construction of [AS] was the
usage of roots of unity, similar to that in Verlinde algebra (when t = q). Even the
simplest ingredient needed there, the refined Verlinde S operator, requires knowing
all Macdonald polynomials at roots of unity that occur in the corresponding per-
fect modules. They are quite involved beyond A1 ; one generally needs any An to
find the superpolynomial. A related problem is that t must be generally an integral
power of q to ensure the existence of the P SL(2, Z)– action in the corresponding
DAHA-module.
There are some formulas for the (coefficients of) Macdonald polynomials for
An (n > 1), but they are very involved theoretically and practically; see e.g. [HHL].
In spite of such obvious difficulties, the authors demonstrated that their calculations
for the simplest torus knots match known or conjectured formulas (upon their
restriction to roots of unity) for the superpolynomials from topology or physics
papers. The superpolynomials are not involved for the simplest torus knots, say
1 + qt + aq for the uncolored trefoil; they are generally very non-trivial apart from
the family T (2m + 1, 2) even in the absence of colors.
Let us mention that the refined S operator taken alone, already quite involved,
results only in the unknot (with the trivial superpolynomial). The whole projective
unitary action of P SL2 (Z) in perfect modules, also called refined (generalized)
Verlinde algebras, is needed in this approach, which is due to Cherednik and A.
Kirillov Jr. This makes using perfect DAHA modules here very difficult, even if
when latest software (like SAGE) is employed for the Macdonald polynomials.
The lift of the formulas at roots of unity q, t to generic q, t is always quite
a challenge (including the Verlinde algebras, i.e. in the unrefined theory), unless
the existence of the superpolynomials and the bounds for the degrees of q, t, a are
a priori known. The rank stabilization (associated with a) has its own challenges
too, even for generic q, t. Also, we do any colors and non-torus iterated knots/links.
These problems were resolved (actually bypassed) in [Ch2], which eliminated

the usage of roots of unity and was written for any torus knots, arbitrary reduced
root systems and dominant weights (colors). It was a significant development and
the beginning of the DAHA-Jones theory. The construction of [AS] results from
that from [Ch2] simply because the projective action of P SL2 (Z) in DAHA re-
duces to that in its perfect modules (a standard theorem from [Ch1]); no other
justification of this connection with [AS] is necessary.
The coincidence with the HOMFLY-PT polynomials (as t = q) for torus knots
was proven in [Ch2] and the connections were conjectured with the (colored) WRT-
invariants for any root systems and reduced uncolored stable Khovanov-Rozansky
polynomials (the Poincarè polynomials of the HOMFLY-PT homology). The latter
are quite involved; not many explicit formulas are known for them. Quite a few un-
colored and colored DAHA superpolynomials were calculated in this paper, as well
as the refined Quantum Group (WRT) invariants for the classical and exceptional
root systems.
As one of the applications, the refined QG invariants of types E6,7,8 for minus-
cule and quasi-minuscule weights and simplest torus knots were calculated in [Ch2],
conjecturally coinciding with the corresponding QG invariants at t = q (confirmed
for E6 by Ross Elliot via the Rosso-Jones formula).
Let us mention that the (projective) action of P SL2 (Z) in DAHA is a gener-
alization of the action of P SL2 (C) in the Heisenberg algebra; DAHA is its certain
deformation. The key obstacle in classical theory of Fourier-Hankel transform is
that there is no action of P SL2 (C) in the Fock representation. This remains the
same in DAHA theory, unless upon the reduction to perfect modules (generalized
Verlinde algebras). Fortunately, such a reduction appeared unnecessary for DAHA-
Jones theory and DAHA superpolynomials.
Concerning the current status of conjectures from [Ch2], practically all “intrin-
sic” ones about the existence and the structure of the DAHA-Jones polynomials
and DAHA-superpolynomials (the A-case) were proved in [Ch3, GN]. The only
conjecture in type A from this paper that remains open by now is the positivity
of the DAHA superpolynomials for rectangle Young diagrams. There are recent
geometric developments here (but no proof so far).
The existence of the DAHA-hyperpolynomials in types B, C, D and their sym-
metries remain open, though the approach from [Ch2] (based on [SV]) can be
extended to this case and there is a sketch of the proof of hyper-duality in [Ch3].
In type D, the Kauffman polynomials (instead of the HOMFLY-PT ones) were
conjectured in [Ch2] to coincide with the DAHA-Jones polynomials as t = q; this
is justified by now.
We note that some of the symmetries of superpolynomials were suggested by
physicists. The most interesting one is the super-duality conjectured in [GS]. This
conjecture became rigorous and for any Young diagram in the DAHA setting [Ch2].
In String Theory, the super-duality is related to the approach to the refined Chern-
Simons-Witten theory via the M5 – theory; the action of C∗ × C∗ there is naturally
associated with parameters q, t. See also e.g. [DGR, AS, DMS] concerning various
physics aspects and formulas.
The DAHA super-duality for torus knots was proven in [GN] (see [Ch3] for
an approach via roots of unity) together with a justification of the stabilization
of the DAHA-Jones polynomials in type A, which was announced in [Ch2] as a
theorem. The switch from the Macdonald polynomials of type An to the so-called
J–polynomials (the stable ones) in [GN] is important. The J–polynomials can be
avoided for knots, but are absolutely necessary for links (our present work).
0.2.2. From torus knots to iterated links. The main demerit of the DAHA-Jones
theory after [AS, Ch2, GN, Ch3] was its restrictions to torus knots. Arbitrary
algebraic knots and links (not only torus ones) are very much needed because of
many reasons.
First of all, this generalization is necessary to employ the technique of physically
inspired theory of the resolved conifold , and its monoidal-type transformations,
used in [Ma] to prove the OS conjecture from [ObS]. Second, there are significant
links to the Fundamental Lemma and related algebraic geometry of rational planar
curves, with various implications within and beyond the Langlands Program.
Third, the topological reasons for the switch from torus knots to any iterated
torus links are quite obvious. This class of links is closed with respect to the
cabling , one of the major operations in Knot Theory. We recall that all algebraic
links are cables of unknot. However iterated torus knots and links are generally
non-algebraic. It is known which ones are algebraic (see [EN]), but this is far from
trivial, especially for links. The present paper is mostly written for any iterated
torus links (not only algebraic), though the positivity conjectures for links generally
require the algebraic ones (and the unreduced setting).
The topological perspective is very important here since the superpolynomials

(of any origin, geometric, algebraic or physical) presumably coincide with the stable
Khovanov-Rozansky polynomials. The theory of such polynomials and HOMFLY-
PT homology is mainly restricted now to uncolored knots (unless for the Khovanov
sl2 homology). There are recent developments, including [WW] and the approach
based on the Howe duality, that allow in principle to incorporate colors, but techni-
cal difficulties remain very serious. The Khovanov-Rozansky theory and categorifi-
cation are of course for all knots, not only for cables, the level DAHA have reached
so far.
The passage to arbitrary torus iterated knots from torus knots was an important
step of the DAHA approach. Let us mention that we expected our theory in [ChD]
to be connected with paper [Sam], but this did not materialize. We failed to
understand the approach used in [Sam] for q = t. In contrast to [ChD], this paper is
for A1 only, but nevertheless his polynomials Jn have significant q, t–denominators,
which is not what can be expected. The polynomiality of DAHA-Jones polynomials
is the key in [Ch2] and further works, including this one. The examples of his J2 –
polynomials for Cab(±5, 2)T (3, 2) from Section 5.2 (ibid.) are very different from
our ones for the same cables as q = t (and we do not understand how they were
obtained and cannot reproduce them ourselves). See Section 4 of [ChD].
0.2.3. Using the Skein. Let us discuss paper [MS]. As t = q, the authors estab-
lish the connection of the Skein of the torus with the Elliptic Hall algebra [SV].
This implies (through quite a few technical steps) the coincidence of the DAHA
superpolynomials of arbitrary torus iterated knots at t = q with the corresponding
HOMFLY-PT polynomials. This was established for any iterated torus knots for
A1 in [ChD] and for any torus knots in [Ch2] in the case of arbitrary An (and
therefore for the HOMFLY-PT polynomials). It is now checked for the Kauffman
polynomials too (this is the DAHA D–case; see [Ch2]).
We note that the exact framing factor (a power of q, a) is missing in [MS] in

contrast to [ChD] (for A1 ). Also, a nonstandard framing in [MS], which makes
their construction directly in terms of Newton’s pairs (vs. the “topological” a, r-
parameters), potentially creates problems with the isotopy equivalence (they refer
to our algebraic proof of the topological invariance).
Using the Seifert framing , which extends the framing used in [MS] (Definition
7.1) from knots to links, the splice diagrams [EN], and the interpretation of the
DAHA-multiplication via link operations, we can extend Section 7.1 of [MS] from
iterated torus knots considered there to arbitrary (colored) torus iterated links .
For instance, the multiplication by Jλ (X) for a Macdonald polynomial is essen-
tially adding an unknot colored by λ; also, applying Jλ (Y ) to a pre-polynomial is
interpreted as adding a meridian. These are standard facts in the theory of skein
(Morton and others) and they are actually the key for the passage from knots to
links (from [ChD] to this paper). Thus, Conjecture 5.3, (i) below (for t = q) be-
comes a theorem if our present paper is combined with [MS]; we will provide the
details elsewhere.
This coincidence is the A–case of the conjectured connections of the DAHA-
Jones polynomials at t = q with the general WRT-invariants for arbitrary root
systems and for any dominant weights. They do not seem very difficult to justify
(including torus iterated links ) using the Rosso-Jones cabling formula [RJ] and
the DAHA shift operator, as it was demonstrated in [Ch2, ChD, ChE]. We have
a sketch of such a justification for An and Ross Elliot can essentially do this for
general root systems (unpublished). Another approach here is via CFT and Ver-
linde algebras, but it seems more involved. Using the skein provides a justification
“without calculations” (thought there were quite a few steps in [MS] to verify),
but this is restricted only to the case of An .
Summary. Let us briefly summarize the main points discussed above. The
passage from knots to arbitrary torus iterated links in our present paper is an
important step in the DAHA-Jones theory. This is absolutely necessary from the
topological-geometric perspective, for instance for reaching the invariants of 3-folds
(via framed links). Also, adding colors in Knot Theory is closely related to the
passage from knots to links. The multi-brunch plane curve singularities are very
important in its own right (including the Fundamental Lemma). One more direction
to be mentioned, is the physical and mathematical theory of topological vertex
based on Hopf links. Algebraically, the DAHA theory of torus iterated links we
present here seems reasonably complete; we use it to approach the q, t–Skein at the
end of the paper.
0.2.4. Splice diagrams. There is a long history of using combinatorial/graphic
presentations of knots, links and related 3-folds. Algebraic links are of great interest
here, since they are in one-to-one correspondence with the plane curve singulari-
ties. Splice diagrams suggested by Neumann and systematically developed in [EN]
appeared a great tool for us. They were aimed at Seifert manifolds (and their
plumbing). This generality is beyond our paper; we assume that the links are in
S3 , not in an arbitrary homology 3-sphere.
The main result of [EN] is that the splice diagrams are isotopy invariants of
the corresponding links and describe all of them. See Theorems 9.2, 9.4 in [EN],
Section 4.3 below and the Appendix A on splice diagrams. We note that our method
does not produce all splice diagrams of solvable type; but all algebraic links can be
reached.
The key operation from [EN] is splicing ; it provides large families of links.
Other operations can be mainly considered as its special cases. The cabling and
twisted unions play the major role in the DAHA approach to torus iterated links.
The other operations (in our context) are erasing components, orientation reversion,
and disjoint sums. The isotopy invariance of our construction is stated in Theorem
4.3,(ii) for DAHA-Jones polynomials (any root systems) and in Theorem 5.1, (ii)
for the superpolynomials.
The justification of the topological symmetries from Theorem 4.3,(ii) is essen-
tially parallel to Theorem 1.2 from [Ch3]. Namely, it is proven there that the
DAHA-Jones polynomials JD (and superpolynomials) are the same for the torus
knots T (r, s) and T (s, r). This coincidence, the triviality of these polynomials for
T (r, 1) and the effect of mirroring (changing q, t to q −1 , t−1 when r → −r) are the
only topological symmetries in the small universe of torus knots. These facts were
deduced in [Ch3] from the properties of the DAHA involutions. We note that the
symmetry T (r, s) ↔ T (s, r) can be far from obvious in other approaches to super-
polynomials (say, when rational DAHA is used for obtaining such polynomials).
When switching to torus iterated knots, one also needs to check that the torus
knot T (r, mr + s) results in the same DAHA-Jones and superpolynomials as the
iterated knots corresponding to the Newton pairs [1, m], [r, s] (the latter is employed
the first). This is the key new symmetry here. Topologically, T (r, mr+s) is obviously
isotopic to Cab(mr+s, r)T (1, m) since T (1, m) is unknot. The corresponding relation
m
for the JD–polynomials readily follows from the commutativity τ− (associated
with [1, m]) with the projection ⇓ onto the polynomial representations, which is
equivalent to the fact that τ− acts in this representation [Ch1]. Other symmetries
are due to applying η, ι, σ 2 inside the DAHA coinvariant; they are compatible with
the projection ⇓ as well. We also constantly use the ϕ–invariance of the coinvariant.
Generalizing the theory from [Ch3, ChD] to arbitrary torus iterated links, the
major (additional) fact needed for the isotopy invariance of our construction is
Theorem 3.9, which states that
- −1 −1
. - −1 . - −1
.
τ− (f ), τ− (g) ev = τ− (f g) ev = τ− f (X)g(X) ev
for arbitrary f = f (X), g = g(X) and the evaluation pairing

{f, g}ev = {f (Y −1 ) g(X) }ev ,
where {·}ev is the DAHA coinvariant.
All other symmetries are either straightforward or follow from this theorem
and the theorems mentioned above for torus iterated knots (based on the action of
ϕ, η, τ− and σ 2 ). For instance, applying ϕ provides switching the components in
the pairs of trees {L, L}; see (4.24) for the exact relation.
The pairs of trees give a very natural way to encode the DAHA invariants,
which corresponds to the consideration of splice diagrams with marked edges in the
terminology of [EN]. The marked edge shows the place where the DAHA coinvariant
will be applied to the last pre-polynomial . I.e. this gives the last step of our
calculation. In the case of a single tree, the last edge is marked (connecting the
last vertex with one of the leaves from it in the language of splice diagrams); the
tree for an iterated knot is simply a path with an arrow at its end and leaves at
the vertices [ChD].
The isotopy invariance of DAHA-Jones polynomials (and superpolynomials)
includes the proof of their independence of the choice of the marked edge; this is
the key new feature when doing links. Assuming that the graph is connected, it
suffices to check that adjacent marked edges give the same. And this follows directly
from the symmetries discussed above, including Theorem 3.9 and the equivalence
of T (r, s) and T (s, r) (the transposition of a leaf and an edge from the same vertex
in the terminology of [EN]).
A typical and instructional example of using Theorem 3.9 is the coincidence of
the
superpolynomials for the link from (9.1) and that for the cable
Cab(2, 3)Cab(−1, 0) T (1, −1) from (9.4). We provide there a direct DAHA de-
duction of the corresponding symmetry and its interpretation in terms of splice
diagrams.

There are many examples of this kind in the paper. For instance, the
link Cab(8, 3)Cab(0, 1) T (2, 1) from (8.8) and (9.10) is considered in full detail.
0.2.5. Conclusion. As we tried to demonstrate, the extension of [Ch2, GN,

Ch3, ChD] from knots to torus iterated links in this paper (any reduced systems
and weights) is an important step with various possible applications in Knot Theory
and theory of plane curve singularities. Using links seems inevitable here for any
self-consistent theory. Even the simplest of them are quite valuable; say, the Hopf
links govern the new theory of DAHA-vertex.
One of the applications of links is the approach to the toric q, t–Skein we suggest
at the end of the paper, based on generalized twisted unions; see Section 9.5. The
DAHA knot operators are certain (symmetric) elements in H H corresponding to
links L. Their matrix elements are essentially the superpolynomials of the links
that are obtained from given L by adding two probe links “on the left and on the
right”. This direction will be continued in our further works.
Due to the novelty of the presented theory, we provide many examples, in-
cluding detailed analysis of algebraic and topological symmetries. The numerical
formulas are important. We selected the most instructional ones; each and every
serves some purpose. The theory of algebraic/iterated links is quite ramified and
we need to demonstrate its range (and related DAHA features) theoretically and
practically. The DAHA approach provides a unique way of obtaining the super-
polynomials for arbitrary colored iterated links; we think that explicit formulas can
help researches from neighboring fields, including various divisions of Macdonald
theory and DAHA theory.
The intrinsic algebraic theory of DAHA-Jones polynomials and the theory of
DAHA superpolynomials in type A is sufficiently well developed by now (we think,
better than other approaches to superpolynomials). This includes the key q, t±1 , a–
polynomiality of DAHA-Jones polynomials, the super-duality and quite a few other
(proven) properties. We expect that the passage to links in this paper is an impor-
tant step toward the theory of refined invariants of Seifert 3–manifolds and related
toric-type surfaces, hopefully including significant applications in Number Theory.
1. Double Hecke algebras

1.1. Affine root systems. Let R = {α} ⊂ Rn be a root system of type
An , . . . ,G2 with respect to a euclidean form (z, z ) on Rn & z, z , W the Weyl group
generated by the reflections sα , R+ the set of positive roots corresponding to fixed
simple roots α1 , . . . , αn ; R− = −R+ . The form is normalized by the condition
(α, α) = 2 for short roots. The weight lattices are P = ⊕ni=1 Zωi , where {ωi }
are fundamental weights: (ωi , αj∨ ) = δij for the coroots α∨ = 2α/(α, α). The root
lattice is Q = ⊕ni=1 Zαi . Replacing Z by Z± = {m ∈ Z, ±m ≥ 0}, we obtain
P± , Q± . See e.g., [Bo] or [Ch1].
def
Setting να == (α, α)/2, the vectors α = [α, να j] ∈ Rn × R ⊂ Rn+1 for α ∈
R, j ∈ Z form the twisted affine root system R ⊃ R, where z ∈ Rn are identified with
def
[z, 0]. We add α0 == [−ϑ, 1] to the simple roots for the maximal short root ϑ ∈ R+ .
The corresponding set R + of positive roots is R+ ∪ {[α, να j], α ∈ R, j > 0}.
The set of the indices of the images of α0 by all automorphisms of the affine
def
Dynkin diagram will be denoted by O (O = {0} for E8 , F4 , G2 ). Let O == {r ∈

O, r = 0}. The elements ωr for r ∈ O are minuscule weights , defined by the
inequalities (ωr , α∨ ) ≤ 1 for all α ∈ R+ . We set ω0 = 0 for the sake of uniformity.
b ∈ P , let
= [α, να j] ∈ R,
Affine Weyl groups Given α
(1.1) z ) = z − (z, α∨ )
sα ( α, b (
z ) = [z, ζ − (z, b)]
= sα , α
for z = [z, ζ] ∈ Rn+1 . The affine Weyl group W + ) is the semidirect
∈R
product W Q of its subgroups W = sα , α ∈ R+ and Q, where α is identified
with
sα s[α, να ] = s[−α, να ] sα for α ∈ R.
is W P , where the corresponding action is
The extended Weyl group W
(1.2) (wb)([z, ζ]) = [w(z), ζ − (z, b)] for w ∈ W, b ∈ P.
Π for Π =
It is isomorphic to W
def
= P/Q. The latter group consists of π0 =id and
the images πr of minuscule ωr in P/Q. Note that πr−1 is πrι and u−1r is ur ι , where
ι is the standard involution (sometimes trivial) of the nonaffine Dynkin diagram,
induced by αi → −w0 (αi ). Generally ι(b) = −w0 (b) = bι , where w0 is the longest
element in W .
The group Π is naturally identified with the subgroup of W of the elements of
the length zero; the length is defined as follows:
def + ).
= |Λ(w)|
l(w) for Λ(w)
== R −1 (−R
+∩w
One has ωr = πr ur for r ∈ O , where ur is the element u ∈ W of minimal length

such that u(ωr ) ∈ P− , equivalently, w = w0 u is of maximal length such that
w(ωr ) ∈ P+ . The elements ur are very explicit. Let w0r be the longest element in
the subgroup W0r ⊂ W of the elements preserving ωr ; this subgroup is generated
by simple reflections. One has:
(1.3) ur = w0 w0r and (ur )−1 = w0r w0 = urι for r ∈ O.
= πr w
Setting w ∈W for πr ∈ Π, w ∈W , l(w)
coincides with the length of
any reduced decomposition of w in terms of the simple reflections si , 0 ≤ i ≤ n.
of the elements of length 0.
Thus Π is a subgroup of W
1.2. Definition of DAHA. We follow [Ch3, Ch2, Ch1]. Let m be the least
natural number such that (P, P ) = (1/m)Z. Thus m = |Π| unless m = 2 for D2k
and m = 1 for B2k , Ck .
The double affine Hecke algebra, DAHA , depends on the parameters q, tν (ν ∈
def ±1/2
{να }) and will be defined over the ring Zq,t == Z[q ±1/m , tν ] formed by polyno-
±1/2
mials in terms of q ±1/m and {tν }. Note that the coefficients of the Macdonald
polynomials will belong to Q(q, tν ).
0 ≤ i ≤ n, we set
= [α, να j] ∈ R,
For α
(1.4) tα = tα = tνα = qαkν , qα = q να , ti = tαi , qi = qαi ,
Also, using here (and below) sht, lng instead of ν, we set
def 1
1
ρk == kα α = ksht ρsht +klng ρlng , ρν = α= ωi .
2 α>0 2 ν =ν ν =ν, i>0
α i
For pairwise commutative X1 , . . . , Xn ,

n
Xili q j if b = [b, j], w(X
def
(1.5) Xb == b ) = Xw(
b) ,
i=1
n
1
where b = li ωi ∈ P, j ∈ Z, w .
∈W
i=1
m
For instance, X0 == Xα0 = qXϑ−1 .

def
Recall that ωr = πr ur for r ∈ O (see above) and that πr−1 = πι(i) , where ι
here is the standard involution of the nonaffine Dynkin diagram. Finally, we set
mij = 2, 3, 4, 6 when the number of links between αi and αj in the affine Dynkin
diagram is 0, 1, 2, 3.
Definition 1.1. The double affine Hecke algebra H H is generated over Zq,t by
the elements {Ti , 0 ≤ i ≤ n}, pairwise commutative {Xb , b ∈ P } satisfying ( 1.5)
and the group Π, where the following relations are imposed:
1/2 −1/2
(o) (Ti − ti )(Ti + ti ) = 0, 0 ≤ i ≤ n;
(i) Ti Tj Ti . . . = Tj Ti Tj . . . , mij factors on each side;
(ii) πr Ti πr−1 = Tj if πr (αi ) = αj ;
(iii) Ti Xb = Xb Xα−1 i
Ti−1 if (b, αi∨ ) = 1, 0 ≤ i ≤ n;
(iv) Ti Xb = Xb Ti if (b, αi∨ ) = 0 for 0 ≤ i ≤ n;
(v) πr Xb πr−1 = Xπr (b) = Xu−1 r (b)
q (ωι(r) ,b) , r ∈ O .
, r ∈ O, the product
∈W
Given w
def
(1.6) Tπr w == πr Til · · · Ti1 , where w
= sil · · · si1 for l = l(w),

does not depend on the choice of the reduced decomposition Moreover,
(1.7) Tv Tw = Tv w whenever l( = l(
v w) for v, w
v ) + l(w) .
∈W
In particular, we arrive at the pairwise commutative elements
def

n
n
def
(1.8) Yb == Yili if b = li ωi ∈ P, Yi == Tωi , b ∈ P.
i=1 i=1
When acting in the polynomial representation V (see below), they are called dif-
ference Dunkl operators.
1.3. The automorphisms. The following maps can be (uniquely) extended

to automorphisms of H
H , where q 1/(2m) must be added to Zq,t (see [Ch1], (3.2.10)–
(3.2.15)) :
(ωr ,ωr )
(1.9) τ+ : Xb → Xb , Ti → Ti (i > 0), Yr → Xr Yr q − 2 ,
(ω ,ω )
τ+ : T0 → q −1 Xϑ T0−1 , πr → q − r2 r
Xr πr (r ∈ O ),
(ωr ,ωr )
(1.10) τ− : Yb → Yb , Ti → Ti (i ≥ 0), Xr → Yr Xr q 2 ,
τ− (Xϑ ) = qT0 Xϑ−1 Ts−1 −1 −1 −1
def
ϑ
; σ == τ+ τ− τ+ = τ− τ+ τ− ,
(1.11) σ(Xb ) = Yb−1 , σ(Yb ) = Tw−1
0
Xb−1
ι Tw0 , σ(Ti ) = Ti (i > 0).
These automorphisms fix tν , q and their fractional powers, as well as the following

anti-involution :
(1.12) ϕ : Xb → Yb−1 , Yb → Xb−1 , Ti → Ti (1 ≤ i ≤ n).
The following anti-involution results directly from the group nature of the
DAHA relations:
(1.13) H = H −1 for H ∈ {Tw , Xb , Yb , q, tν }.
To be exact, it is naturally extended to the fractional powers of q, t:
1
− 2m
1
, q 2m → q − 2m .
1 1
: tν2m → tν
This anti-involution serves the inner product in the theory of the DAHA polynomial
representation.
We will also need the involution:
1
− 2m
1
η : Ti → Ti−1 , Xb → Xb−1 , πr → πr , tν2m → tν , q 2m → q − 2m .
1 1
(1.14)
where 0 ≤ i ≤ n, r ∈ O , b ∈ P . Its actions on Yb is not that uniform:
−1 −1
(1.15) η : Yr → πι(r) Tuι(r) , Yϑ → T0−1 Ts−1
ϑ
,
where ι is the involution of the nonaffine Dynkin diagram; see (1.3).
The involution η extends the Kazhdan–Lusztig involution in the affine Hecke
theory; see [Ch1], (3.2.19–22). Note that
−1 −1 −1
(1.16) ϕτ± ϕ = τ∓ = στ± σ , ητ± η = τ± , ϕσϕ = σ −1 = ηση.
Let us list the matrices corresponding to the automorphisms and anti-automorphisms
1 1
above upon
the natural

projection onto GL 2 (Z), corresponding to t ν = 1 = q 2m .
2m
The matrix αγ βδ will then represent the map Xb → Xbα Ybγ , Yb → Xbβ Ybδ for b ∈ P .

One has:τ+ 1
0
1
1
, τ− 1
1
0
1
,σ 0
−1
1
0
,ϕ 0
−1
−1
0
,η −1
0
0
1
.
Enhanced projective GL2 (Z). The projective GL2 (Z) is the group generated
−1 −1 −1 −1
by τ± , η subject to the relations τ+ τ− τ+ = τ− τ+ τ− , η 2 = 1 and ητ± η = τ± .
The notation will be GL∧2 (Z). The span of τ± is the projective P SL2 (Z) (due to
Steinberg), which is isomorphic to the braid group B3 .
Let us enrich these groups by the following automorphisms of H H. For the pair
of arbitrary characters u, v of Π = P/Q,
(1.17) ζu,v (Xa Tw Yb ) = u(a)v(b)Xa Tw Yb for a, b ∈ P, w ∈ W,
where the order of X, T, Y does not matter here and Tw can be replaced by Tw for
any w∈W . The map Xa Tw Yb → u(a)v(b) can be readily extended to a character
of H
H (its one-dimensional representation).
These automorphisms satisfy the following relations:
(1.18) ϕ ζu,v ϕ = ζv−1 ,u−1 , η ζu,v η = ζu−1 ,v ,
τ+−1 ζu,v τ+ = ζu,uv , τ−
−1
ζu,v τ− = ζ uv,v , σ −1 ζu,v σ = ζv−1 ,u ,
v −1 u −1 v −1 u v −1
def
(1.19) τ+u (τ− ) τ+ = ζu,v σ ζu,v == σ u,v = (τ− ) τ+ (τ− ) for
def −1 −1 def
τ+u == ζu,v τ+ ζu,v v
= τ+ ζ1,u , τ− == ζu,v τ− ζu,v = τ− ζv,1 .
The action of τ± on ζu,v by conjugation is dual to the natural action of SL2 (Z)
in Π2 . Recall that u, v are arbitrary characters of Π. Formulas (1.19) readily follow
from (1.18). We call the group generated by GL∧2 (Z) and all ζu,v the enhanced
projective GL2 (Z) and use the notation GL∧2 (Z)ζ .
The coinvariant. The projective GL2 (Z) and the coinvariant are the main in-
Let us define the latter. Any H ∈ H
gredients of our approach. H can be uniquely
represented in the form c
a,w,b a,w,b X T Y
a w b for w ∈ W , a, b ∈ P (the PBW the-
orem, see [Ch1]). Then the coinvariant is a functional H H → C uniquely defined
via the following substitution in such sums:
{ }ev : Xa → q −(ρk ,a) , Yb → q (ρk ,b) , Ti → ti .

1/2
(1.20)
The key symmetries of the coinvariant are:
(1.21) { ϕ(H) }ev = { H }ev , { η(H) }ev = { H }ev , { ι(H) }ev = { H }ev .
We use here that ι naturally acts in H

H:
(1.22) ι(Xb ) = Xι(b) , ι(Yb ) = Yι(b) , Tiι = Tι(i) , 1 ≤ i ≤ n.
One has {HTw Y b}ev = {H}ev χ(Tw Yb ), where χ is the standard character (one-
dimensional representation) of the algebra HY , by definition generated by Tw , Yb
for w ∈ W, b ∈ P , which sends Yb → q (ρk ,b) , Ti → ti . Therefore {. . .}ev acts via the
def
projection H → H ⇓ == H(1) of H H onto the polynomial representation V, which
is the H
H–module induced from χ; see [Ch1, Ch2, Ch3] and the next section.
1.4. Macdonald polynomials. We will begin with the explicit construction

of the polynomial representation, denoted by V in this work.
Polynomial representation. It was already defined above as an induced represen-
tation. In detail, it is isomorphic to Zq,t [Xb ] as a vector space and the action of
Ti (0 ≤ i ≤ n) there is given by the Demazure-Lusztig operators :
−1/2
)(Xαi − 1)−1 (si − 1), 0 ≤ i ≤ n.
1/2 1/2
(1.23) Ti = ti si + (ti − ti
The elements Xb become the multiplication operators and πr (r ∈ O ) act via the
b ) = Xw(b)
general formula w(X for w . Note that τ− η and ι naturally act in
∈W
the polynomial representation. See formula (1.37) from [Ch3] and (1.37) below for
τ− . We will use the notation τ̇− for this action; it is explicitly defined as follows:

def
(1.24) τ̇− (g) == τ− g(X) (1) for g ∈ V.
As for η:
(1.25) η(f ) = f , where Xb = X−b , (q υ ) = q −υ , (tv ) = t−v for υ ∈ Q.
Also, the automorphisms ζu,1 from (1.17) act in V. They can be represented by
certain translations in x ∈ Cn for X = q x . Since they preserve Yb , we obtain that
simple Y –eigenvectors in V are also ζu,1 –eigenvectors for any character u : Π → C∗ .
u
Thus they are also invariant (up to proportionality) under the action of τ− = τ− ζu,1
from (1.19).
Symmetric Macdonald polynomials. The standard notation for them is Pb (X)
for b ∈ P+ (they are due to Kadell for the classical root systems). The definition
is as follows. Let c+ be such that c+ ∈ W (c) ∩ P+ (it is unique); recall that

Q+ = ⊕ni=1 Z+ αi . For b ∈ P+ ,

Pb − Xa ∈ ⊕b =c+ ∈b−Q+ Q(q, tν )Xc and Pb Xcι μ(X; q, t) = 0
a∈W (b)
∞

def (1−Xα qαj )(1−Xα−1 qαj+1 )
for such c, where μ(X; q, t) == .
α∈R+ j=0
(1−Xα tα qαj )(1−Xα−1 tα qαj+1 )
Here and further f is the constant term of a Laurent series or polynomial f (X);
μ is considered a Laurent series of Xb with the coefficients expanded in terms of
positive powers of q. The coefficients of Pb belong to the field Q(q, tν ). One has
(see (3.3.23) from [Ch1]):
(1.26) Pb (X −1 ) = Pbι (X) = Pb (X), Pb (q −ρk ) = Pb (q ρk )
(α∨ ,b)−1
−ρk −(ρk ,b)
1 − qαj tα Xα (q ρk )
(1.27) = (Pb (q )) = q .
α>0 j=0 1 − qαj Xα (q ρk )
Recall that ι(b) = b = −w0 (b) for b ∈ P .

ι
DAHA provides an important alternative (operator) approach to the P –polynomials;

namely, they satisfy the (defining) relations
Lf (Pb ) = f (q −ρk −b )Pb , Lf == f (Xa → Ya )

def
(1.28)
for any symmetric (W –invariant) polynomial f ∈ C[Xa , a ∈ P ]W . Here b ∈ P+ and
the coefficient of Xb in Pb is assumed 1.
Nonsymmetric Macdonald polynomials. The P –polynomials are t-symmetrizations
of the nonsymmetric Macdonald polynomials Eb ∈ V. For any b ∈ P , we define
them as follows:
−1
(1.29) Ya (Eb ) = q −(a,b+wb (ρk ))
Eb , where wb ∈ W is
a unique element of maximal length such that wb (b) ∈ P+ .
The element wb for b ∈ P+ here is the element of maximal length (an invo-
lution) in the centralizer of b in W . The normalization of E in (1.29) is by
the condition that the coefficient of Xb in Eb is 1. For b ∈ P+ , one has that
Eb − Xb ∈ ⊕b+ =c+∈b−Q+ Q(q, tν )Xc . See [Mac] (for ksht = klng ∈ Z+ ) and (6.14)
from [Ch5] or (3.3.14) from [Ch1]; the differential version is due to Opdam with a
participation of Heckman. The Macdonald conjectures for them were extended from
the symmetric case and justified in [Ch6]; as a matter of fact, these conjectures
become significantly simpler in the nonsymmetric setting.
We note that all monomials in Eb or Pb are in the form Xb+a for a ∈ Q, i.e.
have coinciding images in Π = P/Q. This readily follows from the commutativity
of Yc with the automorphisms ζu,1 from (1.17).
1.5. Evaluation formula. One of the key formulas in this work is the follow-
ing evaluation:
(α∨ ,b)−1
−ρk −(ρk ,b)
(1.30) Eb (q )=q for b ∈ P+ .
α>0 j=1 1 − qαj Xα (q ρk )
For any b ∈ P and wb from (1.29),

1 − q j t X (q ρk )
(1.31) Eb (q −ρk ) = q −(ρk ,b+ ) α α α
,
+ 1 − qαj Xα (q ρk )
{α,j}∈Λb
def ∨ −1
b == {{α > 0, j > 0} | (b+ , α ) > j > 0 for wb (α) ∈ R+ ,
Λ+
(b+ , α∨ ) ≥ j > 0 for wb−1 (α) ∈ R− }.
Formula (1.31) is the Macdonald evaluation conjecture in the nonsymmetric
variant from [Ch6]; see also formulas (6.33), (7.15) from [Ch5] and (3.3.45) from
[Ch1]. The set Λ+
b appears here due to the following (ibid.):
(1.32) + & [−α, jνα ] | {α, j} ∈ Λ+ } = Λ(πb ), where

{R b
πb == bwb−1 w0 , Λ(w)
def + ∩ w
=R + ) for w
−1 (−R .
∈W
Then the DAHA intertwining operators are used to justify (1.31).
def
Spherical normalization. We call Eb◦ == Eb /Eb (q −ρk ) for b ∈ P nonsymmetric
def
spherical polynomials . Accordingly, Pb◦ == Pb /Pb (q −ρk ) for b ∈ P+ , and
1−tα Xα (q ρk )
Pb (q −ρk ) = ΠbR Eb (q −ρk ), ΠbR ==
def
(1.33) ,
1−Xα (q ρk )
α>0,(α,b)>0
(α,b)>0 (α∨,b)−1
(1.34) Pb (q −ρk ) = q −(ρk ,b) .
α>0 j=0 1 − qαj Xα (q ρk )
The product ΠbR becomes over all α > 0 if all ωi appear in b ∈ P+ ; it is then the
Poincaré polynomial ΠR . See formulas (7.15)-(7.19) from [Ch5] concerning these
polynomials with different root lengths. Recall that tα = qαkα here (and there); see
(1.4) above.
The symmetrization. The symmetrization relation between Eb◦ and Pb◦+ is as
follows. For any b ∈ P ,
lsht (w)/2 llng (w)/2
Pb◦+ = (ΠR )−1 P+ (Eb◦ ) for P+ ==
def
(1.35) tsht tlng Tw ,
w∈W
where lsht , llng count correspondingly the number of si for short and long αi in any
reduced decomposition w = sil · · · si1 . We check that the right-hand side here is
proportional to Pb+ and then evaluate at q −ρk by applying the general formula
1/2
{Ti (f )}ev = ti {f }ev for i > 0 and any f (X).
Let us provide the following particular case of (1.35) for the standard Macdo-
nald polynomials:
1 − tα Xα (q ρk )
(1.36) Pb = P+ (Eb ) for b ∈ P+ ,
1 − Xα (q ρk )
α>0,(α,b)=0
For generic dominant b (when all ωi occur in its decomposition), the coefficient of
proportionality here is 1.
Let us also give the formulas for the natural action of τ− in V in the basis
of E–polynomials. See e.g., formula (1.37) from [Ch3]. Recall that we use the
notation τ̇− for this action:

(b+ , b+ )
(1.37) τ̇− (Eb ) = q − 2 −(b+ , ρk )
Eb for b ∈ P.
This readily gives that τ̇− (Pb ) = q −(b,b)/2−(b,ρk ) Pb for b ∈ P+ ; for instance, use
(1.36). We assume that Eb , Pb are well defined.
2. Integral forms
2.1. A preliminary version. The technique of intertwining operators ap-
plied to the E–polynomials results in the following existence criterion . Given
b ∈ B and assuming that q is not a root of unity, the following polynomials are
q, t–integral :

(2.1) Nb Eb◦ for Nb = 1 − qαj tα Xα (q ρk ) ,
{α,j}∈Λ+
b

(2.2) Db Eb for Db = 1 − qαj Xα (q ρk ) .
{α,j}∈Λ+
b
See [Ch7], Corollary 5.3 or (6.13) from [Ch5]. In the case of An , this observation
is due to F. Knop. By integrality (we will mainly omit q, t here), we mean that
their coefficients belong to Zq,t = Z[q ±1 , t±1 ]. Recall that tα = qαkα = q να kα .
These two polynomials obviously coincide because Nb /Db = Eb (q −ρk ); see
(1.30). So we actually have only one statement here. Switching to Pb for b ∈ P+ ,
one obtains the integrality of Nb Pb◦ , Db Pb . Note that all binomials in Nb , Db
contain q and t (both).
Using [Ch5], the existence criterion for Eb◦ can be made much sharper. Namely,
the multiplicities of the binomials (1 − q • t• ) in Nb (generally large) can be reduced
to 1, with minor adjustments for D2m . Also, some of these binomials can be further
reduced to their proper factors. We will justify this mainly following [Ch5]. Such
an improvement was not done there; it is helpful in decomposition theory of V.
2.2. The integrality theorem. Assuming that q, tlng , tsht are generic, the
evaluation formula for Eb (q −ρk ) stabilizes to an infinite product as b ∈ P+ of
integral polynomials as b becomes sufficiently large. Namely, it approaches the
product of ΠR from Part (iii), Theorem 11.8 of [Ch5], which is extended to all
ksht + Z+ , klng + Z+ and then divided by the Poincarè polynomial ΠR .
This product has a canonical presentation E∞ (q −ρk ) = N∞ /D∞ , where the
factors of N ∞ and D∞ are products of binomials in the form
(2.3) (1 − q æ ) for æ = j + jsht ksht + j k , where lng lng
j ∈ N, jsht /νsht ∈ Z+ & jlng /νlng and jsht + jlng > 0.

The exponents æ = j+jsht ksht +jlng klng for the binomials of N∞ are called affine
exponents . There is a deviation from the usage of this term in [Ch5]. Namely, we
omit the binomials with j = 0 and (more importantly) consider the stable case i.e.
do not bound j. Thus the exact name must be affine stable q-dependent exponents ,
but we will simply call them affine. We call an affine exponent æ rational if it is
not divisible by any exponents æ that occur in D∞ .
∞
In the case ADE, he denominator D∞ is very simple: j=1 (1 − tq ). For
j
these and arbitrary root systems, each non-rational æ canonically corresponds to

its divisor æ in D∞ (æ can have multiplicities there). We will provide below
explicit lists. This correspondence will be denoted: æ æ . Importantly, the
multiplicity of each æ in the list of affine exponents is one (including BCFE) unless
for the series æ = 2mk + j in the case of D2m , which is of multiplicity 2.
Note that the binomials in (2.3) are in terms of integral (non-negative) powers
of q, tsht , tlng , so we use æ (in terms of kν ) for the sake of simplicity of notations. For
b ∈ P and α ∈ R+ , we set δα (b) = 1 if wb−1 (α) ∈ R− and 0 otherwise; if b ∈ P− ,
then δα (b) = 1 for any α ∈ R+ .
Theorem 2.1. We fix b ∈ P . Let us consider only the affine exponents æ that

occur in Nb , which means the existence of α ∈ R+ such that æ = kα + (α, ρk ) + j
and (b+ , α∨ )+δα (b) > j > 0. If b ∈ P+ , the inequality (b+ , α∨ ) > 0 and the relation
b = b+ imply that δα (b) = 0 here. Define N b as the product over all such æ of the
following factors:
(a) the binomials (1 − q æ ) for the rational exponents æ,

(b) and the ratios (1 − q æ )/(1 − q æ ) for non-rational æ,
where we use the above correspondence æ æ . There is a reservation for æ =
2mk + j in the case of D2m ; we take the corresponding factor from (a) or (b) twice
if there exist at least two α > 0 such that (α, ρ) = 2m−1 subject to (b+ , α)+δα (b) >
j > 0.
Then we set N b =
def
= gcd(Nb , N∞ /D∞ ), and claim that the coefficients of the

polynomials Eb == Nb Eb◦ are q, t–integral for b ∈ P , as well as those of Pb ==
def def
Nb P for b ∈ P+ . Here we can assume that j ≤ N in N∞ /D∞ for sufficiently

◦
b
large N .
2.3. Using the duality. The justification will be based on the DAHA-duality.
We can assume that the parameters q, tsht , tlng are generic. Indeed, the denomina-
tors of the coefficients of E b can be only in the form (2.3) for certain æ, so it suffices
to establish the absence/reduction of some of them for generic parameters only.
Assuming that q is not a root of unity, the key fact is that for any given values
of tsht , tlng in C∗ , the polynomials Ec become regular (i.e. their coefficients become
regular) for sufficiently large c ∈ P . The latter means that (c+ , αi ) > C for proper
C * 0 and all 1 ≤ i ≤ n.
See formulas (9.16) and (9.17) from [Ch5] (right before Section 10 “The struc-
ture of Vc ”). It is demonstrated there that the generalized Y –eigenspace in V
containing Ec becomes 1-dimensional for c satisfying certain inequalities for its co-
efficients; this readily results in the existence of the corresponding Ec (i.e. in its
q, t–integrality).
Actually Theorem 10.3 there (in Subsection 10.3 “The semisimple submodule”)
establishes that V has a canonical nonzero semisimple submodule. However it was
proven in this work under certain restriction and this (much stronger) fact is not
necessary for our proof. Vice versa, our result here is useful for the decomposition
of the polynomial representation, which we will try to address in other publications.
Then we apply the duality theorem from [Ch6]; see also (6.30) from [Ch5]).
It states that for any b, c ∈ P :
Eb◦ (q c
) = Ec◦ (q b
), b == b + wb−1 (ρk ), Yb (Ec ) = q −(c
,b) Ec ,
def
(2.4)
Pb◦ (q c+ρk ) = Pc◦ (q b+ρk ), Pb (Y −1 )(Pc ) = Pb (q c+ρk )Pc for b, c ∈ P+ .
Note that wb = w0 for b ∈ P− (including b = 0) and therefore b = b − ρk for such
b. Also, b = b + ρk for generic b ∈ P+ (such that (b, αi ) > 0 for i = 1, . . . , n).
The duality is the main motivation of the spherical normalization . All Macdo-
nald’s conjectures can be deduced from the duality. See Proposition 6.6 in [Ch5]
and the end of Section 7 there concerning the deduction of the Evaluation Con-
jecture, Norm Conjecture and Constant Term Conjecture (in this order) from the
duality. One can add to this list the Pieri rules and the difference Mehta-Macdonald
Conjecture from [Ch4], which also result from the duality.
2.4. Concluding the proof. Fixing q, t, b and assuming that c is sufficiently
large (and that q is not a root of unity), we conclude that (Nc /Dc )Eb◦ (q c
) is regular
for such c. This regularity formally results in the regularity of the coefficients of
this polynomial at such q, t; let us demonstrate this.
Not any set of values can be generally used for such an implication, even if it is
arbitrarily large (infinite) and sufficient to “catch” all coefficients of our polynomial.
Say, the non-integral polynomial F (X) = (1 − X)/(1 − q) has integral values at any
X = q n (n ∈ Z); such values are of course sufficient to recover (the coefficients of)
any polynomial F (X). The Lagrange interpolation formula readily clarifies which
sets are good for this.
Let us consider one variable X and assume that a Laurent polynomial F (X) can
be uniquely recovered from its values at certain pairwise distinct points X = q ai .
Then the denominators of the coefficients of F (X) can be only divisors of q ai −q aj =
0, where we treat ai as parameters. Thus the regularity of F (q ai ) implies that for
the coefficients of F if there are no common factors of all differences q ai − q aj .
For instance, let us establish the regularity of (Nc /Dc )Eb◦ for the root system
A1 and sufficiently large c. We set X = Xω for ω = ω1 and take aj = (jω) =
sgn(j)(k + |j|)/2 for i ∈ Z with sufficiently large |j|. Note that sgn(0) = −1 here,
but we need only large |j|. Thus q ai − q aj for i > j is
either tsgn(i)/2 (q i/2 − q j/2 ) for ij > 0
or t1/2 q i/2 − t−1/2 q −j/2 for i > 0 > j.
Let the set J = {j} be with sufficiently large |j| and generic enough for the
recovery of the coefficients of Eb◦ from its values at q c
for c = jω. Let us pick j
here with the same sgn(j); then the corresponding differences in the case ij > 0
are relatively prime to any binomials from (2.3) due to jsht = 1 there. The series
i > 0 > j can be used here too provided i, j are sufficiently large (depending on the
set of æ).
Given b, such a choice of J proves that the coefficients of (Nc /Dc )Eb◦ are
regular. Also, since |j| are sufficiently large, we can replace Nc /Dc by N∞ /D∞ ,
which is its stabilization as |j| → ∞. Since it is q, t–integral (a product of q, t–
polynomials), we conclude that (N∞ /D∞ )Eb◦ is integral too. Finally, we can restrict
ourselves only to the binomials in N∞ that appear in Nb ; we use that all factors
from (a, b) are irreducible for generic t in type A1 .
The case of any root system is actually no different, but we must avoid using
the Lagrange interpolation. Let us take c such that the polynomials Ec exist for
any zeros of Nb . We can assume that c ∈ P− and therefore c = c − ρk . Let us
substitute X → Xq −ρk , Xa → q −(ρk ,a) Xa ; this will not change the (non)regularity
of the coefficients. Upon this rescaling, values of F (X) = (Nc /Dc )Eb◦ (Xq −ρk ) at
q c with sufficiently large c are all regular for such c. We pick sufficiently many of
them ensuring that the coefficients of F (X) can be recovered from its values in this
set (assume its minimality).
Generally the coefficients of F are recovered by applying the inverse of the
matrix transforming the coefficients of F to its values. The entries of this matrix
are powers of q. Therefore the zeros of its (Vandermonde-type) determinant cannot
be zeros of any binomials in (2.3) because æ there always contain ksht or klng .
Then we proceed as for A1 and obtain that the multiplication of Eb◦ by N∞ /D∞
is actually sufficient to make this polynomial q, t–integral. We also know that Nb Eb◦
is integral. Thus we can keep in N∞ /D∞ only the factors that come from the
binomials from Nb , which reduces N∞ /D∞ to N b .
Here we can consider a sufficiently large truncation (N∞ /D∞ )† of N∞ /D∞
and treat it and Nb as polynomials of one variable q with the coefficients that are
rationals in terms of tsht , tlng . Given any fixed t–part (i.e. for any fixed jsht , jlng in
jsht ksht +jlng klng ), the corresponding factors from (a, b) in the theorem are relatively
prime to each other. Indeed, such binomials/ratios must have then different powers
q j and we can use that gcd of the Kummer-type polynomials (1−uq l ) and (1−uq m )
for l = m is 1 if u is not a power of q.
Let us take here a maximal pair {jsht , jlng } (in the poset of pairs). For any
divisor of the corresponding factor from (a, b) that divides gcd(Nb , (N∞ /D∞ )† )
and divides N b , the whole factor from (a, b) must divide both of them. Therefore
one can multiply Eb◦ by the product of all such factors for a given maximal {jsht , jlng }
and then proceed by (double) induction with respect to the (remaining) {jsht , jlng }.
This concludes the proof of the theorem.
2.5. Affine exponents. We will provide here the lists of affine (stable and
q–dependent) exponents. We follow [Ch5], where the formulas must be naturally
adjusted to the stable setting. We use the Coxeter exponents mi (see e.g., [Bo]),
determined from the relation
1 − t1+(α, ρ∨ ) n
1 − t mi +1
(2.5) = , 2ρ∨ = α∨ .
1−t (α, ρ ∨)
i=1
1 − t
α∈R+ α∈R+
This product is the classical Poincarè polynomial , which is for tsht = t = tlng in
terms of our ΠR ; see (1.33). One has that m1 + . . . + mn = |R+ | and (mi +
1) · · · (mn + 1) = |W |; (mi + 1) are called the degrees of the root system R.
The ADE–case. We set tsht = t = tlng , t = q k in the simply-laced case. Then
∞
1 − q j t1+(α,ρ) 1 − q j t mi +1
n ∞
N∞
(2.6) = = .
D∞ j=1
1 − q j t(α,ρ) i=1 j=1
1 − qj t
α∈R+
Therefore the list of affine exponents is

(2.7) ADE : {æ = (mi + 1)k + j + 1 | i = 1, . . . , n, j ∈ Z+ },
unless {(mi + 1)(k + j + 1) æ = k + j + 1, j ∈ Z+ }.
Recall that non-rational exponents æ (the second line) result in the correspond-
ing quotients in N∞ with respect to the map “ ” from æ to their (canonical)

divisors æ from the set of exponents of D∞ .
The case of Bn . The are two series of rational exponents in this case:
(2.8) {æ = 2mklng + 2j + 2 | 2 ≤ m ≤ n, j + 1 ∈ mZ+ },
{æ = 2mklng + 2ksht + 2j + 1 | 0 ≤ m < n, j ≥ 0}.
Here and below j will be always from Z+ . The list of the non-rational æ (with
their canonical divisors æ from D∞ ) is as follows:
(2.9) {4mklng + 2ksht + 2j + 2 2mklng + ksht + j + 1 | 1 ≤ m < n},
(2.10) {m(2klng + 2j + 2) æ = 2klng + j + 1 | 2 ≤ m ≤ n}.
The case of Cn . There are two rational series as well:

(2.11) {æ = mksht + j + 1 | 2 ≤ m ≤ n for j + 1 ∈ mZ+ },
{æ = δm (2mksht + 2klng + 2j + 1) | 0 ≤ m < n, j ≥ 0},
where δm = 2 if m < n/2 and 1 otherwise.
Note that up to proportionality, the rational Cn –exponents coincide with those
of type Bn upon the transposition ksht ↔ klng . This is related to our usage of
“rational” here; in the limit to rational DAHA , the rational exponents (and only
them) can be seen and the proportionality does not matter. So the affine exponents
for B and C coincide in this limit. (We mention that we somewhat fixed (8.3) from
[Ch5]; the main formula (8.6) for all affine exponents in the C–case is correct
here).
The non-rational exponents and their divisors æ (from D∞ ) are as follows in
the case of Cn :
(2.12) {4klng + 2mksht + 2 m (j + 1) 2klng + mksht + m (j + 1)},
where 1 ≤ m < n, m = 1 for odd m and 2 otherwise,
{æ = m(ksht + j + 1) æ = ksht + j + 1} for 2 ≤ m ≤ n.
The case of G2 . There are 3 series of rational exponents:

(2.13) {2ksht + 2j + 1}, {6klng + 6j + 3}, {3klng + 3ksht + 3j + 1, 2},
where j ∈ Z+ as in all formulas here. By series , we mean here the sequences of æ
with coinciding k–components.
The non-rational series are:
{2(ksht + j + 1) ksht + j + 1}, {6(klng + j + 1) 3(klng + j + 1)},
(2.14) {9klng + 3ksht + 3j + 3 3klng + ksht + j + 1} for j ≥ 0.
The case of F4 . In this case, the rational affine exponents are:

{2ksht + 2j + 1}, {3ksht + 3j + 1, 2},
{4klng + 4j + 2}, {6klng + 2j + 2, 4},
{2klng + 4ksht + 2j + 1}, {4klng + 4ksht + 2j + 1, 2, 3},
(2.15) {6klng + 6ksht + 2j + 1}, {8klng + 4ksht + 4j + 2}.
The non-rational F4 –series are:

{2(ksht + j + 1) ksht + j + 1}, {3(ksht + j + 1) ksht + j + 1},
{4(klng + j + 1) 2(klng + j + 1)}, {6(klng + j + 1) 2(klng + j + 1)},
{4klng + 4ksht + 4j + 4 2klng + 2ksht + 2j + 2},
{8klng + 2ksht + 2j + 2 4klng + ksht + j + 1},
{8klng + 4ksht + 4j + 4 2klng + ksht + j + 1},
(2.16) {12klng + 6ksht + 2j + 2 6klng + 3ksht + j + 1}.
Note that the exponents æ = 2klng + ksht + j + 1 from D∞ can potentially

divide 8klng + 4ksht + 4j + 4 and 12klng + 6ksht + 2j + 2, but 6klng + 3ksht + j + 1
can be used only to divide the latter. Thus the correspondence is still unique
in this case. The series æ = ksht + j + 1 and æ = 2(klng + j + 1) are of the same
kind. Such examples exist for other root systems, but all instances come from the
Poincarè polynomial ΠR (upon translations of ksht , klng by Z+ ).
2.6. J -polynomials in type A. In the case of An , let us calculate Nb for

dominant b = ni=1 bi ωi , i.e. with bi ≥ 0 for all i. The usage of Young diagrams
is standard (and convenient) here; they are defined as follows:
λ = λ(b) = {λ1 = b1 + . . . + bn , λ2 = b2 + . . . + bn , . . . , λn = bn }.
We set R+ & α = εlm = εl − m for 1 ≤ l < m ≤ n + 1 in the standard basis
{εl } ∈ Rn+1 . Then
i
(2.17) ωi = ε1 + . . . + εi − (ε1 + . . . + εn+1 ) for i = 1, . . . , n,
n+1
n

n
b= λi εi − |λ| ε1 + . . . + εn+1 /(n + 1) for |λ| = λi ,
i=1 i=1
1

ρ = ω1 + . . . +ωn = (n−1)ε1 + (n−3)ε2 + . . . + (1−n)εn .
2
def
One has: (b, εlm ) = bl + . . . + bm−1 , (b, ρ) = |λ|/2 − λ1 /2. Also, b2 == (b, b) =
n
i=1 λi − |λ| /(n + 1).
2 2
−1
Let us calculate the set Λ+ b from (1.31). Since b ∈ P+ , the root w (α) there
can be only from R+ . Thus:
def
b == {{R+ & α = εlm , j > 0} | bl + . . . + bm−1 > j > 0}.
Λ+
Our construction. In Theorem 2.1, we need to know all possible æ = (α, kρ) +
j = k(m − l) + j for such {α, j}. For l + p − 1 ≤ n, they are as follows:
def
(2.18) {kp + j | 1 ≤ p ≤ n, 0 ≤ j < jp == max(bl + . . . + bl+p−1 )};
l
(2.19) if bn ≥ bn−1 ≥ . . . ≥ b1 , then jp = λn−p+1 for 1 ≤ p ≤ n,
if b1 ≥ b1 ≥ . . . ≥ bn , then jp = λ1 − λp+1 for λn+1 = 0.
The sequences {jp } are obviously the same for b and bι . The coincidence of
b for b and bι is a general fact for any root systems, reflecting that Pb
the factors n
and Pbι (as well as Eb and Ebι for b ∈ P ) have the same q, t–integrality properties.
Note that the Young diagram λ(b ι ) for bι is the complement of λ = λ(b) in the
(n + 1) × λ1 –rectangle (naturally containing λ) rotated by 180◦ .
The inequalities in (2.19) are sufficient but of course not necessary for the
relations jp = bn + . . . + bn−p+1 . For instance, take any non-increasing sequences
(s) (s)
B (s) = {bm ≥ . . . ≥ b1 } of the same size m and such that B (s) ≥ B (s+1) element-
wise. Then jp = λn−p+1 for all p in the sequence {B (1) , B (2) , . . .} of the size
n = m + m + . . ..
Finally, using (2.6,2.7) and the formula mp = p for the Coxeter exponents of
type An , we obtain that
p −1
1 − q j tp
n j
(2.20) b =
N .
p=1 j=1
1 − qj t
b = N
Our theorem states that E b E ◦ and Pb = N
b P ◦ are q, t–integral for any b ∈ P+
b b
in this case.
Classical J–polynomials. Let us provide their definition; we emphasize that only
the An –version (for sln+1 ) of the J–polynomials will be given and used (not that
for gln+1 ):
def

(2.21) Jλ == hλ Pb for λ = λ(b), hλ = (1 − q arm(2) tleg(2)+1 ).
2∈λ
This polynomial is q, t–integral.

