2407.08220v1

CONFORMAL FIELD THEORY OF GAUSSIAN FREE FIELDS
IN A MULTIPLY CONNECTED DOMAIN
TOM ALBERTS, SUNG-SOO BYUN, AND NAM-GYU KANG
Abstract. We implement a version of conformal field theory (CFT) that gives a connection to SLE in a
multiply connected domain. Our approach is based on the Gaussian free field and applies to CFTs with central
charge c ≤ 1. In this framework we introduce the generalized Eguchi-Ooguri equations and use them to derive
arXiv:2407.08220v1 [math-ph] 11 Jul 2024
the explicit form of Ward’s equations, which describe the insertion of a stress tensor in terms of Lie derivatives
and differential operators depending on the Teicmüller modular parameters. Furthermore, by implementing the
BPZ equations, we provide a conformal field theoretic realization of an SLE in a multiply connected domain,
which in particular suggests its drift function, and construct a class of martingale observables for this SLE
process.
1. Introduction and Main results

In his classical work [34], Koebe introduced several classes of the so-called canonical multiply connected
domains. By their very definition, the canonical domains can be uniquely determined by specifying a finite
number of parameters in the moduli space. From the complex geometry viewpoint, these play a role of the
Teichmüller parameters [28] that describe the deformation of a surface within a particular homeomorphism
class. Compared to simply connected domains, one of the principal difficulties arising in a multiply connected
domain is that the non-trivial topology of the domain can significantly impact the behavior of the functions
defined on it. For instance, the Hadamard variational formula, which describes the infinitesimal changes of the
Green’s function under specific perturbations of the underlying domain, requires capturing the deformations of
each boundary component of a multiply connected domain, see e.g. [37].
These same difficulties persist in studying conformal field theory (CFT) in multiply connected domains. In
general, one of the main purposes of CFT is to compute correlation functions of a family of conformal fields,
such as the operator product expansion (OPE) family generated from the basic fields. These describe the
statistical correlations of certain physical observables in a system and can sometimes be expressed in terms of
geometric functions of the underlying domain. In addition to computing correlation functions, CFT also aims
to derive certain functional equations among them, which provide powerful tools to predict the critical behavior
of the statistical physics models. As with the philosophy above, one expects that the description of correlation
functions, as well as their functional equations, reveal significant differences when developing the CFT in a more
complicated geometry as they may depend on its moduli space in a non-trivial way.
In this paper, we implement a version of CFT with central charge c ≤ 1 to present its link to a forward
SLE in a multiply connected domain. The fields that we consider are all constructed from the Gaussian
free field (GFF) with Dirichlet boundary condition, a Gaussian random field whose correlations depend on
the geometry of the domain through its Green’s function. In this theory, there is a well-established way of
computing correlation functions based on Wick’s formula, operator product expansions, and the Coulomb gas
formalism. There is also an infinitesimal calculus for perturbations of CFT correlation functions, which allows
one to express differential equations satisfied by the correlation functions without having explicit expressions
Date: July 12, 2024.

Tom Alberts was supported by NSF grant DMS-1811087, by Simons Foundation grant MPS-TSM-00002584, and by the KIAS
Scholar program. Sung-Soo Byun was supported by the POSCO TJ Park Foundation (POSCO Science Fellowship), by the New
Faculty Startup Fund at Seoul National University and by the National Research Foundation of Korea funded by the Korea govern-
ment (NRF-2016K2A9A2A13003815, RS-2023-00301976). Nam-Gyu Kang was supported by by Samsung Science and Technology
Foundation (SSTF-BA1401-51) and by the National Research Foundation of Korea (NRF-2019R1A5A1028324), and by a KIAS
Individual Grant (MG058103) at Korea Institute for Advanced Study.
1
2 TOM ALBERTS, SUNG-SOO BYUN, AND NAM-GYU KANG
for the functions themselves. Fundamental examples of this calculus include Ward’s equations and the Belavin-
Polyakov-Zamolodchikov (BPZ) equations. The former expresses the perturbations of correlation functions
to the correlation function of the underlying random field when computed after the insertion of a stress
tensor. The latter equates a very specific perturbation of correlation functions to the correlation function of
the underlying random field when computed after the insertion of the one-leg operator. The BPZ equations, in
particular, provide the precise form of the null-vector equation, a differential equation satisfied by the partition
function that encodes important properties of the system. In a simply connected domain, such a theory and its
application to the theory of Schramm Loewner evolution (SLE) has been well-developed, see e.g. [3, 4, 25, 31–
33, 38, 44] and references therein. We also refer to [9, 26, 42] for reviews.
Let us turn our attention to Ward’s equation, which best features the dependence of non-trivial moduli
space. As in classical differential geometry, the Lie derivative of a conformal field describes the infinitesimal
change of the field when it is transported along the flow of a vector field. In CFT, this provides a characteristic
property of a stress tensor (if it exists). It is well known that the Gaussian free field Φ has a stress tensor of
the form
1
A = − J ⊙ J, J = ∂Φ,
2
where ⊙ is Wick’s product, see e.g. [31]. In a simply connected domain, Ward’s equation states that the insertion
of a stress tensor is equivalent to taking a Lie derivative with respect to a proper vector field with a prescribed
pole. For instance, for a holomorphic differential (a conformally covariant field) X in the OPE family of the
GFF with conformal dimension λ, Ward’s equation states that
2
(1.1) 2 E A(ζ)X(z) = kζ (z)∂z + λ kζ′ (z) E X(z), kζ (z) = ,
ζ −z
where all fields are evaluated in the upper-half plane H. Here, the right-hand side of this identity is a Lie
derivative of X with respect to the Loewner vector field kζ in H. Ward’s equation can be generalized to a tensor
product of fields in the OPE family of the GFF.
If the underlying domain is not simply connected, the contributions to Ward’s equation from each boundary
component (except for the boundary component where an SLE grows) must be taken into account. To provide
a clearer illustration let us consider a doubly connected domain as an example. For given modular parameter
r > 0, we take the cylinder {z ∈ C : 0 < Im z < r}/⟨z 7→ z + 2π⟩ as a reference domain. Then in this cylinder
uniformization, Ward’s equation for a tensor product X of fields in the OPE family of the GFF reads as
Z
1
(1.2) 2 E A(ζ)X = Lvζ E X + E A(ξ)X dξ
π γ
(1.3) = Lvζ E X + ∂r E X,
where ζ ∈ [−π, π], γ = [ir − π, ir + π] is the upper boundary component, and vζ is the Loewner vector field
in the cylinder. Compared to (1.1), Ward’s equation of the form (1.2) clearly shows the contribution from a
boundary component. On the one hand, the identity (1.3) reveals an interesting (and indeed far from being
obvious) relation between the boundary integral under the insertion of a stress tensor and the differentiation
with respect to the modular parameter. In addition, this leads to a differential form of the BPZ equation, which
also plays a crucial role in the theory of annulus SLE, cf [10]. Ward’s equation (1.3) was introduced in the
pioneering work of Eguchi and Ooguri [21] in the complex torus of genus one using a path integral formalism.
A particular advantage of considering a doubly connected domain or a complex torus is that it enables explicit
computations using well-known special functions such as Jacobi theta functions. With this strategy, (1.3) was
rigorously proved in [10, 32] using Wick’s calculus and several functional equations between special functions.
Despite the extensive study of the Eguchi-Ooguri version of Ward’s equation for the genus one case, a
corresponding formulation for a multiply connected domain has not, to our knowledge, been proposed in either
the physics or mathematics literature. This slow progress may be attributed, at least in part, to the challenges
arising from the higher dimensional moduli space in such domains and the absence of explicit formulas for the
classical special functions. In this paper, we address this problem by developing a formalism for the Eguchi-
Ooguri version of Ward’s equation in a multiply connected domain. In addition to this, by implementing the
BPZ equations, we introduce the SLE drift function from the viewpoint of CFT and construct a large class of
the associated SLE martingale observables.
CONFORMAL FIELD THEORY IN A MULTIPLY CONNECTED DOMAIN 3
1.1. Basic setup. To introduce our results we first recall some basic notions of multiply connected domains and
CFT. Among the canonical multiply connected domains, the chordal standard domains are the simplest setting
for describing CFT and its applications to SLE. By definition, a chordal standard domain is the upper-half
plane H minus a finite number of horizontal slits [zkl , zkr ], with k = 1, . . . , g and Im zkl = Im zkr > 0. Often we
denote a chordal standard domain by Hg , with the notation emphasizing the number of slits but suppressing
their locations. By a well-known uniformization theorem, any multiply connected domain D is conformally
equivalent to a chordal standard domain.
Our formulation of CFT crucially relies on the Green’s function GD of a domain D, with zero Dirichlet
boundary condition. Letting G e D (·, z0 ) be the harmonic conjugate of GD (·, z0 ) we define the complex Green’s
functions by
GD + iG eD
G+D := , G− +
D = GD .
2
Then a multivalued holomorphic function G+ D (·, z0 ) is defined up to periods. In a general multiply connected
domain the Green’s function can be expressed in terms of a special transcendental function called the Schottky-
Klein prime function [2, 15]. In Subsection 2.2 we present another representation of the Green’s function using
the Riemann theta function and the Schottky double construction.
The basic random field that underlies our CFT is the GFF. We denote it by Φ and consider it as a formal
random function Φ : D → R that satisfies with Dirichlet boundary conditions on each boundary component.
Occasionally we add the subscript (0) to Φ, which indicates that we consider the central charge c = 1 theory.
We extend this to c ≤ 1 later on. The 1- and 2-point correlation functions of Φ are given by
E Φ(ζ) = 0, E Φ(ζ)Φ(z) = 2GD (ζ, z).
As Φ is a Gaussian field these formulas characterize its law, which is conformally invariant due to the conformal
invariance of the Green’s function. The Gaussian law also allows one to apply Wick’s formulas to polynomial
functions of the formal collection of random variables {Φ(z) : z ∈ D} and their derivatives, see [30] for an
exposition of Wick’s formulas.
Correlation functions in CFT are then computed via a combination of Wick’s formulas and the operator
product expansion. OPE is an important binary operation on conformal fields, serving as a tool for analyzing
the algebraic structure of CFT. If a field X is holomorphic, then the OPE is defined as a formal Laurent series
expansion near the diagonal within correlations:
X
X ∗n Y (z) (ζ − z)n ,

X(ζ)Y (z) =: as ζ → z.
In particular, the zeroth coefficient, which we simply denote by X ∗ Y ≡ X ∗0 Y , is called the OPE product of
X and Y . In other words, the field X ∗ Y is obtained by subtracting the divergent terms in the OPE of the
fields X and Y and then taking the limit as ζ → z. This is a standard technique used in field theory to extract
finite results from otherwise divergent calculations. For instance, the OPE of two GFFs is given by
Φ ∗ Φ(z) = 2u(z, z) + Φ(z) ⊙ Φ(z), u(ζ, z) := GD (ζ, z) + log |ζ − z|.
Here the function u(z, z) is called the domain constant for the Dirichlet boundary condition, see e.g. [5, Section
4]. The OPE family of Φ is the algebra over C spanned by the generators 1, all mixed derivatives of Φ, and the
so-called OPE exponentials, under the operations (+, ∗).
To implement CFT with central charge c ≤ 1, one needs to define the central/background charge modification
of Φ. For this purpose, given points qk on a boundary component C0 and a parameter b ∈ R, we consider a
divisor
X
β= βk · qk ,
with the neutrality condition
Z
(1.4) β = bχ
where χ = 2 − 2g is the Euler characteristic of a compact Riemann surface of genus g.

Then in terms of a conformal map w : (D, C0 , q) 7→ (Hg , R, ∞), which is uniquely determined if we impose
hydrodynamic normalization at infinity, we define the background charge modification Φβ of Φ as
X
(1.5) Φβ (z) = Φ(z) − 2b arg w′ (z) − 2 βk Im G+
D (w(qk ), w(z)).
In particular, if β ∞ = bχ · q, then
Φβ∞ (z) = Φ(z) − 2b arg w′ (z).
This form (1.5) of the background charge modification is induced from the Schottky double construction of
the GFF and CFT on a compact Riemann surface of genus g. In particular, the neutrality condition comes from
a version of the Gauss-Bonnet theorem, see Subsection 3.4. This modification gives rise to CFT with central
charge
(1.6) c = 1 − 12b2 .
From Φβ we also construct the OPE family Fβ , the algebra of conformal fields (over C) that is generated by
the constant field 1, all mixed derivatives of Φβ , and the OPE exponentials. The addition of the algebra is the
regular addition of fields, while the multiplication is the OPE multiplication. The OPE family Fβ is the natural
family of fields for which we can develop an infinitesimal calculus for the correlation functions and make the
connection with SLE theory.
To develop the infinitesimal calculus we now recall the definition of the Lie derivatives and a stress tensor.
For a conformal Fock space field X, the Lie derivative Lv X of a smooth vector field v is defined by
d
(Lv X∥ϕ) = (X∥ϕ ◦ ψ−t ),
dt t=0
where ψt is a local flow of v, and ϕ is a given local chart. We further define its C-linear and anti C-linear parts
L±v by
Lv − iLiv Lv + iLiv
L+v := , L−
v := .
2 2
The Lie derivative of a conformal Fock space field can be computed from the field’s transformation law. In
particular, consider a differential X of conformal dimension [λ, λ∗ ], which means that its transformation law
satisfies the covariance rule
X = (h′ )λ (h′ )λ∗ X
e ◦ h,
where X = (X||ϕ) and X̃ = (X||ϕ̃) is the evaluation of X on different charts ϕ, ϕ̃, and h is the holomorphic
transition map between two overlapping charts ϕ, ϕ̃. The Lie derivatives of a (λ, λ∗ )-differential are
L+ X = (v∂ + λv ′ )X,
v L− X = (v̄ ∂¯ + λ∗ v ′ )X.
v
As previously mentioned, the Lie derivatives are closely related to the notion of a stress tensor in CFT. By
definition, a conformal field X has a stress tensor A if
• A is a holomorphic quadratic differential, and
• the residue form of Ward’s identity
I I
1 − 1
L+v X(z) = vAX(z), Lv X(z) = − v̄ ĀX(z)
2πi (z) 2πi (z)
holds for all smooth vector fields v.
It can be shown that all elements of the OPE family Fβ (which is generated from the Gaussian free field Φβ )
have the common stress tensor
1
Aβ := − J ⊙ J + ib∂ − jβ J, jβ = E Jβ .
2
Here Jβ = ∂Φβ and J = ∂Φ. Although Aβ is not a member of the OPE family Fβ , the Virasoro field
1
Tβ := − Jβ ∗ Jβ + ib ∂Jβ = Aβ + E Tβ
2
is in Fβ . The first equality is by definition while the second is a computation in OPE calculus. It can be shown
that the deterministic term E Tβ is a Schwarzian form of order c/12, where c is given by (1.6).
1.2. Main results. We now introduce our results. Let vζ,Hg be the Loewner vector field with sole simple pole
at ζ in a chordal standard domain Hg . In general, the Loewner vector field in a multiply connected domain
can be expressed in terms of the complex excursion reflected Poisson kernel. Let γ : [0, ∞) → Hg be a simple
curve with γ(0) ∈ R. Then by [6, Theorem 3.1] (see also [17, Theorem 6.14]), the family of conformal mappings
gt : Hg \ γt → Hg (t), where Hg (t) is a chordal standard domain, satisfies the chordal Loewner equation
(1.7) ∂t gt (z) = −vξt ,Hg (t) (gt (z)), (g0 (z) = z).
See Subsection 2.3 for more details on the Loewner vector field.
Recall that a chordal standard domain Hg is characterized by the horizontal slits [zkl , zkr ] with k = 1, . . . , g.
In terms of the Teichmüller parameters {zkl , zkr }, we define the differential operator
Xg
(1.8) ∇Hg := vζ (zkl )∂zkl + vζ (z̄kl )∂¯zkl + vζ (zkr )∂zkr + vζ (z̄kr )∂¯zkr .
k=1
Here, we shall use the convention ∇Hg = 0 if g = 0. We further define
X
(1.9) ∇Hg ,q := ∇Hg + vζ (qk )∂qk ,
k
where ∂qk = ∂ + ∂¯ is the differential operator with respect to the real variable qk . Note that while the notation
∇Hg = ∇Hg ,0 treats the underlying domain as a chordal domain Hg without marked points (thus ∇Hg gives the
g ≥ 1 counterpart of the operator ∂r in (1.3) for the g = 1 case), the operator ∇Hg ,q emphasizes the underlying
geometry as (Hg , q), a domain with marked points. The latter is more natural in the context of CFT with
background charge modifications. We intentionally introduce the two operators ∇Hg and ∇Hg ,q , as their origins
in Ward’s equation (1.13) below are different: the first one comes from the generalized Eguchi-Ooguri equation,
whereas the second comes from the residue form of Ward’s identity at each of the points in q. This distinction
also makes the presentation of the proof in Section 4 clearer.
We remark that the operator ∇Hg ,q also naturally appears in the context of the commutation relation of
SLE in a multiply connected domain due to Dubédat, see [18, p.1835].
We have the following form of Ward’s equation.
Theorem 1.1 (Ward’s equations of Eguchi-Ooguri type). Fix a background charge β supported on R,
and assume it satisfies the neutrality condition (1.4). For any tensor product X of fields in the OPE family Fβ
of the GFF Φβ , we have

