prequantized Lagrangian correspondence

This entry is a sub-chapter of geometry of physics. See there for background

The following is effectively a derivation of, and an introduction to, classical mechanics by studying correspondences in what is called (as we will explain) the slice topos over the moduli stack of prequantum line bundles. One such correspondence in this slice topos is precisely a prequantized Lagrangian correspondence and the reader looking for just these should skip ahead to the section The classical action functional prequantizes Lagrangian correspondences. But for completeness and to introduce the technology used here, we start with introducing also more basic concepts, such as phase space etc.



physics, mathematical physics, philosophy of physics

Surveys, textbooks and lecture notes

theory (physics), model (physics)

experiment, measurement, computable physics

Symplectic geometry

Classical Mechanics via Prequantum Correspondences

Model Layer

Here we discuss the notion of prequantized Lagrangian correspondences and how it serves to embed traditional classical mechanics in its formulation as Hamiltonian mechanics and Lagrangian mechanics into the general context of local prequantum field theory and its motivic quantization.

A traditional notion is that of a plain Lagrangian correspondence, which is a Lagrangian submanifold of the Cartesian product of a symplectic manifold and another one, with opposite symplectic structure. As discussed there, plain Lagrangian correspondences naturally generalize symplectomorphisms – hence transformations between phase spaces in physics – via correspondences of symplectic manifolds.

But in prequantum field theory proper, and in particular with an eye towards geometric quantization, one considers the prequantization of these symplectic manifolds by lifting them to prequantum circle bundles with principal connection. The notion of prequantized Lagrangian correspondence is the refinement of that of plain Lagrangian correspondence which does properly respect and reflect this prequantization information: a prequantized Lagrangian correspondence is a Lagrangian subspace as before, but now equipped with an explicit gauge transformation between the pullbacks of the two prequantum circle bundles to the correspondence space.

We discuss below how the concept of prequantized Lagrangian correspondences neatly induces and unifies ingredients and aspects of classical mechanics, notably the Hamiltonian mechanics of symplectic manifolds and the Lagrangian mechanics of action functionals associated to it via the Legendre transform. Specifically, below in The classical action prequantizes Hamiltonian correspondences we see that prequantized Lagrangian correspondences are diagrams which schematically express this data as follows:

spaceoftrajectories initialvalues Hamiltonianevolution phasespace in actionfunctional phasespace out prequantumbundle in prequantumbundle out 2-groupofphases. \array{ && {{space\,of} \atop {trajectories}} \\ & {}^{\mathllap{{initial}\atop {values}}}\swarrow && \searrow^{\mathrlap{{Hamiltonian} \atop {evolution}}} \\ phase\,space_{in} && \swArrow_{{action} \atop {functional}} && phase \,space_{out} \\ & {}_{\mathllap{{prequantum}\atop {bundle}_{in}}}\searrow && \swarrow_{\mathrlap{{prequantum} \atop {bundle}_{out}}} \\ && {{2\text{-}group} \atop {of\,phases}} } \,.

This describes a diagram in what is called the slice topos of smooth sets over the moduli stack of prequantum circle bundles. Once formulated this way, there is an evident refinement to higher moduli stacks of prequantum n-bundles.

For n=2n = 2 we show how the notion of prequantized Lagrangian correspondence is – still naturally in the context of local prequantum field theory – further refined from the context of symplectic manifolds to that of Poisson manifolds. Specifically, this is obtained by realizing that prequantized Lagrangian correspondences are really naturally to be regarded as correspondences-of-correspondences in a 2-category of correspondences, where now the new lower-order correspondences are instead boundary field theories for a 2d Chern-Simons theory (a non-perturbative Poisson sigma-model).

Phase spaces and symplectic manifolds

Given a physical system, one says that its phase space is the space of its possible (“classical”) histories or trajectories. The first two of Newton's laws of motion say that trajectories of physical systems are (typically) determined by differential equations of second order, and therefore these spaces of trajectories are (typically) equivalent to initial value data of 0th and of 1st derivatives. In physics this data (or rather its linear dual) is referred to as the canonical coordinates and the canonical momenta, respectively, traditionally denoted by the symbols “qq” and “pp”. But being coordinates, these are actually far from being canonical in the mathematical sense; all that has invariant meaning is, locally, the surface element dpdq\mathbf{d}p \wedge \mathbf{d}q spanned by a change of coordinates and momenta.

So far this says that a physical phase space is mathematically formalized by a sufficiently smooth manifold XX which is equipped with a closed and non-degenerate differential 2-form ωΩ cl 2(X)\omega \in \Omega^2_{\mathrm{cl}}(X), hence by a symplectic manifold (X,ω)(X,\omega).

The non-degeneracy of a symplectic form encodes the special property (as we will make explicit below) that (time) evolution of coordinates and momenta is uniquely induced by an action functional/Hamiltonian generating the evolution. This is however famously not the case for systems with gauge equivalences, hence such systems which have configurations that are nominally different but nevertheless physically equivalent. Presence of such gauge equivalences is not the exception, but the rule for physical systems, and therefore we want to include this case.

In the presence of gauge equivalences, the phase space form ω\omega is still a closed differential 2-form, it just need not be non-degenerate anymore. While in such a case the pair (X,ω)(X,\omega) could just be called a smooth manifold equipped with a closed differential 2-form}, it is traditional to call this a pre-symplectic manifold in order to amplify the indented use as a model for phase spaces. (Some authors demand that a pre-symplectic form be a closed form with constant rank, but here this technical condition will not be relevant and will not be considered.)


The sigma-model describing the propagation of a particle on the real line \mathbb{R} has as phase space the plane 2=T *\mathbb{R}^2 = T^\ast \mathbb{R} and as symplectic form its canonical volume form. Traditionally the two canonical coordinate functions on this phase space are denoted q,p: 2q,p \;\colon\; \mathbb{R}^2 \longrightarrow \mathbb{R} (called the “canonical coordinate” and the “canonical momentum”, respectively), and in terms of these the symplectic form in this example is ω=dqdp\omega = \mathbf{d} q \wedge \mathbf{d} p.

When dealing with spaces XX that are equipped with extra structure, such as ωΩ cl 2(X)\omega \in \Omega^2_{\mathrm{cl}}(X), then it is useful to have a universal moduli space for these structures, and this will be central for our developments here. So we need a “smooth setΩ cl 2\mathbf{\Omega}^2_{\mathrm{cl}} of sorts, characterized by the property that there is a natural bijection between smooth closed differential 2-forms ωΩ cl 2(X)\omega \in \Omega^2_{\mathrm{cl}}(X) and smooth maps XΩ cl 2 X \longrightarrow \mathbf{\Omega}^2_{\mathrm{cl}} . Of course such a universal moduli spaces of closed 2-forms does not exist in the category of smooth manifolds. But it does exist canonically if we slightly generalize the notion of “smooth set” suitably.


