betaFeedback

WorksMathematics

Lagrangian structures, rr-matrices and Poisson brackets

Abstract

We give a global conceptual viewpoint of two constructions of Poisson structures involving rr-matrices: the first one concerns coupling of generalized dynamical rr-matrices with actions of quasi-triangular Lie bialgebras, the second one is given by the mixed products of Poisson spaces, which combine two Poisson actions of a quasi-triangular Lie bialgebra to create a new one. When the input Lie algebroids are integrable, we construct symplectic groupoids for both types of Poisson structures by taking fibred products of Lie groupoid morphisms equipped with shifted lagrangian structures. We illustrate our results with explicit examples coming from Lie theory and moduli spaces of flat bundles over surfaces.

Introduction

Moment maps are central in symplectic geometry, given their numerous applications to reduction, dynamics, convexity and equivariant cohomology. In Poisson geometry, the very concept of moment map has been extended in order to accommodate various remarkable examples of possibly Lie group or homogeneous space valued moment maps [1, 2, 15]. Given the non linear nature of these generalized moment maps, their existence and equivariance may not be treated by simply using Lie algebra cohomology as in the case of Lie group actions by symplectomorphisms. So it is remarkable that the Lie theoretic approach to Poisson geometry provides a unifying theory for a very wide family of such non linear moment maps.1

In fact, Poisson manifolds may be regarded as infinitesimal objects, in the sense that their cotangent bundles are naturally equipped with a Lie algebroid structure. The underlying Lie bracket is determined by the Poisson bracket {⋅,⋅}\{\cdot,\cdot\} via the formula

[df,dg]=d{f,g}.[df,dg]=d\{f,g\}.

for all pairs of functions ff, gg on the manifold; in fact, we may extend this expression to all 1-forms by using the Leibniz rule. The global objects corresponding to Poisson manifolds are known as symplectic groupoids, these are symplectic manifolds possessing a compatible Lie groupoid structure, encoding the Lie bracket described above. Because of this reason, symplectic groupoids are then termed integrations of Poisson manifolds. It is by looking at these objects that we obtain a very general theory of equivariant moment maps suitable for performing moment map reduction in most important situations.

As a first instance of this phenomenon, we have that Poisson manifolds equipped with Poisson actions of Poisson Lie groups give rise, upon integration, to Hamiltonian actions. For example, let GG be a connected complete Poisson Lie group and G∗G^{*} its simply connected dual group. A Poisson action of GG on an integrable Poisson manifold (M,πM)(M,\pi_{M}) lifts to a Hamiltonian action on its source-simply connected symplectic groupoid Σ(M)\Sigma(M), with moment map

J:Σ(M)⟶G∗J:\Sigma(M)\longrightarrow G^{*}

[8 Thm. 1]. In particular, for an action by Poisson diffeomorphisms we obtain a classical moment map with values in the dual of a Lie algebra [9 Thm. 1.1].

This picture reflects the fact that a Poisson action of a Lie bialgebra (g,g∗)(\mathfrak{g},\mathfrak{g}^{*}) on (M,πM)(M,\pi_{M}) may be encoded in a Lie algebroid morphism

TπM∗M⟶g∗T_{\pi_{M}}^{*}M\longrightarrow\mathfrak{g}^{*}

given by the transpose of the infinitesimal action in the convention of [8 Prop. 1.8]. In the convention of [5 Ex. 3.19] used below, this morphism is −ρ∗-\rho^{*}, where ρ∗\rho^{*} is the transpose of the infinitesimal action ρ:M×g→TM\rho:M\times\mathfrak{g}\to TM. Here TπM∗MT_{\pi_{M}}^{*}M denotes the cotangent Lie algebroid, with anchor α↦πM♯α\alpha\mapsto\pi_{M}^{\sharp}\alpha. We may additionally encode the g\mathfrak{g}-action on MM in a Lie algebroid structure on AM=T∗M⊕(M×g)A_{M}=T^{*}M\oplus(M\times\mathfrak{g}), together with a Lie algebroid morphism to the Drinfeld double d=g⋈g∗\mathfrak{d}=\mathfrak{g}\bowtie\mathfrak{g}^{*} of the Lie bialgebra [16]. The anchor a:AM→TMa:A_{M}\to TM and the morphism φ:AM→d\varphi:A_{M}\to\mathfrak{d} are given by

a(α,u)=πM♯α+ρ(u),φ(α,u)=(u,−ρ∗α).a(\alpha,u)=\pi_{M}^{\sharp}\alpha+\rho(u),\qquad\varphi(\alpha,u)=(u,-\rho^{*}\alpha).

From this viewpoint, the global counterpart of φ\varphi, when AMA_{M} is integrable, is a multiplicative DD-valued moment map

Φ:H⟶D,\Phi:\mathcal{H}\longrightarrow D,

where H⇒M\mathcal{H}\Rightarrow M is the source-simply connected integration of AMA_{M}, DD integrates d\mathfrak{d}, and Φ\Phi is a Lie groupoid morphism which lifts φ\varphi and is g×g\mathfrak{g}\times\mathfrak{g}-equivariant [5 Thm. 5.3]. The groupoid H\mathcal{H} carries a two-form β\beta that fails to be a symplectic form compatible with H\mathcal{H} in a way controlled by the moment map Φ\Phi. Indeed, letting ΩD\Omega_{D} and ΘD\Theta_{D} be, respectively, the Polyakov–Wiegmann two-form on D×DD\times D and the Cartan three-form on DD, β\beta satisfies

ker⁡β∩ker⁡TΦ=0,pr⁡1∗β+pr⁡2∗β−mH∗β=(Φ×Φ)∗ΩD,dβ=−Φ∗ΘD,\ker\beta\cap\ker T\Phi=0,\qquad\operatorname{pr}_{1}^{*}\beta+\operatorname{pr}_{2}^{*}\beta-m_{\mathcal{H}}^{*}\beta=(\Phi\times\Phi)^{*}\Omega_{D},\qquad d\beta=-\Phi^{*}\Theta_{D},

where mHm_{\mathcal{H}} is the groupoid multiplication and pr⁡i\operatorname{pr}_{i} are the projections from the space of composable pairs. In the modern language of shifted symplectic geometry, β\beta is an example of a (2(2-shifted) Lagrangian structure on Φ\Phi [22]. The main utility of such DD-valued moment maps (and of Lagrangian structures in general) is in performing a generalized version of symplectic reduction involving fibred products and quotients by Lie group actions, which accounts for the construction of many symplectic groupoids of interest [5].

Main results

In this note, we consider two further constructions of Poisson manifolds that may be put into the framework of such a theory of multiplicative DD-valued moment maps. These are both associated with the Lie bialgebras (g,g∗)(\mathfrak{g},\mathfrak{g}^{*}) determined by quasi-triangular rr-matrices: these are elements in g⊗g\mathfrak{g} \otimes\mathfrak{g} whose respective symmetric and skew-symmetric parts ss and Λ\Lambda satisfy that ss is g\mathfrak{g}-invariant and both are linked by the classical Yang-Baxter equation. The key fact underlying this work is that ss determines a distinguished lagrangian Lie subalgebra

Rm⊂d×d×d‾,R_m \subset\mathfrak{d} \times\mathfrak{d} \times\overline{\mathfrak{d}},

where the overline indicates the opposite pairing on the corresponding factor. As we shall see, this lagrangian Lie subalgebra is responsible, firstly, for the relationship between Poisson brackets and generalized dynamical rr-matrices pointed out in [24]. Indeed, a generalized dynamical rr-matrix over a Poisson manifold (N,πN)(N,\pi_N) may also be encoded as a Lie algebroid

AN=T∗N⊕(N×g∗),A_N = T^{*}N \oplus(N \times\mathfrak{g}^{*}),

equipped with a distinguished Lie algebroid morphism to d\mathfrak{d} [12, 13]. Therefore, one can combine this Lie algebroid with the one coming from a Poisson action of (g,g∗)(\mathfrak{g},\mathfrak{g}^{*}) on a Poisson manifold (M,πM)(M,\pi_M) to produce a new Poisson bracket, see Proposition 2.1. This construction gives rise to interesting examples, as pointed out already in [19, 24], it includes suitably gauge-fixed Poisson structures on moduli spaces of flat bundles over surfaces. Secondly, we consider the mixed product Poisson structures of [17] and we show that they are also determined by suitable fibred products of Lie algebroids involving RmR_m. The main observation of this work is that then both constructions may be readily promoted to global ones in terms of DD-valued moment maps.

Theorem 1.1. Let s+Λs+\Lambda be a quasitriangular rr-matrix on g\mathfrak{g}, and let AMA_M be the Lie algebroid associated with a Poisson action of the resulting Lie bialgebra on (M,πM)(M,\pi_M). Suppose that AMA_M and the Lie algebroid ANA_N in either of the following cases are integrable.

  1. If ANA_N is associated with a generalized dynamical rr-matrix on (N,πN)(N,\pi_N) with the twist determined by ss, then the coupled Poisson structure (3) on M×NM \times N is integrable.

  1. If ANA_N is associated with a second Poisson action of the same Lie bialgebra on (N,πN)(N,\pi_N), then the mixed product (4) on M×NM \times N is integrable.

