Proof of Symplectic Basis Theorem
We prove by induction on that every symplectic vector space admits a symplectic basis.
Base Case (, ):
Since is non-degenerate, there exist with . Rescaling, we may assume . Set and . Then:
So is a symplectic basis.
Inductive Step: Assume the theorem holds for symplectic spaces of dimension .
Let have dimension . Choose any non-zero . By non-degeneracy, there exists with . Normalizing, choose so that . Set .
Define . Then:
- is a symplectic subspace: is non-degenerate
Consider the symplectic complement:
We claim that and .
Proof of claim:
-
Non-degeneracy of : Suppose satisfies for all . Then for all (by definition of ) and for all , so for all . By non-degeneracy, .
-
Direct sum decomposition: We have and . If , then and for all . Since is non-degenerate, . Thus .
-
Dimension count: By linear algebra, . Thus:
Therefore .
By the inductive hypothesis, admits a symplectic basis .
The union is a symplectic basis for :
- for all (since each is in either or , both isotropic)
- for all (similarly)
- (by construction for or , and by induction for )
This completes the induction.
The proof is constructive: given any non-zero vector, we can algorithmically construct a symplectic basis. The key insight is that symplectic complements of symplectic subspaces are again symplectic, allowing induction.
The proof above can be viewed as a symplectic Gram-Schmidt process: starting from any basis , we construct a symplectic basis by:
- Normalize and find with
- Orthogonalize remaining vectors with respect to using
- Repeat inductively
Unlike the usual Gram-Schmidt for inner products, the symplectic version maintains isotropy of subspaces.
Consider with . Start with .
We need with . Try : . Set , .
The symplectic complement is:
This gives and , so .
Set and . Then is a symplectic basis.
The uniqueness of the standard form (up to symplectomorphism) contrasts sharply with Riemannian geometry, where metrics on a fixed vector space form an infinite-dimensional cone. This rigidity makes symplectic geometry both simpler and more constrained than Riemannian geometry.