betaFeedback

WorksMathematics

Splitting and Mixed Deformations of Toric Generalized Kähler Potentials

Abstract

We study splitting phenomena for extremal toric generalized Kähler structures on products of compact toric manifolds. We prove that every extremal symplectic potential splits when the skew-symmetric parameter BB is block diagonal, without any restriction on AA. We also construct two types of non-splitting examples on CP1×CP1\mathbb{CP}^1\times\mathbb{CP}^1: an explicit constant-generalized-scalar-curvature structure with a splitting potential but a non-product generalized Kähler structure, and a deformation family of constant-generalized-scalar-curvature structures with genuinely non-splitting potentials.

Introduction

The product question studied here concerns canonical geometric structures. In the Kähler geometry, canonical Kähler metrics exhibit remarkable rigidity with respect to product decompositions. Apostolov and the first named author [1] proved a splitting result in extremal Kähler metrics on products of polarized projective manifolds, under certain assumptions involving integral Futaki invariants or automorphism groups. In the toric setting, the first named author [10] removed these additional assumptions and proved that every extremal Kähler metric on a product of compact toric manifolds splits as a product of extremal Kähler metrics on the factors, using the short-time existence of the Calabi flow. These results naturally motivate the investigation of analogous splitting phenomena in generalized Kähler geometry. In the present paper, we establish a splitting theorem for the symplectic potentials of extremal toric generalized Kähler structures when the skew-symmetric parameter BB respects the product decomposition. We also construct two types of examples on CP1×CP1\mathbb{CP}^1 \times\mathbb{CP}^1 illustrating distinct failures of splitting: the first consists of extremal generalized Kähler structures whose symplectic potentials split but whose full generalized Kähler structures do not, while the second consists of extremal generalized Kähler structures whose symplectic potentials themselves fail to split.

Generalized Kähler geometry

Generalized Kähler geometry originated in theoretical physics through the work of Gates, Hull, and Roček [3], who discovered the bi-Hermitian geometry underlying two-dimensional supersymmetric nonlinear sigma models. Their work showed that the target spaces of sigma models with N=(2,2)N = (2, 2) supersymmetry admit a more general geometric structure than ordinary Kähler geometry, involving two integrable complex structures compatible with a common Riemannian metric and a torsion three-form. A new geometric perspective emerged with Hitchin’s introduction of generalized complex geometry [8], which unifies complex and symplectic geometry through structures on the generalized tangent bundle TM⊕T∗MTM \oplus T^*M. Building on Hitchin’s framework, Gualtieri [5, 6] introduced generalized Kähler structures and established their equivalence with the bi-Hermitian geometry arising in the earlier physics literature.

Let MM be a smooth manifold of real dimension 2n2n. The generalized tangent bundle is the direct sum

TM:=TM⊕T∗M,TM := TM \oplus T^*M,

equipped with the natural symmetric bilinear form of signature (2n,2n)(2n,2n), defined by

⟨X+ξ,Y+η⟩=12(ξ(Y)+η(X)),\langle X+\xi,Y+\eta\rangle= \frac{1}{2}\bigl(\xi(Y)+\eta(X)\bigr),

for X,Y∈TMX,Y \in TM and ξ,η∈T∗M\xi,\eta\in T^{*}M. The space of smooth sections of TMTM is endowed with the Courant bracket

[X+ξ,Y+η]C=[X,Y]+LXη−LYξ−12d(ιXη−ιYξ).[X+\xi,Y+\eta]_{C}=[X,Y]+\mathcal{L}_{X}\eta-\mathcal{L}_{Y}\xi-\frac{1}{2}d(\iota_{X}\eta-\iota_{Y}\xi).

A generalized almost complex structure is an endomorphism J:TM→TM\mathcal{J}:TM\to TM satisfying

J2=−Id⁡,⟨Ju,Jv⟩=⟨u,v⟩\mathcal{J}^{2}=-\operatorname{Id},\qquad\langle\mathcal{J}u,\mathcal{J}v\rangle=\langle u,v\rangle

for all u,v∈TMu,v\in TM. Equivalently, J\mathcal{J} is an orthogonal complex structure on the generalized tangent bundle. It is said to be integrable if its +i+i-eigenbundle

LJ=ker⁡(J−iId⁡)⊂TM⊗CL_{\mathcal{J}}=\ker(\mathcal{J}-i\operatorname{Id})\subset TM\otimes\mathbb{C}

is involutive under the complexified Courant bracket, that is,

[Γ(LJ),Γ(LJ)]C⊆Γ(LJ).[\Gamma(L_{\mathcal{J}}),\Gamma(L_{\mathcal{J}})]_{C}\subseteq\Gamma(L_{\mathcal{J}}).

An integrable generalized almost complex structure is called a generalized complex structure.

A generalized Kähler structure is a pair (J1,J2)(\mathcal{J}_{1},\mathcal{J}_{2}) of commuting integrable generalized complex structures,

[J1,J2]=0,[\mathcal{J}_{1},\mathcal{J}_{2}]=0,

such that

G:=−J1J2\mathcal{G}:=-\mathcal{J}_{1}\mathcal{J}_{2}

defines a positive generalized metric. More precisely, G\mathcal{G} is an orthogonal, self-adjoint involution satisfying

G2=Id⁡,⟨Gu,v⟩=⟨u,Gv⟩,\mathcal{G}^{2}=\operatorname{Id},\qquad\langle\mathcal{G}u,v\rangle=\langle u,\mathcal{G}v\rangle,

and the bilinear form

(u,v)G:=⟨Gu,v⟩(u,v)_{\mathcal{G}}:=\langle\mathcal{G}u,v\rangle

is positive definite. Consequently, the generalized tangent bundle admits an orthogonal decomposition

TM=V+⊕V−,TM=V_{+}\oplus V_{-},

where V±V_{\pm} are the ±1\pm1-eigenbundles of G\mathcal{G}. The natural pairing is positive definite on V+V_{+} and negative definite on V−V_{-}.

A generalized metric is equivalently determined by a Riemannian metric gg and a real two-form β\beta. In the chosen splitting of the Courant algebroid, its eigenbundles are given by

V±={X+(β±g)(X):X∈TM},V_{\pm}=\{X+(\beta\pm g)(X):X\in TM\},

where gg and β\beta are regarded as bundle maps from TMTM to T∗MT^{*}M. Thus, the positivity condition on G\mathcal{G} recovers an ordinary Riemannian metric on MM, while the commuting generalized complex structures encode additional complex-geometric data.

Gualtieri’s bi-Hermitian correspondence [5, 6] establishes the equivalence between generalized Kähler structures and the bi-Hermitian geometry originally arising in supersymmetric sigma models. More precisely, upon choosing an isotropic splitting of the underlying exact Courant algebroid, a generalized Kähler structure can be described by a quadruple (g,β,I,J)(g,\beta,I,J), where gg is a Riemannian metric, β\beta is a real two-form, and (I,J)(I,J) are two integrable complex structures compatible with gg, namely,

g(IX,IY)=g(JX,JY)=g(X,Y)g(IX,IY)=g(JX,JY)=g(X,Y)

for all X,Y∈TMX,Y\in TM. The corresponding fundamental two-forms are defined by

ωI(X,Y)=g(IX,Y),ωJ(X,Y)=g(JX,Y).\omega_{I}(X,Y)=g(IX,Y),\qquad\omega_{J}(X,Y)=g(JX,Y).

In the untwisted setting, the integrability conditions for the generalized Kähler structure are equivalent to the torsion identities

−dIcωI=dβ=dJcωJ.-d^{c}_{I}\omega_{I}=d\beta=d^{c}_{J}\omega_{J}.

In particular, the two Hermitian structures (g,I)(g,I) and (g,J)(g,J) have opposite Bismut torsion three-forms, and their common torsion is closed:

ddIcωI=ddJcωJ=0.dd^c_I\omega_I=dd^c_J\omega_J=0.

Thus, both Hermitian structures are pluriclosed (also called strong Kähler with torsion, or SKT). The ordinary Kähler case is recovered when I=JI=J and β=0\beta=0. We adopt the conventions for dcd^c and the identification of differential forms with bundle maps used in [2]. Thus, generalized Kähler geometry extends the classical Kähler framework by incorporating two complex structures, torsion, and holomorphic Poisson geometry, providing a natural setting in which to investigate canonical metrics and their rigidity properties.

Of particular interest is the class of generalized Kähler structures of symplectic type, which admits a description in terms of a symplectic form taming a complex structure; see [5, 6, 2]. More precisely, fixing a complex manifold (M,J)(M,J), let FF be a closed symplectic form satisfying F(X,JX)>0F(X,JX)>0 for every nonzero tangent vector XX. Such a form determines a Riemannian metric gg, a real two-form β\beta, and a second almost complex structure II by

−FJ=g+β,I=−F−1J∗F.-FJ=g+\beta,\qquad I=-F^{-1}J^*F.

If II is integrable, these data define a generalized Kähler structure of symplectic type [2]. In particular, for a fixed complex structure JJ, the entire bi-Hermitian structure is encoded by FF, subject to the additional integrability condition on II.

An important feature of generalized Kähler geometry is its natural connection with Poisson geometry. As observed by Hitchin [9], the commutator of the two complex structures determines a real Poisson bivector

π=12[I,J]g−1.\pi=\frac{1}{2}[I,J]g^{-1}.

whose (2,0)(2,0)-components with respect to II and JJ define holomorphic Poisson structures. In particular, the holomorphic Poisson tensor associated with II and JJ are

πI=π−−1Iπ,πJ=π−−1Jπ.\pi_I=\pi-\sqrt{-1}I\pi,\qquad\pi_J=\pi-\sqrt{-1}J\pi.

This additional Poisson structure plays a central role in the study of generalized Kähler classes and canonical generalized Kähler metrics [2].

Generalized Kähler classes

In classical Kähler geometry, a central role is played by the space of Kähler metrics in a fixed Kähler class α∈H1,1(M,R)\alpha\in H^{1,1}(M,\mathbb{R}). By the ∂∂ˉ\partial\bar\partial-lemma, fixing a reference Kähler form ω0∈α\omega_0\in\alpha, every other Kähler form in the same class can be written as

ωφ=ω0+ddJcφ.\omega_\varphi=\omega_0+dd^c_J\varphi.

where φ∈C∞(M,R)\varphi\in C^\infty(M,\mathbb{R}) is unique up to an additive constant and satisfies ωφ>0\omega_\varphi>0. Thus, a fixed Kähler class admits a global parametrization by an open convex space of smooth potentials. This provides the natural framework for the study of canonical Kähler metrics, including constant scalar curvature and extremal Kähler metrics.

The notion of a generalized Kähler class has emerged through the study of deformations of generalized Kähler structures, particularly the Hamiltonian deformation theory developed by Gualtieri [7] and the deformation classes introduced by Gibson and Streets [4]. Building on these developments, Apostolov, Streets, and Ustinovsky [2] formulated a notion of generalized Kähler class for structures of symplectic type with fixed complex structure JJ, holomorphic Poisson tensor πJ\pi_J, and de Rham cohomology class α\alpha. They further established that the smooth path-connected components of the corresponding space of generalized Kähler structures can be characterized by Hamiltonian potential deformations, providing a natural framework for extending the Calabi program to generalized Kähler geometry.