Here arm(2) is the arm number , which is the number of boxes in the same
row as 2 strictly after it; leg(2) is the leg number , which is the number of boxes
in the column of 2 strictly below it in the –type presentation. Namely, λ1 ≥
λ2 , . . . , λn−1 ≥ λn are the numbers of boxes in the corresponding rows and the ith
row is above the (i + 1)th.
Equivalently:
∗ −1

n λp
(2.22) Jλ = t−(ρ,b) 1−q j t p+1 Pb◦ , p∗ = n−p+1, b ∈ P+ .
p=1 j=0
See, for instance, Theorem 2.1 from [GN]. This makes Jλ very similar to our Pb –
polynomials, but our multipliers are generally greater than those in Jλ ; see below.
Note that the arms and legs do not appear here (in the approach via Pb◦ ). Let us
identify (2.22) and (2.21) for the sake of completeness and because we are doing
the sln+1 –case.
The extra binomial factors necessary to obtain the classical Jλ from Pb◦ =
Pb /Pb (t−ρ ) are as follows. For any box in the (n − p + 1)th row of λ, the t-
power of the corresponding binomial factor from Pb◦ to the classical Jλ must be
(n + 1) − (n − p + 1) + 1 = p + 1. The range of the powers of q in this row is
{0, 1, . . . , λn−p+1 − 1}. So this is what formula (2.22) does.
The product in (2.22) before Pb◦ is {Jλ }ev = Jλ (t−ρ ). It is important that if
one introduces one more variable a = −tn+1 , then this evaluation depends only on
a and the Young diagram λ. Namely,

p −1

n λ
Jλ (t−ρ ) = t−(ρ,b) 1 − q j tn−p+2
p=1 j=0
p −1

n λ
2 −λ1 /4 |λ|/2
(2.23) = (a ) t 1 + q j a t−p+1 .
p=1 j=0
This is one of the main points of the approach from [GN].

Comparing Jλ and Pb . First of all, Macdonald’s arms and legs a(), l()
disappear in the approach from (2.22) due to our division of Pb by its evaluation
at t−kρ . Technically, we use here the so-called co-arms and co-legs , but they are
simply the coordinates of the boxes in λ, which can be directly spelled in terms of
b and the root system.
Our usage of Pb◦ as the starting point for the definition of the J–polynomials
is of course not just to make their definition less combinatorial. It is very natural
because of the construction of the DAHA-polynomials for iterated torus knots.
Also, the spherical normalization Pb◦ is the key for the difference Mehta-Macdonald
formulas from [Ch8], which will be considered below.
Recall that in our construction of Pb , we omit j = 0 in
(2.22), i.e. drop
the binomials containing only t, and also divide the binomials 1 − (q j t)mi +1 by
(1 − q j t) in our construction of the integral form.
Ignoring the binomials without q in the formula for Jλ obviously does not
influence the q, t–integrality; this makes the integral form smaller (but may impact
the symmetries, including the super-duality of the DAHA superpolynomials). The
second reduction is more interesting; it generally diminishes J. For instance, our
construction correctly gives the minimal denominator of Pn in the case of A1 ,
which is (1 − tq n−1 )(1 − tq n−2 ) · · · (1 − tq n−[n/2]
). The J–regularization requires
the multiplication of Pn by the whole product nj=1 (1 − tq n−j ).
With these reservations, the exact match between P and J is upon the relations
jp = λn−p+1 or jp = λ1 − λp+1 (for all p). Then our Pb = N b P ◦ essentially become
b
Jλ for λ = λb ; see (2.19). This conditions are restrictive; generally the binomials
from N b have powers of q larger than those needed in Jλ (with the same t–powers).
Such an increase of our Pb vs. Jλ is mainly because our construction is uniform
and does not use specific features of the DAHA intertwining operators in type An
(quite special). For instance, the root system Am with m equal to the number of
rows of λ = λ(b) is sufficient to find the denominator of Pλ . See Theorem 3.5.1
and Corollary 3.5.2 from [HHL], where the integral forms of E–polynomials were
effectively calculated using the intertwiners, and Corollary 5.1.2 there concerning
the J–polynomials (which are not always optimal as integral forms).
We note that the factors N b coincide for b and bι in our approach, since the
integrality properties of the P –polynomials are unchanged under ι . This cannot be
true in the theory of J–polynomials, which is designed for undetermined ranks (and
aimed at the positivity). By the way, this readily results in a sharper integrality
for a fixed An and sufficiently large diagrams λ if the inequality j < λp∗ in (2.22)
for p ≥ 2 is replaced by min{λp∗ , λ1 − λp+1 }; recall that λp∗ = bn + . . . + bn+1−p

and λ1 − λp+1 = b1 + . . . + bp = λp∗ (b ι ).
Because of the symmetry b → b ι our integral forms Pb are generally larger than
Jλ for λ = λ(b), but not always. Let us provide examples. First of all, Pb = N b P ◦
b
for b = mωn is actually smaller than Jλ for λ = λ(mωn ). Indeed, we do not need
the binomials without q and replace those for non-rational affine exponents by their
certain factors; the A1 –case was discussed above. However, N mω coincides with
1
Nmω in our approach, which makes the former generally greater than Jλ∗ /P ◦ι for
n b
bι = mω1 and λ∗ = λ(b ι ). This ratio is a product of m binomials vs. approximately
n(m − 1) factors in our N mω , a significant increase.
1
Concerning the non-A root systems, the relations of our integral forms to for-
mulas from [RY, OS] can be expected. However our N b cannot be obtained (at
least directly) from the formulas there; we use the spherical normalization. Both
approaches are based on the construction of the E–polynomials via the DAHA in-
tertwiners from [Ch7] (due to Knop and Sahi for An ) and connections are likely.
We note that the J–type integral forms of Pb can be defined for classical root
systems, which is similar to type A, but this will not be discussed here.
To recapitulate, there are two outcomes of our considerations:
a) Jλ can be naturally given entirely via the root system An , if one introduce
them in terms of the spherical normalization Pb◦ of Pb , which allows avoiding the
usage of Young diagrams for this, including Macdonald’s arms and legs , necessary
if Pb is used as the starting point;
b) certain counterparts of J–polynomials can be defined for any root systems
(reasonably close to them, but without the n–stabilization), essentially matching
the Jλ for λ = λ(b), b ∈ P+ (in type An ) such that maxi {bi + . . . + bi−p } is reached
at i = n for any p ≥ 0, where i > p.
3. Topological vertex
3.1. Theta-functions. The DAHA-vertex is closely related to the difference
Mehta-Macdonald formulas, which require the theta-functions . They are defined
for the root system R and will depend in this work on the choice of the character
u : Π = P/Q → C∗ . We denote this group of characters by Π and the trivial
character by 1 . Let
def
def
(3.1) θu (X) == u(b)q (b,b)/2 Xb = ζu,1 (θ) for θ == θ1 .
b∈P
The characters u play here the role of the classical theta- characteristics ; they
were important in [ChF] (necessary for the level-rank duality), though we used a
somewhat different setting there. Namely, they were introduced using the partial
summations θ , where the images of b are taken from subsets ⊂ Π, instead of
our using the characters u here. The difference is that we obtain another basis
in the same space (of theta-functions); using θu is more convenient in the present
work.
The immediate DAHA-motivation of this definition is the following lemma,
where the parameters q, tν are assumed generic.
Lemma 3.1. The formal conjugation by θu−1 in a proper completion of H H or

that for End(V), where V is the polynomial representation, induces the following
DAHA–automorphism: τ+u = τ+ ζ1,u = ζ1,u τ+ . More exactly,
(3.2) (θu )−1 Hθu = τ+u (H) for H ∈ H
H, u ∈ Π .
Proof. This is standard for u = 1 (see [Ch1]), let us outline the enhancement
due to u. We note that for practical calculations, it is convenient to replace here θ
by q −x /2 for X = q x ; all formulas remain the same.
2
The theta-functions above are W –invariant, so it suffices to check that in a

proper completion of V,
T0 (θu ) = τ+u (T0 )(θu ) and Yr (θu ) = τ+u (Yr )(θu ) for r ∈ O.
Recall that τ+u (T0 ) = τ+ (T0 ), so only the latter relation must be checked:

u(b)q −(b,ωr ) Xb q b /2 = q −ωr /2 u(ωr )Xωr θu = τ+ ζ1,u (Yr ) .
2 2
Yr (θu ) =
b∈P
u −1
Combining this lemma with the action (τ̇− ) in V, we obtain that the semi-
u −1
group generated by (τ± ) for u ∈ Π acts in the space linearly generated by the
products Xb θu1 · · · θul for any levels l. We add here q 1/(2m) to the ring of constants
Zq,t , as in Section 1.3. The products θu1 · · · θul are considered as Laurent series
1/2
with the coefficients in Z[tν ][[q 1/(2m) ]]. Thus we switch from V to a certain sub-
space V l of Laurent series. Considering ⊕l V l , one can define there the (projective)
action of GL2 (Z).
This can be used to obtain interesting DAHA–invariants of iterated torus links,
but generally of more involved nature. For instance, we obtain a connection of the
DAHA-vertices for any levels l (see below) with the DAHA-Jones polynomials.
3.2. Mehta-Macdonald identities. We follow Theorem 3.4.5 from [Ch1],

enhancing it by characters u ∈ Π . It is worth mentioning that the passage from
the E–polynomials to the P –polynomials is not always fully clarified in [Ch1]; we
will do this in this work.
def
We denote the constant term of Laurent series f (X) by f . Let μ == μ/μ,
where
∞
(1 − q (ρk ,α)+iνα )2
(3.3) μ = .
α∈R i=1
(1 − tα q (ρk ,α)+iνα )(1 − t−1
α q k
(ρ ,α)+iνα )
+
(z,α∨ )
Recall that qα = q να , q (z,α) = qα , tα = qαkα . This formula is equivalent to the
celebrated Macdonald constant term conjecture (see [Ch1]).
def
The pairing f, g == f g μ for f, g ∈ V plays a major role in the theory; it
is served by the anti-involution of H H from (1.13). Thus the operators Tw , Xa , Yb
are all unitary with respect to it.
We will use Theorem 5.1 from [Ch8] (the first formula) and Theorem 3.4.5 from
[Ch1]. There are other Mehta-Macdonald-type formulas in these works, including
the ones in terms of the Jackson integral ; we will not use them in this work. Recall
that
Eb◦ (q c
) = Ec◦ (q b
), b == b + wb−1 (ρk ), wb (b ) = b+ + ρk ,
def
and Pb◦+ (q c+ +ρk ) = Pc◦+ (q b+ +ρk ), where the latter follows from the former considered
for b− = w0 (b+ ); we use that (b− ) = b− − ρk = w0 (b+ + ρk ).
See (2.4). We will also need (1.36):
lsht (w)/2 llng (w)/2
ΠR Pb◦ = P+ (Eb◦ ) for P+ ==
def
(3.4) tsht tlng Tw .
w∈W
Recall that ΠR is the Poincarè polynomial (with tsht , tlng ).

Theorem 3.2. For b, c ∈ P in the first following formula and b, c ∈ P+ in the
second,
(3.5) u(b + c)Eb◦ Ec◦ θu μ = q (b
,b
)/2+(c
,c
)/2−(ρk ,ρk ) Ec◦ (q b
)θμ ,
u(b + c)Pb◦ Pc◦ θu μ = q b Pc◦ (q b+ρk )θμ .
2
/2+c2 /2+(b+c,ρk )
(3.6)
Here the coefficients of μ are naturally expanded in terms of positive powers of

q and the proportionality factor is a q–generalization of the Mehta -Macdonald
integral:
∞
(ρk ,α∨ )+j
1 − t−1
α qα
(3.7) θμ = (ρ ,α∨ )+j
.
α∈R+ j=1 1 − qα k
Proof. The first formula is obtained from that in [Ch8] using that
(3.8) θu = ζu,1 (θ), ζu,1 (Eb ) = u(b)Eb , and ζu,1 (f (X)) = f (X)
for u ∈ Π and any Laurent series f (X). It results in the following important
lemma.
Lemma 3.3. For an arbitrary Laurent series fc (X) such that ζu,1 (fc ) = u(c)fc
and a W –invariant one gc (X) with the same symmetry,
(3.9) u(b + c)Eb◦ τ̇− (fc ) θu μ = q (b

,b
)/2−(ρk ,ρk )/2 fc (q b
)θμ ,
(3.10) u(b + c)Pb◦ τ̇− (gc ) θu μ = q (b,b)/2+(ρk ,b) gc (q b+ρk )θμ ,
provided the convergence. Here we use the action of τ̇− in the polynomial represen-
tation V; see ( 1.24). We assume that b ∈ P in the first formula and b ∈ P+ in the
second; note that u(b) = u(w(b)) for any w ∈ W, u ∈ Π .
Proof. The polynomials Ec , Pc (or those with ◦) satisfy the condition from the
lemma on fc , gc . Namely,
(3.11) ζu,1 (Ec ) = u(c)Ec , ζu,1 (Pc ) = u(c)Pc .
Formula (1.37) gives that (with or without ◦) :
(c+ , c+ )
τ̇− (Ec ) = q − −(c+
Ec = q −(c
/2+ρk /2) Ec for c ∈ P.
2 2
, ρk )
(3.12) 2
Then we substitute an arrive at (3.9). Formula (3.10) is obtained by the sym-

metrization inside . . . in the left-hand side of (3.9) for symmetric gc and b = b− ∈
P− . Namely, let us use the relations

Tw (f ) , g = f , ((Tw ) (g)) = f , Tw−1 )−1 (g)
for any f (X), g(X). Then (3.4) above considered for b = b− = w0 (b+ ) ∈ P− gives
that

u(b + c) ΠR Pb◦+ τ̇− (gc ) θu μ = u(b + c) P+ Eb◦ τ̇− (gc ) θu μ

=u(b + c)(P+ (1)) Eb◦ τ̇− (gc ) θu μ = u(b + c)ΠR Eb◦ τ̇− (gc ) θu μ
2
= ΠR q (b
,b
)/2−(ρk ,ρk )/2 gc (q b
) θμ = ΠR q b+ /2+(b+ ,ρk ) gc (q b+ +ρk ) θμ .
Dividing by ΠR , we arrive at (3.10). Formula (3.6) from the theorem is its particular
case, which concludes the justification of the theorem.
3.3. Norm-formulas. Let us provide the norm formula for the Macdonald
E–polynomials (see e.g., [Ch6]), the Main Theorem):
(1 − q j t−1 Xα (q ρk ))(1 − q j tα Xα (q ρk ))
(3.13) Eb , Ec = δbc α α
j j
α
.
+ (1 − qα Xα (q ))(1 − qα Xα (q ρk ))
ρ k
{α,j}∈Λb
The corresponding formula for the symmetric polynomials Pb (b = b+ ), the Macdo-

nald norm conjecture, reads as:
∨
(α
,b)−1
(1 − q j+1 t−1 Xα (q ρk ))(1 − q j tα Xα (q ρk ))
(3.14) Pb , Pc = δbc α α α
.
α>0 j=0 (1 − qαj Xα (q ρk ))(1 − qαj+1 Xα (q ρk ))
Note that Pb , Pb = Pbι , Pbι = Pb , Pb ∗ , as well as for Pb◦ , Pb◦ used below.
We will need formula (3.4.1) from [Ch1] and its symmetric variant:
μ−1 (q b
) tα − q j Xα (q ρk )
(3.15) Eb◦ , Ec◦ = δbc −1 −ρ = δbc α
,
μ (q k ) + 1 − qαj tα Xα (q ρk )
{α,j}∈Λb
∨
1 − Xα (q ρk ) (α
,b)−1
tα − qαj+1 Xα (q ρk )
Pb◦ , Pc◦ = δbc
α>0
1 − Xα (q ρk +b ) j=0 1 − qαj tα Xα (q ρk )
δ −1 (q b+ρk ) def
1 − X −1
α
(3.16) = δbc −1 ρk for δ(X) == μ(X) −1 .
δ (q ) α>0
1−t X
α α
Respectively, one takes b, c ∈ P in the first formula and b, c ∈ P+ in the second.

The latter formula follows from the former for b = b+ ∈ P+ . To see this, one can
proceed as follows:
P P+ ◦ ◦
Eb◦ , Pc◦ = Eb◦ (Pc◦ ) , + (1) = Eb (Pc ) , 1
ΠR ΠR
P+
◦
= Eb , Pc◦ = Pb◦ , Pc◦ .
ΠR
Then we switch from μ to δ or expand Pc in terms of the E–polynomials and use
their orthogonality relations; see formula (3.3.15) in [Ch1], which is direct from
the technique of DAHA intertwiners.
Formula (3.16) can be of course deduced from (3.14) and (1.33):
∨
,b)−1 ρk
(α 1−qαj+1 t−1
α Xα (q ) 1−qαj Xα (q ρk )
(3.17) Pb◦ , Pc◦ = δbc ,
α>0 j=0 1−qαj+1 Xα (q ρk ) t−1 j
α −qα Xα (q k )
ρ
though the simplest justification of formula (3.14) for Pb◦ , Pb◦ is via Eb◦ , Eb◦ ,
which is conceptually connected with μ−1 (q b
).
3.4. The case of An . Continuing for any root system, we cancel coinciding
factors in (1.34), but do not perform any further divisions:
(1 − q j ti )
Pb (q −ρk ) = q −(b,ρk ) α α
j i
,
(1 − qα tα )
Then Pb◦ , Pc◦ = δbc q 2(ρk ,b) /(fb fb ) for
(1 − q j ti ) (1 − q j+1 ti−1 )
α α α α
(3.18) fb = , fb = .
(1 − qαj +1 tiα −1 )

(1 − qαj tiα )
Let us now apply this formula to (2.23) in the case of An using

(3.19) Pλ = Jλ /hλ for hλ = (1 − q arm(2) tleg(2)+1 ).
2∈λ
Recall that arm(2) is the number of boxes strictly after 2 (in the same row) and
leg(2) is the number of boxes strictly below it. See (3.19).
Proposition 3.4. For λ = λ(b), ν = λ(c), b, c ∈ P+ in type An , we obtain that
(a2 )λ1 /2 hλ hλ
Pλ◦ , Pν◦ = δλ,ν |λ| for hλ = (1 − q arm(2)+1 tleg(2) ),
t gλ gλ
2∈λ
p −1 p −1

n λ
n λ
(3.20) gλ = 1+q j a t−p+1 , gλ = 1+q j+1 a t−p .
p=1 j=0 p=1 j=0
Proof. We combine (3.18) and (2.23):

p −1
λ1 |λ|
n λ
(3.21) Jλ (t−ρ ) = (a2 )− 4 t 2 1 + q j a t−p+1 for a = −tn+1 .
p=1 j=0
Note that the formulas for our norms of Pb are similar to (but do not coincide
with) those for Macdonald’s stable pairing; see e.g., [GN], formula (6). Under his
pairing, the denominator will be gλ2 :
gλ
Pλ◦ , Pν◦ stab = δλ,ν P ◦ , P ◦ , where gλ = gλ (a → aq/t).
gλ λ ν
An –stabilization. Let us summarize the stabilization properties of Pb (b ∈ P+ )
and the formulas above; this will later result in the a–stabilization of the DAHA-
vertex and DAHA-superpolynomials.
First of all, Pb can be naturally lifted to the root systems gln+1 . Namely,

to Pλ that is a polynomial in terms of positive powers of the variables Xεi (with
the natural action of W = Sn+1 by permutations of i) obtained from Pb by the
substitutions

c
Pb & Xc → Xεc11 +...+cn Xεc22 +...+cn · · · Xεcnn Xε1 · · · Xεn+1 0 ,
n
where c = i=1 ci ωi and c0 ∈ Z+ is adjusted to ensure that the degrees of all
monomials in Pλ are |λ|. Note that c0 = 0 for the leading monomial Xb in Pb . In
the opposite direction, we obtain Pλ by

− |λ|
(3.22) Pb = Xε1 · · · Xεn+1 n+1 Pλ for λ = λ(b),
where we express the right-hand side in terms of Xi = Xωi for i = 1, . . . , n using

(2.17).
Then the passage from gln+m+1 to gln+1 (accordingly, from An+m to An ) is
simply by letting Xi = 0 for i > n + 1. The same stabilization holds for the
E–polynomials (one can use the DAHA-intertwiners for them), but we will not use
this here.
Similarly, θu for gln+1 are as follows:
2
θu = q mi /2 u(m1 + . . . + mn+1 ) Xεm1 1 · · · Xεmn+1
n+1
,
{mi }∈Zn+1
where u is any homomorphism u : Z → C∗ . This definition is stable with respect

to n in the same sense as for Pλ . Then all formulas above can be transformed
to
gln+1 , The definition of μ remains unchanged and switching to ω i = ij=1 εj from
ωi is natural. Then latter is actually necessary mainly to avoid the denominator
(n + 1) in
n b2
= i=1 λ 2
i − |λ|2 /(n + 1) for λ = λ(b); see Section 2.6. Then b2 will
2
become i=1 λi , i.e. will depend only on λ.
We do not actually need DAHA theory for the root system gln+1 in this work.
It is convenient to address the a–stabilization, but we can state and check almost
all stabilization claims within sln+1 , which makes it better compatible with our
consideration of arbitrary (reduced irreducible) root systems R. The passage to
gln+1 may change the formulas only by some q, t–monomial factors, which are
generally ignored in the constructions below, as well as in [Ch2, Ch3, ChD].
Proposition 3.5. Given two Young diagrams λ and μ, the values Pλ (q μ+kρ )
are a–stable, which means that there is a universal expression in terms of q, t, a
such that its value at a = −tn+1 coincides with Pb (q c+kρ ) for λ = λ(b), μ = λ(c),
b, c ∈ P+ for An with n no smaller than the number of rows in λ and in μ. Up
to powers of a1/2 and t1/2 , they are rational function in q, t, a. Also, Pλ◦ (q μ+kρ ),
Pλ , Pλ and Pλ◦ , Pλ◦ are a–stable (in the same sense).
Proof. The claim about the a–stability of Pb (q c+kρ ) is direct from the stability
of Pλ for λ = λ(b). The correction factor from (3.22) does not contribute to these
c+kρ
values. One needs only to know the a–stability of Mb (q ) for the standard
symmetric monomials Mb = c∈W (b) Xc . Other claims follow from this and the
explicit formulas provided above.

3.5. High–level 3j–symbols. We are back to an arbitrary root system R.
We will provide in this work only the symmetric version of DAHA-vertex. The case
of E–polynomials is quite parallel, but we do not see at the moment its applications
(beyond the DAHA theory). For l ∈ Z+ , called the level of theta-function, and for
an unordered set u = (ui , 1 ≤ i ≤ l) ⊂ Π = Hom(Π, C∗ ), we put:

l
Pbu,◦ == Pb◦ θu for θu = θui , Pa◦ Pb◦ θu = c,u ◦
def
(3.23) Ca,b Pc ,
i=1 c∈P+
where a, b, c ∈ P+ . Here θu are considered as Laurent series in terms of Xb with

the coefficients that are formal series in terms of positive powers of q. Analytically,
the convergence of their coefficients is granted for |q| < 1. Here and below we use
that {Pcu,◦ , c ∈ P+ } is a basis in the space V W θu of W –invariant vectors in Vθu ,
1/2
naturally considered over the ring C[[ q 1/(2m) ]] [tν ].
Similarly, Cbc,u will be defined from the relation

c,u
Pbu,◦ == Pb◦ θu = Pb◦1 . . . Pb◦k θu = Cb Pc◦ ,
def
(3.24)
c
where b = (bi , 1 ≤ i ≤ k) ⊂ P+ & c.

We put u = ∅ for l = 0 and, similarly, use ∅ if the corresponding θu is missing
c,∅
in θu . The coefficients Ca,b are a generalization of the classical 3j–symbols (as
tν = qν ).
The key example for us will be l = 1; then we write Pbu,◦ and Ca,b c,u
for u ∈ Π .
Arbitrary levels can be reduced to l = 0, 1, as we will see. Actually l = 1 is formally
sufficient due to the relation
d,∅ c,u
(3.25) c,u
Ca,b = Ca,b Cd,0 for a, b, c ∈ P+ , u = 1 ∈ Π .
d∈P+

d,∅ c,1
To be more exact, Ca,b can be expressed in terms of Ca,b if the inverse of the

c,u
matrix Cd,0 , c, d ∈ P+ is known; recall that 1 is the trivial character of Π = P/Q.
The latter matrix essentially controls the DAHA Fourier transform, so its inverse
can be presented in a similar form; we will not discuss this here. Relation (3.25)
and similar relations for any levels readily follow from the fact that {Pcu,◦ , c ∈ P+ }
is a basis in V W θu .
Proposition 3.6. For any b, c ∈ P+ , using ( 3.6) and ( 3.16):
Pc◦ι
2 2
q b /2+c /2+(b+c,ρk ) ◦ −c−ρk
(3.26) c,u
Cb,0 = Pb◦ θ u μ
= P (q )θμ ,
Pc◦ , Pc◦ u(b−c)Pc◦ , Pc◦ b
(3.27) where Pc◦ , Pc◦ = Pc◦ι , Pc◦ι = δ −1 (q c+ρk )/δ −1 (q ρk )

∨
1 − Xα (q ρk ) (α
,c)−1
tα − qαj+1 Xα (q ρk )
= .
α>0
1 − Xα (q ρk +c ) j=0 1 − qαj tα Xα (q ρk )
def
Similarly, for b = (bi ) ⊂ P+ & c and Pb◦ == i Pb◦i ,
(3.28) Cbc,u = Pb◦ Pc◦ι θu μ /Pc◦ , Pc◦

−1
Pb◦ Pc◦ι θμ τ̇− (Pb◦ Pc◦ι )(q −ρk )θμ
= = .
u(Σi bi − c) Pc◦ , Pc◦ u( i bi − c) Pc◦ , Pc◦
1/2
Therefore Cbc,u belong to C(q 1/(2m) , tν )θμ

; the

denominators
are the products
of binomials from ( 2.3), i.e. in terms of 1 − q • t•sht t•lng with non-negative powers
(strictly positive for q).

The following associativity theorem is direct from the definition of Cbc,u ; we use
that Pc◦ are linearly independent.
Theorem 3.7. (i) Let us represent the sets b, u as unions

= (b1 ), . . . , (bm ) , u
b = (u1 ), . . . , (um ) ,
where the decompositions b and u

are obtained from b and u by adding a proper
number of 0s for the first and ∅s for the second when necessary (at any places) to
make them of the same length m; the size of bi can be different from that of ui .
Then
def
(3.29) Cbc,u = Ξc,
u
==
b
1 2 3 m
Cbc11,u C(c
c2 ,u c3 ,u c,u
2 C(c ,b3 ) · · · C(c
1 ,b ) 2
m .
m−1 ,b )
c1 ,c2 ,...,cm−1 ∈P+
For instance, this gives that Ξc,

u
does not depend on the choice of the decomposi-
b
and u
tions b and depends only
on the sets b,u and
c.
= (u1 ), . . . , (um ) , b = (b), (0), . . . , (0) :
(ii) In particular, for u

(3.30) Ξc,u = c1 ,u1 c2 ,u2 c3 ,u3
Cb,0 Cc1 ,0 Cc2 ,0 · · · Ccc,u m
m−1 ,0
,
b
c1 ,c2 ,...,cm−1 ∈P+
b,u
where Ca,0
are given by formula ( 3.26).

= (b0 , b1 ), (b2 ), . . . , (bm ) . Then
More generally, let b

(3.31) Ξc,
u
= Cbc01,b
,u1 c2 ,u2 c3 ,u3
Cc1 ,b2 Cc2 ,b3 · · · Ccc,u m
m−1 ,bm
b 1
c1 ,c2 ,...,cm−1 ∈P+
c,u
depends only on the (unordered) sets (bi ), (ui ) and c, where Ca,b are given by ( 3.28).
For instance, the 3-vertex associativity conditions hold:
c ,u c,u c ,u c,u c ,u c,u
(3.32) Cb01,b11 Cc1 ,b22 = Cb11,b21 Cc1 ,b20 = Cb01,b21 Cc1 ,b21 .
c1 ∈P+ c1 ∈P+ c1 ∈P+
Note that the order of (u1 , u2 ) influences the summations in (3.32), though the
output does not depend on it. Indeed,
,1 c,1

Cbc01,b
,u1 c,u2
1
Cc1 ,b2 = Cbc01,b 1
Cc1 ,b2 / u1 (b0 + b1 − c1 )u2 (c1 + b2 − c) ,
and the denominator here depends on c1 unless u1 = u2 . Permuting ui in u from
(3.29) leads to quite non-trivial identities. This includes the 2–vertices from (3.30);
the corresponding identities considered under tν = 0 played an important role in
[ChF], with a direct link to the level-rank duality in the case of An .
In a greater detail, formula (3.30) generalizes Rogers-Ramanujan-type sums in
[ChF], which were obtained for tν = 0, b = 0 and for minuscule weights c. Then
the corresponding sums will be q–modular functions under a certain normalization.
The quantities Pb (q c+ρk ) disappear from (3.26) under tsht = 0 = tlng , which signifi-
cantly simplifies the theory. The modular properties of general Ξc, u
are not known,
b
though the case tν = qν is actually similar to tν = 0.
Relations (3.32) are basic for TQFT, where they are (mainly) considered at
roots of unity, which makes the sums involved finite. They result, for instance, in
the WRT invariants of links if the Quantum Group generalization of the 3j-symbols
is taken as the starting point. Our approach is compatible with the passage to roots
of unity, but this will not be discussed in this work.
Another important property of the topological vertex needed in TQFT is its
S3 –symmetry, though this is not always assumed (especially in the refined theo-
c,u c,u
ries). See e.g., [AKMV, GIKV, AFS]. We obviously have that Ca,b = Cb,a . The
invariance with respect to (a, b, c ) → (a, c, b ) holds upon multiplication of our
ι ι
c,u
Ca,b by Pc◦ , Pc◦ , which readily follows from (3.28). Thus we essentially have the
S3 –symmetry for the DAHA-vertex.
3.6. The coinvariant approach. Let us provide an alternative approach

c,u
to level-one Ca,b , which clarifies their S3 –invariance and connects them with the
DAHA polynomials associated with links (with the Hopf links, to be exact).
The key for such an approach is Lemma 3.3. We will begin with the case of
E–polynomials with two sets b = (bi , 1 ≤ i ≤ k), c = (cj , 1 ≤ j ≤ m) ∈ P and (a
single) u ∈ Π . The case of arbitrary sets u (i.e. arbitrary levels) requires general
theory of links, which is beyond this section. Following the notation from (3.24),
let Eb◦ = Eb◦1 · · · Eb◦k (and for c). The same notation will be used without ◦ and
(later) for P –polynomials. Recall from (1.19):
τ+u = τ+ ζ1,u = ζ1,u τ+ , τ−
u
= ζu,1 τ− = ζu,1 τ− , σ u,v = ζuv−1 ,uv σ.
For any g, h ∈ C[Xa ] (in particular, for g = Eb◦ , h = Ec◦ ) and for the coinvariant
{.}ev from (1.20), we set:
def - 1/u
u −1
. - u −1
.
Egh,u == (τ+ )−1 h(Y −1 ) (τ− ) g(X) ev= (τ− ) h(X)g(X) ev ,
- −1
◦ −1 −1
◦ . - −1
◦ .
c,u τ+ Ec (Y ) τ− Eb (X) ev τ− Ec (X)Eb◦ (X) ev
(3.33) Eb = = .
u(Σi bi + Σj cj ) u(Σi bi + Σj cj )
u −1 u
The action of τ± here and below is that in H
H. We use that σ u,u = τ+u (τ− ) τ+ =
ζ1,u2 σ, which gives that

(τ+ )−1 h(Y −1 ) = (τ+ )−1 σ h(X) = (τ+u )−1 σ u,u h = (τ−
u −1
1/u 1/u
) h .
Then the action of ζu,1 on the E–polynomials from (3.11) is applied. Note the
g ↔ h– symmetry of (3.33) and that Ebc,u depends only on the (unordered) set
b ∪ c.
It is worth mentioning that for any g(X),

−1
{φ τ− g(X) }ev = {τ+−1 φ g(X) }ev

= {τ+−1 σ g(X) }ev = {τ− −1 −1
τ+ g(X) }ev = {τ− g(X) }ev .
−1
This clarifies our using the inverses τ± for Ebc,u ; if they are replaced by τ± , then
Y −1 must be replaced there by σ −1 (X).
The same formulas hold for the P –polynomials; we set Pgh,u = Egh,u for sym-
metric g, h. Then for b = (bi ) ⊂ P+ ⊃ (cj ),
- −1
◦ −1 −1
◦ . - −1
◦ .
c,u def τ+ Pc (Y ) τ− Pb (X) ev τ− Pc (X)Pb◦ (X) ev
(3.34) Pb == = .
u(Σi bi + Σj cj ) u(Σi bi + Σj cj )
There is a simplification in the case of a single c ∈ P (or b) due to the following

general identity based on (3.12). Recall that b2 /2 − ρ2k /2 = b2+ /2 + (b+ , ρk ); see
(2.4), (3.12) and Lemma 3.3. For any m ∈ Z, H ∈ H H and g = g(X), one has:
-
. c2 - .
) Ec◦ (X) ev = u(mc)q −m( 2 +(c+ ,ρk )) H · Ec◦ (X) ev ,
+
(3.35) H · (τ−
u m
- u m
◦ . c2 - u m
◦ .
(τ− ) g Ec ev = u(mc) q −m( 2 +(c+ ,ρk )) (τ−
+
) g Ec (X) ev ,
u m
where the second follows from the first for H = (τ− ) (g(X)).
Proposition 3.8. For h = Ec◦ (c ∈ P ) and any g(X),

-
.
Egh,u = u(−c)q c+ /2+(c+ ,ρk ) Ec◦ (Y −1 )(τ−
u −1
2
(3.36) ) g(X) ev ,
q c+ /2+(c+ ,ρk ) - ◦ −1 −1
◦ .
2
Ebc,u= E (Y )τ− Eb (X) ev

u(c +b1 + . . . +bk ) c

◦
◦ ◦ −ρ
2
−1 −1
q c+ /2+(c+ ,ρk ) τ̇− Eb (X) (q c
) τ̇− Ec Eb (q k )
= = ,
u(c + b1 + . . . + bk ) u(c + b1 + . . . + bk )
Ec◦ Eb◦ θu μ
(3.37) Ebc,u = for b = (b1 , . . . , bk ) ⊂ P & c,
θμ
−1 −1

−1
where τ̇− : g(X) → τ− g(X) (1) is the action of τ− in the polynomial represen-
tation V & g(X). In the symmetric case,
- ◦ −1 −1
◦ .
2
q c /2+(c,ρk )
(3.38) Pbc,u= P (Y )τ− Pb (X) ev
u(c +b1 + . . . +bk ) c

◦
◦ ◦ −ρ
2 −1 −1
q c /2+(c,ρk ) τ̇− Pb (X) (q c+ρk ) τ̇− Pc Pb (q k )
= =
u(c + b1 + . . . + bk ) u(c + b1 + . . . + bk )
Pc◦ Pb◦ θu μ ◦ ◦
ι(c),u Pc , Pc
= = C for b = (bi ) ⊂ P+ & c ,
θμ b
θμ
where ι(c) = cι , Cbc,u is defined in ( 3.24); see also formula ( 3.28).
Proof. Applying φ inside the coinvariant (twice, back and forth), we use (3.35)
1/u
for m = −1 and that φτ− u
= τ+ φ :
- ◦ −1 u −1
. - 1/u
.
Ec (Y )(τ− ) g(X) ev = (τ+ )−1 g(Y −1 ) Ec◦ (X) ev
c2 - 1/u
u −1
◦ .
= (u(−c)q 2 +(c+ ,ρk ) )−1 (τ+ )−1 g(Y −1 ) (τ−
+
) Ec (X) ev
c2 - 1/u
u −1
.
= u(c) q −( 2 +(c+ ,ρk )) (τ+ )−1 Ec◦ (Y −1 ) (τ−
+
) g(X) ev .
−1
Then we use formula (3.9) from Lemma 3.3 for b = 0 and f = τ̇− (Ec◦ Eb◦ ), where
±1 ±1
the τ̇− (here and in the next formula) is the action of τ− in V, which is g(X) →
±1

τ− g(X) (1). Namely, one has :
u(c + Σi bi ) τ̇− (f ) θu μ = f (q −ρk )θμ

−1 ◦ −ρ
= u(c + Σi bi ) Ec◦ Eb◦ θu μ = τ̇− (Eb ) (q k )θμ .

This proposition readily results in the following important general property of
the evaluation pairing, which is necessary to connect the DAHA approach with the
splice diagrams of the corresponding link.
Theorem 3.9. For arbitrary
f = f (X), g = g(X) ∈ V and the evaluation
pairing {f, g}ev = {f (Y −1 ) g(X) }ev ,
- −1 −1
. - −1 . - −1
.
(3.39) τ̇− (f ), τ̇− (g) ev = τ̇− (f g) ev = τ− f (X)g(X) ev .
Combining this relation with Lemma 3.3 in the notations there and under the as-
sumptions ζu,1 (f ) = u(b)f , ζu,1 (g) = u(c)g for certain b, c ∈ P and all u ∈ Π , one
has for f, g as above:

(3.40) u(b + c) τ̇− (f ) τ̇− (g) θu μ = {f, g}ev θμ .

- .
Also, { τ̇− (f ) (Y ) τ̇− (g)}ev = τ̇− (f g) ev for f ∈ V W , where V W is the subsub-
space of W –invariants.
def
Proof. We extend {f, g}ev to A, B ∈ H H to {A, B}ev == {ϕ(A)B}ev . Then
{A, B}ev = {B, A}ev and for H(1) ∈ V ⊂ H H&H :

{A, B}ev = {ϕ(A)B(1)}ev = {ϕ B(1) A}ev = {A(1), B(1)}ev ,
- −1 −1
. - −1 −1
. - −1 −1
.
τ̇− (f ) , τ̇− (g) ev = τ̇− (f ), τ− (g) ev = τ− (g), τ̇− (f ) ev
- −1 −1
. -
−1 . -
−1 .
= τ− (g) , τ− (f ) ev= τ+−1 σ(g) , τ− (f ) ev= τ+−1 ϕ(g) , τ− (f ) ev
-
−1 −1 . - −1
. - −1
.
(3.41) = ϕ τ− (g) , τ− (f ) ev = τ− gf ev = τ̇− f g ev .
This gives (3.39). Then use Lemma 3.3 and σ 2 = ι in V W .

It is worth mentioning that the relation
c,u ι(c),u ι(c),u
def Cb
c,u = Pb Pb
(3.42) C = = ◦ ◦ =
b
θμ
Pcι , Pcι Pc◦ , Pc◦
is actually entirely conceptual, as well as its nonsymmetric counterpart. However,
we prefer to obtain it via Lemma 3.3, which is of independent interest. The right-
hand side here is convenient to analyze the action of and η. Using (1.21) and
(1.16):
-
◦ .
τ− Pcι (X)Pb◦ι (X) ev
(3.43) c,u
(Pb ) = c,u ) = C
, (C ι(c),u ,
ι ι b bι
u(Σi bi + Σj cj )
where we apply ι = −w0 to b componentwise.
We obviously used here the rationality of the right-hand side of (3.42); otherwise
q = q −1 must be addressed. Note that one can define
c,u = C c,u /θμ l
C b b
for any level l ≥ 0 and u = (u1 , . . . , ul ). Then the relations from Theorem 3.7

for the C–coefficients will hold for the C–coefficients; we mean those based on the
independence of Ξc, u
on the choice of the decompositions u
b, . However, such
b
coefficients are generally infinite series in terms of (non-negative fractional powers
of) q and applying , which includes q → q −1 , is more involved (though doable
thanks to the q, t–setting).
4. DAHA-Jones theory
4.1. Iterated torus knots. We will begin with torus knots and iterated torus
knots; the iterated links will be considered next.
The torus knots T (r, s) are defined for any integers (including 0 and negative
ones) assuming that gcd(r, s) = 1. One has the symmetries T (r, s) = T (s, r) =
T (−r, −s), where we use “=” for the ambient isotopy equivalence. Also T (r, s) =
if |r| ≤ 1 or |s| ≤ 1 for the unknot , denoted by . See e.g., [RJ, EN, ChD] or Knot
Atlas for the details and the corresponding invariants.
Following [ChD], the [r,s]-presentation of an iterated torus knots will be T (!r,!s)

for two sequences of integers of any signs :
(4.1) !r = {r1 , . . . r }, !s = {s1 , . . . s } such that gcd(ri , si ) = 1;
will be called the length of !r,!s. The pairs [ri , si ] are characteristic or Newton
pairs for algebraic knots (such that ri , si > 0). We will call them Newton pairs too
for arbitrary (possibly negative) ri or si , which is allowed in our considerations.
This presentation, referred to as the [r,s]-presentation , will be exactly the one
needed in the DAHA approach. However it is not optimal for establishing the
symmetries of our polynomials and the justification that our construction depends
only on the corresponding knot/link. We actually need the cable presentation for
this, which is actually from the definition of the corresponding iterated torus knots.
It requires one more sequence of integers (possibly negative):
(4.2) a1 = s1 , ai = ai−1 ri−1 ri + si (i = 2, . . . , m).
See e.g., [EN]. In terms of the cabling discussed below, the corresponding knots
are as follows. First, T (r, s) = Cab(s, r)() (note that we transpose r, s here), and
then we set:

(4.3) T (!r,!s) Cab(!a,!r)() = Cab(a , r ) · · · Cab(a2 , r2 ) T (r1 ,s1 ) .
The first iteration here (application of Cab) is Cab(a1 , r1 ), not that for the last
pair , and then we proceed inverting the natural ordering.
Cabling. Knots and links will be always considered up to ambient isotopy ; we
use “=” for it. The cabling Cab(a, b)(K) of any oriented knot K in (oriented) S3 is
defined as follows; see e.g., [Mo,EN] and references therein. We consider a small 2–
dimensional torus around K and put there the torus knot T (a, b) in the direction of
def
K, which is Cab(a, b)(K) (up to ambient isotopy); we set Cab(!a,!r) == Cab(!a,!r)().
This procedure depends on the order of a, b and orientation of K. We choose
the latter in the standard way, matching the Mathematica package Knot Atlas
[KA]; the parameter a gives the number of turns around K. This construction also
depends on the framing of the cable knots; we take the natural one, associated
with the parallel copy of the torus where a given cable knot sits (its parallel copy
has zero linking number with this knot). It will be the same standard framing for
iterated torus links ; see below. Since the DAHA-invariants are considered in this
work up to powers of q, t, this will be sufficient.
Topological symmetries. By construction, Cab(a, 0)(K) = for any knot K and
Cab(a, 1)(K) = K for any a = 0. Accordingly, we have the following reduction
cases :
(4.4) If ri = 0 for some 1 ≤ i ≤ , then T (!r,!s) =
T ({ri+1 , · · · , r }, {si+1 , · · · , s }), T (!r,!s) = for i = .
(4.5) If ri = 1, si ∈ Z for some i, then T (!r,!s) =
T ({r1 , · · · , ri−1 , ri+1 , · · · , r }, {· · · , si−1 , si+1 , si+2 , · · · }),
where si+1 = si+1 +si ri+1 if i < (no s+1 for i = ).
Let us comment on the last relation; see (4.2). Since ri = 1, one has:
ai = ai−1 ri−1 +si , ai+1 = ai ri+1 +si+1 = ai−1 ri−1 ri+1 + (si+1 + si ri+1 ).
The pairs {ai , ri } are sometimes called topological or cable parameters . Indeed, the
isotopy equivalence of iterated knots generally can be seen only at the level of these
parameters (not at the level of Newton pairs).
Next, the symmetry T (r, s) = T (s, r) results in the following transposition and
reduction properties . For any !r,!s,!a and for m ≥ 0,
(4.6) T (!r,!s) = T ({s1 , r2 , . . . , r }, {r1 , s2 . . . , s }),
(4.7) Cab(a , r )· · · Cab(a1 , r1 )(T (m, 1)) = Cab(a , r )· · · Cab(a1 , r1 )().

Then, switching from a knot K to its mirror image , denoted by K :

Cab(a, b)(K ) = Cab(−a, b)(K) for any a, b with gcd(a, b) = 1,

(4.8) and Cab(−!a,!r) = Cab(!a,!r) , T (!r, −!s) = T (!r,!s) .
We note that the Jones and HOMFLY-PT polynomials for K are obtained from
those for K (can be a link) by the formal conjugation of the parameters, which is
q → q −1 , a → a−1 . This will hold for the DAHA-Jones polynomials and DAHA-
superpolynomials, where the conjugation t → t−1 must be added to that of q, a.
Furthermore, changing the orientation , denoted by “−”, at the ith step, we
obtain that for any 1 ≤ i ≤ ,
(4.9) −Cab(!a,!r) = Cab({. . ., ai−1 ,−ai , ai+1 ,. . .}, {. . ., ri−1 ,−ri , ri+1 ,. . .}),
(−) −i+1
T (!r,!s) = T ({. . . , ri−1 ,−ri , ri+1 , . . .}, {. . ., si−1 ,−si , si+1 ,. . .}).
The second transformation here results in the following cable:
Cab({. . ., ai−1 ,−ai , −ai+1 ,. . .}, {. . ., ri−1 ,−ri , −ri+1 ,. . .}),
which explains the sign.
Changing the orientation of a knot or the simultaneous change of the orienta-
tions of all components of a link (equivalently, applying ι to the weights) does not
influence its Jones and HOMFLY-PT polynomials; so the DAHA-Jones polynomi-
als must remain unchanged under such a transformation (they are). We will mostly
use the symbol ∨ for the change of the orientation in this work.
4.2. From knots to links. Switching to links, we need to add colors to

the cables above, which are dominant weights b. The [r,s]-presentation of a torus
iterated link will be a union of κ colored knots

Υ, (bj )
(4.10) L(r j ,s j ) = {T (!rj ,!sj ), bj ∈ P+ }, j = 1, . . . , κ together with
the incidence matrix Υ = (υj,k ), where 0 ≤ υj,k = υk,j ≤ min{j , k },
implie s that [rij , sji ] = [rik , ski ] for all 1 ≤ i ≤ υj,k and any 1 ≤ j, k ≤ κ.
Here j is the length of !rj = {rij } and !sj = {sji } for 1 ≤ j ≤ κ; we naturally set
υj,j = j .
Subject to this above identification (for i ≤ υj,k ), the pairs [rij , sji ] will be treated
as vertices of a natural graph L determined by Υ; these pairs we be called [r,s]-
labels , of the corresponding vertices. The notation [ , ] will be used exclusively for
such labels [r, s].
The paths are the sequences of vertices with fixed j. The vertices for neigh-
boring i in the same path will be connected by the edges; the graph has a natural
orientation from i to i + 1 in each path. Also, we will add the arrowhead at the
end of each path, which is at i = j .
Such an incidence graph (including the arrowheads) is a union of trees, called
subtrees . Every subtree has at least one base path , the one that intersects all other
paths in this component. It also has a unique initial vertex (corresponding to i = 1
in any base path). The colors bj will be assigned to the arrowheads; thus the j th
path corresponds to the knot T (!rj ,!sj ) colored by bj ∈ P+ . The graph can be empty
(no vertices), then it is a collection of paths that are pure arrowheads. There can
be several arrowheads from the same vertex, but one path always has one arrow.
Topologically and in the DAHA construction, one can (technically) assume that
the graph is a tree by adding an extra initial vertex with the label [1, 0] connected
to its all subtrees.
The a–parameters can be calculated along the paths exactly as we did for the
knots (starting from i = 1, a1 = s1 ); then aji depends only on the corresponding
vertex. The pairs {aji , rij } will be called the cab-labels of the vertices. Actually only
the [r,s]-labels will be needed in the DAHA constructions (we will mostly call them
simply labels and use [ , ] only for them). However the cab-labels are necessary to
explain the topological symmetries (including the DAHA-Jones polynomials).
The torus knot colored by b ∈ P+ will be denoted by Tr,s b
. Respectively,
b b
Caba,r (L), equivalently Cab0,1 Caba,r (L), will be the cable Cab(a, r)(L) of a link L
colored by b. The color can be attached only to the last Cab in the sequence of ca-
bles. In the absence of vertices, the notation is b (the unknot colored by b ∈ P+ ) or
Cab(0, 1)b . We mostly use the same notation L for the graph and the corresponding
link L.
The passage from the [r,s]-presentation to the cab-presentation is

κ
(4.11) L !rj ,!sj , 1 ≤ j ≤ κ Cab(!aj ,!rj ) (),
j=1
where the composition and coproduct of cables is with respect to the tree structure
and Cab(!aj ,!rj ) = · · · Cab(aj2 , r2j )T (r1j , sj1 ) is as in (4.3). In this work, the coproduct

symbol (sometimes omitted) is used when a union of cables is applied to the

same link; this union is disjoint but the result of its application will generally have
nonzero linking numbers. For a tree, the cab-presentation begins as follows:

tree
j

tree
j

j

(4.12) ··· (Cabbaj ,rj Cabaj ,rj ) Cabbaj ,rj Cabb0,1 T (r11 , s11 ) .
3 3 2 2 2 2
j =3 j =2 j =1
The coproduct for j = 1 corresponds to pure arrowheads from i = 1 = j, the

next product is over single edges from the initial point followed by arrowheads.
The 2nd and the 3rd coproducts are with respect to the incidence tree, so must be
understood as follows. The third contains

j k

Cabbaj ,rj Cabbak ,rk Cabaj ,rj if υj,k = 2,
3 3 3 3 2 2
i.e. when the vertices labeled by [r2j , sj2 ]

= [r2k , sk2 ]
are identified in the incidence tree.
If υj,k = 3, i.e. when the corresponding paths are different only by the arrowheads,
this product contains

j k

Cabb0,1 Cabb0,1 Cabaj ,rj Cabaj ,rj .
3 3 2 2
This is similar for the product with j = 2 (when υj,k = 2) and in general.
Here and generally Cab(ai+1 , ri+1 )Cab(ai , ri ) means the composition of cabling
operations (for the standard framing). However, we will frequently omit the symbol
of coproduct between cables, when it is clearly not the composition; for instance
the cables in (4.12) of the same “level” i = 2, 3.
Note that the property Cab(a, 1)(K) = K generally holds only for knots K.
Applying Cab(a, 1) to a disjoint
3m union of knots generally ties them up. For instance,
Cab(−1, 1) of m unknots produces the Hopf m–link for the standard
j=1
framing, which is Sm –symmetric and with pairwise linking numbers all equal to
−1. This corresponds to the tree with one vertex [1, −1] and m colored arrows from
it. Note that applying Cab(0, 1) here simply produces a union of m arrowheads,
without any vertices and edges.
Using the symmetry from (4.5) requires recalculating all s after [rij = 1, sji ∈ Z]
in all paths through it. Relation (4.4) holds for links, but now it must be understood
as deleting all vertices in the paths through a vertex with rijoo = 0 from the first
one (i.e. for all 1 ≤ i ≤ io in any paths containing this vertex). The paths which
share some vertices with those affected remain untouched, so the matrix Υ and
the incidence graph must be recalculated accordingly, which can result in extra
subtrees. Also, (4.8) and (4.9) hold if and “−” are applied to the whole link.
4.3. Splice diagrams. Let us extend the previous construction to the case
of a pair of incidence graphs {L, L}, where the latter can be with or without ∨.
The twisted union of the corresponding links is defined as follows. The cabling
construction provides a canonical embedding of the iterated torus links into the
solid torus. We put the links for L and L into the horizontal solid torus and
the complementary vertical one. The presence of ∨ in L∨ means changing the
orientation of this component; for instance, {, } and {, ∨ } represent uncolored
Hopf 2–links with the linking number lk = −1 and +1 correspondingly.
Recall that the notation L or {L, L} is used for trees and also for the corre-
sponding link L (depending on the context).
Using the language of splice diagrams from [EN], the pair {L, L∨ } corresponds
to ◦1 and ◦1 in L and L connected by an arc :
◦1 · · · ◦1 · · ·
◦1 · · · or ◦1 · · ·∨ ,
where adding ∨ changes the orientation of the link with prime. For instance, the
Hopf links for lk = ±1 are represented as ←→ (for +) and ←→ ∨ .

