Introduction to

Orbital-Free
Density-Functional Theory
Ralf Gehrke
FHI Berlin, February 8th 2005
 Outline
Basics of functional derivatives
I Principles of Orbital-free Density-Functional Theory
• basics of Density-Functional Theory
• motivation for OF-DFT
• basics of the numerical implementation
II Approximations to the kinetic-energy functional
• The Thomas-Fermi Approximation
• The von-Weizsäcker term
• Linear-Response Theory
• Combination of TF and vW
• The Conventional Gradient Expansion
• Average-Density Approximation
Basics of functional derivatives
analogous
function of multiple argument functional
F {y0 , L , y N } F [ y ( x )]

{yi } number y(x ) number

∂F δF
yi dF = dyi δF = δy (x0 )
∂yi y(x)
δy (x0 )

dyi δy (x0 )

xi x x0 x
∂F δF
dF = ∑ dyi δF = F [ y + δy ] − F [ y ] = ∫ dx δy (x )
i ∂yi δy (x )
 Basics of functional derivatives
some rules

δ F [y ] ∂ f
F [ y ( x )] = ∫ dx f ( y ( x )) =
δ y (x ) ∂ y
„Proof“:
δF [ y ] = F [ y + δy ] − F [ y ]
∂f
= ∫ dx f ( y + δy ) − ∫ dx f ( y ) = ∫ dx δy ( x )
1424 3 ∂y
f ( y )+ δy + O (δy )
∂f 2 {
∂y δF
δy ( x )

δ F [y ] ∂ f d ∂f
F [ y ( x )] = ∫ dx f ( y (x ), y′( x )) = −
δ y ( x ) ∂ y dx ∂ y ′

⎛ δF [ y ] ∂f ∂f ∂f d ∂f d ∂f d ∂f ⎞
⎜⎜ r = − ∇ = − − − ⎟⎟
⎝ δy (r ) ∂y ∂∇y ∂y dx1 ∂ (∂y / ∂x1 ) dx2 ∂ (∂y / ∂x2 ) dx3 ∂ (∂y / ∂x3 ) ⎠
Basics of functional derivatives
some rules
„Proof“:

F [ y + δy ] = ∫ dx f ( y + δy, y′ + δy′)
⎛ ∂f ∂f ⎞ ∂f ∂f
= ∫ dx⎜⎜ f ( y ) + δy + δy′ ⎟⎟ = F [ y ] + ∫ dx δy + ∫ dx δy′
⎝ ∂y ∂y′ ⎠ ∂y ∂y′
14243
( y(x ) ⎯⎯⎯→ 0)
x → ±∞ ∫
− dx
d ⎛ ∂f ⎞
⎜ ⎟δy
dx ⎜⎝ ∂y ′ ⎟⎠

⎛ ∂f d ∂f ⎞
δF = ∫ dx⎜⎜ − ⎟⎟δy ( x )
′
y dx ∂y ⎠
⎝1∂44244 3
δF [ y ]
δy ( x )
Basics of functional derivatives
some rules
„chain rule“

δF δF δG ([ y ], x′)
F [G ([ y ( x )], x )] = ∫ dx ′
δy ( x ) δG ([ y ], x′) δy ( x )

some examples
r r r
r r ρ ( r )ρ ( r′) r δJ [ρ ] r ρ ( r ′)
J [ρ ] = ∫∫ d 3 r d 3 r′ r r VH ([ρ ], r ) =
1
δρ ( r ) ∫
r = d 3
r ′ r r
2 r - r′ r - r′

δ 2 J [ρ ] 1
r r = r r
δρ (r )δρ ( r′) r - r′
 Basics of DFT
First Hohenberg-Kohn-Theorem

Energy of groundstate is a mere function of the electron density ρ

r r r
E [ρ ] = F [ρ ] + ∫ d 3 r Vext ( r )ρ ( r )

F [ρ ] = T [ρ ] + E ee [ρ ] = T S [ρ ] + J [ρ ] + E XC [ρ ]
r r
r r ρ (
J [ρ ] = ∫∫ d 3 r d 3 r′ r r
1 r )ρ ( r′)
classical coulomb interaction
2 r - r′

TS [ρ ] kinetic energy of a non-interacting

system with density ρ
EXC [ρ ] exchange-correlation energy
 Basics of DFT
Second Hohenberg-Kohn-Theorem
(Variational Principle)

