Day 1

Basics PAW Hybrids NewDF Optimization HF-PAW
VASP Workshop: Day 1

Martijn Marsman
Computational Materials Physics, University Vienna, and Center for
Computational Materials Science
VASP Workshop at CSC Helsinki 2009
Marsman
PBCs, k-points, plane waves, DFT
Projector Augmented Wave method
Hybrid functionals
New density functionals
Reaching the electronic groundstate
Hartree-Fock in PAW
Marsman
A system of N electrons
1
2
i +
i
H(r
1 , ..., rN ) = E(r1 , ..., rN )
1
(r1 , ..., rN ) = E(r1 , ..., rN )
V (ri ) +
|ri rj |
i=j
Many-body WF storage requirements are prohibitive

N
(#grid points)
Map onto one-electron theory
(r1 , ..., rN ) {1 (r), 2 (r), ..., N (r)}

such as Hohenberg-Kohn-Sham density functional theory
Marsman
Do not need (r1 , ..., rN ), just the density (r):
E[] = Ts [{i []}] + EH [] + Exc [] + EZ [] + U [Z]
(r1 , ..., rN ) =
N
Y
i (ri )
(r) =
N
X
|i (r)|2
EH [] =
1
2
ZZ
(r)(r )
drdr
|r r |
One-electron Kohn-Sham equations

1
+ VZ (r) + VH [](r) + Vxc [](r) i (r) = i i (r)

2
Hartree
Z
(r )
VH [](r) =
dr
|r r |
Exchange-Correlation
Exc [] =???
Vxc [](r) =???
Per definition: Exc = E Ts EH Eext

In practice: Exchange-Correlation functionals are modelled on the uniform
electron gas (Monte Carlo calculations): e.g., local density approximation
(LDA).
Marsman
Translational invariance implies the existence of a good quantum

number, usually called the Bloch wave vector k. All electronic
states can be indexed by this quantum number
|k
In a one-electron theory, one can introduce a second index,
corresponding to the one-electron band n,
|nk
The Bloch theorem states that the one-electron wavefunctions obey
the equation:
nk (r + R) = nk (r)eikR
where R is any translational vector leaving the Hamiltonian
invariant.
k is usually constrained to lie within the first Brillouin zone in
reciprocal space.
Marsman
z
L
4pi/L
b1
b2
a2
b1
b2
y
a1
a3
y
x
x
b3
b3
b1 =
2
a2 a 3
b2 =
2
a3 a 1
= a1 a2 a3
Marsman
b3 =
2
a1 a 2
ai bj = 2ij
The evaluation of many key quantities, e.g. charge density,

density-of-states, and total energy) requires integration over the
first BZ. The charge density (r), for instance, is given by
(r) =
1
BZ
BZ
fnk |nk (r)|2 dk
fnk are the occupation numbers, i.e., the number of electrons that
occupy state nk.
Exploiting the fact that the wave functions at k-points that are
close together will be almost identical, one may approximate the
integration over k by a weighted sum over a discrete set of points
(r) =
n
wk fnk |nk (r)|2 dk,
where the weights wk sum up to one.
Marsman
The intractable task of determining (r1 , ..., rN ) (for N 1023 ) has

been reduced to calculating nk (r) at a discrete set of points {k} in the
first BZ, for a number of bands that is of the order of the number of
electrons per unit cell.
Marsman
Monkhorst-Pack meshes
Idea: equally spaced mesh in Brillouin-zone.

Construction-rule:
kprs = up b1 + ur b2 + us b3
ur =
2rqr 1
r
2qr
= 1, 2, . . . , qr
bi reciprocal lattice-vectors
qr determines number of
k-points in r-direction
Marsman
Example
r
b
quadratic 2-dimensional lattice
q1 = q2 = 4 16 k-points
only 3 inequivalent k-points ( IBZ)
4 k1 =
4 k2 =
8 k3 =
1
BZ
BZ
( 18 ,
( 38 ,
( 38 ,
1
8)
3
8)
1
8)
1 =
2 =
3 =
IBZ
BZ
1
4
1
4
1
2
F (k)dk 14 F (k1 ) + 41 F (k2 ) + 21 F (k3 )
Marsman
k
k
k! r
b
Algorithm
Algorithm:
calculate equally spaced-mesh
shift the mesh if desired
apply all symmetry operations of Bravais lattice to all k-points
extract the irreducible k-points ( IBZ)
calculate the proper weighting
Common meshes: Two choices for the center of the mesh
centered on ( belongs to mesh).
centered around . (can break symmetry !!)
Marsman
Example - hexagonal cell

before
after
symmetrization
>
H
shifted to G
>
>
>
r
>
in certain cell geometries (e.g. hexagonal cells) even meshes break

the symmetry
symmetrization results in non equally distributed k-points
Gamma point centered mesh preserves symmetry
Marsman
r
>
Marsman
Marsman
The total energy

E[, {R, Z}] = Ts [{nk []}] + EH [, {R, Z}] + Exc [] + U ({R, Z})
The kinetic energy
Ts [{nk []}] =
XX
n
wk fnk nk |
1
|nk
2
The Hartree energy
ZZ
eZ (r)eZ (r )
1
EH [, {R, Z}] =
dr dr
2
|r r |
P
where eZ (r) = (r) + i Zi (r Ri )
The electronic charge density
XX
wk fnk |nk (r)|2 dk,
(r) =
n
The Kohn-Sham equations

1
+ VH [eZ ](r) + Vxc [](r) nk (r) = nk nk (r)

2
The Hartree potential
Z
eZ (r )
VH [eZ ](r) =
dr
|r r |
Marsman
Introduce the cell periodic part unk of the wavefunctions

nk (r) = unk (r)eikr
with unk (r + R) = unk (r).
All cell periodic functions are now written as a sum of plane waves
unk (r) =
1
1/2
CGnk eiGr
G
G eiGr
(r) =
nk (r) =
1
1/2
VG eiGr
V (r) =
CGnk ei(G+k)r
In practice only those plane waves |G + k| are included for which

1
|G + k|2 < Ecutoff
2
Marsman
FFT
real space
reciprocal space
G
cut
b2
0
N1 0
5 4 3 2 1 0 1 2 3 4 5
b1
N/2+1
x =n /N
1
1 1
g = n 2 /
1
1
1
FFT
Crnk =
G
N/2
CGnk eiGr CGnk =

Marsman
1
NFFT
Crnk eiGr
r
FFT
real space
reciprocal space
G
cut
b2
0
N1 0
5 4 3 2 1 0 1 2 3 4 5
b1
N/2+1
x =n /N
1
1 1
g = n 2 /
1
1
1
FFT
Crnk =
G
N/2
CGnk eiGr CGnk =

Marsman
1
NFFT
Crnk eiGr
r
FFT
real space
reciprocal space
G
cut
b2
0
N1 0
5 4 3 2 1 0 1 2 3 4 5
b1
N/2+1
x =n /N
1
1 1
g = n 2 /
1
1
1
FFT
Crnk =
G
N/2
CGnk eiGr CGnk =

Marsman
1
NFFT
Crnk eiGr
r
FFT
real space
reciprocal space
G
cut
b2
0
N1 0
5 4 3 2 1 0 1 2 3 4 5
b1
N/2+1
x =n /N
1
1 1
g = n 2 /
1
1
1
FFT
Crnk =
G
N/2
CGnk eiGr CGnk =