Importantly, the transposition L ↔ L does not change the output if ∨ is present
or absent; the total change of the orientation does not influence the link invariants
we will consider.
This is topologically equivalent to combining the trees via an additional vertex
◦ labeled by [1, ±1], i.e. using the tree ◦ ⇒ ◦◦11 ···
···
, where ◦1 , ◦1 from L, L are
connected via such intermediate ◦ . In the DAHA-construction, this is governed
by Theorem 3.9. The corresponding a–parameters in the cab-labels must be then
recalculated, since we now begin with a new vertex ◦. Thus the twisted union (the
arc-connection) actually results in one tree; see (9.4) for a concrete example.
Let us translate more systematically our combinatorial data into the language
of splice diagrams. See [EN] for details. For one tree L, the construction is as above
(including the directions and arrowheads), but we need to switch from the labels
[r, s] in the [r,s]-presentation to the corresponding {a, r} in the cab-presentation.
Graphically, the passage is as follows:
◦
r 1 .
1
→ ◦ ·→ ··
→ ..
[r, s] a H 1H .
Note that we extend [EN] by adding colors to the arrowheads. Also, recall
that a1 = s1 for the initial vertices in any paths .
In the case of two trees {L, L∨ } (note ∨) the connection by an arc described
above corresponds to the following splice diagram:
◦
r11 1
*
' ...
M
[r1 , s1 ] a1 1 H
1 H
j
◦1 → M ·→ ··
→ ◦
◦1 → M →
·→
·· r11 1
*.
[ r 1 , s 1 ] & M
..
1HH

(4.13) a1 1 j,
where M and M are splice diagrams made from graphs M and M. For the sake of
simplicity of this figure, we assume that L, L have only single edges from their initial
vertices ◦1 , ◦1 to M, M, the remaining portions. The general correspondence is
totally similar.
We note that adding ∨ to M (the change of the orientation of this part) in
the language of splice diagrams is as follows. One adds a new trivalent node to the
arc with weights +1 on the edges going to trees and −1 for the third edge, calling
a leaf ; see [EN], Theorem 8.1, Statements 2 and 3 for the details. The nodes with
weights +1 (both) can be deleted from the diagram; in DAHA, this may result in
some q m , which is trivial due to the normalization. Assigning −1 to the leaf as
above changes the orientation of one of the components; it can be M or M, up
to the total change of the orientation. Algebraically, this will correspond to the
DAHA-relations (4.24), (4.25). See also (9.5).
We actually do not need the splice diagrams too much in this work. However,
they are an important and convenient tool for the classification of iterated torus
links, useful to analyze the topological symmetries. The latter are one-to-one with
the symmetries of our DAHA construction. Also, the splice diagrams establish the
connection with 3-folds and can be effectively used for calculating some invariants,
including the linking numbers, with applications to plain curve singularities.
We note that our method can not give all splice diagrams of solvable type,
though all algebraic links can be reached by our construction. See Theorems 9.2, 9.4
in [EN]). For instance, the DAHA-approach does not provide (so far) the following
diagram, corresponding to the granny knot :

◦
31
◦ HH 1
2 1 H
◦ H1 -
311 0
◦ 1
2 .
It represents the connected sum of two trefoils with the same orientation. The
connected sum is the only one from the three operations in [EN] necessary to obtain
arbitrary solvable links in S3 that is generally missing in the DAHA-approach.
These three are disjoint sum, connected sum and cabling. We note that some 3–
component connected sums can be obtained, for instance the connected sum of two
Hopf 2–links (the chain of 3 unknots, where the first and the last are not linked);
see (7.30) below.
Algebraic links. We provide here only basic facts; see [EN] for details and
references, especially Theorem 9.4 there. Generally, one begins with a polynomial
equation f (x, y) = 0 considered in a neighborhood of an isolated singularity 0 =
(x = 0, y = 0). Its intersection with a small 3-dimensional sphere in C2 around
0 is called an algebraic link . Assuming that rij , sji > 0 , any tree LΥ (rj ,sj ) (in the
[r,s]-presentation) corresponds to a germ of plane curve singularity at 0. If these
inequalities hold, the tree will be called positive .
Such germs are unions of unibranch components for the paths of Υ (numbered
by j), which are given as follows:
j j j j j
j j j j

(4.14) y = cj1 xs1 /r1 (1 + cj2 xs2 /(r1 r2 ) 1 + cj3 xs3 /(r1 r2 r3 ) . . . ) at 0.
The parameters cji ∈ C are sufficiently general here. The simplest example is the
equation y rκ = xsκ under gcd(r, s) = 1, which corresponds to the torus link T (rκ, sκ)
with κ knot components isotopic to T (r, s); the pairwise linking numbers here are
all rs. See also Section 8.4.
The unibranch components and the corresponding pairwise linking numbers
uniquely determine the corresponding germ due to the Reeve theorem. The linking
number between the branches corresponding to the paths with the indices j = k is

j
k
(4.15) lk(j, k) = ajio rijo rij rik , where i◦ = υ(j, k).
i=io +1 i=io +1
This formula correctly gives lk(j, k) = 0 if we set aji◦ = 0 when υ(j, k) = 0 (i.e.
allow graphs L here, not only trees), but then the corresponding links will become
non-algebraic. All linking numbers must be strictly positive for algebraic links.
Arbitrary algebraic links can be obtained using this construction and the twisted
union (above) for the pairs of positive algebraic trees with ∨ added to the second
subject to the inequality s1 s1 > r1 r1 for the first vertices of these two trees. Then
{L, L∨ } is called a positive pair .
See [EN] and (4.13) concerning the algebraic links; the linking numbers between
the branches of L and L in their twisted union are (full) products of rij and rik
over i in the corresponding paths.
We note that the theory in [EN] is without colors, as well as that in [ORS].
Attaching colors to the branches requires more involved algebraic-geometric con-
siderations; see [Ma] for the case t = q.
4.4. DAHA-Jones polynomials. They will be defined for an arbitrary (re-

The notations
duced, irreducible) root system R and its twisted affinization R.
and formulas are from the previous sections. The combinatorial data will be the
Υ,(bj )
[r,s]-labeled graphs L(rj ,sj ) from (4.10). Recall that
1 ≤ j ≤ κ, !r j = {rij }, !s j = {sji }, 1 ≤ i ≤ j ,
and Υ is the incidence graph/matrix , and the arrowheads (at the ends of all paths )
are colored by bj ∈ P+ . The incidence graph is not supposed to be connected here
and the paths can contain no vertices; see (4.10). The construction below will be
for two arbitrary such graphs L, L (the second can be empty).
The choice of the integral form of the P –polynomials plays an important role in
the following. Recall from Theorem 2.1 that Pb = N b P ◦ for b ∈ P+ is q, t–integral.
b
This is not always the smallest q, t–integral normalization of Pb . The following
normalization is.
Let us consider the denominators of all coefficients of Pb◦ as polynomials in
terms of q with the coefficients in C(tν ) (actually in Z[tν ]). Then let N̈b be their
least common multiple, LCM . We will assume that its constant term is 1; then it is
def
a polynomial in terms of q with the coefficients from Z[tν ]. The polynomial P̈b ==
N̈b Pb◦ is the minimal q, t–integral form of Pb . Accordingly, we set for b1 , . . . , bm ∈
P+ :

(4.16) (b1 , . . . , bm )∼
ev = LCM Nb1 , . . . , Nbm ,

(b1 , . . . , bm )··ev = LCM N̈b1 , . . . , N̈bm .
Recall that the calculation of (b1 , . . . , bm )∼ ev is entirely combinatorial; it is an explicit

product of binomials in the form of (2.3).
In type A, one can also take here the J–polynomials Jλj for λj = λ(bj ); see
(2.21). Then we set:

(4.17) (b1 , . . . , bm )Jev = (λ1 , . . . , λm )Jev = LCM Jλ1 (tρ ), . . . , Jλm (tρ ) ,
where means here that we normalize LCM by the condition that it is a q, t–

polynomial with the constant term 1. The latter practically means that we ignore
the factors in formula (2.23) for Jλ (tρ ) = Jλ (t−ρ ) before the product there (upon
a = −tn+1 ).
One has the following combinatorially transparent formula:
(4.18) (λ1 , . . . , λm )Jev = (λ1∨ · · · ∨λm )Jev , where
λ1∨ · · · ∨λm is the union of diagrams {λj }.
The latter union is by definition the smallest Young diagram containing all diagrams
λ1 , . . . , λm .
Going back to arbitrary root systems, we can take Pb (b ∈ P+ ) themselves when
(1)
klng = 1 = ksht , equivalently tν = qν . Then the corresponding Pb = Pb (X; tν = qν )
do not depend on q, t for any (reduced, irreducible) root system R. We set in this
case:

(1) (1)
(4.19) (b1 , . . . , bm )(1)
ev = LCM Pb1 (q ρk), . . . , Pbm(q ρk) .
Note that the J–polynomials in the An –case are not minimal integral even when
(1)
t = q. However, using them vs. Pλ for t = q makes the corresponding HOMFLY-
PT polynomials a, q–integral and has other advantages. This is important only for
links; the choice of the integral form does not influence the DAHA-construction for
knots.
We represent torus knots T (r, s) by the matrices γr,s ∈ GL 2 (Z) with the first
column (r, s)tr (tr is the transposition) for r, s ∈ Z provided gcd(r, s) = 1. Let
r,s ∈ GL∧2 (Z) be any pullback of γr,s .
γ
Obviously (r, s) can be lifted to γ of determinant 1 and, accordingly, to γ ∈
P SL∧2 (Z) generated by {τ± }. I.e. the usage of η can be avoided. However, η results
in important symmetries of the JD–polynomials. Similarly, the enhanced GL∧2 (Z)ζ
is actually not needed in what will follow, since all the quantities below (inside
{. . .}ev ) will be ζu,v –eigenvectors. Controlling the corresponding Π –characters is
of importance, but not really significant in the construction below.
Definition 4.1. Ingredients. Let R be a reduced irreducible root system and q
def
not a root of unity. Recall that H → H⇓ == H(1), where the action of H ∈ H H in
V is used. The construction is in terms of two graphs/links from ( 4.10)

Υ,(bj ) Υ,(bj )
(4.20) L = L(rj ,sj ) , L = L( rj , sj ) where bj , bj ∈ P+ ,
1 ≤ j ≤ κ, κ for L, L, !rj = (rij | 1 ≤ i ≤ j ), !rj = (rij | 1 ≤ i ≤ j ).

The j–ranges for !sj ,!sj are those for !rj ,!rj . Let us lift the columns (rij , sji )tr ,
(rij , sji )tr to γij , γij and then to γ
ij , γ
ij ∈ GL∧2 (Z) as above.
The construction also requires the choice of the integral forms Pb for the poly-
nomials Pb . Considering the pairs {i, j} as vertices of the incidence graph for L,
we begin with letting
jj = def j def
P +1 = Pb , γj +1 == id for 1 ≤ j ≤ κ;
j
recall that j = 0 when the j th path contains only one arrowhead and υj,k = 0 if the
corresponding paths do not intersect.
Pre-polynomials. For a given path with the index j, we define the polynomials
j by induction with respect to i, starting with i = j down to i = 0:
P i
υ(k,j)=i

(4.21) j =
P i+1
γ k ki+1 )⇓
(P
i
1≤k≤κ
i.e. the last product is over all paths k that enter (intersect) the path for j exactly
jj = υ(k,j)= Pbk
j
at the index i ≥ 0, including k = j when i = j . Note that P 1≤k≤κ
for a base path j, where this product is over all arrowheads from (originated at) the
vertex {i = j , j}.
j actually depends only on the corresponding
Since i = 0 is allowed in ( 4.21), P 0
j is the product of
subtree for any path j there. If L is the union of subtrees, then P

k k 0
the corresponding polynomials γ 1 )⇓ over all these subtrees (for any choices
1 (P
of paths k there). So we may set P tot j tot

0 = P0 . The polynomial P0 for L is defined
in the same way.
Finale. Using the notations b = (bj ), b = ( bj ), the DAHA-Jones polynomial
for the integral forms Pbj , Pbj and a certain fixed index 1 ≤ jo ≤ κ or jo = ∅ is
(which determines the normalization):

(4.22) R,
JD
jo , Υ, Υ j j jo , Υ, Υ
(rj , sj ) , (rj , sj ) ((b ), ( b ) ; q, t) = JD (rj , sj ) ,( rj , sj ) (b, b ; q, t)
/ 0
= JD jL,o L = = ϕ( P
def tot tot
0 ) P0 /Pbjo (q
−ρk
) ,
ev
where there is no division by Pbjo (q −ρk ) for jo = ∅. Due to the definition of ϕ from
tot ) is simply P
( 1.12), ϕ( P tot (Y −1 ) = P
tot (X → Y −1 ). Taking Y +1 here, we
0 0 0
obtain:
/ 0
(4.23) jL,o L∨ =
JD = P
def tot tot
0 (Y ) P0 /Pbjo (q
−ρk
) .

ev tot
Using automorphisms. We note that the pre-polynomial P0 is invariant with
respect to the action of ζu,1 in the polynomial representation; use (1.18) and (3.8)
to calculate the corresponding character.
The role of ϕ . Applying ϕ inside the coinvariant gives that
(4.24) •L, L = JD
JD •L, L for • = jo , min, ∅,
where the normalizations • must be the same in both sides, i.e. the division must
be by the same Pbjo (q −ρk ) for jo . Topologically, this relation means that adding
the meridian L to (the link associated with) L is isotopic to adding the meridian
L to L.
The operator ϕ( P tot ) does not commute with P tot , which is an operator in
0 0
terms of X. In type A, the commutator relations between such and similar operators
are part of the theory of elliptic Hall algebra (isomorphic to spherical DAHA); see
[SV]. This connection is important for the Hopf links and the DAHA-vertex . See
also some remarks concerning the toric q, t–skein in Section 9.5.
The toric skein algebra was identify with the elliptic Hall algebra for t = q in
[MS] (and with the spherical DAHA). This was used to check the coincidence of our
superpolynomials at t = q with the HOMFLY-PT polynomials for torus iterated
knots, conjectured in [ChD] (justified for any such knots for A1 and torus knots
in [Ch2]). Combining [MS] with our approach here proves Part (i) of Conjecture
5.3.
This implication is of course important, but direct using Rosso-Jones formula
and the DAHA-shift operators is also quite relevant, as it was demonstrated in
[Ch2, ChD, ChE]. We already have a sketch of the proof of Part (i) of Conjecture
5.3 below based on this method, which can be generally extended to any root
systems, i.e. applied to connect arbitrary WRT-invariants with the corresponding
DAHA-Jones polynomials at t = q.
Continuing with the automorphisms, we note that ϕ maps τ± → τ∓ in the
tot , without changing the order of the corresponding tau-matrices.
construction of P 0
This can be used to establish the symmetry T (r, s) ↔ T (s, r), but generally not
inside the cables; the application of ϕ is not compatible with the projection ⇓ onto
the polynomial presentation V. The following ones are.
Automorphisms η, τ− , ι , σ 2 . They are compatible with ⇓ , which fact is the key
in proving the symmetries of the JD–polynomials. For instance, adding the vertex
[r0 = 1, s0 = m] to a tree L results in the same JD–polynomial as the change of

[r1 , s1 ] in L by [r1 , mr1 + s1 ] (for every path, if there are several). This gives (the
key) relations from (4.5); see Part (ii) of Theorem 4.3.
Concerning ι , the definition from (4.23) is actually a particular case of (4.22),
where all colors bj are replaced by ι( bj ). We use that ι preserves the coinvariant
and naturally acts in H H; see (1.20,1.22). Also, ι · γ
=γ · ι for γ ∈ GL2 (Z). We use
these properties of ι and similar ones for η in the theorems below.
Adding ∨. This changes P tot (Y −1 ) = σ( P
tot ) in the definition above by
0 0
tot −1 tot
P0 (Y ) = σ ( P0 ). Equivalently, we can send b → bι . Equivalently, this
operation is the change of the sign of the first label (or labels if it is not a tree) in

L: [ r1j , sj1 ] → [− r1j , − sj1 ].
Since ι commutes with ϕ, we have the relation:
(4.25) •L, L∨ = JD
JD •L∨ , L = JD
•L, L∨ ,
where the normalization • must be the same in all three formulas (say, • = ∅ or
• = min). Indeed, the total change of the orientation of the link represented by

the pair {L, L} must not influence the JD–polynomial. Recall that topologically,

adding ∨ to L is switching the orientation of the corresponding link.
Iterated knots. In the case of torus iterated knots (when there is only one path)
and in the absence of L, we arrive at formula (2.12) from [ChD]:
/
0
(4.26) JDr,s (b; q, t) = γ1 · · · γ
−1 γ (Pb )/Pb (q −ρk ) ⇓ ⇓ · · · .
ev
It includes only one b ∈ P+ and therefore does not depend on the choice of the
integral form Pb of Pb .
The simplest link then is obtained, by adding any number of arrowheads colored
by b1 , · · · , bκ , then Pb /Pb (q −ρk ) must be replaced here by the product
(4.27) 1 = P+1
P 1
· · · P+1
κ
/Pbjo (q −ρk ) = Pb1 · · · Pbκ /Pbjo (q −ρk ).
This particular case is already quite interesting; expressing the products Pb1 · · · Pbκ
as linear combinations of Macdonald polynomials (generalizing the Pieri rules) is a
challenge.
4.5. The polynomiality.

Theorem 4.2. For any choice of the normalization index 1 ≤ jo ≤ κ, the
DAHA-Jones polynomial JD jL,o L defined above is a polynomial in terms of q, tsht , tlng
up to a factor q • t• = q • t•sht t•lng , where the powers • can be rational in the latter. It
does not depend on the particular choice of the lifts γij ∈ GL2 (Z) and γ ij ∈ GL∧2 (Z)
for 1 ≤ i ≤ . Also one can replace in formula ( 4.22) the ratio Pbjo /Pbjo (q −ρk ) by
j
Ebjo /Ebjo (q −ρk ) without changing the output.
Proof. The justification of this and the next theorem almost exactly follow
those in Theorem 1.2 from [Ch3] (for torus knots) and in Theorem 2.1 [ChD] (for
iterated torus knots).
The quantity JD jL,o L can be non-integral with respect to q, tν only if P ◦ =def
=
b
−ρk jo j r s
Pb /Pb (q ) for b = b has a pole in terms of q when = 0, where = (1−q tsht tlng )
for certain j > 0, r, s ≥ 0, r + s > 0. We can assume that the binomial is maximal
such, i.e. that Pb◦ has no singularity at (1 − q jv trv
sht tlng )/ = 0 for any Z & v > 1.
sv
Let us localize and complete the ring of coefficients of H H and the polynomial
±1/(2m) ±1/2
representation V, which is Zq,t = Z[q , tν ], by such , i.e. with respect to

()
the ideal 1 − q v trsht tslng ; the notations will be Zq,t , H
H() , V () . Note that we added
q ±1/(2m) to Zq,t .
We will use the evaluation pairing, defined as follows:
{E, F }ev = E(Y −1 )(F (X))(q −ρk ), E, F ∈ V () .
-
See Theorem 11.8 from [Ch5] and Theorem 1.2 from [Ch3]. We set Rad,p = F ∈
() .
V () | {F, V}ev ∈ p Zq,t for p ∈ Z+ .
Switching from V to the whole H H, we define
def - () .
(4.28) RAD,p == H ∈ H H() | {HHH H H}ev ∈ p Zq,t for p ∈ Z+ .
- . -
Equivalently, RAD,p = H ∈ H H() | H(V () ) ⊂ Rad,p , since Rad,p = F ∈ V () |
() .
{HH(F )}ev ∈ Zq,t ; see Lemma 11.3 from [Ch5].
Here q is not a root of unity. Therefore any Y –invariant submodule of V is
invariant with respect to the natural action of τ̇− of τ− in V. We conclude that ψ
and τ− preserve RAD,p for any p ∈ Z+ (and for generic q). The same holds for η;
see (1.14). Thus the whole GL∧2 (Z) preserves each RAD,r .
def
Let P̄b == l Pb◦ ∈ V () for b = bjo and minimal such l ∈ N. Then P̄b (q −ρk ) ∈
Zq,t due to the normalization of Pb◦ . Since P̄b is an eigenfunction of {Lf } from
l ()
(1.28), one has that P̄b ∈ Rad,l ; see Lemmas 11.4-5 from [Ch5].
This implies that γ (P̄bjo ) ∈ RAD,l for any γ ∈ P SL2 (Z), as well as for any
P̄ = RP̄bjo , where R is any q, t–integral Laurent polynomial.
The projection P̄ = γ(P̄ ) ⇓= γ (P̄ ) ∈
(P̄ )(1) then belongs to Rad,l . Hence γ

RAD,l for any γ ∈ P SL2 (Z) and we can continue this process. One automatically
def
obtains that {P̄ }ev is divisible by l for P̄ == γ
(P̄ ) ⇓ and for any further such
polynomials obtained by this procedure continued for γ , γ , . . . from P SL2 (Z) or
from GL2 (Z).
Here we can multiply each P̄ , P̄ by arbitrary q, t–integral R , R and so on.
Furthermore, we can apply to the final P̄ –polynomial in this chain any operator in
the form Q(Y −1 ) for a q, t–integral Laurent polynomial Q(X) and the output will
have the same divisibility by l as above. Thus we conclude that JD jL,o L has actually
no singularity at = 0, which contradiction is sufficient to claim its q, t–integrality.
j
The independence of JD L,o L of the lifts of the columns (rj , sj )tr to γ
ij and
i i
the possibility to replace Pbjo by Ebjo closely follow the corresponding claims in
Theorem 1.2 from [Ch3].
4.6. Major symmetries. The following theorem is a link counterpart of The-

orem 2.2 from [ChD]. We will comment on its proof (but omit the details); it re-
mains essentially the same as for iterated knots. All following claims hold when L
is replaced by L∨ . Furthermore, formula (3.39) from Theorem 3.9 can be used to
reduce the pairs of graphs to a single tree; see, e.g. an example after (9.4).
Theorem 4.3. (i) Minimal normalization. The q, t–integrality and other claims
from Theorem 4.2 hold for the following modifications of DAHA-Jones polynomials
(which does not require picking jo ):

/ ϕ( P tot 0
tot ) P
(4.29) min
JD min def
(rj , sj ) ,( rj , sj ) (b, b ; q, t) = JD L, L ==
0 0
,
(b, b)∧
ev ev
where the polynomials Pb (b ∈ P+ ) are defined for one of the following integral forms:
Pb = Pb , P̈b , Jλ(b) (for An ), or Pb (when ksht = 1 = klng );
(1)
(4.30)
(b1 , . . . , bm )∧
ev are defined correspondingly. See ( 4.16),( 4.18), ( 4.19).
(ii) Topological symmetries. Following Section 4.2, the polynomial JD jL,o L from
min
Theorem 4.2 and JD L, L from (i), considered up to a monomial factor q • t•sht t•lng ,
depends only on the topological link corresponding to the pair of graphs {L, L}. For
instance, ( 4.4) holds and the reduction of the vertices with r = 1 from ( 4.5) can be
applied in L or in L. Also, the transposition [r1j , sji ] → [sj1 , rij ] from ( 4.6) does not
j min
influence JD L if L = ∅.
L◦ or JD
j min
o or JD
Moreover, JD above become −conjugated if all sji , sji from L and

L change their signs simultaneously; see ( 4.8), ( 4.9). Furthermore, they do not
change when bj → ιπ (bj ) and simultaneously
(4.31) {rij }, {sji } → {. . . , ri−1
j
, −rij , ri+1
j
, . . .}, {. . . , sji−1 , −sji , sji+1 , . . .}
in L and L provided that total numbers of such transformations in every path of
L, L have all the same parity π ∈ Z2 .
(iii) Color exchange. We assume that for given b, b, generic q and certain
(special) tsht , tlng , there exist c = (cj ), c = (cj ) and wj , wj ∈ W satisfying the
relations
∨ (ρw , α∨ ) w ∨
(4.32) q (b+ρk −w(ρk )−w(c) , α) = 1 = qα(b−w(c) , α ) tsht sht tlng(ρlng , α ) ,
for all α ∈ R+ , where w = wj , b = bj , c = cj or w = w, b = bj , c = cj , and we set
def
ν == w(ρν ) − ρν . Then
ρw
(4.33) min
JD min for such q, {tν },
L, L = JD M, M
where M, M correspond to the same (rij , sji ), ( rij , sji ), Υ, Υ but with the colors
bj , bj replaced by cj , cj .
(iv) Specialization q = 1. We now make q = 1, assuming that tν are generic
and using the notation (b1 , . . . , bm )∧
ev from ( 4.16). We switch here to the spherical
polynomials Pb◦ , so the following does not depend on the choice of the integral forms
Pb :

(b1 , . . . , bκ , b1 , . . . , b κ )∧
ev
min
(4.34) ∧ ∧ ∧ κ ∧ (q = 1) JD L, L (q = 1)
(b )ev · · · (b )ev ( b )ev · · · ( b )ev
1 κ 1

κ

κ

j
j
= JD rj , sj b ; q = 1, tν ) JD rj , sj b ; q = 1, tν ),
j=1 j=1

where JD r, s b; q = 1, tν = np=1 JD r, s (ωp ; q = 1, tν )bp ,

for b = np=1 bp ωp ∈ P+ , where the knot JD–polynomials from ( 4.26) are used. See
formula (2.18) in [ChD]. These specializations coincide for the pairs {L, L} and
{L, L∨ }.
Comments. Concerning using “min” in JD min and for the superpolynomials

later, they can be reducible depending on the choice of the integral form. We
expect them to be irreducible for Pb = P̈b , but this is not generally true for other
integral forms, including Pb = Pb and Pb = Jb in the A–case (which are not the
smallest possible).
The justification of the symmetries from Part (ii) is essentially parallel to The-
orem 1.2 from [Ch3]. Let us comment on (4.5). Essentially, one needs to check
here that the torus knot T (mr + s, r) results in the same DAHA-Jones polynomial
as the “2-cable” corresponding to the Newton pairs [m, 1] → [r, s]. Topologically,
T (mr + s, r) is isotopic to Cab(mr + s, r)T (m, 1), since T (m, 1) is unknot. As it
was noted above, the corresponding relation for the JD–polynomials readily fol-
lows from the commutativity τ− m
with ⇓, which simply means that τ− acts in V.
Other symmetries are based on applying η, ι, σ 2 inside the coinvariant; they are
compatible with the projection ⇓ as well. Also, the case of the pair of graphs is
governed by Theorem 3.9.
The last part of the theorem states that the DAHA-Jones polynomials for
iterated torus links become the products of those for the paths under q = 1 (i.e.
over the knots that constitute the link). This is compatible with our conjecture
in [ChD] concerning the Betti numbers of the Jacobian factors for (the germs
of) unibranch planar singularities. The DAHA-superpolynomials in type A under
a = 0, q = 1 occur here. Thus the passage from knots to links does not add much
to our conjecture on the Betti numbers due to (iii); we omit the discussion of the
Jacobian factors and their Betti numbers in this work.
QG invariants. Extending the connection conjectures from [Ch2, ChD], we
expect that for ksht = 1 = klng and the integral form Pb = Pb ,
(1)
j R,Υ,j
(4.35) L,o L (tν → qν ) = JD
JD (rj , sjo) , (rj , sj ) ((bj ), (bj ) ; q, tν → qν )
coincide up to q • with the reduced Quantum Group (WRT) invariants of the cor-
responding iterated torus colored links, where the reduced normalization is for the
weight bjo . To obtain the non-reduced QG-invariants, one takes jo = ∅, i.e. omits
j
L,o L in this case (for kν = 1).
the division by Pbjo (q −ρ ) in JD
(1)
We do not see at the moment any topological meaning of the JD min –polynomials
j j
from part (i) in Theorem 4.3 for such kν and general (b ), ( b ). The division there
is by (b, b)ev , which provides the q–integrality, but results in the polynomials that
(1)
seem too small for the topological interpretation. This is unless all colors are the
same.
We note that the Quantum Group (WRT) invariants are associated with the
root system R; see [Ch2]. The shift operator was used there to deduce this coinci-
dence from [Ste1] in the case of An and torus knots; quite a few confirmations were
provided for other root systems (including special ones). The method was origi-
nated in CFT Conformal Field Theory, and Verlinde algebras. DAHA provides
perfect tools to understand and generalize the latter algebras.
As we discussed above, Part (i) of Conjecture 5.3 can be checked using the
Skein; another approach, which is applicable to verifying the connections with the
QG–invariants for any root systems, is a combination of the DAHA shift operators
with the Rosso-Jones formula .
5. DAHA-superpolynomials
5.1. Main theorem. Following [Ch2, GN, Ch3, ChD], Theorem 4.1 and
Theorem 4.3 can be extended to the DAHA- superpolynomials , the result of the
A
stabilization of JD
n ,jo
An ,min with respect to n → ∞.
L, L (including jo = ∅) or JD L, L
This stabilization was announced in [Ch2] for torus knots; its proof was pub-
lished in [GN]. Both approaches use [SV] and can be extended to arbitrary iterated
knots and links; the Duality Conjecture was proposed in [Ch2] and proven in [GN]
for torus knots; also see [Ch3] for an alternative approach based on the generalized
level-rank duality. The justifications of these claims for iterated torus knots [ChD]
and iterated torus links is essentially parallel to the case of torus knots, though
there are some deviations.
The main change here from knots to links is that the polynomiality of the su-
perpolynomials for links is based on our using the Jλ –polynomials as the integral
form of {Pλ }. Actually {Jλ } were already employed in [GN] for the stabilization
and duality, but the construction of (reduced) JD–polynomials and superpolyno-
mials for knots requires only spherical {Pλ◦ } and does not depend on picking their
integral forms. For links vs. knots, the role of polynomials {Jλ } becomes the key;
without using the J–polynomials the superpolynomials can become rational (not
polynomial) in terms of t.
The sequences !rj , !sj of length j for the graph L and !rj ,!sj of length j for
the graph L will be from the previous sections. We consider only An here, setting
t = tlng = tsht = q k . We will always use below the DAHA-Jones polynomials
j min
o , JD
JD in type An (for sln+1 ) defined in terms of Jλ , unless stated otherwise.
(1)
The integral form Pλ for t = q (i.e. when k = 1) will be needed only when
discussing the connection with the HOMFLY-PT polynomials. Recall that λ = λ(b)
is the Young diagram representing b ∈ P+ .

We will consider P+ & b = ni=1 bi ωi for An as a (dominant) weight for any
Am (for slm+1 ) with m ≥ n − 1, where we set ωn = 0 upon its restriction to An−1 .
The integral form of Pb in (4.22) and (4.29) will be Jλ from (2.21) for λ = λ(b) in
the next theorem. See [GS, Ch2, GN, Ch3, ChD] concerning the version of the
following theorem for torus knots and torus iterated links.
Theorem 5.1. (i) Stabilization. Given the links L, L colored by b = (bj ), b =
(bj ) and the normalization index 1 ≤ jo ≤ κ (including jo = ∅), there exist polyno-
mials from Z[q, t±1 , a]
(5.36) jo = H
H Υ,j o
L, L (rj , sj ),( rj , sj ) (b, b; q, t, a),
L,
H Υ,min j j (b, b; q, t, a)
L = H j j
min
(r , s ),( r , s )
such that for any m ≥ n−1 and proper powers of q, t (possibly rational) :
A ,j
(5.37) H L,mL o (q, t),
jo (q, t, a = −tm+1 ) = ± q • t• JD
L, L
L,
H L (q, t, a = −t
min m+1 A
) = ±q • t• JD
m ,min
L, L (q, t).
by ±q • t•
They are normalized as follows (the hat-normalization). We multiply H
= 0) is not divisible by q and t and that the coefficient of the
to ensure that H(a
= 0) is positive. If JD
minimal power of t in H(a Am –polynomials are considered
under the same hat-normalization, then relations ( 5.37) will automatically hold for
sufficiently large m without any correction factors ±q • t• (and one sufficiently large
m is actually sufficient to fix H uniquely). The a–stabilization and other claims

here hold for HL, L∨ ; see ( 4.23).
(ii) Symmetries. The polynomials H depend only on the isotopy class of the
corresponding torus iterated links. For instance, the symmetries from Part (ii)
of Theorem 4.3 hold for H = H jo , H
min under the normalization from (i) with
the following reservation about ( 4.31); its a–extension holds upon a = a−1 (up to
a• q • t• ). The same claims are for L∨ instead of L, and the a–extension of Theorem
3.9 in the case of the pair of graphs can be reduced to a single graph.
For the color exchange , we impose ( 4.32), and consider wj , wj as elements of
Sm+1 (the Weyl group for Am ) for every m ≥ n. Then for L that is either L or
∨
L (for + = ∅, ∨):
(5.38) H L, L (c, c ; q, t, a) = for such q, t.
L, L (b, b ; q, t, a) = H
Similarly, the specialization relations from ( 4.34) at q = 1 in the case of An hold
for H and such specializations coincide for {L, L} and { L∨ }. Recall that the
product formula there holds when spherical polynomials Pb◦ = Pb /Pb (tρ ) are used in
H.
the formulas for JD,
(iii) Super-duality and dega . We switch from b, b to the corresponding sets of
Young diagrams λ, λ. Let λtr , λtr be their transpositions. Up to powers of q and
t denoted here an below by q • t• , one has:
(5.39) HL, L (λ, λ ; q, t, a) = q • t• H
L, L (λ, λ ; t−1 , q −1 , a);
the same super-duality holds with ∨, i.e. for H L, L∨ .

j j
Let us assume that ri , ri = 0 for i > 1. Then dega H min
L, L ( λ, λ ; q, t, a) and
min
dega H
L,κL∨ ( λ, λ ; q, t, a) are no greater than
κ
j j j
max{1, | r1 |}|r2 · · · rj | |λ | +
j
max{1, | r1j |}| r2j · · · rjj | |λj | − Δ,
j=1 j=1
(5.40) 1 κ 1
min , Δ = |λjo | for H
Δ = |λ ∨. . .∨λ ∨ λ ∨. . .∨ λ κ | for H jo ,
∅ (when Δ = 0),
where |λ| is the number of boxes in λ. This estimate holds for H
ρ
i.e. without divisions by the evaluations at t .
Frequently for {L, L∨ } with ∨ (conjecturally for all positive pairs) the following
improvement of (5.40) holds (with the same Δ):
κ - j . j
dega HL, j j j
j=1 max 1, min{|r1 |, |s1 |} |r2 | · · · |rj | |λ |
min
(5.41) L∨ =
κ - .
+ j=1 max 1, min{| r1j |, | sj1 |} | r2j | · · · | rjj | | λj | + Δ.
5.2. Sketch of the proof. Generally, the stabilization and duality are due to
the switch from Pλ to the modified Macdonald polynomials [GN]. The projective
action of P SL2 (Z) is compatible with such a switch (see there and [SV]). The
transition to the iterated links from torus knots is relatively straightforward; only
one b will be discussed below.
min and for the integral
It suffices to check the stabilization (up to ±q • t• ) for H

form Jλ(b) . Then the stabilization of H follows from Proposition 3.5. Note that
jo
the stabilization holds for Pb = Pb , for instance for the corresponding H ∅ (i.e.
without the division by the evaluation at tρ ). The polynomials Pb generally are not
q, t–integral, so the corresponding H ∅ is generally rational. We note that even if
t = q, the denominators in terms of q do occur in H ∅ for links.

The limit H(a = 0), used in the hat-normalization condition, is the term-wise
intersection (common part) of JD Am for m * 0, under the normalization as in
= 0) with deg t ≤ M can be extracted from a single
(i). Indeed, the terms in H(a
JD Am
if m ≥ M , since a = −tm+1 is beyond this range of t. This explains why the
correction factors ±q • t• are not necessary for large m for the hat-normalization.
Concerning the super-duality and dega H min , the simplest (and actually the
key) particular case is when there are no vertices at all in L and this graph contains
only κ arrowheads. We set symbolically Υ = {·→ · · } (any number of arrows). Let us
→
omit L; replace λ = (λj ) below by λ ∪ λ if L is present. Using (2.23) and taking
into account the normalization from (i),
p −1

n λ
−ρ
Π†λ == 1 + q v a t−p+1 ,
(ρ,b) def
(5.42) t Jλ (t )=
p=1 v=0

·→
(5.43)
H
· · , min
→
(a, q, t) = Π†λ1 · · · Π†λκ /Π†λ1 ∨...∨λκ .
L
The division here will be by Π†λjo for H jo (no division when jo = ∅).
Then (5.42) gives the required formula for dega H in this case. The super-duality
follows here from the relation
Π†λ (a, q, t) = Π†λtr (a, t−1 , q −1 ).
Let us briefly consider the case of {L, L } for L = ·→
· · λ = μ. Then H ∅ is
→ L, L
kρ+λ −1 −ρk −b
essentially Jμ (q ). Indeed, (2.4) and (1.28) give that f (Y )(Pb ) = f (q )Pb
for b ∈ P+ and symmetric f . Then the a–stabilization is due to Proposition 3.5.
The a–degree will be the same as that for Jμ (q kρ ), which is the same with ∨, i.e.
for L∨ instead of L; the corresponding coinvariant is Jμ (q −kρ−λ ) with ∨ (up to the
proper renormalization). These two cases, with ∨ and without, are quite different;
the a–degree of H min
L, L is generally greater than that with ∨ . See (9.8), (7.36) and
(7.34) vs. (9.9), (7.37) and (7.35).
Generalizing, one needs to decompose the pre-polynomials P tot from
tot , P
0 0
(4.22) with respect to the Macdonald polynomial and determine the leading one.
We essentially follow [GN]. Note that it was conjectured in [ChD] that (5.41) gives
exact dega for algebraic knots ; there are examples of non-algebraic knots and links
where this formula for the a–degrees does not work. Without L and in the uncol-
ored case, our formula for dega is nothing but the multiplicity of the corresponding
singularity (Rego). Also, the coincidence of H(q = 1) with ∨ and without is not
difficult to see; the product formula from Part (iii) holds because q = 1 is the case
of the trivial center charge .
Continuing the remarks after Theorem 4.3, we note that the polynomial H min
can be reduced for certain iterated links, so they are not exactly “minimal” a, q, t±1 –
integral. The importance of Jλ (not sharp as integral forms) is mainly due to
the stabilization and super-duality. We omit details here, but let us provide the
following example. If one makes t = q and takes Pλ = Pλ (=Schur polynomials) as
(1)
the integral form, then the corresponding H min

will be not a polynomial in terms
min
λ are of course
of t = q even in the case of 2 unknots, though the corresponding JD
always q–integral. This is important; the polynomial a–stabilization does require
using “non-minimal” integral forms.
See [Ch3] for a systematic consideration of the color exchange condition (5.38)
from (ii). In terms of the Young diagrams, the procedure is as follows. Recall that
n
we associate with cj = i=1 cji ωi the Young diagram
λ(cj ) = {λj1 = cj1 + . . . + cjn , λj2 = cj1 + . . . + cjn−1 , . . . , λjn = cjn , 0, 0, . . .}.
Then we switch to λ(c̄j ) = {λ̄ji = λji − k(i − 1)} for t = q k , apply wj ∈ W to λ(c̄j )
and finally obtain
(5.44) λ(bj ) = {λ̄jwj (i) + k(i − 1)} = {λjwj (i) + k(i − wj (i))}.
Here 1 ≤ j ≤ κ and wj transforms the rows of λj ; we generally set w{λ1 , λ2 , . . . , λn }

= {λw(1) , λw(2) , . . . , λw(n) }. Given k < 0 (it can be fractional), λ(bj ) must be a
Young diagram ; this condition determines which wj can be used. Then we repeat
the same procedure for L and obtain b from c for proper wj ∈ W , 1 ≤ j ≤ κ.
5.3. Super-vertex. Using the definition from (3.34) for trivial u = 1 (which
will be omitted here), one has for any root system R and two sets b = (bj , 1 ≤
j ≤ κ) ⊂ P+ ⊃ b = ( bj , 1 ≤ j ≤ κ),
def -
−1
◦ . - −1
◦ ◦ .
(5.45) Pbb == τ+−1 P◦b (Y −1) τ− Pb (X) ev = τ− Pb P b ev
R R
JD {1,−1}, {1,−1} (b, b; q, t) JD {1,−1} (b ∪ b)
= =
(b1 )ev · · · (bκ )ev ( b1 )ev · · · ( bκ )ev (b1 )ev · · · ( bκ )ev
for any integral form Pb in JD (and the corresponding { }ev ). I.e. P b can be
b
L, L for L = {[r1 = 1, s1 = −1]} = L with the arrowheads colored
interpreted as JD
correspondingly by b and b. Note that we use here the upper indices in (bj ), (bj )
in contrast to the lower ones in (3.34). Due to the division in (5.45), the choice
of the integral form Pb does not really matter, however the corresponding ratio is
generally not a polynomial in terms of q, tν .
Thus the level-one DAHA-vertex appeared directly connected with our invari-
ants for the Hopf links. The case b = c for (a single) c ∈ P+ is of particular
interest. Using (3.38),
◦
ι(c) Pc , Pc◦
(5.46) Pbc = Cb for b ⊂ P+ & c , ι(c) = cι = −w0 (c),
θμ
where Cbc is the DAHA multi-vertex for u = 1 from (3.24,3.28). It is of key
importance for b = (b1 , b2 ) and for b = b.
Corollary 5.2. Let R = An . For Pλ = Jλ , where λ = λ(b), and for the
evaluation (λ)†ev = t(ρ,b) Jλ (t−ρ ) = Π†λ , we set :
∅
H ∅
H
∅,† = {1,−1}, {1,−1} (λ, λ ; q, t) {1,−1} (λ ∪ λ)
(5.47) H L, L = .
(λ1 )†ev · · · (λκ )†ev ( λ1 )†ev · · · ( λκ )†ev (λ1 )†ev · · · ( λκ )†ev
It is a rational function in terms of a, q, t, where the a–degrees of the numerator

and denominator in ( 5.47) coincide (they are a, q, t±1 –polynomials) and are equal
κ j
to κj=1 |λj | + j=1 | λ |. One has:
(5.48) ∅,† = ± q • t• Pb b for m ≥ n − 1,

H L, L
where the extension of the weights b, b to any Am (m ≥ n − 1) is as in ( 5.45)

via the Young diagrams λj = λ(bj ), λj = λ( bj ); recall that we put ωn = 0 upon its
restriction to An−1 .
In particular, the existence of the a–stabilization of Cb /θμ = Pbc /Pc◦ , Pc◦
ι(c)
up to q • t• (depending on m) results from Proposition 3.4. Here 1/Pc◦ , Pc◦ up to

q • t• becomes an a–polynomial of degree 2|λ(c)| with q, t–rational coefficients (up
to q • t• ). Replacing here L by L∨ , we arrive at the a–stabilization of Cbc /θμ
(without ι), satisfying the associativity from Part (ii) of Theorem 3.7 with ui = 1 .

Let us also mention here the relations from (5.47):
∅
H ∅
(5.49) {1,−1}, {1,−1} (λ, λ) = H{1,−1} (λ ∪ λ);
they are quite obvious topologically ; see below. We will provide quite a few examples
3
of the a–stabilization of Cλλ1 ,λ2 /θμ (and that for ι(λ3 )), but will not discuss the
“a–associativity” from (3.32) in this work.
5.4. HOMFLY-PT polynomials. There are two approaches to the unre-

duced HOMFLY-PT polynomials HOM(λ; q, a), via QG (in type A) or using the
corresponding skein relations and the corresponding Hecke algebras. Both are for
any links and colors. See, e.g. [QS] and references there. We provide here only a
sketchy discussion.
Note that the symmetry from Theorem 4.8 in this work corresponds to our
q → q −1 , a → a, λj → (λj )tr . The passage to reduced HOMFLY-PT polynomials
corresponds to our division by Pλjo (q ρ ), so it requires picking one path-component
jo from in a given graph-link. Let us impose the hat-normalization here; the no-
tation will be HOM4 (q, a) or HOM 4 L, L (λ; q, a) for the unreduced ones. We will set
a = −a below.
For links, the q–polynomiality of the unreduced HOMFLY-PT polynomials
does not hold. This is direct from the corresponding normalization of colored
unknots. One has:
κ
(5.50) 4 (q, a) = 1 − a
HOM for κ uncolored unknots.
1−q
4 (q, a) = q • P 1 (q ρ ) · · · Pλκ (q ρ ) for κ unknots upon the a–stabilization
With colors, HOM
(1) (1)
λ
n+1
with a = q (for An ). Note the absence of minus in the latter, in contrast to
this substitution for a in H. The power q • here is adjusted to ensure the hat-
normalization of the left-hand side.
From formulas (2.21),(2.23) for t = q and with such a:
n λp −1

• (1) ρ p=1 j=0 1 − a q j−p+1
(5.51) q Pλ (q ) = , λ = {λp },
2∈λ (1 − q
arm(2)+leg(2)+1 )
where n here is the number of (nonempty) rows in λ. Thus the product on the right-
hand sides j 4
3κin (5.51) over the Young diagrams λ in λ is the value of HOML (λ; q, a)
for L = j=1 .
We note that since we deal only with iterated torus links, the Rosso-Jones
cabling formula is generally sufficient for calculating the corresponding HOMFLY-
PT polynomials; see e.g., [RJ, Mo, ChE]. This can be actually done for arbitrary
root systems (and WRT-invariants of any colored iterated torus links). See [AM,
AMM] and [Ma] about using here the HOMFLY-PT skein relations. Also, paper
[MS] established the identification of the skein algebra of the torus with the Elliptic
Hall algebra for t = q and therefore with the corresponding spherical DAHA; see
[SV].
The algebraic-geometric interpretation of these relations from [Ma] presumably
can be generalized to establish the connection of our superpolynomials to papers
[ObS, ORS]. These two papers are for arbitrary plane curve singularities, but are
restricted to the uncolored case; the main conjecture from [ObS] was extended in
[Ma] to the colored case and proved; this was unrefined , i.e. without t.
Our previous paper [ChD] was in the unibranch case; now we can reach any
multi-branch plane curve singularities (with arbitrary colors). We will not discuss
systematically the relations to [ORS],[Ma] and [Pi] in the present work, though
provide quite a few examples.
The stable Khovanov-Rozansky homology is the slN homology from [KhR1,
KhR2] in the range of N where the isomorphism in Theorem 1 from [Ras] holds
(see also [Kh]). Thus they can be obtained from the triply-graded HOMFLY-
PT homology, assuming that the corresponding differentials are known (they are
generally involved).
Considering links and adding colors makes the KhR–theory significantly more
difficult. Even without such an extension, the HOMFLY-PT homology is known
only for very limited number of examples and no formulas are known for torus iter-
ated knots/links beyond torus knots. The situation is much better for the Khovanov
homology (for sl2 ), though colors-links are a problem even in this case. The categori-
fication theory can generally address arbitrary colors (dominant weights), but the
HOMFLY-PT homology remains quite a challenge; see e.g., [Kh, WW, Rou, Web]
and references there.
Thus we have to restrict Part (iii) of the conjecture below to the uncolored
unreduced case (unless for N = 2). The corresponding Poincarè series, stable
stab
unreduced Khovanov-Rozansky series , will be denoted by KhRL, L (qst , tst , ast ) in
the (topologically) standard parameters; see [ORS] and below. The passage to the
N
Khovanov-Rozansky polynomials for slN , denoted below by KhRL, L , is as follows:
5
(5.52) ast → qst
N
, equivalently, a → tN q/t in our parameters.
This actually depends on the grading used in the theory and there are variations
here in different works.
These relations are applied as such only for sufficiently large (stable) N , other-
wise the theory of differentials is necessary. The differentials correspond to a differ-
ent substitution: a = −tN in DAHA parameters (see the conjecture below). Let us
mention the relation to the Heegaard-Floer homology : N = 0. Also, the Alexander
polynomial of the corresponding singularity is H min (q, q, a = −1)/(1 − q)κ−δκ,1 in
L
the case of one uncolored tree L with κ paths (the number of connected components
in the corresponding cable). This is zeta-monodromy from [DGPS] upon t → q

(unless for the unknot).
We will always impose the hat-normalization from Part (i) of Theorem 5.1
below. The notation will be KhR 4 N . Namely, we divide the first
4 stab and KhR
L, L L, L
polynomial by the smallest power of ast and then divide (both of) them by the
• •
greatest possible qst 4 stab (ast = 0) and KhR
tst making KhR 4 N from Z[qst , tst ] with
L, L L, L
the constant term 1.
Other approaches. The Khovanov-Rozansky theory (unreduced or reduced) is
expected to be connected to the physics superpolynomials based on the theory of
BPS states [DGR, AS, DMS, FGS, GGS]. This theory is not mathematically rig-
orous and the formulas (for small knots/links) are mostly obtained via expected
symmetries. We can prove them in DAHA theory, so the coincidences of our
formulas with physical formulas is not surprising (unless they impose too many
symmetries). For instance, the approach in [DMS] is very algebraic.
See e.g. [Gor, GORS, GN] for an important approach to superpolynomials of
uncolored torus knots based on rational DAHA . It is expected that colors can be
potentially added here (for torus knots); the case of symmetric powers of the fun-
damental representation is in progress, see [GGS]. We will not touch this direction
in this work. Using rational DAHA is connected with the Hilbert schemes of plain
curve singularities and C 2 and with the ORS-polynomials from Part (iv) below, so
this approach is related to our considerations.
Knot operators. Using the Macdonald polynomials instead of Schur functions
in the construction of the so-called knot opertators was suggested in [AS], which
triggered paper [Ch2]. These operators for t = q naturally appear in the approach
to the invariants of torus knots via the Verlinde algebras. This method results
in certain algorithms, but it is justified mathematically by now only for the root
systems A, D. It requires using the roots of unity q (t must be an integral power of
q) and the formula for the refined Verlinde S–operator. In the refined theory, with
the Macdonald polynomials instead of Schur functions, the formula for S becomes
very involved (even for A1 ). One must know all Macdonald polynomials at roots
of unity, which is almost impossible technically (though some formulas for their
coefficients are known).
Furthermore, it must be justified that the final formulas are uniform in terms
of q, i.e. can be lifted to a generic q, t, which is quite a challenge. Formally, such a
lift can be done only if the upper bounds for the degrees of q, t (and a) are known.
So the authors mainly compared their calculations with known/expected formulas.
Only the simplest superpolynomials were discussed in [AS] (mostly uncolored);
they are reproduced in [Ch2] via DAHA.
In more details, the refined S operator is essentially the matrix with the en-
tries Pb (q c+ρk ) for all admissible (which depends on the root of unity) b, c ∈ P+ .
Knowing all Pb (q c+ρk ) is generally a transcendental problem unless for A1 , and
one need the whole projective action of P SL2 (Z) due to [Ki] and the works by
I.Ch. (see [Ch1]). Furthermore, all An are necessary for the stabilization. It is
the price of using the polynomial representation in [Ki, AS] at roots of unity vs.
the direct usage of H H suggested in [Ch2]. These problems were resolved there
(actually bypassed) and the DAHA-Jones theory was extended to any reduced root
systems and dominant weights.
Having said this, DAHA at roots of unity is an important component of the

general theory. The generalized Verlinde algebras , also called perfect DAHA repre-
sentations are one of the main applications of DAHA. They are used in the theory
of the JD–polynomials and DAHA superpolynomials, especially toward 3–folds and
applications in Number Theory. The roots of unity were employed in [Ch3] for the
justification of the super-duality (a sketch).
5.5. Connection Conjecture. From now on, let H L, L (q, t, a)st denote
HL, L (q, t, a) in Theorem 5.1 expressed in terms of the standard topological pa-
rameters (see [Ch2] and Section 1 in [ORS]):
2
t = qst , q = (qst tst )2 , a = a2st tst ,
2
5 5
(5.53) qst = t, tst = q/t, a2st = a t/q.
I.e. we use the substitutions from the first line here to obtain the polynomial
HL, L (q, t, a)st from H
L, L (q, t, a).
We will consider the integral forms Pλ = Jλ and Pλ = Pλ ; the latter is for
(1)
t = q when the Macdonald polynomials coincide with the Schur functions. Also,
H jo is the hat-normalization from Theorem 5.1, where jo = ∅ means that there are
no divisions by the evaluations at q ρ .
Conjecture 5.3. (i) For t = q and Pλ = Pλ , we conjecture that
(1)
(5.54) ∅ (q, t → q, a → −a)st = HOM

H 4 L, L (qst , ast ),
L, L
where the latter is the hat-normalization of the unreduced HOMFLY-PT polynomial

for any pair of graphs {L, L} colored by an arbitrary sequences λ, λ of Young
diagrams. Equivalently,
(5.55) jo (q, t → q, a → −a)st = HOM
H
j
4 L,o L (qst , ast ),
L, L
jo
where HOM 4 L, L (qst , ast ) is the corresponding reduced HOMFLY-PT polynomial. One
can replace L by L∨ in ( 5.54) and ( 5.55).
(ii) Now let {L, L∨ } be a pair of trees such that rij , sji , rij , sji > 0, and s1 s1 >

r1 r1 , where 1 ≤ j ≤ κ, 1 ≤ i ≤ j and 1 ≤ j ≤ κ, 1 ≤ i ≤ j . Here and in (iii)
only L∨ with ∨ (can be empty) is considered and the integral form is Pλ = Jλ .
Then formula ( 5.41) for dega is conjectured to be exact. In the uncolored case, we
also expect the positivity of the following series:
(5.56) min
H L, L∨ (q, t, a)/(1 − t)
κ+κ−1
∈ Z+ [[q, t, a]]
upon the natural t–expansion of this ratio. For any diagrams, it is conjectured to
hold for sufficiently large powers of (1 − t)(1 − q) (provided rij , sji , rij , sji , s1 s1 −

r1 r1 > 0 and with the usage of ∨).
(iii) Furthermore, let λj = = λ(ω1 ) = λj for all j (the uncolored case).
Then {J2 }ev = t1/2 (1 + a)/(a2 )1/4 and we conjecture that for Pλ = Jλ and for the
hat-normalization above:
H ∅ ∨ (1+a) H min ∨
(5.57)
L, L
=
L, L 4 stab ∨ (qst , tst , ast ).
= KhR
L, L
(1 − t)κ+ κ st (1 − t)κ+κ st
The topological setting is unreduced here, so KhRstab are polynomials in terms of
a, q with the coefficients that are generally infinite t–series.
L = Palg , where the latter

stab
(iv) Conjecture 2 from [ORS] states that KhRL,
series is defined there for the corresponding germ of plane curve singularity (see
( 4.14)) in terms of the weight filtration in the cohomology of its nested Hilbert
scheme. Thus the series from ( 5.57) can be also expected to coincide under the
hat-normalization with Palg .
Combining our paper with Section 7.1 from [MS] proves Part (i) (we will post
the details somewhere). Also we can follow here Proposition 2.3 from [Ch2] (for
torus knots), where we used [Ste1]. This approach is based on the DAHA shift
operators and Verlinde algebras.
Instead of using [MS] or the knot operators from CFT and the Verlinde alge-
bras, one can directly apply the Rosso-Jones cabling formula [RJ, Mo, ChE] upon
its relatively straightforward adjustment to iterated links.
This formula used together with the theory of DAHA shift opertor gives a
relatively straightforward way for the justification of Part (i). We have a sketch of
a proof, which follows [Ch2] and especially the case of A1 (iterated torus knots)
considered in detail in Proposition 4.2 from [ChD]. This leads to an exact match,
not only to the coincidence up to proportionality (under the hat-normalization).
The advantage of this approach is that it can be potentially extended to the
WRT invariants for any root systems and for any iterated torus links (following
[ChD]). For torus knots, the connection with the HOMFLY-PT polynomials from
[Ch2] at t = q was extended now to the coincidence of the DAHA hyperpolynomials
of type D at t = q to the Kauffman polynomials; see there.
In the uncolored case, the ratio (1+a) H min /(1 − t)κ+κ in Part (iii) for P = J
L, L
becomes exactly H ∅ for P = P (1) from Part (i) upon t = q. Accordingly, the
L, L
reduced variant of the conjecture from Part (i) becomes in the uncolored case as
follows:
(5.58) HL,
min 2 2 2 2 κ+κ−1
L (qst , qst , −ast )/(1 − qst ) = HOM
reduced
4 L, L (qst , ast ),
where the integral J–form is used in H min .

