Aspects of Symplectic Geometry in Physics Josh Powell
1 Symplectic Geometry In Classical Mechanics
We seek here to use geometry to gain a more solid understanding of physics. This process may seem frighteningly mathematical, but we shouldn’t expect much gain in insight with only a modicum of effort. Following Alexander Pope: “A little learning is a dangerous thing; Drink deep, or taste not the Pierian spring: There shallow draughts intoxicate the brain, And drinking largely sobers us again.”
1.1 Hasty Review of Classical Mechanics Consider a mechanical system with both position degrees of freedom, qi, and momentum degrees of freedom, i pi. The Hamiltonian H(pi, q ) is a function on phase space that governs the dynamics of the system, and in particular of these degrees of freedom. That is, it describes time evolution by specifying their equations of motion. The equations of motion are
i ∂H ∂H q˙ = andp ˙i = − i ∂pi ∂q
We have a system of differential equations which are all first order in time derivatives. Since we know the i expression for the phase space variables’ time derivatives, any function f(pi, q ) of these variables will have a time derivative given by
df X ∂f ∂f = q˙i + p˙ dt ∂qi ∂p i i i X ∂f ∂H ∂f ∂H = − =: {f, H} ∂qi ∂p ∂p ∂qi i i i where in the last line we have defined the Poisson bracket of two functions on phase space. The function i also been explicitly dependent on time, f = f(pi, q , t), then we would have written df ∂f = {f, H} + dt ∂t in order to capture the total time derivative. We have thus seen how the Hamiltonian gives the time evolution of all dynamical quantities in our system.
What if we had chosen another coordinate system on our phase space? Let’s call this new coordinate i i system (Pi,Q ) to distinguish from the previous (pi, q ). The dynamics in the old system are described by varying the action Z X i δ pidq − Hdt = 0
and those in the new come from varying Z X i 0 δ PidQ − H dt = 0
The same dynamics are described in terms of these two coordinate systems if these two integrands differ by i a total differential dF of some function F = F (pi, q , t). We write this as
X i X i 0 dF = pidq − PidQ + (H − H)dt
i i 0 whence we get ∂F/∂q = pi, ∂F/∂Q = −Pi and ∂F/∂t = H − H. We call the function F the generating function of this canonical transformation. If F has no explicity time dependence, it’s an element of F(Q×Q). If it is time dependent, then F ∈ F(Q × Q × R), and the two Hamiltonians differ in a way more than just
1 rewriting the old Hamiltonian in terms of the new coordinates.
There is another common formalism in which the dynamics of a system can be described. Imagine we had i i i i wanted to work not in terms of the variables (pi, q ) but instead (q , q˙ )? Since ∂H/∂pi =q ˙ , we recognize i that what this amounts to is a Legendre transform from pi toq ˙ . Let’s say that the Legendre transform of i i i the Hamiltonian H(pi, q ) is some function L(q , q˙ ). This transformation and its inverse are given by
i ˙i X i i i L(q , q ) = piq˙ − H(pi, q ) with pi = ∂L/∂q˙
i X i i i i H(pi, q ) = piq˙ − L(q , q˙ ) withq ˙ = ∂H/∂pi
i i i The first canonical equation of motion, ∂H/∂pi =q ˙ , is transformed into the tautological statementq ˙ =q ˙ . This comes as no surprise, since it was the equation which specified in the first place the transformation we i are making. The second canonical equation of motion, ∂H/∂q = −q˙i, is transformed into
∂L d ∂L − = 0 ∂qi dt ∂q˙i
which is known as the Lagrange equation of motion. Note that it is second order in time derivatives, unlike the dynamical equations in the Hamiltonian formalism which were both first order. We’ve obtained the Lagrangian formulation of classical mechanics. The two are equivalent descriptions, up to certain technical issues related to when the Legendre transformation is not defined.
1.2 Symplectic Geometry Let M be a smooth manifold of even dimensionality and let Ω be a closed, non-degenerate 2-form on M. We call the pair (M, Ω) a symplectic manifold, and Ω its symplectic structure. The condition that Ω be non-degenerate means that the only vector field X on M satisfying Xy Ω = 0 is the one vanishing uniformly. A non-zero X satisfying this condition would correspond to a zero-mode of the symplectic structure, and would prevent Ω from being invertible. If such a zero-mode exists, it’s possible that it can be projected out and a lower dimensional system thus obtained is a symplectic manifold with non-degenerate symplectic structure. This procedure is known as reduction. One familiar example of this process comes from gauge theories, in which it is sometimes necessary to project out non-physical degrees of freedom. In this case, it’s known as gauge fixing.
