Download - M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

Transcript
Page 1: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

LAX FORMULATION FOR HARMONIC MAPS TO A MODULI OF BUNDLES

RICHARD DERRYBERRY

Abstract. I define an algebraic metric and closed 3-form on a subspace M of the moduli of G-bundles ona complex projective curve C, and show that the resulting two-dimensional σ-model with target M has a

zero-curvature formulation.

1. Introduction

Harmonic maps are mappings between pseudo-Riemannian manifolds that satisfy a certain generalisation ofLaplace’s equation. The exact partial differential equation can be derived as the critical points of an actionfunctional. Usually this functional is taken to be the Dirichlet energy, however I will take the slightly broaderview that the functional merely needs to be the action functional of a classical σ-model:1 e.g. the functionalconsidered in this paper can be found at (5.1).

Harmonic maps from a surface to a Lie group are famously integrable, in the sense that they admit a Lax, orzero-curvature, formulation [Poh76]. There is a rich literature on integrable harmonic map equations, rangingfamiliar coset models (particularly Riemannian symmetric spaces) [Byk16, Zar19] to affine Gaudin models[DLMV19b, DLMV19a] and integrable E-models [LV21]. However, in all of these examples the target spaceis (topologically!) a group manifold or coset space, and the spectral curve is genus zero. The core novelty ofthis paper is the expansion of possible target space topologies to certain moduli spaces of bundles, and theexistence of integrable σ-models with spectral curve of arbitrary genus.

In this paper I will introduce an infinite number of new examples, approximately parametrized by the choiceof a Riemann surface equipped with a holomorphic 1-form with simple zeroes and double poles. The targetspace is an open subspace of the moduli of bundles on the Riemann surface, which I will show may be endowedwith an algebraic metric (so that pseudo-Riemannian manifolds may be obtained by taking a real slice ofthe target space). Algebraic metrics are significantly rarer than their pseudo-Riemannian counterparts, andthe existence of one on the moduli of bundles is certainly unexpected. Another novelty of the construction Iwill present is that the spectral curves of the integrable systems may be of arbitrary genus, and not just ofgenus 0 (as is the case for the previously known examples in the literature). The construction in this paperis based on the physical engineering of a class of new integrable σ-models from 4d Chern-Simons theory byCostello and Yamazaki [CY19], which I will briefly review in Section 2.

1.1. Integrable systems. Having introduced the concept without giving a precise definition, it seems pru-dent to ask to the question: what is an integrable system? A first approximation to a definition might be asfollows: an integrable system is a system of differential equations that can be integrated, or in plain language,solved. Of course this naive definition is deficient: it captures too many systems, while failing to describethe nature or method of solution.

As a warm up to integrable systems in classical field theory, let’s consider integrable systems in classicalmechanics. The dynamics of a system in classical mechanics is described by a path in phase space, a 2n-dimensional symplectic manifold that describes all possible states of the physical system. The symplectic

Date: September 16, 2021.1It is not important for this paper to know what this shade of physical inspiration means, but for the curious: A σ-model is

a (classical or quantum) field theory whose fields are maps σ : Σ→M, where Σ is a ‘spacetime’ manifold, andM is the ‘target’manifold.

1

arX

iv:2

106.

0978

1v2

[m

ath.

AG

] 1

4 Se

p 20

21

Page 2: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

2 RICHARD DERRYBERRY

form gives rise to a Poisson bracket on functions −,−, and f, g = 0 if and only if the function f isconstant along the level sets of g (and vice-versa). The physical system is described by a particular functioncall the Hamiltonian H, and a function that Poisson commutes with H is called a conserved quantity orintegral of motion.

Integrability in this context is the following statement: that one can find n algebraically independent Poissoncommuting conserved quantities. This allows one to solve the system of differential equations determined byH as follows:

• By the Arnold-Liouville theorem, there exists a canonical collection of coordinates on the phasespace called action-angle coordinates [Arn89]. Roughly, the action coordinates parametrize the baseof possible values of our n conserved quantities, while the angle coordinates2 parameterize the fibresover each point in the base.

• In order to have a well-posed problem, we need to supplement our system of differential equationswith 2n pieces of initial data. Start by choosing n of these to be value in our base of conservedquantities. Since they are conserved, the path that describes the solution to our system of differentialequations must stay in the fibre over this point: i.e. we have fixed the action coordinates, and thedynamics is now entirely concentrated in the angle coordinates.

• The path that describes the solution will vary linearly in the angle coordinates. Et voila! We havedetermined the path in phase space that solves our system of differential equations with given initialconditions!

For this paper, I’m interested not in classical mechanical systems, but in classical field theories. Here thesituation is trickier: the phase space of such a theory is infinite dimensional, and so one must try to cook upinfinitely many Poisson commuting functions on phase space and attempt to use these to solve the equationsof motion, for instance via the inverse scattering method [FT07].

The key concept underlying the integrability of these infinite dimensional dynamical systems is that of theLax pair, or zero-curvature, formulation of the system. The basic idea here is to determine a recipe that foreach classical field σ produces a connection D(σ) on spacetime, such that σ satisfies the classical equationsof motion if and only if the connection D(σ) is flat. Integrals of motion may then be obtained by applyinginvariant polynomials to the monodromies of this connection.

As described above, we would still only obtain finitely many integrals of the motion for our infinite dimensionalsystem. To remedy this, one usually produces for each σ not a single connection but a family of connectionsD(σ)z parametrized by z ∈ C, the spectral parameter on an algebraic curve called the spectral curve. Theflatness condition becomes the requirement that D(σ)z is simultaneously flat for all values of z ∈ C. Applyinginvariant polynomials to the monodromies of this family of connection then produces infinitely many integralsof the motion; for instance, if C = C× one can obtain a countable family of commuting conserved quantitiesas the coefficients of a power series expansion of3 Tr HolD(σ)z around z = 0 or z =∞ [FT07].

1.2. An explicit example. For concreteness, let’s suppose that our spacetime is Σ = S1 × R (or in the

algebraic setup of this paper, C× × D). Let M be any space for which the harmonic map equations areintegrable, and let C be the spectral curve; for example, M = G and C = C×. Let Harm(Σ) denote themoduli of solutions to the harmonic map equations.

Then for each field σ ∈ Harm(Σ) there is a family of flat connections D(σ)z, z ∈ C, on spacetime. Wecan take its holonomy around C× (the exact loop does not matter due to flatness), and apply an invariantfunction f ∈ C[G]G to get a number

Ff,z(σ) = f(HolC×D(σ)z) ∈ C.

2If our fibres are compact, they will be real tori.3Or a function derived from this one.

Page 3: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

LAX FORMULATION FOR HARMONIC MAPS TO A MODULI OF BUNDLES 3

As σ varies this gives a function Ff,z : Harm(Σ)→ C for each invariant function f and point z in the spectralcurve. These functions are all conserved by flatness of the connections D(σ), and Poisson commute with eachother [BBT03]. So we’ve found infinitely many commuting conserved quantities.

It remains to ask: what is the space Harm(Σ)? Working on the formal disc, we can expand a solution as aformal power series with coefficients in the algebraic loop group of the target spaceM. Because the equationsof motion are second order, the solution only depends on a point and a first order variation in the loop group.Hence Harm(Σ) ∼= TLM. The metric on M induces a metric on LM, so we can identify this with T ∗LM.This suggests that we have found infinitely many conserved quantities for a quantum mechanical system onthe loop space of M.

Remark 1.1. An important word of caution: I say ‘suggests’ above because it is not clear that the Poissonstructure obtained from the harmonic map equations and the canonical Poisson structure on T ∗LM agree.

1.3. Future directions. The work in this paper represents only the beginning of what could be studied forthese harmonic map equations:

• The existence of a zero-curvature formulation with higher genus spectral curve C suggests that theremay be an action on the space of solutions by the group of maps from C into G. For genus zerothis is the entry point for the loop group action on solutions that allows one to cook up non-trivialsolutions and understand the geometry of the moduli space of harmonic maps [Uhl89]. The hope isthat a similar action could be exploited to similar effect in the higher genus case.

• The 1-loop β-function of the σ-model corresponding to our harmonic map equations is a modifiedRicci flow equation on the target space [CFMP85]. A priori this modified Ricci flow equation isdifficult to understand; however Costello has communicated to me the following conjecture regardingits behaviour:

Conjecture 1 (Costello). Let N denote the moduli space of Riemann surfaces equipped with aholomorphic 1-form ω that has only simple zeroes and double poles, together with a decompositionof the zeroes of ω into two equally sized groups, D1 and D2.

Then the modified Ricci flow on the space of metrics on the target is identical to the flow generatedby a certain flow on N , where:

– The closed periods∮ω are held fixed.

