4
$\begingroup$

I try to optimize and compute frequencies for the system built up from C, O, H, and Co atoms in the following functionals: B2PLYP, B2PLYP-D3, B3LYP, B3LYP-D3, BLYP, BLYP-D3, DSD-PBEP86, PBE, PBE-D3, WB97X-D3. The basis set I use is aug-cc-pVTZ aug-cc-pVTZ/C def2/J with SDD on Co. The input is common and looks like that:

! BLYP D3ZERO aug-cc-pVTZ aug-cc-pVTZ/C def2/J verytightopt freq
%basis
    newECP Co "SDD" end
end
%geom
%MAXCORE 3428
%PAL NPROCS 28 END
* xyzfile 1 1 BLYP-D3.xyz

In each case I got an error:

Nuc. rep. gradient       (SHARK) ... done (  0.0 sec)
HCore & Overlap gradient (SHARK) ... done ( 32.0 sec)
ECP gradient             (SHARK) ... done ( 47.0 sec)
RI-J gradient            (SHARK) ... 
ORCA finished by error termination in SCF gradient
Calling Command: mpirun -np 28  /home/users/john/orca503/orca_5_0_3_linux_x86-64_shared_openmpi411/orca_scfgrad_mpi BLYP-D3.scfgrad.inp BLYP-D3 
[file orca_tools/qcmsg.cpp, line 465]: 
  .... aborting the run

I use single node, which has 2 processors, each 14 cores (2x14) and up to 128GB. The queue system is SLURM and hence I specify there:

#SBATCH --nodes=1
#SBATCH --ntasks=28
#SBATCH --ntasks-per-node=28
#SBATCH --cpus-per-task=1
#SBATCH --mem=128000MB
Improve this question
$\endgroup$
16
  • 1
    $\begingroup$ This could just be my unfamiliarity with Orca input, but what is aug-cc-pvtz/c? It seems like aug-cc-pvtz is the basis set (for C, H, and O) and def2/J is an auxiliary basis. Based on the error message, it seems to be failing during RI gradients, so something about the basis specification may be wrong. Have you tried running this without def2/J just to see if it still gives an error? $\endgroup$
    – Tyberius ♦
    Commented May 31, 2022 at 12:59
  • $\begingroup$ Cross posted on Reddit $\endgroup$
    – Tyberius ♦
    Commented May 31, 2022 at 13:06
  • $\begingroup$ +1 on Tyberius's comment, although I would recommend removing aug-cc-pvtz/c instead, because ground state RI calculations require a /J auxiliary basis set, rather than a /C one. If you provide a /C basis when the program does not require it, this may confuse the program. (Ultimately, for the results to be reliable, one may have to use AutoAux instead of def2/J because def2/J is not designed to work well with aug-cc basis sets) $\endgroup$
    – wzkchem5
    Commented May 31, 2022 at 13:27
  • $\begingroup$ @Tyberius : Will check that $\endgroup$
    – user1247
    Commented May 31, 2022 at 14:30
  • 1
    $\begingroup$ This might be totally unrelated, but shouldn't nprocs be the same as ntasks? I don't remember the details, so how do you calculate maxcore? Looks low. It's been a while for me to do these calculations, but I remember that there was a lot of guidance in the input library. Look at the statistics of slurm to see what is actually used. And is probably a good idea to leave out anything you don't need, like running a Freq job it you are struggling with scf. $\endgroup$
    – Martin - マーチン
    Commented May 31, 2022 at 22:39

0

Reset to default