−
2 EAβ (ζ)X = L+ vζ + L vζ̄ EX
(1.10) 1
Z XI
¯ EAβ (ξ)X dξ + 1

− vζ (ξ) − vζ (ξ) vζ (ξ)EAβ (ξ)X
2πi ∂Hg \R 2πi (qk ) k
where all fields are evaluated in the identity chart of Hg . Here, the Lie derivatives act on neither qk ’s nor the
endpoints {zkl , zkr } of slits. Furthermore, the generalized Eguchi-Ooguri equation
Z
1
¯ EAβ (ξ)X dξ + ∇H EX = 0
(1.11) vζ (ξ) − vζ (ξ) g
2πi ∂Hg \R
holds, as does the following residue form of Ward’s identity at marked points of the background charge
I
1 X X
(1.12) vζ (ξ)EAβ (ξ)X = vζ (qk )∂qk EX.
2πi (qk )
k k
Together, (1.12) and the generalized Eguchi-Ooguri equation (1.11) lead to the Ward equation

−
(1.13) 2 EAβ (ζ)X = L+ vζ + Lvζ̄ EX + ∇Hg ,q EX.
Remark 1. In a multiply connected domain the correlation functions EX depends on the the nodes of X (for
example z1 and z2 if X = Φβ (z1 )Φβ (z2 )), the points qk in the support of the background charge β, and also
implicitly on the modular parameters of the domain. By convention, the Lie derivative terms only act on the
nodes of X in formulas like (1.10) and (1.13). However, it is worth noting that the differential operator ∇Hg ,q
can also be interpreted as a Lie derivative operator, if one so desires. This is achieved by regarding the field
EX as a (0, 0)-differential at the modular parameters, i.e. at the support points {qk } of β and at the endpoints
{zkl , zkr } of the slits (and their complex conjugates). From this viewpoint, the insertion of the stress tensor Aβ
into a field in Fβ in a multiply connected domain genuinely acts as a “total” Lie derivative, acting not only the
nodes of X but also the modular parameters. This viewpoint is consistent with both the physical and geometric
perspectives and aligns with the appearance of such terms in [19] via Itô calculus. However, the way in which
the operator ∇Hg ,q arises in our construction of CFT is far from obvious and requires separate treatment, as is
evident from (1.11) and (1.12).
Remark 2. For the simplest case when X = Φ(z1 )Φ(z2 ) and there is no background charge, Ward equations
give rise to the differential form of the Hadamard’s variational formula
d
GDt (gt (z1 ), gt (z2 )) = − Lvp + ∇Hg GD (z1 , z2 )
dt t=0
under the Loewner flow gt : (Dt , γt ) → (D, ξt ), where p = γ0 . We also mention that a similar form of variational
formula can be found in the context of the resolvent kernel on a compact polyhedral surface (a compact
Riemann surface equipped with a flat metric that has conical singularities), see [35, Proposition 2]. It is also
worth noting that many correlation functions can be expressed in terms of special transcendental functions, such
as Schottky-Klein primes or Riemann theta functions, and Ward’s equation can provide non-trivial functional
equations among these special functions.
A natural extension of Ward’s equations are the so-called BPZ equations. There are several distinct ways of
introducing the BPZ equations; in our theory they come about as a special limiting form of Ward’s equations
applied to a particular representation of the correlation functions. For this representation we introduce the
Wick exponential of a multi-point version of the field Φ. This multi-point field, which we denote by Φ{τ }, is
indexed by double divisors τ = (τ + , τ − ) that satisfy the neutrality condition
Z
τ + + τ − = 0.
We write τ ± = τj± · zj (noting that some τj± may be zero), and define Φ{τ } by
P
X
Φ{τ } ≡ Φ{τ + , τ − } = τj+ Φ+ (zj ) − τj− Φ− (zj ).
Here Φ± are formal 1-point bosonic fields that can be thought of as the holomorphic and anti-holomorphic part
of Φ. They are related as
Φ = Φ+ + Φ− , Φ− = Φ+ .
We define the Wick’s exponential V ⊙ {τ } by
V ⊙ {τ } = e⊙iΦ{τ } .
Then correlation functions of the form E V ⊙ {τ }X are well-defined by Wick calculus. A closely related corre-
by E Oβ {τ }X where Oβ {τ } is the OPE exponential, which we now define. Given a
lation function is given P
background charge β = βk · qk , (βk ∈ R, qk ∈ C0 ) the OPE exponential Oβ {τ } is
O{τ } ≡ Oβ {τ } := e∗iΦβ {τ } ,
where X
− −
Φβ {τ } := τj+ Φ+
β (zj ) − τj Φβ (zj )
and X
Φ+ + ′
β (z) = Φ (z) + ib log w (z) + i βk E Φ+ (qk )Φ+ (z), Φ− +
β = Φβ ,
cf. see [10, Eq.(3.28) and p.32]. Here, a formal correlation E Φ+ (ζ)Φ+ (z) and the OPE exponential are defined
in Subsections 3.3 and 3.4. For any choice of τ the OPE exponential Oβ {τ } is in the OPE family Fβ . In this
paper we use it exclusively as an “insertion” type operator, meaning that we multiply other fields with it before
taking the expectation to produce a correlation function. The main effect of this insertion is to insert level two
degeneracy into the correlation functions (at least for some specific choices of τ ), which introduces second order
differential operators into the limiting form of the Ward equation. In this way we obtain the following form of
BPZ equations [8] in a multiply connected domain.
Theorem 1.2 (BPZ and BPZ-Cardy equations). Fix a background charge β supported on a component
C0 of the multiply connected domain Hg , and assume β satisfies the neutrality condition (1.4). For z ∈ C0 that
is distinct from the support points of β let
Υ(z) := Oβ {α}, α = (a · z − a · ∞, 0),
where a ∈ R satisfies 2a(a + b) = 1 and b ∈ R is determined by the neutrality condition (1.4) of β. Let r0 ≡ rD,0
and r1 ≡ rD,1 be the first two coefficients in the holomorphic part of the Loewner vector field’s Laurent expansion,
i.e.
2
vζ (z) = + r0 (z) + r1 (z)(ζ − z) + o(|ζ − z|), ζ → z.
ζ −z
Then for any tensor product X of fields in the OPE family Fβ , the following holds.
(i) (BPZ equation) We have
1
−
h
′
i
(1.14) 2 ∂z2 EΥ(z)X = Ľ+
vz + Lvz̄ + ∇Hg ,q EΥ(z)X + h1,2 r0 (z) − r1 (z) + 2 E Tβ (z) + r0 (z)∂z EΥ(z)X,
a
where all fields are evaluated in the identity chart of Hg . Here, h1,2 = a2 /2 − ab and the check indicates
that the Lie derivative operator Ľ+vz applies to the punctured domain Hg \ {z}.
(ii) (BPZ-Cardy equation) Let ξ ∈ R and
EΥ(ξ)X
EΥ X := = EV ⊙ {α}X.
EΥ(ξ)
Then we have
1 2 2
+ −

(1.15) ∂ ξ E Υ X + ∂ ξ log EΥ(ξ) − r0 (ξ) ∂ξ E Υ X = Ľv + Lv + ∇ Hg ,q EΥ X,
a2 a2 ξ ξ
where ∂ξ = ∂ + ∂¯ is the differential operator with respect to the real variable ξ.

Remark 3 (Null-vector equation). By letting X = 1 in the BPZ equations (1.14), we obtain the null-vector
equation satisfied by EΥ(z):
1 2 h
′
i
(1.16) ∂z EΥ(z) = ∇ Hg ,q EΥ(z) + h 1,2 r0 (z) − r1 (z) + 2 E Tβ (z) + r0 (z)∂z EΥ(z).
a2
Remark 4 (Comparison with the BPZ equations in a simply connected domain). In a simply connected domain,
the associated Loewner vector field is given by
2
kζ (z) =
ζ −z
in the identity chart of H. Then the BPZ equation reads as
1 2
+ −
X
(1.17) ∂ EΥ(z)X = Ľ + L + v ζ (qk )∂q EX + 2 E Tβ (z) EΥ(z)X.
a2 z kz kz̄ k
k
See [31, Proposition 5.13] and [33, Theorems 7.11 and 7.12] for more about the BPZ equations in a simply
connected domain. Compared to (1.17), the BPZ equations (1.14) in a multiply connected domain contain the
geometric terms
h i
∇Hg + h1,2 r0′ (z) − r1 (z) + r0 (z)∂z EΥ(z)X.
Among these, the Teichmüller term ∇Hg naturally comes from the Eguchi-Ooguri type of Ward’s equation
(Theorem 1.1) and captures the evolution of slits. On the other hand, the remaining terms originate from
the infinitesimal changes of the Lie derivatives in the chordal standard domain with respect to those in the
upper-half plane; more precisely
h i
h1,2 r0′ (z) − r1 (z) + r0 (z)∂z EΥ(z)X = lim L+ +
vζ − Lkζ EΥ(z)X.
ζ→z
See Subsection 4.3 for further details.

Remark 5. The function r0 describes the difference of each side of a small vertical slit attached to the real axis
as it evolves under the Loewner flow. To be more precise, if we define
vbζ (z) := vζ (z) − r0 (z)
so that
2
vbζ (z) = + O(ζ − z), ζ → z,
ζ −z
then the image of each side of a vertical slit under the flow generated by vbζ has the same leading-order length
on R, see [5, 6]. See also Remark 6 below for its appearance in the SLE drift functions.
The Schramm-Loewner evolution SLE(κ, Λ) with positive parameter κ and the drift function Λ ≡ ΛD in a
multiply connected domain D is a conformally invariant law on random curves in D described by the Loewner
equation (1.7) and driving process
√
(1.18) dξt = κ dBt + ΛDt (ξt ) dt.
Throughout the paper, we consider the drift function associated with EΥ(ξ) :
(1.19) Λ(ξ) := κ ∂ξ log EΥ(ξ) − r0 (ξ).
Note that Λ implicitly depends on the domain D, the background charge β, and the marked points in α =
(a · ξ + α0 , 0) that are used to define Υ(ξ). The value of a and κ is determined by the neutrality condition of
β. More precisely, if β satisfies neutrality condition (3.14), then κ and a are determined by the bijections
p p p
b = κ/8 − 2/κ, a = 2/κ.
This gives the central charge
(3κ − 8)(κ − 6)
c = 1 − 12b2 = .
2κ
The relationship between SLE and our CFT is by the following notion of a martingale observable. By defini-
tion, a non-random field M is a martingale-observable for the SLE system (1.7) if the process Mt (z1 , . . . , zn ) =
(M(Dt ,γt ,q) ∥ id)(z1 , . . . , zn ) is a local martingale on the SLE probability space, at least until the first time that
any marked point is swallowed. Note that Mt = (M ∥ gt−1 ) and thus Mt becomes random by its evaluation
on the random charts gt , which in turn inherits its randomness from the stochastic evolution equation of the
driving process ξt . Using the BPZ-Cardy equations (1.15), the next result shows that all fields in Fβ lead to
martingale observables for a certain class of stochastic evolution equations for ξt .
Theorem 1.3 (SLE martingale observables). Let the background charge β and the OPE exponential Υ(z)
be defined as in Theorem 1.2. Suppose that X is a string of fields in Fβ . Then the non-random field
M = EΥ X
is a martingale observable for SLE(κ, ΛD ) with drift function ΛD given by (1.19).
Remark 6 (SLE drift function). In general, there are many candidates for SLE drift functions. Our drift function
(1.19) is motivated by the BPZ-Cardy equation (1.15), which leads to the martingale observable property. Our
drift function is also natural from the viewpoint of the GFF/SLE coupling, see also [19,29]. On the other hand,
Dubédat proposed SLE drift functions using the commutation relation [18, Section 9]. More precisely, in terms
of a partition function Z satisfying a null-vector equation similar to (1.16), the drift function of the form
κ ∂ξ log Z(ξ) − r0 (ξ)
was considered, see [18, pp.1834–1835]. Let us also mention that Zhan [46] used the drift function of the form
κ ∂ξ ∂η GD (ξ, η)
3−
2 ∂ξ GD (ξ, η)
to define the harmonic random Loewner chain in a multiply connected domain. This in particular gives the
scaling limit of the loop-erased random walk [47] for κ = 2.
2. Green’s functions and Loewner vector field