Moreover, in both cases there is an explicit construction of the integration given as follows. Let ΦM:HM→D\Phi_M:\mathcal{H}_M \to D and ΦN:HN→D\Phi_N:\mathcal{H}_N \to D be the Lie groupoid morphisms from the respective source-simply connected integrations of AMA_M and ANA_N determined by the above mentioned Lie algebroid morphisms. This construction relies crucially on a Lie 2-group structure on DD integrating the double, which is guaranteed to exist if DD is simply-connected. In fact, if DD is simply-connected, it turns out that the Lie algebra RmR_m may be promoted to a Lagrangian Lie subgroup of D3D^3 which is also the graph of a groupoid multiplication mD:D⇒Gm_D:D \Rightarrow G, see §3. In the dynamical case, let JL:L→DJ_L:L \to D be the unit map G→DG \to D. In the mixed case, let L=G∗L=G^{*} be the simply connected dual group and let JLJ_L integrate the inclusion of its Lie algebra in d\mathfrak{d}. In either case, the arrows of a symplectic groupoid integrating the indicated bivector are the triples (hM,hN,ℓ)∈HM×HN×L(h_M,h_N,\ell) \in\mathcal{H}_M \times\mathcal{H}_N \times L for which ΦM(hM)\Phi_M(h_M) and ΦN(hN)\Phi_N(h_N) are composable and

mD(ΦM(hM),ΦN(hN))=JL(ℓ).m_D\bigl(\Phi_M(h_M),\Phi_N(h_N)\bigr)=J_L(\ell).

If βM,βN\beta_M,\beta_N are the lagrangian two-forms on HM,HN\mathcal{H}_M,\mathcal{H}_N, the symplectic form is the pullback of pr⁡1∗βM+pr⁡2∗βN\operatorname{pr}_1^*\beta_M+\operatorname{pr}_2^*\beta_N under the projection (hM,hN,ℓ)↦(hM,hN)(h_M,h_N,\ell)\mapsto(h_M,h_N). Schematically, our integration takes the form outlined in Figure 1.

In order to illustrate this result with concrete examples, we revisit a simple construction of dynamical Poisson groupoids studied in [10], and we recover the integration of configuration

Figure 1. Integration of Poisson structures coming from rr-matrices, the arrows labelled by Lie indicate differentiation

spaces of flags by means of decorated moduli spaces conducted in [3]. For comparison, note that Mouquin constructed local symplectic groupoids over mixed product Poisson manifolds [21 Thm. 5.4 and Rem. 5.6]. Our construction refines that observation by giving a corresponding global symplectic groupoid when the two extended Lie algebroids encoding the Poisson action are integrable, without assuming completeness of the dressing actions.

This paper is organized as follows. In §2, we carry out the description of both procedures in terms of Lie algebroids and their fibred products. In § 3, we lift both constructions to the level of Lie groupoids, and we explain how they may give rise to new symplectic groupoids, by using the fact that a transverse intersection of two 2-shifted lagrangians in a 2-shifted symplectic target is 1-shifted symplectic [22 Thm. 2.9], (see [5 Thm. 3.23] for a differential-geometric version of this result). We illustrate our results with explicit symplectic groupoids in both cases.

Acknowledgments. The author thanks H. Bursztyn, M. Cueca and M. Mol for stimulating conversations related to this work.

Poisson brackets from generalized dynamical rr-matrices and mixed products

The central object in this work is given by a Lie algebra equipped with a quasi-triangular rr-matrix. Such objects give rise to a special kind of Lie bialgebra which includes most important examples and, most crucially for us, allows us to perform a classical version of fusion for quantum group representations. This fusion underlies the mixed products of [17] and, as we shall see, the construction of Poisson structures from generalized dynamical rr-matrices in [24].

Let g\mathfrak{g} be a Lie algebra equipped with a quasitriangular rr-matrix s+Λ∈g⊗gs+\Lambda\in\mathfrak{g}\otimes\mathfrak{g}. Here s∈(S2g)gs\in(S^2\mathfrak{g})^{\mathfrak{g}} is its invariant symmetric part, which is possibly degenerate, Λ∈∧2g\Lambda\in\wedge^2\mathfrak{g} is its skew part, and s+Λs+\Lambda solves the classical Yang–Baxter equation:

12[Λ,Λ]=−χ,\frac{1}{2}[\Lambda,\Lambda]=-\chi,

where we write S=s♯:g∗→gS=s^\sharp:\mathfrak{g}^*\to\mathfrak{g} for contraction in the first factor and χ∈∧3g\chi\in\wedge^3\mathfrak{g} is given by

χ(ξ,η,ζ)=⟨ζ,[Sξ,Sη]⟩.\chi(\xi,\eta,\zeta)=\langle\zeta,[S\xi,S\eta]\rangle.

We may use χ\chi to define a Lie algebra structure on d=g⊕g∗\mathfrak{d}=\mathfrak{g}\oplus\mathfrak{g}^*, the double of (g,0,χ)(\mathfrak{g},0,\chi). The Lie bracket is given by

[(u,ξ),(v,η)]=([u,v]−ιξιηχ,ad⁡u∗η−ad⁡v∗ξ),[(u,\xi),(v,\eta)]=([u,v]-\iota_\xi\iota_\eta\chi,\operatorname{ad}^{*}_{u}\eta-\operatorname{ad}^{*}_{v}\xi),

where ad⁡w∗ζ(z)=−ζ([w,z])\operatorname{ad}^{*}_{w}\zeta(z)=-\zeta([w,z]) and its pairing is

⟨(u,ξ),(v,η)⟩=ξ(v)+η(u).\langle(u,\xi),(v,\eta)\rangle=\xi(v)+\eta(u).

The dual Lie algebra of the Lie bialgebra defined by s+Λs+\Lambda sits in this splitting as the lagrangian Lie subalgebra

c={(Λ♯ξ,ξ):ξ∈g∗}.(1)\mathfrak{c}=\{(\Lambda^\sharp\xi,\xi):\xi\in\mathfrak{g}^*\}. \tag*{(1)}

The fusion relation on the double

The double d\mathfrak{d} admits a groupoid structure d⇉g\mathfrak{d} \rightrightarrows\mathfrak{g} compatible with its Lie algebra structure. We describe it following the conventions of [11 Ex. 1.8, (1.9)]. Its source, target and unit are

p−(u,ξ)=u−Sξ,p+(u,ξ)=u+Sξ,e(u)=(u,0).p_{-}(u,\xi)=u-S\xi,\qquad p_{+}(u,\xi)=u+S\xi,\qquad e(u)=(u,0).

Two arrows d1,d2d_{1},d_{2} are composable when p−(d1)=p+(d2)p_{-}(d_{1})=p_{+}(d_{2}). Their multiplication, in this order, is

m(d1,d2)=d1+d2−e(p−(d1)).m(d_{1},d_{2})=d_{1}+d_{2}-e\bigl(p_{-}(d_{1})\bigr).

If we let RmR_{m} be the graph of this multiplication, we have that it is a lagrangian Lie subalgebra of d×d×d‾\mathfrak{d}\times\mathfrak{d}\times\overline{\mathfrak{d}}, where the bar indicates that the pairing is taken with an opposite sign. Write p:Rm→d2p:R_{m}\to\mathfrak{d}^{2} for projection to the pair of composable elements and q:Rm→dq:R_{m}\to\mathfrak{d} for the projection to the output of the multiplication.

For example, if ss is nondegenerate, the map (p+,p−):(u,ξ)↦(u+Sξ,u−Sξ)(p_{+},p_{-}):(u,\xi)\mapsto(u+S\xi,u-S\xi) identifies d⇉g\mathfrak{d}\rightrightarrows\mathfrak{g} with the pair groupoid g×g⇉g\mathfrak{g}\times\mathfrak{g}\rightrightarrows\mathfrak{g}. In terms of its natural coordinates, we have that (x,y)(x,y) is an arrow from yy to xx, the unit at xx is (x,x)(x,x), and the multiplication is (x,y)⋅(y,z)=(x,z)(x,y)\cdot(y,z)=(x,z).

The Lie algebra RmR_{m} gives us the ability to carry out the following operation. Suppose Ai→XiA_{i}\to X_{i}, i=1,2i=1,2, are Lie algebroids equipped with Lie algebroid morphisms φi:Ai→d\varphi_{i}:A_{i}\to\mathfrak{d}. For a lagrangian Lie subalgebra l\mathfrak{l} with inclusion j:l→dj:\mathfrak{l}\to\mathfrak{d}, consider the following two fibred products of Lie algebroids over X1×X2X_{1}\times X_{2} which we assume to be transverse:

Figure 2.2.

Here φfus\varphi_{\mathrm{fus}} is the upper horizontal map in the first square followed by qq. The first square serves to multiply the target elements of φ1×φ2\varphi_{1}\times\varphi_{2}; on the other hand, the second square further constrains the multiplication of a pair of elements to lie in l\mathfrak{l}. In the two examples below, we shall study special situations of such fibred products in which we may identify AlA_{\mathfrak{l}} with the graph of a Poisson tensor.