More precisely, fix a compact complex manifold (M,J)(M,J), a holomorphic Poisson tensor πJ\pi_J, and a cohomology class α∈H2(M,R)\alpha\in H^2(M,\mathbb{R}). Denote by

GKπ,α={F∣(F,J) is a generalized Ka¨hler structure of symplectic type, [F]=α, πJ(F)=πJ}\mathcal{GK}_{\pi,\alpha}=\{F\mid(F,J)\text{ is a generalized Kähler structure of symplectic type},\ [F]=\alpha,\ \pi_J(F)=\pi_J\}

the space of generalized Kähler structures realizing these fixed data. Given a reference structure F0∈GKπ,αF_0\in\mathcal{GK}_{\pi,\alpha}, the smooth path-connected component

GKπ,α0(F0)\mathcal{GK}_{\pi,\alpha}^{0}(F_0)

containing F0F_0 is called a generalized Kähler class [2].

If we fix the complex structure JJ, the holomorphic Poisson tensor πJ\pi_J, and the cohomology class α\alpha, infinitesimal variations within a generalized Kähler class can be represented by

F˙=ddIcφ.\dot{F}=dd_I^c\varphi.

where φ∈C∞(M,R)\varphi\in C^\infty(M,\mathbb{R}) and II is the second complex structure associated with FF. Unlike the ordinary Kähler case, however, the complex structure II generally varies along the deformation. Consequently, the corresponding finite deformation cannot, in general, be expressed simply as F=F0+ddI0cφF=F_0+dd_{I_0}^c\varphi for a single potential.

Instead, following the Hamiltonian deformation approach developed by Gualtieri [7] and the formulation of Apostolov–Streets–Ustinovskiy [2], one considers a smooth family of real-valued functions φt\varphi_t and the associated evolution equations

dFtdt=ddItcφt,It=(Φt)∗I0(Φt)−1.\frac{dF_t}{dt}=dd_{I_t}^c\varphi_t,\qquad I_t=(\Phi_t)_*I_0(\Phi_t)^{-1}.

where Φt\Phi_t is the flow generated by the time-dependent vector field

Xt=−π(dφt).X_t=-\pi(d\varphi_t).

Here π\pi denotes the fixed real Poisson tensor determined by πJ\pi_J. Integrating the first equation gives

Ft=F0+∫0tddIscφs ds.F_t=F_0+\int_0^t dd_{I_s}^c\varphi_s\,ds.

Since each variation is exact, the cohomology class remains unchanged:

[Ft]=[F0]=α.[F_t]=[F_0]=\alpha.

Moreover, the deformation preserves the holomorphic Poisson tensor πJ\pi_J. Provided the taming condition remains satisfied, the family FtF_t therefore stays within the prescribed generalized Kähler class.

Apostolov–Streets–Ustinovskiy [2] establishes that every generalized Kähler structure in the smooth path-connected component of F0F_0 can be obtained through such a generalized Kähler potential deformation. In this sense, smooth real-valued functions play a role analogous to ordinary Kähler potentials, even though the deformation is nonlinear and involves the simultaneous evolution of FtF_t and ItI_t. This provides a geometric characterization of generalized Kähler classes and makes them natural analogues of the classical spaces of Kähler potentials.

When the fixed holomorphic Poisson tensor vanishes, the real Poisson tensor π\pi also vanishes, so the Hamiltonian vector field XtX_t is identically zero. Consequently, Φt\Phi_t is the identity and It=I0=JI_t=I_0=J. The deformation equations reduce to

dFtdt=ddJcφt,\frac{dF_t}{dt}=dd_J^c\varphi_t,

and hence

Ft=F0+ddJc(∫0tφs ds).F_t=F_0+dd_J^c\left(\int_0^t\varphi_s\,ds\right).

Thus, the generalized Kähler class reduces to the ordinary Kähler class,

GK0,α=Kα.\boxed{\mathcal{GK}_{0,\alpha}=\mathcal{K}_\alpha}.

In particular, generalized Kähler classes provide a natural extension of the classical framework for studying canonical metrics within fixed geometric data. Their potential-theoretic description also underlies the generalized Mabuchi functional and the variational formulation of the generalized Calabi problem developed in [2].

Toric generalized Kähler classes and the two skew-symmetric parameters

Let (M2n,T)(M^{2n},T) be a compact toric manifold, where TT is an nn-dimensional real torus with Lie algebra t\mathfrak{t}, and let P⊂t∗≅RnP \subset\mathfrak{t}^{*} \cong\mathbb{R}^{n} be its Delzant moment polytope. We fix a collection of affine functions L={Lr}r=1d\mathcal{L}=\{L_{r}\}_{r=1}^{d} defining the facets of PP, so that

P={x∈t∗:Lr(x)≥0,r=1,…,d}.P=\{x\in\mathfrak{t}^{*}:L_{r}(x)\ge0,\quad r=1,\ldots,d\}.

The primitive inward normals to these facets determine the integral lattice of the torus. Choosing an integral basis, we identify T≅Rn/(2πZ)nT\cong\mathbb{R}^{n}/(2\pi\mathbb{Z})^{n}. On the dense open orbit, we use momentum–angle coordinates

(x,θ)=(x1,…,xn,θ1,…,θn)∈P∘×T.(x,\theta)=(x_{1},\ldots,x_{n},\theta_{1},\ldots,\theta_{n})\in P^{\circ}\times T.

The Abreu–Guillemin description of toric generalized Kähler structures of symplectic type, developed in [2] following earlier work on toric generalized Kähler geometry, associates to such a structure a triple

(u,A,B),(u,A,B),

where uu is a strictly convex symplectic potential on P∘P^{\circ}, defined up to an affine-linear function, and

A,B∈Λ2tA,B\in\Lambda^{2}\mathfrak{t}

are constant real skew-symmetric matrices. The ordinary toric Kähler geometry is recovered when A=B=0A=B=0. More generally, the potential uu determines an associated ordinary toric Kähler structure, called the toric Kähler reduction, while AA and BB encode two additional generalized Kähler deformations.

To describe these data explicitly, write

G=Hess⁡u=(uij).G=\operatorname{Hess}u=(u_{ij}).

In momentum–angle coordinates, the ordinary toric Kähler reduction is determined by the symplectic form

ω=∑i=1ndxi∧dθi\omega=\sum_{i=1}^{n}dx_{i}\wedge d\theta_{i}

and the Riemannian metric

gu=∑i,j=1nuij dxi dxj+∑i,j=1nuij dθi dθj,g_{u}=\sum_{i,j=1}^{n}u_{ij}\,dx_{i}\,dx_{j}+\sum_{i,j=1}^{n}u^{ij}\,d\theta_{i}\,d\theta_{j},

where (uij)=G−1(u^{ij})=G^{-1}. In particular, the corresponding Hessian metric on the interior of the moment polytope is G=Hess⁡uG=\operatorname{Hess}u.

The generalized Kähler structure is obtained by introducing the constant skew-symmetric matrices AA and BB. In the conventions of [2], the complex structure JJ and the symplectic form FF are expressed as

J=∑i,j=1n[(G+A)ij dxi⊗∂∂θj−(G+A)ij−1 dθi⊗∂∂xj].(1)J=\sum_{i,j=1}^{n}\left[(G+A)_{ij}\,dx_{i}\otimes\frac{\partial}{\partial\theta_{j}}-(G+A)^{-1}_{ij}\,d\theta_{i}\otimes\frac{\partial}{\partial x_{j}}\right]. \tag*{(1)}
F=∑i=1ndxi∧dθi+∑i,j=1nBij dxi∧dxj.(2)F=\sum_{i=1}^{n}dx_{i}\wedge d\theta_{i}+\sum_{i,j=1}^{n}B_{ij}\,dx_{i}\wedge dx_{j}. \tag*{(2)}

The matrix G+AG+A is invertible because its symmetric part GG is positive definite. The skew-symmetric parameter AA appears in the expression for JJ, whereas BB appears in the expression for FF. Notice that the sum in the second term of (2) runs over all ordered pairs (i,j)(i,j), without a factor of 1/2. Thus, when n=2n=2 and

B=(0t−t0),B=\begin{pmatrix} 0 & t \\ -t & 0 \end{pmatrix},

we obtain

∑i,j=12Bij dxi∧dxj=2t dx1∧dx2.\sum_{i,j=1}^{2} B_{ij}\,dx_i\wedge dx_j=2t\,dx_1\wedge dx_2.

This normalization will be important in the explicit examples considered later.

Although AA and BB enter the geometric construction in different ways, both contribute to the associated holomorphic Poisson tensor. More precisely, writing

Zi=∂∂θi−−1J∂∂θi,Z_i=\frac{\partial}{\partial\theta_i}-\sqrt{-1}J\frac{\partial}{\partial\theta_i},

Apostolov–Streets–Ustinovskiy [2] shows

πJ=∑i,j=1n(Aij+−1Bij)Zi∧Zj.\pi_J=\sum_{i,j=1}^{n}(A_{ij}+\sqrt{-1}B_{ij})Z_i\wedge Z_j.

Thus the holomorphic Poisson tensor is encoded by the complex skew-symmetric matrix A+−1BA+\sqrt{-1}B. For a fixed complex toric manifold and torus action, these coefficients uniquely determine the invariant holomorphic Poisson tensor. Consequently, fixing the holomorphic Poisson tensor within a toric generalized Kähler class fixes both AA and BB.

Not every strictly convex symplectic potential is admissible for a prescribed BB. Let

uG(x)=12∑r=1dLr(x)log⁡Lr(x)u_G(x)=\frac{1}{2}\sum_{r=1}^{d}L_r(x)\log L_r(x)

denote the Guillemin reference potential. According to [2], the admissible space SA,B(P,L)S_{A,B}(P,L) consists of potentials satisfying the Guillemin boundary condition

u=uG+v,v∈C∞(P‾),u=u_G+v,\qquad v\in C^\infty(\overline{P}),

together with the positivity condition

Hess⁡u+−1B>0\operatorname{Hess}u+\sqrt{-1}B>0

on P∘P^\circ and the corresponding tangential positivity conditions along every face of PP. Here Hess⁡u+−1B\operatorname{Hess}u+\sqrt{-1}B is Hermitian because GG is real symmetric and BB is real skew-symmetric. The additional positivity requirement reflects the condition that the symplectic form FF tames the complex structure JJ, and ensures that the generalized Kähler structure extends smoothly to the compact toric manifold.

Finally, the generalized scalar curvature admits the Abreu-type expression

Gscal⁡(u,A,B)=−∑i,j=1n∂2∂xi∂xj(Hess⁡u+−1B)ij−1(3)\operatorname{Gscal}(u,A,B)=-\sum_{i,j=1}^{n}\frac{\partial^2}{\partial x_i\partial x_j}\left(\operatorname{Hess}u+\sqrt{-1}B\right)^{-1}_{ij} \tag*{(3)}

In particular, the generalized scalar curvature equation depends on the potential uu and the skew-symmetric matrix BB, but is independent of AA. This distinction plays a central role in our splitting results: when BB respects a product decomposition of PP, the extremal equation forces the symplectic potential to split, even when AA has nonzero mixed components. By contrast, mixed components of BB can lead to extremal generalized Kähler structures with genuinely non-splitting symplectic potentials.