Marsman
1
NFFT
Crnk eiGr
r
FFT
real space
reciprocal space
G
cut
b2
0
N1 0
5 4 3 2 1 0 1 2 3 4 5
b1
N/2+1
x =n /N
1
1 1
g = n 2 /
1
1
1
FFT
Crnk =
G
N/2
CGnk eiGr CGnk =

Marsman
1
NFFT
Crnk eiGr
r
Why use plane waves?
Historical reason: Many elements exhibit a band-structure that can

be interpreted in a free electron picture (metallic s and p elements).
Pseudopotential theory was initially developed to cope with these
elements (pseudopotential perturbation theory).
Practical reason: The total energy expressions and the Hamiltonian
H are easy to implement.
Computational reason: The action H| can be efficiently evaluated
using FFTs.
Marsman
Evaluation of H|nk
1
+ V (r) nk (r)
2
using the convention

r|G + k =
1 i(G+k)r
e
G + k|nk = CGnk
1/2
Kinetic energy:
G + k|
1
1
|nk = |G + k|2 CGnk
2
2
NNPLW
Local potential: V = VH [] + Vxc [] + Vext

) Exchange-correlation: easily obtained in real space Vxc,r = Vxc [r ]
) FFT to reciprocal space {Vxc,r }
{Vxc,G }
4
) Hartree potential: Poisson equation in reciprocal space VH,G = |G|
2 G
) add all contributions VG = VH,G + Vxc,G + Vext,G
) FFT to real space {VG }
{Vr }
The action
1 X
Vr Crnk eiGr
NFFT log NFFT
G + k|V |nk =
NFFT r
Marsman
The action of the local potential

2G cut
4 e2
G2
FFT
V
G
3G cut
G cut
add
FFT
R
G
Marsman
(residual vector)
TAGS and links

Sampling the BZ
The KPOINTS file The file that specifies the k-point sampling for a VASP run.
Sampling the Brillouin zone A lecture from the VASP workshop in Vienna (2003).
Defining the structure

The POSCAR file The file that specifies the configuration of the simulation cell.
INCAR tags
PREC=(N)ormal | (A)ccurate
Sets the precision with which densities and potentials are represented (the amount of aliasing one allows, if any),
and the plane wave basis set kinetic energy cutoff (when ENCUT is not set explicitly).
ENCUT (or ENMAX) = [real]
Plane wave basis set kinetic energy cutoff for WFs.
DFT
Introduction to DFT, DFT in depth Lectures from the VASP workshop in Vienna (2003).
Marsman
The PAW method

The number of plane waves needed to describe
tightly bound (spatially strongly localized) states
the rapid oscillations (nodal features) of the wave functions near the
nucleus
exceeds any practical limit, except maybe for Li and H.
The common solution:
Introduce the frozen core approximation:
Core electrons are pre-calculated in an atomic environment and kept
frozen in the course of the remaining calculations.
Use pseudopotentials instead of exact potentials:
) Norm-conserving pseudopotentials
) Ultra-soft pseudopotentials
) The Projector-Augmented-Wave (PAW) method
[P.E. Bl
ochl, Phys. Rev. B 50, 17953 (1994)]
Marsman
rc
Marsman
1111
0000
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
1111
0000
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000 0000
1111
0000
1111
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000 1111
1111
0000
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
1111
0000
Al
Al
111111
000000
000000
111111
000000
111111
000000
111111
000000
111111
000000
111111
000000
111111
000000
111111
exact potential (interstitial region)

Al
3p
3s
000
111
1111
0000
000
111
0000
1111
2p
000
111
0000
1111
000
111
0000
1111
000
111
0000
1111
000
111
2s
0000
1111
000
111
0000
1111
000
111
0000
1111
000
111
0000
1111
000
111
1s
0000
1111
000
111
0000
1111
effective Al atom
111111111
000000000
2p
1s
1111
0000
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
wave-function
2.5
1111
0000
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
2.0
1.5
Marsman
: E= -0.576 R c=1.9
1.0
111111111
000000000
0.5
000000000
111111111
p : E= -0.205 R c=1.9
000000000
111111111
000000000
111111111
000000000
111111111
000000000
111111111
000000000
111111111
0.0
000000000
111111111
000000000
111111111
1
2
3
4
0
000000000
111111111
000000000
111111111
R (a.u.)
000000000
111111111
000000000
111111111
000000000
111111111
000000000
111111111
000000000
111111111
pseudopotential
PAW Al atom
3p
3s
2p and 1s are
nodeless !!!!
nodal structure
is retained
|n = |n +
(|i |i ) pi |n
|n is a pseudo wave function expanded in plane waves

|i , |i , and |pi are atom centered localized functions
the all-electron partial waves |i are obtained as solutions to the
radial scalar relativistic Schr
odinger equation for the spherical
non-spinpolarized atom
1
( + veff )|i = i |i
2
a pseudization procedure yields
|i |i
veff veff
Marsman
pi |j = ij
the pseudo partial waves |k obey

1
+ veff +
2
ij
|pi Dij pj | |k = k 1 +
ij
|pi Qij pj | |k
with the socalled PAW parameters:

Qij = i |j i |j
1
1
Dij = i | + veff |j i | + veff |j
2
2
The all-electron and pseudo eigenvalue spectrum is identical, all-electron
scattering properties are reproduced over a wide energy range.
Marsman
1st s-channel: 1
Mn 4s bound state
2nd s-channel: 2
Mn s non-bound state
1.5
1
0.5
0
-0.5
-1
-1.5
0
1.5
1
0.5
0
-0.5
-1
-1.5
0
rc
rm
rc
rm
Frozen core approximation:

veff [v ] = vH [v ] + vH [Zc ] + vxc [v + c ]
v (r) =
i
veff [v ] = vH [v ] + vH [Zc ] + vxc [v + c ]
v (r) =
i
Marsman
ai |i (r)|2
ai |i (r)|2
|n
Marsman
|n
|n
|i pi |n
Marsman
|n
|n
|i pi |n
|n
|i pi |n +
|i pi |n
Marsman
Character of wavefunction: clm = plm |n
|n
|n
|lm clm +
11111
00000
00000
11111
00000
11111
00000
11111
00000
11111
00000
11111
00000
11111
1111
0000
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
00000
11111
00000
11111
00000
11111
00000 1111
11111
0000
00000
11111
0000
00000 1111
11111
0000
1111
00000
11111
0000
00000 1111
11111
0000
1111
0000
1111
0000
1111
|lm clm
11111
00000
00000
11111
00000
11111
00000
11111
00000
11111
00000
11111
00000
11111
AE
pseudo
+
pseudo-onsite
1111
0000
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
00000
11111
00000
11111
00000
11111
00000 1111
11111
0000
00000
11111
0000
00000 1111
11111
0000
1111
00000
11111
0000
00000 1111
11111
0000
1111
0000
1111
0000
1111
AE-onsite
Same trick works for

Kinetic energy
Exchange correlation energy
Hartree energy
Wavefunctions
Charge density
Marsman
The kinetic energy

For instance, the kinetic energy is given by
Ekin =
fn n |
1
|n
2
By inserting the transformation (i = lm)

|n = |n +
X
(|i |i ) pi |n
i
into Ekin one obtains: Ekin

X
E
1 + E 1 (assuming completeness)
=E
XX
XX
1
1
1
ij i | |j +
ij i | |j
fn n | |n
2
2
2
site (i,j)
site (i,j)
{z
} |
{z
} |
{z
}
1
E
ij is an on-site density matrix:

X
fn n |
pi pj |n
ij =
n
Marsman
E1
For any (quasi) local operator A there exists a PS operator
X
A = A +
|
pi
i |A|j i |A|j
pj |
ij
so that
e A|
e
|A| = |
For instance the PS operator that corresponds to the density operator

|r r| is given by
X
|r r| +
|
pi
i |r r|j i |r r|j
pj |
ij
and the density

|r r|
=
=
e
|r
r|e +
X
ij
e pi
|
i |r r|j i |r r|j
e(r) e1 (r) + 1 (r)
Non-local operators are more complicated

Marsman
pj |e
The Hartree energy

The pseudo-wavefunctions do not have the same norm as the AE
wavefunctions inside the spheres
To deal with long range electrostatic interactions between spheres a
soft
compensation charge is introduced (similar11111
to FLAPW).
00000
11111
00000
11111
00000
00000
11111
0000
00000 1111
11111
1111
0000
00000
11111
0000
00000 1111
11111
0000
00000 1111
11111
0000
1111
00000
11111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
00000
11111
0000
1111
00000
11111
0000
1111
00000
11111
0000
1111
00000
11111
0000
1111
00000
11111
00000
11111
00000
11111
AE
11111
00000
00000
11111
0000
00000 1111
11111
1111
0000
00000
11111
0000
00000 1111
11111
0000
00000 1111
11111
0000
1111
00000
11111
0000
1111
0000
1111
0000
1111
00000
11111
00000
11111
00000
11111
00000 1111
11111
0000
00000
11111
0000
1111
00000
11111
0000
00000 1111
11111
0000
1111
00000 1111
11111
0000
0000
1111
0000
1111
pseudo + compens. pseudo+comp. onsite AE-onsite
E
1 + E1
Hartree energy becomes: EH = E
EH [
+ ]
EH [1 ]
EH [
1 + 1 ] +
sites
1 one-center pseudo charge

charge
Marsman
sites
1 one-center compensation
PAW energy functional

+ E1 E
1
Total energy becomes a sum of three terms: E = E
=
n
1
+ + c ] +
fn n | |n + Exc [
2
EH [
+ ] +
1
E
=
sites
(i,j)
vH [
Zc ] (
(r) + (r)) d3 r + U (R, Zion )
1
1 + + c ] +
ij i | |j + Exc [
2
vH [
Zc ] 1 (r) + (r) d3 r
EH [
1 + ] +
r
E1
=
sites
(i,j)
1
ij i | |j + Exc [1 + c ] +
2
vH [Zc ]1 (r) d3 r
EH [1 ] +
r
Marsman
is evaluated on a regular grid

E
Kohn-Sham functional evaluated in a plane wave basis set
with additional compensation charges to account for the incorrect
norm of the pseudo-wavefunction (very similar to ultrasoft
pseudopotentials).
=
fn n n
pseudo charge density

compensation charge
1 are evaluated on radial grids centered around each ion.

E 1 and E
Kohn-Sham energy evaluated for basis sets {i } and {i }
these terms correct for the shape difference between the pseudo and
AE wavefunctions.
No cross-terms between plane wave part and radial grids exist.
Marsman
The pseudo wave functions |n (plane waves!) are the

self-consistent solutions of
1
+Veff +
2
ij
|pi (Dij +...) pj | |n = n 1+
ij
|pi Qij pj | |n
1
1
1
1
Dij = i | + veff
[1v ]|j i | + veff
[1v ]|j
2
2
1v (r) =
ij
ij i |r r|j
ij =
n
1v (r) =
ij
ij i |r r|j
fn n |pi pj |n
If the partial waves form a complete basis within the PAW spheres,
then the all-electron wave functions |n are orthogonal to the core
states!
Marsman
Accuracy
atomization energy error [kcal/mol]
Subset of G2-1 test set: Deviation PAW w.r.t. GTO, in [kcal/mol].

10
8
6
4
2
0
-2
-4
-6
PBE(G03)-PBE(V)
PBE0(G03)-PBE0(V)
-8
-10
LiH
SO
Si2H6
CH
H2O
SiH4 C2H4 HCN CH3OH P2
BeH
SO2
CN
H2CO CO2
SiO
ClO
CH4
SiH3 C2H2
|EAE | < 1 kcal/mol.

Marsman
Accuracy
Relative PBE bond lengths of Cl2 , ClF, and HCl for various GTO basis
sets specified with respect to plane-wave results:
aug-cc-pVXZ (X= D,T,Q,5)

N.B.: aug-cc-pV5Z basis set for Cl contains 200 functions!
Marsman
|n = |n +
lm
(|lm |lm ) plm |n
|n is the variational quantity of the PAW method.

The PAW method is often referred to as an all-electron method.
Not in the sense that all electrons are treated explicitly, but in the
sense that the valence electronic wave functions are kept orthogonal
to the core states.
Marsman
11111
00000
00000
11111
00000
11111
00000
11111
00000
11111
00000
11111
00000
11111
00000
11111
1111
0000
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
0000 1111
1111
0000
0000 1111
1111
0000
1111
0000
1111
0000
0000 1111
1111
0000
1111
0000
1111
0000
1111
AE
11111
00000
00000
11111
00000
11111
00000
11111
00000
11111
00000
11111
00000
11111
00000
11111
1111
0000
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
1111
0000
0000 1111
1111
0000
0000 1111
1111
0000
1111
0000
1111
0000
0000 1111
1111
0000
1111
0000
1111
0000
1111
pseudo + compens. pseudo+comp. onsite AE-onsite
This general scheme applies to all operators.

Sometimes one may choose to include only parts of the PAW
expressions.
lazy: only implement plane wave part (GW, ...)
efficient: physics of localized orbitals; only spheres (LDA+U,
DMFT, ..., )
Marsman
Hartree-Fock/DFT hybrid functionals
Definition: Exchange correlation functionals that admix a certain amount

of Fock exchange to (a part of) a local or semi-local density functional.
Present a definite improvement over the (semi)-local density
functional description of the properties of molecular systems.
Some hybrid functionals yield an improved description of structural,
electronic, and thermo-chemical properties of small/medium gap
solid state systems.
Marsman
Hartree-Fock Theory
1 (r1 ) 1 (r1 )
1 (r2 ) 2 (r2 )
1
(r1 , ..., rN ) =
..
..
N!
.
.
1 (rN ) 2 (rN )
Slater determinant
N (r1 )
N (r2 )
..
.
N (rN )
1
(r1 , r2 ) = 1 (r1 )2 (r2 ) 1 (r2 )2 (r1 )
2
Pauli exclusion principle: (r1 , r2 ) = 0 for 1 = 2
1
+ VZ (r) + VH [n](r) i (r) + VX (r, r )i (r )dr = i i (r)
2
Orbital dependent:
N
j (r)j (r ) compare to DFT: Vxc [n](r)i (r)
VX (r, r ) =
|r r |
j
No electronic correlation!
Marsman
Hybrid functionals: PBE0, HSE03

PBE0:
1 HF 3 PBE
E + Ex
+ EcPBE
4 x
4
non-empirical: justified using the adiabatic connection formula.
PBE0
Exc
=
J. Perdew, M. Ernzerhof, and K. Burke, J. Chem. Phys. 105, 9982 (1996).
HSE03:
HSE03
Exc
=
3
1 HF,SR
E
() + ExPBE,SR () + ExPBE,LR () + EcPBE
4 x
4
decomposed Coulomb kernel (Savin et. al.):

1
erfc(r) erf(r)
= S (r) + L (r) =
+
r
r
r
semiempirical: is chosen to yield an optimal description of the
atomization energies of the molecules in Poples G2-1 test set.
J. Heyd, G. E. Scuseria, and M. Ernzerhof, J. Chem. Phys. 118, 8207 (2003).
Marsman
Hybrid functionals: B3LYP
B3LYP
EX
LDA
HF
B88
0.8EX
+ 0.2EX
+ 0.72EX
B3LYP
EC
VWN3
LYP
0.19EC
+ 0.81EC
Semiempirical: coefficients chosen to reproduce experimental atomization

energies, electron and proton affinities and ionization potentials of the
molecules in Poples G2 test set and their atomic constituents.
A. D. Becke, J. Chem. Phys. 98, 5648 (1993).