Generalized dynamical r-matrices and Poisson brackets

In mathematical physics, classical dynamical rr-matrices occur as semiclassical limits of dynamical quantum RR-matrices, including Felder’s elliptic examples [7]. This led to an explicit relationship with Lie theory and Poisson geometry. In [13, 14] the dynamical equation was generalized to the case of a Poisson manifold as base, which is the situation relevant for us.

Let (N,πN)(N,\pi_{N}) be Poisson. Choose a basis (ei)i(e_{i})_{i} of g\mathfrak{g}. A generalized dynamical datum with the twist fixed above by χ\chi consists of a tensor ϑ=∑iXi⊗ei\vartheta=\sum_{i}X_{i}\otimes e_{i}, with XiX_{i} vector fields on NN, and a smooth map r:N→∧2gr:N\to\wedge^{2}\mathfrak{g}. Write b=ϑ∗:T∗N→gb=\vartheta^{*}:T^{*}N\to\mathfrak{g}, then we have that bα=∑iα(Xi)eib_{\alpha}=\sum_{i}\alpha(X_{i})e_{i}. Consider the Lie algebroid E=TN⊕(N×g)E=TN\oplus(N\times\mathfrak{g}) with anchor the projection to TNTN, and its bracket on sections defined by

[(X,u),(Y,v)]E=([X,Y],dv(X)−du(Y)+[u,v]g).[(X,u),(Y,v)]_{E}=([X,Y],\mathrm{d}v(X)-\mathrm{d}u(Y)+[u,v]_{\mathfrak{g}}).

where u,v:N→gu,v:N\to\mathfrak{g} and [u,v]g[u,v]_{\mathfrak{g}} is the pointwise Lie bracket [24 (28)], with vector fields acting on g\mathfrak{g}-valued functions by differentiation. We use the same notation for the induced Schouten bracket on Γ(∧∙E)\Gamma(\wedge^{\bullet}E). Put R=πN+∑iXi∧ei+r\mathcal{R}=\pi_{N}+\sum_{i}X_{i}\wedge e_{i}+r. The compatibility condition for these data is

12[R,R]E=−χ.\frac{1}{2}[\mathcal{R},\mathcal{R}]_{E}=-\chi.

In particular, it follows that b:TπN∗N→gb:T^{*}_{\pi_{N}}N\to\mathfrak{g} is a Lie algebroid morphism [14 Def. 4.4 and Thm. 4.5]. Moreover, the dynamical datum defines a Lie algebroid AN=T∗N⊕(N×g∗)A_{N}=T^{*}N\oplus(N\times\mathfrak{g}^{*}) and a Lie algebroid morphism

φN:AN→d,φN(α,ξ)=(bα+r♯ξ,ξ).\varphi_{N}:A_{N}\to\mathfrak{d},\qquad\varphi_{N}(\alpha,\xi)=(b\alpha+r^{\sharp}\xi,\xi).

Similarly, we may construct a Lie algebroid from a Poisson action. Let (M,πM)(M,\pi_M) be a Poisson manifold with a Poisson action of the Lie bialgebra with cobracket δ(u)=[u,Λ]\delta(u)=[u,\Lambda] determined by s+Λ\mathfrak{s}+\Lambda. We denote by ρ:g→X(M)\rho:\mathfrak{g}\to\mathfrak{X}(M) the corresponding action. In the convention of [5 Ex. 3.19], the Poisson action condition is

Lρ(u)πM=−ρ(δ(u)),u∈g.\mathcal{L}_{\rho(u)}\pi_M=-\rho(\delta(u)),\qquad u\in\mathfrak{g}.

The associated Lie algebroid AM=T∗M⊕(M×g)A_M=T^*M\oplus(M\times\mathfrak{g}) has anchor aM(β,u)=πM♯β+ρ(u)a_M(\beta,u)=\pi_M^\sharp\beta+\rho(u). In the splitting d=g⊕g∗\mathfrak{d}=\mathfrak{g}\oplus\mathfrak{g}^* used above, we obtain a Lie algebroid morphism to d\mathfrak{d} given by

φM:AM⟶d,φM(β,u)=(u−Λ♯ρ∗β,−ρ∗β).\varphi_M:A_M\longrightarrow\mathfrak{d},\qquad\varphi_M(\beta,u)=(u-\Lambda^\sharp\rho^*\beta,-\rho^*\beta).

where we denote by ρ∗:T∗M→g∗\rho^*:T^*M\to\mathfrak{g}^* the fibrewise transpose of the action map. We may describe both Lie algebroids uniformly by embedding them as Dirac structures. Denote by TX=TX⊕T∗X\mathbb{T}X=TX\oplus T^*X the standard Courant algebroid of a manifold XX. The embeddings of AMA_M into TM×d\mathbb{T}M\times\mathfrak{d} and ANA_N into TN×d\mathbb{T}N\times\mathfrak{d} take the following form, see [5 (43), (45)]:

(β,u)⟼((πM♯β+ρ(u),β),φM(β,u)),(\beta,u)\longmapsto((\pi_M^\sharp\beta+\rho(u),\beta),\varphi_M(\beta,u)),
(α,ξ)⟼((πN♯α−b∗ξ,α),φN(α,ξ)).(2)(\alpha,\xi)\longmapsto((\pi_N^\sharp\alpha-b^*\xi,\alpha),\varphi_N(\alpha,\xi)). \tag*{(2)}

The Lie brackets on AMA_M and ANA_N are then obtained by restriction of the background Courant brackets, and the anchors are their tangent projections. For generalized dynamical rr-matrices, this Dirac description goes back to [12], see also [5 Ex. 3.20]. Applying construction (2.2) to these Lie algebroids, we recover the Poisson structure of [24 Thm. 2.9]. For a general interpretation of such fibred products of Lie algebroids in terms of Dirac reduction see [5].

Proposition 2.1. Take (A1,A2)=(AM,AN)(A_1,A_2)=(A_M,A_N) as in (2) and l=e(g)\mathfrak{l}=e(\mathfrak{g}) in (2.2). The resulting Lie algebroid is the cotangent algebroid of the Poisson bivector

Π=Q+πN−∑iρ(ei)∧Xi+ρ(r)on M×N.(3)\Pi=Q+\pi_N-\sum_i\rho(e_i)\wedge X_i+\rho(r)\qquad\text{on }M\times N. \tag*{(3)}

Here Q=πM+ρ(Λ)Q=\pi_M+\rho(\Lambda).

Proof. Put dM=φM(β,u)d_M=\varphi_M(\beta,u) and dN=φN(α,ξ)d_N=\varphi_N(\alpha,\xi). The first fibred product in (2.2) imposes p−(dM)=p+(dN)p_-(d_M)=p_+(d_N). The second fibred product requires m(dM,dN)m(d_M,d_N) to be a unit, so the two arrows dMd_M and dNd_N are inverses, which implies that their g∗\mathfrak{g}^*-components sum to zero. Therefore, we have that

ξ=ρ∗β,u=bα+(r+Λ)♯ρ∗β.\xi=\rho^*\beta,\qquad u=b\alpha+(r+\Lambda)^\sharp\rho^*\beta.

Consequently, each covector (β,α)∈T∗(M×N)(\beta,\alpha)\in T^*(M\times N) has a unique lift through the two fibred products in (2.2). Its tangent projection is then given by the anchor

(πM♯β+ρ(bα+(r+Λ)♯ρ∗β), πN♯α−b∗ρ∗β)=Π♯(β,α).(\pi_M^\sharp\beta+\rho(b\alpha+(r+\Lambda)^\sharp\rho^*\beta),\ \pi_N^\sharp\alpha-b^*\rho^*\beta)=\Pi^\sharp(\beta,\alpha).

The resulting Lie algebroid is therefore isomorphic to Gr⁡(Π♯)\operatorname{Gr}(\Pi^\sharp). Indeed, projection to (β,α)(\beta,\alpha) gives the Lie algebroid isomorphism

Ae(g)=((AM×AN)×d2Rm)×de(g)≅TΠ∗(M×N).A_{e(\mathfrak{g})}=((A_M\times A_N)\times_{\mathfrak{d}_2}R_m)\times_{\mathfrak{d}}e(\mathfrak{g})\cong T^*_\Pi(M\times N).

The flip N×M→M×NN\times M\to M\times N takes the Poisson structure in [24 Thm. 2.9, (14)] to Π\Pi.

Example. A gauge-fixed Chern–Simons bracket

We shall illustrate how Proposition 2.1 gives rise to interesting examples associated to the Fock–Rosly Poisson structure on moduli spaces of flat bundles over surfaces, following [19, 24]. Let G=SO+(2,1)⋉R3G = \mathrm{SO}_{+}(2,1) \ltimes\mathbb{R}^{3}, and consider an oriented connected surface Σ\Sigma of genus gg with n≥3n \ge3 punctures. Choose a conjugacy class Ci⊂GC_i \subset G for the holonomy at puncture ii. On the regular locus of P=C1×C2P = C_1 \times C_2, take a local slice N⊂PN \subset P where simultaneous conjugation is locally free. The remaining holonomies of a representation of π1(Σ)\pi_1(\Sigma) belong then to

M=C3×⋯×Cn×G2g.M = C_3 \times\cdots\times C_n \times G^{2g}.

