Basics of DFT and TDDFT

Hardy Gross
E.K.U. Gross
Max-Planck Institute of Microstructure
Physics, Halle
Max-Planck (Saale),ofGermany
Institute
and
Microstructure Physics
Fritz Haber Center
Halle (Saale)
The Hebrew University of Jerusalem
The Hamiltonian of Condensed Matter
(atoms, molecules, clusters, solids)
Hamiltonian for the complete system of Ne electrons with coordinates
r  r r
1 Ne
and Nn nuclei
R with 
 R coordinates
R 1 Nn 
Ĥ T̂n (R )  Ŵnn (R )  T̂e (r )  Ŵee (r )  V̂en (R , r )
Nn
2 Ne
i2 1 N n Z Z 
with T̂n   T̂e   Ŵnn  
 1 2M  i 1 2m 2  , R   R 
 

1 Ne 1 Ne Nn
Z
Ŵee   V̂en   
2 j,k rj  rk j1  1 rj  R 
jk
Hamiltonian for the complete system of Ne electrons with coordinates
r  r r
1 Ne
and Nn nuclei
R with 
 R coordinates
R 1 Nn 
Ĥ T̂n (R )  Ŵnn (R )  T̂e (r )  Ŵee (r )  V̂en (R , r )
Nn
2 Ne
i2 1 N n Z Z 
with T̂n   T̂e   Ŵnn  
 1 2M  i 1 2m 2  , R   R 
 

1 Ne 1 Ne Nn
Z
Ŵee   V̂en   
2 j,k rj  rk j1  1 rj  R 
jk

Stationary Schrödinger equation

Ĥ r , R  E r , R 
Hamiltonian for the complete system of Ne electrons with coordinates
r  r r
1 Ne
and Nn nuclei
R with 
 R coordinates
R 1 Nn 
Ĥ T̂n (R )  Ŵnn (R )  T̂e (r )  Ŵee (r )  V̂en (R , r )
Nn
2 Ne
i2 1 N n Z Z 
with T̂n   T̂e   Ŵnn  
 1 2M  i 1 2m 2  , R   R 
 

1 Ne 1 Ne Nn
Z
Ŵee   V̂en   
2 j,k rj  rk j1  1 rj  R 
jk

Time-dependent Schrödinger equation


i  r, R , t  H r, R  Vlaser r, R , t   r, R , t 
t
 Ne Nn

Vlaser r, R , t    rj   Z R   E f t cos t
 j 1  1 
 Born-Oppenheimer approximation

Neglect nuclear kinetic energy in full Hamiltonian


Tˆ e (r)  W
ˆ (r)  W
ee
ˆ (R)  V
nn en 
ˆ (r, R) Φ BO  r  BO
R ,J J   R ,J  r 
R Φ BO

To be solved for each fixed nuclear configuration R.

J R 
BO
J=1

J=0

R
Adiabatic approximation for the complete molecular wave function:

Ψ BO
r , R   r χ R 
BO
R
BO

find best χBO by minimizing <ΨBO | H | ΨBO > w.r.t. χBO

Nuclear equation
ˆ 1 BO
 Tn (R)  Wnn (R)  
ˆ A υ R  (-i )  B O R 
  M

 BO
 Φ BO
R
*
r  T̂n R Φ BO
R r d r  χ R  Eχ BO
R 

Berry connection

A BO
υ R   R
Φ BO *
r  (-i υ ) Φ R r d r
BO

 BO 
γ BO
C   A R dR is a geometric phase
C

corresponding to interesting topological features of molecules

expansion of 
BO
 
R to second order around equilibrium
positions yields phonon spectrum
Lectures I+II+III: Focus on the electronic-structure problem, for static
and time-dependent systems, but with clamped nuclei
(i.e. Born-Oppenheimer approximation):

H (r, R) Tˆ e (r)  W

ˆ BO ˆ (r)  W
ee
ˆ (R)  V
nn
ˆ (r, R)
en

with fixed nuclear coordinates R .

Lectures I+II+III: Focus on the electronic-structure problem, for static
and time-dependent systems, but with clamped nuclei
(i.e. Born-Oppenheimer approximation):

H (r, R) Tˆ e (r)  W

ˆ BO ˆ (r)  W
ee
ˆ (R)  V
nn
ˆ (r, R)
en

with fixed nuclear coordinates R .

This is still an exponentially hard problem!

