search

dealii / dealii

The development repository for the deal.II finite element library
https://www.dealii.org
Other
1.33k stars 736 forks source link

arpack/step-36_parpack_mf_02 failing #4700

Closed masterleinad closed 7 years ago

masterleinad commented 7 years ago

Currently (4f6efca0da272c90a9cf), I get the following diff

928: * 1,6 ** 928: DEAL::5931.69 928: DEAL::5970.26 928: DEAL::6035.05 928: - DEAL::6035.05 928: DEAL::6099.84 928: DEAL::Ok 928: --- 1,6 ---- 928: + DEAL::5866.90 928: DEAL::5931.69 928: DEAL::5970.26 928: DEAL::6035.05 928: DEAL::6099.84 928: DEAL::Ok

It seems that the double eigenvalue is just reported once and a lower one is found instead.

Unfortunately, CDash never runs this test (or I am still not able to find it there).

davydden commented 7 years ago

Yes, I sometimes see those. Try increasing the number of Arnoldi vectors. Probably has to with round-offs

davydden commented 7 years ago

something like const unsigned int num_arnoldi_vectors = 2*eigenvalues.size() + 20; might make it more robust.