TheoremComplete

Lagrangian Subspace Theorem

TheoremProperties of Lagrangian Subspaces

Let (V,ω)(V, \omega) be a 2n2n-dimensional symplectic vector space and L⊆VL \subseteq V a Lagrangian subspace (i.e., L=LωL = L^\omega). Then:

  1. Dimension: dim⁡L=n=12dim⁡V\dim L = n = \frac{1}{2}\dim V
  2. Complementary Lagrangians: There exists another Lagrangian L′L' such that V=L⊕L′V = L \oplus L' and ω∣L×L′:L×L′→R\omega|_{L \times L'}: L \times L' \to \mathbb{R} is a perfect pairing
  3. Existence: Lagrangian subspaces always exist in any symplectic vector space

Furthermore, the space of all Lagrangian subspaces forms a smooth manifold called the Lagrangian Grassmannian Λ(n)\Lambda(n), diffeomorphic to U(n)/O(n)\text{U}(n)/\text{O}(n).

Lagrangian subspaces are maximal isotropic subspaces, meaning they are the largest subspaces on which the symplectic form vanishes identically. They play a fundamental role in classical mechanics (as configuration spaces) and quantum mechanics (as quantum states in geometric quantization).

DefinitionTransverse Lagrangians

Two Lagrangian subspaces L1,L2L_1, L_2 are transverse if L1∩L2={0}L_1 \cap L_2 = \{0\}, equivalently if L1+L2=VL_1 + L_2 = V.

In this case, every vector v∈Vv \in V has a unique decomposition v=ℓ1+ℓ2v = \ell_1 + \ell_2 with ℓi∈Li\ell_i \in L_i, and the map:

ω(−,−):L1×L2→R\omega(-, -): L_1 \times L_2 \to \mathbb{R}

defines a non-degenerate pairing, establishing isomorphisms L1≅L2∗L_1 \cong L_2^* and L2≅L1∗L_2 \cong L_1^*.

ExamplePosition and Momentum Lagrangians

In (R2n,ω0)(\mathbb{R}^{2n}, \omega_0) with coordinates (q,p)(q, p), consider:

Lq={(q,0):q∈Rn},Lp={(0,p):p∈Rn}L_q = \{(q, 0) : q \in \mathbb{R}^n\}, \quad L_p = \{(0, p) : p \in \mathbb{R}^n\}

Both are Lagrangian subspaces with Lq∩Lp={0}L_q \cap L_p = \{0\} and Lq+Lp=R2nL_q + L_p = \mathbb{R}^{2n}. The pairing is:

ω0((q,0),(0,p))=∑i=1nqipi=⟨q,p⟩\omega_0((q, 0), (0, p)) = \sum_{i=1}^n q_i p_i = \langle q, p \rangle

which identifies LpL_p with the dual space Lq∗L_q^*.

Remark

The condition L=LωL = L^\omega is equivalent to:

  • LL is isotropic: ω∣L=0\omega|_L = 0
  • LL is maximal isotropic: if W⊇LW \supseteq L is isotropic, then W=LW = L
  • dim⁡L=n\dim L = n and ω∣L=0\omega|_L = 0

The third characterization is most commonly used in practice.

The Lagrangian Grassmannian Λ(n)\Lambda(n) parametrizes all Lagrangian subspaces. It has dimension n(n+1)2\frac{n(n+1)}{2} and carries a natural stratification based on the intersection dimensions with a reference Lagrangian.

ExampleLagrangians from Generating Functions

Given a smooth function S:Rn→RS: \mathbb{R}^n \to \mathbb{R}, its graph in phase space:

LS={(q,p):p=∇S(q)}⊆T∗Rn≅R2nL_S = \{(q, p) : p = \nabla S(q)\} \subseteq T^*\mathbb{R}^n \cong \mathbb{R}^{2n}

is a Lagrangian subspace if SS is quadratic. For general SS, LSL_S is a Lagrangian submanifold (not necessarily a linear subspace). This construction relates to classical mechanics where SS is the action.

DefinitionMaslov Index

The Maslov index measures the topology of paths in the Lagrangian Grassmannian. For a path of Lagrangians LtL_t relative to a reference Lagrangian L0L_0, the Maslov index counts (with signs and multiplicity) how many times LtL_t intersects the singular set {L:L∩L0≠{0}}\{L : L \cap L_0 \neq \{0\}\}.

The fundamental group π1(Λ(n))≅Z\pi_1(\Lambda(n)) \cong \mathbb{Z} is generated by the Maslov cycle.

Remark

Any two transverse Lagrangians determine a symplectic basis: if L1=span{e1,…,en}L_1 = \text{span}\{e_1, \ldots, e_n\} and L2=span{f1,…,fn}L_2 = \text{span}\{f_1, \ldots, f_n\} with L1∩L2={0}L_1 \cap L_2 = \{0\}, we can choose bases such that ω(ei,fj)=δij\omega(e_i, f_j) = \delta_{ij}.