To where Knuth paths may lead

Anders Björner

Institut Mittag-Leffler
Stockholm, Sweden

Piteå, 2018
Knuth paths

I The Art of Computer Programming

I Concrete Mathematics (with R. Graham and O. Patashnik)
Knuth paths
Young tableaux

P
Partition λ = (λ1 , . . . , λk ), λ1 ≥ . . . ≥ λk , |λ| = λi , λ ` n

Represented by diagram with λi boxes in row i

Example: λ = (5, 3, 3, 2), |λ| = 13

Young tableaux

1 2 3 5 13
4 7 9
6 8 11
10 12

Standard Young tableau (SYT) of shape λ = filling of the n boxes

of the λ diagram by entries 1, 2, . . . , n so that rows and columns
are strictly increasing.
Young tableaux

1 2 3 5 13
4 7 9
6 8 11
10 12

Standard Young tableau (SYT) of shape λ = filling of the n boxes

of the λ diagram by entries 1, 2, . . . , n so that rows and columns
are strictly increasing.
fλ = number of standard Young tableaux of partition shape λ
Example: f(3,2) = 5
1 2 3 1 2 4 1 2 5 1 3 4 1 3 5
4 5 3 5 3 4 2 5 2 4
RS insertion algorithm for x=35214
Pi Qi

Step1 : 3 1

Step2 : 3 5 1 2

2 5 1 2
Step3 :
3 3

1 5 1 2
Step4 : 2 3
3 4

1 4 1 2
Step5 : 2 5 3 5
3 4
RS correspondence

1 4 1 2
x = 35214 ↔ 2 5 3 5
3 4

x ↔ (P(x), Q(x))

Theorem (Robinson-Schensted) The mapping x 7→ (P(x), Q(x)) is

a bijection between permutations x ∈ Sn and pairs of standard
tableaux of same shape (P, Q).
RS correspondence

1 4 1 2
x = 35214 ↔ 2 5 3 5
3 4

x ↔ (P(x), Q(x))

Theorem (Robinson-Schensted) The mapping x 7→ (P(x), Q(x)) is

a bijection between permutations x ∈ Sn and pairs of standard
tableaux of same shape (P, Q).
Don [TAoCP, Vol. 3]: "RS correspondence has magical properties,
one is lead to suspect witchcraft . . . " For instance,

x −1 ↔ (Q(x), P(x))
RSK correspondence

Theorem (Knuth)
I Vast generalization: Bijections between N-matrices
and pairs of semistandard Young tableaux.
RSK correspondence

Theorem (Knuth)
I Vast generalization: Bijections between N-matrices
and pairs of semistandard Young tableaux.
I Characterization of equivalence relation P(x) = P(y )
via "elementary moves", creating Knuth paths
connecting x and y .
RSK correspondence

Theorem (Knuth)
I Vast generalization: Bijections between N-matrices
and pairs of semistandard Young tableaux.
I Characterization of equivalence relation P(x) = P(y )
via "elementary moves", creating Knuth paths
connecting x and y .

Two important applications, among many:

I Bender-Knuth
I Kazhdan-Lusztig
RSK correspondence

9 5 5 3 2
9 4 4
6 3 1
5 2

I The Bender-Knuth bijective proof of MacMahon’s amazing

formula for the number of plane partitions of n:
1
Σk≥0 PP(k)x k =
(1 − x)(1 − x 2 )2 (1 − x 3 )3 · · ·
Kazhdan-Lusztig representations

24

14 3 2

13

I Proof of the amazing fact that the Kazhdan-Lusztig (1979)

construction of representations for the symmetric groups
produces exactly the irreducible rep’s, correctly labeled.
Kazhdan-Lusztig representations

4321

4312 4231 3421

3412
4132 3241
4213 2431

2413 3214 2341

1432 4123 3142

1342 3124
1423 2314
2143

1243 1324 2134

1234

Bruhat partial order on permutations S4 , generated by inversions.

Kazhdan-Lusztig representations

45123

35124

25134 34125

24135

K-L cell, coming from representation of Hecke algebra

(q-analog of Sn ).
Kazhdan-Lusztig representations

24

14 2

13

K-L polynomial, non-Bruhat edges

Kazhdan-Lusztig representations

2 3
4
1
2

1 4

3 2

4 3

Here, bidirected double edges ↔ elementary Knuth moves

⇒ K-L cell connnected via Knuth paths.
Kazhdan-Lusztig representations
1 2 3
4 5

1 2 4
3 5

1 3 4 1 2 5
2 5 3 4

1 3 5
2 4

The whole picture (almost) can via Knuth paths be translated from
Hecke algebra into Knuth equivalence classes (SYT tableaux of
fixed shape), etc. BUT, ∃ one obstacle to entirely combinatorial
construction: the non-Bruhat edges depend on the K-L
polynomials, and they are difficult ...
Kazhdan-Lusztig representations

24 24

124 23 34 245

134 235
14 25

145 35 13 125

135 135

The K-L graph for λ = (3, 2, 1).