The admissible potentials form a convex set SB(P)S_B(P), independent of AA. The scalar equation is likewise independent of AA. These facts allow a theorem about uu without a block condition on AA; they do not remove AA from the underlying geometry.

Results

Our first result, Theorem 2.4, establishes that every extremal toric generalized Kähler structure on a product toric manifold has a splitting symplectic potential whenever the skew-symmetric parameter BB respects the product decomposition. Notably, no splitting assumption is imposed on the other skew-symmetric parameter AA.

We then construct two types of examples on CP1×CP1\mathbb{CP}^{1} \times\mathbb{CP}^{1} demonstrating distinct failures of splitting. The first is an explicit constant-generalized-scalar-curvature structure with a nonzero mixed component of AA and B=0B = 0. Although its symplectic potential splits, the full generalized Kähler structure does not. The second, established in Theorem 4.2, consists of a one-parameter family of extremal generalized Kähler structures with nonzero mixed components of BB, whose symplectic potentials themselves fail to split. Moreover, Proposition 4.3 provides an exact obstruction to the existence of splitting extremal potentials for nonzero mixed BB.

Together, these results distinguish the splitting of the symplectic potential from that of the full generalized Kähler structure and demonstrate the fundamentally different roles played by the two skew-symmetric parameters AA and BB.

Remark 1.1. Our results are also related to the Yau–Tian–Donaldson type correspondence in toric generalized Kähler geometry. Apostolov, Streets, and Ustinovskiy [2] proved that the existence of an extremal toric generalized Kähler structure implies the existence of an extremal Kähler metric in the same Kähler class. Conversely, they showed that an extremal toric Kähler metric gives rise to extremal generalized Kähler structures for sufficiently small multiples of any prescribed toric holomorphic Poisson tensor. Our splitting theorem complements these results by showing that, when BB is block diagonal, an extremal generalized Kähler structure on a product toric manifold necessarily has a splitting symplectic potential, without any smallness assumption on the Poisson tensor. On the other hand, our examples with mixed BB demonstrate that even an arbitrarily small Poisson deformation can produce extremal generalized Kähler structures with non-splitting symplectic potentials. Thus, while the existence of extremal Kähler and generalized Kähler structures is closely related, their splitting properties can differ substantially.

Acknowledgment

The first-named author would like to thank Vestislav Apostolov for his interest in this work, valuable suggestions, and enlightening discussions.

AI Usage Statement

The authors used AI-assisted tools during the preparation of this manuscript to assist with writing, editing, and improving clarity. The authors have reviewed and verified the content and take full responsibility for the accuracy, originality, and integrity of the paper.

Splitting of extremal potentials for block-diagonal BB

Boundary conditions and conventions

Let P⊂t∗≅RnP \subset\mathfrak{t}^{*} \cong\mathbb{R}^{n} be a compact Delzant polytope defined by affine functions L1,…,LdL_{1}, \ldots, L_{d}, so that

P={x∈Rn:Lr(x)≥0, r=1,…,d}.P = \{x \in\mathbb{R}^{n} : L_{r}(x) \ge0,\ r = 1, \ldots, d\}.

We fix the facet labels and the corresponding Lebesgue measure dxdx on t∗\mathfrak{t}^{*}.

Let B∈Λ2tB \in\Lambda^{2}\mathfrak{t} be a fixed real skew-symmetric matrix. For a symplectic potential uu, set

H=Hess⁡u,KB(u)=H+−1B.H = \operatorname{Hess} u,\qquad K_{B}(u) = H + \sqrt{-1}B.

Since HH is real symmetric and BB is real skew-symmetric, KB(u)K_{B}(u) is Hermitian.

Following [2], we denote by SB(P)\mathcal{S}_{B}(P) the space of admissible symplectic potentials satisfying the following conditions:

  1. u−uG∈C∞(P‾)u-u_{G} \in C^{\infty}(\overline{P});

  2. KB(u)K_{B}(u) is positive definite on P∘P^{\circ};

  3. for every positive-dimensional face Q⊂PQ \subset P, the restriction

(Hess⁡u+−1B)∣TQ\left.(\operatorname{Hess} u+\sqrt{-1}B)\right|_{TQ}

is positive definite on the relative interior Q∘Q^\circ, where the restriction is understood through the tangential Hessian of u∣Qu|_{Q}.

An important feature of these conditions is that they depend on BB, but not on the other skew-symmetric parameter AA. In particular, [2] establishes

SA,B(P,L)=SB(P),\mathcal{S}_{A,B}(P,L)=\mathcal{S}_{B}(P),

and shows that this space is convex. Thus, if u0,u1∈SB(P)u_0,u_1\in\mathcal{S}_{B}(P), then

us=(1−s)u0+su1∈SB(P),0≤s≤1.u_s=(1-s)u_0+su_1\in\mathcal{S}_{B}(P),\qquad0\leq s\leq1.

Define the real symmetric matrix

XB(u)=Re⁡(KB(u)−1).X_B(u)=\operatorname{Re}\left(K_B(u)^{-1}\right).

For later use, we record an equivalent real expression for XB(u)X_B(u):

XB(u)=(H+BH−1B)−1.(4)X_B(u)=\left(H+BH^{-1}B\right)^{-1}. \tag*{(4)}

Indeed, writing KB(u)−1=X+−1YK_B(u)^{-1}=X+\sqrt{-1}Y, with XX symmetric and YY skew-symmetric, the identity

(H+−1B)(X+−1Y)=Id⁡\left(H+\sqrt{-1}B\right)\left(X+\sqrt{-1}Y\right)=\operatorname{Id}

gives

HX−BY=Id⁡,HY+BX=0.HX-BY=\operatorname{Id},\qquad HY+BX=0.

Since HH is invertible, Y=−H−1BXY=-H^{-1}BX, and hence

(H+BH−1B)X=Id⁡,\left(H+BH^{-1}B\right)X=\operatorname{Id},

which proves (4). Notice that BH−1BBH^{-1}B is negative semidefinite, so the positivity of KB(u)K_B(u) imposes a nontrivial restriction on the pair (u,B)(u,B).

We now formulate the extremal equation. Let dσd\sigma denote the boundary measure determined by the facet labels and dxdx. The extremal affine function ℓP∈Aff⁡(P)\ell_P\in\operatorname{Aff}(P) is uniquely characterized by

∫PℓPa dx=2∫∂Pa dσ,for every a∈Aff⁡(P).(5)\int_P \ell_P a\,dx=2\int_{\partial P}a\,d\sigma,\qquad\text{for every }a\in\operatorname{Aff}(P). \tag*{(5)}

Existence and uniqueness follow from the nondegeneracy of the L2(P,dx)L^2(P,dx) inner product on the finite-dimensional space Aff⁡(P)\operatorname{Aff}(P).

The generalized scalar curvature satisfies the corresponding integration-by-parts identity, so its L2L^2-projection onto Aff⁡(P)\operatorname{Aff}(P) is precisely ℓP\ell_P, independently of the choice of admissible potential.

A toric generalized Kähler structure is extremal if and only if its generalized scalar curvature is affine-linear on the moment polytope. Thus, in the present notation, the extremal equation is

SB(u)=ℓP,u∈SB(P).(6)S_B(u)=\ell_P,\qquad u\in\mathcal{S}_B(P). \tag*{(6)}

When ℓP\ell_P is constant, this reduces to the constant generalized scalar curvature equation.

Finally, we emphasize that both the admissible space SB(P)\mathcal{S}_B(P) and the extremal equation (6) are independent of AA. This observation will be important in the splitting argument: the splitting behavior of the symplectic potential is governed by BB, while the parameter AA may still contribute nontrivially to the full generalized Kähler geometry.

Variational facts

Now we define the relative Donaldson-Futaki invariant of (P,L)(P,L):

FP(f)=2∫∂Pf dσ−∫PℓPf dx\mathcal{F}_P(f)=2\int_{\partial P}f\,d\sigma-\int_P\ell_Pf\,dx

and, using any fixed reference u∗∈S0(P)u_*\in\mathcal{S}_0(P), the relative Mabuchi energy:

MB(u)=FP(u)−∫Plog⁡det⁡KB(u) dx+∫Plog⁡det⁡Hess⁡u∗ dx.(7)\mathcal{M}_B(u)=\mathcal{F}_P(u)-\int_P\log\det K_B(u)\,dx+\int_P\log\det\operatorname{Hess}u_*\,dx. \tag*{(7)}

All integrals are finite: the only singularities of the logarithms at facets are integrable logarithms. Reference changes add a constant.

For smooth ff on P‾\overline{P}, the boundary conditions give

∫PSB(u)f dx=2∫∂Pf dσ−∫Ptr⁡(XB(u)Hess⁡f) dx.(8)\int_P S_B(u)f\,dx=2\int_{\partial P}f\,d\sigma-\int_P\operatorname{tr}\left(X_B(u)\operatorname{Hess}f\right)\,dx. \tag*{(8)}

Differentiating (7) and applying (8) yields

DMB(u)[f]=∫P(SB(u)−ℓP)f dx.D\mathcal{M}_B(u)[f]=\int_P\left(S_B(u)-\ell_P\right)f\,dx.

For a segment us=u+sfu_s=u+sf of admissible potentials,

d2ds2MB(us)=∫Ptr⁡[(KB(us)−1/2(Hess⁡f)KB(us)−1/2)2] dx≥0.(9)\frac{d^2}{ds^2}\mathcal{M}_B(u_s)=\int_P\operatorname{tr}\left[\left(K_B(u_s)^{-1/2}(\operatorname{Hess}f)K_B(u_s)^{-1/2}\right)^2\right]\,dx\geq0. \tag*{(9)}

The matrix inside the square is Hermitian; hence the trace of its square is a sum of squares of real eigenvalues. Equality implies Hess⁡f=0\operatorname{Hess}f=0. The boundary estimates implicit in the admissibility conditions justify differentiation and integration.

It follows directly that any solution of (6) globally minimizes MB\mathcal{M}_B on the convex admissible set, and that any two solutions differ by an affine function. This is a uniqueness statement for fixed toric coordinates and fixed BB.

The product averaging argument

In this subsection, we establish our main splitting result for extremal toric generalized Kähler potentials. The proof is based on an averaging procedure over the factors of a product polytope and the convexity properties of the generalized Mabuchi functional.

Let P1⊂Rn1P_1\subset\mathbb{R}^{n_1} and P2⊂Rn2P_2\subset\mathbb{R}^{n_2} be compact Delzant polytopes, and consider their product

P=P1×P2⊂Rn1×Rn2.P=P_1\times P_2\subset\mathbb{R}^{n_1}\times\mathbb{R}^{n_2}.

We equip PP with the product facet labels and write (x,y)(x,y) for its coordinates, where x∈P1x\in P_1 and y∈P2y\in P_2. Denote the volumes of the factors by

V1=∫P1dx,V2=∫P2dy.V_1=\int_{P_1}dx,\qquad V_2=\int_{P_2}dy.