Let ρP\rho_P denote infinitesimal conjugation on PP, and let pN:TP∣N→TNp_N : TP|_N \to TN be the projection along the conjugation orbits. We have that PP inherits a canonical bivector field QPQ_P given as follows. Choose a basis (ea)(e_a) of g\mathfrak{g} and write the symmetric part of the quasitriangular tensor as s=∑a,bsabea⊗ebs = \sum_{a,b}s^{ab}e_a \otimes e_b, where 2s2s is the inverse tensor of the Chern–Simons pairing. For u∈gu \in\mathfrak{g}, let uLu^L and uRu^R be its respective left and right invariant extensions. In what follows, a superscript (i)(i) denotes the lift to the ii-th factor of G×GG \times G. Thus ρi(u)=(uL−uR)(i)\rho_i(u) = (u^L-u^R)^{(i)} generates the action by conjugation on that factor. The canonical (quasi-Poisson) bivector field on PP is

QP=∑a,bsab(eaR(1)∧ebL(1)+eaR(2)∧ebL(2)+ρ1(ea)∧ρ2(eb))∣C1×C2.Q_P = \left.\sum_{a,b}s^{ab}\left(e_a^{R(1)} \wedge e_b^{L(1)} + e_a^{R(2)} \wedge e_b^{L(2)} + \rho_1(e_a) \wedge\rho_2(e_b)\right)\right|_{C_1 \times C_2}.

see [24 Example 2.11, Eq. (15)]. Now we obtain a Poisson structure on the slice via projection πN=(∧2pN)(QP∣N)\pi_N = (\wedge^2 p_N)(Q_P|_N), for the Dirac gauge fixing of [19] we have πN=0\pi_N = 0 [24 Thm. 3.6]. Furthermore, the mixed and orbit components of QPQ_P determine ϑ=∑iXi⊗ei\vartheta= \sum_i X_i \otimes e_i and rr through the decomposition described in [24 Thms. 2.7 and 3.4]:

QP∣N=πN−∑iXi∧ρP(ei)+ρP(r).Q_P|_N = \pi_N - \sum_i X_i \wedge\rho_P(e_i) + \rho_P(r).

We have then that πN\pi_N, ϑ\vartheta and rr give us a generalized dynamical rr-matrix on NN. The Poisson structure πM\pi_M on the space of remaining holonomies M=C3×⋯×Cn×G2gM = C_3 \times\cdots\times C_n \times G^{2g} is given by the Fock–Rosly construction, which is determined by an rr-matrix of the form s−Λs-\Lambda, see [20]. Hence, writing ϑ=∑iXi⊗ei\vartheta= \sum_i X_i \otimes e_i, equation (3) gives the gauge-fixed Poisson bivector on M×NM \times N,

Π=πM+ρ(Λ)+ρ(r)−∑iρ(ei)∧Xi;\Pi= \pi_M + \rho(\Lambda) + \rho(r) - \sum_i \rho(e_i) \wedge X_i;

here rr is evaluated at the point of NN and ρ\rho is the infinitesimal conjugation action on MM [24 Thms. 3.4 and 3.6]. We leave the global study of this example to [4].

The mixed product of two Poisson spaces

Write r=s+Λ=∑axa⊗yar = s + \Lambda= \sum_a x_a \otimes y_a for the fixed quasitriangular tensor. Let (M1,π1)=(M,πM)(M_1,\pi_1) = (M,\pi_M) and (M2,π2)=(N,πN)(M_2,\pi_2) = (N,\pi_N) both be left Poisson spaces for rr, with infinitesimal actions ρ1,ρ2\rho_1,\rho_2. In the left-action convention of [17 §6.3], the mixed product of these Poisson structures is

πmix=π1+π2+∑aρ1(ya)∧ρ2(xa).(4)\pi_{\mathrm{mix}} = \pi_1 + \pi_2 + \sum_a \rho_1(y_a) \wedge\rho_2(x_a). \tag*{(4)}

Consider the corresponding Lie algebroid for each factor Ai=T∗Mi⊕(Mi×g)A_i = T^*M_i \oplus(M_i \times\mathfrak{g}). For each ai=(αi,vi)∈Aia_i = (\alpha_i,v_i) \in A_i, we write ξi=−ρi∗αi\xi_i = -\rho_i^*\alpha_i, where ρi∗\rho_i^* is the transpose of the action. In the splitting d=g⊕g∗\mathfrak{d} = \mathfrak{g} \oplus\mathfrak{g}^* used at the beginning of this section, the Dirac embedding of [5 (43)] is given by

ai⟼((πi♯αi+ρi(vi),αi),(vi+Λ♯ξi,ξi)) in TMi×d.a_i \longmapsto\left((\pi_i^\sharp\alpha_i+\rho_i(v_i),\alpha_i),(v_i+\Lambda^\sharp\xi_i,\xi_i)\right)\ \text{in } TM_i \times\mathfrak{d}.

We abbreviate the component in d\mathfrak{d} by di=ρi(ai)d_i = \rho_i(a_i) in what follows.

Proposition 2.2. Applying the two fibred products in (2.2) successively to A1,A2A_1,A_2 with l=c\mathfrak l=\mathfrak c as in (1) determines the cotangent Lie algebroid of the mixed product (4):

F→pr⁡2A2C→c↓pr⁡1↓p+∘φ2↓↓jA1→p−∘φ1gF→φFd(5)\begin{array}{ccccc} F & \xrightarrow{\operatorname{pr}_2} & A_2 & & C \xrightarrow{} \mathfrak c \\ \downarrow{\operatorname{pr}_1} & & \downarrow{p_+\circ\varphi_2} & & \downarrow\quad\downarrow{j} \\ A_1 & \xrightarrow{p_-\circ\varphi_1} & \mathfrak g & & F \xrightarrow{\varphi_F} \mathfrak d \tag*{(5)} \end{array}

where the projection to the cotangent bundle factor identifies CC with Tπmix∗(M1×M2)T^*_{\pi_{\mathrm{mix}}}(M_1\times M_2).

Proof. Explicitly, we have that FF denotes the Lie algebroid formed by pairs (a1,a2)(a_1,a_2) satisfying p−(d1)=p+(d2)p_-(d_1)=p_+(d_2), this means that:

v2+(Λ♯+S)ξ2=v1+(Λ♯−S)ξ1.v_2+(\Lambda^\sharp+S)\xi_2=v_1+(\Lambda^\sharp-S)\xi_1.

The map φF:F→d\varphi_F:F\to\mathfrak d is then given by φF(a1,a2)=m(d1,d2)\varphi_F(a_1,a_2)=m(d_1,d_2). If we let j:c↪dj:\mathfrak c\hookrightarrow\mathfrak d be the inclusion, we have that C=F×dcC=F\times_{\mathfrak d}\mathfrak c. The condition defining the right square above may be written using the formula

m(d1,d2)=(v+Λ♯(ξ1+ξ2),ξ1+ξ2).m(d_1,d_2)=\bigl(v+\Lambda^\sharp(\xi_1+\xi_2),\xi_1+\xi_2\bigr).

where v=v2−Λ♯ξ1+Sξ1v=v_2-\Lambda^\sharp\xi_1+S\xi_1. Since c\mathfrak c is the graph of Λ♯\Lambda^\sharp, the right square sets v=0v=0. Thus we obtain that

v1=(Λ♯+S)ξ2,v2=(Λ♯−S)ξ1.v_1=(\Lambda^\sharp+S)\xi_2,\qquad v_2=(\Lambda^\sharp-S)\xi_1.

Since each covector (α1,α2)(\alpha_1,\alpha_2) determines a unique element of CC, we may substitute the two values of viv_i in the tangent entries of the Dirac embeddings to obtain that

(π1♯α1−∑aα2(ρ2(xa))ρ1(ya), π2♯α2+∑aα1(ρ1(ya))ρ2(xa)).\left(\pi_1^\sharp\alpha_1-\sum_a\alpha_2(\rho_2(x_a))\rho_1(y_a),\ \pi_2^\sharp\alpha_2+\sum_a\alpha_1(\rho_1(y_a))\rho_2(x_a)\right).

These are exactly the two components of πmix♯(α1,α2)\pi_{\mathrm{mix}}^\sharp(\alpha_1,\alpha_2). Therefore, we may identify CC with Gr⁡(πmix♯)\operatorname{Gr}(\pi_{\mathrm{mix}}^\sharp).

Remark 2.1. The key feature of πmix\pi_{\mathrm{mix}} is that the diagonal g\mathfrak g-action on M1×M2M_1\times M_2 becomes a Poisson action with respect to it [17] and the full Lie algebroid which encodes this action is precisely the fibred product FF above.