A smooth set or smooth 0-type XX is

  1. an assignment to each nn \in \mathbb{N} of a set, to be written X( n)X(\mathbb{R}^n) and to be called the set of smooth maps from n\mathbb{R}^n into XX,

  2. an assignment to each ordinary smooth function f: n 1 n 2f : \mathbb{R}^{n_1} \to \mathbb{R}^{n_2} between Cartesian spaces of a function of sets X(f):X( n 2)X( n 1)X(f) : X(\mathbb{R}^{n_2}) \to X(\mathbb{R}^{n_1}), to be called the pullback of smooth functions into XX along ff;

such that

  1. this assignment respects composition of smooth functions;

  2. this assignment respects the covering of Cartesian spaces by open disks: for every good open cover { nU i n} i\{\mathbb{R}^n \simeq U_i \hookrightarrow \mathbb{R}^n\}_i, the set X( n)X(\mathbb{R}^n) of smooth functions out of n\mathbb{R}^n into XX is in natural bijection with the set {(ϕ i) i iX(U i)| i,jϕ i| U iU j=ϕ j| U iU j}\left\{ (\phi_i)_i \in \prod_i X(U_i) \;|\; \forall_{i,j}\; \phi_i|_{U_{i} \cap U_j} = \phi_j|_{U_{i} \cap U_j} \right\} of tuples of smooth functions out of the patches of the cover which agree on all intersections of two patches.

For more on this see at geometry of physics in the section Smooth sets.

While the formulation of this definition is designed to make transparent its geometric meaning, of course equivalently but more abstractly this says the following:


Write CartSp for the category of Cartesian spaces with smooth functions between them, and consider it as a site by equipping it with the coverage of good open covers. A smooth set or smooth 0-type is a sheaf on this site. The topos of smooth 0-types is the category of sheaves

Smooth0TypeSh(CartSp). \mathrm{Smooth}0\mathrm{Type} \coloneqq \mathrm{Sh}(\mathrm{CartSp}) \,.

In the following we will abbreviate the notation to

HSmooth0Type. \mathbf{H} \coloneqq \mathrm{Smooth}0\mathrm{Type} \,.

The topos of prop. 2 also has another site of definition.


Write SmoothMfdSmoothMfd for the category of smooth manifolds regarded as a site with the standard Grothendieck topology of open covers. There is an equivalence of categories

Sh(SmoothMfd)Smooth0Type. Sh(SmoothMfd) \simeq \mathrm{Smooth}0\mathrm{Type} \,.

The canonical inclusion CartSpSmoothMfdCartSp\hookrightarrow SmoothMfd is readily seen to be a dense subsite. (This is just the statement that – by definition – every smooth manifold may be covered by Cartesian spaces.) The statement hence follows by the comparison lemma.

For the discussion of presymplectic manifolds, we need the following two examples.


Every smooth manifold XSmoothManifoldX \in \mathrm{SmoothManifold} becomes a smooth 0-type by the assignment

X:nC ( n,X). X \;\colon\; n \mapsto C^\infty(\mathbb{R}^n, X) \,.

This construction extends to a full embedding of smooth manifolds into smooth sets

SmoothManifoldH SmoothManifold \hookrightarrow \mathbf{H}

This follows via prop. 1 by the Yoneda lemma.


For pp \in \mathbb{N}, write Ω cl p\mathbf{\Omega}^p_{\mathrm{cl}} for the smooth set whose nn-dimensional plots are smooth differential p-forms on n\mathbb{R}^n:

Ω cl p: nΩ cl p( n) \mathbf{\Omega}^p_{\mathrm{cl}} \colon \mathbb{R}^n \mapsto \Omega^p_{\mathrm{cl}}(\mathbb{R}^n)

and which sends a smooth function f: n 1 n 2f \colon \mathbb{R}^{n_1} \to \mathbb{R}^{n_2} to the pullback of differential forms along this map

n 1 Ω cl p( n 1) f f * n 2 Ω cl p( n 2). \array{ \mathbb{R}^{n_1} &\mapsto& \Omega^p_cl(\mathbb{R}^{n_1}) \\ \downarrow^{\mathrlap{f}} && \uparrow^{\mathrlap{f^\ast}} \\ \mathbb{R}^{n_2} &\mapsto& \Omega^p_cl(\mathbb{R}^{n_2}) } \,.

For more on this example see at geometry of physics in the section Differential forms.

This solves the moduli problem for closed smooth differential forms:


For pp \in \mathbb{N} and XSmoothManifoldSmooth0TypeX \in SmoothManifold \hookrightarrow Smooth0Type, there is a natural bijection

H(X,Ω cl p)Ω cl p(X) \mathbf{H}(X,\mathbf{\Omega}^p_{\mathrm{cl}}) \simeq \Omega^p_{\mathrm{cl}}(X)

between morphisms (of smooth sets)

XΩ cl 2 X \longrightarrow \Omega^2_{cl}

and smooth closed 2-forms

ωΩ cl 2(X) \omega \in \Omega^2_{cl}(X)

on XX.


This follows via prop. 1 by the Yoneda lemma.

So a presymplectic manifold (X,ω)(X,\omega) is equivalently a map of smooth sets of the form

ω:XΩ cl 2. \omega \;\colon\; X \longrightarrow \mathbf{\Omega}^2_{\mathrm{cl}} \,.

Canonical transformations and symplectomorphisms

An equivalence between two phase spaces, hence a re-expression of the “canonical coordinates” and “canonical momenta”, is called a canonical transformation in physics. Mathematically this is a symplectomorphism.


Given two symplectic manifolds (X 1,ω 1)(X_1, \omega_1) and (X 2,ω 2)(X_2, \omega_2) (which might be two copies of one single symplectic manifold), a symplectomorphism between them

f:(X 1,ω 1)(X 2,ω 2) f \;\colon\; (X_1, \omega_1) \longrightarrow (X_2, \omega_2)

is a diffeomorphism

f:X 1X 2 f \;\colon\; X_1 \longrightarrow X_2

of the underlying smooth manifolds, such that the pullback of the second symplectic form along ff equals the first,

f *ω 2=ω 1. f^\ast \omega_2 = \omega_1 \,.

The above formulation of pre-symplectic manifolds as maps into a moduli space of closed differential 2-forms yields the following formulation of symplectomorphisms, which is very simple in itself, but contains in it the seed of an important phenomenon:


A symplectomorphism f:(X 1,ω 2)(X 2,ω 2)f \colon (X_1, \omega_2) \longrightarrow (X_2, \omega_2) as above is, under the identification of prop. 2, equivalently a commuting diagram in H\mathbf{H} of the form

X 1 f X 2 ω 1 ω 2 Ω cl 2. \array{ X_1 && \stackrel{f}{\longrightarrow}&& X_2 \\ & {}_{\mathllap{\omega_1}}\searrow && \swarrow_{\mathrlap{\omega_2}} \\ && \Omega^2_{cl} } \,.

This is the naturality of the Yoneda lemma:

By prop. 2 the identification of ω iΩ cl 2(X i)\omega_i \in \Omega^2_{cl}(X_i) with ω i:XΩ cl 2\omega_i \colon X \to \Omega^2_{cl} is via the natural equivalence

Hom Smooth0Type(X,Ω cl 2)Ω cl 2(X). Hom_{Smooth0Type}(X,\mathbf{\Omega}^2_{cl}) \stackrel{\simeq}{\longrightarrow} \Omega^2_{cl}(X) \,.

