Segfault 11, when running with tasks=400 and inpdftu=5

[Marjana Lezaic]

Dear all,

I have a rather funny problem. I'm trying to get the tensor moment decomposition of the density matrix, for Francesco's example of NiO provided with the Elk code package. It works without problems with the provided input, but if I switch to inpdftu=5, the code crashes with segfault 11 (address not mapped to object at address 0x10).
To make things even more specific, the crash happens only if the tensor moments are calculated without prevously running a self-consistensy cycle, i.e., it will still work if I have

tasks
0
400

or

tasks
1
400,

but not if I have only

tasks
400

(the last one, however, works also fine with inpdftu=2, as initially set in the example). Of course, the fact that I can still run the code with one of the first two options and get what I need is great, but I thought I'd let you know that something odd is going on...

Cheers,
Marjana

[J. K. Dewhurst]

Hi Marjana,

You're quite correct: it's a problem with initialisation in writetm.f90. I'll have it fixed for the next release.

Regards,
Kay.