In general, any diffeomorphism of M which preserves the symplectic structure is known as a sym- plectomorphism. That is, a symplectomorphism is a diffeomorphism between two symplectic manifolds ∗ f :(M, ΩM ) → (N, ΩN ) for which f ΩN = ΩM . We can generate diffeomorphisms (M, Ω) → (M, Ω) by exponentiating a vector field on M. If this vector field, call it X, is to generate a symplectomorphism, then it must satisfy LX Ω = 0. Vector fields satisfying this condition are known as symplectic vector fields. Furthermore, given a function f ∈ F(M), there is Xf ∈ X(M) satisfying
Xf y Ω + df = 0 (∗)
The invertibility of Ω guarantees the existence of Xf so defined. In local coordinates, this vector field has a ab the components (Xf ) = Ω ∂bf. Such a vector field is automatically symplectic: LXf Ω = Xf y dΩ + d Xf y Ω = d(−df) = 0
We call the function f the generating function and the vector field Xf a Hamiltonian vector field. Conversely, given a symplectic vector field X, we can try to find the corresponding generating function by working back- wards. We can of course do this locally, but in order to find a globally defined function that generates X,
2 we need the first de Rham group to be trivial H1(M) = 0. Therefore, obstructions which keep a symplectic vector field from being Hamiltonian arise from H1(M).
Requiring that dΩ = 0 allows one to write Ω = dΘ, at least locally. Depending on the topology of M, this may hold for a globally define 1-form Θ. Θ is known as the symplectic potential. Imagine we have a Hamiltonian vector field Xf generated by f. What is the change in Θ under an infinitessimal flow along Xf ? LXf Θ = Xf y dΘ + d Xf y Θ = Xf y Ω + d Xf y Θ = d Θ(Xf ) − f =: dΛ
where we used the relation (∗). Then Θ is an abelian gauge potential and Ω = dΘ is the corresponding field strength. From this point of view, the flow Ft(Xf ): M → M which preserves Ω but sends Θ → Θ + dΛ is a (symplectomorphic) gauge transformation.
1.2.1 Poisson Brackets
If we have two Hamiltonian vector fields Xf and Xg, their Lie bracket is also Hamiltonian.
L[Xf ,Xg ]Ω = [LXf , LXg ]Ω = 0
What is the corresponding generating function?
[Xf ,Xg]y Ω = (LXf Xg)y Ω
= LXf (Xgy Ω) − Xgy (LXf Ω) = Xf y d Xgy Ω + d Ω(Xg,Xf ) = −d Ω(Xf ,Xg)
Therefore, we define the Poisson bracket of the functions f and g as
{f, g} := Ω(Xf ,Xg) [Xf ,Xg]y Ω + d{f, g} = 0
We could have stopped midway in our derivation of the Poisson bracket and ended up with other equivalent expressions: {f, g} = df(Xg) = LXg f and also {f, g} = −dg(Xf ) = −LXf g. In local coordinates, the Pois- −1 ab a b son bracket takes the form {f, g} = (Ω ) ∂af∂bg. Note in particular that if ξ and ξ are two coordinate functions on phase space, we have {ξa, ξb} = (Ω−1)ab.
The salient features of the Poisson bracket – namely, anti-symmetry and the Jacobi identity – follow readily from the definition.
- Anti-symmetry: {g, f} = Ω(Xg,Xf ) = −Ω(Xf ,Xg) = −{f, g}
- Jacobi Identity: From the second boxed equation above, we see that X{f,g} = −[Xf ,Xg]. From here, we find
{f, {g, h}} = LX{g,h} f = −L[Xg ,Xh]f = −[LXg , LXh ]f
= LXh (LXg f) − LXg (LXh f)
= LXh {f, g} − LXg {f, h} = {{f, g}, h} + {g, {f, h}}
3 Hence, the Poisson bracket and its properties are entirely geometric notions defined on any symplectic man- ifold.
Let Xf be a Hamiltonian vector field on M with generating function f. Then, Xf generates a one- parameter family of diffeomorphisms, its flow, through exponentiation, and is the first-order term in the expansion of an infinitessimal flow. The corresponding infinitessimal change in a function G ∈ F(M) under this infinitessimal diffeomorphism is δG = LXf G = {G, f}.
