There are some questions that deal with proofs that I'd say are pretty clearly computational science. For instance, the question The Lax-Milgram Lemma in FEM with non-homogenous Dirichlet BC seems like a computational science question.
Then, there are questions like the recent How to prove the 2-norm of an invertible matrix is exactly the reciprocal of its minimum singular value? that are usually flagged for closure.
Currently, the FAQ reads: "Mathematics, ask on math.stackexchange.com if it is oriented towards proofs and analytical calculation not of a numerical or computational character." The second question above falls into a little bit of a gray area.
What sort of distinction do we want to draw between mathematics questions that are appropriate for the site versus mathematics questions that should be migrated to Mathematics Stack Exchange? My first hack at this would be to say that "questions that do not demonstrate a clear connection to computational science should be migrated". The first question demonstrates a much clearer connection (it looks like a problem using finite element methods) to computational science than the second (which could be related to computational science -- it is linear algebra, after all -- or it could just be a math question).
My tendency is to be more permissive (we could stand to grow the site a bit). Simultaneously, this site is a community. We rely on the participation of high-reputation users in curating the site, and your input is very important in helping the site grow, so I'd like your opinion on moving forward. A decision on this issue should make it into the FAQ.