It is well known that self-interaction error (SIE) in DFT leads to over-delocalization of unpaired spins in open-shell molecules.
227
Chem. Rev.
(2012),
112,
pp. 289.
Link
This has a variety of deleterious effects, including overstabilization of three-center, two-electron “hemibonds”,
1003
J. Phys. Chem. Lett.
(2021),
12,
pp. 8053.
Link
fractional charges in well-separated chemical moieties (i.e., upon dissociation),
1346
J. Chem. Phys.
(1998),
109,
pp. 2604.
Link
,
1044
J. Chem. Phys.
(2006),
125,
pp. 194112.
Link
and too-low reaction barriers, the latter of which was largely the motivation for the introduction of hybrid density
functionals.
762
J. Phys. Chem. A
(2001),
105,
pp. 2936.
Link
,
547
Chem. Soc. Rev.
(2021),
50,
pp. 8470.
Link
Although various ad hoc self-interaction correction schemes have been introduced
over the years, none of them is entirely satisfactory.
547
Chem. Soc. Rev.
(2021),
50,
pp. 8470.
Link
Density-corrected (DC-)DFT
1236
J. Chem. Theory Comput.
(2019),
15,
pp. 6336.
Link
,
1124
J. Chem. Theory Comput.
(2022),
18,
pp. 817.
Link
,
1096
J. Am. Chem. Soc.
(2022),
144,
pp. 6625.
Link
represents a revival of an old
idea
884
J. Chem. Phys.
(1994),
100,
pp. 6550.
Link
to avoid SIE by evaluating a DFT functional non-self-consistently using self-consistent Hartree-Fock density,
which is SIE-free. Self-consistent iterations at the DFT level are avoided as this would re-introduce SIE into the density. If
represents the user’s chosen density functional and represents the Hartree-Fock functional, then the DC-DFT energy functional
is
(5.73) |
DC-DFT affords barrier heights that are comparable in accuracy to those obtained with hybrid functionals, even if is a semilocal
functional.
546
J. Chem. Phys.
(2008),
128,
pp. 244112.
Link
,
1225
J. Chem. Phys.
(2014),
140,
pp. 18A534.
Link
,
1002
J. Phys. Chem. Lett.
(2022),
13,
pp. 5275.
Link
This does not really reduce the cost of hybrid DFT calculations since the Hartree-Fock
calculation must be iterated to self-consistency, nevertheless DC-DFT may serve as a useful diagnostic tool. If the DC-DFT result with a given
functional is qualitatively different than the self-consistent DFT result with the same functional, then density-driven SIE may be affecting the results.
This diagnostic capacity has been used, for example, to detect unrealistic delocalization of polaron (spin) defects in metal oxides.
1002
J. Phys. Chem. Lett.
(2022),
13,
pp. 5275.
Link
Users of Q-Chem’s implementation of DC-DFT are asked to cite Ref.
1002
J. Phys. Chem. Lett.
(2022),
13,
pp. 5275.
Link
.
Analytic energy gradients for DC-DFT are available,
1002
J. Phys. Chem. Lett.
(2022),
13,
pp. 5275.
Link
but because the
functional is not iterated to self-consistency evaluation of the gradient requires solution
of coupled-perturbed (-vector) equations,
1225
J. Chem. Phys.
(2014),
140,
pp. 18A534.
Link
,
1002
J. Phys. Chem. Lett.
(2022),
13,
pp. 5275.
Link
which makes the gradient somewhat more expensive
than a traditional DFT gradient.
Note: At present, the coupled-perturbed equtions for DC-DFT are solved in serial, meaning that while the SCF iterations are parallelized the -vector iterations are not.
To perform a DC-DFT calculation, set use METHOD in the $rem section to select the functional of choice, and also set DC_DFT = TRUE. Note that because is never diagonalized, any subsequent properties that are computed at the end of the SCF procedure are based on the Hartree-Fock density. This includes one-particle energy levels, Mulliken charges, multipole moments, etc.
DC_DFT
DC_DFT
Controls whether to use DC-DFT.
TYPE:
Boolean
DEFAULT:
FALSE
OPTIONS:
FALSE
Do not do DC-DFT.
TRUE
Iterate the density to self-consistency at the Hartree-Fock level and then perform
evaluate using the functional specified with METHOD.
RECOMMENDATION:
Use if desired. Analytic gradients are available but are a serial bottleneck in the present implementation.