Going beyond Local Density and
Gradient Corrected XC functionals
in Quantum-ESPRESSO
Stefano de Gironcoli
SISSA and INFM DEMOCRITOS
n(r) is a smooth and simple function of r
F[n] is a very non trivial functional of n(r)
Jacob's laddder of Density Functional Theory
LDA and LSDA
GGA : PW91, PBE, revPBE, RPBE, BLYP
META-GGA: PKZB, TPSS,
SIC, DFT+U, hybrids
Van der Waals functionals
Still simple approximations can work reasonably
Ceperley and Alder
Phys. Rev. Lett. 45, 566 - 569 (1980)
From L(S)DA
to GGA
… not a unique recipe
A lot of work is gone in proposing and comparing new
functionals
A few functional are widely used today
PW91, PBE, revPBE, RPBE ...
BLYP
From GGA
to Meta-GGA
Formulation is significantly more complicated and it is not
widely used.
Still ...
Electron Localization Function
ELF can distinguish metallic (ELF=~0.5) from
covalent (ELF=~1.0) regions of the electron density
Meta-GGA has the potential to treat in a different way
correlation in electronically different materials.
Problems with LDA / GGA functionals
Chemical accuracy (1 kcal/mol) is far.
- trends are often accurate for strong bonds (covalent, ionic,metallic)
- weak bonds/small overlaps are problematic
Self interaction cancellation is only approximately
verified in LDA and GGA.
- molecular dissociation limit, TMO & RE and other atom-in-solid system.
Van der Waals interactions are not taken into account
- occasional agreement with exp. from compensation of errors
Self Interaction Error
One important source of error in LDA/GGA that can lead
to qualitatively wrong results is the only approximate
cancellation of self interaction coming from the approximate
treatment of exchange
LDA / GGA can badly fail for TMO and in 4f- systems
Electronic Structure of FeO
SIC, DFT+U, hybrids
Self interaction correction was proposed as early as in 1981
by Perdew-Zunger. Conceptually important but not widely
used.
DFT+U has been introduced by Anisimov, Zaanen and
Andersen as an approximation to treat strongly correlated
materials. It has been more recently been applied also in
more normal system with encouraging results.
Hybrid functionals (like PBE0, B3LYB) mix a fraction of
Self-interaction-free HF with LDA/GGA functionals.
Is the method preferred by chemists.
It is very expensive in a plane-wave basis.
LDA/GGA failure in the atomic limit
A correction is needed to remove spurious self-interaction
Electronic Structure of FeO
Electronic Structure of Ceria
EXPT LDA+U
a0=5.41 A a0=5.41 A
EXPT LDA+U
a0=3.88 A a0=3.89 A
deGironcoli@[Link]
Evaluating the U parameter
In atoms: U is the (wrong) LDA/GGA curvature of the
total energy as a function of occupation number.
In solids: U should be extracted from the curvature of ELDA
with respect to occupation number, after correcting for band
structure effects present also for-non interacting system :
Evaluating the U parameter
[Link] and S. deGironcoli, PRB 71, 035105 (2005)
Hartree-Fock energy
● Hartree-Fock
● Exact Exchange (OEP)
● Hybrid Functionals: HH, B3LYP, PBE0
(range separated) HSE
HF Vx using PWs
FFT pseudo wfc to real space
For each qpoint and each occupied band build “charge density”
FFT charge to [Link] and solve Poisson eq.
FFT back to real space, multiply by wfc and add to result
The q+G=0 divergence
Gygi-Baldereschi PRB 34, 4405 (1986)
Integrable divergence
The q+G=0 divergence
Gygi-Baldereschi PRB 34, 4405 (1986)
The q+G=0 divergence
Gygi-Baldereschi PRB 34, 4405 (1986)
Simple Molecules
Energies in kcal/mol = 43.3 meV
Scaling
● Kinetic energy and local Potential
● Non local potential
● Fock operator
Scaling
● Kinetic energy and local Potential
● Non local potential
● Fock operator
From 10 to 100 times slower than standard case
Scaling
● Kinetic energy and local Potential
● Non local potential
● Fock operator
From 10 to 100 times slower than standard case
Moore's law: computer power doubles every 18 months
( a factor of 10 in 5 yrs)
Scaling
● Kinetic energy and local Potential
● Non local potential
● Fock operator
From 10 to 100 times slower than standard case
Separation of long- and short-range part in X can help
van der Waals
van der Waals interaction is relatively weak
but widespread in nature.
An important source of stability for molecular
solids and physisorption of molecules on
surfaces.
It is due to truly non-local correlation effects.
It is contained in the true XC functional but
LDA/GGA/MetaGGA and Hybrids do not describe it properly.
How to deal with van der Waals ?
- neglect it
- add an empirical damped dispersion correction
Grimme, Tatckenko
- develop a truly non local XC functional
Vdw-DF, vdw-DF2, VV09, VV10
- RPA and beyond RPA
How to deal with van der Waals ?
- add an empirical damped dispersion correction
S. Grimme , J. Comp. Chem 27, 1787 (2006)
How to deal with van der Waals ?
- develop a truly non local XC functional
Phi(n(r),grad n, n(r'), grad n', |r-r'|)
-vdw-DF : Dion et al, PRL 92, 246401 (2006)
-vdW-DF2 : Lee et al, PRB 82, 081101 (2010)
-VV09 : Vydrov and Van Voorhis, PRL 103, 063004 (2009)
-VV10 : Vydrov and Van Voorhis, JCP 133, 244103 (2010)
How to deal with van der Waals ?
- develop a truly non local XC functional
Phi(n(r),grad n, n(r'), grad n', |r-r'|)
SUMMARY
LDA and LSDA
simple and well defined. good geometry, overbinding
GGA : PW91, PBE, revPBE, RPBE, BLYP
many options, improved energetics, good geometry
META-GGA: PKZB, TPSS,
more complicated, not very much used
SIC, DFT+U, hybrids
address the self-interaction error with some drawback
Van der Waals functionals
truly non local, very active
THE END