Questions about long range corrected DFT

Dear users,

I’ve tried to calculate wPBE functional with 6-311g_2d_2p_ basis set.

But the calculate failed with core-dumped messages and I found psi4_scratch directory has about 240Gb scratch files.

Also, when the calculation started, it used very large memories of the cluster (about 90%) even if specifying the amount of memory.

Here is my input file and 10 threads of OpenMP was used.


molecule {
1 2
atomic symbol x y z (100 atoms)
units ANG

set reference uhf
set basis 6-311g_2d_2p_
set scf_type direct
set d_convergence 1e-4
set e_convergence 1e-4
set maxiter 2000
set fail_on_maxiter false

E, wfn = energy(‘wPBE’, return_wfn=True)
oeprop(wfn, ‘MULLIKEN_CHARGES’, title=‘charge_test’)

and last section of output file is

==> Iterations <==

                       Total Energy        Delta E     RMS |[F,P]|

with no energy or delta.

Thank you in advance!

Hm, that’s strange. Have you tried “set scf_type df” for density fitting? That would run a lot faster. It might or might not avoid whatever bug you’re hitting.

Dear Professor Sherrill,

Thank you for the reply! I’ve tried “set scf_type df” but the calculation still use a lot of memory and also failed.

I’ve also did standard pbe and the result is similar. (lots of memory and core-dumped)

Does psi4 have limitation to number of atoms? PBE and wPBE were calculated well with testing O2 system.

100 atoms should be ok for DFT. Maybe if you can send me the input file we can try to figure out what is going on.

The direct computation uses a DF guess to start the computation. If you are constrained on Disk, especially for large computations its recommended to add the option DF_SCF_GUESS False to avoid the initial DF iterations.