Moment map

From HandWiki

In mathematics, specifically in symplectic geometry, the momentum map (or, by false etymology, moment map[1]) is a tool associated with a Hamiltonian action of a Lie group on a symplectic manifold, used to construct conserved quantities for the action. The momentum map generalizes the classical notions of linear and angular momentum. It is an essential ingredient in various constructions of symplectic manifolds, including symplectic (Marsden–Weinstein) quotients, discussed below, and symplectic cuts and sums.

Formal definition

Let M be a manifold with symplectic form ω. Suppose that a Lie group G acts on M via symplectomorphisms (that is, the action of each g in G preserves ω). Let [math]\displaystyle{ \mathfrak{g} }[/math] be the Lie algebra of G, [math]\displaystyle{ \mathfrak{g}^* }[/math] its dual, and

[math]\displaystyle{ \langle, \rangle : \mathfrak{g}^* \times \mathfrak{g} \to \mathbf{R} }[/math]

the pairing between the two. Any ξ in [math]\displaystyle{ \mathfrak{g} }[/math] induces a vector field ρ(ξ) on M describing the infinitesimal action of ξ. To be precise, at a point x in M the vector [math]\displaystyle{ \rho(\xi)_x }[/math] is

[math]\displaystyle{ \left.\frac{d}{dt}\right|_{t = 0} \exp(t \xi) \cdot x, }[/math]

where [math]\displaystyle{ \exp : \mathfrak{g} \to G }[/math] is the exponential map and [math]\displaystyle{ \cdot }[/math] denotes the G-action on M.[2] Let [math]\displaystyle{ \iota_{\rho(\xi)} \omega \, }[/math] denote the contraction of this vector field with ω. Because G acts by symplectomorphisms, it follows that [math]\displaystyle{ \iota_{\rho(\xi)} \omega \, }[/math] is closed (for all ξ in [math]\displaystyle{ \mathfrak{g} }[/math]).

Suppose that [math]\displaystyle{ \iota_{\rho(\xi)} \omega \, }[/math] is not just closed but also exact, so that [math]\displaystyle{ \iota_{\rho(\xi)} \omega = d H_\xi }[/math] for some function [math]\displaystyle{ H_\xi }[/math]. Suppose also that the map [math]\displaystyle{ \mathfrak{g} \to C^\infty(M) }[/math] sending [math]\displaystyle{ \xi \mapsto H_\xi }[/math] is a Lie algebra homomorphism. Then a momentum map for the G-action on (M, ω) is a map [math]\displaystyle{ \mu : M \to \mathfrak{g}^* }[/math] such that

[math]\displaystyle{ d(\langle \mu, \xi \rangle) = \iota_{\rho(\xi)} \omega }[/math]

for all ξ in [math]\displaystyle{ \mathfrak{g} }[/math]. Here [math]\displaystyle{ \langle \mu, \xi \rangle }[/math] is the function from M to R defined by [math]\displaystyle{ \langle \mu, \xi \rangle(x) = \langle \mu(x), \xi \rangle }[/math]. The momentum map is uniquely defined up to an additive constant of integration.

A momentum map is often also required to be G-equivariant, where G acts on [math]\displaystyle{ \mathfrak{g}^* }[/math] via the coadjoint action. If the group is compact or semisimple, then the constant of integration can always be chosen to make the momentum map coadjoint equivariant. However, in general the coadjoint action must be modified to make the map equivariant (this is the case for example for the Euclidean group). The modification is by a 1-cocycle on the group with values in [math]\displaystyle{ \mathfrak{g}^* }[/math], as first described by Souriau (1970).

Hamiltonian group actions

The definition of the momentum map requires [math]\displaystyle{ \iota_{\rho(\xi)} \omega }[/math] to be closed. In practice it is useful to make an even stronger assumption. The G-action is said to be Hamiltonian if and only if the following conditions hold. First, for every ξ in [math]\displaystyle{ \mathfrak{g} }[/math] the one-form [math]\displaystyle{ \iota_{\rho(\xi)} \omega }[/math] is exact, meaning that it equals [math]\displaystyle{ dH_\xi }[/math] for some smooth function

[math]\displaystyle{ H_\xi : M \to \mathbf{R}. }[/math]

If this holds, then one may choose the [math]\displaystyle{ H_\xi }[/math] to make the map [math]\displaystyle{ \xi \mapsto H_\xi }[/math] linear. The second requirement for the G-action to be Hamiltonian is that the map [math]\displaystyle{ \xi \mapsto H_\xi }[/math] be a Lie algebra homomorphism from [math]\displaystyle{ \mathfrak{g} }[/math] to the algebra of smooth functions on M under the Poisson bracket.