1.2.2 Theorem of Darboux Any two symplectic manifolds of the same dimension are locally symplectomorphic, so all their distinguishing features must manifest as global properties. These are the famed Gromov-Witten invariants.
Compare this situation to Riemannian geometry, where symmetries of the manifold must preserve the Riemann curvature tensor. This is a much more restrictive condition and leads to the result that Riemannian manifolds with large symmetry groups are rare.
1.3 Theory of Dynamics Using Geometry Everything up until now was common to all symplectic manifolds. We will now begin to add some physics to the mix. It will be our goal to rephrase the ideas of mechanics in the language of differential geometry on symplectic manifolds. Given an n-dimensional configuration space Q of a system, both TQ and T ∗Q are even-dimensional manifolds on which we will place symplectic structures. We will call these manifolds the system’s velocity phase space and the momentum phase space, respectively. Putting an atlas on the manifold Q allows us to give coordinates to a point q = (qa). We inherit an induced atlas on TQ, whose points are now labelled (q, q˙) withq ˙ ∈ TqQ. But these are the natural variables for the Lagrangian. So we’ve found that the Lagrangian is a function on TQ, the velocity phase space. Likewise, we can denote points on T ∗Q ∗ ∗ by (p, q) with p ∈ Tq Q and recognize that T Q is the manifold on which the Hamiltonian function is defined.
Since we’re doing physics, let’s introduce some terminology common to the physics literature. The following is a “Mathematician ↔ Physicist dictionary” for the relevant terms. Math Term Physics Term symplectomorphism canonical transformation symplectic structure canonical 2-form symplectic potential canonical 1-form mathematics physics without purpose mathematics without rigor physics
1.4 Geometry of Canonical Mechanics A Hamiltonian system (T ∗Q, Ω,H) is a momentum phase space together with a special function H, the Hamiltonian, which governs the time evolution of the system. We’ve already established a natural language in which to describe time evolution: first find the vector field XH generated by the Hamiltonian and then watch the flow of our phase space under this vector field. Recall that the vector field will satisfy Eqn. (∗):
XH y Ω + dH = 0 We’ve already seen how the canonical 1-form transforms under a Hamiltonian vector field’s flow: LXH Θ = d Θ(XH ) − H = dΛ
4 1 where we’ve defined the phase function Λ := Θ(XH ) − H. If Ft is the flow generated by XH , this last line reads ∗ Ft Θ − Θ) LX Θ = lim = dΛ H t→0 t It follows that the difference between Θ and its flow is related to the integral of Λ. Z t ∗ 0 ∗ Ft Θ − Θ = dAt with At := Λ dt ∈ F(T Q) 0 ∗ where the integration is performed along the integral curve of XH from m ∈ T Q to Ft(m). Upon substituting the definition for Λ, we get Z t Z t 0 0 At = Θ(XH ) − H dt = Θ − H dt 0 0 0 The last equality follows from the fact that dt merely represents a step along the integral curve of XH 0 2 passing through the current point, and so Θ(XH )dt = Θ.
What if we evaluated this integral not along an integral curve of XH but along a curve close to it? By how much will the function S change its value? This variational calculation will be facilitated by picking b b coordinates and writing Θ = Θbdξ where the ξ are phase space coordinates. The variation will change our path from {t 7→ ξ(t}) to {t 7→ ξ(t) + η(t)}. To first order in η, the change in the integral will be Z t b a a a 0 δAt = (∂aΘb)dξ η + Θadη − (∂aH)η dt 0 Z t b a a a 0 = (∂aΘb − ∂bΘa)dξ η + d(Θaη ) − (∂aH)η dt 0 t Z t dξb ∂H = Θ ηa − Ω − ηa dt0 a ab 0 a 0 0 dt ∂ξ
Requiring that δAt = 0 for any variation η implies ∂H ∂H Ω ξ˙b = ←→ ξ˙a = (Ω−1)ab , ab ∂ξa ∂ξb a a a a and Θaη (t) − Θaη (0) = 0 ←→ Θaη (t) = Θaη (0)
a These are statements of the equations of motion and conservation of the quantity Θaη . The surface terms vanish if we hold our endpoints fixed – ie, if η(0) = η(t) = 0. But by keeping the surface terms, we can deduce what form Θ must take. This is a commonly used trick and one which we will employ later.
