Implementing DFT

in
Plane­Wave basis

Fabien Bruneval

Service de Recherches de Métallurgie Physique

CEA Saclay
France

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

DFT for periodic systems

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

Outline

A DFT code adapted to periodic systems:

● Self­consistency in KS equations
● Crystal structure
● k­points
● Plane­Waves
● Supercells

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

Solving KS equations
depends on the density
non linear equations

2
∇
h r =− v ion r v H [] r v xc []r 
2

h r  i r =i  i r 

 r =∑ ∣ i r ∣
2

i occ

Energy, Forces, Band structure, Electronic density

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

Solving KS equations
depends on the density
non linear equations

2
−∇
h r = v ion  r v H [ ]r v xc [] r 
2

h  i=i  i

 r =∑ ∣ i r ∣
2

i occ

Energy, Forces, Band structure, Electronic density

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

Solving KS equations
 0 depends on the density
First guess for  r  non linear equations

2
∇
h r =− v ion r v H [] r v xc []r 
2

h  i=i  i Diagonalization

 n n−1
if  ≠  r =∑ ∣i r ∣
2

i occ

Energy, Forces, Band structure, Electronic density

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

Outline

A DFT code adapted to periodic systems:

● Self­consistency in KS equations
● Crystal structure
● k­points
● Plane­Waves
● Supercells

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

Crystal structure
Crystal axis: a1, a2, a3

Direct lattice vector: R=n1 a 1n 2 a2 n3 a 3

Periodic potential: V  r R=V r 

2
Reciprocal lattice axis: b 1= a 2×a3

2
b 2= a 3×a 1

2
b 3= a 1×a2

ReciprocalG=n
lattice: 1 b 1n 2 b2n3 b 3
iG. R
e =1
F. Bruneval Coimbra, 14/04/08
Implementing DFT in Plane­Wave Basis

Bloch theorem and k­points

Bloch theorem:

i k .r
 k i  r =e u k i r 
where k is in the first Brillouin zone
uki(r) is a periodic function with crystal periodicity

Any periodic operator and, in particular, the Hamiltonian, is diagonal in k.

〈 k i∣h∣k ' j〉=∫ d r e i  k ' −k  . r ∗

h r  u r  u k ' j  r 
ki
V
=∑ ∫ d r e i k ' − k  . r R

R 
∗
×h r R u r Ru k ' j r R ki
F. Bruneval Coimbra, 14/04/08
Implementing DFT in Plane­Wave Basis

k­points allow to split the calculations

〈 k i∣h∣k ' j〉= k k ' 〈 k i∣h∣k j〉

The Hamiltonian has blocks of non interacting k­points:

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

Solving KS equations with k­points

 0 depends on the density
First guess for  r  non linear equations

2
∇
h r =− v ion r v H [] r v xc []r 
2
Diagonalizations

h k  k i=k i k
1 1 1 1
i h k k i =k i  k
2 2 2 2
i h k  k i =k i  k
3 3 3 3
i

n  n−1 
if  ≠ 1 2
 r = ∑
N k k ,i occ
∣ k i  r ∣

Energy, Forces, Band structure, Electronic density

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

Brillouin Zone integration

Many quantities of the scheme require averaging in the BZ,
e.g. kinetic term, electronic density:

1
 r = ∑
N k k ∈BZ
∑∣k i r ∣ 2

i occ

1 2
to be exact, it should be  r = ∫
V BZ V
dk ∑
i occ
∣k i  r ∣
BZ

We have to find a set of points in the BZ, which makes the limit as fast as possible:

1 1
∑
N k k ∈BZ

V BZ
∫dk
V BZ

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

Brillouin Zone integration

1 1
∑
Nk k
 ∫
V BZ V
dk
BZ
Monkhorst­Pack technique, Phys. Rev. B 13, 5188 (1976)

2x2 with shift 0.5 0.5 2x2 with shift 0. 0.

