[Trilinos-Users] Problem on recovery of orthogonalization in Anasazi::LOBPCG

[SungHwan Choi]

Dear all,
I run LOBPCG, and as everybody know, failure of orthogonalize might happen.
To my knowledge, the result of recovery should be independent to number of
processors.

Error! Caught LOBPCGRitzFailure at iteration 2
Full orthogonalization is off; will try to recover.

However, as far as I observed, after the above message printed, the same
inputs, running on different number of processors, start to give different
residue values

Do you know why this happened? becuase of above phenomena, total iteration
number is sometimes significantly changed. It is a disaster when studying
parallel efficiency.

Best regards,
Sunghwan Choi
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://trilinos.org/pipermail/trilinos-users/attachments/20161101/4e278bcc/attachment.html>