Lecture I: Ground-state DFT, basic concepts

Lecture II: Approximate xc functionals
Lecture III: Time-dependent DFT, linear response and beyond

Lecture IV: Electron-nuclear coupling beyond Born-Oppenheimer

 Why don’t we just solve the many-body
Schrödinger equation

Example: Oxygen atom (8 electrons)

 
 r1 ,, r8  depends on 24 coordinates

rough table of the wavefunction

10 entries per coordinate:  1024 entries

1 byte per entry:  1024 bytes
5109 bytes per DVD:  21014 DVDs
10 g per DVD:  21015 g of DVDs
= 2109 t of DVDs
Two fundamentally different classes of ab-initio approaches:

• Wave function approaches

-- Configuration interaction
(also stochastic CI)
-- Tensor networks

• “Functional Theories”
Two fundamentally different classes of ab-initio approaches:

• Wave function approaches

-- Configuration interaction
(also stochastic CI)
-- Tensor networks

• “Functional Theories”

Write total energy as functional

of a simpler quantity and minimize
 Motivation
“Functional Theories”

MBPT RDMFT DFT


G( r, r ' , t  t' ) (r, r ') G(r, r ', 0 ) ( r )  ( r, r )
 Motivation
“Functional Theories”

MBPT RDMFT DFT


G( r, r ' , t  t' ) (r, r ') G(r, r ', 0 ) ( r )  ( r, r )
Functional: Functional: Functional:
Φxc[G] Exc[γ] Exc[ρ]
or Σxc[G] or vxc[ρ]
 Motivation
“Functional Theories”

MBPT RDMFT DFT


G( r, r ' , t  t' ) (r, r ') G(r, r ', 0 ) ( r )  ( r, r )
Functional: Functional: Functional:
Φxc[G] Exc[γ] Exc[ρ]
or Σxc[G] or vxc[ρ]
easy (e.g. GW) difficult very difficult
 Motivation
“Functional Theories”

MBPT RDMFT DFT


G( r, r ' , t  t' ) (r, r ') G(r, r ', 0 ) ( r )  ( r, r )
Functional: Functional: Functional:
Φxc[G] Exc[γ] Exc[ρ]
or Σxc[G] or vxc[ρ]
easy (e.g. GW) difficult very difficult
numerically
heavy moderate light
ESSENCE OF DENSITY-FUNTIONAL THEORY

• • Every observable quantity of a

Every observable quantity of a
quantum
quantum system
system can
can be
be calculated
calculated
from
from the
the density
density ofof the
the system
system
ALONE
ALONE
• • The density of particles interacting
The density of particles interacting
with
witheach
eachother
othercan
canbebecalculated
calculatedasas
the
thedensity
densityofofan
anauxiliary
auxiliarysystem
systemofof
non-interacting
non-interactingparticles
particles
ESSENCE OF DENSITY-FUNTIONAL THEORY

• • Every observable quantity of a

Every observable quantity of a
quantum
quantum system
system can
can be
be calculated
calculated
from
from the
the density
density ofof the
the system
system
ALONE
ALONE
• • The density of particles interacting
The density of particles interacting
with
witheach
eachother
othercan
canbebecalculated
calculatedasas
the
thedensity
densityofofan
anauxiliary
auxiliarysystem
systemofof
non-interacting
non-interactingparticles
particles

Hohenberg-Kohn theorem (1964)

Kohn-Sham theorem (1965)
(for the ground state)
 DFT papers per year

DFT:
Jul A Theory
5th, 2017 Full of Holes, Aurora Pribram-Jones, David A. Gross,
KITSKieron Burke, Annual Review of Physical Chemistry (2014). 20
 HOHENBERG-KOHN THEOREM
1—1
1. v(r) (r)
one-to-one correspondence between external potentials v(r) and ground-state
densities (r). In other words: v(r) is a functional of ρ(r)

2. Variational principle
Given a particular system characterized by the external potential v0(r). There
exists a functional, EHK [ρ], such that the solution of the Euler-Lagrange equation

E HK  0
r 
yields the exact ground-state energy E0 and ground-state density 0(r) of this
system

3. EHK[] = F [] + (r) v0(r) d3r

F[] is UNIVERSAL.
compare ground-state densities (r) resulting from
different external potentials v(r).