In this section we present a concise exposition of the fundamental concepts pertaining to Green’s functions
and Loewner vector fields in a multiply connected domain.
2.1. Special functions. We first collect basic facts on fundamental geometric functions in a multiply connected
domain. To make the presentation more concrete we frequently use either chordal standard domains or circular
domains as a model of a multiply connected domain. The circular domains are more convenient for expressing
the geometric functions such as the Schottky-Klein prime function, while the chordal standard domains are
more useful for describing the Loewner evolutions. There is no loss of generality in this approach since every
multiply connected domain is conformally equivalent to a chordal standard domain.
2.1.1. Chordal Standard Domains and their Schottky Doubles. Our presentation also uses the notion of the
Schottky double of the multiply connected domain. If D is a multiply connected domain then, informally, its
Schottky double is obtained by gluing a copy of D to itself along the boundary components. This makes the
Schottky double a compact Riemann surface with genus equal to the number of “holes” in the original domain
D. In many ways the Schottky double is the natural setting for complex analysis, and analytical results on the
original domain are a shadow of results on the double.
2.1.2. Circular Domains and their Schottky Doubles. We call a domain D circular if it is a finitely connected
domain all of whose boundary components are circles. This domain is often employed as a prototypical example
of a multiply connected domain in geometric function theory. We assume that the outer boundary C0 is the
unit circle and write Cj (j = 1, · · · , g) for each boundary component. Let δj ∈ C and rj > 0 be the centres and
radii of the Cj ’s, i.e.
(2.1) Cj = {ζ ∈ C : |ζ − δj | = rj }.
By assumption the δj and rj are such that Cj ⊂ {z : |z| < 1} for each j, and the Cj , j = 1, . . . , g are all disjoint.
We sometimes write r = (r1 , . . . , rg ), δ = (δ1 , . . . , δg ), and C(r, δ) for the domain
C(r, δ) = {z ∈ C : |z| < 1, |z − δj | > rj for j = 1, . . . , g} .
For a circular domain C(r, δ) the Schottky double has a straightforward representation derived from reflection
across the unit circle. For each j = 1, . . . , g let Cj′ be the circle obtained by the reflection of Cj with respect
to C0 , i.e. Cj′ = {ζ ∗ ∈ C : ζ ∈ Cj }, where ζ ∗ = 1/ζ̄. The 2g circles {Cj , Cj′ : j = 1, . . . , g} are known as the
Schottky circles. The complement of the (union of) the Schottky circles is a model of the Schottky double for
the circular domain. This complement is naturally divided into two halves: the part in the interior of the open
unit circle and then the reflection of this interior set through the unit circle. In the usual nomenclature these
two parts are the “front” and “back” sides of the Schottky double. The double has no boundary since each
point ζ ∈ Cj is identified with its reflection ζ ∗ ∈ Cj′ , and points on the unit circle {|z| = 1} are interior points
of the double.
2.1.3. Schottky-Klein prime function. The Schottky-Klein prime function has many uses, but for our purposes
it is a convenient tool for representing meromorphic functions on a compact Riemann surface. We introduce it
in the context of circular domains. For each j = 0, 1, · · · , g, and with δ0 = 0 and r0 = 1, we define Möbius maps
rj2
(2.2) ϕj (ζ) := δ̄j + .
ζ − δj
Note that ϕj conjugates points on the circle Cj , i.e. for ζ ∈ Cj , ζ̄ = ϕj (ζ). Set
rj2 ζ
(2.3) θj (ζ) := ϕ̄j (ζ −1 ) = δj + .
1 − δ̄j ζ
Then it is easy to see that the image of Cj′ under θj is Cj . The Schottky group Θ is defined to be the infinite free
group of mappings generated by compositions of basic Möbius maps θj and their inverses θj−1 , j = 0, 1, · · · , g.
We refer to [40] for more about Schottky group.
From the Schottky group we construct the Schottky-Klein (S-K) prime function ω(ζ, z). The first step
in the construction is to choose a semigroup Θ′′ within the Schottky group Θ. Thus Θ′′ should be closed
under composition but exclude the identity and all inverse maps. For instance, if θ1 , θ1 θ2−1 ∈ Θ′′ , then θ1−1 ,
θ1−1 θ2 ̸∈ Θ′′ . Given Θ′′ the Schottky-Klein prime function is defined by the infinite product
Y (θ̃(ζ) − z)(θ̃(z) − ζ)
(2.4) ω(ζ, z) := (ζ − z) ω
e (ζ, z), ω
e (ζ, z) := .
′′
(θ̃(ζ) − ζ)(θ̃(z) − z)
θ̃∈Θ
′′
The function ω is well defined even though Θ is not determined uniquely. See [13, Sections 4 and 5] and [27]
for more about S-K prime function. The S-K prime function satisfies the rules
1
(2.5) ω(ζ, z) = −ω(z, ζ), ω(1/ζ̄, 1/z̄) = − ω(ζ, z).
ζz
The skew-symmetry of ω follows from (2.4) and the symmetry of ω
e . For the second rule, we refer to [13,
Appendix A].
Example: S-K prime function on an annulus
On the annulus Ar = {z ∈ C : e−r < |z| < 1} the Schottky group has θ1 (ζ) = e−2r ζ as a generator, and
we may take n o
(k)
Θ′′ = θ1 : k ∈ N
as our semigroup. Therefore, the S-K prime function ω in Ar is represented as
∞
Y (e−2rk ζ − z)(e−2rk z − ζ)
(2.6) ω(ζ, z) = (ζ − z) .
(e−2rk ζ − ζ)(e−2rk z − z)
k=1
2.1.4. Riemann theta function. For a general compact Riemann surface M of genus g, let us fix a canonical
basis {aj , bj } of the homology H1 = H1 (M ) satisfying aj · bk = δjk (recall that a · b is the intersection number
of the curves a and b). On a circular domain the aj can be chosen as the circles Cj , j = 1, . . . , g, each of which
is identified with the circles Cj′ . Each bj can be taken as a curve in the complement of the Schottky circles that
connects a ζ ∈ Cj to ζ ∗ ∈ Cj′ . The space of holomorphic 1-differentials on M is g-dimensional, and there is a
basis {ηj } of this space which is uniquely determined by the equations
I
(2.7) ηj = δjk .
ak
Given the basis {ηj } the period matrix τ is defined as

I
τ = {τjk }, τjk = ηj .
bk
It is well known that the period matrix is symmetric and Im τ is positive definite, see e.g. [22, Chapter III.3].
From the period matrix we construct the lattice Λ = Zg + τ Zg in Cg and denote by
TΛ ≡ TgΛ := Cg /Λ
the Jacobi variety of M . For each p0 ∈ M , we write
Z p
Ap0 ≡ A : M → TΛ , p 7→ (η1 , · · · , ηg )T
p0
for the Abel-Jacobi map.

The Riemann theta function Θ(·|τ ) associated with τ is given as
1
X
Θ ≡ Θ(Z|τ ) := e2πi(Z·N + 2 τ N ·N ) , Z ∈ Cg .
N ∈Zg
The theta function is an even, entire function on Cg and has the periodicity properties
1
(2.8) Θ(Z + N ) = Θ(Z), Θ(Z + τ N ) = e−2πi(Z·N + 2 τ N ·N ) Θ(Z), for N ∈ Zg .
2.2. Green’s function with Dirichlet/excursion reflected boundary condition. We now introduce the
Green’s function of a multiply connected domain. In contrast to the last section we first phrase the discussion
on general domains. Throughout we assume that D is a bounded planar domain whose boundary is g + 1
disjoint, smooth Jordan curves Cj , j = 0, 1 · · · , g. We take C0 to be the outermost boundary. A prototypical
example is given by a circular domain.
For a specified boundary condition and each z ∈ D, the Green’s function ζ 7→ GD (ζ, z) is the unique function
such that GD (ζ, z) + log |ζ − z| is harmonic with respect to ζ throughout the region D, including at z, and that
obeys the prescribed boundary condition.
We shall consider two fundamental boundary conditions: the zero Dirichlet boundary condition and the
excursion reflected (ER) boundary condition. The Green’s function with ER boundary condition, also referred
to as the hydrodynamic or modified Green’s function [13,45], is often used in the theory of conformal mappings in
a multiply connected domain [22]. To distinguish Green’s functions with Dirichlet and ER boundary conditions,
we use the notations GDiri
D and GERD , respectively. The boundary conditions are given as follows.
• Dirichlet boundary condition:
(2.9) GDiri
D (ζ, z) = 0 if ζ ∈ Cj , (j = 0, 1, . . . , g).
• Excursion reflected boundary condition:
(
ER
0 if ζ ∈ C0 ,
(2.10) GD (ζ, z) =
γj (z) if ζ ∈ Cj , (j = 1, . . . , g),
where γj ’s are functions depending only on z, and
∂GER
I
D (ζ, z)
dsz = 0, for j = 1, . . . , g.
Cj ∂nz
Here, ∂/∂nz is the derivative in the direction of the outwarding point normal and dsz denotes the
arclength element with integration variable z. These requirements uniquely determine the functions γj .
The Green’s function with Dirichlet boundary conditions measures local times for Brownian motion on D
that is killed when it hits one of the Cj . For the ER Green’s function the associated stochastic process, called
ER Brownian motion, was studied in [12, 17].
2.2.1. Relation between Dirichlet and ER Green’s function. We denote by hj , j = 0, 1, . . . , g, the harmonic
functions on D with boundary conditions
(
1 on Cj ,
(2.11) hj (ζ) =
0 on Ck , k ̸= j.
Note that hj (ζ) corresponds to the probability that a Brownian motion starting at ζ exits D through Cj . It is
well known that they can be expressed in terms of GDiri
D as
∂GDiri
I
1 D (ζ, z)
hj (ζ) = − dsz .
2π Cj ∂nz
From the hj we also extract the collection of integrals
I
1 ∂hj (z)
(2.12) Pkj = dsz ,
2π Ck ∂nz
which are known as the periods of hj . Then it is known that the period matrix P = [Pkj ]gk,j=1 is real, symmetric
and positive-definite. We write h = (h1 , . . . , hg )T . The following relation follows from [17, Proposition 5.2], [15,
Eq.(45)] and [5, Eq.(6)].
Proposition 2.1. In a multiply connected domain D, we have
g
X D E
−1
(2.13) GDiri
D (ζ, z) = GER
D (ζ, z) − γj (z) hj (ζ) = GER
D (ζ, z) − h(ζ), P h(z) ,
j=1
where ⟨·, ·⟩ is the standard inner product.

2.2.2. Green’s functions on circular domains. In [16], the authors express the ER Green’s function defined in a
circular domain in terms of the S-K prime function.
Proposition 2.2. On a circular domain D, we have
1 ω(ζ, z̄ −1 ) ω(ζ −1 , z̄) z ω(ζ, z̄ −1 )
(2.14) GER
D (ζ, z) = log −1 −1
= log .
2 ω(ζ, z) ω(ζ , z ) ω(ζ, z)
and so, by Proposition 2.1,
z ω(ζ, z̄ −1 ) D E
(2.15) GDiri
D (ζ, z) = log − h(ζ), P −1 h(z) .
ω(ζ, z)
The second equality of (2.14) is from the transformation rule (2.5) of the S-K prime function.
Example: Green’s functions on an annulus
Let us return to the annulus Ar = {z ∈ C : e−r < |z| < 1} as a reference domain. From formula (2.6)
for the S-K prime function and Proposition 2.2 we have
∞
z − ζ Y (1 − e−2rk ζz −1 )(1 − e−2rk zζ −1 )
(2.16) GER
Ar (ζ, z) = − log .
1 − ζ z̄ (1 − e−2rk ζ z̄)(1 − e−2rk ζ −1 z̄ −1 )
k=1
It is also easy to see that in Ar ,
log |ζ| 1
h1 (ζ) = − , P = P11 = .
r r
Therefore by (2.13), the Dirichlet Green’s function on Ar is given by
∞
z − ζ Y (1 − e−2rk ζz −1 )(1 − e−2rk zζ −1 ) log |ζ| log |z|
(2.17) GDiri
Ar (ζ, z) = − log − .
1 − ζ z̄ (1 − e−2rk ζ z̄)(1 − e−2rk ζ −1 z̄ −1 ) r
k=1
2.2.3. Domain constant and the conformal radius. Returning to the general setting, as ζ → z the Green’s
functions have the asymptotic expansions
1
(2.18) GDiri
D (ζ, z) = log + dD (z) + o(1),
|ζ − z|
1
(2.19) GER
D (ζ, z) = log + cD (z) + o(1).
|ζ − z|
We recall that the function dD is called the domain constant, whereas cD is called the (logarithmic) conformal
radius, see e.g. [5, Section 4].
2.2.4. Bipolar Green’s function on a compact Riemann surface. We will also need to make use of Green’s
functions on the Schottky double, but for this it is more useful to consider the more general bipolar Green’s
function on a compact Riemann surface. We now recall a representation of the bipolar Green’s function in terms
of the Riemann theta function; the full details can be found in [32].
Let M be a compact Riemann surface of genus g, and let p, q be distinct marked points of M . The bipolar
Green’s function z 7→ Gp,q (z) is determined (up to additive constants) by the requirement that it be harmonic
and have the following asymptotic expansion near marked points: in some/any chart
1
Gp,q (z) = log + O(1) (z → p),
|z − p|
1
Gp,q (z) = − log + O(1) (z → q).
|z − q|
It follows from [32, Appendix A] that the bipolar Green’s function Gp,q is expressed as
Θ(A(z) − A(q) − e) D E
(2.20) Gp,q (z) = log − 2π (Im τ )−1 Im[A(p) − A(q)], Im A(z) ,
Θ(A(z) − A(p) − e)
where the point e is chosen to be an element of TΛ satisfying Θ(e) = 0 and such that neither of the maps
(
Θ(A(z) − A(p) − e)
z 7→
Θ(A(z) − A(q) − e)
is identically zero. Note that the expression (2.20) depends on the choice of e, but not the choice of the base
point p0 of the Abel-Jacobi map A.
Example: The genus one case
We consider the complex torus TΛ := C/Λ of genus one, where Λ = Z + τ Z is the group generated by
z 7→ z + 1, z 7→ z + τ . Here, the modular parameter τ is in the upper-half plane H. The holomorphic
differentials on TΛ form a one-dimensional vector space spanned by dz, therefore the Abel-Jacobi map
is simply the identity, and according to (2.20) the bipolar Green’s function Gp,q on M = TΛ is given by
(up to an additive constant)
θ(z − q) Im(p − q) Im z
(2.21) Gp,q (z) = log − 2π , θ(z) ≡ θ1 (z|τ ).
θ(z − p) Im τ
For an example of a Riemann surface with boundary, consider the case of the Dirichlet and ER Green’s
function on a cylinder.
Example: On a cylinder
In the cylinder
n r o
(2.22) Cr := Sr /⟨z 7→ z + 1⟩, Sr := z ∈ C : 0 < Im z < ,
2π
the Green’s functions are expressed in terms of Jacobi theta function θ(r, z) ≡ θ1 (z| ir
π ) as
θ(r, ζ − z̄) θ(r, ζ − z̄) Im ζ Im z
(2.23) GER
Cr (ζ, z) = log , GDiri
Cr (ζ, z) = log − 4π 2 .
θ(r, ζ − z) θ(r, ζ − z) r
2.3. Loewner vector field in a multiply connected domain. Let D be a multiply connected domain
bounded by g + 1 smooth Jordan curves Cj (j = 0, 1 · · · , g). We define the complex ER Green’s function by
ER+ 1 ER e ER

(2.24) GD := GD + iG D ,
2
where Ge ER is the harmonic conjugate of the ER Green’s function. We denote by
D
ER 1 ∂GER
D (ζ, z)
(2.25) HD (ζ, q) := , (q ∈ C0 )
2 ∂nz z=q
the ER Poisson kernel. Let us also define complex ER Poisson kernel by

ER+ 1 ER ER

(2.26) HD = HD + i H
eD ,
2
where H e ER is the harmonic conjugate of the ER Poisson kernel. We mention that the complex ER Poisson
D
kernel is uniquely defined, see e.g. [17, Proposition 6.4].
The complex ER Poisson kernel can be used to construct a conformal map from D onto a chordal standard
domain. More precisely, the conformal map is given by
ER+
(2.27) f (ζ) := −ia HD (ζ, q) + b, (a, b ∈ R≥0 ),
which maps (C0 ; q) → (R, ∞), see e.g. [14, 17]. In particular, if D is contained in the upper-half plane and
q = ∞, the canonical mapping
ER+
(2.28) g(ζ) := −2iHD (ζ, ∞)
satisfies the hydrodynamic normalization

(2.29) lim (g(ζ) − ζ) = 0.
ζ→∞
See also [17, Proposition 6.4] for a probabilistic representation.