This being natural means that for every morphism X 1fX 2X_1 \stackrel{f}{\longrightarrow} X_2 there is a commuting diagram of the form

Hom(X 2,Ω cl 2) Ω cl 2(X 2) ()f f * Hom(X 1,Ω cl 2) Ω cl 2(X 1) \array{ Hom(X_2,\mathbf{\Omega}^2_{cl}) &\stackrel{\simeq}{\longrightarrow}& \Omega^2_{cl}(X_2) \\ \downarrow^{\mathrlap{(-) \circ f}} && \downarrow^{\mathrlap{f^\ast}} \\ Hom(X_1,\mathbf{\Omega}^2_{cl}) &\stackrel{\simeq}{\longrightarrow}& \Omega^2_{cl}(X_1) }

where on the left we have the pre-composition operation on morphisms and on the right we have, by example 3, the pullback of differential forms.

Consider then the element

(X 2ω 2Ω cl 2)Hom(X 2,Ω cl 2) (X_2 \stackrel{\omega_2}{\to} \mathbf{\Omega}^2_{cl}) \in Hom(X_2,\mathbf{\Omega}^2_{cl})

in the top left set in this diagram. Sending it along the top and right maps yields the pullback of differential forms f *ω 2Ω cl 2(X 1)f^\ast \omega_2 \in \Omega^2_{cl}(X_1). On the other hand, sending it along the left and bottom maps yields the differential form represented by the composite morphism (X 1fX 2ω 1Ω cl 2)(X_1 \stackrel{f}{\to} X_2 \stackrel{\omega_1}{\to}\mathbf{\Omega}^2_{cl}). Commutativity of the above naturality diagram means that these two elements of Ω cl 2(X 1)\Omega^2_{cl}(X_1) coincide. This is the claim to be proven.

Situations like this are naturally interpreted in a slice topos:


For AHA \in \mathbf{H} any smooth set, the slice topos H /A\mathbf{H}_{/A} is the category whose objects are objects XHX \in \mathbf{H} equipped with maps XAX \to A, and whose morphisms are commuting diagrams in H\mathbf{H} of the form

X Y A \array{ X &&\longrightarrow&& Y \\ & \searrow && \swarrow \\ && A }

Write SymplManifold\mathrm{SymplManifold} for the category with presymplectic smooth manifolds as objects and symplectomorphisms, def. 3, betwen them as morphisms


The construction of prop. 2 which sends a smooth symplectic manifold (X,ω)(X,\omega) to the classifying morphism of smooth sets (XωΩ cl 2)(X \stackrel{\omega}{\longrightarrow}\mathbf{\Omega}^2_{cl}) regarded as an object in the slice topos, def. 4 constitutes a full and faithful functor

SymplManifoldSmooth0Type /Ω cl 2 SymplManifold \hookrightarrow Smooth0Type_{/\mathbf{\Omega}^2_{\mathrm{cl}}}

of pre-symplectic manifolds with symplectomorphisms between them into the slice topos of smooth sets over the smooth moduli space of closed differential 2-forms.


By prop. 3.

Trajectories and Lagrangian correspondences

A symplectomorphism clearly puts two symplectic manifolds “in relation” to each other. But it does so also in the formal sense of relations in mathematics. Recall:


For X,YX,Y \in Set two sets, a relation RR between elements of XX and elements of YY is a subset of the Cartesian product set

RX×Y. R \hookrightarrow X \times Y \,.

More generally, for X,YHX, Y \in \mathbf{H} two objects of a topos (such as the topos of smooth sets), then a relation RR between them is a subobject of their Cartesian product

RX×Y. R \hookrightarrow X \times Y \,.

In particular any function induces the relationyy is the image of xx”:


For f:XYf \;\colon\; X \longrightarrow Y a function, its induced relation is the relation which is exhibited by the graph of ff

graph(f){(x,y)X×Y|f(x)=y} graph(f) \coloneqq \left\{ (x,y) \in X \times Y \;|\; f(x) = y \right\}

canonically regarded as a subobject

graph(f)X×Y. graph(f) \hookrightarrow X \times Y \,.

Hence in the context of classical mechanics, in particular any symplectomorphism f:(X 1,ω 1)(X 2,ω 2)f \;\colon\; (X_1, \omega_1) \longrightarrow (X_2, \omega_2) induces the relation

graph(f)X 1×X 2. graph(f) \hookrightarrow X_1 \times X_2 \,.

Since we are going to think of ff as a kind of “physical process”, it is useful to think of the smooth set graph(f)graph(f) here as the space of trajectories of that process. To make this clearer, notice that we may equivalently rewrite every relation RX×YR \hookrightarrow X \times Y as a diagram of the following form:

R i X i Y X Y= R X×Y p X p Y X Y \array{ && R \\ & {}^{\mathllap{i_X}}\swarrow && \searrow^{\mathrlap{i_Y}} \\ X && && Y } \;\; = \;\; \array{ && R \\ && \downarrow \\ && X \times Y \\ & {}^{\mathllap{p_X}}\swarrow && \searrow^{\mathrlap{p_Y}} \\ X && && Y }

reflecting the fact that every element (xy)R(x \sim y) \in R defines an element x=i X(xy)Xx = i_X(x \sim y) \in X and an element y=i Y(xy)Yy = i_Y(x \sim y) \in Y.

Then if we think of R=graph(f)R = graph(f) we may read the relation as “there is a trajectory from an incoming configuration x 1x_1 to an outgoing configuration x 2x_2

graph(f) incoming outgoing X 1 X 2. \array{ && graph(f) \\ & {}^{\mathllap{incoming}}\swarrow && \searrow^{\mathrlap{outgoing}} \\ X_1 && && X_2 } \,.

Notice here that the defining property of a relation as a subset/subobject translates into the property of classical physics that there is at most one trajectory from some incoming configuration x 1x_1 to some outgoing trajectory x 2x_2 (for a fixed parameter time interval at least, we will formulate this precisely in the next section when we genuinely consider Hamiltonian correspondences).

In a more general context one could consider there to be several such trajectories, and even a whole smooth set of such trajectories between given incoming and outgoing configurations. Each such trajectory would “relate” x 1x_1 to x 2x_2, but each in a possible different way. We can also say that each trajectory makes x 1x_1 correspond to x 2x_2 in a different way, and that is the mathematical term usually used:


For X,YHX, Y \in \mathbf{H} two spaces, a correspondence between them is a diagram in H\mathbf{H} of the form

Z X Y \array{ && Z \\ & \swarrow && \searrow \\ X && && Y }

with no further restrictions. Here ZZ is also called the correspondence space.