The global picture: symplectic groupoids from fibred products

To integrate the two pullback diagrams in (2.2), we first integrate d⇒g\mathfrak d\Rightarrow\mathfrak g. To do so we use the standard integration of crossed modules of Lie algebras. Introduce the new coordinates (x,ξ)=(u−Sξ,ξ)(x,\xi)=(u-S\xi,\xi) in which the source map is xx and the target is x+2Sξx+2S\xi. Expressing the coadjoint action as (ad⁡ξ∗)(y)=−ξ([x,y])(\operatorname{ad}^*_\xi)(y)=-\xi([x,y]), we obtain the crossed module of Lie algebras

∂=2S:k→g,\partial=2S:\mathfrak k\to\mathfrak g,

where k\mathfrak k is the vector space g∗\mathfrak g^* with bracket [ξ,η]k=ad⁡∂ξ∗η[\xi,\eta]_{\mathfrak k}=\operatorname{ad}^*_{\partial\xi}\eta, and g\mathfrak g acts on it by the coadjoint action above. In this expression, the pairing becomes

ξ(y)+η(x)+2s(ξ,η).\xi(y)+\eta(x)+2s(\xi,\eta).

Let K,GK,G be the simply connected groups integrating k,g\mathfrak{k},\mathfrak{g}, and use ∂:K→G\partial:K\to G also for the integrated homomorphism. The semidirect product Lie group D=K⋉GD=K\ltimes G is in addition a groupoid over GG, with source gg, target ∂(k)g\partial(k)g for (k,g)(k,g), and unit ϵ(g)=(1,g)\epsilon(g)=(1,g). The presence of this groupoid structure makes DD into a Lie 2-group, a groupoid object in the category of Lie groups. In particular, if SS is invertible, the target and source identify this groupoid with the pair groupoid G×G⇒GG\times G\Rightarrow G.

Equip DD with its standard 2-shifted symplectic form, see [5 Ex. 3.1] for instance. We let θL,θR\theta^{L},\theta^{R} be the respective left and right Maurer–Cartan forms on DD and define

ΩD=−12⟨pr⁡1∗θL,pr⁡2∗θR⟩∈Ω2(D×D),ΘD=−112⟨θL,[θL,θL]⟩∈Ω3(D).\Omega_D=-\frac{1}{2}\langle\operatorname{pr}_1^*\theta^L,\operatorname{pr}_2^*\theta^R\rangle\in\Omega^2(D\times D),\qquad\Theta_D=-\frac{1}{12}\langle\theta^L,[\theta^L,\theta^L]\rangle\in\Omega^3(D).

We say that ΩD\Omega_D is the Polyakov-Wiegmann 2-form and ΘD\Theta_D is the Cartan 3-form. Let RmD⊂D2×D‾R_{m_D}\subset D^2\times\overline{D} be the graph of the groupoid multiplication mDm_D, written as a relation from D2D^2 to DD; here the bar reverses the sign of the shifted symplectic form above. We have that RmDR_{m_D} is a Lie group with Lie algebra Rm\mathfrak{R}_m. In the terminology of [5], this subgroup is 2-shifted lagrangian with zero two-form: we may readily verify its Lie algebra is lagrangian (see e.g. [11 (1.9)]), and the differential forms induced by ΩD\Omega_D and ΘD\Theta_D on the nerve of D2×D‾D^2\times\overline{D} restrict to zero on (the nerve of) RmDR_{m_D} by isotropy. Here the overline indicates a sign reversal in such differential forms for the corresponding factor.

The second idea needed to carry out the integration of (2.2) is that of lagrangian structures and their corresponding notion of composition, which we summarize from [5]. Let Φ:H→D\Phi:\mathcal{H}\to D be a Lie groupoid morphism, where H⇒X\mathcal{H}\Rightarrow X. Write AHA_{\mathcal{H}} for the Lie algebroid of H\mathcal{H}, a:AH→TXa:A_{\mathcal{H}}\to TX for its anchor, and φ=Lie⁡(Φ):AH→d\varphi=\operatorname{Lie}(\Phi):A_{\mathcal{H}}\to\mathfrak{d} for the induced morphism. A two-form β∈Ω2(H)\beta\in\Omega^2(\mathcal{H}) determines a bundle map λβ:AH→T∗X\lambda_\beta:A_{\mathcal{H}}\to T^*X by

λβ(a)(v)=β1x(a,Txϵ(v)),a∈(AH)x,v∈TxX.\lambda_\beta(a)(v)=\beta_{1_x}\bigl(a,T_x\epsilon(v)\bigr),\qquad a\in(A_{\mathcal{H}})_x,\quad v\in T_xX.

where ϵ:X→H\epsilon:X\to\mathcal{H} is the unit map. We say that β\beta defines a 2-shifted lagrangian structure on Φ\Phi if

pr⁡1∗β+pr⁡2∗β−mH∗β=(Φ×Φ)∗ΩD,−dβ=Φ∗ΘD,\operatorname{pr}_1^*\beta+\operatorname{pr}_2^*\beta-m_{\mathcal{H}}^*\beta=(\Phi\times\Phi)^*\Omega_D,\qquad-d\beta=\Phi^*\Theta_D,

and the bundle map

(a,λβ,φ):AH⟶TX⊕(X×d)(a,\lambda_\beta,\varphi):A_{\mathcal{H}}\longrightarrow TX\oplus(X\times\mathfrak{d})

is fibrewise injective with lagrangian image [5 Prop. 3.9]. The first identity above takes place on the space of composable pairs in H\mathcal{H}, with projections pr⁡1,pr⁡2\operatorname{pr}_1,\operatorname{pr}_2 and multiplication mHm_{\mathcal{H}}. When D={1}D=\{1\}, these conditions say that (H,β)(\mathcal{H},\beta) is a quasi-symplectic groupoid with zero background three-form. One of the main results in [5] is that integrable Dirac structures in TX⊕(X×d)TX\oplus(X\times\mathfrak{d}) give rise upon integration to such lagrangian structures.

We recall how to compose two such lagrangian structures, in analogy with usual lagrangian correspondences in classical symplectic geometry. Let D1,D2,D3D_1,D_2,D_3 be Lie groups equipped with 2-shifted symplectic forms as before, and write Di‾\overline{D_i} for DiD_i equipped with the opposite shifted symplectic form. Suppose that

Φ12=(f1,f2):H12⟶D1‾×D2,Φ23=(g2,g3):H23⟶D2‾×D3\Phi_{12}=(f_1,f_2):\mathcal{H}_{12}\longrightarrow\overline{D_1}\times D_2,\qquad\Phi_{23}=(g_2,g_3):\mathcal{H}_{23}\longrightarrow\overline{D_2}\times D_3

carry lagrangian structures β12\beta_{12} and β23\beta_{23}, respectively. If f2f_2 and g2g_2 are transverse, let H13=H12D2 ⁣×H23\mathcal{H}_{13}=\mathcal{H}_{12}\mathbin{{}_{D_2}\!\times}\mathcal{H}_{23}. Then

Φ13=(f1,g3):H13⟶D1‾×D3\Phi_{13}=(f_1,g_3):\mathcal{H}_{13}\longrightarrow\overline{D_1}\times D_3

carries the lagrangian structure

β13=pr⁡1∗β12+pr⁡2∗β23,\beta_{13}=\operatorname{pr}_1^*\beta_{12}+\operatorname{pr}_2^*\beta_{23},

see [5 Thm. 3.23].

Example 3.1 [5]. A recurrent example of this construction is the following. Suppose that the Lie algebra inclusion g∗⊂d\mathfrak{g}^{*} \subset\mathfrak{d} integrates to a Lie group morphism G∗→DG^{*} \to D. Then if β\beta is a lagrangian structure on Φ:H→D\Phi: \mathcal{H} \to D integrating the Lie algebroid of a Poisson action of (g,g∗)(\mathfrak{g},\mathfrak{g}^{*}) on (M,πM)(M,\pi_{M}), we have that H×DG∗\mathcal{H} \mathbin{\times_{D}} G^{*} equipped with the restriction of pr⁡1∗β\operatorname{pr}_{1}^{*}\beta is a symplectic groupoid integrating πM\pi_{M}.

We may apply this composition rule twice to integrate the two fibred products in (2.2). Let Φi:Hi→D\Phi_i : \mathcal{H}_i \to D be the integrating morphisms with two-forms βi\beta_i, and let JL:L→DJ_L : L \to D integrate the inclusion l↪d\mathfrak{l} \hookrightarrow\mathfrak{d}. Write p:RmD→D×Dp : R_{m_D} \to D \times D and q:RmD→Dq : R_{m_D} \to D for the input and output projections of the multiplication graph. The three correspondences are

H1×H2⟶{1},H1×H2→Φ1×Φ2D×D←pRmD→qD←JLL⟶{1}\mathcal{H}_1 \times\mathcal{H}_2 \longrightarrow\{1\}, \qquad\mathcal{H}_1 \times\mathcal{H}_2 \xrightarrow{\Phi_1 \times\Phi_2} D \times D \xleftarrow{p} R_{m_D} \xrightarrow{q} D \xleftarrow{J_L} L \longrightarrow\{1\}

Since RmDR_{m_D} and LL carry lagrangian structures given by the zero 2-form, it follows that the resulting groupoid

KL=((H1×H2)×D×DRmD)×DL\mathcal{K}_L = ((\mathcal{H}_1 \times\mathcal{H}_2) \mathbin{\times_{D \times D}} R_{m_D}) \mathbin{\times_D} L

is equipped with the symplectic form ω=pr⁡1∗β1+pr⁡2∗β2\omega= \operatorname{pr}_{1}^{*}\beta_1 + \operatorname{pr}_{2}^{*}\beta_2, where pr⁡i:KL→Hi\operatorname{pr}_i : \mathcal{K}_L \to\mathcal{H}_i are the projections.