– The periods∫ qpω are held fixed when p, q are both in either D1 or D2.

– When p ∈ D1 and q ∈ D2 the periods∫ qpω satisfy d

∫ qpω∣∣∣ε=0

= 1, where ε is the parameter of

the flow on N .

Some evidence for this conjecture is presented in section 9.1.• Finally, the flat connections on our target space can be interpreted via pushforward as D-modules onBunG(C). We can therefore ask the standard question: what are the corresponding Langlands dualobjects?

1.4. Structure of the paper. The structure of the paper is as follows:

In Section 2 I introduce the core problem of the paper, and give a brief review of the motivating physicalanalysis of Costello and Yamazaki.

Sections 3–5 are concerned with the definition of the field theories introduced by Costello and Yamazaki.These theories are two-dimensional σ-models, and thus require a metric (Section 3) and a closed 3-form(Section 4) be defined on the target space. Once these have been defined, one can write down the actionof the classical field theory and derive its equations of motion (Section 5). In Section 3 I also discuss someproperties of the metric: I calculate its first derivative in a natural collection of coordinates, and consider thesignature of an induced pseudo-Riemannian metric on a real slice of the target space.

Page 4: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

4 RICHARD DERRYBERRY

In Sections 6–7 I consider some novel G-connections that were predicted by Costello and Yamazaki. InSection 6 I show that there are two different flat algebraic G-connections on the target of the σ-model; inSection 7 I use these two connections to build a G-connection on spacetime for every field in the σ-model.

In Section 8 I integrate the previous two topics to prove that the classical equations of motion of the σ-modelshave a zero-curvature formulation.

Finally, in Section 9 I consider in detail the example of CP1 with two marked points. In special cases thisexample recovers the motivating example of harmonic maps into Lie groups.

1.5. Acknowledgements. I sincerely thank my postdoctoral mentor Kevin Costello for his support duringthis project, as well as for suggesting the problem; Benoit Vicedo for a productive discussion on a previousversion of this paper; David Ben-Zvi, with whom I had multiple extremely useful correspondences; and TomMainiero and Sebastian Schulz for comments on a draft version of this paper. Finally, I would like to thankZahavi Derryberry for annotating my notes, and Felix Derryberry for providing typing assistance.

Research at Perimeter Institute is supported in part by the Government of Canada through the Departmentof Innovation, Science and Economic Development and by the Province of Ontario through the Ministry ofColleges and Universities.

2. Setup and background of the problem

Our starting data is a proper curve C over C of genus g, equipped with a meromorphic 1-form ω which hasonly simple zeros and double poles. Let Q be half the divisor of poles of ω, and divide the divisor of zeroesinto two equal degree divisors P1 and P2. We then form the two divisors D1,2 = P1,2 − Q on C of degreeg − 1 which satisfy O(D1 +D2) = KC .

Define BunG(C|Di) ⊂ BunG(C,Q) to be the open substack of the moduli of G-bundles on C trivialised at Qsatisfying the cohomology vanishing condition H•(C; gP (Di)) = 0. Note that by our assumptions and Serreduality,

BunG(C|D1) = BunG(C|D2) =:M.

Finally, let P denote the universal G-bundle on C × BunG(C,Q), and let Ci := C \Di, C0 := C1 ∩ C2.

2.1. Review of the physical problem. My claim, based on the physical analysis of Costello and Yamazakiin [CY19], is that from this initial data we can construct two a priori unrelated doohickeys:

• A two-dimensional classical Lagrangian field theory, specifically a type of σ-model with target M.In particular, our starting data allows us to construct a metric and closed 3-form on M.

• For each field σ in the σ-model, a connection D(σ) on spacetime.

Then the physical analysis in [CY19] concludes, and the main theorem of this paper proves, that a field σ isa solution to the classical equations of motion for the σ-model if and only if D(σ) is a flat connection.

My approach to the problem will be different to that of Costello and Yamazaki. For the sake of history andmotivation, however, let us briefly review their argument:

Costello and Yamazaki begin with the 4d Chern-Simons theory introduced in [Cos13]. This is the four-dimensional theory with action

SCS [A] =1

2π~

∫R2×C

ω ∧ CS(A),(2.1)

where A is a gauge field and CS(A) = tr(A ∧ dA + 23A ∧ A ∧ A). Equipping R2 with a complex coordinate

w, w we may write

A = Awdw +Awdw +Az

Page 5: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

LAX FORMULATION FOR HARMONIC MAPS TO A MODULI OF BUNDLES 5

i.e. we assume that A has no (1, 0)-component in the C-direction. The equations of motion for the theoryare precisely the flatness equations F (A) = 0.

The upshot of [CY19] is that coupling the action (2.1) to various surface defects and compactifying on thecurve C leads to a wide variety of two-dimensional classical field theories which are automatically integrabledue to features inherited from the 4d theory. ‘Integrable’ in this situation means ‘has a Lax/zero curvatureformulation’; hence in this framework the main theorem of this paper comes for free once one knows that the2d theory of interest may be engineered from 4d Chern-Simons theory.

Remark 2.1. In the situation where C = CP1 is the Riemann sphere, this procedure recovers the integrableσ-models originally constructed by different means in [DLMV19b, DLMV19a].

To obtain the theory of interest in the Costello-Yamazaki framework, one begins by imposing certain boundaryconditions on the gauge fields and gauge transformations in order to account for the poles and zeroes of ω:

• At each simple zero of ω, either Aw or Aw has a simple pole.• At each pole of ω, A vanishes.• We only allow gauge transformations that vanish at the poles of ω.

Assume for simplicity that we are studying connections on the trivial principal bundle. Then we can considerthe component Az(w, w) as an R2-family of holomorphic structures on the bundle via the family of operators∂+Az(w, w); this shows us that one of the fields in our compactified theory will be a map R2 → BunG(C,Q).For a large class of such maps—namely, those that land in M—one can further solve uniquely for the fieldsAw and Aw. Thus if we impose this restriction on the target space, we see that our 2d compactification is aσ-model; the Lagrangian of this theory can be determined by inserting the expressions for Aw, Aw in termsof Az back into the original Lagrangian.

In this paper I will show that the formulae for the metric, 3-form, and connections described by Costello andYamazaki are well-defined, and by direct calculation will verify that the conclusion about the zero-curvatureformulation of the classical equations of motion holds.

3. The metric and its properties

3.1. Definition of the metric. First, let us define the metric. I will begin by giving the “cleanest” definitionof the metric as a manifestly algebraic pairing on cohomology groups, before mentioning two cochain modelsthat can be used for computations.

Consider the short exact sequence of sheaves on C,

0→ gP (−Q)→ gP (D1)→ gP ⊗OP1(P1)→ 0

which, by our cohomology vanishing assumption, implies the isomorphism

H1(C; gP (−Q)) ∼= H0(C; gP ⊗OP1(P1)) ∼=

⊕x∈P1

H0(C; gP ⊗Ox(x))

I will describe a pairing that is block diagonal with respect to this direct sum decomposition as follows. Let〈−,−〉 denote a nondegenerate invariant pairing on g. Then if we are working over the field F, we define anF-valued pairing on each block as the following composition:

H0(C; gP ⊗Ox(x))⊗2 H0(C;Ox(x))⊗2 H0(C;Ox(2x)) H0(C;Kx(x)) F〈 〉 ·ω Resx

Note that this is just the value at P ∈ M of an algebraic pairing that can be defined on certain sheaveson C ×M pushed forward along the projection to M. It is symmetric because 〈−,−〉 is symmetric andmultiplication is symmetric in a commutative ring, and nondegeneracy follows from nondegeneracy of thepairing 〈−,−〉. Hence this pairing is an algebraic metric on M.

Page 6: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

6 RICHARD DERRYBERRY

3.1.1. Dolbeault model for the metric. I will now describe the Dolbeault model for the metric, which will beused in later calculations. Consider the diagram

0 Ω0,0C (gP (−Q)) Ω0,0

C (gP (Di))

0 Ω0,1C (gP (−Q)) Ω0,1

C (gP (Di))

ρi

∂ ∂

ρi

∂−1i

Let 〈−,−〉 denote the same nondegenerate invariant pairing as before. We define the metric at the pointP ∈M by

A1, A2 7→ gP (A1, A2) :=

∫C

ω ∧ 〈∂−1

1 ρ1(A1)⊗A2 + ∂−1

1 ρ1(A2)⊗A1〉, A1, A2 ∈ Ω0,1C (gP (−Q)).(3.1)

Remark 3.1. For the rest of the paper I will leave implicit the embeddings ρi.

Proposition 3.1. g defines a metric on M.