We assume throughout this subsection that the constant skew-symmetric matrix BB respects the product decomposition, namely,

B=(B100B2),Bi∈Λ2Rni.B=\begin{pmatrix}B_1&0\\0&B_2\end{pmatrix},\qquad B_i\in\Lambda^2\mathbb{R}^{n_i}.

No corresponding assumption will be imposed on the other skew-symmetric parameter AA.

We first observe that the extremal affine function is compatible with products.

Lemma 2.1. Let P=P1×P2P=P_1\times P_2 be a product of compact Delzant polytopes equipped with the product facet labels. Then

ℓP(x,y)=ℓP1(x)+ℓP2(y).\ell_P(x,y)=\ell_{P_1}(x)+\ell_{P_2}(y).

Proof. Recall that the extremal affine function on a labeled polytope QQ is uniquely characterized by

∫QℓQa dx=2∫∂Qa dσQ,a∈Aff⁡(Q).\int_Q \ell_Q a\,dx = 2\int_{\partial Q} a\,d\sigma_Q,\qquad a\in\operatorname{Aff}(Q).

Set

Vi=Vol⁡(Pi),Si=∫∂Pidσi.V_i=\operatorname{Vol}(P_i),\qquad S_i=\int_{\partial P_i}d\sigma_i.

In particular,

∫PiℓPi dxi=2Si.\int_{P_i}\ell_{P_i}\,dx_i=2S_i.

We claim that

ℓ(x,y)=ℓP1(x)+ℓP2(y)\ell(x,y)=\ell_{P_1}(x)+\ell_{P_2}(y)

satisfies the defining identity for ℓP\ell_P.

Every affine function on PP can be expressed as

a(x,y)=a1(x)+a2(y),a(x,y)=a_1(x)+a_2(y),

where ai∈Aff⁡(Pi)a_i\in\operatorname{Aff}(P_i). The product structure of the volume and boundary measures gives

∫Pℓa=V2∫P1ℓP1a1+V1∫P2ℓP2a2+2S2∫P1a1+2S1∫P2a2.\int_P\ell a=V_2\int_{P_1}\ell_{P_1}a_1+V_1\int_{P_2}\ell_{P_2}a_2 +2S_2\int_{P_1}a_1+2S_1\int_{P_2}a_2.

Using the defining identities for ℓP1\ell_{P_1} and ℓP2\ell_{P_2}, this becomes

∫Pℓa=2V2∫∂P1a1 dσ1+2V1∫∂P2a2 dσ2+2S2∫P1a1+2S1∫P2a2.\int_P\ell a=2V_2\int_{\partial P_1}a_1\,d\sigma_1+2V_1\int_{\partial P_2}a_2\,d\sigma_2 +2S_2\int_{P_1}a_1+2S_1\int_{P_2}a_2.

On the other hand, the facets of PP are precisely E1×P2E_1\times P_2 and P1×E2P_1\times E_2, where EiE_i ranges over the facets of PiP_i. Since the facet measures are the corresponding product measures,

2∫∂Pa dσ=2V2∫∂P1a1 dσ1+2V1∫∂P2a2 dσ2+2S2∫P1a1+2S1∫P2a2.2\int_{\partial P}a\,d\sigma=2V_2\int_{\partial P_1}a_1\,d\sigma_1+2V_1\int_{\partial P_2}a_2\,d\sigma_2 +2S_2\int_{P_1}a_1+2S_1\int_{P_2}a_2.

Consequently,

∫Pℓa=2∫∂Pa dσ\int_P\ell a=2\int_{\partial P}a\,d\sigma

for every a∈Aff⁡(P)a\in\operatorname{Aff}(P). By uniqueness of the extremal affine function,

ℓP=ℓP1+ℓP2.\ell_P=\ell_{P_1}+\ell_{P_2}.

We next introduce an averaging operator on the space of symplectic potentials. For u∈SB(P)u\in S_B(P), define its partial averages by

m1(x)=1V2∫P2u(x,y) dy,m2(y)=1V1∫P1u(x,y) dx,m_1(x)=\frac{1}{V_2}\int_{P_2}u(x,y)\,dy,\qquad m_2(y)=\frac{1}{V_1}\int_{P_1}u(x,y)\,dx,

and set

cu=1V1V2∫Pu(x,y) dx dy.c_u=\frac{1}{V_1V_2}\int_Pu(x,y)\,dx\,dy.

The product averaging operator is defined by

(Tu)(x,y)=m1(x)+m2(y)−cu.(10)(Tu)(x,y)=m_1(x)+m_2(y)-c_u. \tag*{(10)}

The normalization is chosen so that

1V2∫P2(Tu)(x,y) dy=m1(x)\frac{1}{V_2}\int_{P_2}(Tu)(x,y)\,dy=m_1(x)

and

1V1∫P1(Tu)(x,y) dx=m2(y).\frac{1}{V_1}\int_{P_1}(Tu)(x,y)\,dx=m_2(y).

In particular, TT preserves the partial averages of uu, and

T2u=Tu.T^2u=Tu.

Thus TT is a projection onto the space of potentials that split as sums of functions on the two factors. Our first task is to verify that this projection preserves generalized Kähler admissibility.

Lemma 2.2 (Admissibility of averaging). Suppose B=B1⊕B2B=B_1\oplus B_2. If u∈SB(P)u\in S_B(P), then

m1∈SB1(P1),m2∈SB2(P2),m_1\in S_{B_1}(P_1),\qquad m_2\in S_{B_2}(P_2),

and consequently

Tu∈SB(P).Tu\in S_B(P).

Proof. We verify the boundary regularity and positivity conditions separately. Since P=P1×P2P=P_1\times P_2, its Guillemin potential splits:

uG(x,y)=uG,1(x)+uG,2(y).u_G(x,y)=u_{G,1}(x)+u_{G,2}(y).

By admissibility,

u(x,y)=uG,1(x)+uG,2(y)+h(x,y),h∈C∞(P‾).u(x,y)=u_{G,1}(x)+u_{G,2}(y)+h(x,y),\qquad h\in C^\infty(\overline{P}).

Averaging over P2P_2, we obtain

m1(x)=uG,1(x)+c2+1V2∫P2h(x,y) dy,m_1(x)=u_{G,1}(x)+c_2+\frac{1}{V_2}\int_{P_2}h(x,y)\,dy,

where c2c_2 is a constant. Hence

m1−uG,1∈C∞(P1‾).m_1-u_{G,1}\in C^\infty(\overline{P_1}).

The same argument applies to m2m_2.

We next consider positivity in the interior. Write

Hess⁡u+−1B=(Hxx+−1B1HxyHyxHyy+−1B2).\operatorname{Hess}u+\sqrt{-1}B= \begin{pmatrix} H_{xx}+\sqrt{-1}B_1 & H_{xy}\\ H_{yx} & H_{yy}+\sqrt{-1}B_2 \end{pmatrix}.

Since this matrix is positive definite, its principal diagonal blocks are positive definite. Moreover,

Hess⁡m1+−1B1=1V2∫P2(Hxx+−1B1) dy.\operatorname{Hess}m_1+\sqrt{-1}B_1 =\frac{1}{V_2}\int_{P_2}(H_{xx}+\sqrt{-1}B_1)\,dy.

The right-hand side is an average of positive definite Hermitian matrices, and is therefore positive definite. Similarly,

Hess⁡m2+−1B2>0.\operatorname{Hess}m_2+\sqrt{-1}B_2>0.

It remains to verify positivity along the faces. Let E1⊂P1E_1\subset P_1 be a positive-dimensional face. On the relative interior of E1×P2E_1\times P_2, the admissibility of uu implies that the tangential restriction of Hess⁡u+−1B\operatorname{Hess}u+\sqrt{-1}B is positive definite. Restricting further to vectors tangent to E1E_1, and averaging over P2P_2, yields

(Hess⁡m1+−1B1)∣TE1>0.\left.(\operatorname{Hess}m_1+\sqrt{-1}B_1)\right|_{T E_1}>0.

The tangential Hessians are well defined because the restrictions of the smooth parts of the potentials extend to the faces. The corresponding argument applies to every face of P2P_2. Thus mi∈SBi(Pi)m_i\in S_{B_i}(P_i). Finally,

Hess⁡(Tu)+−1B=(Hess⁡m1+−1B100Hess⁡m2+−1B2),\operatorname{Hess}(Tu)+\sqrt{-1}B= \begin{pmatrix} \operatorname{Hess}m_1+\sqrt{-1}B_1 & 0\\ 0 & \operatorname{Hess}m_2+\sqrt{-1}B_2 \end{pmatrix},

which satisfies all the admissibility conditions on the product polytope. □

The next lemma is the central variational ingredient. It shows that averaging decreases the generalized Mabuchi functional, with strict inequality unless the potential already splits.

Lemma 2.3 (Energy decreases under averaging). Suppose B=B1⊕B2B = B_1 \oplus B_2. Then, for every u∈SB(P)u \in\mathcal{S}_B(P),

MB(Tu)≤MB(u).\mathcal{M}_B(Tu) \le\mathcal{M}_B(u).

Moreover, equality holds if and only if

Hess⁡xyu=0\operatorname{Hess}_{xy} u = 0

throughout the interior of PP.

Proof. Recall from (7) that

MB(u)=FP(u)−∫Plog⁡det⁡(Hess⁡u+−1B) dx dy+C,\mathcal{M}_B(u) = \mathcal{F}_P(u) - \int_P \log\det(\operatorname{Hess} u + \sqrt{-1}B)\,dx\,dy + C,

where CC is independent of uu.

We first compare the linear terms. The equality of partial averages implies

∫PTu dx dy=∫Pu dx dy.\int_P Tu\,dx\,dy = \int_P u\,dx\,dy.

More generally, Lemma 2.1 and the partial-average identities give

∫PℓP(Tu) dx dy=∫PℓPu dx dy.\int_P \ell_P(Tu)\,dx\,dy = \int_P \ell_Pu\,dx\,dy.

Furthermore, the boundary of PP is the union of the product facets

∂P=(∂P1×P2)∪(P1×∂P2),\partial P = (\partial P_1 \times P_2) \cup(P_1 \times\partial P_2),

up to sets of boundary measure zero. Thus,

∫∂PTu dσ=∫∂Pu dσ.\int_{\partial P} Tu\,d\sigma= \int_{\partial P} u\,d\sigma.

Consequently,

FP(Tu)=FP(u).(11)\mathcal{F}_P(Tu) = \mathcal{F}_P(u). \tag*{(11)}

It remains to compare the determinant terms. Set

K=Hess⁡u+−1B=(D1CC∗D2),K = \operatorname{Hess} u + \sqrt{-1}B = \begin{pmatrix} D_1 & C \\ C^* & D_2 \end{pmatrix},

where

D1=Hxx+−1B1,D2=Hyy+−1B2,C=Hxy.D_1 = H_{xx} + \sqrt{-1}B_1,\qquad D_2 = H_{yy} + \sqrt{-1}B_2,\qquad C = H_{xy}.

Since KK is positive definite Hermitian, so are D1D_1 and D2D_2.