δE[ρ]
δ
(
δρ (r )
( 3r r
))
r E [ρ ] − µ ∫ d r ρ ( r ) − N = 0 ⇒ r =µ
δρ(r )

„Euler-Lagrange-Equation“

µ chemical potential (negative of first ionization energy of system)

 Basics of DFT
• What is the kinetic energy TS [ρ ]?
→ introduction of a reference-system of non-interacting particles
q −1
r r
ϕ i ϕ j = δ ij ∑ ϕi (r ) = ρ (r ) 1
∑
2
ϕi ∆ ϕi
⇒ TS = −
i =0 2 i
• orbitals {ϕ i } are solutions of single-particle Schrödinger-equations
⎛ 1 KS ⎞
⎜ − ∆ + Veff ⎟ϕi = ε iϕi
⎝ 2 ⎠
(Kohn-Sham-equations)
• effective potential has to be determined self-consistently

r δJ [ρ ] δEXC [ρ ]
V KS
([ρ ], r ) = + + Vext = VH + VXC + Vext
δρ δρ
eff
 Motivation for OF-DFT
• Introduction of orbitals in DFT is not desirable
→ matrix diagonalization scales cubically with basis size
→ many data have to be stored during computation
• the goal is to get rid of KS-orbitals and to express each energy
contribution in terms of the electron charge density

⇒ a functional TS [ρ ] has to be found

• high accuracy for TS [ρ ] is required since TS [ρ ] is of the same order

as E [ρ ] (virial theorem) ↔ EXC [ρ ] is much smaller than E [ρ ]
⇒ many schemes for approximating EXC [ρ ] might be inappropriate
for TS [ρ ]
Basics of numerical implementation
Once an appropriate approximation for TS [ρ ] exists, how can the total
energy E [ρ ] of the groundstate be calculated?

two different approaches:

• direct minimization of the total energy

• self-consistent calculation
Basics of numerical implementation
direct minimization
• energy has to be minimized under the constrained of constant
particle numbers
( 3r r
Π[ρ ] = E [ρ ] − µ ∫ d r ρ (r ) − N )
• variable substitution to ensure positivity of electron density ρ
r r
ϕ (r ) = ρ (r )
• minimization of Π with steepest descend, conjugate-gradients, etc.

δΠ[ρ ] r r δΠ[ρ ]
r =0 ϕ n +1 (r ) = ϕ n (r ) − τ r
δϕ (r ) δϕ (r ) ϕ n ( rr )
Basics of numerical implementation
self-consistent calculation
• add and subtract kinetic energy of a bosonic system

r r r
E [ρ ] = TB [ρ ] + TS [ρ ] + J [ρ ] + EXC [ρ ] + ∫ d 3 r Vext ( r )ρ (r ) − TB [ρ ]

1 3r r r
TB [ρ ] = − ∫ d r ρ ( r )∆ ρ ( r)
2

• insert total energy expression in Euler-Lagrange-equation

r
δTB [ρ ] 1 ∆ϕ ( r )
integration by parts and chain rule ⇒ r = L = − r
δρ (r ) 2 ϕ (r )
Basics of numerical implementation
self-consistent calculation
• Euler-Lagrange-equation is transfered to a Kohn-Sham-like
equation with an additional potential term
r
δE[ρ ] 1 ∆ϕ ( r ) δTS δTB
r =µ ⇒ − r + VH + VXC + Vext + r − r =µ
δρ (r ) 2 ϕ (r ) δρ (r ) δρ (r )

Ĥϕ = µϕ
∆ δTS δTB
Hˆ = − + VH + VXC + Vext + r − r
2 δρ (r ) δρ ( r )

• same solution methods can be used as in the case of the

KS-scheme, but only with one orbital ⇒ no orthogonalization
 KS-scheme OF-scheme

solve poisson-equation solve poisson-equation

r r r r
∆VH ( r ) = −4πρ k ( r ) ∆VH (r ) = −4πρ k (r )

calculate exchange-correlation potential calculate potentials

r r δTˆ δTˆ
VˆXC([ρk ], r ) VˆXC([ρk ], r ) Gˆ ([ρ k ], r ) = Sr ([ρ k ], r ) − Br ([ρ k ], r )
r r r
δρ (r ) δρ (r )

