Introduction to the Theory of

Pseudopotentials

Patrick Briddon
Materials Modelling Group
EECE,
University of Newcastle,
UK.
Contents
Pseudopotential theory.
The concept
Transferability
Norm conservation
Non-locality
Separable form
Non-linear core corrections

Pseudopotentials
A second main plank of modern
calculations
Key idea - only valence electrons
involved in chemical reactions
e.g. Si = [1s
2
2s
2
2p
6
]3s
2
3p
2
Chemical bonding controlled by overlap
of 3s
2
3p
2
electrons with neighbouring
atoms.
Idea: avoid calculating the core states
altogether.
The problem with core states
vary rapidly. This makes
plane wave expansions impossible.
Gaussian expansions difficult
Expensive and hard to do.
oscillate - positions of nodes is
important.
Core states are very hard to describe
accurately. They:
Core states contd.
make the valence states oscillate.
require relativistic treatment.
make the energy very large. This makes
calculations of small changes (e.g.
binding energies) very hard.
Core states:
Empirical Pseudopotentials
Main idea is to look for a form for the
potential V
ps
(r) so that the solutions to:
( ) ( )

+ + + = E
ps
V T r

for a reference system agree with expt.

E.g. get band structures of bulk Si, Ge.
Then, use the potentials to look at SiGe or
SiGe microstructures.
Transferability
Problem: these pseudopotentials cannot
be transferred from one system to
another.
e.g. diamond pseudopotential no good
for graphite, C
60
or CH
4
.
Transferability
Why is this?
the valence charge density is very
different in different chemical situations
- only the core is frozen.
We should not try to transfer the
potential from the valence shell.
Ionic Pseudopotentials
We descreen the pseudopotential:
Split charge density into core and
valence contributions:
( ) ( ) ( ) r r r
v c
n n n + =
( )
( ) density charge valence n
density charge core
v
c
=
=
r
r n
Ionic Pseudopotentials
( ) ( )
( )
( ) | | r
r r
r r
r r
v
xc
v
n V
d n
V V
'

=
}
ps ps
ion
Then construct the transferrable ionic
pseudopotential:
We have subtracted the potential from the
valence density. The remaining ionic
pseudopotential is more transferrable.
Ab Initio pseudopotentials
This approach allows us to generate
pseudopotentials from atomic calculations.
These should transfer to solid state or
molecular environment.
ab initio approach possible.
Look at some schemes for this.
Pseudopotentials that work from H to Pu
by Bachelet, Hamann and Schluter (1982)
Norm Conservation
A key idea introduced in 1980s.

Peviously defined a cutoff radius r
c:
if r > r
c
, V
ps
= V
true
.

Now require +
ps
= +
true
if r > r
c
.

Typically match +
ps
and first two (HSC)
or four (TM) derivatives at r
c


Cutoff Radius
r
c
is a quality parameter NOT an
adjustable parameter.

We do not fit it!

Small r
c
means +
ps
= +
true
for greater
range of r more accurate.

Cutoff Radius

BUT, small r will lead to rapidly varying
+
ps
(eventually it will have nodes).
Use biggest r
c
that leaves results
unchanged.
Generall somewhere between
outermost maximum and node.
Schemes
Kerker (1980)
not widely used

Hamann, Schlter, Chiang, 1982
basis of much future work

Bachelet, Hamann, Schlter, 1982
fitted HSC procedure for all elements
Schemes
Troullier, Martins (1993)
An improvement on BHS
refinement to HSC procedure
widely used today
Vanderbilt (1990)
ultrasoft pseudopotentials
Important for plane waves
widely used today
Schemes
Troullier, Martins (1993)
An improvement on BHS
refinement to HSC procedure
widely used today
Vanderbilt (1990)
ultrasoft pseudopotentials
Important for plane waves
widely used today
Schemes
Hartwigsen, Goedecker, Hutter (1998)
Separable
Extended norm conservation
The AIMPRO standard choice

BUT ...

ALL LOOK COMPLETELY DIFFERENT!
Accuracy
Look at atoms in different reference
configuation.

E.g. C[2s
2
2p
2
] and C[2s
1
2p
3
].

AE = 8.23 eV (all electron)
AE = 8.25 eV (pseudopotential)


Silicon Pseudopotential
Silicon Pseudopotential
Some things to note:

Asymptotic behaviour correct, r>r
c

Non-singular at origin (i.e. NOT 1/r)

Very different s, p, d forms
Pseudo and All electron
Wavefunctions (Si)
Silicon Wavefunctions
Some things to note:

Nodeless pseudo wavefunction, r>r
c

Agree for r>r
c.
Cutoff is around 2.

Smooth not rapidly varying

Log derivative
c
r
dr
R d ln
Non-locality
Norm conserving pseudopotentials are
non-local (semi-local).