By the Schur complement formula,

det⁡K=det⁡D1det⁡(D2−C∗D1−1C).\det K = \det D_1 \det(D_2 - C^*D_1^{-1}C).

Since C∗D1−1CC^*D_1^{-1}C is positive semidefinite, the Fischer determinant inequality yields

det⁡K≤det⁡D1det⁡D2.\det K \le\det D_1 \det D_2.

Equality holds precisely when C=0C = 0. Taking logarithms, we obtain

log⁡det⁡K≤log⁡det⁡D1+log⁡det⁡D2.(12)\log\det K \le\log\det D_1 + \log\det D_2. \tag*{(12)}

We next apply the concavity of log⁡det⁡\log\det on the cone of positive definite Hermitian matrices. For each fixed x∈P1x \in P_1, Jensen’s inequality gives

1V2∫P2log⁡det⁡D1(x,y) dy≤log⁡det⁡(1V2∫P2D1(x,y) dy)=log⁡det⁡(Hess⁡m1(x)+−1B1).\frac{1}{V_2}\int_{P_2}\log\det D_1(x,y)\,dy \le\log\det\left(\frac{1}{V_2}\int_{P_2}D_1(x,y)\,dy\right) = \log\det(\operatorname{Hess}m_1(x)+\sqrt{-1}B_1).

Integrating over P1P_1, we conclude that

∫Plog⁡det⁡D1 dx dy≤V2∫P1log⁡det⁡(Hess⁡m1+−1B1) dx.\int_P\log\det D_1\,dx\,dy \le V_2\int_{P_1}\log\det(\operatorname{Hess}m_1+\sqrt{-1}B_1)\,dx.

Similarly,

∫Plog⁡det⁡D2 dx dy≤V1∫P2log⁡det⁡(Hess⁡m2+−1B2) dy.\int_P\log\det D_2\,dx\,dy \le V_1\int_{P_2}\log\det(\operatorname{Hess}m_2+\sqrt{-1}B_2)\,dy.

Combining these inequalities with (12), we obtain

∫Plog⁡det⁡K dx dy≤V2∫P1log⁡det⁡(Hess⁡m1+−1B1) dx+V1∫P2log⁡det⁡(Hess⁡m2+−1B2) dy=∫Plog⁡det⁡(Hess⁡(Tu)+−1B) dx dy.\begin{aligned} \int_P\log\det K\,dx\,dy &\le V_2\int_{P_1}\log\det(\operatorname{Hess}m_1+\sqrt{-1}B_1)\,dx \\ &\quad+V_1\int_{P_2}\log\det(\operatorname{Hess}m_2+\sqrt{-1}B_2)\,dy \\ &= \int_P\log\det(\operatorname{Hess}(Tu)+\sqrt{-1}B)\,dx\,dy. \end{aligned}

The logarithmic boundary singularities are integrable, so these inequalities extend to the integrals over the full polytopes.

Together with (11), this proves

MB(Tu)≤MB(u).\mathcal{M}_B(Tu)\le\mathcal{M}_B(u).

Finally, if equality holds, the nonnegative difference in the pointwise Fischer inequality must vanish almost everywhere. Hence C=Hxy=0C=H_{xy}=0 almost everywhere, and therefore everywhere in P∘P^\circ by smoothness. Conversely, if Hxy=0H_{xy}=0, then uu is a sum of functions on the two factors, up to a constant, and Tu=uTu=u. Equality follows. □

We are now ready to prove the splitting theorem.

Theorem 2.4 (Splitting of extremal potentials). Let P=P1×P2P=P_1\times P_2 be a product of compact Delzant polytopes, and suppose that B=B1⊕B2B=B_1\oplus B_2. If u∈SB(P)u\in S_B(P) is an extremal toric generalized Kähler potential, that is,

SB(u)=ℓP,S_B(u)=\ell_P,

then uu splits, up to an affine-linear function:

u(x,y)=u1(x)+u2(y)+a(x,y),a∈Aff⁡(P),u(x,y)=u_1(x)+u_2(y)+a(x,y),\qquad a\in\operatorname{Aff}(P),

where

ui∈SBi(Pi),SBi(ui)=ℓPi.u_i\in S_{B_i}(P_i),\qquad S_{B_i}(u_i)=\ell_{P_i}.

No block-diagonal assumption on AA is required.

Proof. Since uu solves the extremal equation, it is a critical point of the relative generalized Mabuchi functional. By the convexity established in (9), it is a global minimizer on the admissible space SB(P)S_B(P).

By Lemma 2.2, the averaged potential TuTu is also admissible. Therefore,

MB(u)≤MB(Tu).\mathcal{M}_B(u)\le\mathcal{M}_B(Tu).

On the other hand, Lemma 2.3 gives

MB(Tu)≤MB(u).\mathcal{M}_B(Tu)\le\mathcal{M}_B(u).

Hence equality holds. The equality characterization in Lemma 2.3 implies

Hess⁡xyu=0.\operatorname{Hess}_{xy} u = 0.

Since P∘=P1∘×P2∘P^\circ= P_1^\circ\times P_2^\circ is connected, this identity implies that uu is the sum of functions depending only on the respective factors. In particular,

u=Tu=m1(x)+m2(y)−cu.u = Tu = m_1(x) + m_2(y) - c_u.

The admissibility of the factors follows from Lemma 2.2.

We next verify the extremal equations on the factors. Since both uu and BB split,

Hess⁡u+−1B=(Hess⁡m1+−1B100Hess⁡m2+−1B2).\operatorname{Hess} u + \sqrt{-1}B = \begin{pmatrix} \operatorname{Hess} m_1 + \sqrt{-1}B_1 & 0 \\ 0 & \operatorname{Hess} m_2 + \sqrt{-1}B_2 \end{pmatrix}.

The generalized Abreu formula therefore gives

SB(u)=SB1(m1)+SB2(m2).S_B(u) = S_{B_1}(m_1) + S_{B_2}(m_2).

Together with Lemma 2.1, this yields

SB1(m1)(x)−ℓP1(x)=−(SB2(m2)(y)−ℓP2(y)).S_{B_1}(m_1)(x) - \ell_{P_1}(x) = -(S_{B_2}(m_2)(y) - \ell_{P_2}(y)).

Since the two sides depend on independent variables, each must be constant. The integration-by-parts identity (8), applied to the constant test function on each factor, shows that these constants vanish. Thus

SBi(mi)=ℓPi,i=1,2.S_{B_i}(m_i) = \ell_{P_i}, \qquad i=1,2.

Finally, the admissible space and the generalized scalar curvature equation are independent of AA. Hence no splitting hypothesis on AA is needed. □

Corollary 2.5. Suppose B=B1⊕B2B = B_1 \oplus B_2. Then an extremal toric generalized Kähler potential exists on P1×P2P_1 \times P_2 if and only if extremal potentials exist on both factors.

Proof. The forward implication follows from Theorem 2.4. Conversely, if ui∈SBi(Pi)u_i \in\mathcal{S}_{B_i}(P_i) solve the corresponding extremal equations, then

u(x,y)=u1(x)+u2(y)u(x,y) = u_1(x) + u_2(y)

belongs to SB(P)\mathcal{S}_B(P), and the additivity of generalized scalar curvature and Lemma 2.1 show that it is extremal. □

Remark 2.6 (Splitting of the geometric structure). The conclusion of Theorem 2.4 concerns the symplectic potential and, consequently, the associated ordinary toric Kähler reduction. It does not necessarily imply splitting of the full generalized Kähler structure. Indeed, even when Hess⁡u\operatorname{Hess} u and BB are block diagonal, mixed components of AA may couple the two factors through the matrix Hess⁡u+A\operatorname{Hess} u + A. If both AA and BB are block diagonal, the full toric generalized Kähler structure splits. The theorem extends inductively to products of finitely many factors.

An explicit non-product example with split potential

Potential splitting and splitting of the full generalized Kähler structure are distinct questions. We now give an intrinsic obstruction to the latter.

Meaning of a GK product

A product of GK structures on two curves has bi-Hermitian tensors

g=g1⊕g2,I=I1⊕I2,J=J1⊕J2,β=pr⁡1∗β1+pr⁡2∗β2.g=g_1\oplus g_2,\qquad I=I_1\oplus I_2,\qquad J=J_1\oplus J_2,\qquad\beta=\operatorname{pr}_1^*\beta_1+\operatorname{pr}_2^*\beta_2.

A GK product decomposition up to diffeomorphism means that a diffeomorphism identifies these tensors with the given structure. The obstruction below also survives closed two-form transforms, since these change β\beta but not g,I,Jg,I,J. We do not claim here that the Riemannian metric alone has no product decomposition.

Lemma 3.1. Every GK structure on a real surface has [I,J]=0[I,J]=0, and hence has zero canonical real Poisson tensor π=12[I,J]g−1\pi=\frac{1}{2}[I,J]g^{-1}. A product of two GK curves therefore has π=0\pi=0 and πJ=0\pi_J=0.

Proof. On a two-dimensional Euclidean vector space the orthogonal complex structures are exactly the two rotations by +π/2+\pi/2 and −π/2-\pi/2. Thus two complex structures Hermitian for the same metric satisfy I=JI=J or I=−JI=-J at every point, and commute in either case. On a product the commutator is the direct sum of the two factor commutators. The assertion follows, as does its invariance under diffeomorphism. □

A concrete cscGK non-product with split potential

There is also a fully explicit example with a fixed numerical skew parameter and no perturbative existence argument. It uses the real skew addition to the complex-structure matrix, denoted by AA, rather than the parameter BB in H+−1BH+\sqrt{-1}B.

Let us consider X=CP1×CP1X=\mathbb{CP}^{1}\times\mathbb{CP}^{1}, the corresponding Delzant polytope is the standard square. We let

w(x,y)=f(x)+f(y),f(s)=12((1+s)log⁡(1+s)+(1−s)log⁡(1−s)).w(x,y)=f(x)+f(y),\qquad f(s)=\frac{1}{2}\bigl((1+s)\log(1+s)+(1-s)\log(1-s)\bigr).

and take

A=(01/10−1/100),B=0.(13)A=\begin{pmatrix}0&1/10\\-1/10&0\end{pmatrix},\qquad B=0. \tag*{(13)}

These data are globally admissible: ww has the standard Guillemin boundary behavior, and admissibility is independent of AA. The matrix in the toric complex-structure ansatz is explicitly

D2w+A=((1−x2)−11/10−1/10(1−y2)−1).D^2w+A=\begin{pmatrix}(1-x^2)^{-1}&1/10\\-1/10&(1-y^2)^{-1}\end{pmatrix}.

By (3), its generalized scalar curvature is

S0(w)=−∂x2(1−x2)−∂y2(1−y2)=4.S_0(w)=-\partial_x^2(1-x^2)-\partial_y^2(1-y^2)=4.

On the other hand, let Xi=∂/∂θiX_i=\partial/\partial\theta_i, i=1,2i=1,2, denote the infinitesimal generators of the real torus action on CP1×CP1\mathbb{CP}^{1}\times\mathbb{CP}^{1}. Define

Zi=Xi−−1JXi,i=1,2.Z_i=X_i-\sqrt{-1}JX_i,\qquad i=1,2.