create hamiltonian create pseudo-hamiltonian

∆ ∆
Hˆ = − + VˆH + VˆXC + Vˆext Hˆ = − + VˆH + VˆXC + Gˆ + Vˆext
2 2

solve solve
Hˆ ϕ i = ε iϕ i , i = 0, L , q − 1 Ĥϕ = µϕ

calculate new electron density calculate new electron density

q −1
ρ~k +1 = ∑ ϕ i 2 ρ k +1 = f (ρ~k +1 , ρ 0 , L , ρ k ) ρ~k +1 = ϕ 2 ρ k +1 = f (ρ~k +1 , ρ 0 , L , ρ k )
i =0

self-consistency no self-consistency no
ρ k +1 − ρ k < ε
achieved? ρ k +1 − ρ k < ε
achieved?
k → k +1 k → k +1
k : interation index
yes yes
q : number of KS-orbitals
 The Thomas-Fermi approximation
• consider a system, consisting of non-interacting, free electrons

( ) (ρ (rr ) = C )
1 r rr
Hˆ = − ∑ ∆ i ϕ k (r ) = C exp ik r k
2
= const.
2 i

• introducing periodic boundary conditions

2π
ki = ni , i = x, y , z
Li
r 8π 3
∆ k = ∆k x ∆k y ∆k z
3

(
k F = 3π 2 ρ )
1/ 3
 The Thomas-Fermi approximation
• kinetic energy can be calculated exactly

1 kF 3 r 2 r 2
kF kF kF
k2 V V V k 5
T = 2∑ = 3 r ∑ ∆ k k ≈ 3 ∫0 d 3
k k = ∫0 dk k 4
= F

k 2 ∆k k 8π 2π 2 2π 2 5

T = V ⋅ C ⋅ ρ 5/3 , C =
3
10
( )
3π 2
2/3

• kinetic energy density t = T / V may be used to approximate the

kinetic energy of a non-homogenous system with sufficiently slowly
varying electron density

r r
T = ∫ d 3 r t [ρ ( r )]
T
t= = Cρ 5 / 3
V
 The Thomas-Fermi approximation

• by construction, the TF approximation is correct in the limit of a

homogenous electron gas
• TF is correct in the limit of infinite nuclear charge (Z → ∞ )

flaws
• infinite charge density at the nucleus
• bad total energies compared to Kohn-Sham
• ( )
algebraical decay of charge density ∝ r −6 instead of
( (
exponential ∝ exp − 2 − 2 µ r ))
• no binding of atoms to form molecules or solids
• no shell structure in atoms
 The von-Weizsäcker term
• originally, von-Weizsäcker derived intuitively a correction to the
Thomas-Fermi approximation to describe the kinetic energy of core
particles to explain mass defects

T = TTF + TvW

r
TvW = ∫ d 3 r t (ρ , ∇ρ )

• the von-Weizsäcker term is exact for one-orbital systems

(e.g. bosonic systems, one or two electrons)
 The von-Weizsäcker term
derivation
• for a one-orbital system, the kinetic energy can be calculated
exactly
1 3r * r r
TvW = − ∫ d r ϕ ( r )∆ ϕ ( r)
2
integration by parts (ϕ (rr ) ⎯⎯⎯→ 0)
r →∞

1 3r r 2 1 3r r 2
= ∫ d r ∇ϕ ( r ) = ∫ d r ∇ ρ ( r )
2 2
r ∂f r r r
∇f ( g ( r )) = ∇g ( r ) ∇ ρ (r ) = ( r)
1
r ∇ ρ
∂g 2 ρ (r )
r 2
1 3r ∇ ρ ( r)
TvW = ∫d r r
8 ρ (r )
 The von-Weizsäcker term
improvements compared to TF
• exponential decay of electronic density
• finite charge at the nucleus
flaws
• in a homogenous system TvW = 0, but for a one-orbital system TTF ≠ 0
⇒ TS = TTF + TvW does not reproduce the two limits

attempts to improve the bridge between the two extremes:

TS = G ( N )TTF + TvW
number of electrons
⎛ empirical parameters
A2 ⎞
G ( N ) = (1 − δ1N − δ 2 N )⎜⎜1 − 1 + 2 ⎟⎟
A1
⎝ N 3 N 3⎠
• experiences indicated that the prefactor of the vW-Term might be
too large
 Linear-Response Theory
Why Linear-Response Theory?
• relationship between Linear Response and functional derivative of
TS can be derived
• correct Linear Response is important to describe charge
oscillations in solids

definition
• small change in potential causes a first-order change of the
charge density
r 3r r r r r r r
( )
δρ r = ∫ d r 1
′ (
χ42 r - r δν r′)
43
′ ) ( δρ (q ) = χ (q )δν (q )
r
δρ ( r )
r
δν ( r ′ ) fourier transformation
 Linear-Response Theory
The Linear-Response Function

