Multithreaded DF-CCSD(T) gradient test fails

1

Hi there,

I was just trying to run the psi4-1.4a2/tests**/dfccsd-t-grad1** test with

set_num_threads(2)

added to the input, and it fails (badly) with psi_1.4_rc1.

All energies (RHF,CCSD,T) are identical, but the gradient changes dramatically compared to a single-threaded run.

Thanks a lot in advance for any hint/advice on that subject.

Udo

2

Is this a self-compiled version or obtained from conda?

3

I can reproduce the errors.
This is concerning and we will have a look at this ASAP.

4

I think we have fixed that problem, which is related to a function in libmints. The new code along with the uhf-ccsd(t) grad is completed, which will be added to Psi4 very soon. We are just adding new tests for new capabilities.

5

Follow the github issue for updates on this matter threading error in DF-CCSD(T) gradient · Issue #2192 · psi4/psi4 · GitHub
It’s a serious bug, so thanks again for reporting this.

6

And, issue solved on the developer version. This was something I accidentally introduced in 1.4rc1. 1.4a2 and 1.3 were both fine.

If you need an up-to-date Psi working right away, you’ll need to compile from source. I can offer assistance on that. If you want to install from conda, I’m pushing to get a new release candidate submitted soon, but that part is out of my hands. (The developer who manages releases has a lot on her to-do list right now!)

Thanks again for the report.

7

Thanks for all you comments.

It is not that urgent, so a can either wait for the next release or compile the code from source.

8

Correction to previous: 1.4a2 was fine, 1.4rc1 would have had this problem as well.

9

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.