M. J. Frisch et al., GAUSSIAN03 Rev. C.02, Gaussian Inc., Wallingford, CT 2004.
Marsman
Computational aspects
ExHF
d3 rd3 r kn
(r)qm (r)K(r, r )qm
(r )kn (r )
kn,qm
with
K(r, r ) =
1
erfc(|r r |)
or
|r r |
|r r |
FFT overlap density to reciprocal space
(r )kn (r )}
(G) = FFT{qm
division by Laplace operator and FFT to real space
V (G) =
evaluate
4e2
|G|2 (G),
and V (r) = FFT{V (G)}
kn
(r)qm (r)V (r)d3 r
N.B.: for all combinations of k, q, n, and m

Marsman
ExHF
d3 rd3 r kn
(r)qm (r)K(r, r )qm
(r )kn (r )
kn,qm
Effort:
(Nbands Nk )(Nbands Nq ) NFFT ln NFFT
Bulk: Nbands Natoms , Nk 1/Natoms
NFFT ln NFFT Natoms
Molecular systems: Nk = 1
3
Nbands Nbands NFFT ln NFFT Natoms
Marsman
Computational aspects: Downsampling

Convergence of ExHF w.r.t. the BZ sampling used to represent VxHF
HSE03
0
0
-15
-30
-45
-60
-75
-90
-105
-120
HSE03
[meV]
-15
Ex
Ex
PBE0
[meV]
PBE0
-30
-45
-60
-75
-90
-105
-120
24 12
nq
24 12
nq
Example: Al (fcc)
Short(er) range in real space = Reduced BZ sampling
Marsman
Downsampling cont.
L
1 band
2 k-points
2L
2 bands
1 k-point
unit cells
1
n
Marsman
k-points
n
1
Assume a maximum interaction range R = mL, then a supercell of twice

this size, i.e. 2m unit cells, correctly incorporates all interactions using
only k = 0
This equivalent to the description one obtains using a single unit cell and
an equidistant 2m sampling of the 1st BZ.
Ergo: reducing the range of the Fock exchange interaction in the HSE
functional allows for the representation of the Fock potential on a courser
grid of k-points.
Marsman
The HSE Fock exchange energy

e2 X
ExSR () =
2wk 2wq fkn fqm
2
kn,qm
ZZ
erfc(|r r |)
kn (r)qm (r)qm (r )kn (r ).
d3 rd3 r
|r r |
The representation of the corresponding short-range Fock potential in
reciprocal space
`
VkSR G, G = k + G|VxSR []|k + G =
X Cqm
(G G )Cqm (G G )
4e2 X
2wq fqm
mq
|k q + G |2
G
2
2
1 e|kq+G | /4 .
Full q-grid:
{q} = {
m1
m2
m3
b1 +
b2 +
b3 |mi = 0, ..., Ni 1}
N1
N2
N3
Downsampled q-grid:
{q}k = {k +
3
X
i=1
mi
Ni
Ci
bi |mi = 0, ...,
1}
Ni
Ci
Marsman
Atomization energies of small molecules

atomization energy error [kcal/mol]
Subset of G2-1 test set: Deviation w.r.t. experiment, in [kcal/mol].

20
10
-10
-20
-30
EXP-PBE(V)
EXP-PBE0(V)
LiH
CH
H2O SiH4 C2H4 HCN CH3OH P2
SO Si2H6
BeH CH4 SiH3 C2H2
CN H2CO CO2
SiO
ClO
SO2
Significant improvement of atomization energies

Marsman
G2-1 test set (55 molecules): atomization energies

ME and MAE w.r.t. experiment, in [kcal/mol]
PBE
PBE0
PBE
PBE0
ME
PAW
GTO
+6.43 +6.85
1.47 1.04
0.42 . . .
0.42 . . .
MAE
PAW GTO
8.57 8.79
3.65 3.42
0.46 . . .
0.49 . . .
3 , 1000 eV cutoff
PAW calculations: only, 101112 A
GTO calculations: aug-cc-pV5Z basis set
J. Paier, R. Hirschl, M. Marsman, and G. Kresse, J. Chem. Phys. 122, 234102 (2005).
Marsman
Solid state systems

Set of test systems
Metals: Na, Mg, Li, Al, Cu, Rh, Pd, Ag
Small gap: Si, GaAs, BP, GaP, SiC, -GaN, C, BN, MgO
Large gap: NaCl, LiCl, NaF, LiF
Lattice constants, bulk moduli
Band gaps
Atomization energies
Transition metal monoxides

Adsorption of CO on d-metal surfaces
Marsman
rel. error of latt. constant [%]
Solid state systems: PBE

2
PBE-VASP
PBE-FLAPW
PBE-G03
1.5
1
0.5
0
-0.5
GaAs MgO
Na
Al
Rh
Overall PAW-GTO agreement good:

MAE(PAW)=0.029
A, MAE(GTO)=0.026
A.
Sizeable discrepancies for Li, Al, C, Si, SiC, GaAs, Rh, and Ag.
Probably due to basis set related inaccuracies in the GTO calculations.
J. Heyd and G. E. Scuseria, J. Chem. Phys. 121, 1187 (2004).
J. Paier, et al., J. Chem. Phys. 124, 154709 (2006).
Marsman
Solid state systems: PBE, PBE0, and HSE03

Lattice constants and bulk moduli
PBE
Solid
a0
...
...
C
3.574
Si
5.469
GaAs 5.752
MgO 4.258
Na
4.200
Al
4.040
Rh
3.830
...
...
All systems:
ME
0.039
MAE 0.045
No metals:
ME
0.048
MAE 0.048
HSE03
a0
B
...
...
3.550 466
5.439
96.5
5.696
69.5
4.212 168
4.225
8.45
4.025
81.1
3.786 285
...
...
a0
...
3.549
5.433
5.672
4.211
4.229
4.012
3.785
...
B
...
467
99.0
72.9
169
8.22
86.0
291
...
12.3
12.4
0.012
0.024
2.6
8.6
0.007
0.022
0.1
7.9
...
...
...
...
13.4
13.4
0.010
0.022
0.4
5.7
0.003
0.019
1.9
5.4
...
...
...
...
B
...
431
87.8
59.9
149
7.80
76.6
254
...
PBE0
Exp.
Values given in
A and GPa, respectively.
Marsman
a0
...
3.567
5.430
5.648
4.207
4.225
4.032
3.798
...
B
...
443
99.2
75.6
165
7.50
79.4
269
...
PBE, PBE0, HSE03, B3LYP: lattice constants
Relative error (%)
4.0
PBE
PBE0
HSE03
B3LYP
3.0
2.0
1.0
0.0
-1.0
-2.0
Na
Al
Li
Rh
Cu
Ag
Pd
Si
GaAs GaP
GaN
BP
SiC
C
BN
NaF
NaCl
LiF
MgO
LiCl
Figure 4. Relative error in the PBE, PBE0, HSE03, and B3LYP lattice constants with respect to experiment.
Marsman
Solid state systems: band gaps
16
DFT
HSE
LiF
Theory (eV)
C BN
2
AlP
Ne
Ar
MgO
ZnS
SiC GaN
CdS
PbTe Si
0.5
ZnO
GaAs
0.25
PbSe
0.5
PbS
1
4
Experiment (eV)
Marsman
16
Solid state systems: Atomization energies
rel. error of AE [%]
-10
-20
-30
PBE
HSE03
B3LYP
-40
-50
ME
MAE
Na
Mg
Pd
PBE
0.045
0.134
Ag
BP
Si
PBE0
0.228
0.286
LiCl NaF LiF
HSE03
0.184
0.252
B3LYP
0.590
0.590
in eV/atom.
Marsman
Hybrid functionals overestimate the exchange splitting in