„inversion“ theorem of functional derivatives:

r r r
r r r
δ ( r′′ - r′) = ∫ d 3 r
δρ ( r ′′ ) δν ( r ) r
(r r
) δν ( r)
δν (r ) δρ (r′) ∫
r r = d 3
r χ r ′′ - r r
δρ
12
(3
r ′)
r r
f ( r - r′ )

fourier transformation:
r r
r ˆ ⎛ δν ( r ) ⎞ ⎛ δν ( r) ⎞
( )
1 = χ q F⎜ ⎜ ⎟
r ⎟ ⇒ F⎜ˆ ⎜ ⎟
r ⎟= r
1
⎝ δρ ( r ′ ) ⎠ ⎝ δρ ( r′) ⎠ χ (q )
 Linear-Response Theory
Linear-Response in the DFT scheme
∆ δEeff
H = − + Veff
ˆ E = TS + Eeff Veff =
2 δρ (r )
δVeff δ 2 Eeff δ 2 (E − TS ) δ 2TS
= = =−
δρ (r′) δρ (r )δρ (r′) δρ (r )δρ (r′) δρ (r )δρ (r′)

δE
r = µ = const.
δρ (r )

1 ⎛ δ 2T ⎞
r = −F
ˆ ⎜ r
S
r ⎟
χ (q ) ⎜ δρ ( r )δρ ( r′)
⎝
⎟
groundstate ⎠
unknown !
 Linear-Response Theory
The Free-Electron-Gas limit of the LR-function

∆ kF ⎛ 1 1 −η 2 ⎛ 1 + η ⎞ ⎞ q
Hˆ 0 = − χ Lindhard = − 2 ⎜⎜ + ln⎜⎜ ⎟⎟ ⎟⎟, η =
2 π ⎝2 4η ⎝ 1 −η ⎠ ⎠ 2k F

⎛ ⎞
⎜ ⎟
⎜ δ TS 2
⎟ 1
Fˆ ⎜ = −
′) ρ ⎟
r r
δρ ( r
⎜ 144244 ) δρ ( r
0 ⎟
χ Lindhard (q )
⎜ 3⎟
f ( r - r′ )
r r
⎝ ⎠
homogenous system
due to symmetry...
 Linear-Response Theory

fourier transformation
weak logarithmic
singularity
cos(2k F r )
lim δρ (r ) ∝
r →∞ r3

„Friedel oscillations“

⇒ singularity is important to describe the physics of a solid properly

 Combination of TF and vW
• goal is to put the combination of TF and vW on a physical basis by
considering the linear-response
TS = TTF + λTvW

• consider a nearly homogeneous electron gas with small fluctuations

r r 3r r
( ) (
ρ r = ρ 0 + ∆ρ r ) ∫ d r ∆ρ (r ) = 0
• kinetic energy may be expanded around the average density

r δTS r 1 3r 3r δ 2TS r r
TS [ρ ] ≈ TS [ρ 0 ] + ∫ d r
3
r ∆ρ ( r ) + ∫∫ d r d r ′ r r ∆ρ ( r )∆ρ ( r ′)
123 δρ ( r ) ρ0 2 δρ (r )δρ (r ′) ρ 0
TTF 1424 3 144244 3
const. ⎛ 1 ⎞
− Fˆ ⎜⎜ ⎟⎟
⎝ χ Lindhard ( q ) ⎠
⇒ linear term vanishes
 Combination of TF and vW
⎧ η2 q • Kompaneets, Pavlovskii (1957)
⎪ 1 − L, small η =
k ⎪ 2k F • Kirzhnits (1957)
χ Lindhard ≈ − F2 ⎨ 3
π ⎪ 1 K • Le Couteur (1964)
⎪⎩ 3η 2 , large η • Stoddart, Beattie, March (1970)

⎧1 − 3λη 2 , small η ⇒ λ = 19
kF 1 kF ⎪
χ TFλvW =− 2 ≈− 2 ⎨ 1
π 1 + 3λη 2
π ⎪ , large η ⇒ λ =1
⎩ 3λη 2
 Combination of TF and vW
• λ = 1 / 9 valid for long wavelength perturbation
→ e.g. appropriate for impurity problems where long wavelength
components of potential are dominant
• λ = 1 valid for short wavelength perturbation
→ e.g. appropriate for perfect lattices
flaws
• bad estimation of total energy:
λ = 1 → overestimation, λ = 1 / 9 → underestimation