An equivalence between two such correspondences is an equivalence Z 1Z 2Z_1 \stackrel{\simeq}{\to}Z_2 that gives a commuting diagram of the form

Z 1 X Y Z 2 \array{ && Z_1 \\ & {}^{}\swarrow && \searrow^{} \\ X &&\downarrow^{\mathrlap{\simeq}} && Y \\ & {}_{}\nwarrow && \nearrow_{} \\ && Z_2 }

Correspondences between XX any YY with such equivalences between them form a groupoid. (See at geometry of physics the section Essence of gauge theory: Groupoids and basic homotopy 1-type theory for more on this.) Hence we write

Corr(H)(X,Y)Grpd. Corr\left(\mathbf{H}\right)(X,Y) \in Grpd \,.

The correspondence induced by the graph of a function f:XYf \colon X \to Y as in example 4 is equivalent, in the sense of def. 7, to the correspondence

X id f X Y. \array{ && X \\ & {}^{\mathllap{id}}\swarrow && \searrow^{\mathrlap{f}} \\ X && && Y } \,.

The equivalence

X id f X Y i X i Y graph(f) \array{ && X \\ & {}^{\mathllap{id}}\swarrow && \searrow^{\mathrlap{f}} \\ X &&\downarrow^{\mathrlap{\simeq}} && Y \\ & {}_{\mathllap{i_X}}\nwarrow && \nearrow_{\mathrlap{i_Y}} \\ && graph(f) }

is induced by

Xgraph(f) X \stackrel{\simeq}{\longrightarrow} graph(f)
x(x,f(x)) x \mapsto (x,f(x))

Moreover, if we think of correspondences as modelling spaces of trajectories, then it is clear that their should be a notion of composition:


Given two consecutive correspondences, then their composite is the correspondence obtained by forming the fiber product of the two coincident morphisms:

( Y 1 Y 2 X 1 X 2 X 3)( Y 1 X 2Y 2 X 1 X 3). \left( \array{ && Y_1 &&&& Y_2 \\ & \swarrow && \searrow && \swarrow && \searrow \\ X_1 && && X_2 && && X_3 } \right) \;\;\;\; \mapsto \;\;\;\; \left( \array{ && Y_1 \circ_{X_2} Y_2 \\ & \swarrow && \searrow \\ X_1 && && X_3 } \right) \,.

Heuristically, the composite space of trajectories Y 1 X 2Y 2Y_1 \circ_{X_2} Y_2 should consist precisely of those pairs of trajectories (f,g)Y 1×Y 2( f, g ) \in Y_1 \times Y_2 such that the endpoint of ff is the starting point of gg. The space with this property is precisely the fiber product of Y 1Y_1 with Y 2Y_2 over X 2X_2, denoted Y 1×X 2Y 2Y_1 \underset{X_2}{\times} Y_2 (also called the pullback of Y 2X 2Y_2 \longrightarrow X_2 along Y 1X 2Y_1 \longrightarrow X_2 and then abbreviated (pb)(pb)):

( Y 1 X 2Y 2 X 1 X 3)=( Z 1×YZ 2 Z 1 (pb) Z 2 X Y Z). \left( \array{ && Y_1 \circ_{X_2} Y_2 \\ & \swarrow && \searrow \\ X_1 && && X_3 } \right) \;\;\; = \;\;\; \left( \array{ && && Z_1 \underset{Y}{\times} Z_2 \\ && & \swarrow && \searrow \\ && Z_1 && (pb) && Z_2 \\ & \swarrow && \searrow && \swarrow && \searrow \\ X && && Y && && Z } \right) \,.

Hence given a topos H\mathbf{H}, correspondences between its objects form a category which composition the fiber product operation, where however the collection of morphisms between any two objects is not just a set, but is a groupoid (the groupoid of correspondences between two given objects and equivalences between them).

One says that correspondences form a (2,1)-category

Corr(H)(2,1)Cat. Corr(\mathbf{H}) \in (2,1)Cat \,.

But for most purposes here, the reader unwilling to enter higher category theory can, to good approximation, pretend that correspondences form an ordinary category.

One reason for formalizing this notion of correspondences so much in the present context that it is useful now to apply it not just to the ambient topos H\mathbf{H} of smooth sets, but also to its slice topos H /Ω cl 2\mathbf{H}_{/\mathbf{\Omega}_{cl}^2} over the universal moduli space of closed differential 2-forms.

To see how this is useful in the present context, notice the following basic observation:


Given a symplectic manifold (X,ω)(X,\omega), then a submanifold

LX L \hookrightarrow X

is called


Let f:X 1X 2f \colon X_1 \to X_2 be a smooth function between smooth manifolds and let

graph(f) X 1×X 2 p 1 p 2 X 1 X 2 \array{ && graph(f) \\ && \downarrow \\ && X_1 \times X_2 \\ & {}^{\mathllap{p_1}}\swarrow && \searrow^{\mathrlap{p_2}} \\ X_1 && && X_2 }

be the induced correspondence. If ω 1\omega_1 and ω 2\omega_2 are symplectic forms on X 1X_1 and X 2X_2, respectively, then p 1 *ω 1p 2 *ω 2p_1^\ast \omega_1 - p_2^\ast \omega_2 is a pre-symplectic form on X 1×X 2X_1 \times X_2, and ff is a symplectomorphism precisely if graph(f)X 1×X 2graph(f) \hookrightarrow X_1 \times X_2 is a Lagrangian submanifold.

To capture this phenomenon, one traditionally sets:


For (X 1,ω 1)(X_1,\omega_1) and (X 2,ω 2)(X_2,\omega_2) two symplectic manifolds (not necessarily of the same dimension), an isotropic correspondence or Lagrangian correspondence between them is a correspondence of the underlying manifolds

R X 1×X 2 p 1 p 2 X 1 X 2 \array{ && R \\ && \downarrow \\ && X_1 \times X_2 \\ & {}^{\mathllap{p_1}}\swarrow && \searrow^{\mathrlap{p_2}} \\ X_1 && && X_2 }

such that the correspondence space RX 1×X 2R \hookrightarrow X_1 \times X_2 is an isotropic submanifold or Lagrangian submanifold, respectively of the product symplectic manifold given by

(X 1×X 2,p 1 *ω 1p 2 *ω 2). (X_1 \times X_2 , p_1^\ast \omega_1 - p_2^\ast \omega_2) \,.

Under the identification of prop. 2, isotropic correspondences as in def. 10 are equivalent to diagrams of smooth sets of the form

R p 1 p 2 X 1 = X 2 ω 1 ω 2 Ω cl 2. \array{ && R \\ & {}^{\mathllap{p_1}}\swarrow && \searrow^{\mathrlap{p_2}} \\ X_1 && \swArrow_= && X_2 \\ & {}_{\mathllap{\omega_1}}\searrow && \swarrow_{\mathrlap{\omega_2}} \\ && \Omega^2_{cl} } \,.

This in turn is equivalent to being a correspondence in the slice topos H /Ω cl 2\mathbf{H}_{/\Omega^2_{cl}}, def. 4, under the identification of prop. 4.


By prop. 3 the commutativity of this diagram says precisely that on RR we have

p 1 *ω 1=p 2 *ω 2 p_1^\ast \omega_1 = p_2^\ast \omega_2


p 1 *ω 1p 2 *ω 2=0. p_1^\ast \omega_1 - p_2^\ast \omega_2 = 0 \,.