d-elements: leads to an increased stability of the spin-polarized
atom.
B3LYP overestimates Exc of localized electrons.
B3LYP fails to describe free electron like behaviour:
LYP underestimates correlation energy in itinerant systems
(does not attain HEG limit!).
J. Paier, M. Marsman, and G. Kresse, J. Chem. Phys. 127, 024103 (2007).
Marsman
Heats of formation
10
Relative error (%)
-10
-20
PBE
PBE0
HSE03
-30
B3LYP
-40
SiC
NaF
NaCl
MgO
LiCl
LiF
Figure 7. Relative error in the PBE, PBE0, HSE03, and B3LYP

heats of formation with respect to experiment.
Marsman
Transition metal monoxides: MnO

4
Total
t2g
Expt
t2g eg
eg
eg
Theory
-10
-8
-6
-4
-2
10
Relative Binding Energy (eV)
Comparison between PBE0 dprojected DOS (bottom) of both Mn

sites together with the experimental
(top) inverse photoemission data
and the difference between on- and
off-resonance photoemission spectra.
DOS (States/eV - atom - spin)
Intensity (arb.)
2
0
2
1
0
-1
-2
2
1
0
-1
-2
0.8
Mn1, d
t2g
eg
t2g
Mn2, d
eg
O, p
PBE0
PBE+U
PBE
0.4
0
-10
-8
-6
-4
-2
Energy (eV)
C. Franchini, V. Bayer, R. Podloucky, J. Paier, and G. Kresse, Phys. Rev. B 72, 045132 (2005).
Marsman
10
Transition metal monoxides cont.
MnO
LDA
HSE03
exp.
A)
a0 (
4.31
4.44
4.45
FeO
LDA
HSE03
exp.
4.17
4.33
4.33
3.26
3.63
3.32/4.2
2.2
2.4
CoO
LDA
HSE03
exp.
4.10
4.26
4.25
2.23
2.67
3.35/4.0
3.4
2.5
NiO
LDA
HSE03
exp.
4.06
4.18
4.17
1.06
1.65
1.64
0.4
4.2
4.0
Marsman
Ms (B )
4.14
4.52
4.58
(eV)
0.4
2.8
3.9
CO adsorption on d-metal surfaces
DFT predicts incorrectly that CO prefers the

hollow site; P. Feibelman et al., J. Phys.
Chem. B 105, 4018 (2001)
unclear why, but the error is relatively large;
best DFT/PBE calculations:
CO@Cu(111): 170 meV
CO@Rh(111):
40 meV
CO@Pt(111):
100 meV
4 layers, c(2 4), = 0.25 ML, asymmetric
A vacuum.
setup, 10
Marsman
CO adsorption on d-metal surfaces cont. I

CO @
Cu(111)
PBE
PBE0
HSE03
exp.
top
0.709
0.606
0.561
0.46-0.52
fcc
0.874
0.579
0.555
hcp
0.862
0.565
0.535
0.165
0.027
0.006
Rh(111)
PBE
PBE0
HSE03
exp.
1.870
2.109
2.012
1.43-1.65
1.906
2.024
1.913
1.969
2.104
1.996
0.099
0.005
0.016
Pt(111)
PBE
PBE0
HSE03
exp.
1.659
1.941
1.793
1.43-1.71
1.816
1.997
1.862
1.750
1.944
1.808
0.157
0.056
0.069
Marsman
CO adsorption on d-metal surfaces cont. II
Hybrid functionals reduce the tendency to stabilize adsorption at the

hollow sites w.r.t. the top site.
reduced CO 2 metal-d interaction
Improved description of the CO LUMO (2 ) w.r.t. the Fermi level

(shifted upwards).
Downshift of the metal d-band center of gravity in Cu(111).
But: Overestimation of the metal d-bandwidth.
A. Stroppa, K. Termentzidis, J. Paier, G. Kresse, J. Hafner, Phys. Rev. B 76, 195440 (2007).
A. Stroppa and G. Kresse. New Journal of Physics 10, 063020 (2008).
Marsman
Conclusions
Implementation:
Only small discrepancies between the PAW and GTO atomization
energies for molecular systems at the PBE and hybrid functional
level.
For solid state systems the PAW and GTO results already differ at
the PBE level; probably basis set related inaccuracies in the GTO
calculations. PAW results validated through a comparison with
FLAPW calculations.
Range decomposition of the Coulomb kernel in the HSE functional
allows for a reduced Brillouin zone sampling in the representation of
the Fock potential; the HSE results remain similar to the PBE0
results.
Marsman
Conclusions cont.
Solid state systems:
The PBE0 and HSE hybrid functionals provide an improved
description of the structural (lattice constants and bulk moduli) and
electronic (band gap) properties of systems with a small/medium
sized band gap.
PBE0, HSE, and B3LYP atomization energies are in overall worse
agreement with experiment than those obtained using the semi-local
PBE density functional, in case of B3LYP even drastically so. This
is mainly due to a worse description of metallic systems.
CO adsorption on d-metal (111) surfaces: hybrid funtionals reduce
the tendency to stabilize adsorption at the hollow sites w.r.t. the
top site.
Marsman
TAGS and links

INCAR tags
Hybrid functionals and Hartree-Fock The VASP manual chapter on hybrid functionals and Hartree-Fock.
LHFCALC Switch on Hybrid and Hartree-Fock type calculations.
HFSCREEN Specifies the range separating parameter in HSE functionals.
ENCUTFOCK Specicifies the FFT grids used in the HF routines.
NKRED, NKREDX, NKREDY, NKREDZ, etc Downsampling the k-point mesh in the representation of the Fock
potential.
GGA-tag Override the type of density functional specified in the POTCAR.
AEXX, AGGAX, AGGAC and ALDAC The fractions of Fock-exchange, gradient corrections to the exchange and
correlation, and the fraction of LDA correlation.
PBE0: LHFCALC = .TRUE.
HSE06: LHFCALC = .TRUE. ; HFSCREEN = 0.2 (with PBE POTCAR, or GGA = PE).
B3LYP: LHFCALC = .TRUE. ; GGA = B3 ; AEXX = 0.2 ; AGGAX = 0.72 ; AGGAC = 0.81 ; ALDAC = 0.19
Hartree-Fock: LHFCALC = .TRUE. ; AEXX = 1.0 ; ALDAC = 0.0 ; AGGAC = 0
A. V. Krukau , O. A. Vydrov, A. F. Izmaylov, and G. E. Scuseria, J. Chem. Phys. 125, 224106 (2006).
Marsman
New density functionals: new GGAs for solids

AM05
PHYSICAL REVIEW B 72, 085108 2005
Functional designed to include surface effects in self-consistent density functional theory