In order to prove this proposition, let me recall that ∂−1

i,P (i = 1, 2) can be expressed in terms of an integralkernel, the Szego kernel SP [CY19, Def. 15.2].

Definition 3.1 (Szego kernel). The Szego kernel for P ∈ M is the unique element SP ∈ H0(C × C; gP ⊗gP ⊗O(D1×C +C ×D2 + ∆C)) such that the residue of SP along the diagonal is the quadratic Casimir forg.

Proof. In order to prove that g defines a metric on M we need to show the following:

(1) gP is smooth in P .(2) gP is gauge invariant, i.e. it vanishes if one of the Ai is a coboundary (hence the expression descends

to a pairing on cohomology).(3) The resulting pairing on cohomology is nondegenerate.

First, the Szego kernel varies algebraically in P [BZB03, Prop. 5.6], and so the expression ∂−1

i,PA variessmoothly with P ∈M. Hence we have that the expression for the metric varies smoothly on M.

For gauge invariance, observe that for φ ∈ Ω0,0C (gP (−Q)) and A ∈ Ω0,1

C (gP (−Q)) we have that ω〈∂−1

1 A, φ〉 is

a smooth function (the poles of ∂−1

1 A and φ cancel with the zeroes of ω, and vice versa). Hence,

gP (A, ∂φ) =

∫C

ω ∧ 〈∂−1

1 A, ∂φ〉+

∫C

ω ∧ 〈A, φ〉 =

∫C

∂(ω〈∂−1

1 A, φ〉)

=

∫∂C=∅

ω〈∂−1

1 A, φ〉 = 0.

Finally, for nondegeneracy we will compute the metric in a convenient basis for the tangent space. SetP1 = p1 + · · · pn, with all pi distinct. Let Di be a coordinate disc around pi with coordinate zi chosen suchthat ω|Di = zidzi, and let ta be a basis of g. Let δ|zi|=ε be the distributional (0, 1)-form defined by∫

g(z, z)dz ∧ δ|z|=ε =

∮|z|=ε

g(z, z)dz

(or a radially symmetric smooth mollification of said form). Trivialise gP on the discs Di, and define

Aia :=

taziδ|zi|=ε on Di,

0 on C \ Di(3.2)

Note that ∂−1

1 Aia = taziδ|zi|≤ε (or, again, an appropriately mollified smooth approximation of the step func-

tion) so that we have

gP (Aia, Ajb) =

∫C

ω ∧ 〈 taziδ|zi|≤ε,

tbzjδ|zj |=ε〉+

∫C

ω ∧ 〈 tbzjδ|zj |≤ε,

taziδ|zi|=ε〉

= δij

∮|zi|=ε

dzizi〈ta, tb〉 = 2πiδijκab

Page 7: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

LAX FORMULATION FOR HARMONIC MAPS TO A MODULI OF BUNDLES 7

where κab = 〈ta, tb〉. Thus, nondegeneracy of the metric follows from nondegeneracy of the pairing 〈−,−〉.

3.1.2. Cech model for the metric. Let us now also consider the following algebraic Cech model of the metric.Consider the Cech complex calculating the derived pushforward over an affine open U ⊂M, with respect tothe open cover C \D1,

∐Di of C:

0 C0(ad(P)(−Q)) C0(ad(P)(Di))⊕

x∈P1ad(P|x×U )⊗ TxC 0

0 C1(ad(P)(−Q)) C1(ad(P)(Di)) 0

ρi

δ δ

ρi

δ−1i

On this open affine, define the metric by the formula

g(A1, A2) =∑x∈D1

Resx(ω〈δ−1

1 ρ1(A1), δ−11 ρ2(A2)〉

).

We would like to show that:

(1) This expression makes sense.(2) This expression vanishes if one of the Ai is a coboundary (hence the expression descends to a pairing

on cohomology).(3) The resulting pairing on cohomology is nondegenerate.

First, the expression δ−11 (A1)⊗ δ−1

1 (A2) is an element of⊕

x∈P1ad(P|x×U )⊗2 ⊗ TxC⊗2. Fix a point x ∈ C.

Then in taking global sections of the exact sequence

0→ KC(−2x)→ KC(−x)→ Ox ⊗ T ∗xC⊗2 → 0

one sees that ω ∈ Γ(OC(−D1 −D2)) ⊂ Γ(OC(−D1)) determines an element ωP1∈⊕

x∈P1T ∗xC

⊗2. Since ω

has strictly first order zeroes at all the points of P1, ωP1is nonvanishing. Thus it trivialises TxC

⊗2, and canbe contracted with δ−1

1 (A1)⊗ δ−11 (A2) to obtain an element of

⊕x∈P1

ad(P|x×U )⊗2. g(A1, A2) is then given

by applying 〈−,−〉 and summing over the points of P1 to obtain a function on U .

This description of the metric makes clear that once again nondegeneracy on cohomology amounts to thenondegeneracy of the invariant pairing 〈−,−〉. It therefore remains to show that the expression vanishes ifone of the Ai is coboundary. But by exactness of the top row in the above diagram, if Ai is coboundary inC1(ad(P)(−Q)) it projects to zero in

⊕x∈P1

ad(P|x×U )⊗ TxC.

3.2. Properties of the metric. Let us now take a moment to explore some of the properties of thismetric. In particular, I am interested in obtaining an expression for the first derivative of the metric in localcoordinates, and an understanding of the (pseudo-)Riemannian metrics induced on real slices of M.

3.2.1. First derivative of the metric. Let Aia be a collection of gP (−Q)-valued (0, 1)-forms which descendto give a basis of H1(C; gP (−Q)). (For concreteness, the reader may wish to consider the basis given by

(3.2).) Then for small enough ~λ = (λia) the expression

∂P+~λ := ∂P + λiaAia

defines a gauge-inequivalent ∂-operator on the smooth bundle underlying P , hence (λia) provides a systemof coordinates in a neighbourhood of P .

Now, we can express ∂−1

P+~λ as a geometric series

∂−1

P+~λ = (∂P + λia[Aia,−])−1 = (1 + λia∂−1

P [Aia,−])−1∂−1

P

= ∂−1

P +∑N≥1

(−1)Nλi1a1 · · ·λiNaN∂−1

P [Ai1a1 ,−]∂−1

P · · · [AiNaN ,−]∂−1

P

Page 8: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

8 RICHARD DERRYBERRY

so that the metric components are locally of the form

gia,jb(~λ) = gP+~λ

(∂

∂λia,∂

∂λjb

)=

∫C

ω ∧ 〈∂−1

P+~λ,1Aia, Ajb〉+

∫C

ω ∧ 〈∂−1

P+~λ,1Ajb, Aia〉

= gP (Aia, Ajb)

− λkc∫C

ω ∧ 〈∂−1

P,1[Akc, ∂−1

P,1Aia], Ajb〉

− λkc∫C

ω ∧ 〈∂−1

P,1[Akc, ∂−1

P,1Ajb], Aia〉+O(λ2).

So we have:

Proposition 3.2. The derivative of g at P is given in local coordinates by

∂gia,jb∂λkc

∣∣∣∣~λ=0

= −∫C

ω ∧ 〈∂−1

P,1[Akc, ∂−1

P,1Aia], Ajb〉 −∫C

ω ∧ 〈∂−1

P,1[Akc, ∂−1

P,1Ajb], Aia〉.

3.2.2. The metric on real slices. The physical system from which this project takes inspiration must of coursehave as its target space some real pseudo-Riemannian manifold, necessitating the choice of a real slice of ourmoduli spaceM. With this in mind, let us explore the induced pseudo-Riemannian metric on slices inducedby a real structure on the curve C and real form of G.

Let ρC : C → C and ρG : G→ G be antiholomorphic involutions, where ρG is also a homomorphism defininga real form of the group. We further require that ρ∗C(ω) = ω. This gives a real structure ρ on M defined bytaking a bundle P with transition functions (ϕij) on some open cover to the bundle with transition functions(ρG ϕij ρC). Fixed points of this action are bundles with transition functions (ϕij) such that there exists

a cochain (si) such that siϕijs−1j = ρG ϕij ρC .

Take the real points M(R) with respect to this structure, and define a pseudo-Riemannian metric by takinggR = 1

2πig. We wish to understand the possible signatures of gR.

We work in a basis modelled on (3.2). Assume that ta is a basis for the real form defined by ρG. Then

ρ∗CAia =ta

ρC(zi)δ|zi|=ε.