Concerning the relation to the plane curve singularities, Proposition 3 from
[ORS] matches the DAHA super-duality and our estimate for dega from Part
(iii) of our Theorem 5.1 restricted to the uncolored case. This is a confirmation of
Part (iv) of the Connection Conjecture. However the relation of our construction
to Palg the stable Khovanov-Rozansky polynomials can not be directly confirmed
at the moment beyond some cases of torus knots; see (5.59) below.
One of the reasons for such an uncertainty is that the positivity of the poly-
nomials H min ∨ (q, t, a) does not hold for links (including uncolored ones) and for
L, L
knots if the corresponding Young diagrams are non-rectangle. We address it in the
positivity conjecture from Part (ii), which however corresponds to the non-reduced
theory. Recall that (5.56) is for any positive pairs {L, L∨ } and can be extended to
any weights. The experiments show that the positivity of the series there almost
always fails when the inequalities from Part (ii) are not satisfied (even in the un-
colored case). This is in contrast to the Khovanov-Rozansky theory, where knots
are arbitrary.
For quite a few colored algebraic links, the division by (1 − t)• is sufficient for
the positivity in Part (ii). However we have examples when the positivity holds
with (1 − q)• and fails for the division by any powers of (1 − t). There are also
examples when only proper powers of (1 − q)(1 − t) ensure the positivity. We did
not reach any conjectures concerning the occurrence of t or q here and the minimal
powers of these corresponding binomials.
One can also try to replace the negative terms −q l tm by q l tm±1 following (5.59)
below for small links (but this is of experimental nature). An obvious problem is
that there are practically no known formulas of the stable KhR–polynomials and
Palg beyond those for some (relatively simple) torus knots.
Therefore Parts (iii, iv) of Conjecture 5.3 are not exactly verifiable conjectures
at the moment, with a reservation concerning the positivity claim (5.56) from (ii),
which is well confirmed in examples. We mention that the relation with the stable
KhR–polynomials was conjectured in [ChD] for pseudo-algebraic iterated torus
knots (with positive DAHA-superpolynomials), not only for algebraic knots. We
also suggested there some procedures of experimental nature, which hopefully may
work for any cables (see below).
Let us address a bit using the differentials. It is generally difficult to calcu-
late KhRn+1 unless for n = 1 (the celebrated Khovanov polynomials). Not many
formulas are known (and all known ones are uncolored so far). The polynomials
KhRstab
L, L are of more algebraic nature, but are actually no simpler. If the polynomial
KhRstab is known, recovering all individual KhRn+1 from it is generally provided
by the theory of differentials ∂n+1 from [Kh, Ras], but this is quite a challenge.
The corresponding homology Ker(∂n+1 )/Im(∂n+1 ) gives KhRn+1 for any n ≥
1. These differentials are generally involved, but their certain algebraic simplifica-
tions, suggested in [Ras, DGR] and developed further in [Ch2], work surprisingly
well for sufficiently small links. The assumption in [Ch2] is that the actual ∂n+1 are
“as surjective as possible” beginning with a = 0. See Conjecture 2.7 and Section
3.6 in [Ch2]; the “smallest” torus knot when (reduced) KhR2 cannot be obtained
this way is T (12, 7). Quite a few examples for torus iterated knots can be found in
[ChD].
Applying this procedure is possible for links, but the unreduced setting create
problems. Each particular homology is finite but there are infinitely many non-
trivial ones. We hope to address this in further works (at least in examples). Let
us consider now a special case.
Recovering KhRn+1 from JDn+1 . Generally, the procedure of obtaining the
KhR–polynomials requires knowing the whole superpolynomials and, moreover, all
differentials. However, the direct recovering reduced KhRn+1 from JDn+1 is not
impossible for small links. This was noted in [Ch2]. Due to the lack of other ways
for justifying Part (iii), this provides at least something, and such an approach is
applicable to any torus iterated links (not only algebraic ones).
Practically it works as follows. For P = J and modulo the operations of
.
changing the sign explained below (we put =), one can expect that
(1+a) H min
L, L . 4 n+1
(5.59) = KhR L, L (q, t)reduced .
(1 − t) κ q →(qt) , t →q , a →−q
2 2 2(n+1)
The latter polynomials can be calculated in few cases for n > 1 and for many
links for n = 1 using of the procedure KhReduced[·][q, t] from [KA] (the hat-
normalization must be applied). If there is no exact match, then the Connec-
tion Conjecture hints that the typical corrections are the substitutions −q l tm →
+q l tm+1 , q l tm → +q l tm+2 and so on. Here −q l tm → +q l tm−1 (etc.) can occur too
for non-algebraic links; see [ChD]. Generally the DAHA part in (5.59) is smaller
than the right-hand side, but in the examples we provide below they have the same
number of terms. Also, see the discussion after Conjecture 2.4 in [ChD]. We will
demonstrate (5.59) for some basic knots.
6. Multiple torus knots

6.1. Preliminary remarks. We will consider examples (mainly of numeri-
cal kind) confirming our theorems and the Connection Conjecture, including the
positivity claim (5.56) for algebraic links. We selected only relatively simple and
instructional ones; however, some formulas are long, which reflects the nature of
this theory. Explicit formulas are of obvious value; they are expected to contain
a lot of geometric information (beyond what we discuss here and know now). We
provide examples only for links here; see [Ch2, Ch3] for DAHA-superpolynomials
for torus knots and [ChD] for iterated torus knots.
We present colored links L in the form of (4.10) and Theorem 5.1:
Υ, (λ1 ,...,λκ )
(6.1) L = L({r1 ,s1 },...,{rκ ,sκ }) , where Υ is presented graphically.
Note that for a given j (the index of the path), we first collect rij in !rj and then
collect the corresponding sji in !sj (i.e. separate rij from sji ). The labels of vertices
are the pairs [rij , sji ], identified with respect to the incidence graph Υ. Practically,
j
- . r1 , . . . , rj
we put !rj ,!sj as , but such a 2-row presentation is graphically
sj1 , . . . , sj
unreasonable in a paper, especially in the indices of L. The square brackets [r, s] is
used only for labels in this work ; also, recall that the first left vertex is [r1j , sj1 ].
We will constantly use the natural diagrams for Υ. For instance,
{Υ, (λj )} = {◦1 → ◦2 →
→
→ , (λ1 , λ2 , λ3 )}
means that κ = 3 and there are two vertices, shown as ◦1,2 with the labels
[r1j , sj1 ], [r2j , r2j ], which are the same for j = 1, 2, 3. So the paths here are differ-
ent only by the arrowheads. Formally, we need to repeat the same !r,!s three times
(for each path) in L, but we will mostly omit coinciding !rj ,!sj if this is clear from
the graph. Thus the corresponding (labeled, colored) tree L will be presented as
→
◦→◦→ , (λ →
. or L ◦→◦
1
,λ2 ,λ3 ) 1
,λ2 ,λ3 )
L- → .- .- - →→ , (λ ..
( {r1 ,r2 },{s1 ,s2 } , {r1 ,r2 },{s1 ,s2 } , {r1 ,r2 },{s1 ,s2 } ) ( {r1 ,r2 },{s1 ,s2 } )
HOMFLY-PT polynomial. The relation (5.55) from the Connection Conjecture,

min and (sometimes)
which is equivalent to (5.54) there, is proven, so we give only H L
min
H
L (q, t → q, a → −a), where Pλ = Jλ .
The corresponding denominators denjo = denjo (L, L) are provided:
(6.2) L,
(H L (q, q, −a)/den
min jo 4 jL,o L (qst , ast ).
)st = HOM
Here ()st in the right-hand side means that one needs to substitute a → a2st , q →
2
qst to compare our polynomials with the usual presentation for the HOMFLY-PT
polynomials. To calculate the latter, we mainly use the software by S. Artamonov
for colored HOMFLY-PT polynomials; see also [AM, AMM].
Generally, in notations from (5.42) and (2.21):

jo =H
H Lmin †
Πλ1 ∨...∨λκ /Π†λjo , hλ = Jλ /Pλ = (1 − q arm(2) tleg(2)+1 ),
L
2∈λ
Π†λjo Jλ j p −1
n λ
(6.3) denjo = , Π†λ = 1−a q v t1−p .
Π†λ1 ∨...∨λκ Pλj t→q
j =jo p=1 v=0
Note that denjo = (1 − q)κ−1 in the uncolored case (for any jo ). We omitted L
here; use (6.3) with λ replaced by λ ∪ λ if it is present. Also, denjo for L, L serve
L, L∨ as well.
The connection of our construction with the HOMFLY-PT polynomials is solid
(it holds in all examples we considered), as well as Part (ii) of the Connection
Conjecture. Parts (iii, iv) are much less convincing at the moment for torus iter-
ated links due to the lack (actually the absence) of formulas for stable Khovanov-
Rozansky polynomials and their algebraic-geometric counterparts from [ORS] is
such a generality. As for the iterated knots , especially for relatively small torus
knots, there are direct confirmations (not too many). The positivity property (5.56)
and its generalizations for any weights hold for algebraic links in all examples we
considered (and powers of (1 − t) are mostly sufficient); this is an indirect confir-
mation of the Connection Conjecture.
The symmetries of DAHA superpolynomials match well those (known and ex-
pected) in geometry/topology/physics. The super-duality is important. It is proven
in [ORS] for uncolored algebraic links within their theory, which is another (in-
direct) confirmation of the Connection Conjecture. We note that this and other
symmetries provide an excellent way to verify our numerical simulations (we always
check all of them).
6.2. Uncolored 2-fold trefoil. The corresponding links will be in this section
uncolored L = T (κ r, κ s). We will mostly provide only minimal H min (for P = J);
L

there will be no L in this section. We begin with the “canonical” example of
T (6, 4).
are as follows:
2-fold trefoil. The [r,s]-presentation and H
1 ≤ j ≤ κ = 2, !rj = 3, !sj = 2, Υ = {◦ ⇒} , λ1 = = λ2 ;
(6.4)
◦⇒, (,) min
T (6, 4) : L = L({3,2},{3,2}) , H L (q, t, a) =
1 − t + qt + q 2 t + q 3 t − qt2 + 2q 4 t2 − q 2 t3 − q 4 t3 + 2q 5 t3 − q 3 t4 − q 5 t4 + 2q 6 t4 − q 4 t5 + q 7 t5 −

q 5 t6 + q 7 t6 − q 6 t7 + q 7 t7 − q 7 t8 + q 8 t8 + a3 q 6 − q 6 t + q 7 t − q 7 t2 + q 8 t2 + a2 q 3 + q 4 + q 5 −

q 3 t + q 5 t + 2q 6 t − q 4 t2 − q 5 t2 + 2q 7 t2 − q 5 t3 − q 6 t3 + q 7 t3 + q 8 t3 − q 6 t4 + q 8 t4 − q 7 t5 + q 8 t5 +

a q + q 2 + q 3 − qt + q 3 t + 3q 4 t + q 5 t − q 2 t2 − q 3 t2 − q 4 t2 + 3q 5 t2 + q 6 t2 − q 3 t3 − q 4 t3 − 2q 5 t3 +

3q 6 t3 + q 7 t3 − q 4 t4 − q 5 t4 − q 6 t4 + 3q 7 t4 − q 5 t5 − q 6 t5 + q 7 t5 + q 8 t5 − q 6 t6 + q 8 t6 − q 7 t7 + q 8 t7 .
The a–degree of H min (q, t, a) is 3, which matches the formula dega = s(2|λ|) −
L
|λ| = 3 from (5.41); see also Part (iii) of Theorem 5.1. The self-duality and other
claims in this theorem hold. For instance, the transposition r ↔ s in T (κr, κs) here
and below does not influence the superpolynomial, though the practical calculations
can be quite different depending on the order of r and s. For instance, the label
2
[3, 2] corresponds to τ+ τ− , but [2, 3] is naturally represented by τ− τ+ τ− .
Note that the series from the positivity claim from (5.56) in Conjecture 5.3
reads H min (q, t, a)/(1 − t) for uncolored 2–links. This positivity holds here and in
L
all examples we considered.
Our superpolynomial matches that suggested in [DMS] in Section 2.8. The
T [4,6]
main factor of −P1 coincides with our one upon the following substitution:
a → A , q → q , t → t2 . I.e. their non-bold A, q, t are essentially the DAHA
2 2
parameters.
HOMFLY-PT polynomial. Recall that Part (i) of the Connection Conjecture
(which is a theorem) claims that
min (q, q, −a)/den)st = HOM
(H
j
4 Lo (qst , ast ),
L
with the denominators explained above.

min (q, q, −a) = 1 − q + q 2 + q 4 − q 5 + 2q 6 −
In this example, den = (1 − q) and H L
2q 7 + 2q 8 − 2q 9 + 2q 10 − q 11 + q 12 + q 14 − q 15 + q 16 − a3 q 6 − q 7 + q 8 − q 9 + q 10 + a2 q 3 +

q 5 + q 7 − q 8 + q 9 + q 11 + q 13 − a q + q 3 + 2q 5 − q 6 + 2q 7 − 2q 8 + 2q 9 − q 10 + 2q 11 + q 13 + q 15 .
Khovanov polynomial. Let us apply the procedure from (5.59) to obtain the
DAHA approximation to the reduced Khovanov polynomial (n = 1). We switch
to the standard parameters: q → (qt)2 , t → q 2 , a → −q 2(n+1) (we add {st, n} but
omit st in qst , tst , ast ). Then
(1+a) H min
L
(6.5) =
(1 − t)2 st,n=1
1 + q 2 + q 4 t2 − q 8 t2 + q 6 t4 + q 8 t4 − q 10 t4 − q 12 t4 + q 8 t6 + q 10 t6 − q 12 t6 − q 14 t6 + 2q 12 t8 −
2q 16 t8 + q 16 t10 − q 18 t10 − q 20 t10 + q 22 t10 + q 20 t12 + q 22 t12 ,
(6.6) 4 2 )reduced =
(KhR L
1 + q 2 + q 4 t2 + q 8 t3 + q 6 t4 + q 8 t4 + q 10 t5 + q 12 t5 + q 8 t6 + q 10 t6 + q 12 t7 + q 14 t7 + 2q 12 t8 +
2q 16 t9 + q 16 t10 + q 18 t11 + q 20 t11 + q 20 t12 + q 22 t12 + q 22 t12 .
Their difference (the second minus the first) is a sum of the terms q l (tm +tm−1 ),
q l (tm − tm−2 ), which exactly matches the expectations concerning (5.59).
6.3. Similar links. Let us consider 2-fold T(2,1). One has:

1 ≤ j ≤ κ = 2, !rj = 2, !sj = 1, Υ = {◦ ⇒} , λ1 = = λ2 ;
(6.7)
◦⇒, (,) min
T (4, 2) : L = L({2,1},{2,1}) , H L (q, t, a) =
1 − t + qt − qt2 + q 2 t2 + a(q − qt + q 2 t).
The a–degree is s(2|λ|) − |λ| = 1. The self-duality reads:

min
q 2 t2 H L
min
(t−1 , q −1 , a) = H L (q, t, a),
and H is obviously positive upon division by (1 − t).
Upon the passage to the standard parameters, formula (6.7) becomes
1 − q 2 + q 4 t2 − q 6 t2 + q 8 t4 + a2 (q 2 t3 − q 4 t3 + q 6 t5 ),
which is the major factor of formula (136) from [DMS].
HOMFLY-PT polynomial. One has:

min
H L (q, q, −a) = 1 − q + q − q + q − a(q − q + q )),
2 3 4 2 3
L
H 4 L (q, a).
(q , q , −a2 )/(1 − q 2 ) = HOM
min 2 2
Khovanov polynomial. Recall that this and other formulas for (reduced) Kho-
vanov polynomials will be in terms of the standard parameters; we add the suffix
{st, n = 1} to remind this. One has:
(1+a) H min
L
(6.8) = 1 + q 2 + q 4 t2 − q 8 t2 + q 8 t4 + q 10 t4 ,
(1 − t)2 st,n=1
(6.9) 4 2 )reduced = 1 + q 2 + q 4 t2 + q 8 t3 + q 8 t4 + q 10 t4 ,
(KhR L
where the difference q 8 (t3 + t2 ) matches our expectations.

2-fold T(1,1) . This is the simplest algebraic torus link.
(6.10)
◦⇒, (,) min
T (2, 2) : L = L({1,1},{1,1}) , H L = 1 − t + qt + aq,
4 L (q, a) = (1 − q + q 2 − aq)/(1 − q).
where HOM
This is for γ = τ− . The hat-normalization of the reduced Khovanov polynomial
is 1 + q 2 + q 4 t2 + q 6 t2 , which exactly coincides with that obtained via (5.59); no
sign-corrections are necessary.
−1
We note that the superpolynomial becomes 1 + a(q + 1/t − q/t) for γ = τ− ,
i.e. for the tree L({1,−1},{1,−1}) .
◦⇒, (,)
2-fold T(m,1) . Let us provide a general formula for L = L({1,m},{1,m}) with
m > 0:
(6.11) min
H L (q, t, a) = (1 − t)(1 + qt + q 2 t2 + q 3 t3 + . . . + q m−1 tm−1 )

+ q m tm + a (1 − t)(q + q 2 t + . . . + q m−1 tm−2 ) + q m tm−1 .
◦⇒, (,)
The corresponding link is isotopic to L({m,1},{m,1}) , so their superpolynomials
must coincide. This is a simplest example of the symmetries of DAHA superpoly-
nomials in Part (ii) of Theorem 5.1. Recall that generally we can transpose r and
s only for the first vertex [r1j , sj1 ] in the presence of iterations. Topologically this is
obvious.
Let us provide a counterpart of this formula for non-algebraic L = L({1,−m},{1,−m})
with m > 0:
(6.12) H min
L (q, t, a) = (1 − q)(1 + qt + . . . + q m−2 tm−2 ) + q m−1 tm−1

1
+ a (1 − q)( + q + q 2 t + . . . + q m−1 tm−2 ) + q m tm−1 .
t
One can change here {1, −m} by {−m, 1}, the superpolynomial will remain the
same. Both formulas, for ± m, satisfy the self-duality (with proper q • t• –multipliers),
which is simple to see directly.
2-fold T(5,2). The [r,s]-parameters and H are as follows:
1 ≤ j ≤ κ = 2, !rj = 5, !sj = 2, Υ = {◦ ⇒} , λ1 = = λ2 ;
(6.13)
◦⇒, (,) min
T (10, 4) : L = L({5,2},{5,2}) , H L (q, t, a) =
1 − t + qt + q 2 t + q 3 t − qt2 + 2q 4 t2 + q 5 t2 + q 6 t2 − q 2 t3 − q 4 t3 + q 5 t3 + q 6 t3 + 2q 7 t3 − q 3 t4 −
q 5 t4 + 3q 8 t4 − q 4 t5 − q 6 t5 − q 8 t5 + 3q 9 t5 − q 5 t6 − q 7 t6 − q 9 t6 + 3q 10 t6 − q 6 t7 − q 8 t7 + 2q 11 t7 −
q 7 t8 − q 9 t8 + q 11 t8 + q 12 t8 − q 8 t9 − q 10 t9 + q 11 t9 + q 12 t9 − q 9 t10 − q 11 t10 + 2q 12 t10 − q 10 t11 +

q 13 t11 − q 11 t12 + q 13 t12 − q 12 t13 + q 13 t13 − q 13 t14 + q 14 t14 + a3 q 6 − q 6 t + q 7 t + q 8 t + q 9 t −
q 7 t2 + 2q 10 t2 − q 8 t3 − q 10 t3 + 2q 11 t3 − q 9 t4 − q 11 t4 + 2q 12 t4 − q 10 t5 + q 13 t5 − q 11 t6 + q 13 t6 −

q 12 t7 + q 13 t7 − q 13 t8 + q 14 t8 + a2 q 3 + q 4 + q 5 − q 3 t + q 5 t + 3q 6 t + 2q 7 t + q 8 t − q 4 t2 − q 5 t2 −
q 6 t2 +2q 7 t2 +3q 8 t2 +3q 9 t2 −q 5 t3 −q 6 t3 −2q 7 t3 +2q 9 t3 +4q 10 t3 −q 6 t4 −q 7 t4 −2q 8 t4 −q 9 t4 +
q 10 t4 + 4q 11 t4 − q 7 t5 − q 8 t5 − 2q 9 t5 − q 10 t5 + 2q 11 t5 + 3q 12 t5 − q 8 t6 − q 9 t6 − 2q 10 t6 + 3q 12 t6 +
q 13 t6 −q 9 t7 −q 10 t7 −2q 11 t7 +2q 12 t7 +2q 13 t7 −q 10 t8 −q 11 t8 −q 12 t8 +3q 13 t8 −q 11 t9 −q 12 t9 +

q 13 t9 + q 14 t9 − q 12 t10 + q 14 t10 − q 13 t11 + q 14 t11 + a q + q 2 + q 3 − qt + q 3 t + 3q 4 t + 2q 5 t + q 6 t −
q 2 t2 − q 3 t2 − q 4 t2 + 2q 5 t2 + 3q 6 t2 + 4q 7 t2 + q 8 t2 − q 3 t3 − q 4 t3 − 2q 5 t3 + q 7 t3 + 5q 8 t3 + 2q 9 t3 −
q 4 t4 −q 5 t4 −2q 6 t4 −q 7 t4 −q 8 t4 +5q 9 t4 +2q 10 t4 −q 5 t5 −q 6 t5 −2q 7 t5 −q 8 t5 −2q 9 t5 +5q 10 t5 +
2q 11 t5 −q 6 t6 −q 7 t6 −2q 8 t6 −q 9 t6 −q 10 t6 +5q 11 t6 +q 12 t6 −q 7 t7 −q 8 t7 −2q 9 t7 −q 10 t7 +q 11 t7 +
4q 12 t7 −q 8 t8 −q 9 t8 −2q 10 t8 +3q 12 t8 +q 13 t8 −q 9 t9 −q 10 t9 −2q 11 t9 +2q 12 t9 +2q 13 t9 −q 10 t10 −

q 11 t10 −q 12 t10 +3q 13 t10 −q 11 t11 −q 12 t11 +q 13 t11 +q 14 t11 −q 12 t12 +q 14 t12 −q 13 t13 +q 14 t13 .
The a–degree of H min (q, t, a) is 3. It is the same as for the trefoil and remains
L
unchanged for all uncolored 2-fold T (2m+1, 2) due to formula (5.41). The positivity
min (q, t, a)/(1 − t) holds. Recall that the division by (1 − t)κ−1 is presumably
of H L
sufficient to ensure the positivity for uncolored algebraic links.
HOMFLY-PT polynomial. Here den = (1 − q) and H min (q, q, −a) = 1 − q +
L
q 2 + q 4 − q 5 + 2q 6 − q 7 + 2q 8 − q 9 + 2q 10 − 2q 11 + 3q 12 − 3q 13 + 3q 14 − 3q 15 + 3q 16 − 2q 17 +

2q 18 −q 19 +2q 20 −q 21 +2q 22 −q 23 +q 24 +q 26 −q 27 +q 28 −a3 q 6 −q 7 +q 8 +q 10 −q 11 +2q 12 −

2q 13 + 2q 14 − 2q 15 + 2q 16 − q 17 + q 18 + q 20 − q 21 + q 22 − a2 −q 3 − q 5 − 2q 7 − 2q 9 − 2q 11 +

q 12 − 2q 13 + 2q 14 − 2q 15 + q 16 − 2q 17 − 2q 19 − 2q 21 − q 23 − q 25 − a q + q 3 + 2q 5 + 2q 7 + 3q 9 −

q 10 + 3q 11 − 2q 12 + 3q 13 − 3q 14 + 3q 15 − 2q 16 + 3q 17 − q 18 + 3q 19 + 2q 21 + 2q 23 + q 25 + q 27 .
Recall that one needs to divide this polynomial by den and then change our a, q
to the standard a2st , qst
2
to compare it with the usual presentation for the reduced
HOMFLY-PT polynomials.
6.4. Uncolored 2-fold T (4, 3). The family T (2m + 1, 2) is well known to
be quite special (topologically and algebraically). Let us provide two uncolored
examples for 2-fold T (3m ± 1, 3).
2-fold T(4,3). The natural notation is T (8, 6). The a–degree of H min (q, t, a)
L
min (q, t, a)/(1 − t) from
is dega = s(2|λ|) − |λ| = 5. The positivity of the series H L
are as follows:
(5.56) holds. The [r,s]-parameters and H
1 ≤ j ≤ κ = 2, !rj = 4, !sj = 3, Υ = {◦ ⇒} , λ1 = = λ2 ;
(6.14)
◦⇒, (,) min
T (8, 6) : L = L({4,3},{4,3}) , H L (q, t, a) =
1 −t+qt+q 2 t+q 3 t+q 4 t+q 5 t−qt2 +q 4 t2 +q 5 t2 +3q 6 t2 +q 7 t2 +q 8 t2 −q 2 t3 −q 4 t3 +3q 7 t3 +
2q 8 t3 + 3q 9 t3 − q 3 t4 − q 5 t4 − q 6 t4 − q 7 t4 + 2q 8 t4 + q 9 t4 + 5q 10 t4 + q 11 t4 − q 4 t5 − q 6 t5 − q 7 t5 −
2q 8 t5 +q 9 t5 +5q 11 t5 +q 12 t5 −q 5 t6 −q 7 t6 −q 8 t6 −2q 9 t6 −q 11 t6 +6q 12 t6 +q 13 t6 −q 6 t7 −q 8 t7 −
q 9 t7 − 2q 10 t7 − q 12 t7 + 5q 13 t7 + q 14 t7 − q 7 t8 − q 9 t8 − q 10 t8 − 2q 11 t8 + 5q 14 t8 − q 8 t9 − q 10 t9 −
q 11 t9 − 2q 12 t9 + q 13 t9 + q 14 t9 + 3q 15 t9 − q 9 t10 − q 11 t10 − q 12 t10 − 2q 13 t10 + 2q 14 t10 + 2q 15 t10 +
q 16 t10 −q 10 t11 −q 12 t11 −q 13 t11 −q 14 t11 +3q 15 t11 +q 16 t11 −q 11 t12 −q 13 t12 −q 14 t12 +3q 16 t12 −
q 12 t13 − q 14 t13 + q 16 t13 + q 17 t13 − q 13 t14 − q 15 t14 + q 16 t14 + q 17 t14 − q 14 t15 + q 17 t15 − q 15 t16 +

q 17 t16 − q 16 t17 + q 17 t17 − q 17 t18 + q 18 t18 + a5 q 15 − q 15 t + q 16 t − q 16 t2 + q 17 t2 − q 17 t3 + q 18 t3
10 11 12 13 14 10
+a q +q +q +q +q −q t+q t+q t+q t+2q t−q t −q t +q t +q t +
4 12 13 14 15 11 2 12 2 14 2 15 2
2q 16 t2 − q 12 t3 − q 13 t3 − q 14 t3 + q 16 t3 + 2q 17 t3 − q 13 t4 − q 14 t4 − q 15 t4 + q 16 t4 + q 17 t4 + q 18 t4 −

q 14 t5 − q 15 t5 + q 17 t5 + q 18 t5 − q 15 t6 − q 16 t6 + q 17 t6 + q 18 t6 − q 16 t7 + q 18 t7 − q 17 t8 + q 18 t8

+a3 q 6 + q 7 + 2q 8 + 2q 9 + 2q 10 + q 11 + q 12 − q 6 t + 2q 9 t + 3q 10 t + 5q 11 t + 3q 12 t + 3q 13 t +
q 14 t − q 7 t2 − q 8 t2 − 2q 9 t2 + q 11 t2 + 6q 12 t2 + 4q 13 t2 + 4q 14 t2 + q 15 t2 − q 8 t3 − q 9 t3 − 3q 10 t3 −
2q 11 t3 − 2q 12 t3 + 5q 13 t3 + 4q 14 t3 + 5q 15 t3 + q 16 t3 − q 9 t4 − q 10 t4 − 3q 11 t4 − 3q 12 t4 − 4q 13 t4 +
4q 14 t4 + 4q 15 t4 + 4q 16 t4 + q 17 t4 − q 10 t5 − q 11 t5 − 3q 12 t5 − 3q 13 t5 − 4q 14 t5 + 5q 15 t5 + 4q 16 t5 +
3q 17 t5 − q 11 t6 − q 12 t6 − 3q 13 t6 − 3q 14 t6 − 2q 15 t6 + 6q 16 t6 + 3q 17 t6 + q 18 t6 − q 12 t7 − q 13 t7 −
3q 14 t7 − 2q 15 t7 + q 16 t7 + 5q 17 t7 + q 18 t7 − q 13 t8 − q 14 t8 − 3q 15 t8 + 3q 17 t8 + 2q 18 t8 − q 14 t9 −

q 15 t9 − 2q 16 t9 + 2q 17 t9 + 2q 18 t9 − q 15 t10 − q 16 t10 + 2q 18 t10 − q 16 t11 + q 18 t11 − q 17 t12 + q 18 t12
3
+a q +q +2q +2q +2q +q +q −q t+2q t+4q t+6q t+5q t+5q t+2q t+q t−
2 4 5 6 7 8 9 3 6 7 8 9 10 11 12
q 4 t2 − q 5 t2 − 2q 6 t2 − q 7 t2 + q 8 t2 + 6q 9 t2 + 7q 10 t2 + 9q 11 t2 + 4q 12 t2 + 2q 13 t2 − q 5 t3 − q 6 t3 −
3q 7 t3 −3q 8 t3 −3q 9 t3 +2q 10 t3 +6q 11 t3 +12q 12 t3 +6q 13 t3 +3q 14 t3 −q 6 t4 −q 7 t4 −3q 8 t4 −4q 9 t4 −
5q 10 t4 − 2q 11 t4 + 2q 12 t4 + 12q 13 t4 + 6q 14 t4 + 3q 15 t4 − q 7 t5 − q 8 t5 − 3q 9 t5 − 4q 10 t5 − 6q 11 t5 −
4q 12 t5 + q 13 t5 + 12q 14 t5 + 6q 15 t5 + 2q 16 t5 − q 8 t6 − q 9 t6 − 3q 10 t6 − 4q 11 t6 − 6q 12 t6 − 4q 13 t6 +
2q 14 t6 + 12q 15 t6 + 4q 16 t6 + q 17 t6 − q 9 t7 − q 10 t7 − 3q 11 t7 − 4q 12 t7 − 6q 13 t7 − 2q 14 t7 + 6q 15 t7 +
9q 16 t7 + 2q 17 t7 − q 10 t8 − q 11 t8 − 3q 12 t8 − 4q 13 t8 − 5q 14 t8 + 2q 15 t8 + 7q 16 t8 + 5q 17 t8 − q 11 t9 −
q 12 t9 −3q 13 t9 −4q 14 t9 −3q 15 t9 +6q 16 t9 +5q 17 t9 +q 18 t9 −q 12 t10 −q 13 t10 −3q 14 t10 −3q 15 t10 +
q 16 t10 + 6q 17 t10 + q 18 t10 − q 13 t11 − q 14 t11 − 3q 15 t11 − q 16 t11 + 4q 17 t11 + 2q 18 t11 − q 14 t12 −

q 15 t12 −2q 16 t12 +2q 17 t12 +2q 18 t12 −q 15 t13 −q 16 t13 +2q 18 t13 −q 16 t14 +q 18 t14 −q 17 t15 +q 18 t15

+a q +q 2 +q 3 +q 4 +q 5 −qt+q 3 t+2q 4 t+3q 5 t+5q 6 t+3q 7 t+2q 8 t+q 9 t−q 2 t2 −q 3 t2 −q 4 t2 +
q 6 t2 + 6q 7 t2 + 6q 8 t2 + 6q 9 t2 + 3q 10 t2 + q 11 t2 − q 3 t3 − q 4 t3 − 2q 5 t3 − 2q 6 t3 − 2q 7 t3 + 3q 8 t3 +
5q 9 t3 +10q 10 t3 +6q 11 t3 +2q 12 t3 −q 4 t4 −q 5 t4 −2q 6 t4 −3q 7 t4 −4q 8 t4 +q 10 t4 +10q 11 t4 +8q 12 t4 +
3q 13 t4 −q 5 t5 −q 6 t5 −2q 7 t5 −3q 8 t5 −5q 9 t5 −2q 10 t5 −2q 11 t5 +9q 12 t5 +8q 13 t5 +3q 14 t5 −q 6 t6 −
q 7 t6 −2q 8 t6 −3q 9 t6 −5q 10 t6 −3q 11 t6 −3q 12 t6 +9q 13 t6 +8q 14 t6 +2q 15 t6 −q 7 t7 −q 8 t7 −2q 9 t7 −
3q 10 t7 − 5q 11 t7 − 3q 12 t7 − 2q 13 t7 + 10q 14 t7 + 6q 15 t7 + q 16 t7 − q 8 t8 − q 9 t8 − 2q 10 t8 − 3q 11 t8 −
5q 12 t8 − 2q 13 t8 + q 14 t8 + 10q 15 t8 + 3q 16 t8 − q 9 t9 − q 10 t9 − 2q 11 t9 − 3q 12 t9 − 5q 13 t9 + 5q 15 t9 +
6q 16 t9 + q 17 t9 − q 10 t10 − q 11 t10 − 2q 12 t10 − 3q 13 t10 − 4q 14 t10 + 3q 15 t10 + 6q 16 t10 + 2q 17 t10 −
q 11 t11 − q 12 t11 − 2q 13 t11 − 3q 14 t11 − 2q 15 t11 + 6q 16 t11 + 3q 17 t11 − q 12 t12 − q 13 t12 − 2q 14 t12 −
2q 15 t12 + q 16 t12 + 5q 17 t12 − q 13 t13 − q 14 t13 − 2q 15 t13 + 3q 17 t13 + q 18 t13 − q 14 t14 − q 15 t14 −

q 16 t14 +2q 17 t14 +q 18 t14 −q 15 t15 −q 16 t15 +q 17 t15 +q 18 t15 −q 16 t16 +q 18 t16 −q 17 t17 +q 18 t17 .
min (q, q, −a) = 1 − q +

HOMFLY-PT polynomial. Here den = (1 − q) and H L
q + q 4 + 2q 6 − q 7 + 3q 8 − q 9 + 3q 10 − q 11 + 4q 12 − 3q 13 + 5q 14 − 3q 15 + 4q 16 − 4q 17 + 5q 18 −
2
4q 19 +4q 20 −3q 21 +5q 22 −3q 23 +4q 24 −q 25 +3q 26 −q 27 +3q 28 −q 29 +2q 30 +q 32 +q 34 −q 35 +

q 36 −a5 q 15 −q 16 +q 17 −q 18 +q 19 −q 20 +q 21 −a4 −q 10 −q 12 −q 13 −q 14 −2q 16 +q 17 −q 18 +

q 19 − 2q 20 − q 22 − q 23 − q 24 − q 26 − a3 q 6 + 2q 8 + q 9 + 3q 10 + q 11 + 5q 12 + 6q 14 − q 15 + 5q 16 −

3q 17 + 5q 18 − 3q 19 + 5q 20 − q 21 + 6q 22 + 5q 24 + q 25 + 3q 26 + q 27 + 2q 28 + q 30 − a2 −q 3 − 2q 5 −
q 6 −3q 7 −2q 8 −5q 9 −2q 10 −7q 11 −2q 12 −7q 13 −q 14 −7q 15 +2q 16 −6q 17 +3q 18 −6q 19 +2q 20 −

7q 21 − q 22 − 7q 23 − 2q 24 − 7q 25 − 2q 26 − 5q 27 − 2q 28 − 3q 29 − q 30 − 2q 31 − q 33 − a q + q 3 + q 4 +
2q 5 + q 6 + 4q 7 + q 8 + 5q 9 + 2q 10 + 5q 11 + q 12 + 7q 13 − q 14 + 6q 15 − 2q 16 + 5q 17 − 3q 18 + 5q 19 −

2q 20 + 6q 21 − q 22 + 7q 23 + q 24 + 5q 25 + 2q 26 + 5q 27 + q 28 + 4q 29 + q 30 + 2q 31 + q 32 + q 33 + q 35 .
Khovanov polynomial. Let us apply the procedure from (5.59) to obtain the
DAHA approximation to the reduced Khovanov polynomial (n = 1). Recall that
we switch to the standard parameters: q → (qt)2 , t → q 2 , a → −q 2(n+1) ; st is
omitted in qst , tst , ast . Then

(1+a) H
min
L
(6.15) =
(1 − t)2 st,n=1
1 + q 2 + q 4 t2 − q 8 t2 + q 6 t4 + q 8 t4 − q 10 t4 − q 12 t4 + q 8 t6 + q 10 t6 − q 12 t6 − q 14 t6 + q 10 t8 +
2q 12 t8 − q 14 t8 − 2q 16 t8 + q 12 t10 + 2q 14 t10 − q 16 t10 − 3q 18 t10 + q 22 t10 + 3q 16 t12 + q 18 t12 −
4q 20 t12 − q 22 t12 + q 24 t12 + q 18 t14 + 2q 20 t14 − 3q 22 t14 − 3q 24 t14 + 2q 26 t14 + q 28 t14 + q 20 t16 +
2q 22 t16 − q 24 t16 − 3q 26 t16 + q 30 t16 + 2q 24 t18 − 3q 28 t18 + q 32 t18 + 2q 28 t20 − q 30 t20 − 2q 32 t20 +
q 34 t20 + q 32 t22 − q 34 t22 − q 36 t22 + q 38 t22 + q 36 t24 + q 38 t24 ,
(6.16) 4 2 )reduced =
(KhR L
1 + q 2 + q 4 t2 + q 8 t3 + q 6 t4 + q 8 t4 + q 10 t5 + q 12 t5 + q 8 t6 + q 10 t6 + q 12 t7 + q 14 t7 + q 10 t8 +
2q 12 t8 + q 14 t9 + 2q 16 t9 + q 12 t10 + 2q 14 t10 + q 16 t11 + 3q 18 t11 + 3q 16 t12 + q 18 t12 + q 22 t12 +
4q 20 t13 + q 22 t13 + q 18 t14 + 2q 20 t14 + q 24 t14 + 3q 22 t15 + 3q 24 t15 + q 20 t16 + 2q 22 t16 + 2q 26 t16 +
q 28 t16 + q 24 t17 + 3q 26 t17 + 2q 24 t18 + q 30 t18 + 3q 28 t19 + 2q 28 t20 + q 32 t20 + q 30 t21 + 2q 32 t21 +
q 32 t22 + q 34 t22 + q 34 t23 + q 36 t23 + q 36 t24 + 2q 38 t24 .
Their difference (the second minus the first) is large, but still a sum of the
terms q l (tm + tm−1 ), q l (t m
− tm−2 ) with
8 2
positive coefficients,

which

confirms the
Connection Conjecture: q t +q 8 t3 + q 10 t4 +q 10 t5 + q 12 t4 +q 12 t5 + q 12 t6 +q 12 t7 +
14 6
q t +q 14 t7 + q 14 t8 +q 14 t9 + 2q 16 t8 +2q 16 t9 + q 16 t10 +q 16 t11 + 3q 18 t10 +3q 18 t11 +
22 10
−q t +2q 22 t12 +q 22 t13 + 4q 20 t12 +4q 20 t13 + −q 24 t12 +4q 24 t14 +3q 24 t15 + 3q 22 t14 +

3q 22 t15 + q 24 t16 +q 24 t17 + −2q 26 t14 +5q 26 t16 +3q 26 t17 + −q 28 t14 +q 28 t16 + −q 30 t16 +

q 30 t18 + 3q 28 t18 + 3q 28 t19 + −q 32 t18 + q 32 t20 + q 30 t20 + q 30 t21 + 2q 32 t20 + 2q 32 t21 +
34 20
−q t + q 34 t22 + q 34 t22 + q 34 t23 + q 36 t22 + q 36 t23 + −q 38 t22 + q 38 t24 .
6.5. Uncolored 2-fold T (5, 3).

(6.17)
◦⇒, (,) min
T (10, 6) : L = L({5,3},{5,3}) , H L (q, t, a) =
1 − t + qt + q 2 t + q 3 t + q 4 t + q 5 t − qt2 + q 4 t2 + q 5 t2 + 3q 6 t2 + 2q 7 t2 + 2q 8 t2 − q 2 t3 − q 4 t3 +
2q 7 t3 + 2q 8 t3 + 5q 9 t3 + 2q 10 t3 + q 11 t3 − q 3 t4 − q 5 t4 − q 6 t4 − q 7 t4 + q 8 t4 + 5q 10 t4 + 4q 11 t4 +
4q 12 t4 − q 4 t5 − q 6 t5 − q 7 t5 − 2q 8 t5 − q 10 t5 + 3q 11 t5 + 3q 12 t5 + 7q 13 t5 + q 14 t5 − q 5 t6 − q 7 t6 −
q 8 t6 − 2q 9 t6 − q 10 t6 − 2q 11 t6 + 2q 12 t6 + q 13 t6 + 8q 14 t6 + 2q 15 t6 − q 6 t7 − q 8 t7 − q 9 t7 − 2q 10 t7 −
q 11 t7 −3q 12 t7 +q 13 t7 +8q 15 t7 +2q 16 t7 −q 7 t8 −q 9 t8 −q 10 t8 −2q 11 t8 −q 12 t8 −3q 13 t8 +8q 16 t8 +
2q 17 t8 − q 8 t9 − q 10 t9 − q 11 t9 − 2q 12 t9 − q 13 t9 − 3q 14 t9 + 8q 17 t9 + q 18 t9 − q 9 t10 − q 11 t10 −
q 12 t10 − 2q 13 t10 − q 14 t10 − 3q 15 t10 + q 16 t10 + q 17 t10 + 7q 18 t10 − q 10 t11 − q 12 t11 − q 13 t11 −
2q 14 t11 − q 15 t11 − 3q 16 t11 + 2q 17 t11 + 3q 18 t11 + 4q 19 t11 − q 11 t12 − q 13 t12 − q 14 t12 − 2q 15 t12 −
q 16 t12 − 2q 17 t12 + 3q 18 t12 + 4q 19 t12 + q 20 t12 − q 12 t13 − q 14 t13 − q 15 t13 − 2q 16 t13 − q 17 t13 −
q 18 t13 + 5q 19 t13 + 2q 20 t13 − q 13 t14 − q 15 t14 − q 16 t14 − 2q 17 t14 + 5q 20 t14 − q 14 t15 − q 16 t15 −
q 17 t15 − 2q 18 t15 + q 19 t15 + 2q 20 t15 + 2q 21 t15 − q 15 t16 − q 17 t16 − q 18 t16 − q 19 t16 + 2q 20 t16 +
2q 21 t16 − q 16 t17 − q 18 t17 − q 19 t17 + 3q 21 t17 − q 17 t18 − q 19 t18 + q 21 t18 + q 22 t18 − q 18 t19 −
q 20 t19 + q 21 t19 + q 22 t19 − q 19 t20 + q 22 t20 − q 20 t21 + q 22 t21 − q 21 t22 + q 22 t22 − q 22 t23 + q 23 t23

+a5 q 15 − q 15 t + q 16 t + q 17 t + q 18 t − q 16 t2 + 2q 19 t2 − q 17 t3 − q 19 t3 + 2q 20 t3 − q 18 t4 − q 20 t4 +

2q 21 t4 − q 19 t5 + q 22 t5 − q 20 t6 + q 22 t6 − q 21 t7 + q 22 t7 − q 22 t8 + q 23 t8

+a4 q 10 +q 11 +q 12 +q 13 +q 14 −q 10 t+q 12 t+2q 13 t+3q 14 t+4q 15 t+2q 16 t+q 17 t−q 11 t2 −q 12 t2 −
q t + 2q 15 t2 + 5q 16 t2 + 4q 17 t2 + 3q 18 t2 − q 12 t3 − q 13 t3 − 2q 14 t3 − 2q 15 t3 + 4q 17 t3 + 4q 18 t3 +
13 2
4q 19 t3 − q 13 t4 − q 14 t4 − 2q 15 t4 − 3q 16 t4 − 2q 17 t4 + 3q 18 t4 + 4q 19 t4 + 4q 20 t4 − q 14 t5 − q 15 t5 −
2q 16 t5 − 3q 17 t5 − 3q 18 t5 + 3q 19 t5 + 4q 20 t5 + 3q 21 t5 − q 15 t6 − q 16 t6 − 2q 17 t6 − 3q 18 t6 − 2q 19 t6 +
4q 20 t6 + 4q 21 t6 + q 22 t6 − q 16 t7 − q 17 t7 − 2q 18 t7 − 3q 19 t7 + 5q 21 t7 + 2q 22 t7 − q 17 t8 − q 18 t8 −
2q 19 t8 − 2q 20 t8 + 2q 21 t8 + 4q 22 t8 − q 18 t9 − q 19 t9 − 2q 20 t9 + 3q 22 t9 + q 23 t9 − q 19 t10 − q 20 t10 −
q 21 t10 + 2q 22 t10 + q 23 t10 − q 20 t11 − q 21 t11 + q 22 t11 + q 23 t11 − q 21 t12 + q 23 t12 − q 22 t13 + q 23 t13

+a3 q 6 +q 7 +2q 8 +2q 9 +2q 10 +q 11 +q 12 −q 6 t+2q 9 t+4q 10 t+7q 11 t+6q 12 t+6q 13 t+3q 14 t+
q t − q 7 t2 − q 8 t2 − 2q 9 t2 − q 10 t2 + 6q 12 t2 + 8q 13 t2 + 12q 14 t2 + 8q 15 t2 + 5q 16 t2 + q 17 t2 −
15
q 8 t3 − q 9 t3 − 3q 10 t3 − 3q 11 t3 − 4q 12 t3 + q 13 t3 + 4q 14 t3 + 14q 15 t3 + 11q 16 t3 + 8q 17 t3 + 2q 18 t3 −
q 9 t4 − q 10 t4 − 3q 11 t4 − 4q 12 t4 − 6q 13 t4 − 3q 14 t4 − 2q 15 t4 + 12q 16 t4 + 12q 17 t4 + 10q 18 t4 +
2q 19 t4 − q 10 t5 − q 11 t5 − 3q 12 t5 − 4q 13 t5 − 7q 14 t5 − 5q 15 t5 − 6q 16 t5 + 9q 17 t5 + 12q 18 t5 +
10q 19 t5 + 2q 20 t5 − q 11 t6 − q 12 t6 − 3q 13 t6 − 4q 14 t6 − 7q 15 t6 − 6q 16 t6 − 7q 17 t6 + 9q 18 t6 +
12q 19 t6 +8q 20 t6 +q 21 t6 −q 12 t7 −q 13 t7 −3q 14 t7 −4q 15 t7 −7q 16 t7 −6q 17 t7 −6q 18 t7 +12q 19 t7 +
11q 20 t7 + 5q 21 t7 − q 13 t8 − q 14 t8 − 3q 15 t8 − 4q 16 t8 − 7q 17 t8 − 5q 18 t8 − 2q 19 t8 + 14q 20 t8 +
8q 21 t8 + q 22 t8 − q 14 t9 − q 15 t9 − 3q 16 t9 − 4q 17 t9 − 7q 18 t9 − 3q 19 t9 + 4q 20 t9 + 12q 21 t9 + 3q 22 t9 −
q 15 t10 − q 16 t10 − 3q 17 t10 − 4q 18 t10 − 6q 19 t10 + q 20 t10 + 8q 21 t10 + 6q 22 t10 − q 16 t11 − q 17 t11 −
3q 18 t11 − 4q 19 t11 − 4q 20 t11 + 6q 21 t11 + 6q 22 t11 + q 23 t11 − q 17 t12 − q 18 t12 − 3q 19 t12 − 3q 20 t12 +
7q 22 t12 + q 23 t12 − q 18 t13 − q 19 t13 − 3q 20 t13 − q 21 t13 + 4q 22 t13 + 2q 23 t13 − q 19 t14 − q 20 t14 −
2q 21 t14 + 2q 22 t14 + 2q 23 t14 − q 20 t15 − q 21 t15 + 2q 23 t15 − q 21 t16 + q 23 t16 − q 22 t17 + q 23 t17