R. Armiento1,* and A. E. Mattsson2,
1Department
of Physics, Royal Institute of Technology, AlbaNova University Center, SE-106 91 Stockholm, Sweden
2
Computational Materials and Molecular Biology MS 1110, Sandia National Laboratories, Albuquerque,
New Mexico 87185-1110, USA
PBEsol
PRL 100, 136406 (2008)
PHYSICAL REVIEW LETTERS
week ending
4 APRIL 2008
Restoring the Density-Gradient Expansion for Exchange in Solids and Surfaces

John P. Perdew,1 Adrienn Ruzsinszky,1 Gabor I. Csonka,2 Oleg A. Vydrov,3 Gustavo E. Scuseria,3 Lucian A. Constantin,4
Xiaolan Zhou,1 and Kieron Burke5
1
Department of Physics and Quantum Theory Group, Tulane University, New Orleans, Louisiana 70118, USA
2
Department of Chemistry, Budapest University of Technology and Economics, H-1521 Budapest, Hungary
3
Department of Chemistry, Rice University, Houston, Texas 77005, USA
4
Donostia International Physics Center, E-20018, Donostia, Basque Country
5
Departments of Chemistry and of Physics, University of California, Irvine, Irvine, California 92697, USA
(Received 13 July 2007; published 4 April 2008)
Better description of lattice constants and bulk moduli, and (jellium)

surface energies.
Marsman
Lattice constants
PHYSICAL REVIEW B 79, 155107 2009
CSONKA et al.
TABLE I. Statistical data for the equilibrium lattice constants of the 18 test solids of Ref. 38 at 0 K calculated from the SJEOS. The
Murnaghan EOS yields identical results within the reported number of decimal places. Experimental low temperature 550 K lattice
constants are from Ref. 56 Li, Ref. 57 Na, K, Ref. 58 Al, Cu, Rh, Pd, Ag, and Ref. 59 NaCl. The rest are based on room temperature
values from Ref. 60 C, Si, SiC, Ge, GaAs, NaF, LiF, MgO and Ref. 57 LiCl, corrected to the T = 0 limit using the thermal expansion from
Ref. 58. An estimate of the zero-point anharmonic expansion has been subtracted out from the experimental values cf. Table II. The
calculated values are precise to within 0.001 for the given basis sets, although GAUSSIAN GTO1 and GTO2 basis-set incompleteness limits
the accuracy to 0.02 . GTO1: the basis set used in Ref. 38. GTO2: For C, Si, SiC, Ge, GaAs, and MgO, the basis sets were taken from
Ref. 41. For the rest of the solids, the GTO1 basis sets and effective core potentials from Ref. 38 were used. The best theoretical values are
in boldface. The LDA, PBEsol, and PBE GTO2 results are from Ref. 14. The SOGGA GTO1 results are from Ref. 15.
MEa
MAEb
MREc %
MAREd %
LDA
LDA
PBEsol
PBEsol
PBEsol
AM05
SOGGA
PBE
PBE
PBE
GTO2
VASP
GTO2
BAND
VASP
VASP
GTO1
GTO2
VASP
BAND
BAND
0.047
0.050
1.07
1.10
0.055
0.050
1.29
1.15
0.022
0.030
0.45
0.67
0.010
0.023
0.19
0.52
0.012
0.023
0.24
0.52
0.029
0.036
0.58
0.80
0.009
0.024
0.19
0.50
0.075
0.076
1.62
1.65
0.066
0.069
1.42
1.48
0.063
0.067
1.35
1.45
0.048
0.052
0.99
1.10
Mean error.
Marsman
TPSS
TAGS and links
INCAR tags
GGA= AM | PS Select the AM05 or PBEsol GGA functional.
Marsman
Electronic optimization
Direct minimization of the DFT functional (Car-Parrinello, modern)
Start with a set of wavefunctions {n (r)|n = 1, .., Ne /2} (random
numbers) and minimize the value of the functional (iteration)
2
Gradient: Fn (r) =
2me
2 + V eff (r, {n (r )}) n n (r)
The Self Consistency Cycle (old fashioned)

Start with a trial density , set up the Schr
odinger equation, and solve it
to obtain wavefunctions n (r)
2
2me
2 + V eff (r, {(r )}) n (r) = n n (r)
as a result one obtains a new charge density (r) =

new Schr
odinger equation iteration
Marsman
n = 1, ..., Ne /2
n
|n (r)|2 and a
Direct mimization vs. SCC

disordered diamond, insulator
disordered fcc Fe, metal

2
energy
n=1,2,4,8
log 10 E-E 0
-6
self.consistent
|
0
-4
n=1
-6
n=8
n=1
self.consistent
-1
-2
-3
-4
direct
CarParrinello
-8
1
log 10 |F-F 0 |
n=1
n=4
-2
direct
CarParrinello
-4
-8
1
log 10 |F-F
11
00
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
L 11
00
11
n=8
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
0
11
0
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
log 10 E-E 0
-2
10
15
iteration
20
-1
-2
-3
-4
forces
10
20
30
iteration
G. Kresse and J. Furthm

uller, Phys. Rev. B 54, 11169 (1996).
Marsman
40
Direct optimization (charge sloshing)

The derivative of the total energy w.r.t. the wave function n | is
X1
X
n +
Hnm (fn fm )|m
|m m | H|
|gn = fn 1
2
m
m
n .
where Hnm = m |H|
unoccupied
charge
occupied
Consider two states
potential
n = ei(kF k)r m = ei(kF +k)r
slowly varying charge

2
4 e / G
strong change in potential
n = n +sm m
= m sn
(r) = 2sRe ei2kr
VH (r) =
2s 4e2
Re ei2kr
|2k|2
The smallest |k| 1/L where L is the largest dimension of the supercell.
the response of the potential VH L2 stable step size s 1/L2
Marsman
A naively straightforward algorithm

Express the Hamiltonian in the plane wave basis set and diagonalize it
H = G|H[]|G
diagonalize H {i , i } i = 1, .., NFFT
Self-consistency
0 H0 1 = f (0 , ) H1 ...
iterate until =
3
BUT: we do not need NFFT one-electron orbitals, at a cost of O(NFFT
) ...
we only need the Nb lowest eigenvectors of H
Iterative diagonalization of H aimed at finding its Nb lowest eigenvectors

(Nb Nel /cell).
Blocked Davidson algorithm, RMM-DIIS, ...
Marsman
The Self-Consistency Cycle

trial-charge in and trial-wavevectors n
set up Hamiltonian H(in )
iterative refinements of wavefunctions {n }
new charge density out =
fn |n (r)|2
two subproblems
optimization of
{n } and in
refinement of
density:
DIIS algorithm
P. Pulay, Chem. Phys. Lett.
73, 393 (1980)
refinement of density in , out new in
no
<
refinement of
wavefunctions:
DIIS or Davidson
algorithm
calculate forces, update ions
Marsman
Key ingredients: Subspace diagonalization and the Residual

Rayleigh-Ritz: the diagonalization of the Nb Nb subspace
X app
X
nm Bmk =
k Snm Bmk
H
m
with
nm = n |H|
m
H
m
Snm = n |S|
P
The Nb eigenvalues/eigenvectors app
and |k = m Bmk |m are the
k
within the
best approximation to the exact Nb lowest eigenvalues of H
subspace spanned by the one-electron orbitals m .
and
The residual vector:

n ,
app S)|
|R(n ) = (H
with app =
(its norm is a measure of the error in the eigenvector).
Marsman
n
n |H|
n
n |S|
The blocked Davidson algorithm

