ABINIT Discussion Forums

Dear all,
I am trying to run abinit parallelized over kpt and bands (version 6.10.2) running on 4 cores. Thereby I am initially reading a previously written WFK file from a previous run "run 1" (I move the ouptut WFK of run 1 to input WFK of run 2). The GS calculation of run 2 finishes nicely with the provided WFK file, but when writing the new WFK file to disk abinit crashes with a segfault. The difference of the two input configurations is a slight change in geometry. Also, this error does not occur when running only kpt-parallelized.

I found that in the file "xderivewrite.F90" at line 572 within the double loop, at least in one case ipos becomes negative which is less than the lower bound of 0. Apparently the array "displace" is passed to the routine at line 874 in the file rwwf.F90. Can anyone reproduce this behaviour? I am not sure what is going wrong there or fixing this alone is the solution to the problem.

I attached the input file to first produce the initial WFK file: abirun.producewfk.in
and then the input file which reads initially the previous WFK file: abirun.in
the PSP files are also attached:
http://www.wfu.edu/~natalie/papers/pwpa ... _abinit.gz
http://www.wfu.edu/~natalie/papers/pwpa ... _abinit.gz

The intel 11.1.072 compiler was used with openmpi 1.4.2

The following configure options were used: FC=mpif90 CXX=mpicxx CC=mpicc FCFLAGS="-O2 -i_dynamic" --with-dft-flavor=none --with-trio-flavor=none --enable-mpi --with-linalg-libs="-L/opt/intel/mkl/10.2.3.029/lib/em64t -lmkl_lapack95_lp64 -lmkl_intel_lp64 -lguide -lmkl_intel_thread -lmkl_core"

Thank you very much in advance for your help!

The issue comes from the following (runing on 24 procs):

- the last band proc restarts with 1067 npw ;
- while the last band proc contain 1069 npw on disk (which change the current npw on this proc to 1069 indeed) ;
- when writing at end of SCF, the write routine use an index of kg points coming from the 1067 npw and not the new 1069 (array is initialised in kpgsph.f90).

So a possible solution would be to recalculate kg indexes after a restart. Any comment from PAW guys ?

Problem solved.
Will be available in the future v6.12.2.

Use of mpi_enreg datastructure had to be rationnalized in inwffil routine and children.