Therefore we have:


For (X 1,ω 2)(X_1, \omega_2) and (X 2,ω 2)(X_2, \omega_2) two symplectic manifolds, there is a full embedding

LagrangianCorrespondences((X 1,ω 1),(X 2,ω 2))Corr(H /Ω cl 2)((X 1,ω 1),(X 2,ω 2)) LagrangianCorrespondences\left(\left(X_1,\omega_1\right), \left(X_2, \omega_2\right)\right) \hookrightarrow Corr\left(\mathbf{H}_{/\mathbf{\Omega}^2_{cl}}\right)\left(\left(X_1,\omega_1\right), \left(X_2, \omega_2\right)\right)

of the Lagrangian correspondences into the space of correspondences between the two manifolds as objects in the slice topos over the universal moduli space of closed differential 2-forms.


The graph of a function f:X 1X 2f \colon X_1\to X_2 between symplectic manifold (X i,ω i)(X_i, \omega_i) is a Lagrangian correspondence precisely if ff is a symplectomorphism.


Under the identification of

graph(f) p 1 p 2 X 1 X 2 \array{ && graph(f) \\ & {}^{\mathllap{p_1}}\swarrow && \searrow^{\mathrlap{p_2}} \\ X_1 && && X_2 }

Hamiltonian (time evolution) trajectories and Hamiltonian correspondences

An important class of symplectomorphisms are the following


Let (X,ω)(X,\omega) be a symplectic manifold. The induced Poisson bracket {,}\{-,-\} takes a smooth function HC (X)H \in C^\infty(X) (the “Hamiltonian”) to the derivation {H,}\{H,-\} on C (X)C^\infty(X). This is equivalently a vector field v HΓTXv_H \in \Gamma T X, the corresponding Hamiltonian vector field.

A Hamiltonian symplectomorphisms from a symplectic manifold (X,ω)(X,\omega) to itself, is a symplectomorphism XXX \to X which is the flow of a Hamiltonian vector field for some parameter “time” tt \in \mathbb{R}

exp(t{H,}):(X,ω)(X,ω). \exp( t \{H,-\}) \;\colon\; (X,\omega) \longrightarrow (X,\omega) \,.

We call a Lagrangian correspondence, def. 10, induced from Hamiltonian symplectomorphisms a Hamiltonian correspondences.


Under the interpretation of correspondences as spaces of trajectories as in example 4, the smooth correspondence space of a Hamiltonian correspondence is naturally identified with the space of classical trajectories of the Hamiltonian dynamics of HH

Fields traj class(t)=graph(exp(t{H,})) Fields_{traj}^{class}(t) = graph\left( \exp(t\{H,-\}) \right)

in that

  1. every point in the space corresponds uniquely to a trajectory of parameter time length tt characterized as satisfying the equations of motion as given by Hamilton's equations for HH;

  2. the two projection maps to XX send a trajectory to its initial and to its final configuration, respectively.


Forming Hamiltonian correspondences consitutes a functor from 1-dimensional cobordisms with Riemannian structure to the category of correspondences in the slice topos:

exp((){H,}):Bord 1 RiemCorr 1(H /Ω 2) \exp((-)\{H,-\}) \;\colon\; Bord^{Riem}_1 \longrightarrow Corr_1(\mathbf{H}_{/\Omega^2})

since for all (“time”) parameter valued t 1,t 2t_1, t_2 \in \mathbb{R} we have a composition (by fiber product) of correspondences exhibited by the following pasting diagram:

graph(exp((t 1+t 2)){H,}) graph(exp(t 1){H,}) (pb) graph(exp(t 2{H,})) X X X Ω cl 2. \array{ &&&& graph\left(\exp\left(\left(t_1+t_2\right)\right) \left\{H,-\right\} \right) \\ && & \swarrow && \searrow \\ && graph\left(\exp\left(t_1\right)\left\{H,-\right\}\right) && (pb) && graph\left(\exp\left(t_2 \left\{H,-\right\}\right)\right) \\ & \swarrow && \searrow & & \swarrow && \searrow \\ X && && X && && X \\ & \searrow &&& \downarrow &&& \swarrow \\ &&&& \Omega^2_{cl} } \,.

The kinetic action and Planck’s constant

To naturally see why there would be any Hamiltonian associated to a (to some) symplectomorphism in the first place, we step back and consider local trivializations or local potentials for symplectic forms. Doing so turns out to give rise to what in physics is called the kinetic action, what in the context of geometric quantization is called prequantization and what in mathematics is called lifting to differential cohomology. All these concepts arise directly from the following simple consideration.

Given a pre-symplectic form ωΩ cl 2(X)\omega \in \Omega^2_{\mathrm{cl}}(X) , by the Poincaré lemma there is a good open cover {U iX} i\{U_i \hookrightarrow X\}_i such that one can find smooth differential 1-forms θ iΩ 1(U i)\theta_i \in \Omega^1(U_i) such that these are local trivializations/potentials for the symplectic form on each patch U iU_i of the cover:

dθ i=ω |U i. \mathbf{d}\theta_i = \omega_{|U_i} \,.

Physically such a 1-form is (up to a factor of 2) a choice of kinetic energy density called a kinetic Lagrangian L kinL_{\mathrm{kin}} (below in example 7 we connect this statement to a maybe more familiar formla):

θ i=2L kin,i. \theta_i = 2 L_{\mathrm{kin}, i} \,.

Consider the phase space ( 2,ω=dqdp)(\mathbb{R}^2, \; \omega = \mathbf{d} q \wedge \mathbf{d} p) of example 1. Since 2\mathbb{R}^2 is a contractible topological space we consider the trivial covering ( 2\mathbb{R}^2 covering itself) since this is already a good covering in this case. Then all the {g ij}\{g_{i j}\} are trivial and the data of a prequantization consists simply of a choise of 1-form θΩ 1( 2)\theta \in \Omega^1(\mathbb{R}^2) such that

dθ=dqdp. \mathbf{d}\theta = \mathbf{d}q \wedge \mathbf{d}p \,.

A standard such choice is

θ=pdq. \theta = - p \wedge \mathbf{d}q \,.

Then given a trajectory γ:[0,1]X\gamma \colon [0,1] \longrightarrow X which satisfies Hamilton's equation for a standard kinetic energy term, then (pdq)(γ˙)(p \mathbf{d}q)(\dot\gamma) is this kinetic energy of the particle which traces out this trajectory.

Given a path γ:[0,1]X\gamma : [0,1] \to X in phase space, its kinetic action S kinS_{\mathrm{kin}} is supposed to be the integral of L kinL_{\mathrm{kin}} along this trajectory. In order to make sense of this in the generality where there is no globally defined θ\theta, there need to be functions g ijC (U iU j,)g_{i j} \in C^\infty(U_i \cap U_j, \mathbb{R}) for each double intersection of patches of the cover, such that these the local θ\theta‘s differ on these double intersection only by the total derivative (de Rham differential d\mathbf{d} ) of these functions:

θ j| U jθ i| U i=dg ij. \theta_j|_{U_j} - \theta_i|_{U_i} = \mathbf{d}g_{i j} \,.