These are holomorphic vector fields of type (1,0)(1,0) associated with the complexified torus action. On the dense open orbit (C∗)2(\mathbb{C}^*)^2, in holomorphic coordinates (z1,z2)(z_1,z_2) adapted to the torus action, they take the form

Zi=2−1zi∂∂zi.Z_i=2\sqrt{-1}z_i\frac{\partial}{\partial z_i}.

Thus the corresponding holomorphic Poisson tensor is

πJ=∑i,j=12AijZi∧Zj=15Z1∧Z2.\pi_J=\sum_{i,j=1}^{2}A_{ij}Z_i\wedge Z_j=\frac{1}{5}Z_1\wedge Z_2.

Equivalently, in holomorphic coordinates,

πJ=−45z1z2∂∂z1∧∂∂z2.\pi_J=-\frac{4}{5}z_1z_2\frac{\partial}{\partial z_1}\wedge\frac{\partial}{\partial z_2}.

This tensor has a nonzero mixed component with respect to the product decomposition. Consequently, the generalized Kähler structure does not split as a product, even though its symplectic potential is a sum of potentials on the two factors.

Mixed-BB deformations with non-split extremal potentials

In the previous section, we constructed an extremal toric generalized Kähler structure on CP1×CP1\mathbb{CP}^{1}\times\mathbb{CP}^{1} with a splitting symplectic potential but a non-splitting generalized Kähler structure, arising from a nonzero mixed component of the skew-symmetric parameter AA. This shows that even when the symplectic potential splits, mixed components of AA can obstruct splitting of the full generalized Kähler structure. To investigate whether the symplectic potential itself must split, we now eliminate this obstruction by assuming that AA respects the product decomposition. In fact, we take A=0A=0 throughout this section and instead introduce a nonzero mixed component of BB. We show that, even under this restriction on AA, there exist extremal toric generalized Kähler structures on CP1×CP1\mathbb{CP}^{1}\times\mathbb{CP}^{1} whose symplectic potentials do not split.

Local deformation from a Kähler extremal metric

4.1. Local deformation from a Kähler extremal metric. We begin by recalling the local deformation theory for extremal toric generalized Kähler structures developed by Apostolov–Streets–Ustinovskiy [2]. Their argument adapts the openness method of LeBrun–Simanca [11] to the generalized Abreu equation. The key observation is that, near an ordinary extremal Kähler metric, the generalized extremal equation can be solved by the implicit function theorem after removing the finite-dimensional kernel arising from holomorphic vector fields.

Let P⊂RnP\subset\mathbb{R}^{n} be a compact Delzant polytope with fixed facet labels, and let u0∈S0(P)u_{0}\in\mathcal{S}_{0}(P) be the symplectic potential of a smooth toric extremal Kähler metric. Thus,

S0(u0)=ℓP,S_{0}(u_{0})=\ell_{P},

where ℓP\ell_{P} is the extremal affine function determined by the labeled polytope.

Fix a real skew-symmetric matrix B∗∈Λ2RnB_{*}\in\Lambda^{2}\mathbb{R}^{n}, and consider the one-parameter family

Bt=tB∗.B_{t}=tB_{*}.

We seek admissible symplectic potentials ut∈SBt(P)u_{t}\in\mathcal{S}_{B_{t}}(P) satisfying

SBt(ut)=ℓP,ut=0=u0.(14)S_{B_{t}}(u_{t})=\ell_{P},\qquad u_{t=0}=u_{0}. \tag*{(14)}

The generalized scalar curvature is given by

SBt(u)=−∑i,j=1n∂2∂xi∂xj(Hess⁡u+−1tB∗)ij−1.S_{B_{t}}(u)=-\sum_{i,j=1}^{n}\frac{\partial^{2}}{\partial x_{i}\partial x_{j}}\left(\operatorname{Hess}u+\sqrt{-1}tB_{*}\right)^{-1}_{ij}.

The equation depends on BtB_{t}, but not on the other skew-symmetric parameter AA.

Proposition 4.1 (Local deformation of extremal potentials). Let u0∈S0(P)u_{0}\in\mathcal{S}_{0}(P) be the symplectic potential of a smooth compact toric extremal Kähler metric, and let B∗∈Λ2RnB_{*}\in\Lambda^{2}\mathbb{R}^{n} be fixed. Then there exists ε>0\varepsilon>0 and a smooth family of admissible potentials

ut∈StB∗(P),∣t∣<ε,u_{t}\in\mathcal{S}_{tB_{*}}(P),\qquad|t|<\varepsilon,

such that

StB∗(ut)=ℓP,ut=0=u0.S_{tB_{*}}(u_{t})=\ell_{P},\qquad u_{t=0}=u_{0}.

The potentials may be normalized by requiring

ut−u0⊥Aff⁡(P)u_{t}-u_{0}\perp\operatorname{Aff}(P)

with respect to the L2(P,dx)L^{2}(P,dx) inner product. With this normalization, the family is locally unique and satisfies

u−t=ut.u_{-t}=u_{t}.

Consequently, it admits an expansion

ut=u0+t2v+O(t4),(15)u_t = u_0 + t^2 v + O(t^4), \tag*{(15)}

where the expansion holds in every fixed finite smooth norm on the compact toric manifold, using the corresponding Kähler-potential identification.

Proof. The local existence statement follows from [2]. We recall the implicit-function-theorem construction and explain why the normalized family can be chosen to depend smoothly on tt.

Let (M,J,ω0)(M,J,\omega_0) be the compact toric Kähler manifold associated with u0u_0. Instead of working directly with symplectic potentials, which have logarithmic singularities along the boundary of PP, we use the corresponding smooth TT-invariant Kähler potentials on MM.

For a sufficiently small invariant function φ\varphi, set

ωφ=ω0+ddJcφ.\omega_\varphi= \omega_0 + dd_J^c\varphi.

and let μφ\mu^\varphi be the corresponding moment map, normalized to have image PP. Following [2], define

Fφ,t=ωφ+t∑i,j=1n(B∗)ij dμiφ∧dμjφ.F_{\varphi,t} = \omega_\varphi+ t\sum_{i,j=1}^{n}(B_*)_{ij}\,d\mu_i^\varphi\wedge d\mu_j^\varphi.

For φ\varphi and tt sufficiently small, this form tames JJ and determines a toric generalized Kähler structure of symplectic type. Under the Legendre-transform identification, its Abreu–Guillemin data are (u,0,tB∗)(u,0,tB_*).

Consider the nonlinear operator

W(φ,t)=Gscal⁡(Fφ,t,J)−ℓP(μφ).W(\varphi,t)=\operatorname{Gscal}(F_{\varphi,t},J)-\ell_P(\mu^\varphi).

The generalized extremal equation is equivalent to

W(φ,t)=0.W(\varphi,t)=0.

At the reference metric,

W(0,0)=0.W(0,0)=0.

We first establish the regularity of this operator. Fix k≥0k\geq0 and 0<γ<10<\gamma<1. On a sufficiently small open neighborhood of (0,0)(0,0), the operator WW extends to a C∞C^\infty map

W:Ck+4,γ(M,R)T×(−ε0,ε0)⟶Ck,γ(M,R)T.W:C^{k+4,\gamma}(M,\mathbb{R})^T\times(-\varepsilon_0,\varepsilon_0)\longrightarrow C^{k,\gamma}(M,\mathbb{R})^T.

At (0,0)(0,0), the linearization of WW in the potential direction is the ordinary Lichnerowicz operator:

DφW(0,0)=Lω0.D_\varphi W(0,0)=\mathcal{L}_{\omega_0}.

Since ω0\omega_0 is extremal, this operator is elliptic and self-adjoint. Its kernel on TT-invariant functions consists of the affine Hamiltonians, namely,

K={a∘μ0:a∈Aff⁡(P)}.\mathcal{K}=\{a\circ\mu^0:a\in\operatorname{Aff}(P)\}.

Let Π0\Pi_0 denote the L2(M,ω0n/n!)L^2(M,\omega_0^n/n!)-orthogonal projection onto K\mathcal{K}. For each k≥0k\geq0, define

Ek=Ck+4,γ(M,R)T∩K⊥,E_k=C^{k+4,\gamma}(M,\mathbb{R})^T\cap\mathcal{K}^{\perp},
Yk=Ck,γ(M,R)T∩K⊥.Y_k=C^{k,\gamma}(M,\mathbb{R})^T\cap\mathcal{K}^{\perp}.

Consider the projected operator

F(φ,t)=(Id⁡−Π0)W(φ,t).\mathcal{F}(\varphi,t)=(\operatorname{Id}-\Pi_0)W(\varphi,t).

It defines a C∞C^\infty map

F:Ek×(−ε0,ε0)⟶Yk.\mathcal{F}:E_k\times(-\varepsilon_0,\varepsilon_0)\longrightarrow Y_k.

Moreover,

DφF(0,0)=Lω0∣Ek.D_\varphi\mathcal{F}(0,0)=\mathcal{L}_{\omega_0}|_{E_k}.

By elliptic Fredholm theory and Schauder estimates,

Lω0:Ek⟶YkL_{\omega_0}: E_k \longrightarrow Y_k

is a bounded linear isomorphism.

The smooth implicit function theorem therefore provides a unique sufficiently small smooth curve

t⟼φt∈Ek,φ0=0,t \longmapsto\varphi_t \in E_k,\qquad\varphi_0=0,

satisfying

(Id⁡−Π0)W(φt,t)=0.(\operatorname{Id}-\Pi_0)W(\varphi_t,t)=0.

We next verify that the projected solution solves the full extremal equation. The integration-by-parts characterization of the extremal affine function implies

ΠtW(φt,t)=0,\Pi_tW(\varphi_t,t)=0,

where Πt\Pi_t is the orthogonal projection onto the affine Hamiltonians associated with the deformed moment map μφt\mu^{\varphi_t}, using the corresponding symplectic volume form.

Since the projected equation implies

W(φt,t)∈K,W(\varphi_t,t)\in K,

and the pairing between the fixed affine Hamiltonians and their deformed counterparts remains nondegenerate for small tt, we conclude that

W(φt,t)=0.W(\varphi_t,t)=0.

Thus φt\varphi_t defines a family of extremal generalized Kähler structures. Elliptic regularity shows that each φt\varphi_t is smooth on MM.

Furthermore, the implicit function theorem can be applied in every higher Hölder space. The resulting local solutions agree by uniqueness in the normalized slice. Consequently, the family depends smoothly on tt in all CkC^k-norms.

The inverse Legendre transform gives the corresponding symplectic potentials utu_t. Since the Legendre-transform identification depends smoothly on the Kähler potential near the reference metric, we obtain

t⟼ut−u0∈C∞(P‾)t\longmapsto u_t-u_0\in C^\infty(\overline{P})

smoothly. The potentials are defined up to affine-linear functions, which we fix by requiring

ut−u0⊥Aff⁡(P).u_t-u_0\perp\operatorname{Aff}(P).

This normalization preserves smooth parameter dependence.

Finally, we establish the symmetry under t↦−tt\mapsto-t. For every admissible potential uu, we have

Hess⁡u−tB∗−1=Hess⁡u+tB∗−1,\operatorname{Hess}u-tB_*\sqrt{-1}=\operatorname{Hess}u+tB_*\sqrt{-1},

and hence

S−tB∗(u)=StB∗(u).S_{-tB_*}(u)=S_{tB_*}(u).

