Message of " Sub-Space-Matrix is not hermitian "

[hatdau]

Hi,
I have the similar problem but only with parallel mode. It finished with no problem if running on ONE node (i.e serial version).

I'm running vasp.5.3 on amd machine

The process stop after first ionic step.

I first got the error "LAPACK: Routine ZPOTRF failed!"
After decrease POTIM, the error become "WARNING: Sub-Space-Matrix is not hermitian in DAV "

I'm running relaxation with following options

Code: Select all

  1 ENCUT?=?400
??2?PREC?=?high
??3?ISMEAR?=?-1
??4?SIGMA?=?0.005
??5?NSW?=?41
??6?
??7?IBRION?=?2
??8?ISIF?=?3
??9?#POTIM?=?0.05
?10?NPAR?=?2
?11?
?12?LMAXMIX?=?4
?13?
?14?ISTART?=?0
?15?ICHARG?=?2

for a simple fcc structure,
kpoints are monkhorst-pack gamma-centered 10x10x10

[huiwen_zhu]

Hi,
I have the similar problem but only with parallel mode. It finished with no problem if running on ONE node (i.e serial version).

I'm running vasp.5.3 on amd machine

Same problem here using vasp.5.4.4 with VTST. The problem occurred when using two nodes (64 cores), but when using one node (32 cores), everything goes smoothly.

Are there any updates on this issue?

[martin.schlipf]

If the behavior changes when you change the number of nodes, I suspect that the number of bands may be the issue. Please compare the two OUTCAR files and check if NBANDS is the same for both calculations. If your system does not have very many electrons at some point you just start increasing the number of bands and not increasing the speed of your calculation. Please review this page to understand a bit better how the parallelization of VASP works.

If NBANDS is not the issue, note that this is a very generic diagonalization error that may have appeared for very different reason then the original post. Please try to simplify the system (smaller unit cell, smaller cutoff, less k-points) to see what is the minimal example that reproduces the issue and then submit the input files of that minimal example for us to review. Furthermore, please state which version of VASP you are using.

[alireza_rezvani]

I have the same problem. It's the KPOINTS problem or possibly the number of the cores. What you can do is go finer KPOINT or decrease the number of cores .