Theorem 3.2. Suppose that the Lie algebroids A1A_1 and A2A_2 in Proposition 2.1 or Proposition 2.2 are integrable. Then the corresponding Poisson manifold is integrable. In both cases a symplectic groupoid is obtained by replacing each Lie algebroid in (2.2) by its groupoid counterpart.

Proof. For each Lie algebroid AiA_i, take its source-simply connected integration Hi⇒Xi\mathcal{H}_i \Rightarrow X_i. The corresponding Dirac embedding determines a lagrangian structure βi\beta_i on the Lie groupoid morphism Φi:Hi→D\Phi_i : \mathcal{H}_i \to D integrating φi\varphi_i [5 Cor. 3.13]. In the case of the Lie algebroids coming from Poisson actions, Φi\Phi_i is, in addition, a multiplicative DD-valued moment map as in [5 Thm. 5.3]. Let P:RmD→D2P : R_{m_D} \to D^2 be the projection to the composable pair and let Φfus\Phi_{\mathrm{fus}} be the projection to the multiplication after the first pullback below. Denoting by JL:L→DJ_L : L \to D the integration of the inclusion l⊂d\mathfrak{l} \subset\mathfrak{d} used in (2.2), the relevant fibred products are

Hfus⟶RmDKL⟶L↓↓P↓↓JLH1×H2→Φ1×Φ2D2Hfus→ΦfusD(6)\begin{aligned} \mathcal{H}_{\mathrm{fus}} &\longrightarrow R_{m_D} & \mathcal{K}_L &\longrightarrow L \\ \downarrow& & \downarrow^{P} & \downarrow& \downarrow^{J_L} \\ \mathcal{H}_1 \times\mathcal{H}_2 &\xrightarrow{\Phi_1 \times\Phi_2} D^2 & \mathcal{H}_{\mathrm{fus}} &\xrightarrow{\Phi_{\mathrm{fus}}} D \tag*{(6)} \end{aligned}

For the Poisson structure coming from the dynamical rr-matrix take JL=ϵ:G→DJ_L = \epsilon: G \to D. For the mixed product Poisson structure take L=G∗L = G^{*}, the simply connected group integrating c\mathfrak{c}, with its integrating immersion JLJ_L. Thus we obtain KL=Hfus×DL\mathcal{K}_L = \mathcal{H}_{\mathrm{fus}} \times_D L.

In Propositions 2.1 and 2.2, each of the two fibred product conditions has the expected codimension dim⁡g\dim\mathfrak{g}. This gives transversality on the source fibres at the units, and right translation implies that transversality holds at every other groupoid element. Thus both pullbacks in (6) are smooth Lie groupoids. By construction, differentiating (6) recovers (2.2):

Lie⁡(RmD)=Rm,Lie⁡(mD)=m,Lie⁡(JL)=j.\operatorname{Lie}(R_{m_D}) = R_m,\qquad\operatorname{Lie}(m_D) = m,\qquad\operatorname{Lie}(J_L) = j.

By [5 Thms. 3.22–3.23, Cor. 3.24], the corresponding lagrangian structures induce on KL\mathcal{K}_L the restriction of pr⁡1∗β1+pr⁡2∗β2\operatorname{pr}_{1}^{*}\beta_1 + \operatorname{pr}_{2}^{*}\beta_2 as discussed above and it integrates the graph of the Poisson structures identified in Proposition 2.1 or Proposition 2.2; hence this 2-form is symplectic. □

Corollary 3.1. Let h⊂g\mathfrak{h} \subset\mathfrak{g} be a Lie subalgebra and NN a connected open subset of h∗\mathfrak{h}^{*} with its linear Poisson structure and π2(N)=0\pi_{2}(N) = 0. A classical dynamical rr-matrix on NN determines an integrable Poisson structure (2.4) for every integrable input AMA_{M}.

Proof. Under the identification T∗N=N×hT^{*}N = N \times\mathfrak{h}, we obtain that the map bb is the inclusion of h\mathfrak{h} in g\mathfrak{g}, hence its transpose is surjective, and so the anchor of ANA_{N} in (2) is surjective. The resulting transitive Lie algebroid has trivial monodromy because π2(N)=0\pi_{2}(N) = 0 [6]. It follows that Theorem 3.2 may be applied.

For iterated mixed products, we need to integrate the full Lie algebroid associated with the Poisson action at each stage.

Corollary 3.2. Suppose that the full Lie algebroids associated with the two Poisson actions in Proposition 2.2 are integrable. Then the first fibred product in (6),

Hfus=(H1×H2)×D×DRmD,\mathcal{H}_{\mathrm{fus}} = (\mathcal{H}_{1} \times\mathcal{H}_{2}) \times_{D \times D} R_{m_{D}},

integrates the full Lie algebroid associated with the diagonal Poisson action on their mixed product.

Proof. Keeping v∈gv \in\mathfrak{g} free in the proof of Proposition 2.2 adds (ρ1(v),ρ2(v))(\rho_{1}(v), \rho_{2}(v)) to the cotangent anchor of the mixed product and also gives the Dirac embedding of the diagonal action. The first fibred product in (6) integrates this embedding.

Example. The rational dynamical rr-matrix for sl2\mathfrak{sl}_{2}

Choose a basis H,E+,E−H, E_{+}, E_{-} of g=sl2(R)\mathfrak{g} = \mathfrak{sl}_{2}(\mathbb{R}) satisfying [H,E±]=±2E±[H, E_{\pm}] = \pm2E_{\pm} and [E+,E−]=H[E_{+}, E_{-}] = H. On the dynamical base N=U={λ∈(RH)∗:q=λ(H)>0}N = U = \{\lambda\in(\mathbb{R}H)^{*} : q = \lambda(H) > 0\} take πU=0\pi_{U} = 0, s=Λ=0s = \Lambda= 0 and

ϑ=∂q⊗H,r(q)=−q−1E+∧E−.\vartheta= \partial_{q} \otimes H, \qquad r(q) = -q^{-1}E_{+} \wedge E_{-}.

Indeed, for R=∂q∧H+f(q)E+∧E−R = \partial_{q} \wedge H + f(q)E_{+} \wedge E_{-}, the coefficient of H∧E+∧E−H \wedge E_{+} \wedge E_{-} in 12[R,R]E\frac{1}{2}[R,R]_{E} is f2−f′f^{2} - f', and all other coefficients vanish. The choice f(q)=−1/qf(q) = -1/q therefore satisfies the dynamical equation in §2.

For a∈Ra \in\mathbb{R} and ξ∈g∗\xi\in\mathfrak{g}^{*}, write R♯(a dq,ξ)=(c∂q,u)R^{\sharp}(a\,dq,\xi) = (c\partial_{q},u). Then

c=−ξ(H),u=aH+q−1(ξ(E−)E+−ξ(E+)E−).c = -\xi(H), \qquad u = aH + q^{-1}\bigl(\xi(E_{-})E_{+} - \xi(E_{+})E_{-}\bigr).

These equations determine a,ξa,\xi uniquely from c,uc,u. Since [R,R]E=0[R,R]_{E} = 0, the map R♯R^{\sharp} identifies the Dirac algebroid AUA_{U} of (2.3) with the gauge algebroid TU⊕(U×g)TU \oplus(U \times\mathfrak{g}) [23 §2]. For a classical dynamical base, AUA_{U} is the dual algebroid of the dynamical Poisson groupoid [5 Ex. 4.9]. Li and Parmentier describe the dual groupoid over a contractible base [10 Thm. 3.2.4].

Let GG be the simply connected group with Lie algebra g\mathfrak{g}. The source-simply connected integration in this example is

HU=U×G×U⟹U.\mathcal{H}_{U} = U \times G \times U \Longrightarrow U.

An arrow (y,g,x)(y,g,x) has source xx and target yy, and the multiplication is given by (z,h,y)(y,g,x)=(z,hg,x)(z,h,y)(y,g,x) = (z,hg,x). Each source fibre is thus of the form U×GU \times G. Write η∈g∗\eta\in\mathfrak{g}^{*} for the covector with η(H)=1\eta(H) = 1 and η(E±)=0\eta(E_{\pm}) = 0, and put ℓ(q)=qη\ell(q) = q\eta. Since s=0s = 0, we can write the double with the group coordinate first as D=G⋉g∗D = G \ltimes\mathfrak{g}^{*}, where Ad⁡g∗ξ=ξ∘Ad⁡g−1\operatorname{Ad}^{*}_{g}\xi= \xi\circ\operatorname{Ad}_{g^{-1}}. Let θGR=dg g−1\theta^{R}_{G} = dg\,g^{-1} and θGL=g−1dg\theta^{L}_{G} = g^{-1}dg denote the matrix Maurer–Cartan forms pulled back to GG from SL2(R)\mathrm{SL}_{2}(\mathbb{R}). The moment map and its lagrangian structure are

ΦU(y,g,x)=(g,ℓ(y)−Ad⁡g∗ℓ(x)),\Phi_{U}(y,g,x) = (g,\ell(y) - \operatorname{Ad}^{*}_{g}\ell(x)),
βU=12 dy∧⟨η,θGR⟩+12 dx∧⟨η,θGL⟩.(7)\beta_{U} = \frac{1}{2}\,dy \wedge\langle\eta,\theta^{R}_{G}\rangle+ \frac{1}{2}\,dx \wedge\langle\eta,\theta^{L}_{G}\rangle. \tag*{(7)}

In order to verify that ΦU\Phi_U is a morphism, we may write it as the product of A=(1,ℓ(y))A=(1,\ell(y)), B=(g,0)B=(g,0) and C=(1,−ℓ(x))C=(1,-\ell(x)), it thus follows that the two factors at the common endpoint cancel under composition. Since the subgroups GG and g∗\mathfrak{g}^{*} are isotropic, the Polyakov–Wiegmann identity for this factorization implies that βU=−(A,B)∗ΩD−(AB,C)∗ΩD\beta_U=-(A,B)^{*}\Omega_D-(AB,C)^{*}\Omega_D. This is the second formula in (3.2) and it implies the two required identities for βU\beta_U.