This means we canot write the action of
potential thus:
( ) l l r V V
l
l
ps

=
=
0

( ) ( ) ( ) ( ) O' O ' = '

*
, ,

lm lm
lm
l
ps
Y Y r r V V r r
( ) ( ) r r + = + V V
( ) ( ) ( ) r r r V r r V
l l
' = ' o ,
Non-locality
Instead we have different potentials for
different atomic states :
( ) l l r V V
l
l
ps

=
=
0

( ) ( ) ( ) ( ) O' O ' = '

*
, ,

lm lm
lm
l
ps
Y Y r r V V r r
( ) ( )
( ) ( ) r r
r r
p p p
s s s
V V
V V
+ = +
+ = +

( ) ( ) ( ) r r r V r r V
l l
' = ' o ,
This is the action of an operator which
my thus be written as
Non-locality
( ) l l r V V
l
l
ps

=
=
0

or
( ) ( ) ( ) ( ) O
'
O
'
=
'

*
, ,

lm lm
lm
l
ps
Y Y r r V V r r
with ( ) ( ) ( ) r r r V r r V
l l
'
=
'
o ,
( ) ( ) r r r r
' '
+
'
= +
}
d ,

V V
Kleinman Bylander Form
Problem: Take matrix elements in the basis
set
i
(r), i=1, N:
( ) ( ) ( ) ( ) ( )
( ) ( ) ( )

}
} }
=
' ' '
=
lm
l
j
lm
i
lm
j
ps
i j i
dr r V r F r F r
d d V d V
* 2
4
,

t
r r r r r r r r r
( ) ( ) ( ) 0 0 d d Y r F
lm i
i
lm
,
}
= r
where
Kleinman-Bylander Form
Problem is: There are N
2
integrals per
atom is the basis set is not localised.
A disaster for plane waves.
Not the best for Gaussians
Recall there is no such things as the
pseudopotential.
Can we chose a form that helps us out?
Kleinman Bylander Form contd
Kleinman and Bylander wrote
So that this time
( ) ( ) ( ) r V r V r r V
l l l
'
=
'
,

=
lm
j
lm
i
lm ij
F F V
*
( ) ( ) ( ) r r d Y r V F
lm l i
i
lm
0 ,
}
=
where
N integrals per atom. Improvement crucial
for plane wave calculations to do 100 atoms
Kleinman Bylander Form contd
The Kleinman and Bylander form
Is called SEPARABLE or sometimes
FULLY NON-LOCAL
( ) ( ) ( ) r V r V r r V
l l l
'
=
'
,
They:
1. Developed a standard pptl e.g. BHS
2. Modified it to make it separable.
The HGH pseudopotentials
HGH pseudopotentials are also fully separable.

They proposed a scheme to generate in this
way directly (i.e. Not a two stage process).

Thus they avoided issues with ghost states
that were initially encountered when trying
to modifuy a previously generated pptl.
The HGH pseudopotentials
HGH pseudopotentials are also fully separable.

They proposed a scheme to generate in this
way directly (i.e. Not a two stage process).

Thus they avoided issues with ghost states
that were initially encountered when trying
to modifuy a previously generated pptl.
Non-Linear Core Corrections
An issue arises when constructing ionic
pseudopotentials:
( ) ( )
( )
( ) | | r
r r
r r
r r
v
xc
v
n V
d n
V V
'

'
=
}
ps ps
ion
We have subtracted the potential coming
from valence charge density.
Non-Linear Core Corrections
contd
OK for Hartree potential as:
( )
( ) ( )
} }
'

+
'

=
r r
r r
r r
r r d n d n
n V
v c
H

However:
( ) ( ) | | ( ) | | r r
v
xc
c
xc
xc
n V n V n V + =
( ) ( ) ( )
3 1 3 1 3 1
v c v c
n n n n + = +
clearly
Non-Linear Core Corrections
contd
This is true if valence and core densities do
not overlap spatially.

i.e. Core states vanish before valence states
significant.

Problem: this just does not always happen.
NLCC contd
Is a problem when it is difficult to decide
what is a core electron and what is a valence
electron.

e.g. Cu: 1s
2
2p
2
2p
6
3s
2
3p
6
4s
2
3d
10

The issue is the 3d electrons a filled shell.
Largely do not participate in bonding. Are
they core ot not?
NLCC contd
What about

e.g. Zn: 1s
2
2p
2
2p
6
3s
2
3p
6
4s
1
3d
10

The same question. What happens if we
look at ZnSe using 3d in the core?

What about ZnO?
Effect of large core
core Val
Aa
0
AlAs [1s
2
2s
2
2p
6
] 3s
2
3p
1
0%
GaAs 3s
2
3p
6
3d
10
4s
2
4p
1
-2%
InAs 4s
2
4p
6
4d
10
5s
2
5p
1
-4%
ZnSe 3s
2
3p
6
3d
10
4s
2
-10%


Non-Linear Core Corrections
contd
A solution is to use a NLCC

Descreen with the potential from the total
density, not just the valence density:
( ) ( )
( )
( ) | | r
r r
r r
r r
tot
xc
v
n V
d n
V V
'

'
=
}
ps ps
ion
Non-Linear Core Corrections
contd
Fixes lattice constant completely for GaAs,
InAs. Good for GaN, ZnSe.

Band structure still be affected. CARE.

NLCC will not work if the states change
shape when moving from atom to solid.
Other properties ma
Summary
The concept of a pseudopotential

A norm conserving pseudopotential

A non-local pseudopotential

A separable pseudopotential.

A nonlinear core correction.
Reading...
Kerker paper

BHS paper

Troullier Martin paper

HGH papers

Louie-Froyen-Cohen paper