CIS calculation failing, bus error

1

I have been trying to obtain the excited state energies and gradients for singlet and triplet states using the CIS method with the DETCI module. The overall goal is to obtain these properties in the presence of an external charge field using the QMMM class.

I am running these calculations on a 34 atom system. A calculation for the first singlet excited state with a small basis (sto-3g) seems to work. Moving to a larger basis (6-311++G**) results in a Bus error before the calculation completes. I believe the error is due to Psi4 running out of memory, as I have assigned Psi4 28 GB of memory, and the job with 32 GB but this limit is exceeded and the job is killed. I have attached the input input.txt and outputs output_1.txt output_2.txt for this calculation.

Looking at previous topics (Regarding CISD calculation) and the documentation, it seems that this behaviour should be expected when using the DETCI module with a large number of basis functions. Are the errors I am getting likely to be due to this?

If this is the case, is there any way to truncate the CISD method used in the FNOCC module to obtain just single excitations?

Alternatively, are there any methods in Psi4 to obtain singlet and triplet state energies and gradients, with a similar cost/accuracy to TD-DFT?

I am using version 1.3.2.

Thank you for your time.

2

You’re not even getting to DETCI. It seems to be the integral setup for HF that is causing problems. DETCI was optimized for full CI computations, so using it for CIS on a system of this many basis functions strikes me as likely to go wrong.

Using FNOCC for CIS isn’t currently supported. I don’t believe the developer in charge of that module has plans to extend it.

Psi does have TDDFT, but it’s not documented, and we still need to get interface parts working… I know it’s rather high on our priority list, but I can’t tell you the specifics. If you’re interested, let me know, and I can get the attention of the developer who is working on that.

Unfortunately, that’s all I can tell you. I’m not familiar enough with the underlying integral machinery to diagnose the issue with integral setup, and I can’t offer a method recommendation either. It’s a very busy few days for the Psi developers, so if needed, I’ll see if I can get the attention of somebody with more experience with these issues when things calm down.

1 Like
3

I would definitely be interested in this, as I was initially going to use TDDFT; CIS was a backup option.

Thank you very much for your answer.

4

I believe the person in charge of TDDFT is @robertodr. Perhaps he can give some insight?