Now, the coordinates zi are such that ω = zidzi, and ρ∗C(ω) = ρC(zi)dρC(zi) = zidzi. This restricts the formof ρC in these coordinates to be ρC(zi) = λizi, where λi = ±1. If λi = +1 then Aia is a tangent vector toM(R); if λi = −1 then iAia is a tangent vector toM(R). Together with the fact that 1

2πig(Aia, Ajb) = δijκabwe have shown the following:

Proposition 3.3. Let Σ(κ) denote the signature of the form 〈−,−〉 restricted to the real form of g definedby ρG. Then the signature of gR is (λ1Σ(κ), . . . , λg−1Σ(κ)).

The following example demonstrates that the sequence (λ1, . . . , λg−1) can in fact take any value in ±1g−1.

Example 3.1. For this example, we make use of the fact that a Riemann surface equipped with a nonzeroholomorphic 1-form is equivalent to a translation surface. An accessible and interesting introduction totranslation surfaces is given by [Wri15]. For the purposes of this paper, however, we simply need to knowthe following facts about the surface C pictured in Figure 1:

• The shaded region, considered as sitting in the complex plane, corresponds to the surface.• The surface is obtained by identifying opposite edges of the rectangle and opposite edges of each

hexagon.• If there are g − 1 hexagons, the resulting surface is of genus g.• The surface C is equipped with the holomorphic 1-form ω induced by the 1-form dz on C.• The corners of the hexagons correspond to simple zeros of ω.• The surface C inherits a real structure ρ from complex conjugation on C.

Page 9: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

LAX FORMULATION FOR HARMONIC MAPS TO A MODULI OF BUNDLES 9

Figure 1. A genus g translation surface with complex conjugation symmetry. Note thereare g − 1 hexagons.

In order to define our metric, we need to choose g−1 of the zeroes of ω, assumed to lie on the real axis, to bethe divisor D1. Let us suppose that from each hexagon we choose either the leftmost corner or the rightmostcorner. We wish to know how ρ acts on coordinate w centred at one of these corners.

Near one of these corners, the local coordinate satisfies wdw = dz, i.e. w is proportional to a square-root ofz. To make sense of this, we need to choose a branch cut for the square-root. Let z = reiθ.

Suppose that we choose the rightmost corner, so that a local model for the corner looks like Figure 2. Thenwe take our branch cut along the negative real axis, so that

w = 2√reiθ/2, −π < θ < π.

Then ρ acts on w by taking θ 7→ −θ, so that ρ(w) = w. Hence in this case, λi = +1.

Figure 2. Taking the branch cut for the square-root along the negative real axis.

Now suppose instead that we choose the leftmost corner, so that a local model for the corner looks like Figure3. Then we take our branch cut along the positive real axis, so that

w = 2√reiθ/2, 0 < θ < 2π.

Now ρ acts on w by taking θ 7→ 2π − θ, so that ρ(w) = −w. Hence in this case, λi = −1.

Thus, by choosing left/rightmost points of the hexagons appropriately, any sequence of ±1 of length g − 1may be achieved.

Page 10: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

10 RICHARD DERRYBERRY

Figure 3. Taking the branch cut for the square-root along the positive real axis.

4. Definition of the 3-form

Next, let us define the 3-form that will appear in the action of our σ-model. I will define this expression onthe Dolbeault complex and then show that it is gauge invariant.

Recall that there is a totally antisymmetric invariant 3-tensor 〈[−,−],−〉 ∈(∧3

g∗)G

. We define the 3-form

Ω on M by the formula

A1, A2, A3 7→∑σ∈S3

(−1)σ∫C

ω ∧ 〈[Aσ(1), ∂−1

1 (Aσ(2))], ∂−1

2 (Aσ(3))〉, A1, A2, A3 ∈ Ω0,1C (gP (−Q)).(4.1)

Proposition 4.1. Equation (4.1) defines a closed 3-form on M.

Proof. Smoothness follows as for the metric by the expression for ∂−1

P in terms of the Szego kernel. Next, wewould like to see that (4.1) vanishes if one of the Ai is a coboundary. Consider the expression

ω〈[∂−1

1 (A1), ∂−1

1 (A2)], ∂−1

2 (A3)〉.

Set A1 = ∂φ, A2 = A,A3 = B. Then the expression ω〈[φ, ∂−1

1 A], ∂−1

2 B〉 is holomorphic, and so we have

0 =

∫C

∂(ω〈[φ, ∂−1

1 A], ∂−1

2 B〉)

= −∫C

ω ∧ 〈[∂φ, ∂−1

1 A], ∂−1

2 B〉 −∫C

ω ∧ 〈[φ,A], ∂−1

2 B〉 −∫C

ω ∧ 〈[φ, ∂−1

1 A], B〉

so that (antisymmetrising the A↔ φ terms)∫C

ω ∧ 〈[∂φ, ∂−1

1 A], ∂−1

2 B〉 −∫C

ω ∧ 〈[A, φ], ∂−1

2 B〉 = −∫C

ω ∧ 〈[φ, ∂−1

1 A], B〉.

Antisymmetrising this expression in the A↔ B terms gives∫C

ω ∧ 〈[∂φ, ∂−1

1 A], ∂−1

2 B〉 −∫C

ω ∧ [A, φ], ∂−1

2 B〉 −∫C

ω ∧ 〈[∂φ, ∂−1

1 B], ∂−1

2 A〉+

∫C

ω ∧ [B,φ], ∂−1

2 A〉

=

∫C

ω ∧ 〈[φ, ∂−1

1 B], A〉 −∫C

ω ∧ 〈[φ, ∂−1

1 A], B〉

=

∫C

ω ∧ 〈[B, ∂−1

1 A], φ〉 −∫C

ω ∧ 〈[A, ∂−1

1 B], φ〉

Page 11: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

LAX FORMULATION FOR HARMONIC MAPS TO A MODULI OF BUNDLES 11

which is precisely the statement that the full antisymmetrisation of the original expression vanishes. Thusthe expression is gauge invariant, and so descends to give a well-defined 3-form on M.

Next, we would like to show that this 3-form is closed. Given a point P ∈M, identify a small neighbourhoodU of P in M with a small neighbourhood of 0 ∈ H1(C; gP (−Q)) by sending the harmonic representatives4

A ∈ Ω0,1(C; gP (−Q)) for elements A ∈ H1(C; gP (−Q)) to the G-bundles with ∂-operator ∂A = ∂P + A.Over U , via this isomorphism, the tangent bundle is trivialised TU ∼= U ×H1(C; gP (−Q)) and the 3-formsends α1, α2, α3 ∈ H1(C; gP (−Q)) to the function on U given by the antisymmetrization of∫

C

ω ∧ 〈[α1, ∂−1

A,1(α2)], ∂−1

A,2(α3)〉

To calculate the exterior derivative of this function, we wish to antisymmetrize the terms which are α4-linearin the expression∫C

ω ∧ 〈[α1, ∂−1

A+α4,1α2]〉, ∂−1

A+α4,2α3〉 −

∫C

ω ∧ 〈[α1, ∂−1

A,1α2], ∂−1

A,2α3〉

= −∫C

ω ∧ 〈[α1, ∂

−1

A,1([α4, ∂−1

A,1α2])], ∂−1

A,2α3〉 −∫C

ω ∧ 〈[α1, ∂−1

A,1α2], ∂−1

A,2([α4, ∂−1

A,2α3])〉+ higher order terms

=

∫C

ω ∧ 〈∂−1

A,1([α4, ∂−1

A,1α2]), [α1, ∂−1

A,2α3]〉 −∫C

ω ∧ 〈∂−1

A,2([α4, ∂−1

A,2α3]), [α1, ∂−1

A,1α2]〉+ h.o.t.

where we have used adjointness of ∂−1

1 and −∂−1

2 (Lemma 4.2 below).

ω is a holomorphic section of KC(−D1 −D2), so we have that

∂(ω ⊗ 〈∂−1

A,1([α4, ∂−1

A,1α2]), ∂A,2([α1, ∂−1

A,2α3])〉)

= −ω ∧ 〈[α4, ∂−1

A,1α2], ∂A,2([α1, ∂−1

A,2α3])〉 − ω ∧ 〈∂−1

A,1([α4, ∂−1

A,1α2]), [α1, ∂−1

A,2α3]〉

= −ω ∧ 〈∂A,2([α1, ∂−1

A,2α3]), [α4, ∂−1

A,1α2]〉 − ω ∧ 〈∂−1

A,1([α4, ∂−1

A,1α2]), [α1, ∂−1

A,2α3]〉

So the term linear in α4 becomes∫C

ω ∧ 〈∂−1

A,1([α4, ∂−1

A,1α2]), [α1, ∂−1

A,2α3]〉+

∫C

ω ∧ 〈∂−1

A,1([α1, ∂−1

A,1α2]), [α4, ∂−1

A,2α3]〉

which is symmetric under the exchange of 1 and 4. Since the symmetric group S4 can be written as a disjointunion S4 = C

∐C · (14) this expression vanishes upon antisymmetrising. Hence, the 3-form is closed.