Representing integer matrices can be read off from the informaton
given by graph.
The FKG inequality

Next topic: A sharper FKG inequality.

Correlation inequality, arose in statistical physics
(Ising ferromagnet model, random cluster model, ... )
The FKG inequality

Next topic: A sharper FKG inequality.

Correlation inequality, arose in statistical physics
(Ising ferromagnet model, random cluster model, ... )
Don [TAoCP Vol 4, Prefascicle 5A, Math. Preliminaries Redux] :

“Many other important inequalities relating to expected

values have been discovered, of which the most significant
for our purposes in this book is the FKG inequality.”
The FKG inequality

Next topic: A sharper FKG inequality.

Correlation inequality, arose in statistical physics
(Ising ferromagnet model, random cluster model, ... )
Don [TAoCP Vol 4, Prefascicle 5A, Math. Preliminaries Redux] :

“Many other important inequalities relating to expected

values have been discovered, of which the most significant
for our purposes in this book is the FKG inequality.”
FKG inequality review
456

356

346 256

345 246 156

245 236
146

Review of concepts: 235 145 136

• L = (∧, ∨) finite distributive lattice 234 135 126

x ∨ (y ∧ z) = (x ∨ y ) ∧ (x ∨ z), and dually
134 125

124

123
FKG inequality review
456

356

346 256

345 246 156

245 236
146

Review of concepts: 235 145 136

• L = (∧, ∨) finite distributive lattice 234 135 126

x ∨ (y ∧ z) = (x ∨ y ) ∧ (x ∨ z), and dually
134 125
• rank function rank : L → Z+ 124

123
FKG inequality review
456

356

346 256

345 246 156

245 236
146

Review of concepts: 235 145 136

• L = (∧, ∨) finite distributive lattice 234 135 126

x ∨ (y ∧ z) = (x ∨ y ) ∧ (x ∨ z), and dually
134 125
• rank function rank : L → Z+ 124

• µ:L→ R+ is log-supermodular if
123

µ(x)µ(y ) ≤ µ(x ∧ y )µ(x ∨ y ), for all x, y ∈ L.

FKG inequality review

def P
For functions k, µ : L → R let Eµ [k] = x∈L k(x)µ(x).

Theorem (Fortuin, Kasteleyn and Ginibre, 1971)

Let L be a distributive lattice, µ : L → R+ log-supermodular, and
g and h increasing functions g , h : L → R+ . Then

Eµ [g ] · Eµ [h] ≤ Eµ [1] · Eµ [gh].

Special cases of the FKG inequality

1. If a1 ≤ . . . ≤ an and b1 ≤ . . . ≤ bn , then
n
X n
X n
X
ai · bi ≤ n · ai bi
i=1 i=1 i=1

(Chebychev)
def
2. Let A and B be two families of subsets of [n] = {1, 2, . . . , n},
both closed under taking subsets.

kAk kBk kA ∩ Bk
· n ≤
2n 2 2n
(Kleitman)
Shows: monotone events are positively correlated.
A polynomial strengthening (q-analog) of FKG
inequality
For functions k, µ : L → R let
def
X
Eµq [k] = k(x)µ(x)q rank(x) ∈ R[q].
x∈L
A polynomial strengthening (q-analog) of FKG
inequality

For functions k, µ : L → R let

def
X
Eµq [k] = k(x)µ(x)q rank(x) ∈ R[q].
x∈L
Theorem (q-FKG)
Let L be a distributive lattice, µ : L → R+ log-supermodular, and g
and h increasing functions g , h : L → R+ . Then

Eµq [g ] · Eµq [h]  Eµq [1] · Eµq [gh]

A polynomial strengthening (q-analog) of FKG
inequality

For functions k, µ : L → R let

def
X
Eµq [k] = k(x)µ(x)q rank(x) ∈ R[q].
x∈L
Theorem (q-FKG)
Let L be a distributive lattice, µ : L → R+ log-supermodular, and g
and h increasing functions g , h : L → R+ . Then

Eµq [g ] · Eµq [h]  Eµq [1] · Eµq [gh]

I.e., coefficient-wise inequality of real polynomials.

Application 1: f -vectors of simplicial complexes

∆1 and ∆2 simplicial complexes on the same vertex set V .

What can be said about the face numbers of intersection ∆1 ∩ ∆2 ?
Application 1: f -vectors of simplicial complexes

∆1 and ∆2 simplicial complexes on the same vertex set V .

