Hello,

This is Themos Tsikas from NAG, Oxford, UK. I am dealing with a user of HECToR, the UK academic Cray machine, who has a problem with Scalapack routine PDGBTRF (banded LU factorization with pivoting). Specifically, the code crashes sometimes with a memory corruption in malloc. Using valgrind, I have tracked it down to IPIV array being written beyond its documented bounds:

* IPIV (local output) INTEGER array, dimension >= DESCA( NB ).

* Pivot indices for local factorizations.

* Users *should not* alter the contents between

* factorization and solve.

In fact, it seems to be treated as if it were of length DESCA(NB)+BWL+BWU, where

* BWL (global input) INTEGER

* Number of subdiagonals. 0 <= BWL <= N-1

*

* BWU (global input) INTEGER

* Number of superdiagonals. 0 <= BWU <= N-1

I've noticed a few posts on the forum reporting bad behaviour from this routine so I am wondering if I've come across a documentation bug or an algorithmic bug. I would appreciate your thoughts on this.

Best Regards

Themos Tsikas