Olle Heinonen

Thanks, Kay. I am running now with OMP_NUM_THREADS=96 and so far the code runs (I recompiled with the the -fcheck compiler flags). I am not sure I understand why OMP_PROC_BIND=false is needed - on some nodes we have, the intersocket communications are slow and I usually want to confine batches of threads or tasks to one socket. I am also a bit mystified by why OMP_STACKSIZE=512M works. I had tried to set the OMP stack size to a much larger number (as my nodes have boatloads of RAM) but that did not...

Hi Kay. Thanks for response. Sorry it took a while to find your response. I get the seg fault with ramdisk .false. and mixsdb=2. The error, as I just noted in a post a few seconds ago, seems to stem from taumt not being allocated: At line 39 of file potks.f90 Fortran runtime error: Allocatable actual argument 'taumt' is not allocated. I am not sure why this shows up - I am palying with LSDA or various GGA flavors for a spin-polarized system, no meta-GGA so I am not sure why this error shows up. I...

Hi. I am running elk on some simple systems that are spin-polarized and have spin-orbit coupling turned on. I get a segmentation fault sometimes (typically for larger systems, > 10-ish atoms) irrespective of using the gnu or Intel coompiler suites. Compiling with gnu flags -fcheck=all shows an error in output: At line 39 of file potks.f90 Fortran runtime error: Allocatable actual argument 'taumt' is not allocated This seems to be an array for kinetic energy density. I get the error irrespective of...

Incidentally, running with Intel+MKL (2022.1.0) gives segmentation errors, apparently by mpi.

Hello elk developers and users. I am fairly new to elk (but not to DFT and all-electron codes). First, kudos to developers - the code is pretty simple to use and not too tedious to install. I am trying to do a optical calculation (RPA - tasks 0, 120 and 121) on a system with 133 atoms and xctype 20 (PBE). They are all light, eg carbon, so not too many electrons. I have scrutinized all suggested ways of dealing with stack memory and threads etc to remove various errors that could arise. At the end,...