One then finds (from the theory of Cech cohomology) that if on triple intersections these functions satisfy

g ij+g jk=g ik g_{ij} + g_{j k} = g_{i k}

then there is a well defined action functional

S kin(γ) S_{\mathrm{kin}}(\gamma) \in \mathbb{R}

obtained by dividing γ\gamma into small pieces that each map to a single patch U iU_i, integrating θ i\theta_i along this piece, and adding the contribution of g ijg_{i j} at the point where one switches from using θ i\theta_i to using θ j\theta_j. Technically this is called the holonomy or parallel transport of the (,+)(\mathbb{R},+)-principal connection which is defined by the data ({θ i},{g ij})(\{\theta_i\}, \{g_{i j}\} ).

However, requiring this condition on triple overlaps as an equation between \mathbb{R}-valued functions makes the local patch structure trivial: if this is possible then one can in fact already find a single θΩ 1(X)\theta \in \Omega^1(X) and functions h iC (U i,)h_i \in C^\infty(U_i, \mathbb{R}) such that θ i=θ| U i+dh i\theta_i = \theta|_{U_i} + \mathbf{d}h_i. This has the superficially pleasant effect that the action is simply the integral against this globally defined 1-form, S kin= [0,1]γ *L kinS_{\mathrm{kin}} = \int_{[0,1]} \gamma^\ast L_{\mathrm{kin}}, but it also means that the pre-symplectic form ω\omega is exact, which is not the case in many important examples. (In more abstract terms what this is saying is that every (,+)(\mathbb{R},+)-principal bundle over a manifolds is trivializable.)

On the other hand, what really matters in prequantum physics is not the action functional S kinS_{\mathrm{kin}} \in \mathbb{R} itself, but the exponentiated action

exp(iS)/(2π), \exp\left( \tfrac{i}{\hbar} S \right) \in \mathbb{R}/(2\pi \hbar)\cdot\mathbb{Z} \,,

which takes values in the quotient of the additive group of real numbers by integral multiples of Planck's constant 2π2\pi \hbar.

In more detail, consider the canonical inclusion

\mathbb{Z} \hookrightarrow \mathbb{R}

of the integers as an addiditve subgroup of the real numbers. Strictly speaking what appears in physics is the real line on which a unit is chosen as part of the identification of mathematical formalism with physical reality, one should really consider all possible additive group homomorphisms \mathbb{Z}\to \mathbb{R}. These are parameterized by

h({0}) h \in (\mathbb{R}- \{0\}) \hookrightarrow \mathbb{R}
()h: (-)\cdot h \;\colon\; \mathbb{Z} \longrightarrow \mathbb{R}

and this “physical unithh is what is called Planck’s constant.

In particular the induced circle group is identified as the quotient of \mathbb{R} by hh \mathbb{Z}, in this sense

U(1)/h U(1) \simeq \mathbb{R}/h \mathbb{Z}

and under this identification its quotient map is expressed in terms of the exponential function exp:z k=0 z kk!\exp \colon z \mapsto \sum_{k = 0}^\infty \frac{z^k}{k!} \in \mathbb{C} as

exp(2πih())=exp(i()):U(1), \exp(2 \pi \tfrac{i}{h}(-)) = \exp(\tfrac{i}{\hbar} (-)) \;\colon\; \mathbb{R} \longrightarrow U(1) \,,


h/2π. \hbar \coloneqq h/2\pi \,.

The resulting short exact sequence is the real exponential exact sequence

0exp(i())U(1)0. 0 \to \mathbb{Z} \longrightarrow \mathbb{R} \stackrel{\exp(\tfrac{i}{\hbar}(-))}{\longrightarrow} U(1) \to 0 \,.

This is the source of the ubiquity of the expression exp(i())\exp(\tfrac{i}{\hbar} (-)) in quantum physics, say in the path integral, where the exponentiated action functional appears as exp(iS)\exp(\tfrac{i}{\hbar} S).

Pre-Quantization and Differential cohomology

By the above discussion, for the exponentiated kinetic action functional to be well defined, one only needs that the equation g ij+g jk=g ikg_{i j} + g_{j k} = g_{i k} on triple intersection holds modulo addition of an integral multiple of Planck's constant h=2πh = 2\pi \hbar.

If this is the case, then one says that the data ({θ i},{g ij})(\{\theta_i\}, \{g_{i j}\}) defines equivalently

on XX, with curvature the given symplectic 2-form ω\omega.

Such data is called a pre-quantization of the symplectic manifold (X,ω)(X,\omega). Since it is the exponentiated action functional exp(iS)\exp(\frac{i}{\hbar} S) that enters the quantization of the given mechanical system (for instance as the integrand of a path integral), the prequantization of a symplectic manifold is indeed precisely the data necessary before quantization.

Therefore, in the spirit of the above discussion of pre-symplectic structures, we would like to refine the smooth moduli space of closed differential 2-forms to a moduli space of prequantized differential 2-forms.

Again this does naturally exist if only we allow for a good notion of “space”. An additional phenomenon to be taken care of now is that while pre-symplectic forms are either equal or not, their pre-quantizations can be different and yet be equivalent:

because there is still a remaining freedom to change this data without changing the exponentiated action along a closed path: we say that a choice of functions h iC (U i,/(2π))h_i \in C^\infty(U_i, \mathbb{R}/(2\pi\hbar)\mathbb{Z}) defines an equivalence between ({θ i},{g ij})(\{\theta_i\}, \{g_{i j}\}) and ({θ˜ i},{g˜ ij})(\{\tilde \theta_i\}, \{\tilde g_{i j}\}) if θ˜ iθ i=dh i\tilde \theta_i - \theta_i = \mathbf{d}h_i and g˜ ijg ij=h jh i\tilde g_{i j} - g_{i j} = h_j - h_i.

This means that the space of prequantizations of (X,ω)(X,\omega) is similar to an orbifold: it has points which are connected by gauge equivalences: there is a groupoid of pre-quantum structures on a manifold XX.

In just the same way then that above we found a smooth moduli space Ω cl 2\mathbf{\Omega}^2_{cl} of closed differential 2-forms, one can find a smooth groupoid (for more on this see at geometry of physics the section Smooth homotopy types ), which we denote

BU(1) connH \mathbf{B}U(1)_{\mathrm{conn}} \in \mathbf{H}

The smooth groupoid BU(1) conn\mathbf{B}U(1)_{\mathrm{conn}} is characterized as follows,

  1. For XX a smooth manifold, maps

    :XBU(1) conn \nabla \colon X \longrightarrow \mathbf{B}U(1)_{conn}

    are equivalent to the above prequantum data ({θ i},{g ij})(\{\theta_i\}, \{g_{i j}\}) on XX;

  2. for 1, 2:XBU(1) conn\nabla_1, \nabla_2 \colon X \longrightarrow \mathbf{B}U(1)_{conn} two such maps, homotopies

    1 X BU(1) conn 2 \array{ & \nearrow \searrow^{\mathrlap{\nabla_1}} \\ X & \Downarrow & \mathbf{B}U(1)_{conn} \\ & \searrow \nearrow_{\mathrlap{\nabla_2}} }

    between these are equivalent to the above gauge transformations ({h i})(\{h_i\}) between this data

(θ 2) i(θ 1) i=dilog(h i). (\theta_2)_i - (\theta_1)_i = \mathbf{d} \tfrac{\hbar}{i} log (h_i) \,.

