Geometry

Symplectic geometry

Geometry built from a form measuring signed area, the natural setting for classical mechanics.

IntuitionIntuition: signed area in phase space

Imagine the position and momentum of a swinging pendulum plotted together as a point $(q,p)$ moving in a plane. As the pendulum swings, this point traces a closed loop, and the area enclosed by the loop stays exactly the same no matter how the pendulum's energy is distributed between position and momentum. Symplectic geometry is the study of spaces equipped with a symplectic form $\omega$ , a rule for measuring this signed area, and of the maps that preserve it exactly — the natural mathematical language of classical mechanics.

An invariant torus in phase space: for an integrable Hamiltonian system, the Liouville–Arnold theorem foliates phase space into tori like this one, each carrying action-angle coordinates

(I,\theta)

on which the flow is a straight-line rotation.

UndergraduateDefinition: symplectic manifolds and Hamiltonian vector fields

Definition: Symplectic manifold

A symplectic manifold is a pair $(M,\omega)$ where $M$ is a smooth manifold of even dimension $2n$ and $\omega$ is a closed, non-degenerate differential $2$ -form: closed means $d\omega=0$ , and non-degenerate means that for every nonzero tangent vector $v$ there is a tangent vector $w$ with $\omega(v,w)\neq 0$ .

d\omega = 0

Closedness ( $d\omega=0$ ) means $\omega$ has no local sources, and is the condition that lets one recover Hamilton's equations consistently from any smooth energy function. Non-degeneracy means $\omega$ sets up a linear isomorphism between tangent vectors and covectors at every point, turning the differential $dH$ of an energy function $H$ into a genuine vector field $X_H$ via the equation below.

\iota_{X_H}\omega = dH

In the standard coordinates $\omega=\sum_{i=1}^n dq_i\wedge dp_i$ , the defining equation $\iota_{X_H}\omega=dH$ unwinds to the familiar Hamilton's equations $\dot q_i=\partial H/\partial p_i$ , $\dot p_i=-\partial H/\partial q_i$ : the abstract symplectic formalism and the classical mechanics formalism are the same statement in different notation.

Symplectic geometry versus Riemannian geometry
Property	Symplectic $(M,\omega)$	Riemannian $(M,g)$
Bilinear form	Skew-symmetric: $\omega(v,w)=-\omega(w,v)$	Symmetric: $g(v,w)=g(w,v)$
Local model	Always the same: $\omega=\sum_{i=1}^n dq_i\wedge dp_i$ (Darboux)	Curvature can vary from point to point
Preserved by isomorphisms	Area/volume $\omega^n$ , not lengths or angles	Lengths, angles and geodesic distance

UndergraduateKey theorems: rigidity and conservation

Theorem: Darboux's theorem

Every symplectic manifold $(M,\omega)$ of dimension $2n$ is locally isomorphic to the standard model: around every point there exist coordinates $(q_1,\dots,q_n,p_1,\dots,p_n)$ in which $\omega=\sum_{i=1}^n dq_i\wedge dp_i$ . In particular, symplectic manifolds carry no local invariants analogous to curvature.

Why is it true?

Riemannian manifolds have curvature, a local invariant that distinguishes a sphere from a plane even in a tiny neighbourhood. Darboux's theorem says symplectic manifolds have no such local fingerprint: seen up close, every symplectic manifold looks exactly like flat phase space $\mathbb{R}^{2n}$ with its standard form. All the interesting content of symplectic topology is therefore global.

Proof

Step 1 (set up the linear problem). By a linear change of coordinates one first arranges that at the chosen point $p_0$ , the form $\omega$ equals the standard form $\omega_0$ exactly at $p_0$ itself, since every non-degenerate skew-symmetric bilinear form on $\mathbb{R}^{2n}$ can be brought to standard form by a suitable basis.

Step 2 (interpolate). Define the family $\omega_t=(1-t)\omega_0+t\omega$ for $t\in[0,1]$ . Since $\omega_0$ and $\omega$ agree at $p_0$ and both are closed and non-degenerate there, $\omega_t$ is also closed and, after shrinking the neighbourhood if necessary, non-degenerate for every $t$ .

Step 3 (solve Moser's equation). Because $\omega$ minus $\omega_0$ is closed and vanishes at $p_0$ , the Poincaré lemma produces a 1-form $\sigma$ with $d\sigma=\omega-\omega_0$ . Moser's trick looks for a time-dependent vector field $X_t$ solving $\iota_{X_t}\omega_t=-\sigma$ ; since $\omega_t$ is non-degenerate this determines $X_t$ uniquely at every point.

Step 4 (integrate the flow). Let $\varphi_t$ be the flow generated by $X_t$ . A direct computation with Cartan's formula gives $\mathcal{L}_{X_t}\omega_t=d\iota_{X_t}\omega_t+\iota_{X_t}d\omega_t=-d\sigma=-(\omega-\omega_0)$ , combined with $\dot\omega_t=\omega-\omega_0$ , so that $\frac{d}{dt}(\varphi_t^*\omega_t)=0$ : the pullback $\varphi_t^*\omega_t$ is constant in $t$ and equals $\omega_0$ for every $t$ .

Step 5 (conclude). Setting $t=1$ gives $\varphi_1^*\omega=\omega_0$ on the shrunk neighbourhood, so the coordinates pulled back along $\varphi_1$ are the desired Darboux coordinates.

Theorem: Liouville's theorem

Let $X_H$ be the Hamiltonian vector field of $H$ on a symplectic manifold $(M,\omega)$ of dimension $2n$ , defined by $\iota_{X_H}\omega=dH$ , and let $\phi_t$ be its flow. Then $\phi_t$ preserves the symplectic volume form: $\phi_t^{*}(\omega^n)=\omega^n$ for every $t$ . In particular the Hamiltonian flow preserves phase-space volume.

Why is it true?

In classical mechanics, Liouville's theorem is the mathematical reason a gas of particles obeying Hamilton's equations cannot spontaneously compress itself into a smaller region of phase space: the flow may stretch and twist the region into a wild shape, but its volume never shrinks or grows. This underlies statistical mechanics and explains why naive numerical integrators that do not respect this conservation law can produce spurious energy drift over long simulations.

Proof

Step 1 (Cartan's magic formula). For any vector field $X$ and form $\omega$ , Cartan's formula gives $\mathcal{L}_X\omega=d(\iota_X\omega)+\iota_X(d\omega)$ . Apply this to $X$ equal to $X_H$ : since $\omega$ is closed, $d\omega=0$ , and since $\iota_{X_H}\omega=dH$ by definition of the Hamiltonian vector field, we get $\mathcal{L}_{X_H}\omega=d(dH)+0$ .

Step 2 (the form itself does not change). The exterior derivative satisfies $d\circ d=0$ for every function, so $d(dH)=0$ . Hence $\mathcal{L}_{X_H}\omega=0$ : the Hamiltonian flow preserves the symplectic form itself, not merely its volume.

Step 3 (pass to the top power). The volume form is $\omega^n=\omega\wedge\cdots\wedge\omega$ ( $n$ factors). The Leibniz rule for the Lie derivative on wedge products gives $\mathcal{L}_{X_H}(\omega^n)=n\,\omega^{n-1}\wedge\mathcal{L}_{X_H}\omega$ , and since $\mathcal{L}_{X_H}\omega=0$ by Step 2, the right-hand side vanishes: $\mathcal{L}_{X_H}(\omega^n)=0$ .

Step 4 (integrate along the flow). If $\phi_t$ is the flow of $X_H$ , then $\mathcal{L}_{X_H}(\omega^n)=0$ means exactly $\frac{d}{dt}\phi_t^{*}(\omega^n)=0$ for every $t$ . Since $\phi_0=\mathrm{id}$ gives $\phi_0^{*}(\omega^n)=\omega^n$ , integrating shows $\phi_t^{*}(\omega^n)=\omega^n$ for all $t$ , which is Liouville's theorem.

AdvancedAdvanced: symplectic rigidity and Gromov's non-squeezing theorem

Gromov proved in 1985 that a symplectic ball $B^{2n}(r)=\{|z|\le r\}$ can be symplectically embedded into the cylinder $B^2(R)\times\mathbb{R}^{2n-2}$ if and only if $r\le R$ — exactly the same condition as for embedding a $2$ -dimensional disk of radius $r$ into a disk of radius $R$ , no matter how large $n$ is. The proof uses pseudo-holomorphic curves and gave birth to the invariants known as symplectic capacities, of which the simplest is the Gromov width $c(B^{2n}(r))=\pi r^2$ , the area of the smallest disk factor a ball can be squeezed through.

c(B^{2n}(r)) = \pi r^2

UndergraduateReal-World Applications and Worked Examples

Because Hamiltonian mechanics is written in symplectic language, symplectic geometry shows up wherever a physical system conserves energy and phase-space structure: celestial mechanics uses it to study the long-term stability of planetary orbits, particle-accelerator design uses symplectic maps to track beams through millions of revolutions without artificial damping or growth, and molecular dynamics software uses symplectic integrators so that simulated molecules do not gain or lose energy purely from numerical error. Optimal-control theory and geometric quantization in mathematical physics also build directly on the symplectic formalism.

Example: Checking that a rotation is symplectic

On $(\mathbb{R}^2,\omega=dq\wedge dp)$ , consider the rotation $\phi_\theta(q,p)=(q\cos\theta-p\sin\theta,\ q\sin\theta+p\cos\theta)$ by a fixed angle $\theta$ . Is $\phi_\theta$ a symplectomorphism, i.e. does it satisfy $\phi_\theta^{*}\omega=\omega$ ?

Solution

Step 1: write the Jacobian. The map is linear with matrix $J=\begin{pmatrix}\cos\theta&-\sin\theta\\ \sin\theta&\cos\theta\end{pmatrix}$ acting on $(q,p)$ .

Step 2: pull back the form. For a linear map on $\mathbb{R}^2$ , $\phi_\theta^{*}(dq\wedge dp)=\det(J)\,dq\wedge dp$ , so it suffices to compute the determinant of $J$ .

Step 3: compute the determinant. $\det(J)=\cos^2\theta+\sin^2\theta=1$ for every $\theta$ , using the Pythagorean identity.

Step 4: conclude. Since $\det(J)=1$ , $\phi_\theta^{*}\omega=\omega$ for every $\theta$ : rotations are symplectomorphisms, exactly matching the geometric fact that rotations preserve area.

Example: Symplectic Euler versus explicit Euler for the harmonic oscillator

For the harmonic oscillator $H(q,p)=\tfrac12(p^2+q^2)$ , compare one step of the symplectic Euler method $q_{n+1}=q_n+hp_n,\ p_{n+1}=p_n-hq_{n+1}$ with one step of the explicit (naive) Euler method $q_{n+1}=q_n+hp_n,\ p_{n+1}=p_n-hq_n$ , for a small step size $h>0$ . Which one is consistent with Liouville's theorem?

Solution

Step 1: write the symplectic Euler map as a function of $(q_n,p_n)$ . Substituting $q_{n+1}=q_n+hp_n$ into the second equation gives $p_{n+1}=p_n-h(q_n+hp_n)=p_n-hq_n-h^2p_n$ .

Step 2: compute its Jacobian. $\frac{\partial(q_{n+1},p_{n+1})}{\partial(q_n,p_n)}=\begin{pmatrix}1&h\\ -h&1-h^2\end{pmatrix}$ , with determinant $1\cdot(1-h^2)-h\cdot(-h)=1-h^2+h^2=1$ for every $h$ .

Step 3: compute the Jacobian of explicit Euler. There $\frac{\partial(q_{n+1},p_{n+1})}{\partial(q_n,p_n)}=\begin{pmatrix}1&h\\ -h&1\end{pmatrix}$ , with determinant $1+h^2$ , which is strictly greater than $1$ for every $h\neq0$ .

Step 4: interpret. By Liouville's theorem the exact flow has Jacobian determinant exactly $1$ at every step. Symplectic Euler matches this exactly regardless of $h$ , so it does not artificially inflate phase-space volume; explicit Euler expands area by a factor $1+h^2$ every step, so energy drifts upward over many steps — this is why long-term simulations use symplectic integrators.

Research frontier as of 2026

Symplectic geometry remains an active research area. Quantitative symplectic topology studies finer invariants than the Gromov width, called symplectic capacities, that measure exactly how 'big' a symplectic manifold is beyond its raw volume; in 2024 Pazit Haim-Kislev and Yaron Ostrover found a convex domain on which two natural capacities disagree, giving a counterexample to a long-standing conjecture of Viterbo from 2000 that all such capacities should coincide for convex bodies (their paper appeared in the Annals of Mathematics in 2026). Another active direction is homological mirror symmetry, Kontsevich's conjecture that the symplectic geometry of a space (encoded in its Fukaya category) is secretly equivalent to the algebraic geometry of a completely different, 'mirror' space — proved in several important cases but still open in general.

Why must a symplectic manifold have even dimension $2n$ ?

For the harmonic oscillator $H(q,p)=\tfrac12(p^2+q^2)$ with the standard form $\omega=dq\wedge dp$ , what is the Hamiltonian vector field $X_H=(\dot q,\dot p)$ defined by $\iota_{X_H}\omega=dH$ ?

According to Liouville's theorem, what does the Hamiltonian flow of a symplectic manifold preserve?

Long-term simulations of the solar system spanning millions of years use symplectic integrators instead of general-purpose methods like standard Runge-Kutta. Why?

References

Dusa McDuff, Dietmar Salamon (2017). Introduction to Symplectic Topology
Mikhail Gromov (1985). Pseudo holomorphic curves in symplectic manifolds
Pazit Haim-Kislev, Yaron Ostrover (2026). A Counterexample to Viterbo's Conjecture · arXiv:2405.16513