The admissibility conditions are also invariant under t↦−tt\mapsto-t.

Therefore, the normalized potentials utu_t and u−tu_{-t} satisfy the same generalized extremal equation for each sufficiently small tt. Local uniqueness implies

u−t=ut.u_{-t}=u_t.

The normalized family is thus smooth and even in tt. In particular, all its odd derivatives at t=0t=0 vanish. Taylor’s theorem gives

ut=u0+t2v+OCk(t4),v=12d2utdt2∣t=0,u_t=u_0+t^2v+O_{C^k}(t^4),\qquad v=\frac{1}{2}\left.\frac{d^2u_t}{dt^2}\right|_{t=0},

for every fixed k≥0k\geq0. □

Non-splitting extremal potentials on CP1×CP1\mathbb{CP}^{1} \times\mathbb{CP}^{1}

We now apply the deformation theory of the previous subsection to construct a family of extremal toric generalized Kähler structures on CP1×CP1\mathbb{CP}^{1} \times\mathbb{CP}^{1} whose symplectic potentials do not split. Starting from the product Fubini–Study metric, we introduce a mixed skew-symmetric parameter BtB_t. We first determine the leading correction to the symplectic potential and then establish an exact obstruction to splitting.

To investigate whether the extremal potentials constructed in Proposition 4.1 split, we examine their leading-order deformation from the product Fubini–Study potential u0u_0.

Let

S(u,t)=StB∗(u),B∗=(01−10).\mathcal{S}(u,t)=S_{tB_*}(u), \qquad B_*=\begin{pmatrix}0 & 1 \\ -1 & 0\end{pmatrix}.

By Proposition 4.1, the normalized family utu_t is smooth and even in tt. Consequently, Taylor’s theorem gives

ut=u0+t2v+O(t4),v=12d2utdt2∣t=0,u_t=u_0+t^2v+O(t^4), \qquad v=\left.\frac{1}{2}\frac{d^2u_t}{dt^2}\right|_{t=0},

where the expansion is understood in the appropriate smooth norms for ut−u0u_t-u_0.

For the fixed potential u0u_0, the generalized scalar curvature is an even function of tt. Since S0(u0)=4S_0(u_0)=4, we may write

S(u0,t)=4−t2Q+O(t4),\mathcal{S}(u_0,t)=4-t^2Q+O(t^4),

where

Q=−12∂2∂t2S(u0,t)∣t=0.Q=-\left.\frac{1}{2}\frac{\partial^2}{\partial t^2}\mathcal{S}(u_0,t)\right|_{t=0}.

The function QQ measures the leading-order change in generalized scalar curvature caused by the deformation of BB, while keeping the symplectic potential fixed.

We now account for the variation of the potential. Taylor expansion of S\mathcal{S} with respect to uu, around u0u_0, gives

S(ut,t)=S(u0,t)+DuS(u0,t)[ut−u0]+O(∥ut−u0∥2).\mathcal{S}(u_t,t)=\mathcal{S}(u_0,t)+D_u\mathcal{S}(u_0,t)[u_t-u_0]+O(\lVert u_t-u_0\rVert^2).

At t=0t=0, the derivative with respect to uu is the linearization of the ordinary Abreu scalar curvature operator:

DuS(u0,0)[v]=Lv.D_u\mathcal{S}(u_0,0)[v]=Lv.

Since S(u,t)\mathcal{S}(u,t) is even in tt for each fixed uu, its derivative with respect to uu is also even in tt. Hence

DuS(u0,t)[v]=Lv+O(t2).D_u\mathcal{S}(u_0,t)[v]=Lv+O(t^2).

Combining the preceding expansions yields

StB∗(ut)=4−t2Q+t2Lv+O(t4)=4+t2(Lv−Q)+O(t4).\begin{aligned} S_{tB_*}(u_t)&=4-t^2Q+t^2Lv+O(t^4)\\ &=4+t^2(Lv-Q)+O(t^4). \end{aligned}

Since utu_t satisfies the generalized extremal equation

StB∗(ut)=4S_{tB_*}(u_t)=4

for all sufficiently small tt, comparison of the coefficients of t2t^2 implies

Lv=Q.(16)Lv=Q. \tag*{(16)}

Thus the leading correction vv is governed by a linear elliptic equation: LL describes the response of the ordinary scalar curvature to variations of the symplectic potential, whereas QQ records the second-order effect of the mixed parameter BtB_t. We compute these quantities explicitly below and show that vv has nonvanishing mixed second derivatives.

The product Fubini–Study structure

Consider the standard labeled square

P=[−1,1]×[−1,1],P=[-1,1]\times[-1,1],

with defining affine functions

L1=1+x,L2=1−x,L3=1+y,L4=1−y.L_1=1+x,\quad L_2=1-x,\quad L_3=1+y,\quad L_4=1-y.

The corresponding Guillemin potential is

u0(x,y)=f(x)+f(y),(17)u_0(x,y)=f(x)+f(y), \tag*{(17)}

where

f(s)=12[(1+s)log⁡(1+s)+(1−s)log⁡(1−s)].f(s)=\frac{1}{2}\left[(1+s)\log(1+s)+(1-s)\log(1-s)\right].

This is the symplectic potential of the product Fubini–Study metric, with the normalization determined by the labeled polytope.

Set

p(x)=1−x2,q(y)=1−y2.p(x)=1-x^2,\qquad q(y)=1-y^2.

Since

f′′(s)=11−s2,f''(s)=\frac{1}{1-s^2},

we have

H0:=Hess⁡u0=(p−100q−1),H0−1=(p00q).H_0:=\operatorname{Hess}u_0= \begin{pmatrix} p^{-1} & 0\\ 0 & q^{-1} \end{pmatrix}, \qquad H_0^{-1}= \begin{pmatrix} p & 0\\ 0 & q \end{pmatrix}.

The ordinary Abreu formula gives

S0(u0)=−p′′(x)−q′′(y)=4.S_0(u_0)=-p''(x)-q''(y)=4.

Thus u0u_0 is a constant-scalar-curvature Kähler potential. The extremal affine function of the labeled square is

ℓP=4.\ell_P=4.

We introduce the family of skew-symmetric matrices

Bt=tJ∗,J∗=(01−10).B_t=tJ_*, \qquad J_*= \begin{pmatrix} 0 & 1\\ -1 & 0 \end{pmatrix}.

For the undeformed potential u0u_0, we obtain

H0+−1Bt=(p−1t−1−t−1q−1),H_0+\sqrt{-1}B_t= \begin{pmatrix} p^{-1} & t\sqrt{-1}\\ -t\sqrt{-1} & q^{-1} \end{pmatrix},

and hence

det⁡(H0+Bt−1)=1pq−t2.\det(H_0+B_t\sqrt{-1})=\frac{1}{pq}-t^2.

Since 0<p,q≤10<p,q\leq1 in the interior of PP, the Hermitian matrix H0+−1BtH_0+\sqrt{-1}B_t is positive definite for ∣t∣≪1|t|\ll1. The corresponding facewise admissibility conditions also hold. Thus u0u_0 remains admissible for sufficiently small tt, although it does not remain extremal when t≠0t\ne0.

By Proposition 4.1, there exists ε>0\varepsilon>0 and a smooth family

ut∈SBt(P),∣t∣<ε,u_t\in\mathcal{S}_{B_t}(P),\qquad|t|<\varepsilon,

satisfying

SBt(ut)=4,ut=0=u0.(18)\mathcal{S}_{B_t}(u_t)=4,\qquad u_{t=0}=u_0. \tag*{(18)}

We choose the affine normalization

ut−u0⊥Aff⁡(P)u_t-u_0\perp\operatorname{Aff}(P)

with respect to L2(P,dx dy)L^2(P,dx\,dy). Also we have

ut=u0+t2v+O(t4),(19)u_t=u_0+t^2v+O(t^4), \tag*{(19)}

where vv is a smooth function on the closed square. The remainder is understood in each fixed CkC^k-norm for the smooth differences of symplectic potentials.

The second-order deformation equation

We next determine the coefficient vv in (19). Note that

XBt(u0)=Re⁡(H0+−1Bt)−1=11−t2pq(p00q).(20)X_{B_t}(u_0)=\operatorname{Re}(H_0+\sqrt{-1}B_t)^{-1}=\frac{1}{1-t^2pq}\begin{pmatrix}p&0\\0&q\end{pmatrix}. \tag*{(20)}

It follows that

SBt(u0)=−∂x2(p1−t2pq)−∂y2(q1−t2pq).S_{B_t}(u_0)=-\partial_x^2\left(\frac{p}{1-t^2pq}\right)-\partial_y^2\left(\frac{q}{1-t^2pq}\right).

Expanding at t=0t=0, we obtain

SBt(u0)=4−t2Q+O(t4),S_{B_t}(u_0)=4-t^2Q+O(t^4),

where

Q(x,y)=(p2)′′(x)q(y)+p(x)(q2)′′(y)=−8+16x2+16y2−24x2y2.(21)Q(x,y)=(p^2)''(x)q(y)+p(x)(q^2)''(y) = -8+16x^2+16y^2-24x^2y^2. \tag*{(21)}

In particular, QQ is not affine-linear. Let LL denote the linearization of ordinary Abreu scalar curvature at u0u_0. Since

δ(H−1)=−H0−1(Hess⁡v)H0−1,\delta(H^{-1})=-H_0^{-1}(\operatorname{Hess}v)H_0^{-1},

we obtain

Lv=∂x2(p2vxx)+2∂x∂y(pqvxy)+∂y2(q2vyy).(22)Lv=\partial_x^2(p^2v_{xx})+2\partial_x\partial_y(pqv_{xy})+\partial_y^2(q^2v_{yy}). \tag*{(22)}

Substituting ut=u0+t2v+O(t4)u_t=u_0+t^2v+O(t^4) into (18) and comparing the coefficients of t2t^2 gives

Lv=Q.(23)Lv=Q. \tag*{(23)}

Explicit solution of the linearized equation

Let

P2(s)=12(3s2−1)P_2(s)=\frac{1}{2}(3s^2-1)

be the second Legendre polynomial. The expression (21) can be rewritten as

Q=163(P2(x)+P2(y))−323P2(x)P2(y).Q=\frac{16}{3}\bigl(P_2(x)+P_2(y)\bigr)-\frac{32}{3}P_2(x)P_2(y).

We use the elementary identities

((1−s2)P2′(s))′=−6P2(s),\bigl((1-s^2)P_2'(s)\bigr)'=-6P_2(s),

and

((1−s2)2P2′′(s))′′=24P2(s).\bigl((1-s^2)^2P_2''(s)\bigr)''=24P_2(s).

Applying (22), we find

L(P2(x))=24P2(x),L(P2(y))=24P2(y).L(P_2(x))=24P_2(x),\qquad L(P_2(y))=24P_2(y).

Moreover,

L(P2(x)P2(y))=(24+24+72)P2(x)P2(y)=120P2(x)P2(y).L(P_2(x)P_2(y))=(24+24+72)P_2(x)P_2(y) =120P_2(x)P_2(y).

