Questions about direct solvers

[zhyxue0]

Hello everyone. I am solving a time-dependent problem, but the assembled matrix A is a constant coefficient matrix. In order to reduce computation time, I don't want to solve Ax = b every time, but calculate the inverse A_inv of A in the first step.
Then at each step x = A_inv * b, how to achieve this?

[dham]

Actually storing Ainv is a bad idea because it will be dense and hence huge. Factorising A into L and U and applying the inverse using them repeatedly is a good idea. LU is the default solver and I think if you pass constant_jacobian=True to the LinearVariationalProblem then the factorisation will be reused, which is what you want.