F. Bruneval Coimbra, 14/04/08
Implementing DFT in Plane­Wave Basis

Equivalence k­points/larger cells

1 unit cell 2 unit cells 4 unit cells

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

Outline

A DFT code adapted to periodic systems:

● Self­consistency in KS equations
● Crystal structure
● k­points
● Plane­Waves
● Supercells

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

Bloch theorem again

i k .r
k i  r =e u k i r 
where k is in the first Brillouin zone
uki(r) is a periodic function with crystal periodicity

uki(r) is periodic and can be expanded in a Fourier series

1 iG. r
u k i  r = ∑
 G
c k i G e

where G is on the reciprocal lattice: G=n1 b 1n 2 b2n3 b 3

1 i k G  .r
 k i  r = ∑
 G
c k i G e

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

Potentials in plane­waves
Z
Example of the ionic potential:
V ion r =
∣r∣
 Z
V ion G=4  2
∣G∣
R­space: V(r) G­space: V(G)

|G|

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

Plane Waves
Reciprocal lattice: Volume of the sphere containing all PW:
G=n1 b 1n 2 b2n3 b 3 4 3
V sphere= G max
3
Volume of occupied by 1 single PW:
3
2 
V PW =


2
G max
E cutoff =
2
3 /2
N G ∝ E cutoff

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

Cutoff for the density

1
 r = ∑
N k k i occ
∗
 k i r  k i r 

1
= ∑ ∑
N k  k i occ G G G ' G
∑ ∗
c k i G c k i G ' ei G −G '  . r

max max
i G 0. r
= ∑ G
 0 e
G0 2G max

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

PW: an orthogonal basis set

The wavefunctions are a linear combination of orthogonal basis functions:

∣k i 〉 = ∑ c k i G ∣kG 〉
∣G∣Gmax
and
1
〈 kG k G ' 〉= ∫ d r e
∣ i G ' −G  .r
=G G '

Variational principle:

E ground state E  Ecutoff = x Ha

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

PW: an orthogonal basis set

Adding more PW or increasing the cutoff energy Gaussian basis sets
makes ALWAYS the result more accurate of quantum­chemisrty:
STO­3G
STO­6G
3­21G
Silicon 6­31G
6­31+G*
6­311+G*
6­311++G**
cc­pVDZ
cc­pVTZ
cc­pVQZ
aug­cc­pVDZ
aug­cc­pVTZ
EPW(Ecutoff)
Eexact aug­cc­pVQZ
Dünning SVP
Dünning DVP
Dünning TVP
Dünning TVPP
F. Bruneval Coimbra, 14/04/08
Implementing DFT in Plane­Wave Basis

PW makes life easier thanks to the FFTs

f r =∑ f G e
iG. r

f G= 1 ∫ d r e−i G . r f r 

1 −i G . r
= ∑
N r r ∈
e f r i 
i
Discrete Fourier Transform
i

It is exact as long as NG = Nr

This means that

f G=DFT −1 [ DFT [ f G ] ]

The fast version of DFT is the famous Fast FT

with scales as O( N log N ) instead of N2.

This enforces the use of regular grid in real space.

F. Bruneval Coimbra, 14/04/08
Implementing DFT in Plane­Wave Basis

Illustration of the use FFTs

r ' 
We need to calculate the Hartree potential: v H r =∫ d r '
∣r −r '∣
FFT
c k i G uki  r 
1
 r = ∑
N k ki
∣u k i  r ∣2
FFT
 G  r 
 G 
vH G=4 
∣G∣2
FFT
vH  G vHr

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

Supercell technique
How to simulate a finite system with periodic boundary conditions?

The cost of the calculation can be problematic,

since
3 /2
N G ∝ E cutoff

Still useful for slabs, wires, etc.

F. Bruneval Coimbra, 14/04/08

Implementing DFT in Plane­Wave Basis

Recap

● Self­consistent loop
● k­points:
– Monkhorst­Pack grid (like 4x4x4 shift 0.5 0.5 0.5)
– Equivalence between k­points and larger cells
● Plane­Waves
– PW are an orthogonal basis set
– Cutoff energy for wavefunctions
– Intensive use of FFT's to increase efficiency

F. Bruneval Coimbra, 14/04/08