+a2 q 3 + q 4 + 2q 5 + 2q 6 + 2q 7 + q 8 + q 9 − q 3 t + 2q 6 t + 4q 7 t + 7q 8 t + 7q 9 t + 7q 10 t + 4q 11 t +
2q 12 t − q 4 t2 − q 5 t2 − 2q 6 t2 − q 7 t2 + 5q 9 t2 + 9q 10 t2 + 14q 11 t2 + 12q 12 t2 + 10q 13 t2 + 3q 14 t2 +
q 15 t2 − q 5 t3 − q 6 t3 − 3q 7 t3 − 3q 8 t3 − 4q 9 t3 − q 10 t3 + 3q 11 t3 + 13q 12 t3 + 16q 13 t3 + 18q 14 t3 +
8q 15 t3 + 3q 16 t3 − q 6 t4 − q 7 t4 − 3q 8 t4 − 4q 9 t4 − 6q 10 t4 − 5q 11 t4 − 4q 12 t4 + 6q 13 t4 + 14q 14 t4 +
23q 15 t4 + 12q 16 t4 + 5q 17 t4 − q 7 t5 − q 8 t5 − 3q 9 t5 − 4q 10 t5 − 7q 11 t5 − 7q 12 t5 − 8q 13 t5 − q 14 t5 +
8q 15 t5 +24q 16 t5 +14q 17 t5 +6q 18 t5 −q 8 t6 −q 9 t6 −3q 10 t6 −4q 11 t6 −7q 12 t6 −8q 13 t6 −10q 14 t6 −
5q 15 t6 +4q 16 t6 +23q 17 t6 +14q 18 t6 +5q 19 t6 −q 9 t7 −q 10 t7 −3q 11 t7 −4q 12 t7 −7q 13 t7 −8q 14 t7 −
11q 15 t7 − 6q 16 t7 + 4q 17 t7 + 24q 18 t7 + 12q 19 t7 + 3q 20 t7 − q 10 t8 − q 11 t8 − 3q 12 t8 − 4q 13 t8 −
7q 14 t8 −8q 15 t8 −11q 16 t8 −5q 17 t8 +8q 18 t8 +23q 19 t8 +8q 20 t8 +q 21 t8 −q 11 t9 −q 12 t9 −3q 13 t9 −
4q 14 t9 − 7q 15 t9 − 8q 16 t9 − 10q 17 t9 − q 18 t9 + 14q 19 t9 + 18q 20 t9 + 3q 21 t9 − q 12 t10 − q 13 t10 −
3q 14 t10 − 4q 15 t10 − 7q 16 t10 − 8q 17 t10 − 8q 18 t10 + 6q 19 t10 + 16q 20 t10 + 10q 21 t10 − q 13 t11 −
q 14 t11 − 3q 15 t11 − 4q 16 t11 − 7q 17 t11 − 7q 18 t11 − 4q 19 t11 + 13q 20 t11 + 12q 21 t11 + 2q 22 t11 −
q 14 t12 −q 15 t12 −3q 16 t12 −4q 17 t12 −7q 18 t12 −5q 19 t12 +3q 20 t12 +14q 21 t12 +4q 22 t12 −q 15 t13 −
q 16 t13 − 3q 17 t13 − 4q 18 t13 − 6q 19 t13 − q 20 t13 + 9q 21 t13 + 7q 22 t13 − q 16 t14 − q 17 t14 − 3q 18 t14 −
4q 19 t14 − 4q 20 t14 + 5q 21 t14 + 7q 22 t14 + q 23 t14 − q 17 t15 − q 18 t15 − 3q 19 t15 − 3q 20 t15 + 7q 22 t15 +
q 23 t15 − q 18 t16 − q 19 t16 − 3q 20 t16 − q 21 t16 + 4q 22 t16 + 2q 23 t16 − q 19 t17 − q 20 t17 − 21 17
2q t +
2q t + 2q t − q t − q t + 2q t − q t + q t − q t + q t
22 17 23 17 20 18 21 18 23 18 21 19 23 19 22 20 23 20

+a q + q 2 + q 3 + q 4 + q 5 − qt + q 3 t + 2q 4 t + 3q 5 t + 5q 6 t + 4q 7 t + 3q 8 t + q 9 t − q 2 t2 − q 3 t2 −
q t + q 6 t2 + 5q 7 t2 + 7q 8 t2 + 10q 9 t2 + 7q 10 t2 + 4q 11 t2 + q 12 t2 − q 3 t3 − q 4 t3 − 2q 5 t3 − 2q 6 t3 −
4 2
2q 7 t3 + q 8 t3 + 3q 9 t3 + 11q 10 t3 + 12q 11 t3 + 11q 12 t3 + 5q 13 t3 + q 14 t3 − q 4 t4 − q 5 t4 − 2q 6 t4 −

3q 7 t4 − 4q 8 t4 − 2q 9 t4 − 2q 10 t4 + 6q 11 t4 + 11q 12 t4 + 17q 13 t4 + 10q 14 t4 + 3q 15 t4 − q 5 t5 − q 6 t5 −
2q 7 t5 − 3q 8 t5 − 5q 9 t5 − 4q 10 t5 − 5q 11 t5 + q 12 t5 + 5q 13 t5 + 18q 14 t5 + 14q 15 t5 + 5q 16 t5 − q 6 t6 −
q 7 t6 − 2q 8 t6 − 3q 9 t6 − 5q 10 t6 − 5q 11 t6 − 7q 12 t6 − 2q 13 t6 + 16q 15 t6 + 15q 16 t6 + 6q 17 t6 − q 7 t7 −
q 8 t7 − 2q 9 t7 − 3q 10 t7 − 5q 11 t7 − 5q 12 t7 − 8q 13 t7 − 4q 14 t7 − 2q 15 t7 + 15q 16 t7 + 15q 17 t7 +
5q 18 t7 − q 8 t8 − q 9 t8 − 2q 10 t8 − 3q 11 t8 − 5q 12 t8 − 5q 13 t8 − 8q 14 t8 − 5q 15 t8 − 2q 16 t8 + 16q 17 t8 +
14q 18 t8 +3q 19 t8 −q 9 t9 −q 10 t9 −2q 11 t9 −3q 12 t9 −5q 13 t9 −5q 14 t9 −8q 15 t9 −4q 16 t9 +18q 18 t9 +
10q 19 t9 + q 20 t9 − q 10 t10 − q 11 t10 − 2q 12 t10 − 3q 13 t10 − 5q 14 t10 − 5q 15 t10 − 8q 16 t10 − 2q 17 t10 +
5q 18 t10 +17q 19 t10 +5q 20 t10 −q 11 t11 −q 12 t11 −2q 13 t11 −3q 14 t11 −5q 15 t11 −5q 16 t11 −7q 17 t11 +
q 18 t11 +11q 19 t11 +11q 20 t11 +q 21 t11 −q 12 t12 −q 13 t12 −2q 14 t12 −3q 15 t12 −5q 16 t12 −5q 17 t12 −
5q 18 t12 +6q 19 t12 +12q 20 t12 +4q 21 t12 −q 13 t13 −q 14 t13 −2q 15 t13 −3q 16 t13 −5q 17 t13 −4q 18 t13 −
2q 19 t13 +11q 20 t13 +7q 21 t13 −q 14 t14 −q 15 t14 −2q 16 t14 −3q 17 t14 −5q 18 t14 −2q 19 t14 +3q 20 t14 +
10q 21 t14 + q 22 t14 − q 15 t15 − q 16 t15 − 2q 17 t15 − 3q 18 t15 − 4q 19 t15 + q 20 t15 + 7q 21 t15 + 3q 22 t15 −
q 16 t16 − q 17 t16 − 2q 18 t16 − 3q 19 t16 − 2q 20 t16 + 5q 21 t16 + 4q 22 t16 − q 17 t17 − q 18 t17 − 2q 19 t17 −
2q 20 t17 + q 21 t17 + 5q 22 t17 − q 18 t18 − q 19 t18 − 2q 20 t18 + 3q 22 t18 + q 23 t18 − q 19 t19 − q 20 t19 −

The a–degree of H min (q, t, a) is 5, the same as that for the 2-fold T (4, 3). The
L
min (q, t, a)/(1 − t) holds.
positivity of the series H L
6.6. Uncolored 3-links. We will begin with the case of the uncolored 3-folded
trefoil defined as follows.
are:
3-fold trefoil. The corresponding [r,s]-presentation and H
1 ≤ j ≤ κ = 3, !rj = 3, !sj = 2, Υ = {◦→
→ }, λ = = λ = λ ;
→
1 2 3
◦→ →
(6.18) → , (,,)
T (9, 6) : L = L({3,2},{3,2},{3,2}) min
, H L (q, t, a) =
1 − 2t + qt + q 2 t + q 3 t + q 4 t + q 5 t + t2 − 2qt2 − q 2 t2 − q 3 t2 + 3q 6 t2 + 2q 7 t2 + q 8 t2 + qt3 − q 2 t3 −
2q 4 t3 − q 5 t3 − 3q 6 t3 + 2q 8 t3 + 4q 9 t3 + q 10 t3 + q 2 t4 − q 3 t4 + q 4 t4 − q 5 t4 − q 6 t4 − 3q 7 t4 − 2q 8 t4 −
3q 9 t4 + 4q 10 t4 + 4q 11 t4 + q 12 t4 + q 3 t5 − q 4 t5 + q 5 t5 − 3q 8 t5 − q 9 t5 − 5q 10 t5 − q 11 t5 + 6q 12 t5 +
3q 13 t5 + q 4 t6 − q 5 t6 + q 6 t6 + q 8 t6 − 2q 9 t6 − q 10 t6 − 4q 11 t6 − 4q 12 t6 + 5q 13 t6 + 3q 14 t6 + q 15 t6 +
q 5 t7 − q 6 t7 + q 7 t7 + q 9 t7 − q 10 t7 − 5q 12 t7 − 5q 13 t7 + 6q 14 t7 + 3q 15 t7 + q 6 t8 − q 7 t8 + q 8 t8 +
q 10 t8 − q 11 t8 + q 12 t8 − 5q 13 t8 − 5q 14 t8 + 5q 15 t8 + 3q 16 t8 + q 7 t9 − q 8 t9 + q 9 t9 + q 11 t9 − q 12 t9 +
q 13 t9 − 5q 14 t9 − 4q 15 t9 + 6q 16 t9 + q 17 t9 + q 8 t10 − q 9 t10 + q 10 t10 + q 12 t10 − q 13 t10 − 4q 15 t10 −
q 16 t10 + 4q 17 t10 + q 9 t11 − q 10 t11 + q 11 t11 + q 13 t11 − q 14 t11 − q 15 t11 − 5q 16 t11 + 4q 17 t11 +
q 18 t11 + q 10 t12 − q 11 t12 + q 12 t12 + q 14 t12 − 2q 15 t12 − q 16 t12 − 3q 17 t12 + 4q 18 t12 + q 11 t13 −
q 12 t13 + q 13 t13 + q 15 t13 − 3q 16 t13 − 2q 17 t13 + 2q 18 t13 + q 19 t13 + q 12 t14 − q 13 t14 + q 14 t14 −
3q 17 t14 + 2q 19 t14 + q 13 t15 − q 14 t15 + q 15 t15 − q 17 t15 − 3q 18 t15 + 3q 19 t15 + q 14 t16 − q 15 t16 +
q 16 t16 −q 17 t16 −q 18 t16 +q 20 t16 +q 15 t17 −q 16 t17 +q 17 t17 −2q 18 t17 +q 20 t17 +q 16 t18 −q 17 t18 −
q 19 t18 +q 20t18 +q 17 t19 −q 18 t19 −q 19 t19 +q 20 t19 +q 18 t20 −2q 19 t20 +q 20 t20 +q 19 t21 −2q 20 t21 +
q 21 t21 + a5 q 15 − 2q 15 t + q 16 t + q 17 t + q 15 t2 − 2q 16 t2 − q 17 t2 + q 18 t2 + q 19 t2 + q 16 t3 − q 17 t3 −

q 18 t3 + q 19 t3 + q 17 t4 − q 18 t4 − q 19 t4 + q 20 t4 + q 18 t5 − 2q 19 t5 + q 20 t5 + q 19 t6 − 2q 20 t6 + q 21 t6

+a4 q 10 + q 11 + q 12 + q 13 + q 14 − 2q 10 t − q 11 t + q 13 t + q 14 t + 3q 15 t + q 16 t + q 10 t2 − q 11 t2 −
2q 12 t2 − 3q 13 t2 − q 14 t2 − q 15 t2 + 3q 16 t2 + 3q 17 t2 + q 18 t2 + q 11 t3 − 3q 14 t3 − 3q 15 t3 − q 16 t3 +
2q 17 t3 + 3q 18 t3 + q 19 t3 + q 12 t4 + q 14 t4 − q 15 t4 − 4q 16 t4 − 2q 17 t4 + 2q 18 t4 + 3q 19 t4 + q 13 t5 +
q 15 t5 − q 16 t5 − 4q 17 t5 − q 18 t5 + 3q 19 t5 + q 20 t5 + q 14 t6 + q 16 t6 − q 17 t6 − 3q 18 t6 − q 19 t6 +
3q 20 t6 + q 15 t7 + q 17 t7 − 3q 18 t7 − q 19 t7 + q 20 t7 + q 21 t7 + q 16 t8 − 3q 19 t8 + q 20 t8 + q 21 t8 +
q 17 t9 − 2q 19 t9 + q 21 t9 + q 18 t10 − q 19 t10 − q 20 t10 + q 21 t10 + q 19 t11 − 2q 20 t11 + q 21 t11

+a3 q 6 + q 7 + 2q 8 + 2q 9 + 2q 10 + q 11 + q 12 − 2q 6 t − q 7 t − 2q 8 t + 2q 10 t + 5q 11 t + 4q 12 t +
4q 13 t + 2q 14 t + q 6 t2 − q 7 t2 − q 8 t2 − 4q 9 t2 − 5q 10 t2 − 5q 11 t2 + q 12 t2 + 4q 13 t2 + 6q 14 t2 +
5q 15 t2 + 2q 16 t2 + q 7 t3 + q 9 t3 − 2q 10 t3 − 4q 11 t3 − 9q 12 t3 − 6q 13 t3 + q 14 t3 + 7q 15 t3 + 7q 16 t3 +
3q 17 t3 + q 18 t3 + q 8 t4 + 2q 10 t4 − q 12 t4 − 7q 13 t4 − 10q 14 t4 − 3q 15 t4 + 7q 16 t4 + 8q 17 t4 + 3q 18 t4 +
q 9 t5 + 2q 11 t5 + q 12 t5 + q 13 t5 − 5q 14 t5 − 11q 15 t5 − 5q 16 t5 + 7q 17 t5 + 7q 18 t5 + 2q 19 t5 + q 10 t6 +
2q 12 t6 + q 13 t6 + 2q 14 t6 − 4q 15 t6 − 11q 16 t6 − 3q 17 t6 + 7q 18 t6 + 5q 19 t6 + q 11 t7 + 2q 13 t7 +
q 14 t7 + 2q 15 t7 − 5q 16 t7 − 10q 17 t7 + q 18 t7 + 6q 19 t7 + 2q 20 t7 + q 12 t8 + 2q 14 t8 + q 15 t8 + q 16 t8 −
7q 17 t8 − 6q 18 t8 + 4q 19 t8 + 4q 20 t8 + q 13 t9 + 2q 15 t9 + q 16 t9 − q 17 t9 − 9q 18 t9 + q 19 t9 + 4q 20 t9 +
q 21 t9 + q 14 t10 + 2q 16 t10 − 4q 18 t10 − 5q 19 t10 + 5q 20 t10 + q 21 t10 + q 15 t11 + 2q 17 t11 − 2q 18 t11 −
5q 19 t11 + 2q 20 t11 + 2q 21 t11 + q 16 t12 + q 18 t12 − 4q 19 t12 + 2q 21 t12 + q 17 t13 − q 19 t13 − 2q 20 t13 +
2q 21 t13 + q 18 t14 − q 19 t14 − q 20 t14 + q 21 t14 + q 19 t15 − 2q 20 t15 + q 21 t15

+a2 q 3 + q 4 + 2q 5 + 2q 6 + 2q 7 + q 8 + q 9 − 2q 3 t − q 4 t − 2q 5 t + 2q 7 t + 6q 8 t + 5q 9 t + 6q 10 t +
3q 11 t + q 12 t + q 3 t2 − q 4 t2 − q 5 t2 − 4q 6 t2 − 5q 7 t2 − 7q 8 t2 + 3q 10 t2 + 9q 11 t2 + 8q 12 t2 + 5q 13 t2 +
q 14 t2 + q 4 t3 + q 6 t3 − 2q 7 t3 − 3q 8 t3 − 10q 9 t3 − 9q 10 t3 − 7q 11 t3 + 5q 12 t3 + 11q 13 t3 + 9q 14 t3 +
4q 15 t3 + q 16 t3 + q 5 t4 + 2q 7 t4 − 6q 10 t4 − 9q 11 t4 − 16q 12 t4 − 3q 13 t4 + 12q 14 t4 + 12q 15 t4 +
6q 16 t4 + q 17 t4 + q 6 t5 + 2q 8 t5 + q 9 t5 + 2q 10 t5 − 3q 11 t5 − 4q 12 t5 − 18q 13 t5 − 11q 14 t5 + 10q 15 t5 +
13q 16 t5 + 6q 17 t5 + q 18 t5 + q 7 t6 + 2q 9 t6 + q 10 t6 + 3q 11 t6 − q 12 t6 − 2q 13 t6 − 17q 14 t6 − 13q 15 t6 +
10q 16 t6 + 12q 17 t6 + 4q 18 t6 + q 8 t7 + 2q 10 t7 + q 11 t7 + 3q 12 t7 − q 14 t7 − 17q 15 t7 − 11q 16 t7 +
12q 17 t7 +9q 18 t7 +q 19 t7 +q 9 t8 +2q 11 t8 +q 12 t8 +3q 13 t8 −2q 15 t8 −18q 16 t8 −3q 17 t8 +11q 18 t8 +
5q 19 t8 + q 10 t9 + 2q 12 t9 + q 13 t9 + 3q 14 t9 − q 15 t9 − 4q 16 t9 − 16q 17 t9 + 5q 18 t9 + 8q 19 t9 + q 20 t9 +
q 11 t10 + 2q 13 t10 + q 14 t10 + 3q 15 t10 − 3q 16 t10 − 9q 17 t10 − 7q 18 t10 + 9q 19 t10 + 3q 20 t10 + q 12 t11 +
2q 14 t11 + q 15 t11 + 2q 16 t11 − 6q 17 t11 − 9q 18 t11 + 3q 19 t11 + 6q 20 t11 + q 13 t12 + 2q 15 t12 + q 16 t12 −
10q 18 t12 + 5q 20 t12 + q 21 t12 + q 14 t13 + 2q 16 t13 − 3q 18 t13 − 7q 19 t13 + 6q 20 t13 + q 21 t13 + q 15 t14 +
2q 17 t14 − 2q 18 t14 − 5q 19 t14 + 2q 20 t14 + 2q 21 t14 + q 16 t15 + q 18 t15 − 4q 19 t15 + 2q 21 t15 + q 17 t16 −
q 19 t16 − 2q 20 t16 + 2q 21 t16 + q 18 t17 − q 19 t17 − q 20 t17 + q 21 t17 + q 19 t18 − 2q 20 t18 + q 21 t18

+a q + q 2 + q 3 + q 4 + q 5 − 2qt − q 2 t + q 4 t + 2q 5 t + 5q 6 t + 4q 7 t + 2q 8 t + q 9 t + qt2 − q 2 t2 −
2q t − 3q 4 t2 − 3q 5 t2 − 4q 6 t2 + q 7 t2 + 6q 8 t2 + 7q 9 t2 + 5q 10 t2 + 2q 11 t2 + q 2 t3 − 2q 5 t3 − 3q 6 t3 −
3 2
7q 7 t3 − 7q 8 t3 − 3q 9 t3 + 6q 10 t3 + 9q 11 t3 + 6q 12 t3 + 2q 13 t3 + q 3 t4 + q 5 t4 − q 7 t4 − 5q 8 t4 −
7q 9 t4 − 11q 10 t4 − 4q 11 t4 + 10q 12 t4 + 11q 13 t4 + 4q 14 t4 + q 15 t4 + q 4 t5 + q 6 t5 + q 7 t5 + q 8 t5 −
3q 9 t5 − 4q 10 t5 − 10q 11 t5 − 13q 12 t5 + 5q 13 t5 + 13q 14 t5 + 7q 15 t5 + q 16 t5 + q 5 t6 + q 7 t6 + q 8 t6 +
2q 9 t6 − q 10 t6 − 2q 11 t6 − 8q 12 t6 − 16q 13 t6 + 2q 14 t6 + 12q 15 t6 + 7q 16 t6 + q 17 t6 + q 6 t7 + q 8 t7 +
q 9 t7 + 2q 10 t7 − 8q 13 t7 − 16q 14 t7 + 2q 15 t7 + 13q 16 t7 + 4q 17 t7 + q 7 t8 + q 9 t8 + q 10 t8 + 2q 11 t8 +
q 13 t8 − 8q 14 t8 − 16q 15 t8 + 5q 16 t8 + 11q 17 t8 + 2q 18 t8 + q 8 t9 + q 10 t9 + q 11 t9 + 2q 12 t9 − 8q 15 t9 −
13q 16 t9 + 10q 17 t9 + 6q 18 t9 + q 9 t10 + q 11 t10 + q 12 t10 + 2q 13 t10 − 2q 15 t10 − 10q 16 t10 − 4q 17 t10 +
9q 18 t10 + 2q 19 t10 + q 10 t11 + q 12 t11 + q 13 t11 + 2q 14 t11 − q 15 t11 − 4q 16 t11 − 11q 17 t11 + 6q 18 t11 +
5q 19 t11 + q 11 t12 + q 13 t12 + q 14 t12 + 2q 15 t12 − 3q 16 t12 − 7q 17 t12 − 3q 18 t12 + 7q 19 t12 + q 20 t12 +
q 12 t13 + q 14 t13 + q 15 t13 + q 16 t13 − 5q 17 t13 − 7q 18 t13 + 6q 19 t13 + 2q 20 t13 + q 13 t14 + q 15 t14 +
q 16 t14 − q 17 t14 − 7q 18 t14 + q 19 t14 + 4q 20 t14 + q 14 t15 + q 16 t15 − 3q 18 t15 − 4q 19 t15 + 5q 20 t15 +
q 15 t16 + q 17 t16 − 2q 18 t16 − 3q 19 t16 + 2q 20 t16 + q 21 t16 + q 16 t17 − 3q 19 t17 + q 20 t17 + q 21 t17 +

q 17 t18 − 2q 19 t18 + q 21 t18 + q 18 t19 − q 19 t19 − q 20 t19 + q 21 t19 + q 19 t20 − 2q 20 t20 + q 21 t20 .
The a–degree of H min (q, t, a) is 5, coinciding with that given by the formula
L
dega = s(3|λ|) − |λ| = 5 from (5.41). The self-duality and other claims in this
theorem hold. We note that the positivity claim from Part (ii) of the Connection
Conjecture 5.3 really requires here κ − 1 = 2. Generally (5.56) reads:
min
H L, L (q, t, a)/(1 − t)
κ−1
∈ Z+ [[q, t, a]].
Taking here (1 − t) is insufficient for the positivity; (1 − t)2 is necessary.
3-fold T(2,1).
◦→ →
(6.19) → , (,,)
T (9, 6) : L = L({2,1},{2,1},{2,1}) min
, H L (q, t, a) =
1 − 2t + qt + q 2 t + t2 − 2qt2 − q 2 t2 + q 3 t2 + q 4 t2 + qt3 − q 2 t3 − q 3 t3 + q 4 t3 + q 2 t4 − q 3 t4 −

q 4 t4 + q 5 t4 + q 3 t5 − 2q 4 t5 + q 5 t5 + q 4 t6 − 2q 5 t6 + q 6 t6 + a2 q 3 − 2q 3 t + q 4 t + q 5 t + q 3 t2 −

2q 4 t2 + q 5 t2 + q 4 t3 − 2q 5 t3 + q 6 t3 + a q + q 2 − 2qt − q 2 t + 2q 3 t + q 4 t + qt2 − q 2 t2 − 3q 3 t2 +

2q 4 t2 + q 5 t2 + q 2 t3 − 3q 4 t3 + 2q 5 t3 + q 3 t4 − q 4 t4 − q 5 t4 + q 6 t4 + q 4 t5 − 2q 5 t5 + q 6 t5 .
Here again the division by (1 − t) is insufficient for the positivity.
7. Hopf links
7.1. Basic constructions. We will provide basic superpolynomials for the
−1
Hopf links. They correspond to taking τ− inside the coinvariant, i.e. this is the
case of multiple T (1, −1). These examples are directly related to the DAHA-vertex
Cbc /θμ = Pb /Pc◦ , Pc◦ from Corollary 5.2. Namely,
ι(c)
∅
H min (λ)
(λ1 ∨. . .∨λκ )†ev H
∅,† = {1,−1} (λ) {1,−1}
(7.1) H =
(λ1 )†ev · · · (λκ )†ev (λ1 )†ev · · · (λκ )†ev
κ
• •
is the stabilization of Pλ = P(λ
λ j
1,...,λκ−1 ) (up to q t ) for λ = (λ ).
The following identity from (3.38) is the key for this connection:
→
◦→
H → ,min min 1 2 3
{1,−1} (λ , λ , λ ) = HL, L (λ , λ ; λ ) for L = ◦ ⇒, L = ◦ →,
1 2 3
where the trees L, L are colored by {λ1 , λ2 } and {λ3 } correspondingly.

Upon adding ∨ to L the last formula (which is the switch from Y −1 to Y in

L), the super-polynomials
min
H 3 ∨ def ⇒,→,min
{1,−1},{1,0} (λ , λ ; λ ) == HL, L∨
1 2
(7.2)
3
provide the a–stabilizations of Cλλ1 ,λ2 Pλ3 , Pλ3 /θμ , which are of obvious interest
since they contain no ι and therefore satisfy straight associativity (with certain
multiplies) from (3.32). We will always take λ3 = in the examples below.

Since the recalculation of the H–polynomials to Pλ and the C–coefficients is
straightforward, we will provide only H. Recall that Proposition 3.4 and (5.42) are
used for this, where
p −1

n λ
(7.3) Π†λ = (λ)†ev = 1 + q v a t−p+1 .
p=1 v=0
jo
Also, den will be provided, which are needed in (6.2):
L
H min
(q, q, −a)/denjo = HOM
j
4 Lo (q, a).
7.2. Colored Hopf 3-links. The first example will be uncolored; it is a direct
continuation of the previous section, namely the case of T (3, −3) in the notations
there. It is self-dual with respect to the super-duality.
◦→ →
(7.4) , , → ,( , , )
: L = L({1,−1},{1,−1},{1,−1}) min
, H L (q, t, a) =
2 2 2 2 3 2 3
1+a q+q 2 + t12 − 2q
t2
+ qt2 + 1t + qt − 2qt +a2 q 3 + t13 − 2q
t3
+ qt3 + tq2 − 2q
t2
+ qt2 + qt + qt − 2qt .
Let us provide its ∨–variant:
(7.5) , ; ∨ ◦⇒,( , ) ◦→,
: L = L({1,−1},{1,−1}) , L = L{1,0} min
, H L, L∨ =
q2

2 − q + a2 q 2 + q
t
− t
− 2t + 2qt + a 3q − q 2 + 1
t
− q
t
− t + q2 t .
We will provide the formulas with ∨ for all further examples:

◦→ →
(7.6) , , : → ,( , , )
L = L({1,−1},{1,−1},{1,−1}) min
, H L (q, t, a) =
q2 q3 2q 3
q2 q3 2q 3
1 + a q2 + q3 + 1
t2
− tq2 − t2
+ t2
+ 1
+ q
− + a2 q 5 + 1
t3
− q
t3
− t3
+ t3
+ q
t2
− t2
+
5
t t t
5 2 3
q
t2
+ q
t
+ q
t
− 2q
t
.
(7.7) , ; ∨ ◦⇒,( , ) ◦→,

: L = L({1,−1},{1,−1}) , L = L{1,0} min
, H L, L∨ =
q2 q4
q2

1 + q − q 2 + a2 q 4 + t
− t
− t − qt + 2q 2 t + a −1 + q + 3q 2 − q 4 + 1
t
− t
− qt + q 4 t .
◦→ →
(7.8) , , : → ,( , , )
L = L({1,−1},{1,−1},{1,−1}) min
, H L (q, t, a) =
q2 q2 q2
q2 2q 2 q3
1 + a q + q2 + 1
t3
− 2q
t3
+ t3
+ 1
t2
− t2
+ q
− + a2 q 3 + 1
t5
− 2q
t5
+ t5
+ q
t3
− t3
+ t3
+
3
t t
q3 q2
q
t2
− t2
+ t
− q
t
.
The last example is dual to (7.6) under t ↔ q −1 . Usually the super-duality is
up to certain q • t• , but for the Hopf link without ∨ , such multipliers are not needed
since the a–constant term is always 1. The constant term is not 1 with ∨. We omit
the corresponding ∨–variant for (7.8), since it is super-dual to (7.7). The next case
will be self-dual, since the diagram is transposition-symmetric.
◦→ →
(7.9) , , : → ,( , , )
L = L({1,−1},{1,−1},{1,−1}) min
, H L (q, t, a) =
2 2 3 2 3
1 + a q2 + q3 + 1
t3
− 2q
t3
+ qt3 + 1
t2
+ q
t2
− 3q
t2
+ qt2 + qt + qt − 2qt + a2 q 5 + t15 − 2q
t5
+
q2 3q 2 3 4
q2 3 4 2 3 4
q5 q3 q4 2q 5

t5
+ q
t4
− t4
+ 3q
t4
− qt4 + tq3 + t3
− 5q
t3
+ t3 + t2 + t2 − t2 + t2 + t + t − t .
3q q q 3q
(7.10) , ; ∨ ◦⇒,( , ) ◦→,

: L = L({1,−1},{1,−1}) , L = L{1,0} min
, H L, L∨ =
q2 q3 q3 4
1 + q − q 2 + a2 q 4 + t2
− t2
+ − qt
− t + q 2 t − qt2 + q 2 t2 + a −q + 5q 2 − 2q 3 + 1
t2
−
2
t

q
t2
− 1
t
+ 3q
t
− 2q
t
− qt − q t + 3q t − q t − q 3 t2 + q 4 t2 .
2 3 4
Note that H ∅,† = H min /(1 + a)2 and den1 = (1 − q)2 in all these examples,
since λ ∨ λ ∨ λ is λ (it contains the remaining two). In the next two examples,
1 2 3 1
den1 = (1 − q)2 (1 − q 2 ).
◦→ →
(7.11) , , : → ,( , , )
L = L({1,−1},{1,−1},{1,−1}) min
, H L (q, t, a) =
2 4 2 3 4
1 + a q2 + q3 + q4 + 1
t2
− 2q t2
+ qt2 + 1t + qt + qt − qt − 2qt + a2 q 5 + q 6 + q 7 + t13 +
3q 2 2q 3 5 3q 46 2 4 5 6 7 2 3 4 6
q
t3
− t3 − t3 + t3 + qt3 − qt3 + tq2 + 2q t2
− 5q
t2
− 2qt2
+ 3q
t2
+ qt2 + qt + 2qt + 2qt − 3qt −
2q 7
2 3 4 5 6 2 3 4 5 6 7 8 3
+ a3 q 9 + tq4 − qt4 − 2q t4
+ 2q t4
+ qt4 − qt4 + qt3 + qt3 − 3q t3
− 2qt3
+ 3q
t3
+ qt3 − qt3 + qt2 +
9
t
4 6 7 8 9 5 6 7 8
q
t2
− 3q
t2
− 2q
t2
+ 2q
t2
+ qt2 + qt + qt + qt − qt − 2qt .
(7.12) , ; ∨ ◦⇒,( ) ◦→,

: L = L({1,−1},{1,−1}) , L = L{1,0}
, min
, H L, L∨ =
q3 4 5 6 4 5 6 7
2 − q 2 + a3 q 7 + t2
− qt2 − qt2 + qt2 + qt + qt − qt − qt − 2t + 2q 2 t + a −2 + 7q 2 + 2q 3 −
2 3 4
3q 4 − q 5 + 2t + qt − 3qt − qt + qt − qt − 2q 2 t + 2q 4 t + q 5 t + a2 −q − q 2 + 2q 3 + 3q 4 + 2q 5 −
2 3 4 2 3 4 5 6
q −q +
6 7 q
t2
− qt2 − qt2 + qt2 + 3qt + 2qt − 4qt − 2qt + qt − q 3 t + q 7 t .
The next 2 cases are dual to the previous ones. We provide them, but then will
omit the examples dual to those already given. Note that
min min
∅,† ( , , ) = HL ( , , ) , H
H ∅,† ( , , ) = HL ( , , ) .
L 2 L
(1 + a) (1 + qa) (1 + a)2 (1 + a/t)
◦→ →
(7.13) , , : → ,( , , )
L = L({1,−1},{1,−1},{1,−1}) min
, H L (q, t, a) =
2 2 2
1 + a q + q2 + 1
t4
− 2q
t4
+ qt4 + t13 − tq3 + t12 + tq2 − 2q t2
+ qt + a2 q 3 + t17 − 2q t7
+ qt7 + t16 −
2 3 2 3 2 3 3 2 3 2 3
3q
t6
+ 3q
t6
− qt6 + t15 − 2q t5
+ qt5 + 2q t4
− 5q
t4
+ 3qt4
+ 2q
t3
− 2q
t3
+ tq2 + 2qt2
− 3q
t2
+ qt + qt +
2 2 3 2 3 2 3 4 3 4
a3 t19 − 2q t9
+ qt9 − tq8 + 2q t8
− qt8 + tq7 − 2qt7
+ qt7 + tq6 − 3qt6
+ 3qt6
− qt6 + tq5 − 2qt5
+ qt5 +
2 3 4 2 3 4 3 4 4
q
t4
− 3q
t4
+ 2q
t4
+ qt3 + qt3 − 2q t3
+ qt2 − qt2 + qt .
(7.14) , ; ∨ ◦⇒,( , ) ◦→,

: L = L({1,−1},{1,−1}) , L = L{1,0} min
, H L, L∨ =
q2 q2
3 2 3 2 3 3
2 − q + a3 q
t5
−
t4 t5
−
+ qt4 + qt3 − qt3 + qt2 − qt2 + qt − 2t2 + 2qt2 + a2 −q + 3q 2 − q 3 +
2 2 2 3 2 3
1
t5
− q
t5
− tq4 + qt4 + 2q
t3
− 2q
t3
+ 3qt2
− 4q
t2
+ qt2 − 1t + 2q + 2qt − qt − qt + q 3 t + a −2 +
2 2
t

7q − 3q 2 + t13 − tq3 + t22 − 3q
t2
+ qt2 + 2q
t
− qt − t + q 2 t − 2qt2 + 2q 2 t2 .
◦→ →
(7.15) , , : → ,( , , )
L = L({1,−1},{1,−1},{1,−1}) min
, H L (q, t, a) =
2 3
q3 q3
q2 q3
1 + a q2 + q3 + 1
t3
− tq3 − qt3 + qt3 + 1
t2
− t2
+ q
− + a2 q 5 + 1
t5
− q
t5
− t5
+ t5
+ q
t3
−
2q 3 q5 q2 5 3 5 t t
t3
+ t3
+ t2
− q
t2
+ qt − qt .
(7.16) , ; ∨ ◦⇒,( , ) ◦→,

: L = L({1,−1},{1,−1}) , L = L{1,0} min
, H L, L∨ =
q2 4 q2
1 + q − q 2 + a2 q 4 + t2
− qt2 − t + q 2 t − qt2 + q 2 t2 + a 2q 2 + q 3 − q 4 + 1
t2
− t2
− 1
+ q
+
q2 q3
t t
t
− t
− qt + q t − q t + q t .
3 3 2 4 2
These 2 examples are self-dual. The degree here is |s|(|λ1 | + |λ2 | + |λ3 |) − |λ1 ∨
∅,† ( , , ) =
λ ∨ λ3 | = 2 + 2 + 1 − 3 = 2, since λ1 ∨ λ2 ∨ λ3 is the 3-hook. Also: H
2
L
min ( , , )/(1 + a)2 , den1 =
2 2
(1−q) (1−q )
H L (1−a/q) .
◦→ →
(7.17) , , : → ,( , , )
L = L({1,−1},{1,−1},{1,−1}) min
, H L (q, t, a) =
2 2 3 2 3 2 3
1+a q 2 +q 3 + t14 − 2q
t4
+ qt4 + t13 − 2q
t3
+ qt3 + t12 + tq2 − qt2 − qt2 + qt + qt − qt +a2 q 5 + t17 − 2q
t7
+
q2 3 4 2 3 2 3 4 2 3 4 5
t7
+ t16 − 2q
t6
+ 2qt6
− qt6 + t15 + tq5 − 4qt5
+ 2qt5
+ 2q
t4
− qt4 − 5q t4
+ 4qt4
+ tq3 + 3qt3
− 3q t3
− 2q
t3
+ qt3 +
2 3 4 5 3 4 5 2 2 3 4 2 3 3
q
t2
+ 3q
t2
− 3q
t2
− qt2 + qt + 2qt − qt +a3 t19 − 2q t9
+ qt9 − qt8 + 2qt8
− qt8 + tq7 − 2q t7
+ qt7 + tq6 − 4q t6
+
4q 4 q5 2q 2 2q 3 2q 4 3q 5 q6 2q 3 3q 4 q5 2q 6 q3 2q 4 3q 5 2q 5 2q 6 q6

t6
− t6
+ t5
− t5
− t5
+ t5
− t5
+ t4
− t4
− t4
+ t4
+ t3
+ t3
− t3
+ t2
− t2
+ t
.
(7.18) , ; ∨ ◦⇒,( , ) ◦→,

: L = L({1,−1},{1,−1}) , L = L{1,0} min
, H L, L∨ =
2 3 4 5 3 5 4 5 5
1 + q − q 2 + a3 qt5 − qt5 − qt4 + qt4 + qt3 − qt3 + qt2 − qt2 + qt − t + qt − 2qt2 + 2q 2 t2 +
2 3 2 3 4 2 3
a2 −q 2 + 3q 4 − q 5 + t15 − tq5 − t14 + 2q
t4
− 2q t4
+ qt4 + tq3 + 2q
t3
− 4q
t3
+ qt3 − tq2 + 4q
t2
− qt2 −
3q 4 5 3 4 5 2
t2
+ qt2 − qt + 5qt − qt − qt − q 3 t + q 5 t + a −3q + 6q 2 + q 3 − 2q 4 + t13 − tq3 + 3q t2
− 4q
t2
+
3 2 3 4
q
t2
− t + t + t − t + t − qt − 2q t + 4q t − q t − 2q t + 2q t .
1 2q 3q 4q q 2 3 4 3 2 4 2
The a–degree is |s|(|λ1 | + |λ2 | + |λ3 |) − |λ1 ∨ λ2 ∨ λ3 | = 3 + 2 + 1 − 3 = 3.

Correspondingly, H ∅,† ( , , ) = H min ( , , )/ (1 + a)2 (1 + aq) , den1 =
L L
(1 − q)2 (1 − q 2 ).
7.3. Hopf 2-links. The previous examples were for the Hopf 3–links. Let us
discuss the Hopf 2–links.

(7.19) ,
◦⇒, ( , ) min
: L = L({1,−1},{1,−1}) , H L (q, t, a) = 1 + a q 2 + 1
− q2
.
t t

(7.20) , min
: H ◦⇒, ( , ; q, t, a) = 1 + a q 2 + 1
− q2
,
t2 t2

(7.21) , min
: H ◦⇒ ( , ; q, t, a) = 1+a q 2 + t12 − tq2 + qt − qt .
2
The last two examples are self-dual, as well as the simplest superpolynomial
H min ( , ) = 1 + a(q + 1/t − q/t), which is actually uncolored T (2, −2), already
◦⇒,
considered in the previous section. The next 2 ones are not self-dual:
(7.22) , :
◦⇒, ( , ) min
L = L({1,−1},{1,−1}) , H L (q, t, a) =
q2 q3
q2 q3 q4 q2 q3 q4 q5

1 + a q2 + q3 + 1
t
+ q
t
− t
− t
+ a2 q 5 + q
t2
− t2
− t2
+ t2
+ t
+ t
− t
− t
.
(7.23) , :
◦⇒, ( , ) min
L = L({1,−1},{1,−1}) , H L (q, t, a) =
q2 q3
q2 q3 q4 q2 q4 q3 q5

1 + a q2 + q3 + 1
t2
− t2
+ q
t
− t
+ a2 q 5 + q
t3
− t3
− t3
+ t3
+ t2
− t2
+ t
− t
.

Here dega = 2 and H ∅,† ( , ) = H
min ( , )/ (1 + a)(1 + aq) . The same
L L
denominator serves H ∅,† ( , ). Otherwise, dega = 1 in the examples above and
L
this denominator is (1 + a). Let us provide den1 from (6.3), needed for obtaining
the reduced hat-normalized HOMFLY-PT polynomials. One has den1 ( , ) =
(1−q)(1−q 2 )
(1−a/q) ; otherwise:
(7.24) den1 (λ1 , ) = (1−q), den1 (λ1 , ) = (1−q)(1−q 2 ) = den1 (λ1 , ),

since λ1 contains λ2 in (7.24).
min )∞ be the leading a–coefficient
Connection to the topological vertex. Let (H

min 1 2 ∞◦⇒

of H
◦⇒ and CDAHA (λ , λ ) = H ◦⇒ (λ , λ ) (t → q), which is proportional to
min num 1 2
ι(λ2 )
Cλ01 ,λ2 and C λ1 Pλ◦2 , Pλ◦2 in this limit. Let us provide CDAH
num
A in the following 2
examples.
(7.25) , min = 1 + aq 2 +

: H 1
− q2

; den1 = (1−q)(1−q 2 )
.
t3 t3 1−aq

1 − a( q13 − 1
+ q 2 ) (1 − aq)
=
HOM
q
, CDAHA = 1 − q + q ,
num 2 5
(1 − q)(1 − q 2 )
(7.26) , min
: H ◦⇒ ( , ; q, t, a) =
2
2 2 2
, ; q, q, −a) =
1+a q+ t14 − tq4 + t13 − tq3 + qt +a2 t17 − tq7 − tq6 + qt6 + tq4 − qt4 + tq3 − qt3 + qt , H(

1− a 1+ q14 − q12 + q + a2 q17 − q16 − q15 + q14 + q13 − 1q + q , den1 = (1 − q)(1 − q 2 ), CDA num
HA =
( , ; q, a) = 1 − q − q 2 + q 3 + q 4 − q 6 + q 8 .
num 1 2 •
Vertex amplitudes. We claim that CDAH A (λ , λ ) coincide up to q with the
1 2 tr
numerators of CVA (λ , (λ ) ), from Section “The vertex amplitudes” of [AKMV].
We denote them by CVA to avoid confusion with our own C. Note the transposition
here; by the way, if tr is applied to λ1 instead of λ2 , then q must be changed to q −1 .
This coincidence includes λ2 = ∅, when the corresponding weight is b = 0 (denoted
by dot in [AKMV]); they numerators are pure powers q • in this case.
This claim can be deduced from (3.26) and we checked it numerically in all
examples we considered. The denominators of CVA (λ1 , (λ2 )tr ) are the products of
binomials in the form (1 − q • ) and are actually the matter of normalization of the
2–vertex. Such vertex amplitudes are connected with the stable Macdonald theory,
which generally corresponds to taking a = 0 in our approach. Here we take the top
power of a instead of a = 0, since the conjugation is actually needed to connect our
C with CVA . See e.g., [AFS].
However, the connection of our C and CVA fails for the 3–vertex, with ∨ or
without (i.e. with ι in C or without). Our formulas for (the numerators of)
H min (λ1 , λ2 )top (t → q) seem different from the corresponding ones in [AKMV]
◦⇒
for 3–links. It is possible that we have two different theories (based on different
kinds of Mehta-Macdonald identities). Recall that our approach is for arbitrary
root systems ι = −w0 naturally emerges. It is in contrast to the theory from
[AKMV, AFS], including the K–theoretical Nekrasov partition functions [Nek];
it heavily involves the transposition { }tr of diagrams, which certainly requires sta-
ble An . Thus at this moment, such a coincidence of two theories seems a special
feature of the 2–vertex.
Actually the Macdonald polynomials are needed for the refined topological
(physics) vertex only a little; the skew Schur functions are the main ingredient.
See e.g., [AFS]; the appearance of stable Macdonald polynomials in the Nekrasov
instanton sums from formulas (5.1-3) there is via the evaluation of these polynomials
at tρ , which is a very explicit (and simple) product. Replacing the skew Schur
function by the corresponding Macdonald polynomials does not actually influence
the partition sums.
The Hopf links naturally emerge and play an important role in [AKMV],
[GIKV] and other papers on topological vertex. However we do not quite un-
derstand the connection of the superpolynomials for Hopf links with the (refined)
topological vertex considered in [GIKV]; these directions look like two parallel but
different theories to us (though both are from the same physics source). In DAHA
theory, these two are really closely related, the DAHA-vertex is part of the DAHA-
superpolynomials for links. However our approach seems (so far) different from the
physics theory of the topological vertex. Mathematically, the DAHA-vertex seems
almost inevitable, to combine the topological vertex and superpolynomials in one
theory (what we do).
Khovanov polynomials for Hopf links. Let us comment a bit on Section 3.1 from
[GIKV] on the Hopf 2–link for the fundamental representation, denoted by L2a1 in
[KA]. We will use (5.59). Recall that the standard parameters there were modified
vs. those in (5.53) as follows:
(7.27) t → q 2 , q → (qt)2 , a → −a2 (a → −q 2(n+1) for An ).
Only the substitution for a is different here; it is exactly the one from the HOMFLY-
PT reduction vs. a → a2 t in the Connection Conjecture. Such a modification of
a is very natural, since it automatically provides the divisibility by (1 − t)κ (see
below). It allows to deal directly with H min , which is a polynomial in terms of
a, q, t±1 .
Let us switch in what will follow to new q, t, a from (7.27). Then the uncolored
superpolynomial suggested in [GIKV] is essentially obtained from our one by the
conjugation q, t, a → q −1 , t−1 , a−1 up to some simple factor. Upon the substitution
a = q 2 , it results in the Khovanov polynomial of this link (L2a1), which is (1 +
q 2 )(1 + q 4 t2 ) = 1 + q 2 + q 4 t2 + q 6 t2 up to q • t• –proportionality.
More systematically, in the new parameters above, our superpolynomial reads:

HL2a1 = 1 − a2 (1/q 2 − t2 + q 2 t2 ). We obtain, which matches (41) in [GIKV], that:
min
1−a2
(7.28) L2a1
H min 4 N (q, t) for N = 2, 3, 4, 5.
= q • t• KhR L2a1
(1−q 2 )2 a →q N
Let us provide the left-hand side for N = 5: 1 + 2q 2 + 3q 4 + 4q 6 + 4q 8 + 3q 10 + 2q 12 +

q 14 + q 10 t2 + q 12 t2 + q 14 t2 + q 16 t2 + q 18 t2 .
Let us apply the same procedure to the Hopf 3–link, denoted by L6n1 =
Link[ 6, N onAlternating, 1 ] in [KA]; we use the procedure KhReduced[·]. We
generally expect that

1 − a2
L6n1 . 4 N (q, t) for N * 0.
(7.29) H min
= q • t• KhR
(1 − q 2 )3 a →q N
L6n1
For N = 2, we obtain the following :

1 − a2 min
H (a = q 2 ) =(2 + 3q 2 + q 4 − q 2 t2 + q 6 t2 + q 8 t4 + q 10 t4 )q 2 t2 ,
(1 − q 2 )3 L6n1
4 2 (q, t) =(2 + 3q 2 + q 4 + q 2 t + q 6 t2 + q 8 t4 + q 10 t4 )/q.
KhR L6n1
The change −q t → q 2 t (they are outlined) is a typical correction due to the

2 2
nontrivial differential ∂2 . Note that the t–degree in this correction diminishes (by
1); it always increases (here by 1) in such corrections in the examples of algebraic
links we considered.
We see that it is not impossible to obtain the reduced Khovanov polynomials
4 2 and other Khovanov- Rozansky polynomials directly from H
KhR min , i.e. without
the division by (1 − t) κ−1
. However we do not know how far this and similar
procedures can go for links. In the examples we provide, managing the negative
terms in H min is always by −q l tm → q l tm±1 , q l tm → q l tm±2 and so on. This
is always (in examples) in the same direction for a given link, which matches the
discussion of the usage of differentials after the Connection Conjecture on [ChD].
Employing the differentials provides sharper match here, but, as we have seen, even
a “recovery” of reduced KhR 4 n+1 from a single H min for the corresponding An is
not impossible (for small links).
7.4. Algebraic Hopf links. The matrix τ− , which played the key role above,
will be replaced by τ+ in this section. Recall that according to Part (ii) of the
Connection Conjecture 5.3, we expect the properties of the superpolynomials for
positive pairs of trees {L, L∨ } (with ∨ and satisfying the inequalities for r, s in
Part (ii)) to be the same as for a single positive tree L. Recall that the positivity
of {L, L∨ } is sufficient for the corresponding link to be algebraic. It is necessary if
the trees L, L are reduced (minimal modulo the equivalence); see [EN].
Before considering the algebraic links, let us provide the simplest non-algebraic
twisted union , which is that for two disjoint (unlinked) unknots. I.e. Υ is ⇒ or,
equivalently, ◦ ⇒ with the vertex labeled by {1, 0}. Such a union is the result of
adding the meridian having the linking numbers +1 with both unknots. This is the
3-chain, which is obviously non-algebraic. Its superpolynomial is as follows:
(7.30)
◦⇒, ( , ) ◦→,
3-chain : L = L({1,0},{1,0}) , L = L{1,0} min
, H L, L∨ (q, t, a) =

1 + a2 q 2 − 2t + 2qt + t2 − 2qt2 + q 2 t2 + a 2q − 2qt + 2q 2 t = (1 + aq − t + qt)2 .
The square here is not surprising, since the 3-chain is the connected sum of two
(positive) Hopf 2–links, with the DAHA-superpolynomials (1 + aq − t + qt). Such
a product formula holds if the meridian, which is a colored unknot, is added to an
arbitrary disjoint union of links. Let us justify
this.
−1 −1
Recall the general formula {P, Q}ev = {τ− τ̇− (P )τ̇− (Q) }ev from Theorem
3.9 for P, Q ∈ V W . Let Q = P tot and P = P tot be from (4.22) for any graphs
0 0
L, L. Due to the symmetry (4.5), the components of the link for {L, L∨ } is a

certain union of those for L and L.