There is a universal curvature map, a morphism of smooth groupoids

F:BU(1) connΩ cl 2 F \;\colon\; \mathbf{B}U(1)_{\mathrm{conn}} \longrightarrow \mathbf{\Omega}^2_{\mathrm{cl}}

which is such that for :XBU(1) conn\nabla \colon X \longrightarrow \mathbf{B}U(1)_{conn} a U(1)U(1)-principal connection, the composite

F :XBU(1) connF ()Ω cl 2 F_\nabla \;\colon\; X \stackrel{\nabla}{\longrightarrow} \mathbf{B}U(1)_{conn} \stackrel{F_{(-)}}{\longrightarrow} \mathbf{\Omega}^2_{cl}

is its curvature 2-form.

Hence this is the map that sends ({θ i},{g ij})(\{\theta_i\}, \{g_{i j}\}) to ω\omega with ω| U i=dθ i\omega|_{U_i} = \mathbf{d}\theta_i.



A prequantization of a symplectic manifold (X,ω)(X,\omega) is – if it exists – a choice of circle group-principal connection \nabla on XX whose curvature 2-form is the given symplectic form

F =ω. F_\nabla = \omega \,.

In terms of the classifying morphism of differential forms as in prop. 2 this reads as follows.


Given a presymplectic manifold (X,ω)(X,\omega), regarded equivalently as an object (XωΩ cl 2)H /Ω cl 2(X \stackrel{\omega}{\longrightarrow} \mathbf{\Omega}^2_{cl}) \in \mathbf{H}_{/\mathbf{\Omega}^2_{cl}} by prop. 4, then a prequantization of (X,ω)(X,\omega), def. 12, is equivalently a choice of lift \nabla in

X BU(1) conn ω F () Ω cl 2. \array{ X &\stackrel{\nabla}{\longrightarrow}& \mathbf{B}U(1)_{conn} \\ & {}_{\mathllap{\omega}}\searrow & \downarrow^{\mathrlap{F_{(-)}}} \\ && \mathbf{\Omega}^2_{cl} } \,.

Phrased this way, there is an evident concept of prequantization of Lagrangian correspondences:


Given prequantized symplectic manifolds (X i, i)(X_i,\nabla_i) as in prop. 10, and given a Lagrangian correspondence as in prop. 6, then a prequantization of this correspondence is a lift of the whole diagram through the universal curvature map of prop. 9:

Z X 1 X 2 ω 1 ω 2 Ω cl 2 Z X 1 X 2 1 2 BU(1) conn F Ω cl 2. \array{ && Z \\ & \swarrow && \searrow \\ X_1 && && X_2 \\ & {}_{\mathllap{\omega_1}}\searrow && \swarrow_{\mathrlap{\omega_2}} \\ && \mathbf{\Omega}^2_{cl} } \;\;\;\; \mapsto \;\;\;\; \array{ && Z \\ & \swarrow && \searrow \\ X_1 && \swArrow_{\simeq} && X_2 \\ & {}_{\mathllap{\nabla_1}}\searrow && \swarrow_{\mathrlap{\nabla_2}} \\ && \mathbf{B}U(1)_{conn} \\ && \downarrow^{\mathrlap{F}} \\ && \mathbf{\Omega}^2_{cl} } \,.

This means in words that a prequantized Lagrangian correspondence is a prequantization of the in- and out-going symplectic manifolds together with a choice of equivalence/gauge transformation between the two prequantum circle bundles pulled back to the correspondences space.


By duality in the smmetric monoidal (2,1)-category of correspondences, a prequantized Lagragian correspondence is equivalently a diagram of the form

Y * X 1×X 2 2 BU(1) conn \array{ && Y \\ & \swarrow && \searrow \\ \ast && \swArrow && X_1 \times X_2 \\ & \searrow && \swarrow_{\mathrlap{\nabla_2- \nabla}} \\ \\ && \mathbf{B}U(1)_{conn} }

hence a trivialization of the product of one prequantum bundle with the negative (the inverse under tensor product) of the other, on the correspondence space.

Hamiltonian flows, the Legendre transform and the Hamilton-Jacobi action


Consider the phase space ( 2,ω=dqdp)(\mathbb{R}^2, \; \omega = \mathbf{d} q \wedge \mathbf{d} p) of example 1 equipped with its canonical prequantization by θ=pdq\theta = p \, \mathbf{d}q from example 7,

Then smooth 1-parameter flows of this data via prequantized correspondences, def. 13,

tX f t X θ F t θ BU(1) conn t \;\;\;\; \mapsto \;\;\;\; \array{ X && \stackrel{f_t}{\longrightarrow} && X \\ & {}_{\mathllap{\theta}} \searrow & \swArrow_{ F_t } & \swarrow_{\mathrlap{\theta}} \\ && \mathbf{B}U(1)_{conn} }

are in bijection with smooth functions H: 2H \colon \mathbb{R}^2 \longrightarrow \mathbb{R}.

This bijection works by regarding HH as a Hamiltonian, def. 11, and assigning the flow f t=exp(t{H,})f_t = \exp(t \{H,-\}) of its Hamiltonian vector field

tX exp(t{H,}) X θ exp(iS t) θ BU(1) conn, t \;\; \mapsto \;\; \array{ X && \stackrel{\exp(t \{H,-\})}{\longrightarrow} && X \\ & {}_{\mathllap{\theta}} \searrow & \swArrow_{\exp( \tfrac{i}{\hbar} S_t )} & \swarrow_{\mathrlap{\theta}} \\ && \mathbf{B}U(1)_{conn} } \,,

where the prequantization is given by

  • S t: 2S_t \;\colon\; \mathbb{R}^2 \longrightarrow \mathbb{R} is the Hamilton-Jacobi action of the classical trajectories induced by HH,

  • which is the integral S t= 0 tLdtS_t = \int_{0}^t L \, d t of the Lagrangian LdtL \,d t induced by HH,

  • which is the Legendre transform of the Hamiltonian

    LpHpH: 2. L \coloneqq p \frac{\partial H}{\partial p} - H \;\colon\; \mathbb{R}^2 \longrightarrow \mathbb{R} \,.

By prop. 8 the prequantum filler of the diagram is given by a function F t=exp(iS t)F_t =\exp(\tfrac{i}{\hbar} S_t) satisfying

f t *θθ=dS t. f_t^\ast \theta - \theta = -\mathbf{d}S_t \,.