The Loewner vector field vζ (z) is the analytic function in z with
∂GER
D (z, ζ)
Im vζ (z) = − , (ζ ∈ C0 ),
∂nζ
see e.g. [6, Theorem 3.1]. Then in terms of the ER Poisson kernel, we have
ER+
(2.30) vζ (z) ≡ vζ,D (z) := 2iHD (z, ζ).
Let γ : [0, ∞) → D be a simple curve with γ(0) ∈ R and gt be the canonical mapping from D \ γ[0, t] onto
D. Write
Ωt = gt (D \ γ[0, t]), ξt = gt (γ(t)).
Then by [6, Theorem 3.1] (see also [17, Theorem 6.14]), the family of conformal mappings gt satisfies the chordal
Loewner equation (1.7), which we recall is
∂t gt (z) = −vξt ,Ωt (gt (z)), g0 (z) = z.
Equation (1.7) was derived in [36] (resp., [5]) for the annulus (resp., disc) with circular slits. See also [11].
Example: Loewner vector field on a simply connected domain
In the upper-half plane H with Dirichlet boundary conditions we have

ζ − z̄ + 1 ζ − z̄
(2.31) GH (ζ, z) = log , GH (ζ, z) = log .
ζ −z 2 ζ −z
On the other hand, by (2.31), we have

1 d z − x + iy z̄ − x − iy i i
HH (z, x) = log = −
2 dy z − x − iy z̄ − x + iy y=0 z−x z−x
and z − x + iy
1 d i
HH+ (z, x) = log = .
2 dy z − x − iy y=0 z−x
Therefore in this case, the equation (1.7) gives rise to the well-known chordal Loewner equation
2
∂t gt (z) = , g0 (z) = z.
gt (z) − ξt
Example: Loewner vector field on an annulus
Note that by (2.17), we have

1 θ(ζ − z̄)
GER,+
Cr (ζ, z) = log .
2 θ(ζ − z)
Then by using
θ′ (z) 1 θ′′′ (0)
= ζ(r, z) + z,
θ(z) 3 θ′ (0)
where ζ(r, z) is the Weierstrass zeta function, we have
1 θ′′′ (0)

1 d θ(z − x + iy)
HCER,+ (z, x) = log = i ζ(z − x) + (z − x) .
r
2 dy θ(z − x − iy) y=0 3 θ′ (0)
This gives that in the cylinder Cr ,
2 θ′′′ (0)
vp (z) = −2 ζ(z − p) − (z − p).
3 θ′ (0)
By [41, Eq.(23.9.3)], ζ(z) has the asymptotic expansion

1
ζ(z) = + O(z 3 ) as z → 0,
z
hence it follows that as p → z,
2 2 θ′′′ (0)
vp (z) = + (p − z) + O(|p − z|2 ).
p−z 3 θ′ (0)
2.3.1. Asymptotics of vζ in a chordal standard domain. In a general chordal standard domain with ζ ∈ R the
Loewner vector field has the asymptotic expansion
2
(2.32) vζ (z) = + rD,0 (z) + rD,1 (z)(ζ − z) + o(|ζ − z|), z → ζ,
ζ −z
see e.g. [17, Proposition 6.4] and [18, p.1834]. In particular,
2
(2.33) vζ′ (z) = + r ′
D,0 (z) − rD,1 (z) + O(|ζ − z|), z → ζ.
(ζ − z)2
The z-dependence of the coefficients in (2.32) comes from the fact that the (ER) Green’s function in the chordal
standard domain is not translation invariant, i.e. GER ER
D (ζ, z) ̸= GD (ζ + t, z + t).
3. Gaussian free fields in a multiply connected domain

3.1. GFFs and their correlation functions. Let D be a multiply connected domain bounded by g + 1
smooth Jordan curves Cj , j = 0, 1 · · · , g. We assume C0 is the outermost boundary component.
Let Φ be the mean zero GFF with zero Dirichlet boundary condition and 2-point correlation function
(3.1) E Φ(ζ)Φ(z) = 2GDiri
D (ζ, z).
Although Φ cannot be realized as a random function, we write it as a (correlation functional-valued) function

(or a Fock space field) since we are only interested in studying its correlation functions, which are well defined.
For example, in a cylinder Cr , by (2.23),
2
θ(r, ζ − z̄) Im ζ · Im z
(3.2) E Φ(ζ)Φ(z) = log − 8π 2 .
θ(r, ζ − z) r
In a general circular domain C, it follows from (2.15) that
2
z ω(ζ, z̄ −1 ) D E
(3.3) E Φ(ζ)Φ(z) = log − 2 h(ζ), P −1 h(z) .
ω(ζ, z)
To work with the Gaussian free field on the Schottky double we first turn to the more general setting of
the GFF on a compact Riemann surface. On a compact Riemann surface M ≡ Mg of genus g, the Gaussian
free field Ψ is defined as a bi-variant Fock space field (z, z0 ) 7→ Ψ(z, z0 ), see [32, Section 2]. The correlation
functions of Ψ are given in terms of the bipolar Green’s function as

E Ψ(ζ, z)Ψ(ζ̃, z̃) = 2 Gζ,z (ζ̃) − Gζ,z (z̃) = 2 Gζ̃,z̃ (ζ) − Gζ̃,z̃ (z) .
For example, when g = 1, by (2.21),
2
θ(ζ̃ − z)θ(z̃ − ζ) Im(ζ − z) · Im(ζ̃ − z̃)
(3.4) E Ψ(ζ, z)Ψ(ζ̃, z̃) = log − 4π .
θ(ζ̃ − ζ)θ(z̃ − z) Im τ
In general, by (2.20),
2
Θ(A(ζ̃) − A(z) − e)Θ(A(z̃) − A(ζ) − e)
E Ψ(ζ, z)Ψ(ζ̃, z̃) = log
(3.5) Θ(A(ζ̃) − A(ζ) − e)Θ(A(z̃) − A(z) − e)
D E
− 4π (Im τ )−1 Im[A(ζ) − A(z)], Im[A(ζ̃) − A(z̃)] .
3.2. Schottky double construction of GFFs. In this subsection, we explain that GFF in a multiply con-
nected domain can be constructed from GFF on a compact Riemann surface via the Schottky double. For the
case g = 0, such a construction was extensively studied in [33]. As a pre-cursor to the general setting we give
the standard example for the case g = 1.
Example: GFF on a cylinder via the GFF on a torus
Recall (2.22) for the representation of a cylinder Cr . The Schottky double of the cylinder Cr is a complex
torus Tr := C/Λ, where Λ = Z + ir π Z. Note that by (3.2) and (3.4),
2
θ(r, ζ − z̄) Im ζ · Im z 1
E Φ(ζ)Φ(z) = log − 8π 2 = E Ψ(ζ, ζ̄)Ψ(z, z̄).
θ(r, ζ − z) r 2
Therefore we have the Schottky double relation
1
Φ(z) = √ Ψ(z, z̄)
2
within correlations.
To describe the Schottky doubly relation on higher genus surfaces, it is convenient to take the reference
domain as the circular domain C = C(r, δ) in (2.1). Recall that Cj′ is the reflection of Cj with respect to C0 ,
and that the image of Cj′ under θj is Cj . Moreover for ζ ∈ Cj′ , we have θj (ζ) = ζ ∗ = 1/ζ.
We will denote by F the region of the complex plane that is exterior to the 2g circles {Cj , Cj′ |j = 1, · · · , g}.
This is called the fundamental region associated with the Schottky group generated by the Möbius maps
{θj |j = 1, · · · , g}. Here, the two halves of F (one is the reflection of the other one with respect to the unit
circle) can be interpreted as two sides of a symmetric compact Riemann surface Cb known as the Schottky double
of C. The bi-variant field Ψ(z, z0 ) exists on the Schottky double (see the last section) and can be used to define
the Dirichlet GFF in the circular domain C in the following way.
Proposition 3.1. Within correlations, the Dirichlet GFF on a circular domain C satisfies the Schottky double
relation
1
Φ(z) = √ Ψ(z, z ∗ ).
2
Formally speaking, the idea behind Proposition 3.1 is just an instance of the method of images. This makes
the statement intuitively clear, but we prove it by verifying that the correlation functions of the two sides match.
To accomplish this we independently compute the Dirichlet and ER Green’s functions in a circular domain.
Lemma 3.2. In a circular domain D, we have
1 Θ(A(z) − A(ζ ∗ ) − e)Θ(A(z ∗ ) − A(ζ) − e)
(3.6) GER
D (ζ, z) = log ,
2 Θ(A(z) − A(ζ) − e)Θ(A(z ∗ ) − A(ζ ∗ ) − e)
where z ∗ = 1/z̄ and
D E
−1
(3.7) GDiri ER
D (ζ, z) = GD (ζ, z) − π (Im τ ) Im[A(ζ ∗ ) − A(ζ)], Im[A(z ∗ ) − A(z)] .
To prove Lemma 3.2 first requires a few basic facts about the period matrices P and τ . We explain this
relation next, but on first reading it is enough to simply accept (3.8) as fact and return to this argument later.
On the Schottky double Cb take the circle Cj (which is identified with Cj′ ) as a j-th a-cycle of C. b Also, for
′ ∗
ζ ∈ Cj , take any line joining ζ and θj (ζ) = ζ as a j-th b-cycle. Let {ηj |j = 1, · · · , g} be the basis of the space of
all holomorphic 1-differentials on Cb which is uniquely determined by (2.7). We now recall ηj = dvj and express
the integral of the first kind {vj | j = 1, · · · , g} in terms of harmonic functions {hj (ζ) | j = 0, 1, · · · , g} given by
(2.11). Recall that the period matrix P = (Pjk )gj,k=1 of hj is given by (2.12). Since hj vanishes on C0 , one can
extend hj to the fundamental region F as
(
hj (ζ) on C
hj (ζ) =
e
∗
−hj (ζ ) on F \C.
Let vbj be the holomorphic function on F whose imaginary part is e

hj . Note that
vbj (ζ ∗ ) = vbj (ζ), ζ ∈ C.
Then it follows that
1 −1
(v1 , · · · , vg )T = −
P (b v1 , · · · , vbg )T ,
2π
where P is given by (2.12), see e.g. [15]. Moreover, the period matrix τ = (τjk )gj,k=1 with τjk = bk dvj is related
H
to P as
i
(3.8) τ = P −1 , (Im τ )−1 = πP .
π
Proof of Lemma 3.2. By the characteristic property of the Green’s function, all we need to show is that for
ζ ∈ Cj ,
1 Θ(A(z) − A(ζ ∗ ) − e)Θ(A(z ∗ ) − A(ζ) − e)
log = π (Im τ )−1 Im[A(ζ ∗ ) − A(ζ)], Im[A(z ∗ ) − A(z)] .
2 Θ(A(z) − A(ζ) − e)Θ(A(z ∗ ) − A(ζ ∗ ) − e)
For any ζ ∈ C, we have
Z ζ∗ Z ζ∗ T T
∗
A(ζ ) − A(ζ) = dv1 , · · · , dvg = v1 (ζ ∗ ) − v1 (ζ), · · · , vg (ζ ∗ ) − vg (ζ)
ζ ζ
T T
(3.9) = v1 (ζ) − v1 (ζ), · · · , vg (ζ) − vg (ζ) = −2i Im v1 (ζ), · · · , vg (ζ)
i −1
= P h(ζ) = τ h(ζ).
π
The last equality used (3.8). In particular, for ζ ∈ Cj , the definition of the hk gives that h(ζ) = ej for
ej = (0, · · · , 1, · · · , 0)T , and so
A(ζ ∗ ) − A(ζ) = τ ej .
For ζ ∈ Cj this gives rise to
D E D E
π (Im τ )−1 Im[A(ζ ∗ ) − A(ζ)], Im[A(z ∗ ) − A(z)] = π ej , Im[A(z) − A(z ∗ )] .
On the other hand, it follows from (2.8) that

Θ(A(z) − A(ζ ∗ ) − e) Θ(A(z) − A(ζ) − e − A(ζ ∗ ) + A(ζ))
=
Θ(A(z) − A(ζ) − e) Θ(A(z) − A(ζ) − e)
Θ(A(z) − A(ζ) − e − τ ej ) h 1 i
= = exp − 2πi ⟨A(z) − A(ζ), ej ⟩ − τjj
Θ(A(z) − A(ζ) − e) 2
and
Θ(A(z ∗ ) − A(ζ) − e) h
∗ 1 i
= exp 2πi ⟨A(z ) − A(ζ), ej ⟩ − τjj .
Θ(A(z ∗ ) − A(ζ ∗ ) − e) 2
Therefore we obtain
1 Θ(A(z) − A(ζ ∗ ) − e)Θ(A(z ∗ ) − A(ζ) − e) D
∗
E
log = π e j , Im[A(z) − A(z )] ,
2 Θ(A(z) − A(ζ) − e)Θ(A(z ∗ ) − A(ζ ∗ ) − e)
which completes the proof. □
Remark 7. As a consequence of Lemma 3.2, we have the functional relations:

2
z ω(ζ, z̄ −1 ) Θ(A(z) − A(ζ ∗ ) − e)Θ(A(z ∗ ) − A(ζ) − e)
log = log
ω(ζ, z) Θ(A(z) − A(ζ) − e)Θ(A(z ∗ ) − A(ζ ∗ ) − e)
and D E D E
h(ζ), P −1 h(z) = π (Im τ )−1 Im[A(ζ ∗ ) − A(ζ)], Im[A(z ∗ ) − A(z)] .
Proof of Proposition 3.1. We need to check

1
(3.10) E Φ(ζ)Φ(z) = E Ψ(ζ, ζ ∗ )Ψ(z, z ∗ ).
2
By definition, the identity (3.10) is equivalent to
1
(3.11) GDiri (Gζ,ζ ∗ (z) − Gζ,ζ ∗ (z ∗ )).
D (ζ, z) =
2
Then the desired identity follows from Lemma 3.2 and the discussion in Section 3.1. □
3.3. Chiral fields and OPE exponentials. Chiral fields are defined by taking the holomorphic derivative of
Φ (which complexifies it) and integrating it along paths. More precisely, for a path γ in D we define
Z
Φ+ (γ) = J, J = ∂Φ
γ
and write Φ (ζ, z) for the collection of Φ (γ), where γ is a path from z to ζ. Then (ζ, z) 7→ Φ+ (ζ, z) is a
+ +
bi-variant multivalued field with correlation

E Φ+ (ζ, ζ0 )Φ(z) = G+ +
D (ζ, z) − GD (ζ0 , z),
where G+
D is the complex Green’s function defined in Subsection 1.1. Furthermore
E Φ+ (ζ, ζ0 )Φ+ (z, z0 ) = log λ(ζ, ζ0 ; z, z0 ) + π⟨(Im τ )−1 (A(ζ) − A(ζ0 )), (A(z) − A(z0 ))⟩,
where λ is the (generalized) cross-ratio,
Θ(A(p̃) − A(q) − e)Θ(A(q̃) − A(p) − e)
λ(p, q; p̃, q̃) = .
Θ(A(p̃) − A(p) − e)Θ(A(q̃) − A(q) − e)
It is well known that the multivalued function λ(p, q; p̃, q̃) is independent of e, see [22, Section VII.6]. It can be
rewritten in terms of the prime form as
E(p̃, q)E(q̃, p)
λ(p, q; p̃, q̃) =
E(p̃, p)E(q̃, q)
by using the uniqueness of the cross-ratio (e.g. see [43]) and the periodicity properties of the prime form below.
The (Schottky-Klein) prime form E (see [24, 39] for its definition and basic properties) on a compact Riemann
surface is the skew-symmetric bi-differential with periodicity properties
E(p + aj , q) = E(p, q),
Z q
τjj
E(p + bj , q) = exp − 2πi + ωj E(p, q)
2 p
and asymptotic behavior
z − z′
E(p, q) = √ √ 1 + O (z − z ′ )2
dz dz ′
in a local chart that contains both p and q, where z is the image of p for that chart and z ′ the image of q. Its
conformal dimension is −1/2 with respect to each variable.
For calculations it is sometimes convenient to introduce a formal 1-point field z 7→ Φ+ (z) such that Φ+ (ζ, z) =
Φ (ζ) − Φ+ (z). The 1-point field Φ+ would then have correlation
+
E Φ+ (ζ)Φ+ (z) = − log E(ζ, z) + π⟨(Im τ )−1 A(ζ), A(z))⟩.