We must have just varied the action. Upon comparison to §1.1, we find that our previous unwitting P a choice of canonical 1-form was Θ = pa dq and so
X a Ω = dΘ = dpa ∧ dq is our canonical 2-form on the momentum phase space. Note the symplectic structure given above will give the following values of the Poisson brackets of phase space coordinates:
{qa, qb} = 0
{pa, pb} = 0 a {q , pb} = δab
1Secretly, this is the pullback of the Langrangian L ∈ F(TQ) under the Legendre transform τ : T ∗Q → TQ, so I give this function a name close to L but not quite! 2In canonical coordinates, to be introduced shortly, this equality reads: pqdt˙ 0 = p dq.
5 In this coordinate system on T ∗Q, the matrix expression for Ω is particularly simple. 0 I (Ωab) = −I 0
∗ ∗ At is a function on T Q. Given a point m ∈ T Q, we can flow it under XH for a given time t and integrate the value of Θ − H dt0 along this curve. This statement amounts to saying that the dynamical history of a mechanical system can be determined by the initial configuration and conjugate momenta to the configuration coordiantes. But there is another view in some situations. Much like a “shooting problem” in the study of differential equations, we might be able to specify our initial and final configurations and determine which (unique) history bridges the two in time t. Such a view would make the action a function on the space Q × Q. Viewed as such, it’s known as the Hamilton two point characteristic function. In some cases, the configuration space allows many different dynamical histories to interpolate from the ini- tial to the final configuration in the given time. This situation is more complicated and will be discussed later.
We will see in §1.4.3 why these expressions for Θ and Ω are standard, at least within the context of non-relativistic mechanics. Coordinates which render the symplectic structure’s matrix expression in the above form are known as canonical coordinates. Looking at which changes of basis give rise to another set of canonical coordinates leads one to consider the symplectic groups Sp(2n; R).
1.4.1 Rehash of Λ and H In general, given an action integral S = R dt Λ, the canonical 1-form Θ and the canonical 2-form Ω, we can determine the Hamiltonian H of the system by solving the two equations
XH y Ω + dH = 0 XH y Θ − H = Λ
a If we assume a canonical coordinate system (pa, q ) has been put on our cotangent bundle, these become the following especially nice relationship
X ∂H Λ = pa − H ∂pa which becomes a Legendre transform just as soon as we dignify ∂H/∂pa to be new coordinates to take the place of the pa’s.
1.4.2 Liouville’s Theorem We can define a volume form, ie a top form, on (T ∗Q, Ω) by noting the following. Since dim T ∗Q = 2n and 2 ∗ −n Vn 2n ∗ a Ω ∈ Ω (T Q), we can consider Ω := (2π) Ω ∈ Ω (T Q). Picking canonical coordinates (pa, q ) on T ∗Q, we see that the volume form is expressed as
n a ^ dpa ∧ dq = Ω 2π a=1
This volume form is preserved under Ft, the flow generated by XH , since Ω is preserved.
n ∗ −n ^ ∗ Ft Ω = (2π) Ft Ω = Ω
6 1.4.3 An Intrinsic Definition of Θ We motivated the typical form Θ takes, up to gauge transformations, through comparison with the expression of the action stated in §1.1. The following is another characterization which explains why we dub this framework “canonical.” Θ ∈ Ω1(T ∗Q) is uniquely determined by the condition that for any 1-form α ∈ Ω1(Q) on our configuration space α∗Θ = α, where we view α as a map α: Q → T ∗Q to take the pullback. Define the basepoint projection π : T ∗Q → Q : (p, q) 7→ q. Since α is a section of the contangent bundle over Q, we know that π ◦ α = idQ. We can invert ∗ ∗ the above expression to solve explicitly for Θ. At each point p ∈ T Q we have Θp = π p, or equivalently