Lemma 4.2. For A,B ∈ Ω0,1C (gP (−Q)),

∫Cω ∧ 〈∂−1

1 A,B〉 = −∫Cω ∧ 〈A, ∂−1

2 B〉.

Proof. Observe that ω〈∂−1

1 A, ∂−1

2 B〉 is holomorphic. So

0 = −∫C

∂(ω〈∂−1

1 A, ∂−1

2 B〉)

=

∫C

ω ∧ 〈A, ∂−1

2 B〉+

∫C

ω ∧ 〈∂−1

1 A,B〉.

Remark 4.1. As for the metric, the above analysis of the 3-form can also be performed in the Cech model.

5. The action and its variation

Let us now consider the action of the two-dimensional sigma model built from the metric (3.1) and 3-form(4.1), namely5

S[σ] =

∫D2

‖dσ‖2dvolD2 +1

3

∫D2×R≥0

σ∗(Ω)(5.1)

4Or any other choice of basis, as per our calculation of the derivative of the metric.5The factor of 1

3could be absorbed into the definition of the 3-form; leaving it explicit here will simplify some formulae later.

Page 12: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

12 RICHARD DERRYBERRY

where D2 is a two-dimensional disc and σ is an extension of σ to D2 × R≥0.6

Call the first term of (5.1) Sg and the second term SΩ. We will examine the variation of each of these terms.Let η be a constant metric on the disc D2 (we will eventually take this to be the metric dt1dt2).

Since our spacetime is the disc we can work in coordinates (λia) on the target space – all of our equationswill be derived using these coordinates.

Proposition 5.1. Let Σ(t1, t2, τ) be a family of maps from D2 to M with Σ(t1, t2, 0) = σ. Then

d

∣∣∣∣τ=0

Sg[Σ] =

∫D2

∂Σkc

∂τ

∣∣∣∣τ=0

ηαβ(−2σ∗(gia,kc)

∂2σia

∂tα∂tβ− σ∗

(∂gkc,jb∂λia

+∂gia,kc∂λjb

− ∂gia,jb∂λkc

)∂σia

∂tα

∂σjb

∂tβ

)dt1∧dt2.

Proof. A standard exercise in the calculus of variations.

Proposition 5.2. With notation as in Proposition 5.1, the variation of SΩ is

d

∣∣∣∣τ=0

SΩ[Σ] =

∫D2

∂Σkc

∂τ

∣∣∣∣τ=0

σ∗(Ωkc,ia,jb)εαβ ∂σ

ia

∂tα

∂σjb

∂tβdt1 ∧ dt2.

Proof. Let Σ be an extension of the family of maps Σ from Proposition 5.1 to D2 × R≥0. Since Ω is closed,∫∂(D2×R≥0×[0,T ])

Σ∗(Ω) = 0 =

∫D2×R≥0

Σ∗T (Ω)−∫D2×R≥0

σ∗(Ω)−∫D2×[0,T ]

Σ∗Ω.

So, ∫D2×R≥0

Σ∗T (Ω)−∫D2×R≥0

σ∗(Ω) =

∫D2×[0,T ]

Σ∗(Ωia,jb,kc)dΣia ∧ dΣjb ∧ dΣkc

= 3

∫D2×[0,T ]

∂Σkc

∂τΣ∗(Ωkc,ia,jb)dDΣia ∧ dDΣjb ∧ dτ

where dD is the de Rham differential on D2. The result now follows by taking ddT

∣∣T=0

and dividing by 3.

Putting these two variational formulae together, we obtain:

Corollary 5.3. The equations of motion for the action S[σ] are

σ∗(gia,kc)ηαβ ∂2σia

∂tα∂tβ+

1

2σ∗(∂gkc,jb∂λia

+∂gia,kc∂λjb

− ∂gia,jb∂λkc

)ηαβ

∂σia

∂tα

∂σjb

∂tβ− 1

2σ∗(Ωia,jb,kc)ε

αβ ∂σia

∂tα

∂σjb

∂tβ= 0,

or in expanded form for the metric dt1dt2,

2σ∗(gia,kc)∂2σia

∂t1∂t2+

1

2σ∗(∂gkc,jb∂λia

+∂gia,kc∂λjb

− ∂gia,jb∂λkc

)(∂σia

∂t1

∂σjb

∂t2+∂σia

∂t2

∂σjb

∂t1

)− 1

2σ∗(Ωia,jb,kc)

(∂σia

∂t1

∂σjb

∂t2− ∂σia

∂t2

∂σjb

∂t1

)= 0.

6. Connections on the target space

In this section I will show that there are two natural families of algebraic connections on the target spaceM, parametrised by C0, and each related to one of the divisors D1 or D2.

6Since Ω is closed, the classical equations of motion are insensitive to the choice of extension.

Page 13: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

LAX FORMULATION FOR HARMONIC MAPS TO A MODULI OF BUNDLES 13

6.1. What is a connection? Let πM : C ×M→M be the projection. Then by the defining condition ofM,

RπM∗(ad(P)⊗ π∗CO(Di)) = 0,

where πC : C ×M→ C is the projection to C.

Consider ∆M :M→M×M, the diagonal map. This embedding corresponds to an ideal sheaf I onM×M,and we can consider the quasicoherent sheaf of algebras on M

0→ Ω1M → p1∗

(OM×M/I2

)︸ ︷︷ ︸J1(OM)

→ OM → 0

where p1 :M×M→M is the first projection (and similarly for p2). Note that J1(OM) = p1∗p∗2(OM).

Set ∆(1) := SpecM×M(OM×M/I2) ⊂M×M. Then a connection on an object7 B →M is an isomorphism

φ : (p∗1B)|∆(1) ' (p∗2B)|∆(1) which restricts to the identity on M.

6.2. The connection of interest. Let U = Spec(R) ⊂M be an affine patch and consider trying to definea connection on P|C0×U relative to C0. Start by taking

P|C×U → C × Uand consider a square-zero extension

0→ J → R′ → R→ 0, U ′ := Spec(R′).

Equivalence classes of lifts of P|C×U together with its trivialisation on Q to C × U ′ are parametrised by

R1πU∗(ad(P|C×U )(−Q))⊗R J.Set supp(Di) = xi,1, . . . , xi,n and let Di.j be the formal disc around the point xi,j (similarly D×i,j for the

punctured formal disc). Consider the cover of C given by

Ui =

Ci,n⋃j=1

Di,j

.

The above cohomology group can be calculated using the Cech complex for this cover:

C0(Ui; ad(P)(−Q)) C1(Ui; ad(P)(−Q))

Γ(Ci; ad(P)(−Q))⊕⊕n

1 Γ(Di,j ; ad(P))⊕n

1 Γ(D×i,j ; ad(P))

δ

Consider the exact sequence on C

0→ OC(−Q)→ OC(Di)→ ODi(Di)︸ ︷︷ ︸⊕g−1j=1 Txi,jC

→ 0

which yields the exact sequences on C × U0→ ad(P|C×U )(−Q)→ ad(P|C×U )⊗ π∗UOC(Di)→ ad(P|C×U )⊗ π∗UODi(Di)︸ ︷︷ ︸⊕g−1

1 ad(P|xi,j×U )⊗Txi,jC

→ 0

and so a sequence of Cech complexes with exact rows

0 C0(Ui; ad(P|C×U )(−Q)) C0(Ui; ad(P|C×U )⊗ π∗UOC(Di))⊕n

1 ad(P|xi,j×U )⊗ Txi,jC 0

0 C1(Ui; ad(P|C×U )(−Q)) C1(Ui; ad(P|C×U )⊗ π∗UOC(Di)) 0

δ ∼

7This is deliberately a bit vague: one can consider connections on objects in any Grothendieck fibration over our category ofspaces, see [nLa21]. We will not need this generality, however—blessedly, we restrict ourselves to vector bundles.

Page 14: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

14 RICHARD DERRYBERRY

The dashed arrow gives a map that takes a lift of ad(P|C×U )(−Q) to C × U ′ to a trivialisation of that liftwith first-order poles on Di.

Given two lifts, P, P ′, taking their difference under the dashed arrow gives a unique isomorphism

P|Ci×U ' P ′|Ci×Uwith first-order poles on Di.

So given the two pulllbacks of ad(P|C0×U )(−Q) to the square-zero neighbourhood, we obtain a uniqueisomorphism between them with first-order poles on Di, call this φi,U .

The above diagram is compatible with restricting the affine open V ⊂ U , and the corresponding maps φi,Vare unique; thus the collection φi,UU⊂M glues to give a connection φi on ad(P|C0×M)(−Q) relative toC0 ×M→ C0.

