[DFTB-Plus-User] Crash with libdftD4 libs-dftD3

[Bruno Madebène]

Hello DFTB+ users.

I just started to play with DFTB+, last stable version installed with 
anaconda on a CENTOS 7 computer (48core, 392GB RAM)

I try to play with a 20Ang box of 266 water molecules with xtb 
hamiltonian. And  I encountered few problems.

- First one:  from vasp postcar coordinates file, only the 384 first 
atoms are used. So i just switch to generic coordinates.

- Second one: Crash without error message at step 0 depending on the 
number of atoms:
Works below 576 atoms.
Crash for more than 600 atoms.

This is not a problem with periodic, cluster, type of calculation. Works 
fine with less than 576 atoms.
I tried with several proc number, 4 to 48, and also try to play with 
memory. Same problem

 From system errors, i found segmentation fault from libdftD4 and libdftD3.

[31307100.723080] dftb+[272846]: segfault at 7fffaac4e780 ip 
00007f0d665a1413 sp 00007fffaac4e780 error 6 in 
libdftd4.so.3.4.0[7f0d6658c000+40000]
[31307112.677172] dftb+[272944]: segfault at 7ffe9a4e6040 ip 
00007fb0519c9b9b sp 00007ffe9a4e6040 error 6 in 
libs-dftd3.so.0.6.0[7fb0519ac000+32000]

So, does the pb comes from the code? Or binaries ?

I was lazy by simply recovering the binaries with anaconda. Do you think 
I should try to recompile from sources?

Thanks and have a nice day !

Bruno Madebène

Assistant Professor

Sorbonne Université.
Laboratoire Monaris UMR8233
Paris France