If the action of G on (M, ω) is Hamiltonian in this sense, then a momentum map is a map [math]\displaystyle{ \mu : M\to \mathfrak{g}^* }[/math] such that writing [math]\displaystyle{ H_\xi = \langle \mu, \xi \rangle }[/math] defines a Lie algebra homomorphism [math]\displaystyle{ \xi \mapsto H_\xi }[/math] satisfying [math]\displaystyle{ \rho(\xi) = X_{H_\xi} }[/math]. Here [math]\displaystyle{ X_{H_\xi} }[/math] is the vector field of the Hamiltonian [math]\displaystyle{ H_\xi }[/math], defined by

[math]\displaystyle{ \iota_{X_{H_\xi}} \omega = d H_\xi. }[/math]

Examples of momentum maps

In the case of a Hamiltonian action of the circle [math]\displaystyle{ G = \mathcal{U}(1) }[/math], the Lie algebra dual [math]\displaystyle{ \mathfrak{g}^* }[/math] is naturally identified with [math]\displaystyle{ \mathbb{R} }[/math], and the momentum map is simply the Hamiltonian function that generates the circle action.

Another classical case occurs when [math]\displaystyle{ M }[/math] is the cotangent bundle of [math]\displaystyle{ \mathbb{R}^3 }[/math] and [math]\displaystyle{ G }[/math] is the Euclidean group generated by rotations and translations. That is, [math]\displaystyle{ G }[/math] is a six-dimensional group, the semidirect product of [math]\displaystyle{ SO(3) }[/math] and [math]\displaystyle{ \mathbb{R}^3 }[/math]. The six components of the momentum map are then the three angular momenta and the three linear momenta.

Let [math]\displaystyle{ N }[/math] be a smooth manifold and let [math]\displaystyle{ T^*N }[/math] be its cotangent bundle, with projection map [math]\displaystyle{ \pi : T^*N \rightarrow N }[/math]. Let [math]\displaystyle{ \tau }[/math] denote the tautological 1-form on [math]\displaystyle{ T^*N }[/math]. Suppose [math]\displaystyle{ G }[/math] acts on [math]\displaystyle{ N }[/math]. The induced action of [math]\displaystyle{ G }[/math] on the symplectic manifold [math]\displaystyle{ (T^*N, \mathrm{d}\tau) }[/math], given by [math]\displaystyle{ g \cdot \eta := (T_{\pi(\eta)}g^{-1})^* \eta }[/math] for [math]\displaystyle{ g \in G, \eta \in T^*N }[/math] is Hamiltonian with momentum map [math]\displaystyle{ -\iota_{\rho(\xi)} \tau }[/math] for all [math]\displaystyle{ \xi \in \mathfrak{g} }[/math]. Here [math]\displaystyle{ \iota_{\rho(\xi)}\tau }[/math] denotes the contraction of the vector field [math]\displaystyle{ \rho(\xi) }[/math], the infinitesimal action of [math]\displaystyle{ \xi }[/math], with the 1-form [math]\displaystyle{ \tau }[/math].

The facts mentioned below may be used to generate more examples of momentum maps.

Some facts about momentum maps

Let [math]\displaystyle{ G, H }[/math] be Lie groups with Lie algebras [math]\displaystyle{ \mathfrak{g}, \mathfrak{h} }[/math], respectively.

