caliarim

Hi, are -M_bar and/or K_bar SPD?

If I am not wrong, inv(M_tilde)*K_tilde = [0,I;-inv(M_bar)*K_bar,-inv(M_bar)*C_bar]. So, you can turn it into a regular eigenvalue problem, provided that M_bar is not singular.

Yes, the problem could be related with #79 (force the initial vector to be in the range). In #142 there is a further modification which should manage rank deficient matrices....

I verified that arpack 3.5.0 with a dummy generalized problem with the zero input matrix and the identity matrix B stops with "starting vector is zero".

I think I found a workaround at the time, give me a week or so.

This is my proposal ``` diff --git a/SRC/dgetv0.f b/SRC/dgetv0.f index 1d6dc01..9635720 100644 --- a/SRC/dgetv0.f +++ b/SRC/dgetv0.f @@ -258,6 +258,8 @@ c %-----------------------------------------% c | Back from computing OP*(initial-vector) | c...

Thanks, I will investigate why the problem is still there for the symmetric case.

This is the error before my patch ``` Error with _saupd, info = -9 Check documentation in _saupd ``` and this is the result after ``` Ritz values and relative...

Sorry, version.h is not valid fortran file. It cannot be directly used by X{n,s}aupd.f.

You have msglvl .gt. 0 somewhere.