Reduction in CIOVERLAP size

[Priyam Kumar De]

Hello all,

I am trying to run surface hopping simulation for a system with large number of basis. This leads to great calculation time for CIOVERLAP. However, when using ncore and ndisc to reduce the size of total orbitals to be used for CIOVERLAP, there is some error and the calculation stops. 

&cioverlap part of user_config.nml -

cio_options = -s transmomin -a -t 5e-4 -e -1 -i 148

The error is - 

cp: cannot create regular file 'overlap/': Not a directory

At line 1037 of file ../src/mod_cioverlap.f90

Fortran runtime error: End of record

Can you please help me with this and mention methods to reduce computation time for CIOVERLAP. 

Thanks,

Priyam

[Priyam Kumar De]

Hello all,

I am also trying to reduce orbital number using OrbWin option in %cis. This orca calculation is working, however some error is showing when newton-x tried to execute commands and the job gets aborted. 

Is there any other details need to provide when using OrbWin?

Thanks,

Priyam

[Rafael Souza]

Dear Priyam,

I messaged the developer of CIOverlap and did some tests myself. Changing the number of frozen and virtual orbitals can reduce computational time, but I didn't see much difference in my reduced tests. Changing the -t (threshold) to 5e-3 significantly improved performance, but it affected the overlap results, and I don't know to what extent this will affect the dynamics.

Then I started looking into the code to see possible improvements. I am still testing, but it has worked well so far. You can download the executable and replace it in your current installation to see if it works in your case/system.

You can download it from here: https://gitlab.com/light-and-molecules/cioverlap/-/blob/speedup/bin/cioverlap?ref_type=heads
You have to replace the cioverlap executable in the newton-x-package/CIOVERLAP/bin/ folder. I would definitely keep a backup of the cioverlap executable you already have.

You should be able to call it exactly the same, and it may be a little faster, but this one you can run in parallel by setting the OMP_NUM_THREADS environment variable:
export OMP_NUM_THREADS=8

I would use the same number of processors you are using for the ORCA calculation.

This was compiled in a fairly recent system (Ubuntu 24), so you may have a problem with the GLIBC version in an older system. Let me know.

Regarding the OrbWin, I never tried that; I don't know how it affects the number of expected orbitals. If you share the error message, we can try to reproduce it.

Best regards,
Rafael

[Priyam Kumar De]

Dear Rafael,

I really appreciate your effort to help me with this problem.

I tried to run with the newer cioverlap file, unfortunately the cluster I am using doesn't have updated GLIBC. This particular error was printed in the output files. So I have contacted the IT helpdesk to update the GLIBC version and waiting for their response. 

Regarding the OrbWin, it is a special function that can be used in ORCA to reduce the number of orbitals used for CIS calculations. Normally, ORCA take orbitals within energy range of -3 to +3 hartree for CIS calculation. This range considers huge number of orbitals and CIOVERLAP takes all those orbitals. Now, using this OrbWin, I can choose the range of the orbital required for CIS calculations. 

OrbWin[0] = 120,-1,-1,160               - this option is used in the TDDFT section of ORCA input file. 

The number of orbital used for cis calculation can be seen in the orca output file in TDDFT calculation section. 

I believe, with use of this option, the .cis file generated by orca is a different than normal scenario and newton-X is not able to read that. Details of this option can be found the webpage https://www.faccts.de/docs/orca/6.0/manual/contents/detailed/tddft.html

Please let me know if you need any other information.

Best Regards,

Priyam

[Rafael Souza]

Dear Pryiam,

I am still to check the OrbWin, but you can find a version of cioverlap compiled in the cluster here; it should support a very old version of GLIBC. You will find it here:
https://gitlab.com/light-and-molecules/cioverlap/-/blob/speedup/bin/cioverlap_cluster?ref_type=heads

Rename it and replace the cioverlap you currently have.

Best regards,
Rafael

[Priyam Kumar De]

Dear Rafael,

Thank you for this file. It worked with current GLIBC version and I got sufficient speedup in CIOVERLAP calculations. 

Best Regards,

Priyam