A linear symplectic space is a 2n-dimensional vector space $V$ with a symplectic two form $\omega.$ On this vector space $V$ is a canonical basis $(e_1,...,e_n,f_1,...f_n)$ with $\omega(e_i,f_j) = \delta_{i,j}$ and $\omega(e_i,e_j)= \omega(f_i,f_j)=0.$

Now, a Lagrangian subspace $L$ of $V$ is an n-dimensional subspace satisfying $L^{\perp} = L$ where the complement is taken with respect to the form $\omega.$

The theorem states now: Any Lagrangian subspace $L$ has an orthogonal complement $K$ that is also a Lagrangian subspace such that $L \oplus K = V.$

The proof starts like this: Let $L_0=(e_1,...,e_n)$ and set $M = L \cap L_0.$ Then $M$ is a $k \le n$ dimensional subspace of $L_0$. As any $k-$ dimensional subspace of $L_0$ has a $n-k$ dimensional orthogonal complement, we find such a $n-k$ dimensional space $T$ such that $M \oplus T = L_0.$

Now, the proof we had in the lecture says: Without loss of generality $T= \operatorname{span}\{e_1,...,e_{n-k}\}$ and I don't really understand why we are able to say this wlog here?


So, I think what the author is doing is picking a new symplectic basis for his/her proof at this step. Instead of using the standard symplectic basis, they would like to use a symplectic basis $\{\tilde e_i, \tilde f_i\}$ where the terms $\tilde e_1, \ldots, \tilde e_{n-k}$ span the space $T$. So, how do we construct the new basis $\tilde e_i$?

  • Let's start by constructing the $\tilde e_i$. We can produce an orthonormal basis $\{\tilde e_i\}$ for $L$ with the first $n-k$ terms spanning $L_0$ by using Gram-Schmidt.
  • Now, how do we construct the $\tilde f_i$? Write $\tilde e_i=\sum_k a^k_i e_k$. Then if $\tilde f_i=\sum_k a^k_i f_k$, $$ \omega(\tilde e_i, \tilde f_j)=\omega\left( \sum_k a^k_i e_k, \sum_k a^k_j f_k\right)$$ which is $\sum_k a^k_i a^k_j = \tilde e_i \cdot \tilde e_j= \delta_{ij}$.

So, we've produced a basis with the desired properties. The author therefore assumes that we've been working with this new basis from the start of the proof.

  • $\begingroup$ How can $L_0$ be spanned by $n-k$ terms when $L_0$ is a n-dimensional space? $\endgroup$ – Surfaceint Aug 20 '15 at 12:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.