Hi Long and Miroslav, Just to clarify: for collinear calculations the first nstfv orbitals are spin-up and the next nstfv orbitals are spin-down. For non-collinear calculations the orbitals are not separable into pure up and down but rather general 2-spinor functions. Regards, Kay.
Hi, The GW calculation in Elk is one-shot G0W0 in the Matsubara frequency representation. It is very important to converge all the relevant parameters for the GW calculation. Here is an example for Ni which is properly converged: tasks 0 10 600 610 630 spinpol .true. bfieldc 0.0 0.0 0.1 reducebf 0.85 highq .true. ! Matsubara frequency cut-off wmaxgw 15.0 ! response function G-vector cut-off gmaxrf 3.5 ! temperature of many-body system (determines Matsubara frequency spacing) tempk 1000.0 wplot 3000...
Further update: I've contacted Susi Lehtola and he requests that you file a ticket on the Libxc Gitlab tracker with a reproducer.
Update: I've now reproduced the issue by using Libxc version 7.1.2. This version does not fail gracefully with negative sigma for r2SCAN like version 7.0.0. This should be fixed in Libxc but we may have make a work-around in Elk for the time being. Thanks and regards, Kay.
Hi Ricardo, I ran your example and it works fine for me. I've attached my INFO.OUT file. Perhaps your stack size was insufficient. Try using the script elk/elk.sh to run the code: #!/bin/bash export OMP_NUM_THREADS=32 export OMP_PROC_BIND=false export OMP_STACKSIZE=512M ulimit -Ss 524288 ~/elk/src/elk Best wishes, Kay.
Elk version 11.2.3 released
The f03 interface.
Hi Marcin, Libxc 7.0.0 has worked fine with Elk for over a year now (since version 10.5.16). Regards, Kay.