By defining Φ− = Φ+ we then obtain the decomposition
Φ = Φ+ + Φ− ,
in the sense that all correlations against Φ can be replaced by correlations against Φ+ + Φ− .
We now define the OPE exponential of the field Φ+ (ζ, z). As explained in [32, 33], the OPE exponential is
a conformal field defined via a power series expansion of the exponential function, but for which multiplication
is replaced by the operator product expansion. For σ ∈ R we denote the OPE exponential by
+
O{σ · z − σ · z0 } = e∗iσΦ (z,z0 )
.
See [32, 33] for the precise power series expansion. Each term in the expansion of the OPE exponential is an
element of the Fock space family generated by Φ, therefore so too is the sum. Although the OPE exponential
appears to be more complicated than the Wick exponential of Φ+ (z, z0 ), they turn out to be the same up to a
factor of a deterministic conformal field. This is shown in [32, Proposition 5.3], which we now recall.
P R
Proposition 3.3 (Proposition 5.3 of [32]). Let X = j τj Φ(zj ) for a divisor τ satisfying τ = 0. Then for
α∈C
α2
e∗αX = exp E[X ∗ X] e⊙αX .
2
Thus the OPE exponential can be expressed entirely in terms of the Wick exponential and the zero order
coefficient in the OPE expansion of the field X with itself. The operator product expansion is computed by
using Wick’s formula to compute the expansion of
X X
τj Φ(zj ) τj Φ(zj + ϵj ) ,
j j
as the support points ϵj all converge to the origin. The same methods as in Proposition 3.3 can be straight-
forwardly extended to apply this calculation to the field Φ+ (z, z0 ), the only difference being that Φ+ (z, z0 ) is
specified if an integral of the current field is taken along a given path and so we must specify how the two paths
γ, γ̃ are related in the product
Φ+ (z, z0 )Φ+ (z̃, z̃0 ).
We work with the natural choice that the endpoints (z̃, z̃0 ) approach (z, z0 ) while the paths γ, γ̃ stay disjoint
but close to each other. For this choice it can be shown by straightforward calculation that c+ (z, z0 ) depends
only on the choice of γ (as the prime form does, see [24, Chapter 1]) and
1 π
c+ (z, z0 ) := E Φ+ ∗ Φ+ (z, z0 ) = log E(z, z0 ) + ⟨(Im τ )−1 (A(z) − A(z0 )), (A(z) − A(z0 ))⟩.
2 2
+ 1
Furthermore c (z, z0 ) is a PPS form of order (− 2 , 0) with respect to z and z0 . Therefore, by the same reasoning
as in Proposition 3.3 we have
2 +
c (z,z0 ) ⊙iσΦ+ (z,z0 )
(3.12) O{σ · z − σ · z0 } = e−σ e .
As a conformal field the OPE exponential is a holomorphic differential with conformal dimension σ 2 /2 at z, z0 ,
and satisfies
+ 2 + 1 n π o
E e∗iσΦ (z,z0 ) = e−σ c (z,z0 ) = exp − σ 2
⟨(Im τ )−1
(A(z) − A(z0 )), (A(z) − A(z 0 )) .
E(z, z0 )σ2 2
Finally we introduce the OPE exponential of more complicated linear combinations of the fields Φ± . Recall
that in the introduction we defined Φ{τ } as the linear combination
X
Φ{τ } = Φ{τ + , τ − } = τj+ Φ+ (zj ) − τj− Φ− (zj ),
for any double divisor τ = (τ + , τ − ) (τ ± = τj± · zj ) satisfying the neutrality condition

P
Z
τ + + τ − = 0.
The OPE exponential O{τ } is then defined by

O{τ } := e∗iΦ{τ } .
Although these OPE exponentials depend on many more points than O{σ · z − σ · z0 }, it can be shown that
+ − + −
(3.13) O{τ } = O(τ1 ) (z1 )O(τ1 ) (z1 ) · · · O(τn ) (zn )O(τn ) (zn ),
where O(τ ) (z) = O{τ · z − τ · z0 }. The result does not depend on the choice of a base point z0 due to the
neutrality condition. This representation shows that the multiplication rule
O{τ 1 }O{τ 2 } = O{τ 1 + τ 2 }.
holds. This shows that while the fields O{τ } may appear to be more complicated, they can be entirely expressed
in terms of products of the simpler field O{σ · z − σ · z0 }. Products of the latter fields can be computed using
equation (3.12) and the Wick rule
e⊙αX e⊙βY = eαβ Cov(X,Y ) e⊙(αX+βY ) .
For a more detailed discussion of OPE exponentials in a simply connected domain see [33].
3.4. Background charge modification. To implement the c < 1 version of our conformal field theory we
introduce the background charge modification. It is simplest to introduce on the Schottky double Cb and then
transfer the results back to the circular domain C. The presentation here is a specialization of the results found
in [32, Section 6]. We remind the reader of the following basic facts:
• The surface Cb is covered by a collection of charts (ϕ, U ), U ⊂ Cb open, where ϕ : U → C. The
transition map between overlapping charts is always assumed to be holomorphic. A non-random field is
an assignment of a smooth function to each chart. For a non-random field ψ and charts ϕ, ϕ, e we write
ψ = (ψ ∥ ϕ) and ψe = (ψ ∥ ϕ)
e for the assignment of ψ to each of the charts.
• A non-random field ψ is called a PPS(µ, ν) form if ψ satisfies the transformation law
ψ = ψ̃ ◦ h + µ log h′ + ν log h′ ,
where h is the transition map between two overlapping charts ϕ, ϕ. e
+ −
• A non-random field ψ is called a harmonic form if ψ = ψ + ψ for some (multivalued) holomorphic
form ψ + and (multivalued) anti-holomorphic form ψ − .
• A (harmonic) PPS form ψ is simple if ∂ ∂ψ ¯ is a finite linear combination of Dirac delta measures.
• For a given point q ∈ C,
b we call ψq a basic form if it is a simple form with a sole singularity at q,
1 ¯ q = −πδq ,
ψq (z) ∼ log , equivalently ∂ ∂ψ
|z − q|2
in any chart.
Simple PPS forms have the following properties. These results are proved in Theorem 6.4, Proposition 6.6,
and Corollaries 6.3 and 6.5 of [32].
Proposition 3.4. Throughout let M be a compact Riemann surface with genus g ̸= 1 and χ = 2 − 2g. Then
(1) For any q ∈ M there exists a unique basic form ψq (unique up to an additive constant), and it is a
PPS(1/χ, 1/χ) form.
(2) A finite linear combination of basic forms is a simple form.
(3) Let ψ be a simple PPS(ib, ib) form on a compact Riemann surface of genus g. Then
Z
i ¯ = b(2 − 2g) = bχ.
(3.14) ∂ ∂ψ
π
P P
(4) Suppose that g ̸= 1 and a finite atomic measure β = βk δqk on the surface satisfies βk = b(2 − 2g).
Then there is a unique (up to an additive constant) simple PPS(ib, ib) form ψβ satisfying
i ¯
(3.15) β = ∂ ∂ψ β.
π
(5) Given q ∈ M and β of part (iv), any ψβ in part (iv) satisfies
X
ψβ (z) − ψβ (z0 ) = ibχ(ψq (z) − ψq (z0 )) + 2i βk (Gqk ,q (z) − Gqk ,q (z0 )),
k
and the right hand side does not depend on the choice of q.
In the case g = 1 the last formula specializes to
X
ψβ (z) − ψβ (z0 ) = 2ib log |ω(z)| − 2ib log |ω(z0 )| + 2i βk (Gqk ,q (z) − Gqk ,q (z0 )),
k
where ω is a holomorphic form (equivalently, a (1, 0) differential).
Given a simple PPS(ib, ib) form ψ, the corresponding background charge β is defined by
i ¯ X X
β = ∂ ∂ψ = βk δqk = βk · qk .
π
Here, we think of β as a measure. In the algebraic geometry literature it would often be called a divisor.
Equation (3.14) is referred to as the neutrality condition of the simple form ψ, it is also called the neutrality
condition of the corresponding background charge β. Given the background charge β, we will typically define
b as the unique constant such that
Z X
(3.16) β= βk = b(2 − 2g) = bχ.
k
We write ψq+ , ψq− for holomorphic part and anti-holomorphic part of a basic simple form ψq . This gives the
decomposition
ψq = ψq+ + ψq− .
Let β ± be background charges (atomic divisors), which are allowed to satisfy different neutrality conditions
X
βk± = b± χ,
k
although in the rest of this paper they will always be the same. The divisors β ± keep track of modifications of
the field Ψ via linear combinations of the basic forms ψq± . More precisely, we define the field Ψβ+ ,β− (z, z0 ) via
X X
βk+ ψq+k (z) − ψq+k (z0 ) − i βk− ψq−k (z) − ψq−k (z0 ) .

Ψβ+ ,β− (z, z0 ) = Ψ(z, z0 ) + i
k k
Using these definitions for the fields on a compact Riemann surface, we now define the field Φβ . The idea is
to define Φβ and its mean φβ as a “symmetric” case of Ψβ+ ,β− on the Schottky double. The special case that
we use is simply β + = β − = β, which is an example of the method of images. We assume that the background
charge β is supported on C0 , and all charges in β are real (i.e. β(qk ) ∈ R for all k). We now define
X
(3.17) Φβ = Φ + φβ , φβ (z) = E Ψβ,β (z, z ∗ ) = −2 βk Im(ψq+k (z) − Ψ+ ∗
qk (z )).
k
This definition makes Φβ a PPS(ib, −ib) form with φβ = E Φβ . An equivalent way of defining it is
1
Φβ (z) = √ Ψ√2β,√2β (z, z ∗ ),
2
which makes the method of images representation more transparent, and matches the representation in Propo-
sition 3.1. In a multiply connected domain D, this leads to the formula presented in equation (1.5), i.e.
X
φβ (z) = −2b arg w′ (z) − 2 βk Im G+D (w(qk ), w(z)),
where w : (D, C0 , q) → (Hg , R, ∞) is a conformal map satisfying a hydrodynamic normalization near w(q) = ∞.
Here, we recall that G+D is the complex Green’s function. In particular, in a circular domain, it follows from
Lemma 3.2 that

+ 1 Θ(A(z) − A(1/z̄0 ) − e)Θ(A(1/z) − A(z̄0 ) − e)
GD (z, z0 ) = log
4 Θ(A(z) − A(z0 ) − e)Θ(A(1/z) − A(1/z0 ) − e)
πi D E
+ (Im τ )−1 (A(z) + A(1/z)), Im[A(1/z̄0 ) − A(z0 )]
2
up to periods. In addition, we define formal 1-point fields Φ± β as
Φ± ± ±
β = Φ + φβ ,
− ∗ − −
where φ+ + + ∗
β (z) = ψβ (z) − ψβ (z ), and φβ (z) = ψβ (z) − ψβ (z ). The current field Jβ is defined as
X
−
Jβ := ∂Φβ = J + ȷβ , ȷβ (z) = i βk ∂z (G+
D (qk , z) − GD (qk , z)).
Based on the definition of the current field we have the following.

Proposition 3.5. The field Φβ has a stress tensor

Aβ := A(0) + ib∂ − ȷβ J,
and its Virasoro field is given as
1
Tβ := − Jβ ∗ Jβ + ib ∂Jβ .
2
Proof. The proof is the same as the one in [10, Proposition 3.5] and [33, Theorem 5.2]. □
Given the field Φβ we can also consider its OPE exponential. In [32, Lemma 5.2] it is proved that for a
smooth function f and a mean zero Gaussian conformal field X (say X = Φ(z) − Φ(z0 )) that
e∗(X+f ) = ef e∗ X.
Consequently, if we denote the OPE exponential of Φ+
β (z, z0 ) by
+
Oβ {τ · z − τ · z0 } = e∗iτ Φβ (z,z0 )
then we have the formula
+ +
Oβ {τ · z − τ · z0 } = eiτ (φβ (z)−φβ (z0 )) O{τ · z − τ · z0 }.
The last formula combined with the OPEP product formula (3.13) us to compute the exponential of Φβ {τ }
R +OPE −
+ − ± n ±
for a double divisor τ = (τ , τ ), τ = j=1 τj · zj with neutrality condition τ + τ = 0. Recalling that
Oβ {τ } = e∗iΦβ {τ } , the precise formula is that
(τ + ) (τ − ) (τ + ) (τ − )
Oβ {τ } = Oβ 1 (z1 )Oβ 1 (z1 ) · · · Oβ n (zn )Oβ n (zn ),
(τ )
where we set Oβ (z) = Oβ {τ · z − τ · z0 }. As before this representation does not depend on the choice of the
base point z0 . If the support of τ contains the point at infinity, or if the support of τ intersects the support of
β, then a rooting procedure or a rescaling procedure is needed. See [31, Section 12.3] or [33] for more details.
4. Ward’s and BPZ equations

In this section we prove Theorem 1.1, which establishes the Eguchi-Ooguri version of Ward’s equation in
canonical multiply connected domains, and Theorem 1.2, which takes an OPE limit of the Ward equation to
obtain the BPZ and BPZ-Cardy equation. The latter makes use of the level two degeneracy property of vertex
exponentials, and is the key step in establishing the connection between conformal field theory and SLE.
4.1. Ward’s identity. We begin by introducing the global Ward’s functional, which plays a key role in under-
standing the relationship between stress tensors and vector fields. To define Ward’s functional, we consider a
meromorphic vector field v defined on D, which is continuous up to its boundary. Ward’s functionals Wϵ± (v)
are then given by
Z ZZ
1 1 ¯
(4.1) Wϵ+ (v) = W + (v; Dϵ ) = vAβ − (∂v)A β, Wϵ− (v) = Wϵ+ (v),
2πi ∂Dϵ π Dϵ
S
where Dϵ = D \ B(qj , ϵ). Note here that we have excluded the nodes of β, where the integrand vAβ has
singularities.
Since v is meromorphic the term ∂v ¯ is zero except at the poles of v, and so ∂v
¯ is interpreted in the sense of
distributions. For a chordal standard domain D, the boundary ∂D is the real axis together with a collection of
contours encircling each horizontal slit . By [31, Proposition 5.9], it follows that for X in the OPE family Fβ ,
(4.2) L+ +
v X = Wϵ X, L− −
v X = Wϵ X.
Here, the Lie derivatives do not act on the qk variables. The next lemma gives a representation of the stress
tensor in terms of the Ward’s functional. In the next lemma and the rest of the paper we will frequently use
that for a chordal standard domain D
Z Z
f (ξ) dξ = f (ξ) dξ,
∂D ∂D
which follows because ∂D consists of R and a collection of horizontal slits.

Lemma 4.1. In the identity chart of a chordal standard domain D, we have
Z I
+ − 1
¯
1 X
2Aβ (ζ) = W (vζ ) + W (vζ̄ ) − vζ (ξ) − vζ (ξ) Aβ (ξ) dξ + vζ (ξ)Aβ (ξ)
2πi ∂D\R 2πi (qk ) k
within correlations. Here, ∂D \ R is a combination of small contours encircling each slit of D.
Proof. For ζ ∈ D, we define the reflected vector field
vζ♯ (z) := vζ (z̄) = vζ̄ (z).
By using the definition (4.1), we have
Z Z Z
1 ¯ ζ )Aβ + 1 1
Wϵ+ (vζ ) = − (∂v v ζ Aβ , Wϵ+ (vζ♯ ) = vζ♯ Aβ .
π Dϵ 2πi ∂Dϵ 2πi ∂Dϵ
The second equation uses that vζ♯ is holomorphic on D, which follows from vζ being meromorphic on D with
¯ = Āβ (ξ) on D, we have
sole pole at ζ. Since Aβ (ξ)
Z Z
− + ♯ 1 1 X ¯ β (ξ)
¯ dξ
Wϵ (vζ̄ ) = Wϵ (vζ ) = v (ξ)Aβ (ξ) dξ + vζ (ξ)A
2πi ∂D\R ζ̄ 2πi ∂B(q k ,ϵ)∩H
k
Z XZ
1 ¯ β (ξ) dξ − 1
=− vζ (ξ)A vζ (ξ)Aβ (ξ) dξ,
2πi ∂D\R 2πi ∂B(qk ,ϵ)∩H−
k
where H− := {ξ : Im ξ < 0} and, by convention, the curve ∂B(qk , ϵ) ∩ H− is oriented in the counterclockwise
direction. Then it follows from ∂v ¯ ζ = −2πδζ that
Z Z XI
1 ¯ ζ )Aβ + 1