  1. Let [math]\displaystyle{ \mathcal{O}(F), F \in \mathfrak{g}^* }[/math] be a coadjoint orbit. Then there exists a unique symplectic structure on [math]\displaystyle{ \mathcal{O}(F) }[/math] such that inclusion map [math]\displaystyle{ \mathcal{O}(F) \hookrightarrow \mathfrak{g}^* }[/math] is a momentum map.
  2. Let [math]\displaystyle{ G }[/math] act on a symplectic manifold [math]\displaystyle{ (M, \omega) }[/math] with [math]\displaystyle{ \Phi_G : M \rightarrow \mathfrak{g}^* }[/math] a momentum map for the action, and [math]\displaystyle{ \psi : H \rightarrow G }[/math] be a Lie group homomorphism, inducing an action of [math]\displaystyle{ H }[/math] on [math]\displaystyle{ M }[/math]. Then the action of [math]\displaystyle{ H }[/math] on [math]\displaystyle{ M }[/math] is also Hamiltonian, with momentum map given by [math]\displaystyle{ (\mathrm{d}\psi)_{e}^* \circ \Phi_G }[/math], where [math]\displaystyle{ (\mathrm{d}\psi)_{e}^* : \mathfrak{g}^* \rightarrow \mathfrak{h}^* }[/math] is the dual map to [math]\displaystyle{ (\mathrm{d}\psi)_{e} : \mathfrak{h} \rightarrow \mathfrak{g} }[/math] ([math]\displaystyle{ e }[/math] denotes the identity element of [math]\displaystyle{ H }[/math]). A case of special interest is when [math]\displaystyle{ H }[/math] is a Lie subgroup of [math]\displaystyle{ G }[/math] and [math]\displaystyle{ \psi }[/math] is the inclusion map.
  3. Let [math]\displaystyle{ (M_1, \omega_1) }[/math] be a Hamiltonian [math]\displaystyle{ G }[/math]-manifold and [math]\displaystyle{ (M_2, \omega_2) }[/math] a Hamiltonian [math]\displaystyle{ H }[/math]-manifold. Then the natural action of [math]\displaystyle{ G \times H }[/math] on [math]\displaystyle{ (M_1 \times M_2, \omega_1 \times \omega_2) }[/math] is Hamiltonian, with momentum map the direct sum of the two momentum maps [math]\displaystyle{ \Phi_G }[/math] and [math]\displaystyle{ \Phi_H }[/math]. Here [math]\displaystyle{ \omega_1 \times \omega_2 := \pi_1^*\omega_1 + \pi_2^*\omega_2 }[/math], where [math]\displaystyle{ \pi_i : M_1 \times M_2 \rightarrow M_i }[/math] denotes the projection map.
  4. Let [math]\displaystyle{ M }[/math] be a Hamiltonian [math]\displaystyle{ G }[/math]-manifold, and [math]\displaystyle{ N }[/math] a submanifold of [math]\displaystyle{ M }[/math] invariant under [math]\displaystyle{ G }[/math] such that the restriction of the symplectic form on [math]\displaystyle{ M }[/math] to [math]\displaystyle{ N }[/math] is non-degenerate. This imparts a symplectic structure to [math]\displaystyle{ N }[/math] in a natural way. Then the action of [math]\displaystyle{ G }[/math] on [math]\displaystyle{ N }[/math] is also Hamiltonian, with momentum map the composition of the inclusion map with [math]\displaystyle{ M }[/math]'s momentum map.

Symplectic quotients

Suppose that the action of a compact Lie group G on the symplectic manifold (M, ω) is Hamiltonian, as defined above, with momentum map [math]\displaystyle{ \mu : M\to \mathfrak{g}^* }[/math]. From the Hamiltonian condition it follows that [math]\displaystyle{ \mu^{-1}(0) }[/math] is invariant under G.

Assume now that 0 is a regular value of μ and that G acts freely and properly on [math]\displaystyle{ \mu^{-1}(0) }[/math]. Thus [math]\displaystyle{ \mu^{-1}(0) }[/math] and its quotient [math]\displaystyle{ \mu^{-1}(0) / G }[/math] are both manifolds. The quotient inherits a symplectic form from M; that is, there is a unique symplectic form on the quotient whose pullback to [math]\displaystyle{ \mu^{-1}(0) }[/math] equals the restriction of ω to [math]\displaystyle{ \mu^{-1}(0) }[/math]. Thus the quotient is a symplectic manifold, called the Marsden–Weinstein quotient, symplectic quotient or symplectic reduction of M by G and is denoted [math]\displaystyle{ M/\!\!/G }[/math]. Its dimension equals the dimension of M minus twice the dimension of G.

Flat connections on a surface

The space [math]\displaystyle{ \Omega^1(\Sigma, \mathfrak{g}) }[/math] of connections on the trivial bundle [math]\displaystyle{ \Sigma \times G }[/math] on a surface carries an infinite dimensional symplectic form

[math]\displaystyle{ \langle\alpha, \beta \rangle := \int_{\Sigma} \text{tr}(\alpha \wedge \beta). }[/math]

The gauge group [math]\displaystyle{ \mathcal{G} = \text{Map}(\Sigma, G) }[/math] acts on connections by conjugation [math]\displaystyle{ g \cdot A := g^{-1}(dg) + g^{-1} A g }[/math]. Identify [math]\displaystyle{ \text{Lie}(\mathcal{G}) = \Omega^0(\Sigma, \mathfrak{g}) = \Omega^2(\Sigma, \mathfrak{g})^* }[/math] via the integration pairing. Then the map

[math]\displaystyle{ \mu: \Omega^1(\Sigma, \mathfrak{g}) \rightarrow \Omega^2(\Sigma, \mathfrak{g}), \qquad A \; \mapsto \; F := dA + \frac{1}{2}[A \wedge A] }[/math]

that sends a connection to its curvature is a moment map for the action of the gauge group on connections. In particular the moduli space of flat connections modulo gauge equivalence [math]\displaystyle{ \mu^{-1}(0)/\mathcal{G} = \Omega^1(\Sigma, \mathfrak{g}) /\!\!/ \mathcal{G} }[/math] is given by symplectic reduction.

See also


  1. Moment map is a misnomer and physically incorrect. It is an erroneous translation of the French notion application moment. See this mathoverflow question for the history of the name.
  2. The vector field ρ(ξ) is called sometimes the Killing vector field relative to the action of the one-parameter subgroup generated by ξ. See, for instance, (Choquet-Bruhat DeWitt-Morette)