At the unit (q,1,q)(q,1,q), the algebroid element (c ∂q,u)(c\,\partial_q,u) is represented by v=(c,−u,0)v=(c,-u,0) in the right-invariant convention of [5 §2]. Using the same identification in the double, we obtain

Lie⁡(ΦU)(c ∂q,u)=(u,−cη−ad⁡ℓ(q)∗ξ)=(u,ξ),βU(v,(Y,0,Y))=η(u)Y=aY(Y∈R).\operatorname{Lie}(\Phi_U)(c\,\partial_q,u)=(u,-c\eta-\operatorname{ad}^{*}_{\ell(q)}\xi)=(u,\xi),\qquad\beta_U(v,(Y,0,Y))=\eta(u)Y=aY\quad(Y\in\mathbb{R}).

Together with the anchor c ∂qc\,\partial_q, these are exactly the components of the embedding in (2.3) corresponding to the Dirac structure associated to a dynamical rr-matrix. Therefore, we have that (HU,ΦU,βU)(\mathcal{H}_U,\Phi_U,\beta_U) integrates the Dirac structure of r(q)r(q).

In order to apply Theorem 3.2, we may take any integrable Lie algebroid AM=T∗M⊕(M×g)A_M=T^{*}M\oplus(M\times\mathfrak{g}) corresponding to a GG-action on MM by Poisson automorphisms. Let βM\beta_M be the corresponding lagrangian structure and denote by ΦM(k)=(gM(k),νM(k))\Phi_M(k)=(g_M(k),\nu_M(k)) the components of its moment map, which takes values in D=G⋉g∗D=G\ltimes\mathfrak{g}^{*}. The two conditions expressed by (3.1) now read

gM(k)=g,νM(k)+ℓ(y)−Ad⁡g∗ℓ(x)=0.g_M(k)=g,\qquad\nu_M(k)+\ell(y)-\operatorname{Ad}^{*}_{g}\ell(x)=0.

It follows that the pairs (k,(y,g,x))(k,(y,g,x)) satisfying these equations constitute a groupoid over M×UM\times U integrating (2.4), whose symplectic form is the restriction of pr⁡1∗βM+pr⁡2∗βU\operatorname{pr}_{1}^{*}\beta_M+\operatorname{pr}_{2}^{*}\beta_U.

Example. Integration of the flag variety

Let GG be a connected complex semisimple group with opposite Borel subgroups B=B+B=B_{+} and B−B_{-} with maximal torus T=B+∩B−T=B_{+}\cap B_{-}. Let UU be the unipotent radical of BB, and denote the quotient maps by q±:B±→Tq_{\pm}:B_{\pm}\to T. Choosing an invariant nondegenerate bilinear form on g\mathfrak{g}, an orthonormal basis (hi)(h_i) of Lie⁡(T)\operatorname{Lie}(T), and dual root vectors E±αE_{\pm\alpha} for positive roots α\alpha determined by BB, we obtain the standard quasi-triangular rr-matrix

rst=12∑ihi⊗hi+∑α>0E−α⊗Eα,Λst=12(rst−rst21).r_{\mathrm{st}}=\frac{1}{2}\sum_i h_i\otimes h_i+\sum_{\alpha>0}E_{-\alpha}\otimes E_{\alpha},\qquad\Lambda_{\mathrm{st}}=\frac{1}{2}(r_{\mathrm{st}}-r_{\mathrm{st}}^{21}).

where the superscript 2121 interchanges the tensor factors. We use the Poisson tensor πG=ΛstR−ΛstL\pi_G=\Lambda_{\mathrm{st}}^{R}-\Lambda_{\mathrm{st}}^{L}, following the convention in [3 §2.1], this amounts to using r=−rstr=-r_{\mathrm{st}} in (2.5). The projection G→G/BG\to G/B determines a Poisson structure on the quotient that we denote by πG/B\pi_{G/B}. Since the symmetric part of rr is nondegenerate, we obtain that the double is the pair groupoid D=G×G⇒GD=G\times G\Rightarrow G. With r=−rstr=-r_{\mathrm{st}}, the endpoint map (p+,p−)(p_{+},p_{-}) takes the pairing given by evaluation to the one given by the negative of the chosen form on the first copy of g\mathfrak{g} and its positive on the second copy.

The groupoid integrating the Lie algebroid corresponding to the g\mathfrak{g}-action on G/BG/B is the decorated annulus of [5 Ex. 5.26], [3 proof of Prop. 3.1]. Explicitly, the space of arrows is HB=G×(G×BU)\mathcal{H}_B=G\times(G\times_B U), where G×BUG\times_B U is the quotient by b⋅(g,u)=(gb−1,bub−1)b\cdot(g,u)=(gb^{-1},bub^{-1}). The multiplication of two arrows h1=(a1,[g1,u1])h_1=(a_1,[g_1,u_1]) and h2=(a2,[g2,u2])h_2=(a_2,[g_2,u_2]) is given by

(a1,[g1,u1])(a2,[a1−1g1,u2])=(a1a2,[g1,u1u2]);(a_1,[g_1,u_1])(a_2,[a_1^{-1}g_1,u_2])=(a_1a_2,[g_1,u_1u_2]);

choosing representatives with g2=a1−1g1g_2=a_1^{-1}g_1. An element (a,[g,u])(a,[g,u]) has source, target and moment map

s(a,[g,u])=a−1gB,t(a,[g,u])=gB,ΦB(a,[g,u])=(gug−1,a).s(a,[g,u])=a^{-1}gB,\qquad t(a,[g,u])=gB,\qquad\Phi_B(a,[g,u])=(gug^{-1},a).

The lagrangian structure on ΦB\Phi_B is the two-form βB\beta_B determined as follows. Let κ\kappa denote the chosen invariant bilinear form on g\mathfrak{g}, with the pairing on D=G×GD=G\times G being (−κ,κ)(-\kappa,\kappa). Consider the quotient map

q:G×G×U⟶HB,q(a,g,u)=(a,[g,u]),q:G\times G\times U\longrightarrow\mathcal{H}_B,\qquad q(a,g,u)=(a,[g,u]),

and put C=gug−1C=gug^{-1}. Then we may define βB\beta_B via

q∗βB=12κ(C−1dC∧(dgg−1−daa−1))−12κ(g−1dg∧duu−1).q^*\beta_B=\frac{1}{2}\kappa\left(C^{-1}dC\wedge\left(dg g^{-1}-da a^{-1}\right)\right)-\frac{1}{2}\kappa\left(g^{-1}dg\wedge du u^{-1}\right).

Graphically, we may depict HB\mathcal{H}_B as the space of representations of the fundamental groupoid of the annulus. One boundary carries a flag with its monodromy lying in the unipotent radical of the stabilizer of that flag while its other boundary has framings on two marked points. In the disc of Figure 2 we have a flag over one marked point and a framing on the other, parallel transport from the flag to the framed point identifies the flag with a point of G/BG/B. Changing that frame gives the residual GG-action.

The flag annulus and its base

Figure 2. The flag annulus and its base. The two maps retain the flag in the target frame and in the source frame, respectively.

The Poisson structure on G/BG/B is integrated by the restriction

ΦB−1(G∗)⇒G/B,G∗={(b+,b−)∈B+×B−:q+(b+)q−(b−)=1}.\Phi_B^{-1}(G^*)\Rightarrow G/B,\qquad G^*=\{(b_+,b_-)\in B_+\times B_-:q_+(b_+)q_-(b_-)=1\}.

Here G∗G^* denotes the indicated embedded integration of the dual Lie algebra. The residual moment map in [3 proof of Prop. 3.1] lists the two holonomies in the reverse order.