¯ Aβ (ξ) dξ − 1

W + (vζ ) + W − (vζ̄ ) = − (∂v vζ (ξ) − vζ (ξ) vζ (ξ)Aβ (ξ) dξ
π D 2πi ∂D\R 2πi (qk )
k
Z I
1
¯
1 X
= 2Aβ (ζ) + vζ (ξ) − vζ (ξ) Aβ (ξ) dξ − vζ (ξ)Aβ (ξ) dξ.
2πi ∂D\R 2πi (qk ) k
□
Together, Lemma 4.1 and equation (4.2) complete the proof of equation (1.10) in Theorem 1.1. Equation
(1.12) in Theorem 1.1 is proved in [33, Lemma 7.2]. Although the statement in [33] is originally formulated for
the g = 0 case, the proof applies to our current setup as well since it pertains to a local property independent
of the global underlying geometry. We restate the lemma here for the reader’s convenience.
Lemma 4.2 (Lemma 7.2 of [33]). We have
I
(4.3) E vζ Aβ X = vζ (qk )∂qk E X,
(qk )
where X is a tensor product of fields in Fβ whose nodes do not contain any of the qk points in β.
To complete the proof of Theorem 1.1 only requires showing the generalized Eguchi-Ooguri equation (1.11).
This is done in the next section. Once the generalized Eguchi-Ooguri equations are proved, the final equation
(1.13) follows directly from (1.10) – (1.12).
4.2. Generalized Eguchi-Ooguri equation. In this subsection we give the proof of the generalized Eguchi-
Ooguri equations (1.11), thereby completing the proof of Theorem 1.1. We start by showing that (1.11) holds
for the OPE family F(b) , i.e. when the background charge modification only happens at one point, and then
extend it to the case of general background charge for a divisor β that satisfies (3.16).
Lemma 4.3. For any X ∈ F(b) , the generalized Eguchi-Ooguri equation
Z
1
¯ EA(ξ)X dξ + ∇H EX = 0
(4.4) vζ (ξ) − vζ (ξ) g
2πi ∂D\R
holds in the identity chart of a chordal standard domain D.
Proof. First recall that F(b) is the OPE family by generated by Φ(b) , which is simply Φβ in the special case
β = bχ · ∞. Placing the background charge at infinity means that Φ(b) = Φ in the standard identity chart.
Let W be the family of conformal fields that satisfy the equation (4.4). In order to prove that F(b) ⊂ W, it
suffices to show that any tensor product of GFFs and their derivatives is an element of W, and that the family
W is closed under OPE products. We break the proof into several parts.
• Proof that Φ(b) (z) ∈ W. We first show that in the identity chart the identity
Z Z
1 1
(4.5) vζ (ξ)EA(ξ)X dξ + v (ξ)EA(ξ)X dξ + ∇Hg EX = 0
2πi ∂D\R 2πi ∂D\R ζ̄
holds for X = Φ(b) (z) and A = A(b) . Note that we wrote the term ∇Hg EX to show the similarity with (4.4), but
that in this case EX = E Φ(b) = 0 so that term can be ignored. For the remainder, recall that in the identity
chart we have j(b) ≡ 0 and so
1
A ≡ A(b) = − J ⊙ J + ib∂J.
2
Then by Wick’s formula, we have
(4.6) EA(b) (ξ)Φ(b) (z) = ib∂ξ EJ(ξ)Φ(z) = 2ib∂ξ2 GD (ξ, z).
By residue calculus, and using the asymptotic ζ → z expansions
2 1 1
vζ (z) = + O(1), ∂ζ GD (ζ, z) = − + O(1),
ζ −z 2ζ −z
we can evaluate the contour integrals as
Z Z
1 1 1 1 ′
vζ (ξ)∂ξ2 GD (ξ, z) dξ = −2∂ζ2 GD (ζ, z) + vζ′ (z), vζ̄ (ξ)∂ξ2 GD (ξ, z) dξ = v (z).
2πi ∂D\R 2 2πi ∂D\R 2 ζ̄
By combining the last two formulas with the Wick expansion (4.6) we obtain
Z Z
1 b
vζ (ξ)EA(b) (ξ)Φ(b) (z) dξ = vζ (ξ)∂ξ2 GD (ξ, z) dξ = bi − 4∂ζ2 GD (ζ, z) + vζ′ (z) ,
2πi ∂D\R π ∂D\R
Z Z
1 b
vζ̄ (ξ)EA(b) (ξ)Φ(b) (z) dξ = v (ξ)∂ξ2 GD (ξ, z) dξ = bi vζ̄′ (z).
2πi ∂D\R π ∂D\R ζ̄
Therefore the identity (4.5) for X = Φ(b) (z) is equivalent to the identity
(4.7) 4∂ζ2 GD (ζ, z) = vζ′ (z) − vζ̄′ (z).
This identity is proved separately in Lemma 4.5 below.

• Proof that Φ(z1 )Φ(z2 ) ∈ W. We first compute the correlation EA(b) (ξ)Φ(b) (z1 )Φ(b) (z2 ) as
EA(b) (ξ)Φ(b) (z1 )Φ(b) (z2 ) = − 12 E(J ⊙ J)(ξ)Φ(z1 )Φ(z2 ) + ibE∂J(ξ)Φ(z1 )Φ(z2 ),
which uses that Φ(b) ≡ Φ in the identity chart. The last term is zero since the fields ∂J and Φ are mean zero.
For the first term we have by Wick’s formula that
EA(ξ)Φ(z1 )Φ(z2 ) = −EJ(ξ)Φ(z1 )EJ(ξ)Φ(z2 ) = −4∂ξ GD (ξ, z1 )∂ξ GD (ξ, z2 ).
By combining the latter with the residue calculus we obtain
Z Z
1 2
vζ (ξ)EA(ξ)Φ(z1 )Φ(z2 ) dξ = − vζ (ξ)∂ξ GD (ξ, z1 )∂ξ GD (ξ, z2 ) dξ
2πi ∂D\R πi ∂D\R

= 8∂ζ G(ζ, z1 )∂ζ GD (ζ, z2 ) + 2 vζ (z1 )∂z1 + vζ (z2 )∂z2 GD (z1 , z2 ).
Similarly, we have
Z Z
1 2
vζ̄ (ξ)EA(ξ)Φ(z1 )Φ(z2 ) dξ = − v (ξ)∂ξ GD (ξ, z1 )∂ξ GD (ξ, z2 ) dξ
2πi ∂D\R πi ∂D\R ζ̄

= 2 vζ̄ (z1 )∂z1 + vζ̄ (z2 )∂z2 GD (z1 , z2 )

= 2 vζ (z̄1 )∂¯z1 + vζ (z̄2 )∂¯z2 GD (z1 , z2 ).
For p ∈ R, let us write
∂
P (z) := GD (ζ, z).
∂n ζ=p
for the Poisson kernel. The Hadamard’s variational formula for the Loewner chains is given by
d
(4.8) GDt (gt (z1 ), gt (z2 )) = −P (z1 )P (z2 ),
dt t=0
see [29]. Recall that the motion of the moduli under the Loewner flow [6, Lemma 4.1] can be written as
d d
(4.9) gt (zkl ) = −vξt (gt (zkl )), gt (zkr ) = −vξt (gt (zkr )).
dt dt
Here, ξt = gt (γ(t)). Note that the Green’s function depends on the underlying domain D, which in turn can be
interpreted as a function of the moduli {zkl , zkr , z̄kl , z̄kr }, i.e.
GD (ζ, z) ≡ GD (ζ, z; {zkl , zkr , z̄kl , z̄kr }).
Denoting
∂k = ∂zk , (k = 1, 2), ∂kl = ∂zkl , ∂kr = ∂zkr , (k = 1, . . . , g),
it follows from (4.9) that
d
− GDt (gt (z1 ), gt (z2 )) = vξt (gt (z1 ))∂1 + vξt (gt (z2 ))∂2 GDt (gt (z1 ), gt (z2 ))
dt
Xg
+ vξt (gt (zkl ))∂kl + vξt (gt (z̄kl ))∂¯kl + vξt (gt (zkr ))∂kr + vξt (gt (z̄kr ))∂¯kr GDt (gt (z1 ), gt (z2 )).
k=1
Since the Green’s function is a scalar field (in other words, a (0, 0)-differential) the Lie derivative formula for
scalar fields (see [31, Proposition 4.1]) gives
d
GDt (gt (z1 ), gt (z2 )) = − Lvp + ∇Hg GD (z1 , z2 ),
dt t=0
where ∇Hg is given by (1.8). Using this and the fact p ∈ R, the Hadamard’s formula (4.8) can be rewritten as

Lvp + ∇Hg GD (z1 , z2 ) + 4∂ζ GD (ζ, z1 )∂ζ GD (ζ, z2 ) = 0.
ζ=p
The harmonic extension of this identity for ζ ∈ D gives rise to

−
L+
vζ + L vζ̄ + ∇ Hg GD (z1 , z2 ) + 4∂ζ GD (ζ, z1 )∂ζ GD (ζ, z2 ) = 0.
Combining all of the above, Eguchi-Ooguri equation (4.5) for X = Φ(z1 )Φ(z2 ) follows.
• Proof that Φ(z1 ) · · · Φ(z2n ) ∈ W. Note that by Wick’s formula, we have
X
E[A(ξ)Φ(z1 ) · · · Φ(z2n )] = E[A(ξ)Φ(zj )Φ(zk )]E[Xjk ],
j̸=k
where
Xjk = Φ(z1 ) · · · Φ(zk−1 )Φ(zk+1 ) · · · Φ(zj−1 )Φ(zj+1 ) · · · Φ(z2n ), (j ̸= k).
Then by (4.4) for the products of two GFFs, it follows that
Z
1
¯ EA(ξ)X dξ = −
X
(4.10) vζ (ξ) − vζ (ξ) ∇Hg E[Φ(zj )Φ(zk )] E[Xjk ].
2πi ∂D\R
j̸=k
On the other hand, by Wick’s formula, we have

XY
(4.11) E[Φ(z1 ) · · · Φ(z2n )] = E[Φ(zik )Φ(zjk )],
P k
where the sum is taken over all partitions P of the set {1, · · · , 2n} into disjoint pairs {il , jl }. Then by Leibniz’s
rule and rearranging the terms, we have
XY X ∇ E[Φ(z )Φ(z )]
Hg ik jk
∇Hg E[Φ(z1 ) · · · Φ(z2n )] = E[Φ(zik )Φ(zjk )]
E[Φ(zik )Φ(zjk )]
P k k
X X Y
= ∇Hg E[Φ(zj )Φ(zk )] E[Φ(zil )Φ(zjl )] ,
j̸=k Pjk l
where Pjk is the set of all partitions of {1, . . . , 2n} \ {j, k} into disjoint pairs. By another application of (4.11)
the inner sum over Pjk is simply EXjk . Combining this with (4.10) we obtain (4.4) for X = Φ(z1 ) · · · Φ(z2n ).
• W is closed under OPE products. Suppose that we have the following OPE
X
X(ζ)Y (z) = Cn (z)(ζ − z)n , ζ → z.
n
If both X and Y are in W, as ζ → z,

Z Z !
1 ¯ EA(ξ)X(ζ)Y (z) dξ =
X 1 ¯ EA(ξ)Cn (z) dξ (ζ − z)n

vζ (ξ) − vζ (ξ) vζ (ξ) − vζ (ξ)
2πi ∂D\R n
2πi ∂D\R
and X
∇Hg E[X(ζ)Y (z)] = ∇Hg E[Cn (z)](ζ − z)n .
n
By comparing the coefficients of the two expansions above, it follows that Cn ∈ W. □
Remark 8. By Lemma 4.3 with X = Φ(z1 )Φ(z2 ), we have

Z
1
¯ ∂ξ GD (z1 , ξ)∂ξ GD (ξ, z2 ) dξ = ∇H GD (z1 , z2 ).
vζ (ξ) − vζ (ξ) g
πi ∂D\R
This gives
Z
1
¯ ∂ξ G± (z1 , ξ)∂ξ GD (ξ, z2 ) dξ = ∇H G± (z1 , z2 ).
(4.12) vζ (ξ) − vζ (ξ) D g D
πi ∂D\R
The last formula will be used in the proof of the next proposition.
We now extend the generalised Eguchi-Ooguri equation to a non-trivial background charge β.
Proposition 4.4. For any X ∈ Fβ , the generalized Eguchi-Ooguri equation
Z
1
¯ EAβ (ξ)X dξ + ∇H EX = 0
(4.13) vζ (ξ) − vζ (ξ) g
2πi ∂D\R
holds in the identity chart of a chordal standard domain D.
Proof. The proof proceeds in several steps, as in the proof of Lemma 4.3.
• Identity (4.13) holds for Φβ . We show the identity for
X
−
X = Φβ = Φ + φβ , φβ (z) = i βk G+
D (qk , z) − GD (qk , z) .
Recall that
Aβ := A(0) + ib∂ − ȷβ J.
Applying Wick’s formula, we have

h i
EAβ (ξ)Φβ (z) = E ib∂ξ − ȷβ (ξ) J(ξ) Φ(z) + φβ (z)
h i
= ib∂ξ E J(ξ) Φ(z) + φβ (z) − ȷβ (ξ) EJ(ξ)Φ(0) (z)

= ib∂ξ − ȷβ (ξ) EJ(ξ)Φ(0) (z).
• Reduction to an integral identity. Since E Φβ = φβ , the last computation means that we want to show
Z
1 ¯ (ib∂ξ − jβ (ξ))EJ(ξ)Φ(0) (z) dξ + ∇H φβ = 0.

(4.14) vζ (ξ) − vζ (ξ) g
2πi ∂D\R
However by means of the asymptotic ζ → z expansions
2 1 1
vζ (z) = + O(1), ∂ζ2 G(ζ, z) = + O(1),
ζ −z 2 (ζ − z)2
and the residue calculus, we have

Z
1 1
vζ (ξ)∂ξ2 GD (ξ, z) dξ = −2∂ζ2 GD (ζ, z) + vζ′ (z)
2πi ∂D\R 2
and
Z
1 1
vζ̄ (ξ)∂ξ2 GD (ξ, z) dξ = − vζ̄′ (z).
2πi ∂D\R 2
This gives
Z
1
¯ ∂ξ EJ(ξ)Φ(0) (z) dξ = −4∂ 2 GD (ζ, z) + v ′ (z) − v ′ (z) = 0,
vζ (ξ) − vζ (ξ) ζ ζ ζ̄
2πi ∂D\R
with the last equality following from Lemma 4.5. The above implies that showing (4.14) is equivalent to showing
Z
1 ¯ jβ (ξ)EJ(ξ)Φ(0) (z) dξ = ∇H φβ .

(4.15) vζ (ξ) − vζ (ξ) g
2πi ∂D\R
• Proof of (4.15). This identity is relatively straightforward. We have that

Z
1
¯ ȷβ (ξ) EJ(ξ)Φ(0) (z) dξ
vζ (ξ) − vζ (ξ)
2πi ∂D\R
Z
1
¯ ȷβ (ξ) ∂ξ GD (ξ, z) dξ
= vζ (ξ) − vζ (ξ)
πi ∂D\R
Z
2X
¯ ∂ξ G+ (qk , ξ) − G− (qk , ξ) ∂ξ G+ (ξ, z) dξ
= βk vζ (ξ) − vζ (ξ) D D
π ∂D\R
X
−
= 2i βk ∇Hg G+ D (qk , z) − GD (qk , z) = ∇Hg φβ (z).
The second equality uses the definition of jβ , while the third uses (4.12). This proves (4.14) and (4.15).
• Identity (4.13) holds for X = Φβ (z1 )Φβ (z2 ). We now prove the identity (4.13) for X = Φβ (z1 )Φβ (z2 ). By
expanding out Aβ and φβ we obtain
Z
1
¯ EAβ (ξ)Φβ (z1 )Φβ (z2 ) dξ
2πi ∂D\R
Z
1
¯ EA(0) (ξ)Φ(0) (z1 )Φ(0) (z2 ) dξ
= vζ (ξ) − vζ (ξ)
2πi ∂D\R
Z
φβ (z2 )
¯ (ib∂ξ − ȷβ (ξ)) EJ(ξ)Φ(0) (z1 ) dξ
− vζ (ξ) − vζ (ξ)
2πi ∂D\R
Z
φβ (z1 )
¯ (ib∂ξ − ȷβ (ξ)) EJ(ξ)Φ(0) (z2 ) dξ
− vζ (ξ) − vζ (ξ)
2πi ∂D\R
= ∇Hg E Φ(0) (z1 )Φ(0) (z2 ) + φβ (z2 )∇Hg φβ (z1 ) + φβ (z1 )∇Hg φβ (z2 )
= ∇Hg E Φβ (z1 )Φβ (z2 ).
The second equality is a combination of Lemma 4.3 and identity (4.14), and the third equality is a Leibniz
formula.
• Identity (4.13) holds for X = Φβ (z1 ) · · · Φβ (zn ). To handle the general case n > 2, for each k ≤ n, let
k
XY n−k
Y
Ik := φβ (zil ) Φ(0) (zjl ) ,
l=1 l=1
where the sum is over all partitions of {1, . . . , n} into a set of set k and the complementary set of size n − k.
Then we have
Φβ (z1 ) · · · Φβ (zn ) = I0 + · · · + In .
Let us write
k
XY n−k
Y
Ak := φβ (zil ) ∇Hg E Φ(0) (zjl ),
l=1 l=1
n
X X k−1
Y n−k
Y
Bk := ∇Hg φβ (zj ) φβ (zil )E Φ(0) (zjl ).
j=1 l=1 l=1
Then it follows from the Leibniz rule that ∇Hg EIk = Ak + Bk . On the other hand, we have
Z
1
¯ EAβ (ξ)Ik dξ = Ck + Dk ,
2πi ∂D\R
where
Z
1
¯ EA(ξ)Ik dξ,
Ck := vζ (ξ) − vζ (ξ)
2πi ∂D\R
Z
1
¯ E(ib∂J(ξ) − jβ (ξ)J(ξ))Ik dξ.
Dk := − vζ (ξ) − vζ (ξ)
2πi ∂D\R
Lemma 4.3 gives that Ak = Ck , and an application of Wick’s formula and (4.14) gives that Bk = Dk . This
completes the proof. □
We finish this section with a proof of the crucial Green’s function identity (4.7). Our proof is based on
properties of the bipolar Green’s function Gp,q on the Schottky double.
Lemma 4.5. Let D be a chordal standard domain and G = GDiri be the Dirichlet Green’s function for D.
Then in the identity chart of D we have
4∂ζ2 G(ζ, z) = vζ′ (z) − vζ′ (z),
where vζ (z) = −4∂ζ GER,+ (z, ζ).
Proof. For ζ ∈ R we have defined the vector field vζ by (2.30), which is equivalent to
vξ (z) = −4 ∂ζ |ζ=ξ GER,+ (z, ζ).
With this representation of vζ in hand, proving the lemma is equivalent to showing