Proposition 7.1. Let Q = P L,tot , R = P

M,tot for arbitrary graphs L, M and
0 0
N be the disjoint union of L, M. We take → as L; equivalently, L = ◦→ where
◦ has the label [1, 0]; it will be colored by any Young diagram λ. The corresponding
H{N , L∨ } will be normalized for λ, i.e. for jo being the index of L (we put below λ
instead). So this is not the standard min–normalization. Then
(7.31) λ ∨ = H
H λ ∨ Hλ
{N , L } {L, L } {M, L∨ } .
λ ∨ = {QR, ι(P ◦ )}ev = QR(q λ+kρ ) = H

Proof. H λ ∨ Hλ
{N , L } {L, L } {M, L∨ } .
λ

At the level of HOMFLY-PT polynomials (upon a → −a, t → q), relation (7.31)
becomes a well-known property of the connected sum of L, M,
where by tilde, we

mean the twist with unknot. This sum is N and (7.31) holds. We cannot comment
on the validity of such a product formula in Khovanov-Rozansky theory.
Unusual positivity factors. Let us demonstrate that the division by powers (1−t)•
can be insufficient to ensure the positivity in Part (ii) for colored algebraic links.
We begin with
◦→
→ ,( , , ) min
(7.32) , 2× : L = L{1,1}
→
, H L (q, t, a) =
1 + a3 q 9 − t − qt + q 2 t + q 3 t − t2 + qt2 − q 2 t2 − q 3 t2 + 2q 4 t2 + t3 + qt3 − q 2 t3 − 2q 3 t3 − q 4 t3 +
q 5 t3 +q 6 t3 −qt4 +q 2 t4 +2q 3 t4 −2q 4 t4 −q 5 t4 +q 6 t4 +q 3 t5 −2q 5 t5 +q 7 t5 −q 3 t6 +q 4 t6 +2q 5 t6 −

2q 6 t6 −q 7 t6 +q 8 t6 +a2 q 5 +q 6 +q 7 −q 5 t−q 6 t+q 7 t+q 8 t−q 5 t2 +q 9 t2 +q 5 t3 −2q 7 t3 +q 9 t3 +
2
a q +q 3 +q 4 −q 2 t−2q 3 t+2q 5 t+q 6 t−q 2 t2 −q 4 t2 −q 5 t2 +2q 6 t2 +q 7 t2 +q 2 t3 +2q 3 t3 −q 4 t3 −

4q 5 t3 − q 6 t3 + 2q 7 t3 + q 8 t3 − q 3 t4 + q 4 t4 + 2q 5 t4 − 2q 6 t4 − q 7 t4 + q 8 t4 + q 5 t5 − 2q 7 t5 + q 9 t5 .
Here dega = 3 = | | + 2| |−|
| as it is conjectured. This degree remains the
same for the triple ( , 2 × ); we will omit the formula for H min for the latter.
The corresponding series H min
/(1 − t) are non-positive for any p ≥ 0 (for both).
p
min /(1 − q)3 are positive and 3 is the smallest here (for both).
However, the series H
Thus (1 − t)• is sufficient to ensure the positivity for ( , 2× ), but any power of
(1−t) is insufficient for 2 × .
The next case demonstrates that (1−t)(1−q) may be really needed.
(7.33) , : L = L{1,1}
◦⇒, ( , ) min
, H L (q, t, a) =
3 6
1+ a tq −2t+2qt+t2 −4qt2 +2q 2 t2 +q 3 t2 +3qt3 −5q 2 t3 +2q 3 t3 −qt4 +3q 2 t4 −4q 3 t4 +2q 4 t4 +
3 4
q 3 t5 −2q 4 t5 +q 5 t5 +a2 −q 3 +q 4 +q 5 + qt + qt −2q 4 t+q 5 t+q 6 t−q 5 t2 +q 6 t2 +a −q +2q 2 +

q 3 + qt −4q 2 t+3q 3 t+q 4 t+q 2 t2 −5q 3 t2 +3q 4 t2 +q 5 t2 +q 2 t3 +q 3 t3 −4q 4 t3 +2q 5 t3 −q 5 t4 +q 6 t4 .
The a–degree remains 3 = | |+| |−| |. The corresponding series H min /(1−
t) is non-positive for any p ≥ 0. Since H
p min
is self-dual, the same non-positivity
min /((1−t)(1−q))2 is positive (as it was
holds for (1 − q)• . However, the series H
conjectured).
7.5. Twisted T (2κ, κ). Let us discuss now in greater detail the case of two
trees when L is uncolored ◦ → with trivial r = 1, s = 0. I.e. this will be the twisted
union with the uncolored unknot. In our notations, it is for L = L◦→, 1,0 . We will
add “prime” to the corresponding notation/name of the link without showing L
(with or without ∨). Only the multiples of T (2, 1) will be considered in this section.
T(4,2)-prime for 3-hook and 1-box.
(7.34) T (4, 2) : L = L2,1

◦⇒, ( , )
, L = L◦→, min
, H L, L (q, t, a) =
1,0
,

1 − t + qt − 2qt2 + 2q 2 t2 + qt3 − 2q 2 t3 + q 3 t3 − q 3 t4 + q 4 t4 + a3 q 4 − 3q 5 + 4q 6 − 3q 7 + q 8 −
q3 4 3 4 6 3 4 5 6 7 4 5 6
t4
+ qt4 + 2q
t3
− 3q
t3
+ qt3 − qt2 + 4q
t2
− 5q
t2
+ qt2 + qt2 − 3qt + 8qt − 5qt − q 6 t + 2q 7 t − q 8 t +
2 2 4 2 3 4
a2 q 2 − 7q 3 + 11q 4 − 3q 5 − 2q 6 + q 7 − tq4 + qt4 + 2q t3
− 3q t3
+ qt3 − tq2 + 5q
t2
− 4qt2
− 2q
t2
+
q5 q6 4q 2 10q 3 3q 4 4q 5 6
t2
+ t2
− + − − + qt + q 3 t − 7q 4 t + 10q 5 t − 4q 6 t + q 4 t2 − 4q 5 t2 + 5q 6 t2 −
t t t t
2 3
3q 7 t2 + q t − q t + 2q t
8 2 6 3 7 3
− q 8 t3 + a −3q + 6q 2 − q 4 + t12 − tq2 − 1t + 3q − qt − qt + qt −
t
6q 2 t + 6q 3 t − q 5 t + q 2 t2 − 6q t + 6q t − q t + q t − 3q t + 3q t − q t − q t + q t .
3 2 4 2 5 2 3 3 4 3 5 3 6 3 5 4 6 4
(7.35) T (4, 2),∨ : L = L2,1

◦⇒, ( , )
, L = L◦→, min
, H L, L∨ (q, t, a) =
1,0
,
1−2t+qt+q 2 t+t2 −3qt2 −q 2 t2 +2q 3 t2 +q 4 t2 +3qt3 −2q 2 t3 −4q 3 t3 +2q 4 t3 +q 5 t3 −qt4 +3q 2 t4 +
q 3 t4 −7q 4 t4 +2q 5 t4 +2q 6 t4 −q 2 t5 +2q 3 t5 +3q 4 t5 −7q 5 t5 +2q 6 t5 +q 7 t5 −q 3 t6 +2q 4 t6 +3q 5 t6 −
7q 6 t6 + 2q 7 t6 + q 8 t6 − q 4 t7 + 2q 5 t7 + q 6 t7 − 4q 7 t7 + 2q 8 t7 − q 5 t8 + 3q 6 t8 − 2q 7 t8 − q 8 t8 + q 9 t8 −

q 6 t9 +3q 7 t9 −3q 8 t9 +q 9 t9 +q 8 t10 −2q 9 t10 +q 10 t10 +a2 q 5 −2q 5 t+q 6 t+q 7 t+q 5 t2 −3q 6 t2 +q 7 t2 +

q 8 t2 + 3q 6 t3 − 5q 7 t3 + q 8 t3 + q 9 t3 − q 6 t4 + 3q 7 t4 − 3q 8 t4 + q 9 t4 + q 8 t5 − 2q 9 t5 + q 10 t5 + a q 2 +
q 3 −2q 2 t−q 3 t+2q 4 t+q 5 t+q 2 t2 −2q 3 t2 −4q 4 t2 +3q 5 t2 +2q 6 t2 +3q 3 t3 +q 4 t3 −9q 5 t3 +3q 6 t3 +
2q 7 t3 −q 3 t4 +2q 4 t4 +4q 5 t4 −10q 6 t4 +3q 7 t4 +2q 8 t4 −q 4 t5 +2q 5 t5 +4q 6 t5 −9q 7 t5 +3q 8 t5 +q 9 t5 −

q 5 t6 +2q 6 t6 +q 7 t6 −4q 8 t6 +2q 9 t6 −q 6 t7 +3q 7 t7 −2q 8 t7 −q 9 t7 +q 10 t7 +q 8 t8 −2q 9 t8 +q 10 t8 .
The a–degree in the first formula is 3, i.e. greater than that in the second (with
∨): dega = s(| | + | |) + max{ s, 1}| | − | | = 2. This is an important confir-
mation of the role of ∨ in our theorems/conjectures. With ∨ here, the positivity
min
of H L, L∨ (q, t, a)/(1 − t) holds, but it fails for (1 − t) . The term (1 − t) agrees
3 2 3
with the informal observation above that the proper power of (1 − t) here can be

(1 − t)κ+ κ−1 times (1 − t), which is the extra correction factor for .
T(6,3)-prime for 3 boxes.
◦→
→ ,( ,
min
T (6, 3) , L = L◦→,
, )
(7.36) , , : L = L2,1→ 1,0 ,H L, L (q, t, a) =
6 7 6 7 8
1+a4 −3q 6 +6q 7 −3q 8 − 3q t2
+ 3q
t2
+ 6qt − 9qt + 3qt −2t+qt+q 2 t+t2 −2qt2 −q 2 t2 +q 3 t2 +
q 4 t2 +qt3 −q 2 t3 −q 3 t3 +q 4 t3 +q 2 t4 −q 3 t4 −q 4 t4 +q 5 t4 +q 3 t5 −2q 4 t5 +q 5 t5 +q 4 t6 −2q 5 t6 +q 6 t6 +
4 5 6 7 3 4 5 6 7 8
a3 −2q 3 −2q 4 +3q 5 +12q 6 −13q 7 +3q 8 − qt2 − 3q + 3q + qt2 + qt + 3qt + 5qt − 13q + 3qt + qt +
t2 t2 t
q t−q t−5q t+3q t+5q t−3q t+q t −q t −2q t +3q t −q t +q t −2q t +q t +
3 4 5 6 7 8 4 2 5 2 6 2 7 2 8 2 5 3 6 3 7 3
3 6 2 3 4 5 7
a2 −2q −q 2 +9q 4 +4q 5 −8q 6 +q 8 − qt2 + qt2 + qt + qt + 3qt − 2qt − 4qt + qt +qt−q 2 t−4q 3 t−
6q 4 t + 10q 5 t + 4q 6 t − 4q 7 t + q 2 t2 + q 3 t2 − 3q 4 t2 − 6q 5 t2 + 9q 6 t2 − 2q 7 t2 + q 3 t3 + q 4 t3 − 4q 5 t3 +

3q 7 t3 −q 8 t3 +q 4 t4 −q 5 t4 −q 6 t4 +q 7 t4 +q 5 t5 −2q 6 t5 +q 7 t5 +a −2+2q+2q 2 +3q 3 −q 4 −q 5 + 1t −
q3
+t−4qt−2q 2 t+7q 4 t−q 5 t−q 6 t+2qt2 −2q 2 t2 −3q 3 t2 −3q 4 t2 +7q 5 t2 −q 6 t2 +2q 2 t3 −q 3 t3 −
t
3q t +3q 6 t3 −q 7 t3 +2q 3 t4 −2q 4 t4 −2q 5 t4 +2q 6 t4 +2q 4 t5 −4q 5 t5 +2q 6 t5 +q 5 t6 −2q 6 t6 +q 7 t6 .
4 3
◦→
→ ,( ,
min
T (6, 3),∨, , L = L◦→,
, )
(7.37) , : L = L2,1→ 1,0 ,H L, L∨ (q, t, a) =
1 − 3t + qt + q 2 t + q 3 t + 3t2 − 3qt2 − 2q 2 t2 − 2q 3 t2 + 2q 4 t2 + q 5 t2 + q 6 t2 − t3 + 3qt3 + q 3 t3 −

5q 4 t3 − q 5 t3 + 2q 6 t3 + q 7 t3 − qt4 + 2q 2 t4 − q 3 t4 + 4q 4 t4 − 2q 5 t4 − 6q 6 t4 + 2q 7 t4 + 2q 8 t4 − q 2 t5 +
2q 3 t5 −2q 4 t5 +3q 5 t5 +q 6 t5 −6q 7 t5 +2q 8 t5 +q 9 t5 −q 3 t6 +2q 4 t6 −2q 5 t6 +3q 6 t6 +q 7 t6 −6q 8 t6 +
2q 9 t6 +q 10 t6 −q 4 t7 +2q 5 t7 −2q 6 t7 +3q 7 t7 −2q 8 t7 −q 9 t7 +q 10 t7 −q 5 t8 +2q 6 t8 −2q 7 t8 +4q 8 t8 −

5q 9 t8 +2q 10 t8 −q 6 t9 +2q 7 t9 −q 8 t9 +q 9 t9 −2q 10 t9 +q 11 t9 −q 7 t10 +2q 8 t10 −2q 10 t10 +q 11 t10 −

q 8 t11 +3q 9 t11 −3q 10 t11 +q 11 t11 −q 9 t12 +3q 10 t12 −3q 11 t12 +q 12 t12 +a3 q 6 −3q 6 t+q 7 t+q 8 t+
q 9 t+3q 6 t2 −3q 7 t2 −2q 8 t2 +q 9 t2 +q 10 t2 −q 6 t3 +3q 7 t3 −4q 9 t3 +q 10 t3 +q 11 t3 −q 7 t4 +2q 8 t4 −

2q 10 t4 +q 11 t4 −q 8 t5 +3q 9 t5 −3q 10 t5 +q 11 t5 −q 9 t6 +3q 10 t6 −3q 11 t6 +q 12 t6 +a2 q 3 +q 4 +q 5 −
3q 3 t−2q 4 t−q 5 t+3q 6 t+2q 7 t+q 8 t+3q 3 t2 −2q 5 t2 −7q 6 t2 +q 7 t2 +3q 8 t2 +2q 9 t2 −q 3 t3 +2q 4 t3 +
2q 5 t3 + 4q 6 t3 − 9q 7 t3 − 4q 8 t3 + 4q 9 t3 + 2q 10 t3 − q 4 t4 + q 5 t4 + 6q 7 t4 − 6q 8 t4 − 4q 9 t4 + 3q 10 t4 +
q 11 t4 −q 5 t5 +q 6 t5 +6q 8 t5 −9q 9 t5 +q 10 t5 +2q 11 t5 −q 6 t6 +q 7 t6 +4q 9 t6 −7q 10 t6 +3q 11 t6 −q 7 t7 +
q 8 t7 + 2q 9 t7 − 2q 10 t7 − q 11 t7 + q 12 t7 − q 8 t8 + 2q 9 t8 − 2q 11 t8 + q 12 t8 − q 9 t9 + 3q 10 t9 − 3q 11 t9 +

q 12 t9 + a q + q 2 + q 3 − 3qt − 2q 2 t − q 3 t + 3q 4 t + 2q 5 t + q 6 t + 3qt2 − 2q 3 t2 − 7q 4 t2 − 2q 5 t2 +
4q 6 t2 +3q 7 t2 +q 8 t2 −qt3 +2q 2 t3 +2q 3 t3 +4q 4 t3 −4q 5 t3 −10q 6 t3 +2q 7 t3 +4q 8 t3 +q 9 t3 −q 2 t4 +
q 3 t4 + 5q 5 t4 + 2q 6 t4 − 12q 7 t4 + 4q 9 t4 + q 10 t4 − q 3 t5 + q 4 t5 − q 5 t5 + 4q 6 t5 + 4q 7 t5 − 12q 8 t5 +
2q 9 t5 +3q 10 t5 −q 4 t6 +q 5 t6 −q 6 t6 +4q 7 t6 +2q 8 t6 −10q 9 t6 +4q 10 t6 +q 11 t6 −q 5 t7 +q 6 t7 −q 7 t7 +
5q 8 t7 − 4q 9 t7 − 2q 10 t7 + 2q 11 t7 − q 6 t8 + q 7 t8 + 4q 9 t8 − 7q 10 t8 + 3q 11 t8 − q 7 t9 + q 8 t9 + 2q 9 t9 −

2q 10 t9 − q 11 t9 + q 12 t9 − q 8 t10 + 2q 9 t10 − 2q 11 t10 + q 12 t10 − q 9 t11 + 3q 10 t11 − 3q 11 t11 + q 12 t11 .
The a–degree in the first formula is 4, which is (again) greater than that in the
second : dega = s(3| |)+max{ s, 1}| |−| | = 3. The positivity of H min
L, L (q, t, a)/(1−
t)3 holds, but fails for (1 − t)2 , this agrees with κ + κ − 1 = 3 from the Connection
Conjecture. Both polynomials are self-dual and become (1+a)3 as q = 1. Using the
spherical polynomials would result at 1 for q = 1 (since all components are unknots);
the factor (1 + a)4−1 is due to using J–polynomials and the min–normalization.
The corresponding plane curve singularity is:
- .
C = {y 8 + x4 = 0}, Link(C) = 4 × , lki =j = 2 .
Its Alexander polynomial is 1 + q 4 − 2q 8 − 2q 12 + q 16 + q 20 . See Section 8.4

below for details.
This link is isotopic to uncolored (4 × Cab(2, 1))T (1, 1). The corresponding
→
◦→◦ →
→ , (4× )
HM (q, t, a) for M = L{1,1},{1,1}
min →
coincides with that from (7.37), which also
→
→
coincides with H min (q, t, a) for N = L◦ →
→
, (4× )
. Its link is (4 × Cab(0, 1))T (2, 1),
N {2,1}
isotopic to the previous two.
Let us check directly the coincidence of the superpolynomials for the original
{L, L} and N above. For E = Eλ , P = Pλ , where λ ∈ P, P+ respectively, one has:
(7.38) {E(Y −1 ) τ+−1 τ−

−1
(Q)}ev {τ+−1 τ−
−1
(EQ)}ev , Q ∈ V,
{ P (Y ) τ+ τ− (Q) }ev { τ+ τ− (P Q) }ev for Q ∈ V W,
though one can assume that Q ∈ V in both relations. By , we mean the equality
up to a factor q • t• . The second relation follows from the first upon applying η inside
the coinvariant. See (1.16), (1.21) and Section 1.3 for this and other properties we
will need.
−1 −1 −1
Let us check the first equality. We use the identity τ+ τ− τ+ = σ = τ− τ+ τ− ,
−1 −1 −1
its corollary τ+ τ− τ+ = τ− τ+ τ− , the ϕ–invariance of {·}ev , and also the relation
ϕ(Q) = σ(Q). One has:

{E(Y −1 ) (τ+−1 τ−
−1
)(Q)}ev = {ϕ E(Y −1 ) (τ+−1 τ−
−1
τ+ )(Q) }ev
={(τ+−1 τ− τ+ )(τ−
−1 −1
τ+ τ− )(Q) E}ev = {(τ+−1 τ− (τ+ τ−
−1 −1
τ+ )τ− )(Q) E}ev
−2 −2
={τ− (Q) E}ev {τ− (QE)}ev = {(τ+−1 τ−
−1
)(QE)}ev .
8. Further examples
8.1. On cable notations. Let us somewhat simplify the notations. If the tree
is one path extended by several arrowheads at its end, then the notation without
the duplication of the sets (!rj ,!sj ) can be used, since the corresponding paths are
different only by the arrowheads (i.e. by colors). More generally, in the case of
one base path with several arrowheads going from its vertices (not only from its
very last vertex), we will provide the [r,s]-parameters only for this path. The other
paths will be its portions that and the arrowheads before the final vertex.
For example, for the tree Υ = {◦ →
→
→◦→} with 2 vertices and 2 arrowheads, the
◦
link L{3,2},{−2,1} means combinatorially and topologically:
→◦→
→
◦
1 2
1 2
- , (λ ,λ ) . -
L →◦→ . ∼ Cab(−11, 2)λ Cab(0, 1)λ T (3, 2),
( {3,2},{−2,1} , {3,2} )
where we count the paths from the bottom to the top. Recall that the labels
[r, s] corresponding here to !r = {3, 2},!s = {−2,

1} are

[3, −2] and [2, 1], so they
3 2
can be naturally considered as the columns of −2 1
; these columns become the
first columns of the corresponding γ. To avoid any misunderstanding in some
examples, we will continue duplicating {!r,!s} for the paths different only by the

arrowheads. Also, the symbol will be mainly omitted below if the confusion
with the composition of cables is impossible.
8.2. Colored 2-links. Actually we have already considered quite a few col-
ored examples in Section 7. Let us add some simple colors to the 2–fold trefoil.
min are as follows:

2-column and 2-row for trefoil. The [r,s]-presentation and H
1 ≤ j ≤ κ = 2, !rj = 3, !sj = 2, Υ = {◦ ⇒} , λ1 = , λ2 = ;
(8.1)
◦⇒, ( , ) min
T (6, 4) : L = L({3,2},{3,2}) , H L (q, t, a) =
1 + qt + q 2 t + q 3 t − t2 + q 3 t2 + 3q 4 t2 + q 5 t2 − qt3 − q 2 t3 − q 3 t3 + 3q 5 t3 + 2q 6 t3 + q 7 t3 − q 3 t4 −
3q 4 t4 − q 5 t4 + 3q 6 t4 + 3q 7 t4 + q 8 t4 − 3q 5 t5 − 2q 6 t5 + q 7 t5 + 3q 8 t5 + 3q 9 t5 − 3q 6 t6 − 3q 7 t6 +
2q 8 t6 + 3q 9 t6 + 2q 10 t6 − 2q 7 t7 − 3q 8 t7 − q 9 t7 + 3q 10 t7 + 3q 11 t7 − 3q 8 t8 − 3q 9 t8 + q 10 t8 +
3q 11 t8 + 2q 12 t8 − 2q 9 t9 − 3q 10 t9 − q 11 t9 + 3q 12 t9 + 3q 13 t9 − 3q 10 t10 − 3q 11 t10 + 2q 12 t10 +
3q 13 t10 + q 14 t10 − 2q 11 t11 − 3q 12 t11 + q 13 t11 + 3q 14 t11 + q 15 t11 − 3q 12 t12 − 2q 13 t12 + 3q 14 t12 +
2q 15 t12 − 3q 13 t13 − q 14 t13 + 3q 15 t13 + q 16 t13 − 3q 14 t14 + 3q 16 t14 − q 14 t15 − q 15 t15 + q 16 t15 +
q 17 t15 − q 15 t16 + q 17 t16 − q 15 t17 + q 17 t17 − q 16 t18 + q 18 t18
15 10 11 12 13
+a5 q t − q 15 t + q 17 t − q 17 t3 + q 19 t3 + a4 q 12 + q 13 + q 14 + q t + q t + q t + q t − q 10 t −
q t − q t + 2q t + 2q t − q t − q t + 2q t + q t − q t − 2q t − q t + 2q 16 t3 +
11 12 14 15 12 2 13 2 15 2 16 2 13 3 14 3 15 3
2q 17 t3 − q 14 t4 − 2q 15 t4 + 2q 17 t4 + q 18 t4 − q 15 t5 − q 16 t5 + q 18 t5 + q 19 t5 − q 16 t6 − q 17 t6 +
q 18 t6 + q 19 t6 − q 17 t7 + q 19 t7 − q 17 t8 + q 19 t8
6 7 8 9 10
+a3 q 8 + 3q 9 + 3q 10 + 2q 11 + q 12 + qt + qt + 2qt + qt + q t − q 6 t − q 7 t − 2q 8 t − q 9 t + 3q 10 t +
4q t+4q t+q t−q t −3q t −3q t +q t +5q t +5q 13 t2 +2q 14 t2 −4q 10 t3 −4q 11 t3 −
11 12 13 8 2 9 2 10 2 11 2 12 2
q 12 t3 + 4q 13 t3 + 6q 14 t3 + q 15 t3 − 3q 11 t4 − 6q 12 t4 − 2q 13 t4 + 4q 14 t4 + 6q 15 t4 + 2q 16 t4 − 3q 12 t5 −
5q 13 t5 − 2q 14 t5 + 4q 15 t5 + 5q 16 t5 + q 17 t5 − 3q 13 t6 − 6q 14 t6 − q 15 t6 + 5q 16 t6 + 4q 17 t6 + q 18 t6 −
3q 14 t7 − 4q 15 t7 + q 16 t7 + 4q 17 t7 + 2q 18 t7 − 4q 15 t8 − 3q 16 t8 + 3q 17 t8 + 3q 18 t8 + q 19 t8 − 3q 16 t9 −
q 17 t9 + 3q 18 t9 + q 19 t9 − q 16 t10 − 2q 17 t10 + q 18 t10 + 2q 19 t10 − q 17 t11 + q 19 t11 − q 17 t12 + q 19 t12
3 4 5 6
+a2 2q 5 +3q 6 +4q 7 +2q 8 +q 9 + qt + qt + qt + qt −q 3 t−q 4 t−q 5 t+4q 7 t+7q 8 t+4q 9 t+2q 10 t−
2q 5 t2 − 3q 6 t2 − 4q 7 t2 + q 8 t2 + 8q 9 t2 + 8q 10 t2 + 4q 11 t2 + q 12 t2 − q 6 t3 − 4q 7 t3 − 7q 8 t3 − 2q 9 t3 +
7q 10 t3 + 9q 11 t3 + 5q 12 t3 − 3q 8 t4 − 9q 9 t4 − 6q 10 t4 + 5q 11 t4 + 10q 12 t4 + 6q 13 t4 + q 14 t4 − 2q 9 t5 −
9q 10 t5 − 7q 11 t5 + 4q 12 t5 + 10q 13 t5 + 6q 14 t5 − 2q 10 t6 − 9q 11 t6 − 9q 12 t6 + 4q 13 t6 + 10q 14 t6 +
5q 15 t6 +q 16 t6 −2q 11 t7 −9q 12 t7 −7q 13 t7 +5q 14 t7 +9q 15 t7 +4q 16 t7 −2q 12 t8 −9q 13 t8 −6q 14 t8 +
7q 15 t8 + 8q 16 t8 + 2q 17 t8 − 2q 13 t9 − 9q 14 t9 − 2q 15 t9 + 8q 16 t9 + 4q 17 t9 + q 18 t9 − 3q 14 t10 −
7q 15 t10 +q 16 t10 +7q 17 t10 +2q 18 t10 −4q 15 t11 −4q 16 t11 +4q 17 t11 +4q 18 t11 −q 15 t12 −3q 16 t12 +
3q 18 t12 + q 19 t12 − 2q 16 t13 − q 17 t13 + 2q 18 t13 + q 19 t13 − q 17 t14 + q 19 t14 − q 17 t15 + q 19 t15

+a q 2 + 2q 3 + 2q 4 + q 5 + qt − qt + q 3 t + 2q 4 t + 5q 5 t + 3q 6 t + q 7 t − q 2 t2 − 2q 3 t2 − 2q 4 t2 +
6q 6 t2 + 7q 7 t2 + 3q 8 t2 + q 9 t2 − q 3 t3 − 2q 4 t3 − 5q 5 t3 − 3q 6 t3 + 4q 7 t3 + 8q 8 t3 + 6q 9 t3 + q 10 t3 −
q 5 t4 − 6q 6 t4 − 7q 7 t4 + 2q 8 t4 + 8q 9 t4 + 7q 10 t4 + 2q 11 t4 − 5q 7 t5 − 8q 8 t5 − 2q 9 t5 + 8q 10 t5 +
9q 11 t5 + 2q 12 t5 − 5q 8 t6 − 9q 9 t6 − 2q 10 t6 + 7q 11 t6 + 8q 12 t6 + 2q 13 t6 − 4q 9 t7 − 9q 10 t7 − 5q 11 t7 +
7q 12 t7 + 9q 13 t7 + 2q 14 t7 − 5q 10 t8 − 9q 11 t8 − 2q 12 t8 + 8q 13 t8 + 7q 14 t8 + q 15 t8 − 4q 11 t9 −
9q 12 t9 − 2q 13 t9 + 8q 14 t9 + 6q 15 t9 + q 16 t9 − 5q 12 t10 − 8q 13 t10 + 2q 14 t10 + 8q 15 t10 + 3q 16 t10 −
5q 13 t11 −7q 14 t11 +4q 15 t11 +7q 16 t11 +q 17 t11 −6q 14 t12 −3q 15 t12 +6q 16 t12 +3q 17 t12 −q 14 t13 −
5q 15 t13 + 5q 17 t13 + q 18 t13 − 2q 15 t14 − 2q 16 t14 + 2q 17 t14 + 2q 18 t14 − q 15 t15 − 2q 16 t15 + q 17 t15 +

2q 18 t15 − q 16 t16 + q 18 t16 − q 17 t17 + q 19 t17 .
The a–degree of H min (q, t, a) is 5, which matches the formula dega = s(| | +
L
| |) − | | = 5 from (5.41). The self-duality and other claims in this theorem
hold. For instance, the transposition 6 ↔ 4 in T (6, 4) does not influence the
superpolynomial.
min (q, t, a)/(1 − t) from (5.56) holds; though it was conjec-
The positivity of H L
tured with such a power of (1 − t) only in the uncolored case.
Concerning the HOMFLY-PT polynomials, the following holds:
L
H min 4 jLo =1 (q, a).
(q, q, −a)/den1 = HOM
(1−q)(1−q 2 )
Here den1 ( , )= (1−a/q) . Otherwise in this section:
den1 (λ1 , ) = (1 − q), den1 (λ1 , ) = (1 − q)(1 − q 2 ) = den1 (λ1 , ).

Note that den1 depends only on the colors and therefore is the same as for the
colored Hopf links in (7.24).
2-fold T(2,1) . One has:
1 ≤ j ≤ κ = 2, !rj = 2, !sj = 1, Υ = {◦ ⇒} , λ1 = , λ2 = ;
(8.2)
◦⇒, ( , ) min
T (4, 2) : L = L({2,1},{2,1}) , H L (q, t, a) =
1 − t2 + q 2 t2 − q 2 t4 + q 4 t4 + a(q 2 − q 2 t2 + q 4 t2 ).
The a–degree is s(|λ1 | + |λ2 |) − |λ1 ∨λ2 | = 1. The self-duality and other claims
hold, including the positivity upon the division by (1 − t). Interestingly, (8.2) is
the result of the substitution q → q 2 , t → t2 in the superpolynomial at (6.7) for

λ1 = = λ2 :
min
H , = 1 − t + qt − qt + q t + a(q − qt + q t).
2 2 2 2
For the same tree and λ1 = , λ2 = :

(8.3) min (q, t, a) = 1 − t2 + qt2 − qt4 + q 2 t4 + a(q − qt2 + q 2 t2 ).
H
,
This is the result of the substitution t → t2 in the previous formula.

T [4,2]
The superpolynomial from (8.3) is the major factor of −P[1,1],1 in (237) from
[DMS] upon the substitution q → q 2 , t → t2 , a → A2 . Colored superpolynomials
for T (6, 4) are not given in [DMS].
2-column and 1 box for trefoil. Let us somewhat reduce the previous example;
we will consider the diagrams λ1 = , λ2 = for the trefoil:
(8.4)
◦⇒, ( , ) min
T (6, 4) : L = L({3,2},{3,2}) , H L (q, t, a) =
1+2qt−t2 +2q 2 t2 +q 3 t2 −2qt3 +q 2 t3 +2q 3 t3 −2q 2 t4 +3q 4 t4 −q 2 t5 −q 3 t5 +2q 4 t5 +q 5 t5 −q 3 t6 −
2q 4 t6 +3q 5 t6 −q 3 t7 −q 4 t7 +q 5 t7 +q 6 t7 −q 4 t8 −2q 5 t8 +3q 6 t8 −q 4 t9 −q 5 t9 +q 6 t9 +q 7 t9 −q 5 t10 −
q 6 t10 + 2q 7 t10 − q 5 t11 − q 6 t11 + 2q 7 t11 − q 6 t12 + q 8 t12 − q 6 t13 + q 8 t13 − q 7 t14 + q 8 t14 − q 7 t15 +
7 4 5 6
q 8 t15 −q 8 t17 +q 9 t17 +a4 qt −q 7 t+q 8 t−q 8 t3 +q 9 t3 +a3 q 5 +q 6 + qt + qt + qt −q 4 t−q 5 t+
q t+2q t−q t +2q t −2q t +2q t −q t −q t +2q t −2q t +q t +q t −q 7 t6 +q 9 t6 −
6 7 5 2 7 2 6 3 8 3 6 4 7 4 8 4 7 5 8 5 9 5
2 3 4
q 8 t7 +q 9 t7 −q 8 t8 +q 9 t8 +a2 2q 3 +2q 4 +q 5 + qt + qt + qt −q 2 t−q 3 t+2q 4 t+3q 5 t+q 6 t−2q 3 t2 −
q t +4q t +3q t −3q t −q t +4q t +q t −q t −4q 5 t4 +2q 6 t4 +4q 7 t4 −2q 5 t5 −3q 6 t5 +
4 2 5 2 6 2 4 3 5 3 6 3 7 3 4 4
4q 7 t5 +q 8 t5 −q 5 t6 −4q 6 t6 +2q 7 t6 +3q 8 t6 −2q 6 t7 −2q 7 t7 +4q 8 t7 −q 6 t8 −3q 7 t8 +3q 8 t8 +q 9 t8 −

2q 7 t9 +q 8 t9 +q 9 t9 −q 7 t10 −q 8 t10 +2q 9 t10 −q 8 t11 +q 9 t11 −q 8 t12 +q 9 t12 +a q +2q 2 +q 3 + qt −
qt+2q 2 t+3q 3 t+q 4 t−qt2 −2q 2 t2 +2q 3 t2 +5q 4 t2 +q 5 t2 −2q 2 t3 −2q 3 t3 +3q 4 t3 +3q 5 t3 −3q 3 t4 −
3q 4 t4 +6q 5 t4 +2q 6 t4 −q 3 t5 −3q 4 t5 +q 5 t5 +4q 6 t5 −2q 4 t6 −5q 5 t6 +5q 6 t6 +2q 7 t6 −q 4 t7 −3q 5 t7 +
4q 7 t7 − 2q 5 t8 − 4q 6 t8 + 5q 7 t8 + q 8 t8 − q 5 t9 − 3q 6 t9 + 2q 7 t9 + 2q 8 t9 − 2q 6 t10 − 2q 7 t10 + 4q 8 t10 −

q 6 t11 − 2q 7 t11 + 3q 8 t11 − 2q 7 t12 + q 8 t12 + q 9 t12 − q 7 t13 + q 9 t13 − q 8 t14 + q 9 t14 − q 8 t15 + q 9 t15 .
The a–degree of H min (q, t, a) is dega = s(| | + | |) − | | = 4. This super-

L
polynomial is dual to that for λ1 = , λ2 = , which we will omit. The positivity

of the series HL (q, t, a)/(1 − t) holds.
min
8.3. 3-hook for trefoil. Let us conclude this set of examples with λ1 =
, λ2 = for the trefoil:
(8.5)
◦⇒, ( , ) min
T (6, 4) : L = L({3,2},{3,2}) , H L (q, t, a) =
1 − t + 2qt + q 2 t + q 3 t − 3qt2 + q 2 t2 + q 3 t2 + 4q 4 t2 + q 5 t2 + qt3 − 4q 2 t3 − q 3 t3 − 3q 4 t3 + 6q 5 t3 +
3q 6 t3 + q 7 t3 + 2q 2 t4 − 3q 3 t4 − q 4 t4 − 9q 5 t4 + 6q 6 t4 + 5q 7 t4 + 2q 8 t4 + 2q 3 t5 − 2q 4 t5 + 2q 5 t5 −
13q 6 t5 +2q 7 t5 +6q 8 t5 +5q 9 t5 +2q 4 t6 −2q 5 t6 +4q 6 t6 −14q 7 t6 +5q 9 t6 +6q 10 t6 +2q 5 t7 −2q 6 t7 +
6q 7 t7 − 14q 8 t7 − 3q 9 t7 + 4q 10 t7 + 7q 11 t7 + 2q 6 t8 − 2q 7 t8 + 6q 8 t8 − 14q 9 t8 − 2q 10 t8 + 4q 11 t8 +
6q 12 t8 +2q 7 t9 −2q 8 t9 +7q 9 t9 −14q 10 t9 −3q 11 t9 +5q 12 t9 +5q 13 t9 +2q 8 t10 −2q 9 t10 +6q 10 t10 −
14q 11 t10 +6q 13 t10 +2q 14 t10 +2q 9 t11 −2q 10 t11 +6q 11 t11 −14q 12 t11 +2q 13 t11 +5q 14 t11 +q 15 t11 +
2q 10 t12 −2q 11 t12 +4q 12 t12 −13q 13 t12 +6q 14 t12 +3q 15 t12 +2q 11 t13 −2q 12 t13 +2q 13 t13 −9q 14 t13 +
6q 15 t13 + q 16 t13 + 2q 12 t14 − 2q 13 t14 − q 14 t14 − 3q 15 t14 + 4q 16 t14 + 2q 13 t15 − 3q 14 t15 − q 15 t15 +
q 16 t15 +q 17 t15 +2q 14 t16 −4q 15 t16 +q 16 t16 +q 17 t16 +q 15 t17 −3q 16 t17 +2q 17 t17 −q 17 t18 +q 18 t18
q 15

+a5 −q 15 + q 16 + t
− 2q 16 t + 2q 17 t + q 16 t2 − 2q 17 t2 + q 18 t2 − q 18 t3 + q 19 t3 + a4 −q 10 +
10 11 12 13
q 12 +q 13 +2q 14 + q t + q t + q t + q t −2q 11 t−2q 12 t+q 14 t+4q 15 t+q 11 t2 −q 12 t2 −3q 13 t2 +
4q 16 t2 + q 12 t3 − 4q 14 t3 − q 15 t3 + 4q 17 t3 + q 13 t4 − 4q 15 t4 + q 17 t4 + 2q 18 t4 + q 14 t5 − 3q 16 t5 +

q 18 t5 + q 19 t5 + q 15 t6 − q 16 t6 − 2q 17 t6 + q 18 t6 + q 19 t6 + q 16 t7 − 2q 17 t7 + q 19 t7 − q 18 t8 + q 19 t8
6 7 8 9 10
+a3 −q 6 + 4q 9 + 3q 10 + 3q 11 + q 12 + qt + qt + 2qt + qt + q t − 2q 7 t − 2q 8 t − 5q 9 t + 4q 10 t +
4q 11 t + 6q 12 t + 2q 13 t + q 7 t2 − q 8 t2 − q 9 t2 − 11q 10 t2 + q 11 t2 + 5q 12 t2 + 8q 13 t2 + 4q 14 t2 + q 8 t3 +
2q 10 t3 − 14q 11 t3 − 4q 12 t3 + 4q 13 t3 + 9q 14 t3 + 4q 15 t3 + q 9 t4 + 5q 11 t4 − 15q 12 t4 − 7q 13 t4 +
4q 14 t4 + 9q 15 t4 + 4q 16 t4 + q 10 t5 + 6q 12 t5 − 14q 13 t5 − 7q 14 t5 + 4q 15 t5 + 8q 16 t5 + 2q 17 t5 +
q 11 t6 + 6q 13 t6 − 15q 14 t6 − 4q 15 t6 + 5q 16 t6 + 6q 17 t6 + q 18 t6 + q 12 t7 + 5q 14 t7 − 14q 15 t7 + q 16 t7 +
4q 17 t7 + 3q 18 t7 + q 13 t8 + 2q 15 t8 − 11q 16 t8 + 4q 17 t8 + 3q 18 t8 + q 19 t8 + q 14 t9 − q 16 t9 − 5q 17 t9 +

4q 18 t9 +q 19 t9 +q 15 t10 −q 16 t10 −2q 17 t10 +2q 19 t10 +q 16 t11 −2q 17 t11 +q 19 t11 −q 18 t12 +q 19 t12
3 4 5 6
+a2 −q 3 + 2q 5 + 3q 6 + 5q 7 + 2q 8 + q 9 + qt + qt + qt + qt − 2q 4 t − 3q 5 t − q 6 t + 2q 7 t +
10q 8 t + 5q 9 t + 3q 10 t + q 4 t2 − q 5 t2 − 5q 6 t2 − 7q 7 t2 − 6q 8 t2 + 13q 9 t2 + 10q 10 t2 + 7q 11 t2 +
q 12 t2 + q 5 t3 + q 6 t3 − 3q 7 t3 − 10q 8 t3 − 17q 9 t3 + 11q 10 t3 + 13q 11 t3 + 10q 12 t3 + q 13 t3 +
q 6 t4 + 2q 7 t4 + q 8 t4 − 8q 9 t4 − 27q 10 t4 + 6q 11 t4 + 15q 12 t4 + 12q 13 t4 + 2q 14 t4 + q 7 t5 + 2q 8 t5 +
3q 9 t5 − 4q 10 t5 − 31q 11 t5 + 2q 12 t5 + 16q 13 t5 + 12q 14 t5 + q 15 t5 + q 8 t6 + 2q 9 t6 + 4q 10 t6 −
2q 11 t6 − 33q 12 t6 + 2q 13 t6 + 15q 14 t6 + 10q 15 t6 + q 16 t6 + q 9 t7 + 2q 10 t7 + 4q 11 t7 − 2q 12 t7 −
31q 13 t7 + 6q 14 t7 + 13q 15 t7 + 7q 16 t7 + q 10 t8 + 2q 11 t8 + 4q 12 t8 − 4q 13 t8 − 27q 14 t8 + 11q 15 t8 +
10q 16 t8 + 3q 17 t8 + q 11 t9 + 2q 12 t9 + 3q 13 t9 − 8q 14 t9 − 17q 15 t9 + 13q 16 t9 + 5q 17 t9 + q 18 t9 +
q 12 t10 + 2q 13 t10 + q 14 t10 − 10q 15 t10 − 6q 16 t10 + 10q 17 t10 + 2q 18 t10 + q 13 t11 + 2q 14 t11 −
3q 15 t11 − 7q 16 t11 + 2q 17 t11 + 5q 18 t11 + q 14 t12 + q 15 t12 − 5q 16 t12 − q 17 t12 + 3q 18 t12 + q 19 t12 +

q 15 t13 − q 16 t13 − 3q 17 t13 + 2q 18 t13 + q 19 t13 + q 16 t14 − 2q 17 t14 + q 19 t14 − q 18 t15 + q 19 t15

+a −q + 2q 2 + 2q 3 + 2q 4 + q 5 + t − 3q 2 t + q 3 t + 2q 4 t + 6q 5 t + 4q 6 t + q 7 t + q 2 t2 − 5q 3 t2 −
q
3q 4 t2 − 4q 5 t2 + 8q 6 t2 + 10q 7 t2 + 4q 8 t2 + q 9 t2 + 2q 3 t3 − 4q 4 t3 − 4q 5 t3 − 14q 6 t3 + 3q 7 t3 +
14q 8 t3 + 9q 9 t3 + 2q 10 t3 + 3q 4 t4 − 2q 5 t4 − 21q 7 t4 − 6q 8 t4 + 14q 9 t4 + 13q 10 t4 + 4q 11 t4 + 3q 5 t5 −
q 6 t5 + 5q 7 t5 − 22q 8 t5 − 15q 9 t5 + 12q 10 t5 + 17q 11 t5 + 5q 12 t5 + 3q 6 t6 − q 7 t6 + 8q 8 t6 − 21q 9 t6 −
19q 10 t6 + 9q 11 t6 + 17q 12 t6 + 5q 13 t6 + 3q 7 t7 − q 8 t7 + 10q 9 t7 − 20q 10 t7 − 22q 11 t7 + 9q 12 t7 +
17q 13 t7 + 4q 14 t7 + 3q 8 t8 − q 9 t8 + 10q 10 t8 − 20q 11 t8 − 19q 12 t8 + 12q 13 t8 + 13q 14 t8 + 2q 15 t8 +
3q 9 t9 − q 10 t9 + 10q 11 t9 − 21q 12 t9 − 15q 13 t9 + 14q 14 t9 + 9q 15 t9 + q 16 t9 + 3q 10 t10 − q 11 t10 +
8q 12 t10 − 22q 13 t10 − 6q 14 t10 + 14q 15 t10 + 4q 16 t10 + 3q 11 t11 − q 12 t11 + 5q 13 t11 − 21q 14 t11 +
3q 15 t11 + 10q 16 t11 + q 17 t11 + 3q 12 t12 − q 13 t12 − 14q 15 t12 + 8q 16 t12 + 4q 17 t12 + 3q 13 t13 −
2q 14 t13 −4q 15 t13 −4q 16 t13 +6q 17 t13 +q 18 t13 +3q 14 t14 −4q 15 t14 −3q 16 t14 +2q 17 t14 +2q 18 t14 +

2q 15 t15 − 5q 16 t15 + q 17 t15 + 2q 18 t15 + q 16 t16 − 3q 17 t16 + 2q 18 t16 − q 18 t17 + q 19 t17 .
The a–degree of H min (q, t, a) is s(| | + | |) − | | = 5. This superpolynomial

L
is self-dual. The positivity of the series H min (q, t, a)/(1 − t)2 holds. Here (1 − t)2
L
is minimal necessary, since the 3-hook alone requires the division by (1 − t) for the
positivity and we have another (1 − t) because we do a 2–link. This is of course an
informal argument, not a justification.
Generally, Part (ii) of the Connection Conjecture claims that proper powers of
(1 − t)(1 − q) are necessary for arbitrary colors for positive pairs of trees {L, L∨ },
i.e. those satisfying the inequalities r, r, s, s > 0 and s1 s1 > r1 r1 . This is supposed
to be similar to Corollary 4.10 from [EGL] concerning the positivity of unreduced
HOMFLY-PT polynomials. See also Example 4.22 there concerning the 3–hook
and Section 5.4 devoted to the 3–hook in [Ch2]. However the powers (1 − t)• are
insufficient for algebraic links, which cannot be seen in the theory of HOMFLY-PT
polynomials (where t = q). We do not have any conjectures at the moment about
the minimal powers of (1 − t) and (1 − q) necessary for the positivity (assuming

that Part (ii) is true).
8.4. Alexander polynomials. The uncolored reduced hat-normalized

HOMFLY-PT polynomial is H min (q, q, −a)/(1−q)κ−1 , recalculated to the standard
L
a, q. Therefore the corresponding Alexander polynomial must be H min (q, q, −1)/(1−
L
κ−δκ,1
q) , where δ is the Kronecker symbol. Without δκ,1 , this will be the multi-
variable Alexander polynomial for coinciding parameters. For algebraic links, this
polynomial is directly connected with the zeta function of monodromy of the Milnor
fiber of the corresponding singularity. Here κ is the number of connected compo-
nent of the link, which is 2 in the example above. Generally, the substitution
a → −1 is the passage to the Heegaard-Floer homology. We will provide in this
section the equations of the corresponding singularities (at x = 0, y = 0) and their
Alexander polynomials, the zeta-monodromy from [DGPS] upon t → q (unless for
the unknot).
Note that the value of the Alexander polynomial at q = 1 for any 2–link (a
link with 2 components) is ± lk for the linking number lk. Thus this polynomial
uniquely determines the (topological) type of singularity in this case, assuming
that the corresponding knot components are known. Therefore H must uniquely
determine such singularities.
Let us consider the uncolored Cab(7, 1)T (6, 4) for Υ = {◦ → ◦ ⇒} with 2
arrowheads. We set !r = {3, 1},!s = {2, 1}. The [r,s]-labels of the vertices are
[3, 2] and [1, 1] (the first and the second). As above, we put T (6, 4) instead of
(Cab(2, 3)Cab(2, 3))().
(8.6) Cab(7, 1)T (6, 4) : L = L{3,1},{2,1}

◦→◦⇒, ( , ) min
, H L (q, t, a) =
1 − t + qt + q 2 t + q 3 t − qt2 + 2q 4 t2 − q 2 t3 − q 4 t3 + 2q 5 t3 − q 3 t4 − q 5 t4 + 2q 6 t4 − q 4 t5 − q 6 t5 +

2q 7 t5 − q 5 t6 + q 8 t6 − q 6 t7 + q 8 t7 − q 7 t8 + q 8 t8 − q 8 t9 + q 9 t9 + a3 q 6 − q 6 t + q 7 t − q 7 t2 + q 8 t2 −
3
q t + q t + a q + q + q − q t + q t + 2q t − q t − q t + 2q 7 t2 − q 5 t3 − q 6 t3 + 2q 8 t3 −
8 3 9 3 2 4 5 3 5 6 4 2 5 2

q 6 t4 − q 7 t4 + q 8 t4 + q 9 t4 − q 7 t5 + q 9 t5 − q 8 t6 + q 9 t6 + a q + q 2 + q 3 − qt + q 3 t + 3q 4 t + q 5 t −
q t − q t − q t + 3q t + q t − q t − q t − 2q t + 3q 6 t3 + q 7 t3 − q 4 t4 − q 5 t4 − 2q 6 t4 +
2 2 3 2 4 2 5 2 6 2 3 3 4 3 5 3

3q 7 t4 +q 8 t4 −q 5 t5 −q 6 t5 −q 7 t5 +3q 8 t5 −q 6 t6 −q 7 t6 +q 8 t6 +q 9 t6 −q 7 t7 +q 9 t7 −q 8 t8 +q 9 t8 .
The a–degree of H min (q, t, a) is dega = s1 r2 | | + s1 | | − | | = 2 × 1 + 2 − 1 = 3.
L
The corresponding singularity is:
- .
C : {(x3 +y 2 )2 +x7 = 0}, Link(C) = T (3, 2), T (3, 2), lk= 7 .
The Alexander polynomial is 1 + q 4 + q 6 + q 8 + q 10 + q 12 + q 16 ; use [DGPS]. This
example is of interest because the link of (x3 +y 2 )2 +x7 +x5 y = 0 is Cab(13, 2)T (3, 2),
which is the simplest non-torus algebraic knot ; its Alexander polynomial is 1 − q +
q 4 − q 5 + q 6 − q 7 + q 8 − q 9 + q 10 − q 11 + q 12 − q 15 + q 16 .
We note that the term x5 y here is minimal in the following sense. If it is
replaced
- by x4 y, then.the corresponding Link((x3 + y 2 )2 + x7 + x4 y = 0) becomes
T (4, 3), T (1, 0), lk= 4 , i.e. not a knot.
Now let us discuss the simplest algebraic link obtained by linking T (3, 2) and
T (5, 3). It will be Cab(1, 1)(T (3, 2)T (5, 3)) for Υ = {◦ ⇒◦◦⇒ } with !r1 = {1, 2},!s1 =
{1, 1} and !r2 = {1, 3},!s2 = {1, 2}. The [r,s]-labels of the vertices are [1, 1] and
{[2, 1], [3, 2]} (for the first ◦ and for the remaining two vertices).

◦ ⇒ ◦ ⇒, ( , )
(8.7) Cab(3, 2)Cab(5, 3) T (1, 1) : L- ◦ .- min
., H L =
{1,2},{1,1} , {1,3},{1,2}
1−t+qt+q t+q t+q t−qt +2q t +2q t +q t −q 2 t3 −2q 4 t3 +q 5 t3 +3q 6 t3 +2q 7 t3 −q 3 t4 −

2 3 4 2 4 2 5 2 6 2
2q 5 t4 +3q 7 t4 +2q 8 t4 −q 4 t5 −2q 6 t5 +3q 8 t5 +q 9 t5 −q 5 t6 −2q 7 t6 +q 8 t6 +2q 9 t6 −q 6 t7 −2q 8 t7 +

2q 9 t7 + q 10 t7 − q 7 t8 + q 10 t8 − q 8 t9 + q 10 t9 − q 9 t10 + q 10 t10 − q 10 t11 + q 11 t11 + a4 q 10 + q 11 t +

a3 q 6 + q 7 + q 8 + q 9 − q 6 t + q 7 t + 2q 8 t + 2q 9 t + 2q 10 t − 2q 7 t2 + 2q 9 t2 + 2q 10 t2 + q 11 t2 − 2q 8 t3 +

2q 10 t3 + q 11 t3 − 2q 9 t4 + q 10 t4 + q 11 t4 − q 10 t5 + q 11 t5 + a2 q 3 + q 4 + 2q 5 + q 6 + q 7 − q 3 t + q 5 t +
4q 6 t + 4q 7 t + 3q 8 t + q 9 t − q 4 t2 − 2q 5 t2 − 2q 6 t2 + 4q 7 t2 + 6q 8 t2 + 4q 9 t2 + q 10 t2 − q 5 t3 − 2q 6 t3 −
4q 7 t3 + 3q 8 t3 + 6q 9 t3 + 3q 10 t3 − q 6 t4 − 2q 7 t4 − 4q 8 t4 + 4q 9 t4 + 4q 10 t4 + q 11 t4 − q 7 t5 − 2q 8 t5 −

2q 9 t5 +4q 10 t5 +q 11 t5 −q 8 t6 −2q 9 t6 +q 10 t6 +2q 11 t6 −q 9 t7 +q 11 t7 −q 10 t8 +q 11 t8 +a q +q 2 +
q 3 +q 4 −qt+q 3 t+3q 4 t+4q 5 t+2q 6 t+q 7 t−q 2 t2 −q 3 t2 −2q 4 t2 +2q 5 t2 +6q 6 t2 +5q 7 t2 +2q 8 t2 −
q 3 t3 −q 4 t3 −4q 5 t3 −q 6 t3 +6q 7 t3 +6q 8 t3 +2q 9 t3 −q 4 t4 −q 5 t4 −4q 6 t4 −2q 7 t4 +6q 8 t4 +5q 9 t4 +
q 10 t4 −q 5 t5 −q 6 t5 −4q 7 t5 −q 8 t5 +6q 9 t5 +2q 10 t5 −q 6 t6 −q 7 t6 −4q 8 t6 +2q 9 t6 +4q 10 t6 −q 7 t7 −

q 8 t7 − 2q 9 t7 + 3q 10 t7 + q 11 t7 − q 8 t8 − q 9 t8 + q 10 t8 + q 11 t8 − q 9 t9 + q 11 t9 − q 10 t10 + q 11 t10 .
The a–degree is s11 r21 +s21 r22 −1 = 3+2−1 = 4. The positivity of H min (q, t, a)/(1−
L
t) holds. The corresponding singularity is given by the equation (y +x3 )(y 2 +x3 ) =
5
0 with the linking number lk = 6; its Alexander polynomial is 1 + q 6 + q 7 + q 13 +

q 14 + q 20 .
Our last example will be for the uncolored tree Υ = {◦ →
→◦→} with the labeled
vertices [r11 = 2, s11 = 1], [r21 = 3, s12 = 2] for the long path and [r12 = 2, s21 = 1] for the
short one (equal to [r11 , s11 ] in the tree).

