Probable CPU time per iteration step

Dear developers,

I am running my MRCC tasks using the following methods: CCSDT and CCSDTQ.

In both cases, I have seen the next lines:

• CCSDTQ:
  Code:

  Probable CPU time per iteration step (hours): 83516.07
• CCSDT:
  Code:

  Probable CPU time per iteration step (hours): 88.57

I have stopped the first calculation, and only CCSDT was continued. I was surprised when I have noticed that each iteration step was running for about 10 minutes.

So, my questions are:
What does this metric show?
Can I extrapolate CCSDT results to CCSDTQ calculation? I mean the each CCSDTQ iteration step will be done approximately 83516.07/88.57*10=9429.39 minutes.

Best regards,
Igor

You should forget this metric.
But a CCSDTQ calculation is not feasible with 160 orbitals (your previous example). You use a smaller basis for CCSDTQ.

Thanks!

Which count of orbitals can be computed using CCSDTQ method?
I understand that it depends on my computer, and currently, I am using 1x Xeon Gold 6146 (24 threads x 3700 MHz) with enough memory.

And one additional question:
Sometimes I have seen that MRCC runs in one thread enough long while (LR-)CC iterations are going. What does MRCC do at these moments?

Best regards,
Igor

Which count of orbitals can be computed using CCSDTQ method?
<100.

Sometimes I have seen that MRCC runs in one thread enough long while (LR-)CC iterations are going. What does MRCC do at these moments?
It is calculating, e.g., the DIIS vectors or performing I/O operations.

Dear Igor,
I think the largest number of spatial orbitals I did with CCSDTQ was 321, but it was H2O with frozen core:


In your other question you were running Ne2 with all electrons correlated, which is 20 electrons. The cost increases rapidly as you add electrons, so what Mihály was suggesting was to do CCSDTQ in a smaller basis set, then add (CCSDTQ-CCSDT) from the small basis set, to the CCSDT result at the bigger basis set. You can also try this with CCSDT(Q) instead of CCSDT.

With best wishes!
Nike

Dear Prof. Kallay,

Thank you for your reply! I will decrease the basis set and try to calculate my system again.

> It is calculating, e.g., the DIIS vectors or performing I/O operations.
Is it possible to parallelize these one-threaded calculation operations, e.g., DIIS, which perform when CC calculation is?

Best regards,
Igor