m_ritz_val[nev_new] with nev_new=m_ncv

[jdbancal]

Hey @yixuan,

Me again ;-) I have some trouble with GenEigsSolver : the method nev_adjusted sometimes tries to access m_ritz_val[nev_new] with a value of nev_new which is larger than the size of the array, thus resulting in a fatal error.

From what I can see, the m_ritz_val vector contains m_ncv elements. At the same time, in the beginning of function nev_adjusted it looks like nev_new can take values up to m_ncv (first value m_nev, and then potentially increase up to m_ncv in the loop). When this happens, m_ritz_val[nev_new] leads to a segmentation fault...

The occurence of this behavior seems to be linked with the decrease of the value of nconv between two iterations. At least both things happen jointly. Is the algorithm safe if nconv decreases for some iterations (I guess only momentarily...)? Is it normal to allow for nev_new to take values up to m_ncv?

Cheers!

[yixuan]

If you encountered such situations then this should be a bug. Let me take a look.

[yixuan]

Hi @jdbancal, I think this bug has been fixed. Would you like to have a try of the most recent version?

[jdbancal]

Hey @yixuan, I tested the code which was giving a problem on the previous version of the library, and it works fine now :-) It looks like you fixed the problem. Thanks!

[yixuan]

Great! I think we can close this issue now.