Denote the connection associated to the divisor D1 by ∇+ and the connection associated to the divisor D2 by∇−. The diagram above also tells us how to derive the connection 1-form in local coordinates (λia) centredat P ∈M: writing D1 = d+α = d+αiadλ

ia and D2 = d+ β = d+ βiadλia, the connection components are

precisely given by the singular gauge trivialisations of the basis elements, i.e.

αia(~λ) = ∂−1

P+~λ,1Aia, βia(~λ)= ∂−1

P+~λ,2Aia.(6.1)

7. The induced connection and flatness equation

Now, given the connections ∇± and a field σ : D2 →M, we wish to define an induced connection D(σ) on D2

as follows. Recall that we have coordinates t1, t2 on D2, which are the null directions for our metric dt1dt2,and set ∂α = ∂

∂tα.

Definition 7.1. The induced connection D(σ) is defined by the equations

D(σ)∂1 := (σ∗∇+)∂1 , D(σ)∂2 := (σ∗∇−)∂2 .(7.1)

In terms of the connection forms on the target we can write these equations as

D(σ)∂1 :=∂

∂t1+

[σ∗α

(∂

∂t1

),−], D(σ)∂2 :=

∂t2+

[σ∗β

(∂

∂t2

),−].

so that in local coordinates we can write

D(σ) = d+ σ∗α

(∂

∂t1

)dt1 + σ∗β

(∂

∂t2

)dt2.

Proposition 7.1. D(σ) is flat if and only if

σ∗(βia − αia)∂2σia

∂t1∂t2+ σ∗

(∂βjb∂λia

− ∂αia∂λjb

+ [αia, βjb]

)∂σia

∂t1

∂σjb

∂t2= 0.(7.2)

Proof. D(σ) has curvature

F (σ)

dt1 ∧ dt2=

∂t1

(σ∗β

(∂

∂t2

))− ∂

∂t2

(σ∗α

(∂

∂t1

))+

[σ∗α

(∂

∂t1

), σ∗β

(∂

∂t2

)].

In local coordinates (λia) near P ∈M we write

α = αiadλia, β= βiadλ

ia, σ = (σia(t1, t2)),(7.3)

so that

σ∗α = αia(σ(t1, t2))dσia = σ∗(αia)∂σia

∂tαdtα,

Page 15: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

LAX FORMULATION FOR HARMONIC MAPS TO A MODULI OF BUNDLES 15

and similarly for σ∗β. Then

∂t1

(σ∗β

(∂

∂t2

))=

∂t1

(βia(σ(t1, t2))

∂σia

∂t2

)=∂βia∂σjb

∂σjb

∂t1

∂σia

∂t2+ βia(σ(t1, t2))

∂2σia

∂t1∂t2

= σ∗(∂βia∂λjb

)∂σjb

∂t1

∂σia

∂t2+ σ∗(βia)

∂2σia

∂t1∂t2

and similarly for α, so that

F (σ)

dt1 ∧ dt2=

∂t1

(σ∗β

(∂

∂t2

))− ∂

∂t2

(σ∗α

(∂

∂t1

))+

[σ∗α

(∂

∂t1

), σ∗β

(∂

∂t2

)]= σ∗

(∂βia∂λjb

)∂σjb

∂t1

∂σia

∂t2+ σ∗βia

∂2σia

∂t1∂t2

− σ∗(∂αia∂λjb

)∂σjb

∂t2

∂σia

∂t1− σ∗αia

∂2σia

∂t1∂t2+

[σ∗αia

∂σia

∂t1, σ∗βjb

∂σjb

∂t2

]= σ∗(βia − αia)

∂2σia

∂t1∂t2+ σ∗

(∂βjb∂λia

− ∂αia∂λjb

+ [αia, βjb]

)∂σia

∂t1

∂σjb

∂t2.

I am soon going to want to compare the flatness equation to the equations of motion; for this reason, it isuseful to “re-symmetrise” the flatness equation to place it in a form that more closely resembles the desiredresult. Namely, we take the expression (7.2) for F (σ) and rewrite it as

σ∗(βia − αia)∂2σia

∂t1∂t2+ σ∗

(∂βjb∂λia

− ∂αia∂λjb

+ [αia, βjb]

)∂σia

∂t1

∂σjb

∂t2

= σ∗(βia − αia)∂2σia

∂t1∂t2

+ σ∗(∂βjb∂λia

− ∂αia∂λjb

+ [αia, βjb]

)(∂σia

∂t1

∂σjb

∂t2+

1

2

∂σjb

∂t1

∂σia

∂t2− 1

2

∂σjb

∂t1

∂σia

∂t2

)= σ∗(βia − αia)

∂2σia

∂t1∂t2

+1

2σ∗(∂βjb∂λia

− ∂αia∂λjb

+ [αia, βjb]

)(∂σia

∂t1

∂σjb

∂t2+∂σjb

∂t1

∂σia

∂t2

)+

1

2σ∗(∂βjb∂λia

− ∂αia∂λjb

+ [αia, βjb]

)(∂σia

∂t1

∂σjb

∂t2− ∂σjb

∂t1

∂σia

∂t2

)= σ∗(βia − αia)

∂2σia

∂t1∂t2

+1

4σ∗(∂βjb∂λia

− ∂αia∂λjb

+ [αia, βjb] +∂βia∂λjb

− ∂αjb∂λia

+ [αjb, βia]

)(∂σia

∂t1

∂σjb

∂t2+∂σjb

∂t1

∂σia

∂t2

)+

1

4σ∗(∂βjb∂λia

− ∂αia∂λjb

+ [αia, βjb]−∂βia∂λjb

+∂αjb∂λia

− [αjb, βia]

)(∂σia

∂t1

∂σjb

∂t2− ∂σjb

∂t1

∂σia

∂t2

)

8. Flatness and the equations of motion

Given a field σ : D2 →M, there are two objects we can now consider:

• The action (5.1) and corresponding equations of motion (Cor. 5.3).• The induced connection on spacetime, D(σ) (Def. 7.1).

Page 16: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

16 RICHARD DERRYBERRY

The following is the main theorem of this paper:

Theorem 8.1. The field σ satisfies the equations of motion for S[σ] if and only if the connection D(σ) isflat.

Proof. We proceed by comparing the coefficients of ∂2σia

∂t1∂t2, ∂σ

ia

∂t1∂σjb

∂t2+ ∂σjb

∂t1∂σia

∂t2, and ∂σia

∂t1∂σjb

∂t2− ∂σjb

∂t1∂σia

∂t2in

the equations of motion and flatness equations. We see that we wish to compare

Equations of motion Flatness

2gia,kc βia − αia(8.1)

1

2

(∂gkc,jb∂λia

+∂gia,kc∂λjb

− ∂gia,jb∂λkc

)1

4

(∂βjb∂λia

− ∂αia∂λjb

+ [αia, βjb] +∂βia∂λjb

− ∂αjb∂λia

+ [αjb, βia]

)(8.2)

−1

2Ωia,jb,kc

1

4

(∂βjb∂λia

− ∂αia∂λjb

+ [αia, βjb]−∂βia∂λjb

+∂αjb∂λia

− [αjb, βia]

)(8.3)

Note that, by the same calculations as we performed for the derivative of the metric,

∂αia∂λjb

∣∣∣∣~λ=0

= −∂−1

P,1[Ajb, ∂−1

P,1Aia],∂βjb∂λia

∣∣∣∣~λ=0

= −∂−1

P,2[Aia, ∂−1

P,2Ajb]

Since the equations of motion and flatness equations have well-defined coordinate-free descriptions, to com-pare these equations it is enough to compare them at every point P ∈M in coordinates centred at P . Thusfrom now on I will omit the

∣∣~λ=0

symbol from my calculations – all terms are to be implicitly evaluated atthe origin of the coordinate system.

Notice that a priori there are many (infinitely!) more “flatness equations”, since the equations of motion areindexed by (ia, kc) while the flatness equations are indexed by (ia, z ∈ C). To obtain a matching number ofequations, we take the flatness equations and pair them with each basis vector Akc, i.e. we take∫

C

ω ∧ 〈−, Akc〉.

Doing this for (8.1) gives∫C

ω ∧ 〈βia − αia, Akc〉 =

∫C

ω ∧ 〈∂−1

2 Aia, Akc〉 −∫C

ω ∧ 〈∂−1

1 Aia, Akc〉

= −∫C

ω ∧ 〈∂−1

1 Aia, Akc〉 −∫C

ω ∧ 〈Aia, ∂−1

1 Akc〉

= −gia,kc = −1

2(2gia,kc) .