→
(8.8)
◦→◦→, ( , )
Cab(8, 3)Cab(0, 1) T (2, 1) : L = L{2,3},{1,2} , H min
L (q, t, a) =
1 − t + qt + q 2 t + q 3 t − qt2 + 2q 4 t2 + q 5 t2 + q 6 t2 − q 2 t3 − q 4 t3 + q 5 t3 + 2q 6 t3 + 2q 7 t3 − q 3 t4 −
q 5 t4 + q 7 t4 + 3q 8 t4 − q 4 t5 − q 6 t5 + q 8 t5 + 3q 9 t5 − q 5 t6 − q 7 t6 + q 9 t6 + 2q 10 t6 − q 6 t7 − q 8 t7 +
2q 10 t7 + q 11 t7 − q 7 t8 − q 9 t8 + q 10 t8 + q 11 t8 − q 8 t9 − q 10 t9 + 2q 11 t9 − q 9 t10 + q 12 t10 − q 10 t11 +

q 12 t11 −q 11 t12 +q 12 t12 −q 12 t13 +q 13 t13 +a3 q 6 −q 6 t+q 7 t+q 8 t+q 9 t−q 7 t2 +q 9 t2 +2q 10 t2 −

q 8 t3 + 2q 11 t3 − q 9 t4 + q 11 t4 + q 12 t4 − q 10 t5 + q 12 t5 − q 11 t6 + q 12 t6 − q 12 t7 + q 13 t7 + a2 q 3 +
q 4 + q 5 − q 3 t + q 5 t + 3q 6 t + 2q 7 t + q 8 t − q 4 t2 − q 5 t2 − q 6 t2 + 3q 7 t2 + 4q 8 t2 + 3q 9 t2 − q 5 t3 −
q 6 t3 − 2q 7 t3 + q 8 t3 + 4q 9 t3 + 4q 10 t3 − q 6 t4 − q 7 t4 − 2q 8 t4 + q 9 t4 + 4q 10 t4 + 3q 11 t4 − q 7 t5 −
q 8 t5 − 2q 9 t5 + q 10 t5 + 4q 11 t5 + q 12 t5 − q 8 t6 − q 9 t6 − 2q 10 t6 + 3q 11 t6 + 2q 12 t6 − q 9 t7 − q 10 t7 −

q 11 t7 + 3q 12 t7 − q 10 t8 − q 11 t8 + q 12 t8 + q 13 t8 − q 11 t9 + q 13 t9 − q 12 t10 + q 13 t10 + a q + q 2 +
q − qt + q t + 3q t + 2q t + q t − q t − q t − q t + 2q t + 4q t + 4q t + q t − q 3 t3 −
3 3 4 5 6 2 2 3 2 4 2 5 2 6 2 7 2 8 2
q 4 t3 − 2q 5 t3 + 3q 7 t3 + 6q 8 t3 + 2q 9 t3 − q 4 t4 − q 5 t4 − 2q 6 t4 − q 7 t4 + 2q 8 t4 + 6q 9 t4 + 2q 10 t4 −
q 5 t5 − q 6 t5 − 2q 7 t5 − q 8 t5 + 2q 9 t5 + 6q 10 t5 + q 11 t5 − q 6 t6 − q 7 t6 − 2q 8 t6 − q 9 t6 + 3q 10 t6 +
4q 11 t6 − q 7 t7 − q 8 t7 − 2q 9 t7 + 4q 11 t7 + q 12 t7 − q 8 t8 − q 9 t8 − 2q 10 t8 + 2q 11 t8 + 2q 12 t8 − q 9 t9 −

q 10 t9 − q 11 t9 + 3q 12 t9 − q 10 t10 − q 11 t10 + q 12 t10 + q 13 t10 − q 11 t11 + q 13 t11 − q 12 t12 + q 13 t12 .
The a–degree of H min (q, t, a) is dega = s1 r2 | | + s1 | | − | | = 3 + 1 − 1 = 3.

L

The positivity of HL (q, t, a)/(1 − t) holds. The corresponding singularity is (y 8 +
min
x3 )(y 2 +x) = 0 with lk = 6 and the Alexander polynomial 1+q 4 +q 10 +q 14 +q 20 +q 24 .

Recall that the latter is zeta-monodromy from [DGPS] as t → q.
8.5. Two different paths. Let us extend the previous example. The tree
will be now uncolored Υ = ◦ ⇒◦◦⇒ with the vertices labeled by
[r11 = 2, s11 = 1], [r21 = 3, s12 = 2], and [r12 = 2, s21 = 1], [r22 = 2, s22 = 1].
The first vertices [r1i , si1 ] from the corresponding paths are identified in this tree.
Since we have two different paths, the simplified notations for L from the previous
example cannot be used now; both sets of labels (for both paths) must be shown.

◦ ⇒ ◦ ⇒, ( , )
(8.9) Cab(8, 3)Cab(5, 2) T (2, 1) : L- ◦ .- ., H min
L =
{2,3},{1,2} , {2,2},{1,1}
1−t+qt+q t+q t+q t−qt +q t +2q t +2q t +q 7 t2 +q 8 t2 −q 2 t3 −q 4 t3 +q 6 t3 +2q 7 t3 +

2 3 4 2 4 2 5 2 6 2
3q 8 t3 +3q 9 t3 +q 10 t3 −q 3 t4 −q 5 t4 −q 6 t4 +2q 9 t4 +5q 10 t4 +3q 11 t4 +q 12 t4 −q 4 t5 −q 6 t5 −q 7 t5 −

q 8 t5 − q 9 t5 + 5q 11 t5 + 5q 12 t5 + 2q 13 t5 − q 5 t6 − q 7 t6 − q 8 t6 − q 9 t6 − 2q 10 t6 − 2q 11 t6 + 4q 12 t6 +
6q 13 t6 + 3q 14 t6 − q 6 t7 − q 8 t7 − q 9 t7 − q 10 t7 − 2q 11 t7 − 3q 12 t7 + 3q 13 t7 + 7q 14 t7 + 3q 15 t7 −
q 7 t8 − q 9 t8 − q 10 t8 − q 11 t8 − 2q 12 t8 − 3q 13 t8 + 3q 14 t8 + 6q 15 t8 + 2q 16 t8 − q 8 t9 − q 10 t9 − q 11 t9 −
q 12 t9 − 2q 13 t9 − 3q 14 t9 + 4q 15 t9 + 5q 16 t9 + q 17 t9 − q 9 t10 − q 11 t10 − q 12 t10 − q 13 t10 − 2q 14 t10 −
2q 15 t10 + 5q 16 t10 + 3q 17 t10 − q 10 t11 − q 12 t11 − q 13 t11 − q 14 t11 − 2q 15 t11 + 5q 17 t11 + q 18 t11 −
q 11 t12 − q 13 t12 − q 14 t12 − q 15 t12 − q 16 t12 + 2q 17 t12 + 3q 18 t12 − q 12 t13 − q 14 t13 − q 15 t13 −
q 16 t13 +3q 18 t13 +q 19 t13 −q 13 t14 −q 15 t14 −q 16 t14 +2q 18 t14 +q 19 t14 −q 14 t15 −q 16 t15 −q 17 t15 +
q 18 t15 +2q 19 t15 −q 15 t16 −q 17 t16 +2q 19 t16 −q 16 t17 −q 18 t17 +q 19 t17 +q 20 t17 −q 17 t18 +q 20 t18 −

q 18 t19 + q 20 t19 − q 19 t20 + q 20 t20 − q 20 t21 + q 21 t21 +a4 q 10 − q 10 t + q 11 t + q 12 t + q 13 t + q 14 t −
q 11 t2 + 2q 14 t2 + 2q 15 t2 + q 16 t2 − q 12 t3 − 2q 14 t3 + q 15 t3 + 3q 16 t3 + 2q 17 t3 − q 13 t4 − 2q 15 t4 +
3q 17 t4 + 2q 18 t4 − q 14 t5 − 2q 16 t5 + 3q 18 t5 + q 19 t5 − q 15 t6 − 2q 17 t6 + q 18 t6 + 2q 19 t6 − q 16 t7 −

2q 18 t7 + 2q 19 t7 + q 20 t7 − q 17 t8 + q 20 t8 − q 18 t9 + q 20 t9 − q 19 t10 + q 20 t10 − q 20 t11 + q 21 t11

+a3 q 6 + q 7 + q 8 + q 9 − q 6 t + q 8 t + 2q 9 t + 4q 10 t + 3q 11 t + 2q 12 t + q 13 t − q 7 t2 − q 8 t2 −
q 9 t2 + 4q 11 t2 + 6q 12 t2 + 6q 13 t2 + 4q 14 t2 + q 15 t2 − q 8 t3 − q 9 t3 − 2q 10 t3 − 3q 11 t3 + 4q 13 t3 +
9q 14 t3 + 8q 15 t3 + 3q 16 t3 − q 9 t4 − q 10 t4 − 2q 11 t4 − 4q 12 t4 − 3q 13 t4 + q 14 t4 + 9q 15 t4 + 10q 16 t4 +
4q 17 t4 − q 10 t5 − q 11 t5 − 2q 12 t5 − 4q 13 t5 − 5q 14 t5 − q 15 t5 + 9q 16 t5 + 10q 17 t5 + 3q 18 t5 −
q 11 t6 − q 12 t6 − 2q 13 t6 − 4q 14 t6 − 5q 15 t6 − q 16 t6 + 9q 17 t6 + 8q 18 t6 + q 19 t6 − q 12 t7 − q 13 t7 −
2q 14 t7 − 4q 15 t7 − 5q 16 t7 + q 17 t7 + 9q 18 t7 + 4q 19 t7 − q 13 t8 − q 14 t8 − 2q 15 t8 − 4q 16 t8 − 3q 17 t8 +
4q 18 t8 + 6q 19 t8 + q 20 t8 − q 14 t9 − q 15 t9 − 2q 16 t9 − 4q 17 t9 + 6q 19 t9 + 2q 20 t9 − q 15 t10 − q 16 t10 −
2q 17 t10 − 3q 18 t10 + 4q 19 t10 + 3q 20 t10 − q 16 t11 − q 17 t11 − 2q 18 t11 + 4q 20 t11 − q 17 t12 − q 18 t12 −


+a2 q 3 + q 4 + 2q 5 + q 6 + q 7 − q 3 t + 3q 6 t + 4q 7 t + 5q 8 t + 4q 9 t + 2q 10 t + q 11 t − q 4 t2 − q 5 t2 −
2q 6 t2 +2q 8 t2 +7q 9 t2 +9q 10 t2 +9q 11 t2 +5q 12 t2 +2q 13 t2 −q 5 t3 −q 6 t3 −3q 7 t3 −2q 8 t3 −3q 9 t3 +
2q 10 t3 + 8q 11 t3 + 14q 12 t3 + 12q 13 t3 + 6q 14 t3 + q 15 t3 − q 6 t4 − q 7 t4 − 3q 8 t4 − 3q 9 t4 − 5q 10 t4 −
4q 11 t4 +2q 12 t4 +14q 13 t4 +17q 14 t4 +10q 15 t4 +2q 16 t4 −q 7 t5 −q 8 t5 −3q 9 t5 −3q 10 t5 −6q 11 t5 −
7q 12 t5 −4q 13 t5 +12q 14 t5 +20q 15 t5 +11q 16 t5 +2q 17 t5 −q 8 t6 −q 9 t6 −3q 10 t6 −3q 11 t6 −6q 12 t6 −
8q 13 t6 −7q 14 t6 +11q 15 t6 +20q 16 t6 +10q 17 t6 +q 18 t6 −q 9 t7 −q 10 t7 −3q 11 t7 −3q 12 t7 −6q 13 t7 −
8q 14 t7 − 7q 15 t7 + 12q 16 t7 + 17q 17 t7 + 6q 18 t7 − q 10 t8 − q 11 t8 − 3q 12 t8 − 3q 13 t8 − 6q 14 t8 −
8q 15 t8 − 4q 16 t8 + 14q 17 t8 + 12q 18 t8 + 2q 19 t8 − q 11 t9 − q 12 t9 − 3q 13 t9 − 3q 14 t9 − 6q 15 t9 −
7q 16 t9 + 2q 17 t9 + 14q 18 t9 + 5q 19 t9 − q 12 t10 − q 13 t10 − 3q 14 t10 − 3q 15 t10 − 6q 16 t10 − 4q 17 t10 +
8q 18 t10 + 9q 19 t10 + q 20 t10 − q 13 t11 − q 14 t11 − 3q 15 t11 − 3q 16 t11 − 5q 17 t11 + 2q 18 t11 + 9q 19 t11 +
2q 20 t11 − q 14 t12 − q 15 t12 − 3q 16 t12 − 3q 17 t12 − 3q 18 t12 + 7q 19 t12 + 4q 20 t12 − q 15 t13 − q 16 t13 −
3q 17 t13 − 2q 18 t13 + 2q 19 t13 + 5q 20 t13 − q 16 t14 − q 17 t14 − 3q 18 t14 + 4q 20 t14 + q 21 t14 − q 17 t15 −

q 18 t15 −2q 19 t15 +3q 20 t15 +q 21 t15 −q 18 t16 −q 19 t16 +2q 21 t16 −q 19 t17 +q 21 t17 −q 20 t18 +q 21 t18

+a q + q 2 + q 3 + q 4 − qt + q 3 t + 2q 4 t + 4q 5 t + 3q 6 t + 2q 7 t + q 8 t − q 2 t2 − q 3 t2 − q 4 t2 + 3q 6 t2 +
5q 7 t2 + 7q 8 t2 + 6q 9 t2 + 3q 10 t2 + q 11 t2 − q 3 t3 − q 4 t3 − 2q 5 t3 − 2q 6 t3 + q 8 t3 + 7q 9 t3 + 11q 10 t3 +
9q 11 t3 + 4q 12 t3 + q 13 t3 − q 4 t4 − q 5 t4 − 2q 6 t4 − 3q 7 t4 − 2q 8 t4 − 3q 9 t4 + q 10 t4 + 10q 11 t4 +
14q 12 t4 + 9q 13 t4 + 3q 14 t4 − q 5 t5 − q 6 t5 − 2q 7 t5 − 3q 8 t5 − 3q 9 t5 − 5q 10 t5 − 4q 11 t5 + 6q 12 t5 +
16q 13 t5 + 13q 14 t5 + 4q 15 t5 − q 6 t6 − q 7 t6 − 2q 8 t6 − 3q 9 t6 − 3q 10 t6 − 6q 11 t6 − 7q 12 t6 + 2q 13 t6 +
16q 14 t6 + 15q 15 t6 + 4q 16 t6 − q 7 t7 − q 8 t7 − 2q 9 t7 − 3q 10 t7 − 3q 11 t7 − 6q 12 t7 − 8q 13 t7 + q 14 t7 +
16q 15 t7 +13q 16 t7 +3q 17 t7 −q 8 t8 −q 9 t8 −2q 10 t8 −3q 11 t8 −3q 12 t8 −6q 13 t8 −8q 14 t8 +2q 15 t8 +
16q 16 t8 + 9q 17 t8 + q 18 t8 − q 9 t9 − q 10 t9 − 2q 11 t9 − 3q 12 t9 − 3q 13 t9 − 6q 14 t9 − 7q 15 t9 + 6q 16 t9 +
14q 17 t9 +4q 18 t9 −q 10 t10 −q 11 t10 −2q 12 t10 −3q 13 t10 −3q 14 t10 −6q 15 t10 −4q 16 t10 +10q 17 t10 +
9q 18 t10 + q 19 t10 − q 11 t11 − q 12 t11 − 2q 13 t11 − 3q 14 t11 − 3q 15 t11 − 5q 16 t11 + q 17 t11 + 11q 18 t11 +
3q 19 t11 − q 12 t12 − q 13 t12 − 2q 14 t12 − 3q 15 t12 − 3q 16 t12 − 3q 17 t12 + 7q 18 t12 + 6q 19 t12 − q 13 t13 −
q 14 t13 − 2q 15 t13 − 3q 16 t13 − 2q 17 t13 + q 18 t13 + 7q 19 t13 + q 20 t13 − q 14 t14 − q 15 t14 − 2q 16 t14 −
3q 17 t14 + 5q 19 t14 + 2q 20 t14 − q 15 t15 − q 16 t15 − 2q 17 t15 − 2q 18 t15 + 3q 19 t15 + 3q 20 t15 − q 16 t16 −
q 17 t16 − 2q 18 t16 + 4q 20 t16 − q 17 t17 − q 18 t17 − q 19 t17 + 2q 20 t17 + q 21 t17 − q 18 t18 − q 19 t18 +

q 20 t18 + q 21 t18 − q 19 t19 + q 21 t19 − q 20 t20 + q 21 t20 .
The a–degree of H min (q, t, a) is dega = s1 r1 | | + s2 r2 | | − | | = 3 + 2 − 1 = 4.
L 1 2 1 2

The positivity of HL (q, t, a)/(1 − t) holds.
min
The last one will be uncolored Cab(5, 2)T (4, 2) for Υ = {◦ → ◦ ⇒}. The
[r,s]-presentation is as follows:
1 ≤ j ≤ κ = 2, !rj = {2, 2}, !sj = {1, 1}, Υ = {◦ → ◦ ⇒} , λ1 = = λ2 .
-
Now we. -can the avoid.duplication of notation; {2, 2}, {1, 1} in L will mean ( {2, 2},
{1, 1} , {2, 2}, {1, 1} ) for 2 paths. So the labels are [2, 1] and [2, 1]. The cable is
Cab(5, 2)T (4, 2) for each path.
(8.10) Cab(5, 2)T (4, 2) : L = L{2,2},{1,1}

◦→◦⇒, ( , ) min
, H L (q, t, a) =
1 − t + qt + q 2 t + q 3 t − qt2 + 2q 4 t2 + q 5 t2 + q 6 t2 − q 2 t3 − q 4 t3 + q 5 t3 + q 6 t3 + 2q 7 t3 − q 3 t4 −
q 5 t4 + 3q 8 t4 − q 4 t5 − q 6 t5 − q 8 t5 + 3q 9 t5 − q 5 t6 − q 7 t6 − q 9 t6 + 3q 10 t6 − q 6 t7 − q 8 t7 + 2q 11 t7 −
q 7 t8 − q 9 t8 + q 11 t8 + q 12 t8 − q 8 t9 − q 10 t9 + q 11 t9 + q 12 t9 − q 9 t10 − q 11 t10 + 2q 12 t10 − q 10 t11 +

q 13 t11 − q 11 t12 + q 13 t12 − q 12 t13 + q 13 t13 − q 13 t14 + q 14 t14 + a3 q 6 − q 6 t + q 7 t + q 8 t + q 9 t −
q 7 t2 + 2q 10 t2 − q 8 t3 − q 10 t3 + 2q 11 t3 − q 9 t4 − q 11 t4 + 2q 12 t4 − q 10 t5 + q 13 t5 − q 11 t6 + q 13 t6 −

q 12 t7 + q 13 t7 − q 13 t8 + q 14 t8 + a2 q 3 + q 4 + q 5 − q 3 t + q 5 t + 3q 6 t + 2q 7 t + q 8 t − q 4 t2 − q 5 t2 −
q 6 t2 +2q 7 t2 +3q 8 t2 +3q 9 t2 −q 5 t3 −q 6 t3 −2q 7 t3 +2q 9 t3 +4q 10 t3 −q 6 t4 −q 7 t4 −2q 8 t4 −q 9 t4 +
q 10 t4 + 4q 11 t4 − q 7 t5 − q 8 t5 − 2q 9 t5 − q 10 t5 + 2q 11 t5 + 3q 12 t5 − q 8 t6 − q 9 t6 − 2q 10 t6 + 3q 12 t6 +
q 13 t6 −q 9 t7 −q 10 t7 −2q 11 t7 +2q 12 t7 +2q 13 t7 −q 10 t8 −q 11 t8 −q 12 t8 +3q 13 t8 −q 11 t9 −q 12 t9 +

q 13 t9 + q 14 t9 − q 12 t10 + q 14 t10 − q 13 t11 + q 14 t11 + a q + q 2 + q 3 − qt + q 3 t + 3q 4 t + 2q 5 t + q 6 t −
q 2 t2 − q 3 t2 − q 4 t2 + 2q 5 t2 + 3q 6 t2 + 4q 7 t2 + q 8 t2 − q 3 t3 − q 4 t3 − 2q 5 t3 + q 7 t3 + 5q 8 t3 + 2q 9 t3 −
q 4 t4 −q 5 t4 −2q 6 t4 −q 7 t4 −q 8 t4 +5q 9 t4 +2q 10 t4 −q 5 t5 −q 6 t5 −2q 7 t5 −q 8 t5 −2q 9 t5 +5q 10 t5 +
2q 11 t5 −q 6 t6 −q 7 t6 −2q 8 t6 −q 9 t6 −q 10 t6 +5q 11 t6 +q 12 t6 −q 7 t7 −q 8 t7 −2q 9 t7 −q 10 t7 +q 11 t7 +
4q 12 t7 −q 8 t8 −q 9 t8 −2q 10 t8 +3q 12 t8 +q 13 t8 −q 9 t9 −q 10 t9 −2q 11 t9 +2q 12 t9 +2q 13 t9 −q 10 t10 −

q 11 t10 −q 12 t10 +3q 13 t10 −q 11 t11 −q 12 t11 +q 13 t11 +q 14 t11 −q 12 t12 +q 14 t12 −q 13 t13 +q 14 t13 .
The a–degree of H min (q, t, a) is s1 r2 (| |+| |)−| | = 3. This polynomial is self-

L
dual; all uncolored ones are self-dual. The positivity of the series H min (q, t, a)/(1 −
L
κ−1
t) holds here and in further (uncolored) examples; recall that κ is the number
of paths (components of the link), which is 2 in the considered case.
The plane curve singularity for (8.10) is
- .
C = {(y 5 + x2 )(y 5 − x2 ) = 0}, Link(C) = T (5, 2), T (5, 2), lk = 10 .
The knots here (and the corresponding branches of this singularity) and the link-
ing number lk between them (equivalently, the intersection multiplicity) uniquely
determine the germ of this curve. The Alexander polynomial is 1 + q 4 + q 8 + q 10 +
min (q, q, −1)/(1 − q)2 .
q 12 + q 14 + q 16 + q 18 + q 22 + q 26 , which coincides with H L
9. Generalized twisted union

Continuing Section 7.5, we will provide further examples of the pairs of trees
{L, L} and {L, L∨ }. They will be mostly for the uncolored unknot taken as L,
i.e. for
L = L◦→,

1,0 .
We will simply add “prime” to the corresponding knot/link L (followed by ∨ if
it is present), to indicate using this very L. Recall that H min
L, L∨ is obtained by the
substitution X → Y instead of X → Y −1
in HL, L . Topologically, this is adding
min
the meridian to L. Note that all examples of Hopf links from Section 7 can be
restated in terms of such pairs due to (3.34).
9.1. Uncolored trefoil-prime. The [r,s]-presentation for such L is as follows:

1 ≤ j ≤ κ = 2, !rj = 3, !sj = 2, Υ = {◦ →} , λ1 = ;
(9.1) ◦→,
T (3, 2) : L = L{3,2} min
, H L, L (q, t, a) =
q2 q3 q3
q2

1 + a2 q 2 − t2
+ t2
+ q
t
− t
+ qt + a 2q + 1
t
− t
+ q2 t .
Adding ∨. The positivity of the series H min

L, L (q, t, a)/(1 − t) from (5.56) in
Conjecture 5.3 fails (and for any power of (1 − t)), but holds for the ∨–variant of
this superpolynomial (as conjectured):
(9.2) T (3, 2),∨ : L = L{3,2}

◦→, min
, H L, L∨ (q, t, a) =

1 − t + qt + q 2 t − qt2 + q 3 t2 − q 2 t3 + q 3 t3 − q 3 t4 + q 4 t4 + a2 q 3 − q 3 t + q 4 t + a q + q 2 −

qt + 2q 3 t − q 2 t2 + q 4 t2 − q 3 t3 + q 4 t3 .
The a–degree of H min
L, L∨ (q, t, a) is given by the formula dega = max{s, 1}| | +

max{ s, 1}| | − | | = 2 + 1 − 1 = 2. It is the same as in the previous example.
Both polynomials result at q = 1 in (1 + a)(1 + a + t); the second factor is the
specialization at q = 1 for the uncolored trefoil.
The series H min
L, L∨ (q, t, a)/(1 − t) is positive; the corresponding germ of the
singularity is
- .
C = {(y 3 + x2 )(y 3 + x) = 0}, Link(C) = T (3, 2), T (3, 1), lk = 3 ,
where T (3, 1) = , and the Alexander polynomial is 1 + q 3 + q 6 .
We note that the superpolynomial for (9.2) does not coincide with

→
(9.3)
◦→◦→, ( , )
Cab(3, 2)Cab(1, 1) T (1, 1) : L = L{1,1},{1,2} min
, H L (q, t, a) =
1 + a2 q 3 − t + qt + q 2 t − qt2 + q 2 t2 − q 2 t3 + q 3 t3 + a(q + q 2 − qt + q 2 t + q 3 t − q 2 t2 + q 3 t2 ).
The corresponding singularity is C = {(y 3 + x2 )(y + x) = 0} with the same link
components {T (3, 2), } as above but with lk = 2 and the Alexander polynomial
1 + q 4 . The self-duality and other claims in Theorem 5.1 hold in these cases.
Splice interpretation. Let k ∈ Z. Recall that formula (3.39) provides the coinci-
dence of H min from (9.2) with that for

◦ ⇒ ◦ ⇒, ( , )
(9.4) Cab(2, 3)Cab(k,−1) T (1,−1) : L- ◦ .- ..
{1,3},{−1,5} , {1,0},{−1,k−1}
Explicitly, the following DAHA-identity establishes this coincidence:

- −1
1−k 2 . - .
τ− τ− σ (P ) ⇓ τ− τ+ τ− 2
(P ) ⇓ ev = q • t• P ι ⇓ , τ+ τ−
2
(P ) ⇓ ev ,
where we use the commutativity of τ− with the projection H⇓= H(1) of H ∈ H H
onto V, i.e. the fundamental fact that τ− acts in V; this action was denoted by τ̇−
in (3.39). Recall that P ι = P (X −1 ), L is ◦ →, and the colors are trivial (shown by
in the diagrams below).
Let us translate the latter identity into the language of splice diagrams. The
corresponding diagrams for k = 1 are:
◦
31
' -
[3, 2] 2 1
◦ →
◦
◦ → ;
11
[1, 0] & -
(A) 0 1
◦
31
◦ -
[3, 5]
1
1 1 2 1
[1, −1] ◦ ⇒◦◦ ⇒ ◦ ◦
[−1, 0] −1 H 1 HH−11 .
H -
(B) 1 1
Recall that the colors are assigned to the arrowheads; also we will use ∨ for
the change of orientation of the corresponding component. Generally, any leaves
of weight 1 can be deleted without changing the link (see here and below [EN],
Theorem 8.1, Statements 2,3). The situation with leaves of weight −1 is somewhat
more involved. Namely, one can change weights of two edges from the same node
to their negatives together with the simultaneous change of orientations of all com-
ponents starting with these edges. Starting with the diagram (B), this results in
the following (isotopic) equivalence ≈≈ :
◦ ◦
31 31
◦ - ◦ -

1
1 1 2 1 1
1 1 2 1
◦ ◦ ≈≈ ◦ ◦
H
−1 1 HH−1 1 H HH−1
H1 - −1 H1 - ∨
1 1 1 1
◦
31
◦ - ◦

1 1 2 1
1 31
≈≈ ◦ ◦ ≈≈ -
HH .
1 1 2 1
−1 HH1 -
1 −1
(9.5)
At the last step here, we used Statement 3 of Theorem 8.1 from [EN] three
times. Applying it one more time, we conclude that the diagram/link in (A) is
equivalent to that in (B).
9.2. Colored/iterated examples. Let us begin with the trefoil-prime for

the 2-row:
(9.6) T (3, 2) ◦→,
: L = L{3,2} min
, H L, L (q, t, a) =
6 8 5 6 7 8 4 6
1+a3 q 7 − qt2 + qt2 + qt + qt − qt − qt +q 2 t+q 3 t+q 4 t2 +a2 q 4 +3q 5 +q 6 −q 7 −q 8 − qt2 + qt2 +
q2 3 4 5 6 8 4
t
+ qt + qt − qt − 3qt + qt +q 6 t+q 7 t +a 2q 2 +2q 3 +q 4 −q 5 −q 6 + 1t − qt +2q 4 t+2q 5 t+q 6 t2 .
It is super-dual to the superpolynomial for . The positivity of the series

H min (q, t, a)/(1 − t) p
fails for p = 1 and any p > 1 (due to the absence of ∨). At
L, L
q = 1, it equals (1 + a)(1 + a + t)2 , which also coincides with the specialization for
the corresponding ∨–polynomial:
(9.7) T (3, 2),∨ : L = L{3,2}

◦→, min
, H L, L∨ (q, t, a) =
1 − t + q 2 t + q 3 t + q 4 t − q 2 t2 − q 3 t2 + q 5 t2 + 2q 6 t2 − q 4 t3 − q 5 t3 + q 7 t3 + q 8 t3 − q 6 t4 − q 7 t4 +

q 8 t4 + q 9 t4 − q 8 t5 + q 10 t5 + a3 q 9 − q 9 t + q 11 t + a2 q 5 + q 6 + q 7 − q 5 t − q 6 t + 2q 8 t + 2q 9 t −
2 3 4
q t −q t +q t +q t −q t +q t +a q +q +q −q 2 t−q 3 t+2q 5 t+3q 6 t+q 7 t−q 4 t2 −
7 2 8 2 10 2 11 2 9 3 11 3

2q 5 t2 − q 6 t2 + q 7 t2 + 3q 8 t2 + q 9 t2 − q 6 t3 − 2q 7 t3 + 2q 9 t3 + q 10 t3 − q 8 t4 − q 9 t4 + q 10 t4 + q 11 t4 .
The a–degree of H min |+

L, L∨ (q, t, a) is given by the formula dega = max{s, 1}|
max{ s, 1}| |−| | = 4+1−2 = 3. See ((5.40) and (5.41). This polynomial is dual
to the superpolynomial for . The positivity of the series H min
L, L∨ (q, t, a)/(1 − t)
holds.
Iterated T(2,1)-prime. Let us replace the additional arrowhead in the example
from (8.8) by the prime-construction (twisted union with one box) for the starting
!r = {2, 3},!s = {1, 2}. Recall that all prime examples are for L = L◦→,
1,0 .
(9.8) ◦→◦→,
Cab(8, 3)(T (2, 1)) : L = L{2,3},{1,2} min
, H L, L (q, t, a) =
1 + qt + q 2 t + q 2 t2 + q 3 t2 + q 4 t2 + q 3 t3 + q 4 t3 + q 5 t3 + q 4 t4 + q 5 t4 + q 5 t5 + q 6 t5 + q 6 t6 +
6 8 6 7 8 9
q 7 t7 + a4 q 7 − q 8 − qt2 + qt2 + qt − qt − qt + qt + q 8 t − q 9 t + a3 q 4 + 2q 5 + 2q 6 − 4q 7 − 2q 8 +
q4 5 7
q3 q4 6 7
3q 8
2q 9 − t2
− qt2 + 2q
t2
+ + − 3qt − 2qt + + q 5 t + 2q 6 t + 2q 7 t − 3q 8 t + q 6 t2 + 2q 7 t2 −
t t
2 t
3 6 2
q 9 t2 + q t + q t − q t + q t
7 3 8 3 9 3 8 4
+ a q + 3q + 3q 4 − 4q 6 − 2q 7 + 2q 8 − qt2 + qt2 + qt + qt +
2 3
3 4 5 6 7
q
− 2qt − 2qt − qt + 2qt + q 3 t + 3q 4 t + 5q 5 t + q 6 t − 4q 7 t − q 8 t + q 9 t + q 4 t2 + 3q 5 t2 + 5q 6 t2 −
t
2q t + q 5 t3 + 3q 6 t3 + 3q 7 t3 − 2q 8 t3 + q 6 t4 + 3q 7 t4 + q 8 t4 − q 9 t4 + q 7 t5 + q 8 t5 + q 8 t6 +
8 2
3
a 2q + 2q 2 + q 3 − q 4 − q 5 + 1t − qt + 2q 2 t + 3q 3 t + 3q 4 t − 2q 6 t + 2q 3 t2 + 3q 4 t2 + 4q 5 t2 − q 7 t2 +

2q 4 t3 + 3q 5 t3 + 3q 6 t3 − q 7 t3 + 2q 5 t4 + 3q 6 t4 + q 7 t4 − q 8 t4 + 2q 6 t5 + 2q 7 t5 + 2q 7 t6 + q 8 t7 .
It is self-dual; the positivity of the series H min
L, L (q, t, a)/(1 − t) fails for any
p
p ≥ 0, which failure is generally expected without ∨. As in (7.34), the a–degree of

H min
L, L (q, t, a), which is 4, is greater than dega = s1 r2 | | + max{ s, 1}| | − | | =
3 + 1 − 1 = 3.
The ∨–version is generally with X replaced by Y in the pre-polynomial Ptot

0
for L. This pre-polynomial is J (up to a renormalization). We will omit the
corresponding
(9.9) H min
L, L∨ (q, t, a) for Cab(8, 3)(T (2, 1))
,∨
: L = L ◦→◦→, . {2,3},{1,2}
min (q, t, a) for the polynomial from (8.8). The latter was defined
It coincides with H M
(and calculated) for

→
◦→◦→, ( , )
(9.10) Cab(8, 3)Cab(0, 1) T (2, 1) : M = L{2,3},{1,2} .
Let us demonstrate this coincidence theoretically. Setting P = P /{P }ev , P ι =
ι(P ), Q = τ+ τ−2
(P ) ⇓= τ+ τ−
2
(P )(1) ∈ V,

−1 ι 2
τ+ τ− (P Q) = τ+ τ− ϕσ (P )Q = ϕτ− (P ι )τ+ τ− (Q),

2 ι

{τ+ τ− (P Q)}ev = {ϕ τ̇− (P ) τ+ τ− (Q) ⇓ }ev {P ι , τ+ τ− (Q) ⇓}ev ,
where τ̇− is the action of τ− in V and is the equality up to q • t• . Modulo this
min
equivalence, the last quantity coincides with H ι
L, L∨ . Recall that P can be replaced
by Eωn /{Eωn }ev here and the E–polynomials are τ̇− –invariant up to ; see (1.37)
and also (7.38).
Continuing with (9.8),(9.9), let us provide their values at q = 1:
min
H min
L, L (q = 1, t, a) = HL, L∨ (q = 1, t, a) =

1 + a 1 + 2t + 3t2 + 3t3 + 2t4 + 2t5 + t6 + t7

+a2 (1 + 2t+ 2t2 + t3 + t4 ) + a(2 + 4t + 5t2 + 4t3 + 3t4 + 2t5 + t6 )

min
= 1+a H L (q = 1, t, a).
Recall that factor (1 + a) here is due to our using J instead of spherical P ◦
(which would give 1 for L = ); the division by the LCM of the evaluations of
J–polynomials in Hmin is only by one J (tρ ) here.
The second factor is relatively long. It is the evaluation at q = 1 of H 8,3
min
( )
for the torus knot T (8, 3). Indeed, Cab(8, 3)(T (2, 1)) is isotopic to T (8, 3) or T (3, 8),
where the latter presentation is somewhat more convenient practically; the corre-
sponding γ3,8 corresponds to τ− 2 2
τ+ τ− . For the sake of completeness, H 8,3
min
( )=
1 + qt + q 2 t + q 2 t2 + q 3 t2 + q 4 t2 + q 3 t3 + q 4 t3 + q 5 t3 + q 4 t4 + q 5 t4 + q 5 t5 + q 6 t5 + q 6 t6 +

q 7 t7 + a2 q 3 + q 4 t + q 5 t + q 5 t2 + q 6 t2 + q 6 t3 + q 7 t4 + a q + q 2 + q 2 t + 2q 3 t + q 4 t + q 3 t2 +

2q 4 t2 + 2q 5 t2 + q 4 t3 + 2q 5 t3 + q 6 t3 + q 5 t4 + 2q 6 t4 + q 6 t5 + q 7 t5 + q 7 t6 .
Double-prime T(2,1). Let us provide at least one example of twisted union with

L colored by . We will use double-prime then:
(9.11) T (4, 2),∨, : L = L2,1

◦⇒, ( , )
, L = L◦→, min
, H L, L∨ (q, t, a) =
1,0
1 − t + q 2 t − t2 − q 2 t2 + q 3 t2 + q 4 t2 + t3 − q 2 t3 − q 3 t3 + q 5 t3 + q 2 t4 − q 3 t4 − 2q 4 t4 + 2q 6 t4 +
q 3 t5 −2q 5 t5 +q 7 t5 +q 4 t6 −2q 6 t6 +q 8 t6 +q 5 t7 −q 6 t7 −q 7 t7 +q 8 t7 +q 6 t8 −q 7 t8 −q 8 t8 +q 9 t8 +

q 7 t10 − q 8 t10 − q 9 t10 + q 10 t10 + a2 q 5 − q 5 t + q 7 t − q 5 t2 + q 8 t2 + q 5 t3 − 2q 7 t3 + q 9 t3 + q 7 t5 −

q 8 t5 − q 9 t5 + q 10 t5 + a q 2 + q 3 − q 2 t − q 3 t + q 4 t + q 5 t − q 2 t2 − q 3 t2 − q 4 t2 + q 5 t2 + 2q 6 t2 +
q 2 t3 + q 3 t3 − q 4 t3 − 3q 5 t3 + 2q 7 t3 + q 4 t4 − q 5 t4 − 2q 6 t4 + 2q 8 t4 + 2q 5 t5 − q 6 t5 − 3q 7 t5 + q 8 t5 +

q 9 t5 + q 6 t6 − q 7 t6 − q 8 t6 + q 9 t6 + q 7 t7 − q 8 t7 − q 9 t7 + q 10 t7 + q 7 t8 − q 8 t8 − q 9 t8 + q 10 t8 .
The a–degree is 2 = | |+| |+| |−| |. The positivity of H min 2
L, L (q, t, a)/(1−t)
−1
holds (as for all uncolored diagrams). The super-duality (q ↔ t ) is true.
9.3. Generalized twisting. The following generalization of twisted unions

{L, L} and {L, L∨ } is very natural algebraically.

Recall that the H–polynomials for the standard twisted union are essentially
{P (Y )Q(X)}ev for the pre-polynomials P = P
−1 tot and Q = P tot from (4.22)
0 0

defined via the graphs L and L.
The division by the evaluations is used here; it can be for jo as in (4.22) or the
one used in the construction of H min from Theorem 4.3, where the division is by
the LCM of the evaluations of all J–polynomials involved in both, L and L.
When Y −1 is changed here to Y , we add ∨ to L. The twisting here is the
application σ ±1 : X → Y ∓1 to P (corresponding to L). Let us replace σ ±1 by
ξ for an arbitrary ξ ∈ P SL2 (Z). Let (α, β)tr be the first column of ξ. Here and
below we will omit hat in ξ (which means the standard lift of ξ to an automorphism
of H H).
Given ξ ∈ P SL2 (Z) and graphs L, L (labeled, with arrowheads colored by

Young diagrams), we define ξH min
L, L (q, t, a) as {ξ P (X) Q(X)}ev in the normal-
ization of Theorem 4.3. It is a polynomial in terms of q, t±1 , a depending only on
{α, β}. We say that L is ξ–twisted by L. Note that ξH min (q, t, a) = ξH min (q, t, a) =
∅, L L
ξ min
HL, ∅ (q, t, a).
This construction can be reduced to the σ −1 –twisted union. Namely,
ξ min L, M = [β, α] ·→

(9.12) HL, L (q, t, a) =H min
M∨ (q, t, a) for · · L,
→
i.e. for M obtained from L by adding [β, α] as the first vertex (connected with
all paths). If α, β > 0 and αs1 > βr1 or β = 1, α = 0 and s1 s1 > r1 r1 , then
we call {ξ, L, L∨ } positive if the trees L, L are positive. This is sufficient for the
corresponding link to be algebraic (necessary if L, L are reduced [EN]). Note that
we obtain {L, L∨ } for β = 1, α = 0. Also, if ∨ is omitted, then [β, α] in (9.12)
must be changed to [−β, −α] and the resulting link becomes non-algebraic.
To
justify

(9.12), let ξ = ϕ(ξ) = ϕ ξϕ, which is the conjugation of ξ by the
matrix 1 0 . Then for any P, Q ∈ V W and P ι = ι(P ),
0 1

−1 ι
−1 ι
(9.13)
ξ(P )(Q) = ϕ ξσ
(P ) (Q), {ξ(P ) Q}ev = {ϕ ξσ (P ) Q}ev ,
−1 (P ι ), Q}ev = {Q, ξσ
= {ξσ −1 (P ι ) ⇓}ev = {ξσ
−1 (P ι ) ⇓, Q}ev ,
−1 is (β, α)tr . This gives the required.

where the first column of ξσ
Thus we can use Theorem 5.1. For instance, the super-duality from (5.39) holds
Upon the switch to the spherical P ◦ , the product formula (4.34) for
for such ξH. λ
the specializations at q = 1 is also true (in the case of An ), where we add [β, α] as
the first vertex to (all paths in) L. Upon such a modification of L, the estimate
(5.40) for dega holds too. Also, the positivity claim from Part (ii) of Conjecture
5.3 can be extended to positive {ξ, L, L∨ }.
Let us combine (9.13) with Theorem 3.9. Recall that τ̇− (P ) = τ− (P ) ⇓ =
def −1 −1

τ− (P ) (1) ∈ V W . For any P, Q ∈ V W and for ζ == τ− ξσ = α−β β ∗
∗
,

(9.14) τ̇− (P ι Q) = τ− ϕ σ −1 (P ) Q ⇓ = ϕ(τ− (P )) τ̇− (Q) ,

−1
τ̇− ζ(P ι ) ⇓ τ̇− (Q) = ϕ ϕ ξ(P ) ⇓ (Q),
/
−1 ι 0
(9.15) τ̇− ζ(P ) ⇓ τ̇− (Q ) = {ξ(P )(Q)}ev .
ev
∅ ∅
Therefore ξ(P )(ι(Q )) serves as a pre-polynomail for the tree

[1, 1] ⇒ [β, α−β] ·→
· · L, [1, −1] ·→

(9.16) → · · L , P ∅ ∼ L, Q∅ ∼ L,
→
where ∼ means that P ∅ , Q∅ are pre-polynomials for L, L. However, it does not
coincide with the standard pre-polynomial associated with this tree in (4.22) (and
through this work).
9.4. Some examples. The first will be for ξ with the first column (3, 2)tr .
2
We will write ξ = γ3,2 then; τ+ τ− will do. We take two unknots L, L colored by
, ; the second is where ξ is applied.
ξ min
(9.17) ξ = γ3,2 , L = → , L = → , HL, L (q, t, a) =
4 7
1 + a t3q + qt + qt2 − t3 + q 2 t3 + q 3 t3 − qt4 + q 2 t4 + q 3 t4 − qt5 + q 3 t5 + q 4 t5 − q 2 t6 + 2q 4 t6 −
q 2 t7 − q 3 t7 + q 4 t7 + q 5 t7 − q 3 t8 + q 5 t8 − q 3 t9 + 2q 5 t9 − q 4 t10 + q 6 t10 − q 4 t11 + q 6 t11 − q 4 t12 +
4 5 6 6
q 6 t12 − q 5 t15 + q 7 t15 + a3 −q 4 + q 6 + q 7 + qt3 + qt3 + qt2 + qt + q 7 t + q 7 t2 − q 5 t3 + q 7 t3 +
2 3 4 3 4
a2 −q 2 + 2q 4 + 2q 5 + qt3 + qt2 + qt2 + qt + qt − q 3 t + q 4 t + 2q 5 t + q 6 t − q 3 t2 + 2q 5 t2 + 2q 6 t2 −
2q t + 3q t − 2q t + 2q t + q t − q t + q 6 t5 + q 7 t5 − q 5 t6 + 2q 7 t6 − q 5 t7 + q 7 t7 − q 5 t8 +
4 3 6 3 4 4 6 4 7 4 4 5

q 7 t8 +a 2q 2 +q 3 + tq2 + qt −qt+q 2 t+2q 3 t+q 4 t−qt2 +2q 3 t2 +2q 4 t2 −2q 2 t3 +3q 4 t3 +q 5 t3 −
q t − 2q 3 t4 + 2q 4 t4 + 3q 5 t4 − 2q 3 t5 + 3q 5 t5 + q 6 t5 − q 3 t6 − q 4 t6 + 2q 5 t6 + 2q 6 t6 − 3q 4 t7 +
2 4

3q 6 t7 − 2q 4 t8 + 2q 6 t8 − q 4 t9 + q 6 t9 + q 7 t9 − q 5 t10 + q 7 t10 − q 5 t11 + q 7 t11 − q 5 t12 + q 7 t12 .
The positivity of the series ξH min
L, L (q, t, a)/(1 − t) holds. This polynomial is
ξ min
super-dual to the polynomial H for the pair of diagrams , . Using the
evaluation formula at q = 1 for the latter and the q ↔ 1/t-duality, we obtain that
ξ min
HL, L (q, t = 1, a) = (1 + a)(1 + a + aq)3 , which is true indeed for the polynomial
in (9.17). The value ξH min
L, L (q = 1, t, a) is (1 + a) times the evaluation at q = 1 for
the trefoil and ω3 ; the latter is irreducible.
The next example will be for the uncolored trefoil taken as L.
(9.18) ξ = γ3,2 , L = L◦→, , L = L◦→, , ξH L,
min
L (q, t, a) =
{3,2} {1,0}
1−t+q 2 t+q 3 t+q 4 t+q 5 t−q 2 t2 −q 3 t2 +2q 6 t2 +2q 7 t2 +q 8 t2 −q 4 t3 −q 5 t3 −q 6 t3 −q 7 t3 +q 8 t3 +

3q 9 t3 +q 10 t3 −q 6 t4 −q 7 t4 −q 8 t4 −2q 9 t4 +q 10 t4 +3q 11 t4 +q 12 t4 −q 8 t5 −q 9 t5 −q 10 t5 −2q 11 t5 +
2q 12 t5 +3q 13 t5 −q 10 t6 −q 11 t6 −q 12 t6 +2q 14 t6 +q 15 t6 −q 12 t7 −2q 13 t7 +q 14 t7 +2q 15 t7 −2q 14 t8 +

2q 16 t8 − q 15 t9 + q 17 t9 + a4 q 14 − q 14 t + q 16 t − q 16 t2 + q 18 t2 + a3 q 9 + q 10 + q 11 + q 12 − q 9 t −
q t + q t + 2q t + 2q t − q t − 2q t − q t + 2q t + 2q t − q 13 t3 − 2q 14 t3 + q 16 t3 +
10 12 13 14 11 2 12 2 13 2 15 2 16 2

q 17 t3 +q 18 t3 −q 15 t4 −q 16 t4 +q 17 t4 +q 18 t4 −q 16 t5 +q 18 t5 +a2 q 5 +q 6 +2q 7 +q 8 +q 9 −q 5 t−
q t − q t + q t + 3q t + 4q t + 3q t + q t − q t − 2q t − 3q t − 2q 10 t2 + 2q 11 t2 + 4q 12 t2 +
6 7 8 9 10 11 12 7 2 8 2 9 2
4q 13 t2 + q 14 t2 − q 9 t3 − 2q 10 t3 − 4q 11 t3 − 3q 12 t3 + 2q 13 t3 + 4q 14 t3 + 3q 15 t3 + q 16 t3 − q 11 t4 −
2q 12 t4 − 4q 13 t4 − q 14 t4 + 4q 15 t4 + 3q 16 t4 + q 17 t4 − q 13 t5 − 3q 14 t5 − q 15 t5 + 2q 16 t5 + 2q 17 t5 +

q 18 t5 − 2q 15 t6 − q 16 t6 + 2q 17 t6 + q 18 t6 − q 16 t7 + q 18 t7 + a q 2 + q 3 + q 4 + q 5 − q 2 t − q 3 t + q 5 t +
3q 6 t+4q 7 t+2q 8 t+q 9 t−q 4 t2 −2q 5 t2 −2q 6 t2 −2q 7 t2 +2q 8 t2 +5q 9 t2 +4q 10 t2 +2q 11 t2 −q 6 t3 −
2q 7 t3 −3q 8 t3 −4q 9 t3 +5q 11 t3 +4q 12 t3 +2q 13 t3 −q 8 t4 −2q 9 t4 −3q 10 t4 −5q 11 t4 +q 12 t4 +6q 13 t4 +
3q 14 t4 + q 15 t4 − q 10 t5 − 2q 11 t5 − 3q 12 t5 − 3q 13 t5 + 3q 14 t5 + 5q 15 t5 + q 16 t5 − q 12 t6 − 3q 13 t6 −

2q 14 t6 +2q 15 t6 +3q 16 t6 +q 17 t6 −2q 14 t7 −2q 15 t7 +2q 16 t7 +2q 17 t7 −q 15 t8 −q 16 t8 +q 17 t8 +q 18 t8 .
The super-duality is with the ξH min –polynomial for the diagrams , . The
ξ min min
positivity is for HL, L (q, t, a)/(1 − t). One has: ξH 3
L, L (1, t, a) = (1 + a)(1 + a + t) .
◦
Concerning the latter product, the switch to polynomials Pλ in Part (iv) of Theorem
4.3 means the division of the latter product by (1 + a), which corresponds to the
intersection of , . Since the tree L coincides with L extended by [3, 2], and
= 2ω1 , we indeed arrive at (1 + a + t)3 .