(4.16) ∂ζ2 G(ζ, z) = −∂ζ ∂z GER,+ (z, ζ) − ∂z GER,+ (z, ζ)
By means of the bipolar Green’s function on the Schottky double we have

G(ζ, z) = GDiri (ζ, z) = Gz,z∗ (ζ) − Gz,z∗ (ζ ∗ ),
from which it follows that
∂ζ2 G(ζ, z) = ∂ζ2 Gz,z∗ (ζ).
By representation (3.6) of the ER Green’s function we have
−∂ζ ∂z GER,+ (z, ζ) = 21 ∂ζ ∂z log Θ(A(z) − A(ζ) − e) =: Ωz (ζ),
so that
−∂ζ ∂z GER,+ (z, ζ) − ∂z GER,+ (z, ζ) = Ωz (ζ) − Ωz∗ (ζ).
Let ω1 = ∂ζ2 Gz,z∗ (ζ) dζ and ω2 = (Ωz (ζ) − Ωz∗ (ζ)) dζ. We will prove (4.16) by showing that ω0 := ω1 − ω2 = 0
on the Schottky double. As a first step, note that both ω1 and ω2 have double poles at z, z ∗ , and no poles
elsewhere, and that consequently ω0 has no singularities on the Schottky double. Hence it is a holomorphic
differential. Moreover we have that I I
ω1 = d (∂ζ Gz,z∗ (ζ)) = 0
aj aj
for every cycle aj , and by definition of Ωz (ζ) we have

I
ω2 = ∂η |η=z (log θ(ζ + aj − η) − log θ(ζ − η)) − ∂η |η=z∗ (log θ(ζ + aj − η) − log θ(ζ − η)) = 0,
aj
where θ(z) = Θ(A(z) − e) for short. The right hand side is zero by periodicity of θ around the aj cycles. Thus
ω0 = ω1 − ω2 is a holomorphic differential with vanishing aj periods. Consequently ω0 ≡ 0, by a well known
theorem in complex analysis, see [22, Proposition III.3.3]. □
Remark 9. The quantity Ωz (ζ) dζ dz is often known as the fundamental normalized bidifferential. See [23,
Chapter 2] and [20, Section 2.3] for more. It also makes an appearance in the recent work [29], which studies
scaling limits of Ising model correlation functions in multiply connected domains.
4.3. BPZ equations. In this section we prove the BPZ equations for Fβ on a chordal standard domain. The
BPZ equations may be regarded as a limiting form of the Ward equation as the argument of the Virasoro field
Tβ approaches the node of a field X ∈ Fβ . The limit is in the sense of the operator product expansion between
the Virasoro field and X. Operator product expansion with Tβ is expressed in terms of the Virasoro generators
Ln ≡ Lβ β
n . Recall that for n ∈ Z, the Virasoro generator Ln ≡ Ln is an operator that acts on fields X ∈ Fβ via
I
1
(4.17) Lβn (z) := (ζ − z)n+1 Tβ (ζ) dζ, i.e. Lβn X = Tβ ∗(−n−2) X.
2πi (z)
Equivalently, the OPE expansion between Tβ and X can be expressed in terms of the Virasoro nodes as
(Lβ0 X)(z) (Lβ

−1 X)(z)
(4.18) Tβ X(z) = · · · + 2
+ + (Lβ
−2 X)(z) + · · · .
(ζ − z) ζ −z
To express the BPZ equations we also introduce the vector field kζ which has a single pole at ζ, defined as
2
kζ (z) = .
ζ −z
Proposition 4.6. For Y ∈ Fβ and X = X1 · · · Xn , (Xj ∈ Fβ ), we have

−
2 E Tβ ∗ Y (z)X = E Y (z)L+
vz X + Lvz̄ E Y (z)X + ∇Hg ,q E Y (z)X
(4.19)
+

+ lim L+ vζ − Lkζ E Y (z)X + 2 E Tβ (z) E Y (z)X,
ζ→z
where all fields are evaluated in the identity chart of the chordal standard domain. In particular, for a holomor-
phic differential Y ∈ Fβ with conformal dimension h, we have
−
2 E Tβ ∗ Y (z)X = E Y (z)L+vz X + Lvz̄ E Y (z)X + ∇Hg ,q E Y (z)X
(4.20) h
′
i
+ h rD,0 (z) − rD,1 (z) + 2 E Tβ (z) + rD,0 (z)∂z E Y (z)X,
where rD,0 and rD,1 are given by (2.32).

Proof. It follows from Ward’s OPE [31, Proposition 5.3] that
1 +
Singζ→z Aβ (ζ)Y (z)X = Lkζ Y (z) X.
2
On the other hand, by Theorem 1.1,

−
(4.21) 2 EAβ (ζ)Y (z)X = L+
vζ + Lvζ̄ E Y (z)X + ∇Hg ,q E Y (z)X.
We now subtract the singular part from the both sides of (4.21) when ζ → z. Consequently, we obtain

− +
2 EAβ ∗ Y (z)X = E Y (z)L+ +
vz X + Lvz̄ E Y (z)X + ∇Hg ,q E Y (z)X + lim Lvζ − Lkζ E Y (z)X.
ζ→z
Then the identity (4.19) follows from Tβ = Aβ + E Tβ .

Recall that if the field Y is a differential, we have

′
2h 2
L+vζ Y (z) = hvζ (z) + vζ (z)∂z Y (z), L+
kζ Y (z) = + ∂ z Y (z),
(ζ − z)2 ζ −z
see [31, Proposition 4.1]. It now follows from (2.32) and (2.33) that

+
h
′ 2 2 i
lim L+ vζ − L kζ Y (z) = lim h v ζ (z) − + v ζ (z) − ∂z Y (z)
ζ→z ζ→z (ζ − z)2 ζ −z
h i
′
= h rD,0 (z) − rD,1 (z) + rD,0 (z)∂z Y (z).
This completes the proof. □
Proposition 4.6 is especially fruitful when the field Y is Virasoro primary. By definition, a Virasoro primary
field X ∈ Fβ is a differential of conformal dimension (λ, λ∗ ). Then by [31, Proposition 7.5], it follows that
(4.22) Lβ
0 X = λX, Lβ
−1 X = ∂X, Lβ
nX = 0 for all n ≥ 1.
Similar identities also hold for X̄. Furthermore, X is called current primary if there exist q, q∗ such that
J0β X = −iqX, J0β X̄ = iq ∗ X̄, Jnβ X = Jnβ X̄ = 0 for all n ≥ 1,
∗
where the numbers q, q are called charges of X and the current generators are given by
I
1
Jnβ (z) := (ζ − z)n Jβ (ζ) dζ, i.e. Jnβ X = Jβ ∗(−n−1) X.
2πi (z)
Proposition 4.7 (Cf. Proposition 11.2 [31]). Let O ∈ Fβ be a current primary field with charges q, q∗ . Then
if 2q(b + q) = 1, we have
1
(4.23) Lβ
−2 O = (Lβ )2 O.
2q 2 −1
By combining Proposition 4.7 with Proposition 4.6 we can now give the proof of Theorem 1.2, the BPZ-Cardy
equations for a chordal standard domain.
Proof of Theorem 1.2. Recall that the field Υ(z) is defined by
Υ(z) := O{α}, α = (a · z − a · ∞, 0),
where a ∈ R satisfies 2a(a + b) = 1 and b ∈ R is determined by the neutrality condition (1.4) of β. Then Υ(z)
is primary with dimensions λ = a2 /2 − ab = h1,2 and current primary with charge a, see [31, Proposition 12.4].
Thus by combining the level two equation (4.23) with (4.17) and (4.22), we obtain
1 2
Tβ ∗ Υ(z) = ∂ Υ(z).
2a2 z
Then the BPZ equation (1.14) follows from Proposition 4.6.
Now to derive the BPZ-Cardy equation (1.15), first rewrite the BPZ equation (1.14) in terms of EΥ X via
1 2 1h 2 2
i
∂z E Υ X(z)EΥ(z) = EΥ(z)∂ z EΥ X(z) + 2∂ z E Υ X(z)∂z EΥ(z) + E Υ X(z)∂z EΥ(z)
a2 h a2 i
+ − ′
= Ľvz + Lvz̄ + ∇Hg ,q + h1,2 rD,0 (z) − rD,1 (z) + 2 E Tβ (z) + rD,0 (z)∂z EΥ X(z)EΥ(z).
On the other hand, by multiplying EΥ X to the null-vector equation (1.16), we have

1 2
h
′
i
E Υ X∂ z EΥ(z) = E Υ X ∇ Hg ,q + h1,2 rD,0 (z) − rD,1 (z) + 2 E Tβ (z) + rD,0 (z)∂ z EΥ(z).
a2
Subtracting these equations gives rise to
1 2

EΥ(z)∂ z EΥ X(z) + 2∂ z EΥ X(z)∂z EΥ(z)
a2
−
= EΥ(z) Ľ+ vz + Lvz̄ + ∇Hg ,q + rD,0 (z)∂z EΥ X(z).
Therefore we conclude (1.15). □
5. SLE Martingale Observables

In this section we prove Theorem 1.3, that the correlation functions of all fields in the family Fβ are
martingale observables for an associated SLE process. The main ingredient is the BPZ-Cardy equation (1.15),
which was derived without any reference to SLE. Theorem 1.3 gives an infinite family of martingale observables
for the SLE process, and the martingale can depend on any finite collection of marked points in the domain.
Note that the BPZ-Cardy equation allows us to verify the martingale observable property without having
explicit formulas for the martingale. In fact we typically do not even have an explicit formula for the Loewner
vector field vξ or the stochastic differential equation (1.18). Nonetheless the martingale observable property
is verified by a relatively straightforward application of Ito’s formula, thereby demonstrating the power of the
CFT approach.
Proof of Theorem 1.3. Using the BPZ-Cardy equation (1.15), the proof follows along the same lines in the
literature [31, Section 14.3], [33, Section 8.3] and [10, Section 5.3]. We give a shortened version here.
Recall that M is the non-random conformal field
EΥ(ξ)X
M = EΥ X = ,
EΥ(ξ)
where Υ(ξ) = Oβ {a · ξ − a · ∞}. Thus M depends on the marked point ξ, the marked points in β, the nodes of
X , and the modular parameters zkl , zkr of the chordal standard domain. Denote by z the modular parameters
and q the marked points of β. Define a function m(ξ, z, q, t) by
m(ξ, z, q, t) := (Rξ,z,q ∥ gt−1 ), Rξ,z,q = EΥ X .

We need to prove that Mt := m(ξt , z(t), q(t), t) is a martingale, where ξt is the solution to the stochastic
differential equation defined by (1.18) and (1.19). By Ito’s formula and (1.19) we have
1 ∂m
dMt = (∂ξ m)(ξt , z(t), q(t), t) dξt + (∂ξ2 m)(ξt , z(t), q(t), t) d⟨ξ⟩t + (ξt , q(t), t) dt
2 ∂t
X
+ żk (t)∂zk m(ξt , z(t), q(t), t) + żk (t)∂¯zk m(ξt , z(t), q(t), t) dt
k
X
+ q̇k (t)∂qk m(ξt , z(t), q(t), t)
k
√

κ ∂m
= κ(∂ξ m) dBt + κ ∂ξ log EΥ(ξ) − r0 (ξ) ∂ξ m + ∂ξ2 m + − ∇Hg ,q m dt.
2 ∂t
The drift term vanishes by the BPZ-Cardy equations, we will now explain why. By definition of m the time
derivative ∂m/∂t is given by
∂m d −1
(ξ, z, q, t) = EX ∥ gt+s .
∂t ds s=0
Let fs,t = gt+s ◦ gt−1 . Then ∂s fs,t = −vξt+s ◦ fs,t by the Loewner equation, with f0,t = id. We deduce that
fs,t = id − svξt + o(s), s ↓ 0.
Since gt+s = fs,t ◦ gt , the definition of the Lie derivative gives
d −1 d
EX ∥ gt+s = EX ∥ gt−1 ◦ fs,t
−1
= −Ľvξ m ∥ gt−1 .
ds s=0 ds s=0
Taken together and invoking the BPZ-Cardy equations (1.15) (with ξ on the real line), the last three calculations
show that the drift term of dMt vanishes. □
We conclude with some example of martingale observables. We first discuss the bosonic observable for these
SLE systems, i.e. the one-point function of the modified Gaussian field Φβ . Here the modification comes in
two different forms: one in the background charge β and another through the OPE exponential Υ(ξ). This
modification affects both the structure of the martingale observable and the associated SLE system. It is the
analogue of the Schramm-Sheffield observable for chordal SLE(κ, ρ) systems in simply connected domains.
Example: Bosonic observable
The simplest one-point observable is obtained by P taking X = Φβ (z), typically for z in the interior of
the chordalP standard domain. Recall that β = k βk · qk for qk ∈ R and the βk satisfy the neutrality
condition βk = bχ. Here, we have Υ(ξ) = Oβ {a · ξ − a · ∞, 0}. Then an exercise in Wick’s calculus
shows that
EΥ Φβ (z) = φβ+a·ξ−a·∞ (z).
As a conformal field, the latter depends on the marked points qk , ξ, and z. It is a [0, 0]-differential in
all of the q-variables and a PPS(ib, −ib) form in the z variable. By Theorem 1.3, it is a martingale
observable for the SLE process with driving function
√
dξt = κ dBt + ΛDt (ξt ) dt,
with the drift function Λ(ξ) given by
Λ(ξ) := κ∂ξ log EΥ(ξ) − r0 (ξ).
Here Dt = gt (D0 \γ[0, t]), where γ is the SLE curve. Thus Dt is yet another chordal standard domain,
and the field ΛDt is evaluated in this domain. The nature of the OPE exponential Υ(ξ) makes it difficult
to give an exact expression for the term EΥ(ξ), even in the standard coordinate chart. It only depends
on the divisor β and the marked point ξ. Similarly, the conformal field EΥΦβ (z) only depends on these
divisors and ξ, along with the point z ∈ D0 . The coordinate chart gt−1 is well-defined around these
marked points, so by evaluating the field EΥ Φβ (z) along the charts gt−1 the martingale in coordinates is
X
t 7→ −2b arg gt′ (z) + 2 βk Im G+
D (gt (qk ), gt (z))
k
by (1.5). For z on the real line (but distinct from the qk and ξ) it can be shown that EΥ Φβ (z) is piecewise
constant.
In the multiply connected setting it is very difficult to provide any explicit formulas for the martingales,
other than in terms of complicated special functions. Nonetheless we can see interesting functional relations
arise between the correlation functions, even without explicit expressions available for them.
Example: A functional relation for a sequence of correlation functions derived from Tβ
In the introduction we argued that CFT can be used to derive functional equations between correlation
functions. Via the Virasoro field Tβ we can use the Eguchi-Ooguri form of Ward’s equation to derive a
recursive sequence of correlation functions. For the setup, let D be a chordal P standard domain and fix
a sequence of distinct points z1 , z2 , . . . ∈ D ∪ R. FixPa background charge β = k βk · qk where one has
qk ∈ R and the βk satisfy the neutrality condition βk = bχ. Assume that the qk are all distinct from
the zj . We will consider the conformal fields
Vn = EΥ(ξ)Xn , Xn = Tβ (z1 ) . . . Tβ (zn ),
where in this case Υ(ξ) = Oβ {a · ξ − a · ∞, 0} and we recall that a is a solution to 2a(a + b) = 1. Here ξ
is distinct from the qk and zj . As a (non-random) conformal field, Vn is
• a [0, 0]-differential at the qk ,
• a Schwarzian form of order c/12 = 1/12 − b2 at the zj , j = 1, . . . , n,
• and an [h1,2 , 0]-differential at ξ, where h1,2 = a2 /2 − ab.
It will also transform as a differential at infinity, but in these calculations we only consider conformal
maps that fix infinity and have the hydrodynamic normalization there, so the transformation rule at
infinity does not enter in. More precisely, the Loewner vector field has a triple zero at infinity and the
Lie derivative at infinity in Ward’s equation does not appear in the identity chart of a chordal standard
multiply connected domain. Note the sequence Xn+1 is built up by successive multiplications by the
stress tensor Tβ , i.e. there is the recursive relation Xn+1 = Tβ (zn+1 )Xn . At the same time, the Ward
equation relates multiplication by the stress tensor to the Lie derivative. By the Ward equation (1.13),
and recalling that Aβ = Tβ − E Tβ , we obtain the recursive equation
(5.1) 2Vn+1 = 2Vn E Tβ (zn+1 ) + (L+
vz + ∇Hg ,q )Vn .
n+1
In a simply connected domain, this recursion formula coincides with that for an avoiding probability of
the SLE(8/3) curve (in the c = 0 case), see [31, Section 14.5] and [25]. This is closely related to the
restriction property and we refer to [1] for a related recent work in a multiply connected domain. In
principle, (5.3) can be solved given the one-point function E Tβ (z) and an initial value V0 = EΥ(ξ).
Although Vn is not a martingale observable, Theorem 1.3 states that Mn := Vn /EΥ(ξ) is a martingale
observable for an SLE process, when evaluated on the random charts gt−1 . By dividing through (5.1) by
EΥ(ξ) and using the Leibniz rule we obtain
Mn
(5.2) 2Mn+1 = 2Mn E Tβ (zn+1 ) + (L+ vzn+1 + ∇Hg ,q )Mn + (L+ + ∇Hg ,q )EΥ(ξ).
EΥ(ξ) vzn+1
In the higher genus case that we consider both will be expressed in terms of complicated special functions,
which makes it difficult to find a closed form solution to the recursive equation. The transformation
rules for Vn allow us to compute the Lie derivative term as