Example. Integration of flag configurations

For m≥1m\ge1, we denote by F~m=G×B⋯×BG\widetilde{F}_m=G\times_B\cdots\times_B G the quotient of GmG^m by the intermediate Bm−1B^{m-1}-action. At the jj-th position, we have that b∈Bb\in B sends (gj,gj+1)(g_j,g_{j+1}) to (gjb,b−1gj+1)(g_jb,b^{-1}g_{j+1}). Put Fm=F~m/BF_m=\widetilde{F}_m/B for the quotient of the remaining right action on the last factor. Using the expressions pj=g1⋯gjp_j=g_1\cdots g_j for all j=1…m+1j=1\ldots m+1, we obtain the two identifications

Fm⟶(G/B)m,[g1,…,gm]⟼(p1B,…,pmB),F_m\longrightarrow(G/B)^m,\qquad[g_1,\ldots,g_m]\longmapsto(p_1B,\ldots,p_mB),
F~m+1⟶(G/B)m×G,[g1,…,gm+1]⟼(p1B,…,pmB,pm+1).\widetilde{F}_{m+1}\longrightarrow(G/B)^m\times G,\qquad[g_1,\ldots,g_{m+1}]\longmapsto(p_1B,\ldots,p_mB,p_{m+1}).

In these coordinates, the quotient Poisson structures are iterated mixed products [3 §2.2 and Thm. 2.1], [18 Prop. A.6].

To integrate FmF_m, one takes mm copies of HB\mathcal{H}_B, which is represented by the decorated annulus of Figure 2. According to Corollary 3.2, one needs to take the fibred product of mm such groupoids along matching components of their moment maps. Explicitly, this amounts to the following construction. For each hj∈HBh_j\in\mathcal{H}_B we write ΦB(hj)=(aj,bj)∈D=G×G\Phi_B(h_j)=(a_j,b_j)\in D=G\times G and define the fibred product with residual moment map

Em={(h1,…,hm):bj=aj+1 (1≤j<m)},Φm(h1,…,hm)=(a1,bm).E_m=\{(h_1,\ldots,h_m):b_j=a_{j+1}\ (1\le j<m)\},\qquad\Phi_m(h_1,\ldots,h_m)=(a_1,b_m).

Geometrically, these equations describe the gluing of successive annuli along boundary arcs. The resulting surface is a sphere with m+1m+1 holes. Each of the mm holes is labelled with a flag and the monodromy around the hole is constrained to lie in the unipotent radical of the stabilizer of the flag. The remaining boundary component carries two marked points and its two arc holonomies serve to define Φm\Phi_m, see Figure 3.

Successive gluing of decorated annuli

Figure 3. Successive gluing of decorated annuli integrates the full Lie algebroid of the Poisson action on FmF_m

By Theorem 3.2 (or Example 3.1), a symplectic groupoid integrating the mixed Poisson structure is

Φm−1(G∗)⟹Fm.\Phi_m^{-1}(G^*) \Longrightarrow F_m.

For m=1m=1 this gives the groupoid over G/BG/B mentioned above.

In order to integrate F~m+1\widetilde{F}_{m+1}, attach to the same mm decorated annuli the unrestricted annulus HG\mathcal{H}_G integrating the last GG-factor, which is described in [5 Prop. 5.23]. This gives a sphere with m+2m+2 holes, see Figure 4.

Integration of the Lie algebroid encoding the Poisson action

Figure 4. Integration of the Lie algebroid encoding the Poisson action on F~m+1\widetilde{F}_{m+1}

There are now two free boundaries with marked points at which to impose suitable lagrangian conditions. Write (G∗)∨={(v,u):(u,v)∈G∗}(G^*)^\vee=\{(v,u):(u,v)\in G^*\}. Following the convention of [3 Rem. 3.2], these two ordered pairs of arc holonomies are constrained to lie in G∗G^* and (G∗)∨(G^*)^\vee, respectively. This constraint determines an integration of F~m+1\widetilde{F}_{m+1}. Figure 5 shows the two surfaces for m=2m=2. The construction with the last GG-factor recovers the symplectic groupoid of [3 Prop. 3.1].

Surfaces for the two symplectic groupoids with two flag factors

Figure 5. Surfaces for the two symplectic groupoids with two flag factors. Each UU denotes a flag boundary with unipotent monodromy. On each boundary with two dots the subgroup label constrains the ordered pair of arc holonomies.

References

  1. [1]Alekseev A., Kosmann-Schwarzbach Y., Manin pairs and moment maps, J. Differential Geom. 56 (2000), 133–165, arXiv:math/9909176.
  2. [2]Alekseev A., Malkin A., Meinrenken E., Lie group valued moment maps, J. Differential Geom. 48 (1998), 445–495, arXiv:dg-ga/9707021.arxiv.org/abs/dg-ga/9707021
  3. [3]Álvarez D., Morita equivalences, moduli spaces, and flag varieties, Int. Math. Res. Not. 2026 (2026), rnag039, arXiv:2504.09293.DOI
  4. [4]Álvarez D., Bursztyn H., Cueca M., Quasi-hamiltonian groupoids and moduli spaces of flat bundles over surfaces.
  5. [5]Álvarez D., Bursztyn H., Cueca M., Shifted lagrangian structures in Poisson geometry, 2026, arXiv:2605.29117.arxiv.org/abs/2605.29117
  6. [6]Crainic M., Fernandes R.L., Integrability of Lie brackets, Ann. of Math. 157 (2003), 575–620, arXiv:math/0105033.DOI
  7. [7]Etingof P., Varchenko A., Solutions of the quantum dynamical Yang–Baxter equation and dynamical quantum groups, Comm. Math. Phys. 196 (1998), 591–640.
  8. [8]Fernandes R.L., Iglesias Ponte D., Integrability of Poisson–Lie group actions, Lett. Math. Phys. 90 (2009), 137–159, arXiv:0902.3558.arxiv.org/abs/0902.3558
  9. [9]Fernandes R.L., Ortega J.P., Ratiu T.S., The momentum map in Poisson geometry, Amer. J. Math. 131 (2009), 1261–1310.arxiv.org/abs/0705.0562
  10. [10]Li L.C., Parmentier S., On dynamical Poisson groupoids I, Mem. Amer. Math. Soc. 174 (2005), arXiv:math/0209212.DOI
  11. [11]Li-Bland D., Ševera P., Symplectic and Poisson geometry of the moduli spaces of flat connections over quilted surfaces, in Mathematical Aspects of Quantum Field Theories, Springer, 2015, 343–411, example 1.8 and (1.9) in the preprint are Example 8 and (9) in the book, arXiv:1304.0737.
  12. [12]Liu Z.J., Xu P., Dirac structures and dynamical r-matrices, Ann. Inst. Fourier 51 (2001), 835–859.DOI
  13. [13]Liu Z.J., Xu P., Dynamical r-matrices coupled with Poisson manifolds, Prog. Theor. Phys. Suppl. 144 (2001), 133–140.DOI
  14. [14]Liu Z.J., Xu P., The local structure of Lie bialgebroids, Lett. Math. Phys. 61 (2002), 15–28, arXiv:math/0210104.DOI
  15. [15]Lu J.H., Momentum mappings and reduction of Poisson actions, in Symplectic Geometry, Groupoids, and Integrable Systems, Mathematical Sciences Research Institute Publications, Vol. 20, Editors P. Dazord, A. Weinstein, Springer, New York, 1991, 209–226.DOI
  16. [16]Lu J.H., Poisson homogeneous spaces and Lie algebroids associated to Poisson actions, Duke Math. J. 86 (1997), 261–304, arXiv:q-alg/9503003.arxiv.org/abs/q-alg/9503003
  17. [17]Lu J.H., Mouquin V., Mixed product Poisson structures associated to Poisson Lie groups and Lie bialgebras, Int. Math. Res. Not. 2017 (2017), 5919–5976, arXiv:1504.06843.arxiv.org/abs/1504.06843
  18. [18]Lu J.H., Mouquin V., Yu S., Configuration Poisson groupoids of flags, Int. Math. Res. Not. 2023 (2023), 18035–18107, arXiv:2109.03724.arxiv.org/abs/2109.03724
  19. [19]Meusburger C., Schönfeld T., Gauge fixing and classical dynamical r-matrices in ISO(2,1)-Chern–Simons theory, Comm. Math. Phys. 327 (2014), 443–479, arXiv:1203.5609.DOI
  20. [20]Mouquin V., The Fock–Rosly Poisson structure as defined by a quasi-triangular r-matrix, SIGMA 13 (2017), 063, 13 pages, arXiv:1610.01782.
  21. [21]Mouquin V., Local Poisson groupoids over mixed product Poisson structures and generalised double Bruhat cells, J. Symplectic Geom. 19 (2021), 993–1045, arXiv:1908.04044.
  22. [22]Pantev T., Toën B., Vaquié M., Vezzosi G., Shifted symplectic structures, Publ. Math. Inst. Hautes Études Sci. 117 (2013), 271–328, arXiv:1111.3209.arxiv.org/abs/1111.3209
  23. [23]Xu P., Triangular dynamical r-matrices and quantization, Adv. Math. 166 (2002), 1–49, arXiv:math/0005006.DOI
  24. [24]Xu X., Generalized classical dynamical Yang–Baxter equations and moduli spaces of flat connections on surfaces, Comm. Math. Phys. 341 (2016), 523–542, arXiv:1407.2071v4.

Paper details

vlnc:W_agkm9y47
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC0-1.0
Usage

views · 0

downloads · 0

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

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents