A new kind of geometry on an even-dimensional manifold
You arrive at this rung fluent in two languages already. From the Riemannian track you know how a metric g equips each tangent space with an inner product — a symmetric bilinear form — out of which lengths, angles, geodesics, and curvature flow. From the forms track you know the differential form, built from the alternating algebra, and the meaning of a form being closed. Symplectic geometry is what happens when you keep the second language and deliberately throw away the symmetry. Instead of a symmetric form you put a skew-symmetric one on each tangent space, and instead of pointwise rigidity you get something stranger.
Concretely, a symplectic form omega on a manifold M is a differential 2-form satisfying two conditions. First, it is closed: d omega = 0, in the sense you learned for closed and exact forms. Second, it is non-degenerate: at each point p, the skew form omega_p on T_p M has trivial radical — if omega_p(v, w) = 0 for every w, then v = 0. A manifold M together with such an omega is a symplectic manifold, written (M, omega). Non-degeneracy is the analogue of a metric being positive-definite, but skew rather than symmetric.
The model: a symplectic vector space
All the geometry lives, pointwise, in one linear-algebra picture, so it pays to nail it down first. Take R^(2n) with coordinates (q_1, ..., q_n, p_1, ..., p_n) — think 'positions q and momenta p' — and define the standard form omega_0 = sum over i of dp_i ^ dq_i. On basis vectors it pairs each q-direction with its matching p-direction by ±1 and pairs everything else by 0. This is the linear model: a real vector space with a non-degenerate skew form is a symplectic vector space, and omega_0 is its canonical shape.
The crucial linear fact — the linear Darboux lemma — is that this is the only shape. Any non-degenerate skew form on a finite-dimensional real space admits a basis (a symplectic basis e_1, ..., e_n, f_1, ..., f_n) in which it looks exactly like omega_0: omega(e_i, f_j) = delta_ij, with all other pairings zero. The proof is a skew version of Gram-Schmidt — pick any e_1, find a partner f_1 with omega(e_1, f_1) = 1, peel off the plane they span, and recurse on the omega-orthogonal complement. There is no continuous knob to turn, no curvature, no 'eigenvalues of omega' to measure: every symplectic vector space of dimension 2n is isomorphic to every other.
This is already the headline in miniature. Contrast it with a metric: two inner products on the same space can genuinely differ once you remember the standard metric (their relative 'eigenvalues' are invariants — that is curvature's seed). A skew form has no such invariants. Symplectic geometry will inherit this floppiness wholesale: there is nothing to measure locally. Whatever content the subject has must be global.
The canonical example: cotangent bundles
Where do symplectic manifolds come from? The richest source — the one that motivated the whole field through classical mechanics — is the cotangent bundle T*M of any smooth manifold M. Points of T*M are pairs (q, p) where q is a point of M and p is a covector at q (a momentum). On T*M there lives a canonical 1-form, the tautological or Liouville 1-form theta, defined with no choices at all, and the canonical symplectic form is omega = −d theta. Because it is exact, it is automatically closed; one checks it is non-degenerate. So every configuration space M auto-generates a phase space (T*M, −d theta).
Local picture on T*M, coords (q_1..q_n, p_1..p_n): tautological 1-form theta = sum_i p_i dq_i symplectic form omega = -d theta = sum_i dq_i ^ dp_i Darboux normal form (any symplectic 2n-manifold, locally): omega = dq_1 ^ dp_1 + dq_2 ^ dp_2 + ... + dq_n ^ dp_n
There are non-cotangent symplectic manifolds too, and they matter. The 2-sphere S^2 carries its area form, which is closed (top degree on a surface) and non-degenerate, making (S^2, area) symplectic — yet S^2 is not a cotangent bundle. More generally every Kähler manifold, every complex projective variety, is symplectic via its Kähler form. So the field straddles two worlds: the floppy mechanics side (cotangent bundles) and the rigid complex-algebraic side, which the neighbouring Kähler track explores. Both obey the same Darboux theorem locally.
The Darboux theorem: no local invariants
Now the headline, stated honestly with its hypotheses. The Darboux theorem: if (M, omega) is a symplectic manifold of dimension 2n, then around every point p there exist coordinates (q_1, ..., q_n, p_1, ..., p_n) — Darboux coordinates — in which omega = sum of dq_i ^ dp_i, exactly the flat model omega_0. Not approximately, not to first order: on the nose, on a whole neighbourhood. Every symplectic manifold is locally indistinguishable from R^(2n) with its standard form.
Let the contrast with Riemannian geometry land, because it is the whole point. There is no Darboux theorem for metrics: you generally cannot flatten g to the Euclidean form near a point — the obstruction is precisely the curvature tensor, which you cannot make vanish by choosing coordinates. Metrics have a local invariant (curvature) and so a rich local theory. Symplectic forms have none. Two symplectic manifolds of the same dimension look identical through any small enough window. Whatever distinguishes them is global, topological, large-scale — never local.
Why the proof works: Moser's trick
It is worth seeing why Darboux is true, because the mechanism — Moser's trick — recurs everywhere in symplectic geometry (it powers Weinstein's neighbourhood theorem in guide 3). The idea is to interpolate. Near p, write the linear model omega_0 = sum dq_i ^ dp_i (achievable on the tangent space by the linear Darboux lemma) and your given omega_1 = omega. Connect them by the straight path omega_t = (1−t) omega_0 + t omega_1, which is closed for all t and, on a small enough neighbourhood, stays non-degenerate.
- Arrange omega_0 and omega_1 to agree at the point p (use linear Darboux on T_p M), so their difference vanishes at p.
- Write the difference as omega_1 − omega_0 = d sigma for some 1-form sigma vanishing at p — possible because the difference is closed and we are local (the Poincaré lemma).
- Solve, for each t, the equation omega_t(X_t, -) = −sigma for a time-dependent vector field X_t. Non-degeneracy of omega_t is exactly what makes this solvable — it lets you invert the form.
- Flow along X_t from time 0 to 1. Cartan's magic formula gives d/dt of the pullback (phi_t)*omega_t = 0, so the flow phi_1 pulls omega_1 back to omega_0 — those are your Darboux coordinates.
The single load-bearing idea is that non-degeneracy lets you convert a form into a vector field and then flow. A closed form whose cohomology class does not change can be straightened by a diffeomorphism. Hold onto that pattern; it is the engine of the whole subject. A map that does the straightening — a diffeomorphism pulling one symplectic form back to another — is a symplectomorphism, the symplectic notion of equivalence, and Darboux says locally everyone is symplectomorphic to the standard model.
What the rest of the rung builds on this
With (M, omega) and Darboux coordinates in hand, the non-degeneracy of omega becomes a dictionary. It sets up a perfect pairing between vector fields and 1-forms: given a function H, the equation omega(X_H, -) = dH defines a unique Hamiltonian vector field X_H. That single move — turning energy H into a flow — is classical mechanics, and it is the entire subject of guide 2, along with the Poisson bracket and Liouville's volume-preservation theorem. Everything starts from omega being invertible, which is just non-degeneracy wearing working clothes.
Non-degeneracy also makes 'size' make sense for subspaces. A subspace on which omega restricts to zero is called isotropic; a maximal such subspace, of dimension exactly n inside the 2n, is Lagrangian. The Lagrangian submanifold — half-dimensional, omega vanishing on it — turns out to be the central object of the whole field (graphs of closed 1-forms, the zero section of T*M, fixed-point sets), and it headlines guide 3 along with Weinstein's theorem, itself another Moser-trick payoff. Guide 4 brings symmetry: when a group acts preserving omega, the moment map records conserved quantities and symplectic reduction shrinks phase space.
One honest caveat to carry forward, and a reminder of scope. The caveat: this guide states Darboux and sketches its mechanism, but a careful proof — defining the Liouville form invariantly, checking non-degeneracy of omega_t, justifying the flow exists — is a few pages of a graduate text, not a paragraph; treat the sketch as the idea, not the verification. The scope: signs and even the very definition omega = +d theta versus −d theta differ between authors, so when you open a book, read its conventions on page one and stay loyal — most early confusion in this subject is a misplaced sign, not a misunderstood concept.