(interpolation leads to λ = 1 / 5 )

• still no shell structures in atoms

 Conventional Gradient Expansion
Basic Idea
• include higher order gradient corrections to the kinetic energy
functional by expanding the Lindhard function to higher orders

kF ⎛ η 2 η 4 ⎞
χ Lindhard ≈ − 2 ⎜⎜1 − − − L⎟⎟ T = T0 + T2 + T4 + L
π ⎝ 3 15 ⎠
1
TTF TvW
9

T4 [ρ ] =
1 3r (
⎛ 2
1 3⎜ ∇ ρ )
2
9∇ 2 ρ (∇ρ ) (∇ρ )
2 4 ⎞
⎟
540(3π 2 )
23 ∫d r ρ ⎜ ρ2 −
8ρ 3
+
3ρ 4 ⎟
⎝ ⎠
(Hodges, 1973)
T6 [ρ ] = L (extremely complicated...) (Murphy, 1981)
 Conventional Gradient Expansion
flaws
• not suitable for isolated systems
with exponential decay of charge
density
→ kinetic energy potential δT
diverges for order four and δρ
higher
→ T diverges for order six

⇒ in self-consistent calculations,
charge density shows wrong
decay behaviour
• linear response is wrong, since
expansion does only converge for
η <1
⇒ CGE is of no practical use
 Average-Density Approximation
Basic idea
• include non-local effects to the kinetic energy functional

example: (Garcia-Gonzales et al., PRA 54, 1897 (1996) )

r r r
TS [ρ ] = − TTF [ρ ] + TvW [ρ ] + ∫ d 3 r ρ ( r )t0 (ρ~ ( r ))
3 8
5 5
3r r r r 3r r rr
T [ρ ] = d r ρ ( r )t (ρ ( r ))
TF ∫ 0 ρ (r ) = d r ρ (r )Ω(r , r ′)
~
∫′ ′

• correct linear-response behaviour is enforced by appropriate chose

rr
of the weight function Ω( r , r ′)

⎛ δ 2T ⎞!
Fˆ ⎜ ⎟ 1
r
S
r = −
⎜ δρ ( r )δρ ( r ′) ⎟ χ Lindhard (q )
⎝ ρ0 ⎠
Average-Density Approximation
Kr

Kohn-Sham
rr r r
symmetric ADA Ω ( r , r ′ ) = Ω ( r ′, r )
rr r r
non-symmetric ADA Ω ( r , r ′ ) ≠ Ω ( r ′, r )
TFλvW
⇒ non-local functionals can improve numerical results significantly

(Garcia-Gonzales et al., PRA 54, 1897 (1996) )

 Summary
• goal is to get rid of the Kohn-Sham-orbitals to
reduce the computational time
• trade-off are less acurate results due to the
necessity of approximating the kinetic energy
functional
• state-of-the-art are non-local functionals which
improve the results of the classical functionals
significantly
 Literature
• Wang, Y.A., Carter, E.A. 2000, „Orbital-free kinetic energy density functional theory“, Theoretical
Methods in Condensed Phase Chemistry, S.D. Schwartz, Ed. Kluwer, 117-184
• P. Garcia-Gonzales, J.E. Alvarellos and E. Chacon, „Kinetic-energy density functional: Atoms and
shell structure“, Phys. Rev. A, 54, 1897 (1996)
• M. Levy, J.P. Perdew and V. Sahni, „Exact differential equation for the density and ionization
energy of a many-particle system“, Phys. Rev. A, 30, 2745 (1984)
• C.H. Hodges, „Quantum Corrections to the Thomas-Fermi Approximation-The Kirzhnits Methods“,
Can. J. Phys., 54, 1428 (1973)
• R.G. Parr, S. Liu, A.A. Kugler and A. Nagy, „Some identities in density-functional theory“,
Phys. Rev. A, 52, 969 (1995)
• Y. Wang and R.G. Parr, „Construction of exact Kohn-Sham orbitals from a given electron density“,
Phys. Rev. A, 47, R1591 (1993)
• T. Gal and A. Nagy, „A method to get an analytical expression for the non-interacting kinetic
energy density functional“, J. Mol. Struc. 501-501, 167 (2000)