Examples of iterated type. It will be ξH = {ξ γ (J )⇓ J ◦ }ev for γ = γ2,1 upon
the hat-normalization. Recall that Jλ◦ = Pλ◦ = Pλ /Pλ (tρ ).
ξ min
(9.19) ξ = γ3,2 , L = L◦→, ◦→,
{1,0} , L = L{2,1} , HL, L (q, t, a) =
1 − t + qt + q 2 t + q 3 t + q 4 t − qt2 + 2q 4 t2 + 2q 5 t2 + q 6 t2 − q 2 t3 − q 4 t3 + q 5 t3 + 3q 6 t3 + 2q 7 t3 −
q 3 t4 − q 5 t4 + 3q 7 t4 + 3q 8 t4 − q 4 t5 − q 6 t5 + 3q 8 t5 + 2q 9 t5 − q 5 t6 − q 7 t6 + 3q 9 t6 + q 10 t6 − q 6 t7 −
q 8 t7 + q 9 t7 + 2q 10 t7 − q 7 t8 − q 9 t8 + 2q 10 t8 + q 11 t8 − q 8 t9 + q 11 t9 − q 9 t10 + q 11 t10 − q 10 t11 +

q 11 t11 −q 11 t12 +q 12 t12 +a4 q 10 +q 11 t+q 12 t2 +a3 q 6 +q 7 +q 8 +q 9 −q 6 t+q 7 t+2q 8 t+2q 9 t+
2q t − q t + 2q t + 3q t + 2q t − q t + 2q t + 2q 11 t3 + q 12 t3 − q 9 t4 + 2q 11 t4 + q 12 t4 −
10 7 2 9 2 10 2 11 2 8 3 10 3

q 10 t5 + q 11 t5 + q 12 t5 − q 11 t6 + q 12 t6 + a2 q 3 + q 4 + 2q 5 + q 6 + q 7 − q 3 t + q 5 t + 4q 6 t + 4q 7 t +
3q t + q t − q t − q t − q t + 4q t + 6q 8 t2 + 5q 9 t2 + q 10 t2 − q 5 t3 − q 6 t3 − 2q 7 t3 + 3q 8 t3 +
8 9 4 2 5 2 6 2 7 2
7q 9 t3 + 5q 10 t3 + q 11 t3 − q 6 t4 − q 7 t4 − 2q 8 t4 + 3q 9 t4 + 6q 10 t4 + 3q 11 t4 − q 7 t5 − q 8 t5 − 2q 9 t5 +
4q 10 t5 + 4q 11 t5 + q 12 t5 − q 8 t6 − q 9 t6 − q 10 t6 + 4q 11 t6 + q 12 t6 − q 9 t7 − q 10 t7 + q 11 t7 + 2q 12 t7 −

q 10 t8 + q 12 t8 − q 11 t9 + q 12 t9 + a q + q 2 + q 3 + q 4 − qt + q 3 t + 3q 4 t + 4q 5 t + 2q 6 t + q 7 t − q 2 t2 −
q t −q t +2q t +6q t +5q t +2q 8 t2 −q 3 t3 −q 4 t3 −2q 5 t3 +6q 7 t3 +7q 8 t3 +3q 9 t3 −q 4 t4 −
3 2 4 2 5 2 6 2 7 2
q 5 t4 − 2q 6 t4 − q 7 t4 + 6q 8 t4 + 7q 9 t4 + 2q 10 t4 − q 5 t5 − q 6 t5 − 2q 7 t5 − q 8 t5 + 6q 9 t5 + 5q 10 t5 +
q 11 t5 − q 6 t6 − q 7 t6 − 2q 8 t6 + 6q 10 t6 + 2q 11 t6 − q 7 t7 − q 8 t7 − 2q 9 t7 + 2q 10 t7 + 4q 11 t7 − q 8 t8 −

q 9 t8 − q 10 t8 + 3q 11 t8 + q 12 t8 − q 9 t9 − q 10 t9 + q 11 t9 + q 12 t9 − q 10 t10 + q 12 t10 − q 11 t11 + q 12 t11 .
This polynomial is self-dual. The positivity of ξH min

L, L (q, t, a)/(1 − t) holds. One
ξ min
has: HL, L (q = 1, t, a) = (1 + a) 1 + 3a + 3a + a + 3t + 7at + 5a2 t + a3 t + 4t2 + 8at2 +
2 3

5a2 t2 +a3 t2 +4t3 +8at3 +4a2 t3 +4t4 +6at4 +2a2 t4 +3t5 +4at5 +a2 t5 +2t6 +2at6 +t7 +at7 +t8 ,
where the second factor is that for Cab(13, 2)T (3, 2), which appears when making
L empty.
min
Let us provide the identity from (9.12) in this case. The polynomial ξH L, L (q, t, a)
from (9.19) coincides with
min
H ◦→ ◦→◦→
(9.20) ∨ (q, t, a) for L = L{1,0} , L = L{{2,2},{3,1}} .
L, L
The latter corresponds to Cab(13, 2)T (2, 3) for [2, 3] → [2, 1] → . Note [2, 3] here.
min –polynomial due to applying P (Y )
The label [ 3, 2 ] would result in a different H
to the corresponding pre-polynomial.
Making now ξ = γ1,1 , say taking ξ = τ− , we obtain:
ξ min
(9.21) ξ = γ1,1 , L = L◦→, ◦→,
{1,0} , L = L{2,1} , HL, L (q, t, a) = 1 + a2 q 3

−t + qt + q 2 t − qt2 + q 2 t2 − q 2 t3 + q 3 t3 + a q + q 2 − qt + q 2 t + q 3 t − q 2 t2 + q 3 t2 ,
which is self-dual. The positivity of ξH min

L, L (q, t, a)/(1 − t) holds. One has:
ξ min
HL, L (1, t, a) = (1+a)(1+a+t). Taking empty L here will result in Cab(3, 2)T (1, 1),
i.e. in the trefoil; this explains the factor (1 + a + t).
This polynomial can be obtained by our usual construction. Namely, we have
the following DAHA identities:
min
H =H min min
[2,3]→ , [1,0]→ ∨ = H[1,1]→[2,1]→ .
τ−
[1,0]→ , [2,1]→

Recall that the first polynomial is {τ− τ+ τ− (J )⇓ J ◦ }ev upon the hat-normalization.
Up to q • t• , the remaining two correspond to:

{J (Y ) τ− τ+ τ− (J ◦ ) ⇓ }ev = {J (Y ) τ− τ+ τ− (J ◦ )⇓ ⇓ }ev ,
which are identical due the commutativity of τ− with ⇓. Their coincidence with
the first one is essentially the verification of the fact that our twisted construction
depends only on the topological type of the corresponding link.
9.5. Toward the Skein. The generalized twisting can be used for the follow-
ing topological characterization of the pre-polynomials. Let P ∅ be a pre-polynomial,
where ∅ means that it is obtained without the division by the evaluations of the
J–polynomials involved. We mostly need P = P min in this work, which is P ∅ di-
vided by the total LCM of these evaluations. The corresponding superpolynomial
is Hmin= {P min }ev ; we will mostly drop the hat-normalization in this section. Due
to this normalization, P ∅ and P min are actually needed only up to q, t–monomial
factors in the rest of this work.
Theorem 9.1. (i) Let P ∅ be the standard pre-polynomial for a tree L from
( 4.22) and ξ ∈ P SL2 (Z) be as above. Using the non-degeneracy of { , }ev , the
pre-polynomial ξ(P ∅ ) ⇓ for the tree [α, β] → L can be uniquely determined via
the coinvariants {ξ(P ∅ )(Jλι )}ev for all diagrams λ.
The latter coinvariants are the
∅
superpolynomials HM for the trees M = [1, 1] ⇒ [β, α−β] → L, [1, −1] → λ .
(ii) Therefore all pre-polynomials (for any trees and colors) can be uniquely
recovered if all H–invariants are known. Topologically, this gives that the standard
ore-polynomials ( 4.22) are invariants of the corresponding links considered (natu-
rally) in the solid torus. This means that the symmetries from Sections 4.1, 4.2
hold unless they involve the first vertex [r1 , s1 ]. For instance, ( 4.7) holds, but ( 4.6)
does not.
Note that the class of pre-polynomials is closed with respect to the (usual)
multiplication, where the renormalization (the division by the LCM of all evalua-
tions) is necessary if Hmin are considered. Indeed, the pure product corresponds to
the union of the trees, which can be considered as one tree by connecting them to
the additional (initial) vertex [1, 0]. Generally, the reduction to a single tree is not
always reasonable. For instance, this may result in non-positive labels for algebraic
pairs of trees {L, L∨ }, as it is clear from (9.16).
The final step here would be a similar topological understanding of the knot
operators themselves in terms of the topological interpretation of their matrix el-
ements. These operators are K = ξ(P ) (and then by induction) in the notation
from Theorem 9.1; the matrix elements for such K are {Jμ , K(Jλι )}ev for any λ, μ.
Similar to the pre-polynomials, the knot operators can be expected invariants of
the links in the solid torus, but now we have to remove the middle circle too.
Equivalently, they correspond to links in the torus times an interval, T × I.
Indeed, considering {Jμ , K(Jλι )}ev with “free ends” λ, μ means topologically
that the corresponding trees are subtrees of undermined larger ones, extending them
in both directions (below i = 0 and beyond i = j ). Accordingly, the twisted unions
can be applied in both directions of the corresponding links. This requires fixing
the meridian as for pre-polynomials and also the middle circle. This topological
restriction gives that the symmetries from (4.6) and (4.7) must not generally hold
(now both) for the knot operators.
This approach can be expected to provide an identification of the spherical
DAHA with the toric q, t–skein algebra , the Skein, which is for T × I. This will
hopefully help to obtain H–polynomials for arbitrary links (not only cables of ),
though this problem can be approached independently.
Our way to understand the Skein via twisted unions (following Theorem 9.1)
is global (not via local skein relations), but there is a relatively direct connection
with the relations of the Elliptic Hall Algebra from [SV]. Such an approach to the
Skein is not at all restricted to An and can be readily extended to any root systems.
The a–stabilization for exceptional root systems will/may be lost (though we found
some in [ChE]), but anything else is expected to hold. The refined WRT-invariants
(for any Quantum Groups) will play the role of DAHA-superpolynomials.
This program is partially connected with paper [ChE] devoted to the q, t–
composite theory in annulus ×I. The Macdonald polynomials for composite parti-
def
tions [λ, μ] == ι(λ) + μ and their a–stabilization were used, instead of the products
Pλ Pμ here. This construction is actually for links too (for double torus knots),
but the corresponding superpolynomials are (naturally) smaller than those in the
present work. For instance, the uncolored composite case is for the weight ω1 + ωn
upon the stabilization at n → ∞. We do P22 or P2 P2ι here.
Seifert 3-folds. It seems that the expected interpretation of the spherical DAHA
as Skein(T × I) will require Seifert manifolds.
The twisted unions (with ∨) can be naturally considered as links in the lens
spaces. We place two links for L and L in the standard horizontal solid torus and
its vertical complement in S3 and then perform the corresponding twist at their
common boundary, the 2-dimensional torus. The lens spaces and Seifert manifolds
seem really necessary if arbitrary sequences ξ! = (ξi , 1 ≤ i ≤ m) of matrices from
P GL2 (Z) are considered for m > 1. Given a sequence of pre-polynomials P!∅ =
(Pi∅ ), the corresponding twisted knot operator and pre-polynomial are

(9.22) K ∅ = ξ1 (P1∅ ) · · · ξm (Pm
∅
), P ∅ = ξ1 (P1∅ ) · · · ξm (Pm
∅
) ⇓.
∅
Accordingly, we set H∅ = ξ HP = {P ∅ }ev and define Kmin , P min , Hmin upon the
division by the LCM of all J–polynomials involved.
The theory of Khovanov-Rozansky polynomials of the links in the lens spaces
and Seifert 3–folds is not developed. However see [Ste2] and references there for the
mathematics and physics approaches to the HOMFLY-PT theory there. The gen-
eralized ORS polynomials can be expected then for the germs of curve singularities
in the toric surfaces associated with these manifolds (and their plumbing).
Using Seifert manifolds is highly desirable anyway, especially if S3 is really

insufficient for a topological interpretation of the knot operators K, pre-polynomials
P for m > 1 and H–invariants for m > 2.
Recall that the latter can be reduced to the invariants in S3 for m = 2 due

to (9.12). Namely {ξ(Q), ξ (P )}ev for pre-polynomials P, Q and ξ, ξ ∈ P SL(2, Z)
are associated with proper links in S3 . We note that this bilinear form (in P, Q)
is symmetric when ξ = ϕ(ξ), which intersects the higher-level DAHA coinvariants
from [ChM].
The DAHA theory of K, P, H from (9.22) is similar to that for m = 2. They
depend only on the first columns (αi , βi )tr of ξi and the hat-normalized H min are
±1
a, q, t –polynomials satisfying the super-duality. Also, if Pi are for positive trees
such that s1 > r1 > 0, then the conditions αi > βi > 0 seem sufficient (not always
necessary) for the positivity of H min /((1−q)(1−t))M for M large enough.
Conceptually, such H min
is the dynamical m–point correlation function in
the DAHA theory; dynamical, because the parameters are weights. The theory
of the corresponding KZ-type difference equations, closely related to the so-called
A–polynomials , is in progress. An interpretation of generalized H min (for any
m) within the S –theory is not impossible, as well as that for the toric q, t–skein
3
algebra, but it seems that Seifert manifolds will be needed here. Anyway, the
interpretation of the knot operators from (9.22) and their iterations topologically
and physically is a challenge for the new theory we present in this work.
Appendix A. Links and splice diagrams

In this appendix, we remind the main topological constructions from [EN],
especially the definition and basic properties of the splice diagrams. We reduce the
generality and always assume that the links are in S3 , not in an arbitrary homology
3-sphere.
The key operation there is splicing ; it provides a large family of links. Other
operations can be mainly considered as its special cases. This includes the cabling,
unions and twisted unions, which play the major role in the DAHA approach to
torus iterated links.
A.1. Links, cables and splices. A link in S3 is a disjoint union {Si |i ∈ I} of

curves Si : S → S called components.
1 3
oriented simple closed

3 6(smooth) It will be
denoted by L = S , i∈I Si . Since we deal only with the links in S3 , we mostly
6
omit S3 in this notation and write L = i∈I Si . A knot is a link with a single
component. We consider links up to isotopies (smooth homotopies), ignoring the
order of components. However the orientation of the components will matter. The
unknot will be denoted by , the border of a (standard) disk D2 ⊂ S3 .
6
For each component Si of a link L = i∈I Si , there is a system of solid
tori N (Si ) ⊂ S3 such that Si ⊂ N (Si ), N (Si ) ∩ Sj = ∅ (j = i) and there is a
homeomorphism N (Si ) → D2 × S1 for each i mapping Si to “the middle circle”
{a} × S1 ⊂ D2 × S1 , a ∈ D2 . Such N (Si ) is called a tubular neighborhood of the
component Si .
Since ∂N (Si ) is homeomorphic to the torus S1 × S1 , the first homology is
H1 (∂N (Si ), Z) = Z2 and there is a topologically distinguished basis there. Let
l (·, ·) be the linking number. There is a single pair (M, L), M, L ∈ H1 (∂N (Si ), Z)
such that i) M is contractible in N (Si ), ii) l (Si , M) = 1, iii) L is homotopically

equivalent to Si in N (Si ), and iv) l (Si , L) = 0. We call M the meridian and L the
longitude .
A framed link is a link with a choice of an element Fi ∈ H1 (∂N (Si ), Z) for each
component Si such that Fi is homotopically equivalent Si in N (Si ). This element is
called framing. One can always equip a component with the longitude as its framing
and we call such a choice topological framing. However in the DAHA-approach, a
different framing is generally needed (see below).
Given two framings Fi and F i of the same component Si , one can find k ∈ Z

such that Fi − Fi = kMi , where Mi is the meridian of Si . Let us call k the difference
of framings Fi and Fi .
We now briefly present the operations on 6 links.

Erasing components. Given a link 6 L = i∈I Si and a subset of its components
J ⊂ I, one can consider a new link i∈I\J Si .
6
Orientation reversion. For a link L = i∈I Si , one can reverse
6 the orientation
6 for
a subset J ⊂ I of components. This link will be denoted by i∈I\J Si ∪ j∈J (Sj∨ ).
We will need the following general operation on manifolds. Let us define the
connected sum Σ " Σ of two (connected) manifolds Σ and Σ of the same dimension.
We select two open balls U ⊂ Σ and U ⊂ Σ and then paste Σ \ U and Σ \ U
along ∂U and ∂U . This does not depend (up to a homeomorphism) on the choice
of U and U . Note that S3 " S3 = S3 .

6 6
Disjoint sum. For two links L = S3 , i∈I Si and L = S3 , j∈J Sj , their
disjoint sum is the link
⎛ ⎞
7 7
L + L = ⎝S3 " S3 , Si ∪ Sj ⎠ .
i∈I j∈J
The components Si and Sj here are the images of the corresponding components
of the original links under the natural maps S3 \ U → S3 " S3 and S3 \ U → S3 " S3
from the definition of S3 " S3 . The open balls U and U must not intersect the
components of L and L ; L + L does not depend on choice of these domains.
Connected sum. Now let us pick two components Si0 , Sj 0 in L, L (i0 ∈ I,
j0 ∈ J) and define the connected sum of links L and L along Si0 and Sj 0 :

7 7
(A.1) L" L (Si0 , Sj 0 ) = S3 " S3 , Si0 " Sj 0 ∪ Si ∪ Sj .
i∈I\{i0 } j∈J\{j0 }
We identify Si and Sj for i = i0 , j = j0 with the images of the corresponding

components of the original links under the natural maps S3 \ U → S3 " S3 and
S3 \ U → S3 " S3 . The definition of S3 " S3 is above. Let us pick the open balls U
and U intersecting only with Si , Sj and such that the intervals U ∩ Si0 = I and
U ∩ Sj 0 = I are non-empty. Then we glue Si0 \ U to Si0 \ U with respect to the
orientation of Si0 and Sj 0 in S3 " S3 . This gives Si0 " Sj 0 needed in (A.1). One can
check that L" L (Si0 , Sj 0 ) does not depend on the choice of U and U ; however it
of course depends on the choice of components Si0 and Sj 0 .
Note that both operations, L + L and L" L , result in links in S3 since S3 " S3 =
S . The next operation is the key in DAHA theory.
3
Cabling. For coprime k, l ∈ Z, there exists a unique up to homotopy oriented

simple curve Cabk,l (Si ) ⊂ ∂N (Si ) such that it is homotopy equivalent to kM +
lL. Here k, l ∈ Z can be arbitrary, then Cabk,l (Si ) ⊂ ∂N (Si ) will be a union
(unique up to homotopy) of d = gcd(k, l) non-intersecting oriented connected closed
components; each of them is homology equivalent6 to (k/d)M + (l/d)L.
The cable of type (k, l), k, l ∈ Z of a link L = i∈I Si along the component Si0
is the link 7
S
Cabk,li0 L = Cabk,l (Si ) ∪ Si .
i∈I\{i0 }
The knot Cabk,l ( ) is the torus knot (k, l) denoted by T (k, l) (by T (l, k) in the
paper).
The (k, 1)-cabling preserves a link (up to isotopy) and (k, −1)-cabling reverses
the orientation of the cabled component:
7 7
Si0 Si0
Cabk,1 L = L, Cabk,−1 Si = Si∨0 ∪ Si .
i∈I i∈I\{i0 }
The cabling operations along different components commute with each other
because different components have non-intersecting tubular neighborhoods.
Splice. This is actually the most general operation on links; all previous ones
can be obtained as its proper6specializations. 6
We take two links L = i∈I Si and L = j∈J Sj and fix one component in
each: Si0 and Sj 0 , where i0 ∈ I and j0 ∈ J. The splice of the links L and L along
Si0 and Sj 0 , denoted by
⎛ ⎞
Si0 Sj 0 7 7
L L = ⎝Σ, Si ∪ Sj ⎠ ,
i∈I\{i0 } j∈J\{j0 }
is as follows. We glue together S \ N (Si0 ) and S3 \ N (Sj 0 ) along ∂N (Si0 ) and

3
∂N (Sj 0 ), where the meridian of ∂N (Si0 ) is identified with the longitude of ∂N (Sj 0 )
and vice versa. As above, Si and Sj (i = i0 , j = j0 ) are considered as the images
of the corresponding components under the inclusions S3 \ N (Si0 ) → Σ and S3 \
N (Sj 0 ) → Σ. We will need only cases when Σ is diffeomorphic to S3 , so the splice
(the union of Si , Sj above) will be a link in S3 .
This operation is L ↔ L –symmetric. We will need its non-symmetric gener-
alization. It will depend now on the framing Fi0 of Si0 . The splice of L and L
along Si0 and Sj 0 deformed by framing Fi0 is as follows. Now only the longitude
of Sj 0 will be identified with the meridian of Si0 . The meridian of Sj 0 will not
be identified with the corresponding longitude; it will be now identified with the
framing Fi0 . We set:
Si0 Sj 0
(A.2) L −−→ L .
Here Fi0 is determined from the context. If Fi0 is the topological framing,
then the deformed splice coincides with the standard one. The difference can be
only because of non-trivial choices of the framing. Note that the splice operations
(deformed or non-deformed) commute with each other.
A.2. Splice diagrams. Splice diagrams provide a convenient way to deal with
links made by splices of “canonical” links in Seifert manifolds (here in S3 ).
Seifert fibrations. A standard fibered solid torus corresponding to a pair of
coprime integers (k, l) (k > 0) is the topological space D2 × [0, 1] with its (two)
borders identified by relations (x, 1) ∼ (ρl/k x, 1), where ρα is the rotation by angle
2πα in D2 around the center O ∈ D2 . The fibration is inherited from D2 × [0, 1] in
the following way. For each x ∈ D2 \ O, the segments {ρi/k x} × [0, 1] form a circle.
These circles will be called regular fibers. The segment O × [0, 1] is a circle too,
called a singular fiber of multiplicity k if k = 1 (if k = 1 it is regular). Thus we
have a fibration with the S1 –fibers.
A Seifert fibration of a 3-manifold Σ is a continuous map π : Σ → B onto
2-manifold B satisfying the following. For each point b ∈ B, there exists its neigh-
borhood Ub such that the fibration π −1 (Ub ) is isomorphic to the interior of the
standard fibered solid torus. Singular fibers of multiplicity k ≥ 2 correspond then
the singular fibers above. This definition does not depend on the choice of the
isomorphism.
The following fibrations of S3 with S1 -fibers are important for us. For each un-
ordered pair of coprime integers α1 , α2 (αi ≥ 2), there exists a projection πα1 ,α2 : S3 →
S2 with exactly two singular fibers with multiplicities α1 , α2 . This property uniquely
determines such a fibration. Following [EN], let us provide some details.
62
Topological description. Let B = S2 \ i=1 Ui be the 2-sphere without two
disjoint domains Ui , E = S1 × B → B be the trivial 62S -bundle
1
over it. The border
E consists of two tori, one for each Ui : ∂E = i=1 T i . The required space is
obtained by gluing E with two solid tori D2 × S1 (one for each Ti ) along border.
Fixing a section σ : B → E, the fiber Hi in Ti ⊂ E and Qi = Ti ∩ σ(B) form a basis
in H1 (Ti , Z) for i = 1, 2.
The fibers are as follows. Since αi are pairwise coprime, there exist βi (1 ≤ i ≤
2) such that
β1 α2 + β2 α1 = 1.
For each i, we then glue Ti with the border of D2 × S1 in such a way that the
homology class of αi Qi + βi Hi becomes zero. The resulting manifold is S3 with the
inherited S1 -fibration. Note that there is a flexibility with the choices of σ and βi
here, but the output does not depend on this up to a diffeomorphism.
Analytic description. Let S3 ⊂ C2 be the standard unit sphere |Z1 |2 + |Z2 |2 = 1
for Zi ∈ C. The fibers will be then the orbits of the following S1 -action:
t(Z1 , Z2 ) = (tα2 Z1 , tα1 Z2 ),
where t ∈ C, |t| = 1. The singular fiber of multiplicity αi are exactly the in-
tersections of the sphere with the hyperplanes Zi = 0. This gives the required.
The cases with αi ≤ 2. In the construction of the fibration πα1 ,α2 , we imposed
the conditions αi ≥ 2 (mainly to ensure that the fibrations have two singular fibers).
This restriction can be omitted. We have the following 3 cases.
(i) If αi = 1 for one of i = 1, 2 or both, then the corresponding fiber is regular
and such an extension is obvious.
(ii) If α1 < 0 or α2 < 0, then the resulting fibration of S3 will be the one
for {|α1 |, |α2 |} upon the inversion of the orientation of S3 (as a 3-manifold) when
α1 α2 < 0.
(iii) If αi = 0, then αj = ±1 for the remaining index j due to gcd(αi , αj ) = 1,

so the fiber for αj is regular. The multiplicity of the singular fiber is then 0 and it is
the unknot; the regular fibres are the meridians of the singular one. The S1 -action
with respect to this fibration is not free, but there is another S1 -action on S3 whose
fixed point are precisely the points of the singular fiber.
Seifert links. Let us provide a series of “canonical” links, which will be then
used as starting points for the operations above. The Seifert link is a link with the
exterior that admits the Seifert fibration. It is known [EN], that every Seifert link
in S3 is a collection of fibers in the Seifert fibration of S3 . Let (α1 , . . . , αk ) be a
k–tuple of non-negative integers (k ≥ 2) with αj = 1 for j = i, i for a certain pair
1 ≤ i = i ≤ k such that gcd(αi , αi ) = 1. Then it defines the link
7
k
Lα1 ,...,αk = Sl ,
l=1
where Sl are the fibers of the Seifert fibration παi ,αi : S3 → S2 . Namely, if l = i, i ,
they are the singular fibers of multiplicity αi , αi and regular fibers otherwise. Note
that if there are several choices of such a pair αi , αi (then one of them must be 1),
the link Lα1 ,...,αk does not depend on such a choice.
We will represent Lα1 ,...,αk as follows:
... ... ... ...
α@
I αi 6αk α@
I αi 6
αk
1 1
@ @1
(A.3) with = +1 or simply .
The reflection −Lα1 ,...,αk of Lα1 ,...,αk will then correspond to = −1.
If we do not assume that all αi are non-negative, the corresponding link can
be defined by the relations
... ... ... ∨ ...
α@
I αi 6 αk αI
@ −αi 6 αk
1
@ ∼ 1
@
(A.4) δ ,
where δ = −1; by ∨, we mean the reversion of the orientation of the corresponding
component. Here i can be arbitrary, so (A.4) allows to define links for any k-tuples
(α1 , . . . , αk ) such that
(A.5) αj = ±1 for j = i, i , where 1 ≤ i = i ≤ k, gcd(αi , αi ) = 1.
Let us discuss the framing. In DAHA theory, only regular components are
needed, which are regular fibers (those of multiplicity ±1). Any fibers have tubular
neighborhoods with unions of Seifert fibres as their borders. The regular fibers
are homotopic to the fibers belonging to the border of the corresponding tubular
neighborhood. This supplies them with a canonical framing, called Seifert framing.
For irregular components, one may choose any framing (say, topological), but such
components are not actually needed in our work.
The topological framing is the most common choice. However the Seifert fram-
ing appeared the one serving the DAHA invariants. Let us comment on the differ-
ence.
The regular components Si and Sj of a given link ±Lα1 ,...,αk have the linking
i . . . α
number lk = ±α1 . . . α j . . . αk (it results from the formula for linking numbers
in [EN], Chapter III, Section 10). Here α i means that αi is omitted. Let us add a
(regular) component Sl to a regular component Sl in ±Lα1 ,...,αk using the Seifert
framing. The resulting link will be ±Lα1 ,...,αk ∪ Sl , which is ±Lα1 ,...,αk ,1 for αl = 1
and ∓Lα1 ,...,αk ,−1 for αl = −1 by definition. The linking number between Sl and
Sl will be ±αl · α1 . . . α
l . . . αk = ±α1 . . . αk , which is the difference between the
Seifert framing and the topological framing (when the linking number between S
and S would be zero).
Splice diagrams. They are trees or disjoint unions of trees, where the vertices
can be nodes labeled by ±1, unlabeled nodes, called leaves, and arrowheads ; the
pairs {a labeled node, an edge from it} are decorated by integers, called weights.
In greater detail, the vertices are as follows.
An arrowhead, at the end of an edge:
-
S
or - ∨ .
S
It is a vertex of valence 1. The arrowheads are topologically interpreted as the

components of the resulting links. We mostly put the names of the components
over the arrowheads, S or S∨ in these examples. Here ∨ stands for the change
of orientation of the component S; see below. The S–labels are necessary in the
operations on links.
A leaf, an unlabeled vertex of valence 1:
◦.
A general vertex, a node labeled by ±1:
... ...
α@ αi αk
1
@
.
This must be a vertex of valence ≥ 3 with the label = ±1 in it. Every adjacent
edge has an integer weight αi such that 1(α1 , . . . , αk ) satisfies A.5. As in (A.3), we
replace the node with = +1 in it by .
Given a splice diagram Γ, the corresponding link L(Γ) is constructed as follows.
A general vertex with the label and the weights α1 , . . . , αk at the adjacent
edges is interpreted as the link Lα1 ,...,αk . It has the standard orientation for = 1,
which must be reversed for = −1. The adjacent edges give the components of
this link; the weights show their multiplicities (as singular fibers).
An edge that does not have the arrowhead is interpreted as the splice of two
links associated with the vertices at these ends (along the corresponding compo-
nents). The arrowheads provide the components of L(Γ); if there is ∨ there, the
orientation of the corresponding component must be reversed.
A leaf alone is interpreted as the unknot. Replacing an arrowhead by a leaf
topologically means deleting the corresponding component. The disjoint sum of
graphs corresponds to the disjoint sum of links. See also below.
The links that are constructed by splice diagrams form the class of graph links.
Note that we restrict our definitions to a subclass sufficient to work with links in
S3 . See [EN] for arbitrary graph links.
A.3. Operations on links. The operations from section A.1 for the links
associated with such diagrams naturally result in operations on splice diagrams.
The graphs of links L will be presented as boxes with L inside in the figures below.
Sometimes we show the arrows from this box corresponding to the components of
the link labeled by S, S and so on. We will use the operations S ∨ , S" S etc., defined
above and all previous notations for the link operations will be used. We will show
the name of the initial component (before the transformation) in the figures below;
the corresponding arrowhead is replaced as follows.
The erasure of a component. We replace the arrow labeled by S in the graph of
L by the leaf. This corresponds to erasing component S:
L → L ◦ .
S
The orientation reversion of a component S of a link L. The new (changed)

component is denoted by S ∨ :
◦
−11 S ∨
L - .
S x y
The weights x and y here are arbitrary; the link does not depend on them, only the
Seifert framing does. So one can choose them to obtain the desired Seifert framing.
The connected sum of two links L1 and L2 along S and S . The new (changed)
component is denoted by S" S .
S"S
06
S 1 S
L1 1 1 L2
.
The cable of type (a, r) along any component S of a link L. The new component
is called Caba,r (S):
◦
r
S 1 Caba,r (S)
-
L a 1
.
From the definition of the splice diagrams, it is obvious that the spice of two
graph links is a graph link. The same is true for the disjoint sum. Moreover, the
following theorem holds.
Theorem A.1. The set of all graph links in S3 is the set of all solvable links,
i.e. all links which can be constructed from the unknot by any number of cablings,
disjoint and connected sums.
The proof is in [EN] (Theorem 9.2).
A.4. Equivalent diagrams. Now we provide all relations between splice di-
agrams. We denote the link made from L by reversing of all orientations of com-
ponents as L∨ , which is adding ∨ to the components that have none or deleting it
for the components with ∨.
Theorem A.2. ([EN], Theorem 8.1) The following relations hold:

(i) L = L∨
(i.e. when the orientations of all components are reversed);
(ii)
L1 L1
@ α1 @ α1
.. @ .. @
. α L0 ∼ . δ L∨
0 −α 0
αk αk 0
Lk Lk
for δ = −;
(iii)
L1 L1
@ α1 @ α1
.. @ .. @
. ◦ ∼ . if k > 2,
1
αk αk
Lk Lk
⎧
L1 ⎪
⎪
@ α1 ⎪
⎨ L1 L2 if = 1
@
◦ ∼ ;
1 ⎪
α2 ⎪
⎪
L2 ⎩ L1 L∨
2 if = −1
(iv)
L1 L1 ◦
@ α1
.. @ ..
. ◦ ∼ .
0
αk
Lk Lk ◦
(the disjoint sum in the right-hand side);

(v) ◦ ◦
∼ L
L
(the disjoint sum in the left-hand side);

(vi) If α0 α0 = γδα1 . . . αk α1 . . . αl define = ±1 such that
α0 = γα1 . . . αl ,
α0 = δα1 . . . αk .
(vi ) If γδ = 1, then
L1 L1 L1 L1
@ α1 α1

@α1 α1
.. @ α .. .. @ ..
. γ α 0 δ . ∼ . .
0
αk αl@@ @

Lk Ll Lk αk αl @ Ll
.
(vi ) If γδ = −1, then
L1 L1 L1 L∨
@ α1 α1

@α1 α1 1
.. @ α .. .. @ ..
. γ α 0 δ . ∼ . .
0
αk αl@@ @

Lk Ll Lk αk αl @ L∨l
.
(such exist, since α0 is prime to α1 . . . αk andα0is prime to α1
. . . αk .)
Splice diagrams are called equivalent if they can be obtained from each other by
these relations [EN]. We have the following corollary, one of the key in the DAHA
approach; see below.
Corollary A.3. For any general vertex and arbitrary i = ±1,
1 1
a 1 .. a1 ar1 ..
. ∼ .
r @
l@ r @
l@ ,
where = 1 . . . l .
We call a splice diagram minimal if no equivalent diagram has fewer edges. We
call a splice diagram Γ normal if
(i) Γ is minimal; (ii) all edge weights are non-negative;
(iii) if an edge weight is zero, the adjacent vertex has the label +1.
Theorem A.4. (Corollary 8.3 in [EN]). Up to the transformation Γ → Γ∨
(see above), there is a unique normal form for any splice diagram.
A.5. Connection with DAHA. We begin with the general description of

algebraic links from Theorem 9.4 in [EN]. All algebraic links are graph links; see
Appendix to Chapter I in [EN]. The following theorem describes all of them.
Theorem A.5. The solvable link L(Γ) is an algebraic link if
(a) all edge weights and vertex labels in Γ are positive;
(b) α0 α0 > α1 . . . αk α1 . . . αl for every edge as follows:

@ α1 α α1
.. @1 0 1 ..
. α0 .
αk αl@@ .
Vice versa, if L is an algebraic link then its normal form graph satisfies the
above conditions.
These conditions automatically holds for the positive trees from our paper. The
twisted union for {L, L∨ } there corresponds to
@1 1
@1 a2 1
a1
r1 r@
2@ .
Condition (b) from the theorem then becomes a1 a2 > r1 r2 (if the minimality
holds). This is the positivity condition for the pairs {L, L∨ } in our work; the
positive pairs are sufficient to obtain arbitrary algebraic links.
Seifert framing. We will briefly discuss the framing needed in the DAHA ap-
proach. We need not only splices of links, but also deformed splices, defined in
(A.2) at the end of Section A.1; they are denoted by an edge marked by an arrow
in the middle: - , not related to the arrowheads we use for components. This
is a certain extension of the definition of the splice diagram. They arrowed edges
correspond to deformed splices in the same way as the ordinary edges correspond
to usual splices.
Let L be a link with a chosen component S in it. If the framing on S differs
by D from the topological one, then the following links are equivalent:
1 1
S- r1 .. S r1 ..
L s @ . ∼ L a @.
1@ 1@
,
where a = s + rD. This is clear from the definitions.
If S is a regular component of a Seifert link, then it has the Seifert framing.
One can use it to simplify the relation for the decomposition in Corollary A.3, which
can be presented as follows:
1 1 1
a 1 .. a1 ar1 .. a 1 - 0 1 ..
. ∼ 1 . ∼ 1 .
r 1@@ r 1@@ r 1@@
◦ ◦ ◦
This is important for the DAHA-approach. Recall the (a, r, 1)-vertex corre-
sponds to the projective P SL2 (Z)–action and (0, 1, . . . 1)–vertex corresponds to
the multiplication of pre-polynomials. The usage of the Seifert framing completely
clarifies the passage from the (r, s)-pairs to (a, r)-pairs in the DAHA-approach.
Namely,
◦ ◦ ◦ ◦

r1 S
- 1 - ∼
r r1 S r1
L L - ,

a 1 s 1 a 1 a 1
where a = s + a r r for the Seifert framing on S. This is precisely the recursive

formula for the switch from the (r, s)-pairs to (a, r)-pairs in the DAHA-approach.
The arrows at the right ends in this figure are arrowheads showing the components.
Thus applying γr,s to a pre-polynomial in DAHA theory corresponds to splic-
ing L → Ls,r,1 in topology. See the main body of our work for this and other
connections. The Seifert framing is exactly the one which corresponds to the
DAHA-approach. This results in the exact topological interpretation of the DAHA-
superpolynomials of torus iterated links, i.e. without using the hat-normalization
(ignoring factors q • t• ). This will be discussed somewhere.
Acknowledgements. We are grateful to Mikhail Khovanov, Andrei Negut,
David Rose for useful discussions, Semen Artamonov for our using his software
for calculating colored HOMFLY-PT polynomials, and the referee for important
suggestions. Special thanks to Aaron Lauda for his attention to this work and
helpful discussions on paper [MS]. The first author thanks Andras Szenes and
University of Geneva for the invitation and hospitality. I.D. acknowledges partial
support from the RFBR grants 13-02-00478, 14-02-31446-mol-a, NSh-1500.2014.2

and the common grant 14-01-92691-Ind-a.
References
[AKMV] M. Aganagic, A. Klemm, M. Mari–no, and C. Vafa, The topological vertex, Comm. Math.
Phys. 254 (2005), no. 2, 425–478, DOI 10.1007/s00220-004-1162-z. MR2117633
[AS] M. Aganagic, and S. Shakirov, Knot homology from refined Chern-Simons theory,
arXiv:1105.5117v1 [hep-th], 2011.
[AMM] A. Anokhina, A. Mironov, A. Morozov, and And. Morozov, Colored HOMFLY polyno-
mials as multiple sums over paths or standard Young tableaux, Adv. High Energy Phys.
(2013), Art. ID 931830, 12. MR3127275
[AM] A. S. Anokhina and A. A. Morozov, Cabling procedure for the colored HOMFLY polyno-
mials, Theoret. and Math. Phys. 178 (2014), no. 1, 1–58, DOI 10.1007/s11232-014-0129-
2. Russian version appears in Teoret. Mat. Fiz. 178 (2014), no. 1, 3–68. MR3302459
[AFS] H. Awata, and B. Feigin, and J. Shiraishi, Quantum algebraic approach to refined topo-
logical vertex, arXiv:1112.6074 [hep-th] (2011).
[Bo] N. Bourbaki, Groupes et algèbres de Lie, Ch. 4–6, Hermann, Paris (1969).
[Ch1] I. Cherednik, Double affine Hecke algebras, London Mathematical Society Lecture Note
Series, vol. 319, Cambridge University Press, Cambridge, 2005. MR2133033
[Ch2] I. Cherednik, Jones polynomials of torus knots via DAHA, Int. Math. Res. Notices,
2013:23 (2013), 5366–5425.
[Ch3] I. Cherednik, DAHA-Jones polynomials of torus knots, arXiv:1406.3959 [math.QA]
(2014).
[Ch4] I. Cherednik, Macdonald’s evaluation conjectures and difference Fourier transform, In-
vent. Math. 122 (1995), no. 1, 119–145, DOI 10.1007/BF01231441. MR1354956
[Ch5] I. Cherednik, Nonsemisimple Macdonald polynomials, Selecta Math. (N.S.) 14 (2009),
no. 3-4, 427–569, DOI 10.1007/s00029-009-0493-1. MR2511191
[Ch6] Ivan Cherednik, Nonsymmetric Macdonald polynomials , IMRN 10 (1995), 483–515.
[Ch7] I. Cherednik, Intertwining operators of double affine Hecke algebras, Selecta Math.
(N.S.) 3 (1997), no. 4, 459–495, DOI 10.1007/s000290050017. MR1613515
[Ch8] Ivan Cherednik, Difference Macdonald-Mehta conjecture, IMRN 10 (1997), 449–467.
[ChD] Ivan Cherednik, and I. Danilenko, DAHA and iterated torus knots, arXiv:1408.4348
[math.QA] (2014).
[ChF] I. Cherednik and B. Feigin, Rogers-Ramanujan type identities and Nil-DAHA, Adv.
Math. 248 (2013), 1050–1088, DOI 10.1016/j.aim.2013.08.025. MR3107536
[ChM] I. Cherednik and X. Ma, Spherical and Whittaker functions via DAHA I, Selecta Math.
(N.S.) 19 (2013), no. 3, 737–817, DOI 10.1007/s00029-012-0110-6. MR3101119
[ChE] , and R. Elliot, Refined composite invariants of torus knots via DAHA, Preprint:
arXiv: 1503.01441, to appear in ”les Annales de la Faculté des sciences de Toulouse:
Mathématiques (2015).
[DGPS] W. Decker, and G.M. Greuel, G. Pfister, and H. Schönemann, Singular 4-0-2 – A com-
puter algebra system for polynomial computations (2015).
[DMS] P. Dunin-Barkowski, A. Mironov, A. Morozov, A. Sleptsov, and A. Smirnov, Super-
polynomials for torus knots from evolution induced by cut-and-join operators, J. High
Energy Phys. 3 (2013), 021, front matter+85. MR3046744
[DGR] N. M. Dunfield, S. Gukov, and J. Rasmussen, The superpolynomial for knot homologies,
Experiment. Math. 15 (2006), no. 2, 129–159. MR2253002
[EN] D. Eisenbud and W. Neumann, Three-dimensional link theory and invariants of plane
curve singularities, Annals of Mathematics Studies, vol. 110, Princeton University Press,
Princeton, NJ, 1985. MR817982
[EGL] P. Etingof, E. Gorsky, and I. Losev, Representations of rational Cherednik alge-
bras with minimal support and torus knots, Adv. Math. 277 (2015), 124–180, DOI
10.1016/j.aim.2015.03.003. MR3336085
[FGS] H. Fuji, S. Gukov, and P. Sulkowski, Super-A-polynomial for knots and BPS states,
Nuclear Phys. B 867 (2013), no. 2, 506–546, DOI 10.1016/j.nuclphysb.2012.10.005.
MR2992793
[Gor] E. Gorsky, Arc spaces and DAHA representations, Selecta Math. (N.S.) 19 (2013), no. 1,
125–140, DOI 10.1007/s00029-012-0096-0. MR3022754
[GN] E. Gorsky and A. Neguţ, Refined knot invariants and Hilbert schemes (English, with
English and French summaries), J. Math. Pures Appl. (9) 104 (2015), no. 3, 403–435,
DOI 10.1016/j.matpur.2015.03.003. MR3383172
[GORS] E. Gorsky, A. Oblomkov, J. Rasmussen, and V. Shende, Torus knots and the rational
DAHA, Duke Math. J. 163 (2014), no. 14, 2709–2794, DOI 10.1215/00127094-2827126.
MR3273582
[GGS] E. Gorsky, and S. Gukov, and M. Stosic, Quadruply-graded colored homology of knots,
arXiv:1304.3481 (2013).
[GIKV] S. Gukov, A. Iqbal, C. Kozçaz, and C. Vafa, Link homologies and the refined topological
vertex, Comm. Math. Phys. 298 (2010), no. 3, 757–785, DOI 10.1007/s00220-010-1045-4.
MR2670927
[GS] S. Gukov, and M. Stosic, Homological algebra of knots and BPS states,
arXiv:1112.0030v1 [hep-th] (2011).
[HHL] J. Haglund, M. Haiman, and N. Loehr, A combinatorial formula for nonsym-
metric Macdonald polynomials, Amer. J. Math. 130 (2008), no. 2, 359–383, DOI
10.1353/ajm.2008.0015. MR2405160
[Kh] M. Khovanov, Triply-graded link homology and Hochschild homology of Soergel bimod-
ules, Internat. J. Math. 18 (2007), no. 8, 869–885, DOI 10.1142/S0129167X07004400.
MR2339573
[KhR1] M. Khovanov and L. Rozansky, Matrix factorizations and link homology, Fund. Math.
199 (2008), no. 1, 1–91, DOI 10.4064/fm199-1-1. MR2391017
[KhR2] M. Khovanov and L. Rozansky, Matrix factorizations and link homology. II, Geom.
Topol. 12 (2008), no. 3, 1387–1425, DOI 10.2140/gt.2008.12.1387. MR2421131
[Ki] A. A. Kirillov Jr., On an inner product in modular tensor categories, J. Amer. Math.
Soc. 9 (1996), no. 4, 1135–1169, DOI 10.1090/S0894-0347-96-00210-X. MR1358983
[KA] The Knot Atlas, http://katlas.org/.
[Mac] I. G. Macdonald, Affine Hecke algebras and orthogonal polynomials, Astérisque 237
(1996), Exp. No. 797, 4, 189–207. Séminaire Bourbaki, Vol. 1994/95. MR1423624
[Ma] D. Maulik, Stable pairs and the HOMFLY polynomial, arXiv:1210.6323 [math.AG]
(2012).
[Mo] H. R. Morton, The coloured Jones function and Alexander polynomial for torus
knots, Math. Proc. Cambridge Philos. Soc. 117 (1995), no. 1, 129–135, DOI
10.1017/S0305004100072959. MR1297899
[MS] H. R. Morton, and P. Samuelson, The HOMFLY-PT skein algebra of the torus and the
elliptic Hall algebra, arXiv:1410.0859v2 [math.QA] (2015).
[Nek] N. A. Nekrasov, Seiberg-Witten prepotential from instanton counting, Adv. Theor. Math.
Phys. 7 (2003), no. 5, 831–864. MR2045303
[ORS] A. Oblomkov and V. Shende, The Hilbert scheme of a plane curve singularity and the
HOMFLY polynomial of its link, Duke Math. J. 161 (2012), no. 7, 1277–1303, DOI
10.1215/00127094-1593281. MR2922375
[ObS] A. Oblomkov and V. Shende, The Hilbert scheme of a plane curve singularity and the
HOMFLY polynomial of its link, Duke Math. J. 161 (2012), no. 7, 1277–1303, DOI
10.1215/00127094-1593281. MR2922375
[OS] D. Orr, and M. Shimozono, Specializations of nonsymmetric Macdonald-Koornwinder
polynomials, arxiv:1310.0279v2 (2013).
[Pi] J. Piontkowski, Topology of the compactified Jacobians of singular curves, Math. Z. 255
(2007), no. 1, 195–226, DOI 10.1007/s00209-006-0021-3. MR2262728
[QS] H. Queffelec, and A. Sartori, A note on the gl(m|n) link invariants and the HOMFLY-
PT polynomial, arxiv 1506.03329 (2015).
[RY] A. Ram and M. Yip, A combinatorial formula for Macdonald polynomials, Adv. Math.
226 (2011), no. 1, 309–331, DOI 10.1016/j.aim.2010.06.022. MR2735761
[Ras] J. Rasmussen, Some differentials on Khovanov-Rozansky homology, Geom. Topol. 19
(2015), no. 6, 3031–3104, DOI 10.2140/gt.2015.19.3031. MR3447099
[RJ] M. Rosso and V. Jones, On the invariants of torus knots derived from quantum groups,
J. Knot Theory Ramifications 2 (1993), no. 1, 97–112, DOI 10.1142/S0218216593000064.
MR1209320
[Rou] R. Rouquier, Khovanov-Rozansky homology and 2-braid groups, arXiv: 1203.5065 (2012).
[Sam] P. Samuelson, A topological construction of Cherednik’s sl2 torus knot polynomials,
arXiv:1408.0483 [math.QA] (2014).
[SV] O. Schiffmann and E. Vasserot, The elliptic Hall algebra, Cherednik Hecke alge-
bras and Macdonald polynomials, Compos. Math. 147 (2011), no. 1, 188–234, DOI
10.1112/S0010437X10004872. MR2771130
[Ste1] S. Stevan, Chern-Simons invariants of torus links, Ann. Henri Poincaré 11 (2010), no. 7,
1201–1224, DOI 10.1007/s00023-010-0058-z. MR2753387
[Ste2] Sébastien Stevan, Torus Knots in Lens Spaces & Topological Strings, arXiv: 1308.5509v2
[hep-th] (2014).
[Web] B. Webster, Knot invariants and higher representation theory, arXiv: 1309.3796 (2013).
[WW] B. Webster, and G. Williamson, A geometric construction of colored HOMFLYPT ho-
mology, arXiv:0905.0486v3 (2010).
Department of Mathematics, University of North Carolina at Chapel Hill, Chapel

Hill, North Carolina 27599
E-mail address: chered@email.unc.edu
Department of Mathematics, Columbia University, 2990 Broadway, New York, New

York 10027
E-mail address: danilenko@math.columbia.edu
Selected Published Titles in This Series
684 Anna Beliakova and Aaron D. Lauda, Editors, Categorification in Geometry,
Topology, and Physics, 2017
683 Anna Beliakova and Aaron D. Lauda, Editors, Categorification and Higher
Representation Theory, 2017
682 Gregory Arone, Brenda Johnson, Pascal Lambrechts, Brian A. Munson,
and Ismar Volić, Editors, Manifolds and K-Theory, 2017
681 Shiferaw Berhanu, Nordine Mir, and Emil J. Straube, Editors, Analysis and
Geometry in Several Complex Variables, 2017
680 Sergei Gukov, Mikhail Khovanov, and Johannes Walcher, Editors, Physics and
Mathematics of Link Homology, 2016
679 Catherine Bénéteau, Alberto A. Condori, Constanze Liaw, William T. Ross,
and Alan A. Sola, Editors, Recent Progress on Operator Theory and Approximation in
Spaces of Analytic Functions, 2016
678 Joseph Auslander, Aimee Johnson, and Cesar E. Silva, Editors, Ergodic Theory,
Dynamical Systems, and the Continuing Influence of John C. Oxtoby, 2016
677 Delaram Kahrobaei, Bren Cavallo, and David Garber, Editors, Algebra and
Computer Science, 2016
676 Pierre Martinetti and Jean-Christophe Wallet, Editors, Noncommutative
Geometry and Optimal Transport, 2016
675 Ana Claudia Nabarro, Juan J. Nuño-Ballesteros, Raúl Oset Sinha, and Maria
Aparecida Soares Ruas, Editors, Real and Complex Singularities, 2016
674 Bogdan D. Suceavă, Alfonso Carriazo, Yun Myung Oh, and Joeri Van der
Veken, Editors, Recent Advances in the Geometry of Submanifolds, 2016
673 Alex Martsinkovsky, Gordana Todorov, and Kiyoshi Igusa, Editors, Recent
Developments in Representation Theory, 2016
672 Bernard Russo, Asuman Güven Aksoy, Ravshan Ashurov, and Shavkat
Ayupov, Editors, Topics in Functional Analysis and Algebra, 2016
671 Robert S. Doran and Efton Park, Editors, Operator Algebras and Their
Applications, 2016
670 Krishnendu Gongopadhyay and Rama Mishra, Editors, Knot Theory and Its
Applications, 2016
669 Sergiı̌ Kolyada, Martin Möller, Pieter Moree, and Thomas Ward, Editors,
Dynamics and Numbers, 2016
668 Gregory Budzban, Harry Randolph Hughes, and Henri Schurz, Editors,
Probability on Algebraic and Geometric Structures, 2016
667 Mark L. Agranovsky, Matania Ben-Artzi, Greg Galloway, Lavi Karp, Dmitry
Khavinson, Simeon Reich, Gilbert Weinstein, and Lawrence Zalcman, Editors,
Complex Analysis and Dynamical Systems VI: Part 2: Complex Analysis, Quasiconformal
Mappings, Complex Dynamics, 2016
666 Vicenţiu D. Rădulescu, Adélia Sequeira, and Vsevolod A. Solonnikov, Editors,
Recent Advances in Partial Differential Equations and Applications, 2016
665 Helge Glöckner, Alain Escassut, and Khodr Shamseddine, Editors, Advances in
Non-Archimedean Analysis, 2016
664 Dihua Jiang, Freydoon Shahidi, and David Soudry, Editors, Advances in the
Theory of Automorphic Forms and Their L-functions, 2016
663 David Kohel and Igor Shparlinski, Editors, Frobenius Distributions: Lang-Trotter
and Sato-Tate Conjectures, 2016
662 Zair Ibragimov, Norman Levenberg, Sergey Pinchuk, and Azimbay Sadullaev,
Editors, Topics in Several Complex Variables, 2016
For a complete list of titles in this series, visit the

AMS Bookstore at www.ams.org/bookstore/conmseries/.
684 CONM
Categorification • Beliakova and Lauda, Editors
The emergent mathematical philosophy of categorification is reshaping our view of modern
mathematics by uncovering a hidden layer of structure in mathematics, revealing richer and
more robust structures capable of describing more complex phenomena. Categorification is
a powerful tool for relating various branches of mathematics and exploiting the common-
alities between fields. It provides a language emphasizing essential features and allowing
precise relationships between vastly different fields.
This volume focuses on the role categorification plays in geometry, topology, and physics.
These articles illustrate many important trends for the field including geometric represen-
tation theory, homotopical methods in link homology, interactions between higher repre-
sentation theory and gauge theory, and double affine Hecke algebra approaches to link
homology.
The companion volume (Contemporary Mathematics, Volume 683) is devoted to cate-
gorification and higher representation theory.
ISBN 978-1-4704-2821-1
9 781470 428211
CONM/684
AMS

(Anna Beliakova, Aaron D. Lauda) Categorification (B-Ok - Xyz)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Anna Beliakova, Aaron D. Lauda) Categorification (B-Ok - Xyz)

Uploaded by

Copyright:

Available Formats

684

American Mathematical Society

American Mathematical Society

Library of Congress Cataloging-in-Publication Data

The emergent mathematical philosophy of categoriﬁcation is reshaping our view

This volume was inspired by the conference Categoriﬁcation in Algebra, Geom-

Geometry and categorification

• In Section 2, we consider categories of coherent sheaves and algebraic K-

for any short exact sequence

(5) · · · −→ Li+1 f ∗ G −→ Li f ∗ E −→ Li f ∗ F −→ Li f ∗ G −→ Li f ∗ E −→ Li−1 f ∗ E −→ · · ·

sheaf has a finite length resolution.

point x ∈ X and a complete flag V1 ⊂ V2 ⊂ · · · ⊂ Vn = Ex . The cohomology ring is

This sum is well-defined because we have that

Thus, we have that

The Chern character of Vλ , considered as a T-equivariant coherent sheaf on a

3. The function-sheaf correspondence

consider the ètale topology.

reasonable schemes, this is the same by [Mil13, 10.2].

polynomials with integral coeﬃcients)9 . We can consider the base-change of this

defines a map from the Grothendieck group K(Sh(X ⊗ Fp )) to Q -valued functions on

made to work for any variety over the complex numbers.

Theorem 3.7. For any map between Fp -schemes f : X → Y, we have Φ(n) ( f! F ) =

particular, if f : X → Spec Fp , and F = Q , then we find that

While it might seem strange, this theorem is thoroughly topological in nature:

Now, consider the diﬀerence

which is isomorphic to H∗ (Fl(m)) tensored with the constant sheaf on Gr(r ⊂ r +

modulo the action of the group

This is shown in [KS12, 7.3.5] to be equivalent to the more general definition.

where ∩ is the usual intersection product of homology classes.

equivalent to H 0 /hH 0 having support on the scheme M ×N M.

• Nakajima quiver varieties [Nak94, Nak98] (also discussed below), which

4.2. Quiver varieties. Nakajima defined a remarkable set of varieties attached

with various minor caveats.

Theorem 4.10. We have a surjective algebra map U(g) → Htop BM

Department of Mathematics, University of Virginia, Charlottesville, Virginia

A geometric realization of modiﬁed quantum algebras

Abstract. A geometric construction of Lusztig’s modiﬁed quantum algebra

2010 Mathematics Subject Classiﬁcation. Primary 17B37, 14L30, 14F05, 14F43.

(6) Up to an assumption, the complexes involved in this construction are

1λ 1λ = δλ,λ 1λ , ∀λ, λ ∈ X;

where we set m = 1 − i · j in the q-Serre relations (U̇e) and (U̇f) and

Let A = Z[v, v −1 ] be the subring of Laurent polynomials in Q(v). Let AU̇ be

A sequence, (En , fn )n∈N , of smooth resolutions of X:

complex L ∈ D + (X) satisfying e∗ (L) ē∗ (K).

−D (Ēn |en ) is deﬁned to be the category whose

Similarly, we deﬁne the bounded (resp. bounded above; unbounded) G-equi-

distinguished triangle for all n ∈ N.

Recall from [BL94, Part I, 3.2] that there exists a functor

In fact, one can deﬁne a morphism,

Suppose that D is a fourth triangulated category with a thick subcategory

This lemma follows since by Lemma 3.16 we have F2 F1 = Q ◦ f2 f1 ◦ Q! =

Elements in EΩ (D, V ) will be denoted by X = (x, q) where x (resp. q) is in the ﬁrst

structure inherited from that of DG b

Fi . So Qj∗ Qj (K) ∈ Ni . Thus, Qi Qj∗ Qj (K) = 0. If K1 → K → K2 → is a distin-

In general, let us order the vertex set I as i1 , · · · , in . Deﬁne a sequence of thick

where f1 and f2 are the ranks of EΩ (D, V s , V t ) and EΩ (D, V 1 , V 2 , V 3 ) over

Consider the following cartesian diagram

where r123 = r12 + r23 . Let F1 = EΩ∪Ω (D, V 1 , V 3 ) and

= W! (Y2∗ K ⊗ Y1∗ L ⊗ T3!

Let F2 = EΩ∪Ω (D, V 2 ) × EΩ\Ω (D, V 2 ). Thus, F = F1 × F2 . Each component

(29) = W! Δ! (Δ∗ Y2∗ K ⊗ Δ∗ Y1∗ L ⊗ S ∗ L13 )[r13 ]

By combining the above analysis, we have

where the second equality is due to the following computation.

deﬁned in exactly the same manner and Fi,a

For u ≥k v ≥ w, deﬁne the 2-isomorphism Fu,v,w : F (ϕv,w ) ×F (v) F (ϕu,v ) →

F )(ϕu,w ) = F (ϕu,w ) F (ϕu,w ) is deﬁned to be Fu,v,w on the ﬁrst component