Use of MPI in ORCA

[Priyam Kumar De]

I am trying to run ORCA using multiple processor while running Newton-X for surface hopping. I got this error in the orca output - 

This version of Spack (openmpi ~legacylaunchers schedulers=slurm)

is installed without the mpiexec/mpirun commands to prevent

unintended performance issues. See https://github.com/spack/spack/pull/10340

for more details.

I am submitting this job using Slurm. 

However, when I am running without %pal option in orca, the dynamics runs. 

I also tried to use wrapper file for orca job submission, but I am not sure where to make changes in Newton-X file to use the wrapper file. 

Thanks,

Priyam

[Rafael Souza]

Hello,

It seems to be related to running ORCA in a non-standard way, since you need a wrapper to call ORCA.

If you have to use a personalized script to run ORCA, you can use the generic interface to an external script. The script will have to call ORCA, parse the energy and gradient, and put it in the format specified.

Best regards,
Rafael

[Priyam Kumar De]

Hello,

Thank you for the response. I will try to use the ORCA wrapper using the generic interface. 

But is it possible to run ORCA using multiple processor (%pal option in ORCA input file) during surface hopping dynamics in Newton-X. Because, initially I tried doing it without any wrapper file and that gave the error mentioned previously. Please let me know if there is specific ways to run ORCA without wrapper (in presence of %pal option in ORCA). 

Regards,

Priyam

[Rafael Souza]

Hello,

Yes, you can run NX-NS dynamics using ORCA in parallel. On Newton-X's side, it doesn't know about the parallelization of ORCA. As long as the parallelization keyword is present in the ORCA input, you run the calculation with NX-NS the same way you would run a serial calculation.

The error you showed seems to be happening at the ORCA level. Can you run the input file of the JOB_AD folder with ORCA the normal way?

• $ORCA/orca orca.inp > orca.out

As long as you can run a single point the normal way, there should be no problem for the dynamics, parallel or not.

Best regards,

Rafael

[Priyam Kumar De]

Hello,

I have tried running the orca job from JOB_AD folder and it ran perfectly (using %pal option). 

Is it possible that the error is due to Newton-X job submission using slurm? If I use ntasks=n (n is the nprocs in orca input file), the md.log file shows info and error n times. As if n-number of newton-X is running simultaneously. And if ntasks=1 is used, Newton-X tries to run, but the error comes in orca.out file - 

This version of Spack (openmpi ~legacylaunchers schedulers=slurm)

is installed without the mpiexec/mpirun commands to prevent

unintended performance issues. See https://github.com/spack/spack/pull/10340

for more details.

And the last line shows - [file orca_main/main_driver_cis.cpp, line 114]: Error (MAIN_CIS): No orbitals were found in the gbw file

Upon checking, there is something in the orca.gbw file.

Regards,

Priyam

[Rafael Souza]

Hello,

So ORCA, in parallel by itself, runs fine;

Does Newton-X with ORCA in serial also work in this machine?

Newton-X is unaware of the ORCA parallelization, so I didn't expect it to make a difference.

Can you share here or send me the user_config.nml file, the orca input, the md.log and possibly the log file with this error?

Best regards,

Rafael

[Priyam Kumar De]

Hello,

I made some changes in the job submission script with help of chatGPT and it is working fine now. I think, submitting jobs using srun is the problem. Without srun, it ran correctly.

Regards,

Priyam