Next, let’s compare the terms in (8.2). Writing the equations of motion explicitly gives

1

2

(∂gkc,jb∂λia

+∂gia,kc∂λjb

− ∂gia,jb∂λkc

)=

1

2

∫C

ω ∧ 〈[Ajb, ∂−1

1 Aia], ∂−1

2 Akc〉︸ ︷︷ ︸(I)

+1

2

∫C

ω ∧ 〈[Ajb, ∂−1

1 Akc], ∂−1

2 Aia〉︸ ︷︷ ︸(II)

+1

2

∫C

ω ∧ 〈[Aia, ∂−1

1 Ajb], ∂−1

2 Akc〉︸ ︷︷ ︸(III)

+1

2

∫C

ω ∧ 〈[Aia, ∂−1

1 Akc], ∂−1

2 Ajb〉︸ ︷︷ ︸(IV )

− 1

2

∫C

ω ∧ 〈[Akc, ∂−1

1 Aia], ∂−1

2 Ajb〉︸ ︷︷ ︸(V )

− 1

2

∫C

ω ∧ 〈[Akc, ∂−1

1 Ajb], ∂−1

2 Aia〉︸ ︷︷ ︸(V I)

,

Page 17: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

LAX FORMULATION FOR HARMONIC MAPS TO A MODULI OF BUNDLES 17

while the corresponding term for the flatness equations gives∫C

ω ∧ 〈14

(∂βjb∂λia

− ∂αia∂λjb

+ [αia, βjb] +∂βia∂λjb

− ∂αjb∂λia

+ [αjb, βia]

), Akc〉

=1

4

∫C

ω ∧⟨−∂−1

2 [Aia, ∂−1

2 Ajb] + ∂−1

1 [Ajb, ∂−1

1 Aia] + [∂−1

1 Aia, ∂−1

2 Ajb]

−∂−1

2 [Ajb, ∂−1

2 Aia] + ∂−1

1 [Aia, ∂−1

1 Ajb] + [∂−1

1 Ajb, ∂−1

2 Aia], Akc

⟩= −1

4

∫C

ω ∧ 〈[Aia, ∂−1

1 Akc], ∂−1

2 Ajb〉︸ ︷︷ ︸(IV )

−1

4

∫C

ω ∧ 〈[Ajb, ∂−1

1 Aia], ∂−1

2 Akc〉︸ ︷︷ ︸(I)

+1

4

∫C

ω ∧ 〈[Akc, ∂−1

1 Aia], ∂−1

2 Ajb〉︸ ︷︷ ︸(V )

−1

4

∫C

ω ∧ 〈[Ajb, ∂−1

1 Akc], ∂−1

2 Aia〉︸ ︷︷ ︸(II)

− 1

4

∫C

ω ∧ 〈[Aia, ∂−1

1 Ajb], ∂−1

2 Akc〉︸ ︷︷ ︸(III)

+1

4

∫C

ω ∧ 〈[Akc, ∂−1

1 Ajb], ∂−1

2 Aia〉︸ ︷︷ ︸(V I)

= −1

2

[1

2

(∂gkc,jb∂λia

+∂gia,kc∂λjb

− ∂gia,jb∂λkc

)]where we have freely made use of the adjointness of ∂

−1

1 and −∂−1

2 , and the underbraced indices indicatehow to match the flatness and equation of motion terms.

Finally, consider the flatness expression in (8.3):∫C

ω ∧ 〈14

(∂βjb∂λia

− ∂αia∂λjb

+ [αia, βjb]−∂βia∂λjb

+∂αjb∂λia

− [αjb, βia]

), Akc〉

=1

4

∫C

ω ∧⟨−∂−1

2 [Aia, ∂−1

2 Ajb] + ∂−1

1 [Ajb, ∂−1

1 Aia] + [∂−1

1 Aia, ∂−1

2 Ajb]

+∂−1

2 [Ajb, ∂−1

2 Aia]− ∂−1

1 [Aia, ∂−1

1 Ajb]− [∂−1

1 Ajb, ∂−1

2 Aia], Akc

⟩=

1

4

∫C

ω ∧ 〈[Aia, ∂2]−1Ajb], ∂−1

1 Akc〉︸ ︷︷ ︸(23)

−1

4

∫C

ω ∧ 〈[Ajb, ∂−1

1 Aia], ∂−1

2 Akc〉︸ ︷︷ ︸(12)

+1

4

∫C

ω ∧ 〈[∂−1

1 Aia, ∂−1

2 Ajb], Akc〉︸ ︷︷ ︸(123)

−1

4

∫C

ω ∧ 〈[Ajb, ∂−1

2 Aia], ∂−1

1 Akc〉︸ ︷︷ ︸(132)

+1

4

∫C

ω ∧ 〈[Aia, ∂−1

1 Ajb], ∂−1

2 Akc〉︸ ︷︷ ︸(1)

−1

4

∫C

ω ∧ 〈[∂−1

1 Ajb, ∂−1

2 Aia], Akc〉︸ ︷︷ ︸(13)

=1

4

∑s∈S3

(−1)s∫C

ω ∧ 〈[As(ia), ∂−1

1 As(jb)], ∂−1

2 As(kc)〉

= −1

2

(−1

2Ωia,jb,kc

)where the underbraced incides now indicate which term corresponds to which element of the symmetric groupS3. But now observe that we have ∫

C

ω ∧ 〈Flatness, Akc〉 = −1

2EOMkc

for all basis elements Akc, completing the proof of the theorem.

Page 18: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

18 RICHARD DERRYBERRY

9. Example: CP1 with two marked points

Let us now consider an example where we can write things out fairly explicitly: the case of CP1 with twomarked points.

Let C = CP1 equipped with coordinate z and 1-form

ω =(z − p1)(z − p2)

z2dz, p1 6= p2, pi 6= 0.(9.1)

Note that ω has second order poles at 0 and ∞, and simple zeroes at p1 and p2. Let Q denote the divisor0 +∞. We wish to consider BunG(CP1, Q), the moduli of G-bundles on CP1 with a trivialisation on Q.

Assume that G is simple simply-connected, and consider the cohomology vanishing condition

H0(CP1, gP (D1)) = 0,

where we choose D1 = p1 − 0 −∞. In this example, the cohomology vanishing condition implies that thebundle P is trivializable.8 So, fix a trivialized bundle with trivialisations at 0 and ∞, P = CP1 ×G. We canalways act on the bundle to simultaneously change the trivialisations at 0 and ∞, so fix the trivialisation at∞ to agree with the trivialisation of the bundle. Our remaining degree of freedom is the trivialisation of thebundle at 0, and so we see that M ∼= G. As a check on this, observe that we have an agreement of tangentspaces

TPM = H1(C; g⊗O(−0−∞)) ∼= g⊗H1(C;O(−2)) ∼= g.

Note that in this example we can explicitly determine the form of the Szego kernel. For the trivial G-bundleon CP1 this will be a g⊗ g-valued meromorphic function S(z, z′) on CP1 × CP1 with

• simple pole along z = z′ with residue the Casimir element c ∈ g⊗ g,• simple poles at z = p1 and z′ = p2, and• simple zeroes at z = 0,∞, z′ = 0,∞.

Given these conditions, it is easy to check that the kernel is

S(z, z′) =zz′

(z − z′)(z − p1)(z′ − p2)c(9.2)

Now, we want to calculate the metric and 3-form onM. SinceM has a transitive G×G-symmetry, given byleft and right G-multiplication on the trivialisation at 0, and the expressions for the metric and 3-form areindependent of the trivialisation at ∞, it suffices to calculate at the basepoint ∗ of M given by the trivialtrivialisation at ∞; i.e. we are looking at multiples of the usual bi-invariant metric and 3-form on G.

Let ta be a basis for g, and take as representatives for a basis of T∗M the forms

Aa = (p1 − p2)z

(z − p1)(z − p2)δ|z−p1|=εta

These correspond to the basis ta under the identification ofM withG, which we can see in the following way:a tangent vector toM is represented by a form A ∈ Ω0,1(CP1;O(−0−∞))⊗ g, and if it is nonzero it has noantiderivative in Ω0,0(CP1;O(−0−∞))⊗g. It does however have an antiderivative χ ∈ Ω0,0(CP1;O(−∞))⊗g—i.e. we allow the antiderivative to be nonzero at 0—and then χ(0) ∈ g is precisely the infinitesimal variationof the framing at 0.

Observing that

(p1 − p2)z

(z − p1)(z − p2)=

p1

z − p1− p2

z − p2

we have that Aa = ∂χa for

χa = −(

p1

z − p1δ|z−p1|≥ε +

p2

z − p2δ|z−p1|≤ε

)ta

8To see this, suppose E = O(i1)⊕ · · ·O(in) with i1 + · · ·+ in = 0. Then E ⊗O(−1) has vanishing H0 if and only if all ofthe ik < 1, which forces i1 = · · · = in = 0.

