Page 1 of 1

Seg fault using UPF pseudopotentials

Posted: Tue Feb 05, 2013 8:11 pm
by Tom Poole
Hello,

I've recently attempted to reproduce some Quantum Espresso calculations using Abinit, but have had little success performing calculations using the UPF format of pseudopotential. Versions 7.0.5 and 6.12.3 both segfault inside the SCF loop if a UPF pseudopotential is used. This is easily reproducible (for me at least) by taking the bulk Silicon example from the tutorials and switching the 14si.pspnc file for the 14-Si.nlcc.UPF file, also found in the bundled tutorial material.

The culprit seems to be the section of vtowfk.F90 at line 864 which is included via a compiler #if/#else directive. When using UPF pseudopotentials a segfault is encountered when calling cg_zdotc. This can be avoided by executing the presently excluded branch.

I've not delved into the mechanics of this fix, so I can't suggest how the current code is failing or even if this is a sensible thing to do. The results obtained after this procedure seem correct.

Yours,

Tom

Re: Seg fault using UPF pseudopotentials

Posted: Tue Feb 05, 2013 8:39 pm
by Alain_Jacques
Hi Tom.

Recompile abinit with --enable-zdotc-bugfix="yes"

Any improvement?

Kind regards,

Alain

Re: Seg fault using UPF pseudopotentials

Posted: Wed Feb 06, 2013 1:49 am
by Tom Poole
Hi Alain,

That has fixed things. For the benefit of others experiencing this problem (who also hadn't noticed the well-named option), the required addition is --enable-zdot-bugfix="yes" (no c).

Thank you for your help.

Tom