Take a subset of all bands {n |n = 1, .., N } {k1 |k = 1, .., n1 }
Extend this subset by adding the (preconditioned) residual vectors
to the presently considered subspace
{k1 /gk1 = K(H app S)k1 |k = 1, .., n1 }
Raighley-Ritz optimization (sub-space rotation) in the 2n1
dimensional subspace {k1 /gk1 } to determine the n1 lowest
eigenvectors {k2 |k = 1, .., n1 }.
Extend the subspace with the residuals from {k2 }
{k1 /gk1 /gk2 = K(H app S)k2 |k = 1, .., n1 }
Raighley-Ritz optimization {k3 |k = 1, .., n1 }
etc ...
The optimized {km |k = 1, .., n1 } replace {n |n = 1, .., n1 }
Move on to subset {k1 |k = n1 + 1, .., n2 }, ..., etc, ...
After treating all bands: Raighley-Ritz optimization of {n |n = 1, .., N }
Marsman
Charge density mixing (RMM-DIIS)

Minimization of the norm of residual vector
R[in ] = out [in ] in
with out (r) =
occupied
|R[in ]| min
wk fnk |nk (r)|2
DIIS algorithm is used for the optimization of the norm of the residual vector.
Linearization of R[in ] around sc (linear response theory)
R[] = J( sc )
with the charge dielectric function J
leads to
J = 1 |{z}
U
4e2
q2
R[in ] = out [in ] in = J(in sc )
Marsman
Divergence of the dielectric function

Eigenvalue spectrum of J determines convergence
J=1 U
4e2
q2
Broader eigenvalue spectrum slower convergence

For insulators and semi-conductors, the width of the eigenvalue
spectrum is constant and system size independent ( )!
For metals the eigenvalue spectrum diverges, its width is
proportional to the square of the longest dimension of the cell:
Short wavelength limit J 1 (no screening)
Long wavelength limit J 1/q2 L2 (metallic screening)
Complete screening in metals causes charge sloshing
Marsman
The dielectric matrix

0.4
Use a model dielectric function

that is a good initial approximation
for most systems
AMIX
0.3
0.2
0.1
G1q
2
AMIX
, AMIN)
max( qq2 +BMIX
AMIN
0
3
2
G (1/A )
This is combined with a convergence accelerator.

The initial guess for the dielectric matrix is improved using
information accumulated in each electronic (mixing) step (DIIS).
Marsman
Return to direct optimization: Why?
Pure DFT functional depends only on the density
1
+ Veff [](r) + Vext (r) n (r) = n n (r)
2
DFT-Hartree-Fock Hybrid functional depends explicitly on the wave functions
Z
occ
X
m (r )n (r )
1
+ Veff [](r) + Vext (r) n (r)+C
dr = n n (r)
m (r)
2
|r r |
m
so density-mixing will not work (reliably).
Unfortunately we know direct optimization schemes are prone to charge
sloshing for metals and small-gap systems.
Marsman
Mixed scheme
The gradient of the wave functions is given by
X1
X
n +
|m m | H|
|gn = fn 1
Hnm (fn fm )|m
2
m
m
n
with Hnm = m |H|
A search direction towards the groundstate w.r.t. unitary transformations
between the orbitals within the subspace spanned by wave functions can
be found from perturbation theory
Unm = nm s
Hnm
Hmm Hnn
but this is exactly the term that is prone to charge sloshing!

Solution: Use density mixing to determine the optimal unitary
transformation matrix Unm .
Marsman
Optimal subspace rotation

kl = l |H[]|
Define a Hamilton matrix H

k
where H[]
= T + Vext + Veff [] + VXnl [{}, {f }]
kl
Determine the subspace rotation matrix V that diagonalizes H
Recompute the (partial) occupancies {f }
P
The transformed orbitals l Vnl l and partial occupancies {f } define a
new charge density

mix and
and iterate the above until a stable point is found sc
Hsc
nm = m |H[sc ]|n defines the optimal subspace rotation
Unm = nm s
Hsc
nm
Hsc
nn
Hsc
mm
N.B.: we do not update the orbital dependent part of the Hamiltonian

VXnl [{}, {f }]
Marsman
The full mixed scheme

The iterative optimization of the wavefunctions cycles through the
following steps:
1
construct the Hamiltonian, H, from the current wavefunctions and

partial occupancies, and calculate H| ;
inner loop: determine the self-consistent Hamiltonian, Hsc , defining
the preconditioned direction for the subspace rotation U.
minimization along the preconditioned search direction, defined by
n , U, and a gradient acting on the partial
(1 m |m m |)H|
occupancies. For instance by means of a conjugate-gradient
algorithm.
This loop is repeated until the change in the free energy from one
iteration to the next drops below the required convergence threshold
Ethr (usually 104 eV).
Marsman
It works: fcc Fe
2
10
1 cell
2 cells
4 cells
10
10
-1
log10 | E-E0|
10
-2
10
-3
10
-4
10
-5
10
-6
10
-7
10
-8
10 0
10
20
30
Iteration
The convergence behaviour of HSE03 calculations using the improved direct minimization procedure (solid lines)
and a standard conjugate gradient algorithm (dotted lines). Calculations on single, double, four times, and eight
times repeated cells are marked with circles (
), diamonds (), squares ( ), and triangles (), respectively.
Marsman
TAGS and links
INCAR tags
ALGO The INCAR tag that sets which algorithm is used for the electronic minimization.
IALGO and LDIAG Same as above, but more to choose from (ALGO is the preferred tag). For the direct optimizers
(ALGO =All | Damped) LDIAG = .TRUE. swithches on the density mixer in the determination of the subspace
rotation matrix.
Mixing tags The VASP manual chapter on the settings for the density mixer.
TIME Time step in the direct optimization method ALGO = Damped, and trial time step for the conjugate
gradient direct optimizer ALGO = All.
Electronic optimization Lecture from the VASP workshop in Vienna (2003).
Marsman
Hartree-Fock within the PAW formalism

In principle we would want to follow the original scheme
solve the non-spinpolarized spherical atom within the
(scalar relativistic) Hartree-Fock approximation
compute all-electron partial waves
pseudize ... etc etc ...
Unfortunately this is already problematic in the second step, solving
1
( + vH + vX )|i = i |i
2
with
at.ref
r|
vX |i
a (r)
a
a (r )i (r )
dr
|r r |
= numerically unstable for unbound partial waves with < 0.

Marsman
The HF-PAW method (cont.)

A pragmatic approximate solution:
keep the partial waves and projector functions (|i , |i , and |pi )
obtained with DFT.
orthogonalize the all-electron partial waves |i with respect to the
Hartree-Fock core states: |i |i .
compute the PAW parameters using |i and Hartree-Fock core

states:
Dij
and
1
cv
1
HF
1
1
i | + vH
[1v ] + vH
[HF
Zc ] + vX [{ij }] + vX [{c }]|j
2
1
1
1
1
[1v ] + vH
[Zc ] + vX
[{ij }]|j
i | + vH
2
Qij = i |j i |j
1
1
HF
The most important step is in this scheme: vH
[DFT
Zc ] vH [Zc ].
Marsman

Bulk equilibrium volumes for GaAs and Si
Ga
Valence electrons
As
4s2 4p1
3d10 4s2 4p1
3p6 3d10 4s2 4p1
4s2 4p3
4s2 4p3
3d10 4s2 4p3
3 )
GaAs: 0 (A
PAW
HF-PAW
46.06
47.07
47.74
48.79
47.80
47.80
Valence electrons
Si
3 )
Si: 0 (A
PAW
HF-PAW
3s2 3p2
2s2 2p6 3s2 3p2
41.44
41.91
3s and 3p levels in the spherical Si atom.
3s
3p
3s2 3p2
PAW
HF-PAW
-15.967
-15.918
-2.761
-2.756
-13.205
-13.161
2s2 2p6 3s2 3p2

