wB97X-D gave different energy by QChem

nakatamaho · December 25, 2020, 3:27am

Dear developers,
I noticed wB97X-D calculation by QChem gave different energy to other quantum chemistry packages.
Ne cc-pvdz/wb97x-d

Psi4: -128.8874883792256
-128.887480845 [*]
GAMESS: -128.8874855766
Qchem: -128.88384634 [!]
[!] different

github.com

psi4/psi4/blob/master/samples/dft-custom-hybrid/input.dat

#! DFT (hybrids) test of implementations in: hybrid_superfuncs.py

##ORCA v.4.0.1, cc-pVDZ, RI-MP2, no FC, GRID7
# PW6B95
# WB97X-D3
# HF-3c
# PBEh-3c
# PBE0
##Gaussian 09: 5D SP Int(Grid=99770)
# wb97x
# SOGGA11-X
##Gaussian 16: cc-pvdz 5D SP int=(grid=99770)
# MN15

# MISSING: 
#          "wpbe"     : build_wpbe_superfunctional,
#          "wpbe0"    : build_wpbe0_superfunctional,
#          "b5050lyp" : build_b5050lyp_superfunctional,

# more tests in tests/gcp/

This file has been truncated. show original

Note that I can reproduce the same energies among QC packages for Ne b3lyp/cc-pvdz as follows:
Ne cc-pvdz/b3lyp
Psi4: -128.90946300805294
GAMESS: -128.9094629934
Qchem: -128.90946758

I uploaded inputfiles and output files here: https://www.dropbox.com/s/uc25gi1f7np40pi/Ne.qchemdifferentwb97xd.tar.gz?dl=0

Thank you

epif · December 28, 2020, 7:46pm

The energy difference is caused by different grids used in Q-Chem and other packages for ωB97X-D. Since Q-Chem 4.4, the default grid for range-separated functionals in Q-Chem is SG-2 (https://onlinelibrary.wiley.com/doi/abs/10.1002/jcc.24761).

Using XC_GRID=1 will invoke SG-1 in Q-Chem (default prior to 4.4) and yield an energy consistent with the other results listed here.

nakatamaho · December 29, 2020, 1:42am

Thank you very very much! I could reproduce the results!

jherbert · December 30, 2020, 6:30am

Just as a clarification, default is SG-2 for functionals based on wB97, since those are more challenging to integrate, but not for all range-separated functionals. Something like LRC-wPBE would still default to SG-1.