By standard Lie theory a smooth such 1-parameter flow is fixed by its derivative by tt. For the above equation this yields

vθ=dL \mathcal{L}_v \theta = -\mathbf{d}L


  1. vΓ(TX)v \in \Gamma(T X) is the vector field of the flow tf tt\mapsto f_t;

  2. v\mathcal{L}_v is the Lie derivative along vv;

  3. LStL \coloneqq \frac{\partial S}{\partial t}.

By Cartan's magic formula this equation is equivalent to

ι vω=dLdι vθ. \iota_v \omega = -\mathbf{d}L - \mathbf{d} \iota_v \theta \,.

This is the symplectic form of Hamilton's equations for vv and says that

HLι vθ H \coloneqq - L - \iota_v \theta

is a Hamiltonian that makes vv a Hamiltonian vector field. The correction term is

ι vθ =ι v(pdq) =p vq . \begin{aligned} \iota_v \theta &= \iota_v ( p \, \mathbf{d}q ) \\ & = p \partial_v q \\ \end{aligned} \,.

But since vv is Hamiltonian, this is given by one component of Hamilton's equations ι v(dpdq)=dH\iota_v (\mathbf{d}p \wedge \mathbf{d}q) = \mathbf{d}H saying that vq=Hp\partial_v q = \frac{\partial H}{\partial p}.

Hence in summary the flow is Hamiltonian and the pre-quntum filler is the choice of Hamiltonian HH specified by

St=L=pHpH. \frac{\partial S}{\partial t} = L = p \frac{\partial H}{\partial p} - H \,.

In particular, this induces a functor

exp(iS):Bord 1 RiemCorr 1(H /BU(1) conn). \exp(\tfrac{i}{\hbar} S) \;\colon\; Bord_1^{Riem} \longrightarrow Corr_1(\mathbf{H}_{/\mathbf{B}U(1)_{conn}}) \,.

In summary, prop. 11 and remark 2 say that a prequantized Lagrangian correspondence is conceptually of the following form

spaceoftrajectories initialvalues Hamiltonianevolution phasespace in actionfunctional phasespace out prequantumbundle in prequantumbundle out 2groupofphases. \array{ && {{space\,of} \atop {trajectories}} \\ & {}^{\mathllap{{initial}\atop {values}}}\swarrow && \searrow^{\mathrlap{{Hamiltonian} \atop {evolution}}} \\ phase\,space_{in} && \swArrow_{{action} \atop {functional}} && phase \,space_{out} \\ & {}_{\mathllap{{prequantum}\atop {bundle}_{in}}}\searrow && \swarrow_{\mathrlap{{prequantum} \atop {bundle}_{out}}} \\ && {{2-group} \atop {of\,phases}} } \,.

The proof of prop. 11 recovers, from general abstract input, precisely all the ingredients known in physics as canonical transformations.

The proposition says that the slice topos H /BU(1) conn\mathbf{H}_{/\mathbf{B}U(1)_{conn}} unifies classical mechanics in its two incarnations as Hamiltonian mechanics and as Lagrangian mechanics, where the relation between the two via the Legendre transform is exhibited by the homotopies that fill diagrams in the slice topos over BU(1) conn\mathbf{B}U(1)_{conn}.

Hamiltonian\leftarrow Legendre transform \rightarrowLagrangian
Lagrangian correspondenceprequantizationprequantized Lagrangian correspondence

Heisenberg group and Poisson bracket

Above we have interpreted maps f:XYf \colon X \to Y as correspondences between XX and YY by taking the correspondence space to be the graph of ff. There is also another natural way to regard maps as correspondences: we may simply take XX as the correspondence space, take the left map out of it to be the identity and the right map to be ff itself:

(XfY)( X id f X Y). \left( X \stackrel{f}{\longrightarrow} Y \right) \;\; \mapsto \;\; \left( \array{ && X \\ & {}^{\mathllap{id}}\swarrow && \searrow^{\mathrlap{f}} \\ X && && Y } \right) \,.

Consider now those correspondences which are equivalences (isomorphisms) in the category of correspondences Corr 1(H)Corr_1(\mathbf{H}). If we forget the smooth structure on everything and consider just correspondences of the underlying sets, hence Corr 1(Set)Corr_1(Set), then it is easy to see that under the cardinality map correspondences are given by matrices with cardinality entries and composition of correspondence by fiber product induces matrix multiplication.

Therefore for a correspondence to be an equivalence-transformation it has to be of the form above, induced by a direct map, which in addition is an equivalence f:XYf \colon X \stackrel{\simeq}{\longrightarrow} Y.


Let (X,ω)(X,\omega) be a symplectic manifold and choose any prequantization (L,)(L,\nabla), thought of, via remark \ref{PrequantizationIsLiftThroughCurvatureBaseChange}, as an object in the slice (2,1)-topos, H /BU(1) conn\nabla \in \mathbf{H}_{/\mathbf{B}U(1)_{conn}}. Then

See (hgp 13)

For some reason, the quantomorphism group which is the Lie integration of the Poisson bracket is less famous than the Heisenberg group that sits inside it:


Suppose that (X,ω)(X,\omega) itself has the structure of a group (for instance if (X,ω)(X,\omega) is a symplectic vector space such as ( 2n, ip idq i)(\mathbb{R}^{2n}, \sum_i p_i \mathbf{d}q^i) ), then the subgroup of the quantomorphism group whose underlying diffeomorphisms are given by the action of XX is the Heisenberg group of XX.

Hamiltonian actions and moment maps

For GG a Lie group, a Hamiltonian action of GG on (X,ω)(X,\omega) is equivalently an action by prequantized Lagrangian correspondences, hence a group homomorphism

GAut (Corr 1(H /BU(1) conn)). G \longrightarrow \mathbf{Aut}_\nabla(Corr_1(\mathbf{H}_{/\mathbf{B}U(1)_{conn}})) \,.

The Lie differentiation of this is the corresponding moment map.

See (hgp 13)

Semantic Layer

We now discuss the above constructions more abstractly in cohesive topos theory.

under construction

given VV then the delooping BAut(V)\mathbf{B} \mathbf{Aut}(V) of its automorphism group Aut(V)\mathbf{Aut}(V) is the 1-image factorization of the name *VType\ast \stackrel{\vdash V}{\longrightarrow} Type of VV

*BAut(V)Type. \ast \stackrel{}{\longrightarrow} \mathbf{B}\mathbf{Aut}(V) \hookrightarrow Type \,.

for the slice over BU(1) conn\mathbf{B}U(1)_{conn} this needs to be subjected to differential concretification


Syntactic Layer

We now discuss the above constructions yet more abstractly in homotopy type theory.


prequantization is lift through dependent sum along the universal curvature map of prop. 9

F ():H /BU(1) connH /Ω cl 2 \underset{F_{(-)}}{\sum} \;\colon\; \mathbf{H}_{/\mathbf{B}U(1)_{conn}} \longrightarrow \mathbf{H}_{/\mathbf{\Omega}^2_{cl}}



As far as it is not covered by traditional material, the above discussion is taken from

For more and related references see there and see at motivic quantization.

Last revised on November 14, 2014 at 08:04:00. See the history of this page for a list of all contributions to it.