n h n
′
X i c X ′′′
L+
vzn+1 Vn = vzn+1 (ξ)∂ξ Vn + h1,2 vzn+1 (ξ)Vn + vzn+1 (zj )∂zj + 2vz′ n+1 (zj ) Vn + v (zj )Vn,j ,
j=1
12 j=1 zn+1
where Vn,j = EΥ(ξ)Xn,j and Xn,j = Tβ (z1 ) . . . T̂β (zj ) . . . Tβ (zn ). Here, the hat indicates that term is
left out of the product. Combining the above, it follows that

X
′
c X ′′′
(5.3) 2V n+1 = 2E T (z
β n+1 ) + v (z )∂
zn+1 j zj + 2v zn+1 j (z ) + Dn [ξ, q] Vn + vzn+1 (zj )Vn,j ,
j
12 j
where
Dn [ξ, q] := h1,2 vz′ n+1 (ξ) + vzn+1 (ξ)∂ξ + ∇Hg ,q .
After normalizing (5.3) the sequence Mn satisfies
X
′
2Mn+1 = 2E Tβ (zn+1 ) + vzn+1 (zj )∂zj + 2vzn+1 (zj ) + Dn [ξ, q] Mn
j
(5.4)
c X ′′′ Dn [ξ, q]EΥ(ξ)
+ v (zj )Mn,j + Mn ,
12 j zn+1 EΥ(ξ)
where Mn,j = Vn,j /EΥ(ξ). This formula agrees with (5.2) since each Mn is a [0, 0]-differential at the qk
and at ξ, and a Schwarzian form of order c/12 at z1 , . . . , zn , whereas EΥ(ξ) is an [h1,2 , 0]-differential at
ξ.
References
[1] J. Aru and P. Bordereau, SLE and its partition function in multiply connected domains via the Gaussian Free Field and
restriction measures, arXiv:2405.20148.
[2] H. F. Baker, Abelian functions: Abel’s theorem and the allied theory of theta functions, Cambridge University Press, 1995.
[3] M. Bauer and D. Bernard, SLEκ growth processes and conformal field theories, Phys. Lett. B 543 (2002), 135–138.
[4] M. Bauer and D. Bernard, Conformal field theories of stochastic Loewner evolutions, Comm. Math. Phys. 239 (2003),
493–521.
[5] R. O. Bauer and R. M. Friedrich, On radial stochastic Loewner evolution in multiply connected domains, J. Funct. Anal. 237
(2006), 565–588.
[6] R. O. Bauer and R. M. Friedrich, On chordal and bilateral SLE in multiply connected domains, Math. Z. 258 (2008), 241–265.
[7] B. Bayraktaroglu, K. Izyurov, T. Virtanen and C. Webb, Bosonization of primary fields for the critical Ising model on
multiply connected planar domains, arXiv:2312.02960.
[8] A. A. Belavin, A. M. Polyakov and A. B. Zamolodchikov, Infinite conformal symmetry in two-dimensional quantum field
theory, Nuclear Phys. B 241 (1984), 333–380.
[9] F. Bracci, Filippo, M. D. Contreras and S. Dı́az-Madrigal and A. Vasilév, Classical and stochastic Löwner-Kufarev equations,
In Harmonic and complex analysis and its applications, Trends Math., pages 39–134. Birkhäuser/Springer, Cham, 2014.
[10] S.-S. Byun, N.-G. Kang and H.-J. Tak, Conformal field theory for annulus SLE: partition functions and martingale-
observables, Anal. Math. Phys. 13 (2023), no. 1, 87 pp.
[11] Z.-Q. Chen, M. Fukushima and T. Murayama, Stochastic Komatu-Loewner Evolutions, World Scientific. 2023.
[12] Z.-Q. Chen, M. Fukushima and S. Rohde, Chordal Komatu-Loewner equation and Brownian motion with darning in multiply
connected domains, Trans. Amer. Math. Soc. 368 (2016), 4065–4114.
[13] D. Crowdy and J. Marshall, Analytical formulae for the Kirchhoff-Routh path function in multiply connected domains, Proc.
R. Soc. Lond. Ser. A Math. Phys. Eng. Sci. 461 (2005), 2477–2501.
[14] D. Crowdy and J. Marshall, Conformal mappings between canonical multiply connected domains, Comput. Methods Funct.
Theory 6 (2006), 59–76.
[15] D. Crowdy and J. Marshall, Computing the Schottky-Klein prime function on the Schottky double of planar domains, Comput.
Methods Funct. Theory 7 (2007), 293–308.
[16] D. Crowdy and J. Marshall, Green’s functions for Laplace’s equation in multiply connected domains, IMA J. Appl. Math. 72
(2007), 278–301.
[17] S. Drenning, Excursion reflected Brownian motion and Loewner equations in multiply connected domains, ProQuest LLC,
Ann Arbor, MI, 2011. Thesis (Ph.D.)–The University of Chicago.
[18] J. Dubédat, Commutation relations for Schramm-Loewner evolutions, Comm. Pure Appl. Math. 60 (2007), 1792–1847.
[19] J. Dubédat, SLE and the free field: partition functions and couplings, J. Amer. Math. Soc. 22 (2009), 995–1054.
[20] J. Dubédat, SLE and Virasoro representations: localization, Comm. Math. Phys. 336 (2015), 695–760.
[21] T. Eguchi and H. Ooguri, Conformal and current algebras on a general Riemann surface, Nuclear Phys. B 282 (1987),
308–328.
[22] H. Farkas and I. Kra, Riemann surfaces, Graduate Texts in Mathematics. Springer New York, 1992.
[23] J. Fay, Theta functions on Riemann surfaces, Lecture Notes in Mathematics. Springer, 1973.
[24] J. Fay, Kernel functions, analytic torsion, and moduli spaces, Mem. Amer. Math. Soc. 96 No. 464 (1992), vi+123.
[25] R. Friedrich and W. Werner, Conformal restriction, highest-weight representations and SLE, Comm. Math. Phys. 243 (2003),
105–122.
[26] I. A. Gruzberg, Stochastic geometry of critical curves, Schramm-Loewner evolutions and conformal field theory, J. Phys. A
39 (2006), 12601–12655.
[27] D. A. Hejhal, Theta functions, kernel functions and Abelian integrals, volume 129. American Mathematical Soc., 1972.
[28] J. H. Hubbard, Teichmüller theory and applications to geometry, topology, and dynamics. Vol. 1, Matrix Editions, Ithaca,
NY, 2006.
[29] K. Izyurov and K. Kytölä, Hadamard’s formula and couplings of SLEs with free field, Probab. Theory Related Fields 155
(2013), 35–69.
[30] S. Janson, Gaussian Hilbert spaces, volume 129 of Cambridge Tracts in Mathematics. Cambridge University Press, Cambridge,
1997.
[31] N.-G. Kang and N. G. Makarov, Gaussian free field and conformal field theory, Astérisque 353 (2013), viii+136.
[32] N.-G. Kang and N. G. Makarov, Calculus of conformal fields on a compact Riemann surface, arXiv:1708.07361.
[33] N.-G. Kang and N. G. Makarov, Conformal field theory on the Riemann sphere and its boundary version for SLE,
arXiv:2111.10057.
[34] P. Koebe, Abhandlungen zur Theorie der konformen Abbildung. (German) IV. Abbildung mehrfach zusammenhängender
schlichter Bereiche auf Schlitzbereiche, Acta Math. 41 (1916), 305–344.
[35] A. Kokotov and D. Korotkin, Tau-functions on spaces of abelian differentials and higher genus generalizations of Ray-Singer
formula, J. Differential Geom. 82 (2009), 35–100.
[36] Y. Komatu, On conformal slit mapping of multiply-connected domains, Proc. Japan Acad. 26 (1950), 26–31.
[37] I. Krichever, A. Marshakov and A. Zabrodin, Integrable structure of the Dirichlet boundary problem in multiply-connected
domains, Comm. Math. Phys. 259 (2005), 1–44.
[38] K. Kytölä, On conformal field theory of SLE(κ, ρ), J. Stat. Phys. 123 (2006), 1169–1181.
[39] D. Mumford, Tata Lectures on Theta. II. Progress in Mathematics, vol. 43. Birkhäuser Boston Inc., Boston (1984). Jacobian
theta functions and differential equations, With the collaboration of C. Musili, M. Nori, E. Previato, M. Stillman and H.
Umemura
[40] D. Mumford, C. Series and D. Wright, Indra’s pearls: The vision of Felix Klein, Cambridge University Press, 2002.
[41] F. W. Olver, D. W. Lozier, R. F. Boisvert and C. W. Clark, NIST handbook of mathematical functions, US Department of
Commerce, National Institute of Standards and Technology, 2010.
[42] E. Peltola, Toward a conformal field theory for Schramm-Loewner evolutions, J. Math. Phys. 60 (2019), 103305.
[43] C. Poor, Fay’s Trisecant Formula and Cross-Ratios, Proc. Am. Math. Soc. 114 (1992), 667–671.
[44] I. Rushkin, E. Bettelheim, I. Gruzberg and P. Wiegmann, Critical curves in conformally invariant statistical systems, J.
Phys. A 40 (2007), 2165–2195.
[45] M. Schiffer, Recent advances in the theory of conformal mapping, Appendix to: R. Courant, Dirichlet’s Principle, Conformal
Mapping and Minimal Surfaces, 1950.
[46] D. Zhan, Random Loewner chains in Riemann surfaces, PhD thesis, California Institute of Technology, 2004.
[47] D. Zhan, The scaling limits of planar LERW in finitely connected domains, Ann. Probab. 36 (2008), 467–529.
Department of Mathematics, University of Utah, Salt Lake City, UT 84112, USA, School of Mathematics, Korea
Institute for Advanced Study, Seoul, 02455, Republic of Korea
Email address: alberts@math.utah.edu
Department of Mathematical Sciences and Research Institute of Mathematics, Seoul National University, Seoul
151-747, Republic of Korea
Email address: sungsoobyun@snu.ac.kr
School of Mathematics, Korea Institute for Advanced Study, Seoul, 02455, Republic of Korea
Email address: namgyu@kias.re.kr

2407.08220v1

Uploaded by

Document Information

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

2407.08220v1

Uploaded by

Copyright:

Available Formats

CONFORMAL FIELD THEORY OF GAUSSIAN FREE FIELDS

IN A MULTIPLY CONNECTED DOMAIN

TOM ALBERTS, SUNG-SOO BYUN, AND NAM-GYU KANG

1. Introduction and Main results

Date: July 12, 2024.

where χ = 2 − 2g is the Euler characteristic of a compact Riemann surface of genus g.

where ∂ξ = ∂ + ∂¯ is the differential operator with respect to the real variable ξ.

See Subsection 4.3 for further details.

2. Green’s functions and Loewner vector field

Example: S-K prime function on an annulus

Given the basis {ηj } the period matrix τ is defined as

for the Abel-Jacobi map.

where ⟨·, ·⟩ is the standard inner product.

Example: Green’s functions on an annulus

Example: The genus one case

the ER Poisson kernel. Let us also define complex ER Poisson kernel by

satisfies the hydrodynamic normalization

See also [17, Proposition 6.4] for a probabilistic representation.

Example: Loewner vector field on a simply connected domain

In the upper-half plane H with Dirichlet boundary conditions we have

Example: Loewner vector field on an annulus

Note that by (2.17), we have  

By [41, Eq.(23.9.3)], ζ(z) has the asymptotic expansion

3. Gaussian free fields in a multiply connected domain

Although Φ cannot be realized as a random function, we write it as a (correlation functional-valued) function

Example: GFF on a cylinder via the GFF on a torus

Let vbj be the holomorphic function on F whose imaginary part is e

On the other hand, it follows from (2.8) that

Remark 7. As a consequence of Lemma 3.2, we have the functional relations:

Proof of Proposition 3.1. We need to check

bi-variant multivalued field with correlation

E Φ+ (ζ)Φ+ (z) = − log E(ζ, z) + π⟨(Im τ )−1 A(ζ), A(z))⟩.

for any double divisor τ = (τ + , τ − ) (τ ± = τj± · zj ) satisfying the neutrality condition

The OPE exponential O{τ } is then defined by

Based on the definition of the current field we have the following.

Proposition 3.5. The field Φβ has a stress tensor

4. Ward’s and BPZ equations

which follows because ∂D consists of R and a collection of horizontal slits.

(4.7) 4∂ζ2 GD (ζ, z) = vζ′ (z) − vζ̄′ (z).

This identity is proved separately in Lemma 4.5 below.

On the other hand, by Wick’s formula, we have

If both X and Y are in W, as ζ → z,

Remark 8. By Lemma 4.3 with X = Φ(z1 )Φ(z2 ), we have

Applying Wick’s formula, we have

However by means of the asymptotic ζ → z expansions

and the residue calculus, we have

• Proof of (4.15). This identity is relatively straightforward. We have that

By means of the bipolar Green’s function on the Schottky double we have

for every cycle aj , and by definition of Ωz (ζ) we have

(Lβ0 X)(z) (Lβ

Proposition 4.6. For Y ∈ Fβ and X = X1 · · · Xn , (Xj ∈ Fβ ), we have

where rD,0 and rD,1 are given by (2.32).

Then the identity (4.19) follows from Tβ = Aβ + E Tβ .

Proof of Theorem 1.2. Recall that the field Υ(z) is defined by

Υ(z) := O{α}, α = (a · z − a · ∞, 0),

On the other hand, by multiplying EΥ X to the null-vector equation (1.16), we have

Therefore we conclude (1.15). □

5. SLE Martingale Observables

m(ξ, z, q, t) := (Rξ,z,q ∥ gt−1 ), Rξ,z,q = EΥ X .

Example: Bosonic observable

Example: A functional relation for a sequence of correlation functions derived from Tβ

rules for Vn allow us to compute the Lie derivative term as

You might also like

Note that by (2.17), we have