SAPT - computing time questions

Dear colleagues,

I intend to calculate the interaction energy between two hexamethyldisiloxane molecules (see below)

My system has 52 atoms, and I intend to calculate SAPT at the levels sSAPT0/jun-cc-pVDZ, SAPT2+/aug-cc-pVDZ and SAPT2+(3)dmp2/aug-cc-pVTZ to obtain the Lennard-Jones parameters for the atoms of this molecule.

Given my little experience in SAPT, could I ask a few questions?

1 - Is calculating the SAPT energy of a system of this size (i.e., 52 atoms) something viable or is the computation time too high? I say this because I would have to calculate the energy for several pairs of hexamethyldisiloxane for my purpose. Did any colleague calculate the energy of a system similar to this size and could share the estimated time spent on the calculation?

2 - What would be a good starting guess for the number of processors that I could use to calculate SAPT for a system of this size (i.e., 52 atoms)?

Thanks in advance.

The system size is routinely accessible with SAPT2+(3) and TZ basis with psi4.
I’d recommend jun-cc-pVTZ instead of the full aug-cc-pVTZ, the heavy augmentation is not needed for accurate results. Especially on the H atoms. This will save a lot of basis functions.

The scaling beyond say 16 threads is not good but still worth it if the node is free. Running multiple SAPT2+(3) on a single node will slow things down because disk writing/reading may limit the calculations.

Difficult to state a time estimate.

1 Like

@hokru

Thanks for reply.

I cited these 3 SAPT methods (sSAPT0/jun-cc-pVDZ, SAPT2+/aug-cc-pVDZ and SAPT2+(3)dmp2/aug-cc-pVTZ) based on this article https://doi.org/10.1063/1.4867135

I ran these 3 levels of SAPT for methane pairs with 1 processor and the calculations were quick. For SAPT2+(3)dmp2/aug-cc-pVTZ I spent about 5 minutes at each energy point.

One last question: would a system of this size (i.e., 52 atoms) generate temporary files of more than 100 gb?

Many thanks.

It’s possible more than 100 gb are needed. Hard to estimate well.

@hokru

Thanks for reply.

I tried 3 levels of SAPT (i.e., sSAPT0/jun-cc-pVDZ, SAPT2+/aug-cc-pVDZ and SAPT2+(3)dmp2/aug-cc-pVTZ) for my 52 atom system and only sSAPT0/jun-cc-pVDZ just ran correctly.

The others gave a problem of lack of memory.

I will try to work with sSAPT0/jun-cc-pVDZ.