What can be said about the face numbers of intersection ∆1 ∩ ∆2 ?

fi (∆) = number of i-dimensionalPfaces

f -vector/f -polynomial f∆ (q) = i≥0 fi−1 (∆) q i ,
Application 1: f -vectors of simplicial complexes

∆1 and ∆2 simplicial complexes on the same vertex set V .

What can be said about the face numbers of intersection ∆1 ∩ ∆2 ?

fi (∆) = number of i-dimensionalPfaces

f -vector/f -polynomial f∆ (q) = i≥0 fi−1 (∆) q i ,

Theorem
f∆1 (q) · f∆2 (q)  (1 + q)|V | · f∆1 ∩∆2 (q)

I.e., coefficient-wise inequality of real polynomials.

Note: q = 1 gives Kleitman’s inequality
Appl. 2: Positive correlation wrt Plancherel measure

Young’s lattice Y
Partitions ordered by inclusion of diagrams. A distributive lattice.
Appl. 2: Positive correlation wrt Plancherel measure

For the symmetric group G = Sn :

• set of irreducible representations Irr(Sn ) ↔ {partitions of n}
• deg(λ) = fλ , for partitions λ
Appl. 2: Positive correlation wrt Plancherel measure

For the symmetric group G = Sn :

• set of irreducible representations Irr(Sn ) ↔ {partitions of n}
• deg(λ) = fλ , for partitions λ
fλ2
• Function Irr(G ) → R given by λ 7→ n! known as Plancherel
measure on Irr(G ).
• Extend to measure on all partitions (thus on Y ) by poissonization

def θ|λ| fλ2

λ 7→ πθ (λ) = e −θ
(|λ|!)2

Here θ > 0 is a real parameter.

Appl. 2: Positive correlation wrt Plancherel measure

Asymtotics of these measures much studied. E.g.,

I measure concentration phenomenon known (limit shape of
typical partition diagram),

Early references: Logan-Shepp, Vershik-Kerov,

Borodin-Okounkov-Olshanski, . . .
Appl. 2: Positive correlation wrt Plancherel measure

Asymtotics of these measures much studied. E.g.,

I measure concentration phenomenon known (limit shape of
typical partition diagram),
I close connections to question of "shape" of random
permutation (e.g. longest increasing subsequence),

Early references: Logan-Shepp, Vershik-Kerov,

Borodin-Okounkov-Olshanski, . . .
Appl. 2: Positive correlation wrt Plancherel measure

Asymtotics of these measures much studied. E.g.,

I measure concentration phenomenon known (limit shape of
typical partition diagram),
I close connections to question of "shape" of random
permutation (e.g. longest increasing subsequence),
I and to distribution of eigenvalues of random matrices.

Early references: Logan-Shepp, Vershik-Kerov,

Borodin-Okounkov-Olshanski, . . .
Appl. 2: Positive correlation wrt Plancherel measure

Proposition
Suppose that 0 ≤ s ≤ t. Then the function

fλt
λ 7→
(|λ|!)s

is log-supermodular on the Young lattice.

Proof. Use hook-length formula. 2

Appl. 2: Positive correlation wrt Plancherel measure

Proposition
Suppose that 0 ≤ s ≤ t. Then the function

fλt
λ 7→
(|λ|!)s

is log-supermodular on the Young lattice.

Proof. Use hook-length formula. 2

s = 1, t = 2 case Plancherel measure

s = t = 2 case its poissonization
Appl. 2: Positive correlation wrt Plancherel measure

Let µ be Plancherel measure or its poissonization.

For functions k : Y → R+ let
def
X
Fµq [k] = k(λ) µ(λ)q |λ| ∈ R+ [[q]].
λ∈Y
Appl. 2: Positive correlation wrt Plancherel measure

Let µ be Plancherel measure or its poissonization.

For functions k : Y → R+ let
def
X
Fµq [k] = k(λ) µ(λ)q |λ| ∈ R+ [[q]].
λ∈Y
Theorem
If g , h : Y → R+ are increasing functions (w.r.t. containment of
partition shapes), then

Fµq (g ) · Fµq (h)  Fµq (1) · Fµq (gh).

I.e., coefficient-wise inequality of real formal power series.
Appl. 2: Positive correlation wrt Plancherel measure

Let µ be Plancherel measure or its poissonization.

For functions k : Y → R+ let
def
X
Fµq [k] = k(λ) µ(λ)q |λ| ∈ R+ [[q]].
λ∈Y
Theorem
If g , h : Y → R+ are increasing functions (w.r.t. containment of
partition shapes), then

Fµq (g ) · Fµq (h)  Fµq (1) · Fµq (gh).

I.e., coefficient-wise inequality of real formal power series.
Proof of Theorem. From the q-FKG inequality and
log-supermodular proposition. truncating the formal power series to
polynomials on lower intervals. 2
Happy birthday!