Page 19: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

LAX FORMULATION FOR HARMONIC MAPS TO A MODULI OF BUNDLES 19

and so χa(0) = ta.

Note that these also have singular antiderivatives

∂−1

1 Aa = (p1 − p2)z

(z − p1)(z − p2)δ|z−p1|≤εta, ∂

−1

2 Aa = −(p1 − p2)z

(z − p1)(z − p2)δ|z−p1|≥εta,

so that

gab =

∫CP1

ω ∧ 〈∂−1

1 Aa, Ab〉+

∫CP1

ω ∧ 〈∂−1

1 Ab, Aa〉 = κab

∮|z−p1|=ε

dz

z − p1

(p1 − p2)2

z − p2= 2πi(p1 − p2)κab

and∫CP1

ω ∧ 〈[Aa, ∂−1

1 Ab], ∂−1

2 Ac〉 =1

3

∮|z−p1|=ε

dz

(z − p1)2

z

(z − p2)2〈[ta, tb], tc〉(p1 − p2)3

=2πi

3κcdf

dab

d

dz

(z

(z − p2)2

)∣∣∣∣z=p1

· (p1 − p2)3 = −2πi

3κcdf

dab(p1 + p2).

Since this expression is already antisymmetric it is equal to Ωabc. This matches the expression for Ω obtainedby a different derivation in [CY19, §10.2.3].

Remark 9.1. This example shows that the main theorem of this paper can be considered a vast generalisationof the well-known result that the harmonic map equation for surfaces mapping to Lie groups has a Lax pairformulation [Poh76]. It would be interesting to see whether this can be leveraged into an understanding ofthe moduli space of solutions to the classical equations of motion, as was done for harmonic maps to Liegroups in [Uhl89].

9.1. The 1-loop β-function. Finally, I will give a sketch of some evidence for Conjecture 1. Recall thatthis conjecture of Costello claims that the one-loop β-function may be identified with a vector field inducedby a flow on the moduli space of Riemann surfaces equipped with a holomorphic 1-form ω of the type wehave been considering, together with a decomposition of the zeroes of ω into two equally sized groups, D1

and D2.

In our situation, the constraints on the flow are that:

• P1 =∮|z|=ε ω is fixed, and

• P2 =∫ p2p1ω varies to first order in the flow parameter ε as P2 → P2 + ε.

We can achieve this by flowing the points p1, p2 linearly,

p1 → p1 + εc and p2 → p2 − εc(9.3)

and making a judicious choice of constant c.

It is straightforward to calculate that the periods are given by

P1 = −2πi(p1 + p2),(9.4)

P2 = 2(p2 − p1) +P1

2πilog

(p2

p1

)mod P1.(9.5)

Under the flow (9.3) we have that

P2 → P2 + εc(p1 − p2)2

p1p2+O(ε2),

so our constraints are satisfied if we take c = p1p2(p1−p2)2 . Similarly, under this flow the metric coefficients vary

as

gab → gab + 4πiεcκab =

(1 + ε

2p1p2

(p1 − p2)3

)gab.

Page 20: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

20 RICHARD DERRYBERRY

By the definition of the one-loop β-function, we would therefore like to show that

βab?=

2p1p2

(p1 − p2)3gab.

According to [CFMP85], the one-loop β-function for the metric is

1

α′βµν = Ricµν −

1

4HµλκH

λκν ,(9.6)

where H is proportional to Ω, H = cΩ. I will pin down the specific value of c shortly.

The Ricci curvature is invariant under constant scalings of the metric, and so Ricab = − 14κab. We therefore

have that

− 4

α′βab = κab + c2gcegdfΩacdΩbef

= κab +1

(2πi)2(p1 − p2)2

(2πi)2(p1 + p2)2c2

9κceκdfκdgf

gacκfhf

hbe︸ ︷︷ ︸

−κab

=

(1− c2

9

(p1 + p2)2

(p1 − p2)2

)κab

To determine c we observe that limp1→0p2→∞

ω

p1 + p2is the 1-form that gives rise to the WZW model [CY19], and

that the β-function of that theory vanishes. The result of rescaling the 1-form is a constant rescaling of themetric and 3-form. The Ricci tensor is invariant under constant rescalings, as is the expression gcegdfΩacdΩbef(the rescaling of the 3-forms cancels with the rescaling of the inverse metrics), so the β-function is unchangedby rescalings and we may simply pass to the p1 → 0, p2 →∞ limit

1− c2

9

(p1 + p2)2

(p1 − p2)2→ 1− c2

9= 0

i.e. c = 3. We then have

− 4

α′βab =

(1− (p1 + p2)2

(p1 − p2)2

)κab =

(p1 − p2)2 − (p1 + p2)2

(p1 − p2)2κab = − 4p1p2

(p1 − p2)2κab

so that

βab = α′p1p2

(p1 − p2)2κab =

α′

2πi

p1p2

(p1 − p2)3gab

which gives the desired result upon setting α′ = 4πi.

Remark 9.2. An alternative proof of the conjecture in this example – in fact, a proof of the conjecture for anarbitrary number of double poles on the Riemann sphere9 – can be deduced from the results of [DLSS21], inwhich the 1-loop RG flow of the models introduced in [DLMV19b, DLMV19a] is completely determined.

References

[Arn89] V. I. Arnol’d. Mathematical methods of classical mechanics, volume 60 of Graduate Texts in Mathematics. Springer-

Verlag, New York, second edition, 1989. Translated from the Russian by K. Vogtmann and A. Weinstein.[BBT03] Olivier Babelon, Denis Bernard, and Michel Talon. Introduction to classical integrable systems. Cambridge Mono-

graphs on Mathematical Physics. Cambridge University Press, Cambridge, 2003.[Byk16] Dmitri Bykov. Complex structures and zero-curvature equations for σ-models. Physics Letters B, 760:341–344,

2016.

[BZB03] David Ben-Zvi and Indranil Biswas. Theta functions and Szego kernels. Int. Math. Res. Not., (24):1305–1340, 2003.

[CFMP85] C. G. Callan, D. Friedan, E. J. Martinec, and M. J. Perry. Strings in background fields. Nuclear Phys. B, 262(4):593–609, 1985.

[Cos13] Kevin Costello. Supersymmetric gauge theory and the yangian, 2013.[CY19] Kevin Costello and Masahito Yamazaki. Gauge Theory And Integrability, III. 8 2019.

9Many thanks to Benoit Vicedo for this reference and observation.

Page 21: M C M arXiv:2106.09781v1 [math.AG] 17 Jun 2021

LAX FORMULATION FOR HARMONIC MAPS TO A MODULI OF BUNDLES 21

[DLMV19a] F. Delduc, S. Lacroix, M. Magro, and B. Vicedo. Assembling integrable σ-models as affine Gaudin models. J. HighEnergy Phys., (6):017, 86, 2019.

[DLMV19b] F. Delduc, S. Lacroix, M. Magro, and B. Vicedo. Integrable coupled σ models. Phys. Rev. Lett., 122:041601, Jan

2019.[DLSS21] Francois Delduc, Sylvain Lacroix, Konstantinos Sfetsos, and Konstantinos Siampos. RG flows of integrable σ-models

and the twist function. J. High Energy Phys., (2):Paper No. 065, 43, 2021.

[FT07] Ludwig D. Faddeev and Leon A. Takhtajan. Hamiltonian methods in the theory of solitons. Classics in Mathematics.Springer, Berlin, english edition, 2007. Translated from the 1986 Russian original by Alexey G. Reyman.

[LV21] Sylvain Lacroix and Benoıt Vicedo. Integrable E-models, 4d Chern-Simons theory and affine Gaudin models. I.

Lagrangian aspects. SIGMA Symmetry Integrability Geom. Methods Appl., 17:Paper No. 058, 45, 2021.[nLa21] nLab authors. Grothendieck connection. http://ncatlab.org/nlab/show/Grothendieck%20connection, April 2021.

[Poh76] K. Pohlmeyer. Integrable Hamiltonian systems and interactions through quadratic constraints. Comm. Math. Phys.,

46(3):207–221, 1976.[Uhl89] Karen Uhlenbeck. Harmonic maps into Lie groups: classical solutions of the chiral model. J. Differential Geom.,

30(1):1–50, 1989.[Wri15] Alex Wright. Translation surfaces and their orbit closures: an introduction for a broad audience. EMS Surv. Math.

Sci., 2(1):63–108, 2015.

[Zar19] K. Zarembo. Integrability in sigma-models. In Integrability: from statistical systems to gauge theory, pages 205–247.Oxford Univ. Press, Oxford, 2019.