It follows that

v(x,y)=29(P2(x)+P2(y))−445P2(x)P2(y)(24)v(x,y)=\frac{2}{9}\bigl(P_2(x)+P_2(y)\bigr)-\frac{4}{45}P_2(x)P_2(y) \tag*{(24)}

satisfies Lv=QLv=Q.

This solution is orthogonal to Aff⁡(P)=span⁡{1,x,y}\operatorname{Aff}(P)=\operatorname{span}\{1,x,y\}. It is also unique under this normalization. Indeed, integration by parts gives

∫PwLw dx dy=∫P(p2wxx2+2pq wxy2+q2wyy2) dx dy.(25)\int_P wLw\,dx\,dy=\int_P\left(p^2w_{xx}^2+2pq\,w_{xy}^2+q^2w_{yy}^2\right)\,dx\,dy. \tag*{(25)}

The boundary terms vanish because pp and qq vanish at the corresponding endpoints. Consequently, Lw=0Lw=0 implies Hess⁡w=0\operatorname{Hess}w=0 in the interior of PP, so ww is affine-linear.

We have therefore obtained the following second-order expansion:

ut(x,y)=u0(x,y)+t2[29(P2(x)+P2(y))−445P2(x)P2(y)]+O(t4).(26)u_t(x,y)=u_0(x,y)+t^2\left[\frac{2}{9}\left(P_2(x)+P_2(y)\right)-\frac{4}{45}P_2(x)P_2(y)\right]+O(t^4). \tag*{(26)}

The crucial feature of this expansion is the mixed term

−445t2P2(x)P2(y).-\frac{4}{45}t^2P_2(x)P_2(y).

Differentiating, we obtain

∂x∂yut=−45t2xy+O(t4).(27)\partial_x\partial_yu_t=-\frac{4}{5}t^2xy+O(t^4). \tag*{(27)}

At the point (x,y)=(1/2,1/2)(x,y)=(1/2,1/2),

∂x∂yut(1/2,1/2)=−15t2+O(t4),\partial_x\partial_yu_t(1/2,1/2)=-\frac{1}{5}t^2+O(t^4),

which is nonzero for all sufficiently small t≠0t\ne0. Since the mixed Hessian of any potential of the form

u(x,y)=u1(x)+u2(y)+a(x,y),a∈Aff⁡(P),u(x,y)=u_1(x)+u_2(y)+a(x,y),\qquad a\in\operatorname{Aff}(P),

vanishes identically, this proves that the deformed potentials utu_t do not split. We summarize the construction in the following theorem.

Theorem 4.2 (Perturbative non-splitting). Let P=[−1,1]2P=[-1,1]^2 be the standard labeled moment polytope of CP1×CP1\mathbb{CP}^1\times\mathbb{CP}^1, and let

Bt=t(01−10).B_t=t\begin{pmatrix}0&1\\-1&0\end{pmatrix}.

For sufficiently small ∣t∣|t|, there exists a smooth family of admissible toric generalized constant-scalar-curvature Kähler potentials ut∈SBt(P)u_t\in S_{B_t}(P) satisfying

ut=0=u0,SBt(ut)=4.u_{t=0}=u_0,\qquad S_{B_t}(u_t)=4.

With the affine normalization described above, this family has the expansion (26). In particular,

Hess⁡xyut≠0\operatorname{Hess}_{xy}u_t\ne0

for every sufficiently small nonzero tt. Thus the associated symplectic potentials do not split with respect to the prescribed product decomposition.

4.2.4. An exact obstruction to splitting. The preceding calculation proves non-splitting for sufficiently small parameters by exhibiting a nonzero mixed term in the perturbation expansion. We now prove a stronger statement: for any nonzero mixed parameter, no splitting potential can solve the generalized cscK equation, provided it satisfies the admissibility conditions.

Proposition 4.3 (Exact non-splitting obstruction). Let

Bt=t(01−10),t≠0,B_t=t\begin{pmatrix}0 & 1 \\ -1 & 0\end{pmatrix}, \qquad t\ne0,

on the standard labeled square P=[−1,1]2P=[-1,1]^2. There is no admissible symplectic potential of the form

u(x,y)=f1(x)+f2(y)u(x,y)=f_1(x)+f_2(y)

modulo affine-linear functions satisfying

SBt(u)=4.S_{B_t}(u)=4.

Proof. Suppose that such a potential exists. Set

a(x)=1f1′′(x),b(y)=1f2′′(y).a(x)=\frac{1}{f_1^{\prime\prime}(x)}, \qquad b(y)=\frac{1}{f_2^{\prime\prime}(y)}.

The Guillemin boundary conditions imply that aa and bb extend smoothly to the closed intervals, are positive in their interiors, and satisfy

a(±1)=0,a′(−1)=2,a′(1)=−2,a(\pm1)=0,\qquad a'(-1)=2,\qquad a'(1)=-2,
b(±1)=0,b′(−1)=2,b′(1)=−2.b(\pm1)=0,\qquad b'(-1)=2,\qquad b'(1)=-2.

Indeed, for example,

f1′′(x)=11−x2+h1′′(x),f_1^{\prime\prime}(x)=\frac{1}{1-x^2}+h_1^{\prime\prime}(x),

where h1h_1 is smooth on [−1,1][-1,1]. Consequently,

a(x)=1−x21+(1−x2)h1′′(x),a(x)=\frac{1-x^2}{1+(1-x^2)h_1^{\prime\prime}(x)},

which gives the claimed boundary behavior.

By admissibility,

1−t2a(x)b(y)>01-t^2a(x)b(y)>0

in the interior. The real part of the inverse Hermitian matrix is

XBt(u)=11−t2ab(a00b).X_{B_t}(u)=\frac{1}{1-t^2ab}\begin{pmatrix}a & 0 \\ 0 & b\end{pmatrix}.

Hence the generalized scalar curvature equation takes the form

−∂x2(a1−t2ab)−∂y2(b1−t2ab)=4.-\partial_x^2\left(\frac{a}{1-t^2ab}\right)-\partial_y^2\left(\frac{b}{1-t^2ab}\right)=4.

We evaluate this identity on the edge y=1y=1. Since b(1)=0b(1)=0, the first term reduces to −a′′(x)-a^{\prime\prime}(x). Moreover,

∂y2(b1−t2ab)∣y=1=b′′(1)+2t2a(x)(b′(1))2.\left.\partial_y^2\left(\frac{b}{1-t^2ab}\right)\right|_{y=1}=b^{\prime\prime}(1)+2t^2a(x)(b'(1))^2.

Using b′(1)=−2b'(1)=-2, we obtain

a′′(x)+8t2a(x)=−4−b′′(1).(28)a^{\prime\prime}(x)+8t^2a(x)=-4-b^{\prime\prime}(1). \tag*{(28)}

Similarly, evaluating the equation on the edge x=1x=1 gives

b′′(y)+8t2b(y)=−4−a′′(1).(29)b^{\prime\prime}(y)+8t^2b(y)=-4-a^{\prime\prime}(1). \tag*{(29)}

Integrating (28) over [−1,1][-1,1], and using

a′(1)−a′(−1)=−4,a'(1)-a'(-1)=-4,

we find

b′′(1)=−2−4t2∫−11a(x) dx.b^{\prime\prime}(1)=-2-4t^2\int_{-1}^{1}a(x)\,dx.

Similarly,

a′′(1)=−2−4t2∫−11b(y) dy.a^{\prime\prime}(1)=-2-4t^2\int_{-1}^{1}b(y)\,dy.

On the other hand, evaluating (28) at x=1x=1, where a(1)=0a(1)=0, yields

a′′(1)+b′′(1)=−4.a^{\prime\prime}(1)+b^{\prime\prime}(1)=-4.

Combining these three identities gives

t2(∫−11a(x) dx+∫−11b(y) dy)=0.t^2\left(\int_{-1}^{1}a(x)\,dx+\int_{-1}^{1}b(y)\,dy\right)=0.

Since a,b>0a,b>0 in the interior and t≠0t\ne0, this is impossible. The desired contradiction proves the proposition. □

Remark 4.4. Theorem 4.2 produces exact generalized cscK structures for all sufficiently small nonzero tt, while (26) describes only their leading deformation terms. The proposition shows that non-splitting holds independently of this asymptotic expansion, for every nonzero parameter at which an admissible cscGK solution exists.

The construction does not supply a closed formula for utu_t or a numerical lower bound for the existence interval. Moreover, varying BtB_t changes the holomorphic Poisson tensor, so the resulting structures generally belong to different generalized Kähler classes.

References

Hongnian Huang, Department of Mathematics and Statistics, University of New Mexico, Albuquerque, NM 87131, USA

Email address: [email protected]

Li Sheng, Department of Mathematics, Sichuan University, Chengdu 610065, Sichuan, China

Email address: [email protected]

References

  1. [1]Vestislav Apostolov and Hongnian Huang. “A Splitting Theorem for Extremal Kähler Metrics”. In: Journal of Geometric Analysis 25.1 (2015), pp. 149–170.
  2. [2]Vestislav Apostolov, Jeffrey Streets, and Yury Ustinovskiy. “The Riemannian and symplectic geometry of the space of generalized Kähler structures”. In: (2023). arXiv: 2302 . 07314 [math.DG].
  3. [3]Jr. Gates S. James, C. M. Hull, and M. Roček. “Twisted Multiplets and New Supersymmetric Non-linear σ-Models”. In: Nuclear Physics B 248.1 (1984), pp. 157–186.
  4. [4]Matthew Gibson and Jeffrey Streets. “Deformation Classes in Generalized Kähler Geometry”. In: Complex Manifolds 7 (2020), pp. 241–256. DOI: 10.1515/come-2020-0101.arxiv.org/abs/2005.03062
  5. [5]Marco Gualtieri. “Generalized Complex Geometry”. PhD thesis. University of Oxford, 2004. arXiv: math/0401221.DOI
  6. [6]Marco Gualtieri. “Generalized Kähler geometry”. In: Communications in Mathematical Physics 331 (2014), pp. 297–331.
  7. [7]Marco Gualtieri. “Generalized Kähler Metrics from Hamiltonian Deformations”. In: Geometry and Physics: Volume II. 2018.arxiv.org/abs/1807.09704
  8. [8]Nigel Hitchin. “Generalized Calabi–Yau Manifolds”. In: The Quarterly Journal of Mathematics 54.3 (2003), pp. 281–308.
  9. [9]Nigel Hitchin. “Instantons, Poisson Structures and Generalized Kähler Geometry”. In: Communications in Mathematical Physics 265.1 (2006), pp. 131–164.DOI
  10. [10]Hongnian Huang. “A Splitting Theorem on Toric Manifolds”. In: Mathematical Research Letters 20.2 (2013), pp. 273–278.DOI
  11. [11]Claude LeBrun and Santiago R. Simanca. “Extremal Kähler metrics and complex deformation theory”. In: Geometric and Functional Analysis 4 (1994), pp. 298–336.DOI

Paper details

vlnc:W_bf8kgggk
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 8

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

AI-assisted tools

View statement
Looking for related papers…
Contents