(r)

v(r)

QUESTION: Are the ground-state densities coming from

different potentials always different?
 G
A Ã
 (r)
v(r)  (r1…rN)

single-particle ground-state
ground-state
potentials having densities
wavefunctions
nondegenerate
ground state

Hohenberg-Kohn-Theorem (1964)

G: v(r) 
G: v(r)   (r)
(r) is
is invertible
invertible
Proof

Step 1: Invertibility of map A

Solve many-body Schrödinger equation for the external potential:

V̂ 
E  T̂  Ŵ ee

N
T̂  
 v r j     Wee r1 ... rN   constant
j1 

This is manifestly the inverse map: A given  uniquely yields the

external potential.
Step 2: Invertibility of map Ã

Given: two (nondegenerate) ground states , ’ satisfying

Ĥ E Ĥ T̂  Ŵ  V̂

with
Ĥ'  ' E '  ' Ĥ' T̂  Ŵ  V̂'

to be shown:   '   '

 

  = ’
’ 

cannot happen
Use Rayleigh-Ritz principle:

 E   Ĥ    ' Ĥ  '   ' H 'V  V'  '

E' d r' r v r   v ' r 
3

 E'   ' Ĥ '  '   Ĥ ' 

E  d 3rr v ' r   v r 

Reductio ad absurdum:
Assumption  = ’. Add  and   E + E’ < E + E’
 Consequence

Every quantum mechanical observable is completely

determined by the ground state density.

Proof:     v       i 
G 1
  solveS.E .

Hermitian operator B̂ Bi    i  B̂  i 

e.g. excitation spectrum: Ei[ρ]

What is a FUNCTIONAL?

E[]
(r) R
functional

set of functions set of real numbers

Generalization:
 
v r   v    r  functional depending parametrically on r

  
r1 ...rN         r1 ...rN 
  or on r1...rN 
QUESTION:

How to calculate ground state density o r  of a given
 system
(characterized by the external potential Vo  vo r  )
without recourse to the Schrödinger Equation?

Theorem:

There exists a density functional EHK[] with properties

i) EHK[] > Eo for   o
ii) EHK[o] = Eo
where Eo = exact ground state energy of the system

E HK  0
Thus, Euler equation r
yields exact ground state density o.
proof:

formal construction of EHK[] :

 A~  1
for arbitrary ground state density r     

define: E HK    T̂  Ŵ  V̂o 

> Eo for   o
= Eo for  = o q.e.d.

EHK[] = d3r (r) vo(r) +  T̂  Ŵ 

F[] is universal
An Example: Explicit algorithm to construct the HK map
vs ρ for non-interacting particles

(– h2 +2 vs(r) ) i = i i  i* ·

2m
i

N N
h2  2

i=1
i*(– 2m )i + vs(r)(r) =i 
=1
ii(r)2

1 N 2 ) )
ii(r)2– i* (- h2m
2
 vs(r) = r(r) ·
i=1
( i

Iterative procedure
r0(r) given (e.g. from experiment)
Start with an initial guess for vs(r) (e.g. LDA potential)
h2  2
solve (– 2m + vs(r) ) i = i i

1 N
h2  2
vs (r) = r ( r) ·
new
( ii(r)2– i* (- 2m ) i)
0 i=1

solve SE with vsnew and iterate, keeping ρ0(r) fixed

Consequence: The orbitals are functionals of the density: i[ρ]
 KOHN-SHAM EQUATIONS

Rewrite HK functional: EHK[] = TS[] + (r) v0(r) d3r + EH[] + Exc[]

where S T     j
d 3
r 
    r   2
/ 2   j    r 
is the kinetic energy functional of non-interacting particles


E HK  0 yields the Kohn-Sham equations:
r 

 / 2  v0 r   vH [] r   v xc [] r   j r  j  j r 

 2

The orbitals from these equations yield the true density of the interacting system
Walter Kohn: “The KS equations are an
exactification of the Hartree mean-field equation”
Exc[ρ] is a universal functional of the density which, in practice, needs
to be approximated.
The functional Exc[] is universal:

Curse or blessing?
The functional Exc[] is universal:

Curse or blessing?

Only ONE functional needs to

be approximated
 The functional Exc[] is universal:

Curse or blessing?

Only ONE functional needs to

be approximated

Functional can be systematically improved, i.e. results will

improve -on average- for all systems. Systematic improvement
for a single given system is not possible.
The five levels of (TD)DFT

Level 1: Basic Theorems

Level 2: Deduce exact features of the xc functional

Level 3: Find approximate functionals for v xc   r ' t '  rt 

Level 4: Write code that solves the (TD)KS equations

Level 5: Run code for interesting systems/questions