PAW
HF-PAW
-15.895
-15.908
-2.749
-2.756
-13.146
-13.151
All energies in eV.
Marsman
HF ref.
-15.906
-2.760
-13.147
41.86
41.93

The remaining discrepancies are related to the fact that the one-center
expansions of the wave functions inside the PAW spheres are not
complete enough (particularly further away from the nucleus)
n (r) =
i
i (r) pi |n
n (r) =
i
i (r) pi |n
This causes problems for at least two reasons:

1
The core-valence exchange interaction is computed between the

Hartree-Fock core states {HF
c } and the one-center expansion of
only.
The point where the Hartree potentials arising from the all-electron
and pseudo ionic cores match moves into regions where the
one-center expansions are not good. This is bound to happen to
1
1
HF
some degree since we adjust vH
[DFT
Zc ] vH [Zc ] and leave
1
vH [Zc ] fixed.
Marsman

P
ei |r r|
ej pej |en
|en (r)|2 = ij en |e
pi
e = e1
Marsman
0.12
0.10
0.08
0.06
0.04
0.02
0.00
0.0
PS PW
PS one center
0.5
1.0
1.5
2.0
2.5
3.0

P
ei |r r|
ej pej |en
pi
e = e1
ij
en |e
pi i |r r|j pej |en =
1 = 1HF
0.12
0.10
0.08
0.06
0.04
0.02
0.00
0.0
0.12
0.10
2 0.08
|ref
HF (r)|
0.06
0.04
0.02
0.00
0.0
Marsman
PS PW
PS one center
0.5
1.0
1.5
2.0
2.5
3.0
HF ref
AE one center
0.5
1.0
1.5
2.0
2.5
3.0

P
ei |r r|
ej pej |en
pi
e = e1
ij
en |e
pi i |r r|j pej |en =
1 = 1HF
e e +
0.12
0.10
0.08
0.06
0.04
0.02
0.00
0.0
0.12
0.10
2 0.08
|ref
HF (r)|
0.06
0.04
0.02
0.00
0.0
1HF
Marsman
0.12
0.10
0.08
0.06
0.04
0.02
0.00
0.0
PS PW
PS one center
0.5
1.0
1.5
2.0
2.5
3.0
HF ref
AE one center
0.5
1.0
1.5
2.0
2.5
3.0
HF ref
PAW AE total
0.5
1.0
1.5
2.0
2.5
3.0
TAGS and links
INCAR tags
LRHFATM = .TRUE. Changes the DFT AE core charge density (1 ) to a Hartree-Fock one.
Marsman
Talks and examples
Handson sessions
Handson
Handson
Handson
Handson
Session
Session
Session
Session
1
2
3
4
Lectures
SESSION
SESSION
SESSION
SESSION
SESSION
SESSION
SESSION
SESSION
SESSION
SESSION
SESSION
SESSION
1:
2:
3:
4:
5:
6:
7:
8:
10:
11:
12:
13:
Introduction to Computational Materials Science

Introduction to DFT
Pseudopotentials I
Pseudopotentials II
Sampling the Brillouin zone
Ionic relaxation methods
Electronic relaxation methods
Computational Platforms
Accuracy and Validation of results
Pseudopotential Data Base
DFT in depth
Unpaired electrons in DFT
Marsman
Some literature
Hybrid functionals in VASP

The Perdew-Burke-Ernzerhof exchange-correlation functional applied to the G2-1 test set using a plane-wave basis
set, J. Paier, R. Hirschl, M. Marsman, and G. Kresse, J. Chem. Phys. 122, 234102 (2005).
Screened hybrid density functionals applied to solids, J. Paier, M. Marsman, K. Hummer, G. Kresse, I. C. Gerber,
and J. G. Angy
an, J. Chem. Phys. 124, 154709 (2006); Erratum to the previous, J. Chem. Phys. 125, 249901
(2006).
Why does the B3LYP hybrid functional fail for metals, J. Paier, M. Marsman, and G. Kresse, J. Chem. Phys.
127, 024103 (2007).
Hybrid functionals applied to extended systems, M. Marsman, J. Paier, A. Stroppa, and G. Kresse, J. Phys.:
Condens. Matter 20, 064201 (2008).
Marsman

Day 1

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Day 1

Uploaded by

Copyright:

Available Formats

Basics PAW Hybrids NewDF Optimization HF-PAW

VASP Workshop: Day 1

VASP Workshop at CSC Helsinki 2009

VASP Workshop: Day 1

Basics PAW Hybrids NewDF Optimization HF-PAW

PBCs, k-points, plane waves, DFT

Projector Augmented Wave method

New density functionals

Reaching the electronic groundstate

VASP Workshop: Day 1

Basics PAW Hybrids NewDF Optimization HF-PAW

Many-body WF storage requirements are prohibitive

(r1 , ..., rN ) {1 (r), 2 (r), ..., N (r)}

VASP Workshop: Day 1

Basics PAW Hybrids NewDF Optimization HF-PAW

Do not need (r1 , ..., rN ), just the density (r):

E[] = Ts [{i []}] + EH [] + Exc [] + EZ [] + U [Z]

One-electron Kohn-Sham equations

+ VZ (r) + VH [](r) + Vxc [](r) i (r) = i i (r)

Vxc [](r) =???

Per definition: Exc = E Ts EH Eext

VASP Workshop: Day 1

Basics PAW Hybrids NewDF Optimization HF-PAW

Translational invariance implies the existence of a good quantum

VASP Workshop: Day 1

Basics PAW Hybrids NewDF Optimization HF-PAW

VASP Workshop: Day 1

Basics PAW Hybrids NewDF Optimization HF-PAW

The evaluation of many key quantities, e.g. charge density,

fnk |nk (r)|2 dk

wk fnk |nk (r)|2 dk,

where the weights wk sum up to one.

VASP Workshop: Day 1

Basics PAW Hybrids NewDF Optimization HF-PAW

The intractable task of determining (r1 , ..., rN ) (for N 1023 ) has

VASP Workshop: Day 1

Basics PAW Hybrids NewDF Optimization HF-PAW

Idea: equally spaced mesh in Brillouin-zone.

VASP Workshop: Day 1

Basics PAW Hybrids NewDF Optimization HF-PAW

F (k)dk 14 F (k1 ) + 41 F (k2 ) + 21 F (k3 )

VASP Workshop: Day 1

Basics PAW Hybrids NewDF Optimization HF-PAW

VASP Workshop: Day 1

Basics PAW Hybrids NewDF Optimization HF-PAW

Example - hexagonal cell

in certain cell geometries (e.g. hexagonal cells) even meshes break

VASP Workshop: Day 1

Basics PAW Hybrids NewDF Optimization HF-PAW

VASP Workshop: Day 1

Basics PAW Hybrids NewDF Optimization HF-PAW

VASP Workshop: Day 1

Basics PAW Hybrids NewDF Optimization HF-PAW

The total energy

The Hartree energy

The Kohn-Sham equations

+ VH [eZ ](r) + Vxc [](r) nk (r) = nk nk (r)

VASP Workshop: Day 1

Basics PAW Hybrids NewDF Optimization HF-PAW

Introduce the cell periodic part unk of the wavefunctions

In practice only those plane waves |G + k| are included for which

VASP Workshop: Day 1

Basics PAW Hybrids NewDF Optimization HF-PAW