Brief History of DFT

• First speculated 1926 [Before Schrödinger]
– Thomas‐Fermi (kinetic energy) and Dirac (exchange 
energy, 1931) formulas
– Slater Exchange Approximation 1951 [First LDA to Vx(ρ)
• Officially born in 1964 with Hohenberg‐ Kohn’s 
original proof+ Kohn‐Sham Model 1965
• GEA/GGA formulas available later 1980’
• Becoming popular later 1990’
• Pinnacled in 1998 with a Chemistry Nobel prize
Let us recall The ORIGINAL HF   Orbital Equations

h1 χ i ( x1 ) +
⎡N ⎛ 1 ⎞ ⎤
⎢ ∑ ∫ χ j ( x2 ) ⎜ ⎟ χ j ( x2 )dx2 ⎥ χ i ( x1 )
*

⎣ j =1 ⎝ r12 ⎠ ⎦
⎡N ⎛ 1 ⎞ ⎤
− ⎢ ∑ ∫ χ j ( x2 ) ⎜ ⎟ χ i ( x2 )dx2 ⎥ χ j ( x1 )
*

⎣ j =1 ⎝ r12 ⎠ ⎦
= ε i χ i ( x1 )
Let   us   consider  Hartree‐Fock  Orbital Equations : The Coulomb  Potential

⎡ * 1 ⎤
J j ( x1 ) χ i ( x1 ) = ⎢ ∫ χ j ( x2 )( ) χ j ( x2 )dx2 ⎥ χ i ( x1 )
⎣ r12 ⎦
In HF orbital equations, we have:
N
⎡ * 1 ⎤
−∑ ⎢ ∫ χ j ( x2 )( ) χ j ( x2 )dx2 ⎥ χ i ( x1 );
j =1 ⎣ r12 ⎦
Coulomb potential is a LOCAL Potential
because at any given point "x1 " it is the same
for all Fock orbitals χ i ( x1 ). It's source charge
arises from a Coulomb Charge located at "x2 ".
Let   us   consider  Hartree‐Fock  Orbital Equations : The non‐local Exchange  Potential

⎡ 1 ⎤
K j ( x1 ) χ i ( x1 ) = ⎢ ∫ χ j ( x 2 ) ( ) χ i ( x 2 ) dx 2 ⎥ χ j ( x1 )
*

⎣ r12 ⎦
In HF orbital equations, we have:
N
⎡ 1 ⎤
− ∑ ⎢ ∫ χ j ( x 2 ) ( ) χ i ( x 2 ) dx 2 ⎥ χ j ( x1 ) =
*

j =1 ⎣ r1 2 ⎦
⎡ 1 ⎤
χ χ 2{ iχ χ 1 }⎥
* *
N ⎢ j ( x 2 ) ( ) i ( x 2 ) dx ( x 1 ) i ( x )
r12
= −∑ ⎢∫ ⎥ χ j ( x1 )
j =1 ⎢ {χ i ( x1 ) χ i ( x1 )}
*
⎥
⎢⎣ ⎥⎦
⎡ 1 ⎤
N ⎢∫
χ *
i ( x1 ) χ *
j ( x 2 ) ( ) χ j ( x1 ) χ i ( x 2 ) dx 2 ⎥
r12
= −∑ ⎢ ⎥ χ i ( x1 )
j =1 ⎢ χ i ( x1 ) χ i ( x1 )
*
⎥
⎢⎣ ⎥⎦ i
The non‐local HF Exchange  Potential corrects The Coulomb Potential such that:  (1) 
Self Interaction is removed &  (2) The reference elec. “1” in χ i moves in the Mean
Electrostatic Field due to “N-1” other electrons.
⎡ 1 ⎤
K j ( x1 ) χ i ( x1 ) = ⎢ ∫ χ j ( x 2 ) (
*
) χ i ( x 2 ) d x 2 ⎥ χ j ( x1 )
⎣ r1 2 ⎦
⎡ 1 ⎤
N ⎢ ∫
χ *
i ( x 1 ) χ *
j ( x 2 ) ( ) χ j ( x 1 ) χ i ( x 2 ) d x 2 ⎥
r1 2
= -∑ ⎢ ⎥ χ i ( x1 )
j =1 ⎢ χ i ( x1 ) χ i ( x1 )
*
⎥
⎢⎣ ⎥⎦
L IK E S P IN S

∴ T h e H F E x . P o t. is N O N L O C A L . A t a n y g iv e n
lo c a tio n " x 1 ", th is p o te n tia l d u e to o th e r e le c tro n s
a t " x 2 " is d iffe re n t fo r d iffe re n t H F o rb ita ls
d u e to " χ i ( x 2 ) " in it. It's s o u rc e c h a rg e is th e
E x c h a n g e C h a rg e lo c a te d a t " x 2 " . T h is c h a rg e
c o rre c ts th e C o u lo m b C h a rg e w h ic h d is re g a rd s
F e rm i C o rre la tio n [P a u li E x c lu s io n fo r lik e -s p in s ]
EXCHANGE CHARGE DENSITY CORRECTS
(REMOVES/SUBTRACTS) THE COULOMB
CHARGE DENSITY SUCH THAT THE LIKE SPIN
ELECTRONS ARE AVOIDED BEING AT THE
SAME LOCATION IN SPACE.

THE INTEGRAL OF EXCHANGE CHARGE

DENSITY IS = 1 Electronic Charge.
HF    “Exchange   Charge Density” Its   PROPERTIES 
N χ ( x1 ) χ ( x2 ) χ j ( x1 ) χ i ( x2 )dx2
* *

−e ∑
i j

j =1 χ i ( x1 ) χ i ( x1 )
*

When x1 = x2 the exchange charge density becomes:

N χ ( x1 ) χ ( x1 ) χ j ( x1 ) χ i ( x1 )dx2
* *

e∑ = ρ + ( x1 )
i j

j =1 χ i ( x1 ) χ i ( x1 )
*

The radius of exchange hole, r0 ,

4π 3 1
→( ρ(r))
1/3
r0 ρ(r) =1; ∴
3 r0

At x1 ρ(x1)
HF    “Exchange   Charge Density” & Its   Orbital Average which makes
Exchange Potential  from    nonlocal → local  

N χ ( x1 ) χ ( x2 ) χ j ( x1 ) χ i ( x2 )dx2
* *

−e∑ i j

j =1 χ i ( x1 ) χ i ( x1 )
*

Weightage,Probability that el at x1 is in χ i =
χ i* ( x1 ) χ i ( x1 )
Now sum over all “i “  orbitals.
⎡ N
⎤
⎢ ∑ χ k ( x1 ) χ k ( x1 ) ⎥
*

⎣ k =1 ⎦ Use a dummy index “i=k”  →
N N

∑
k =1
∑ k 1 j ( x2 ) χ j ( x1 ) χ k ( x2 )dx2
χ *

j =1
( x ) χ *

−e N

∑ k ( x1 ) χ k ( x1 )
χ *

k =1
The  Slater Exchange Approximation to HF   Orbital Equations
h1 χ i ( x1 ) +
⎡ N ⎛ 1 ⎞ ⎤
⎢ ∑ ∫ χ j ( x2 ) ⎜ ⎟ χ j ( x2 )dx2
*
⎥
⎢ j =1 ⎝ 12 ⎠
r ⎥
⎢ N
⎥
⎢ ∑ χ k ( x1 ) χ k ( x1 )
*
⎥
⎢ k =1
⎥ χ i ( x1 ) = ε i χ i ( x1 )
⎢ N N ⎛ 1 ⎞ ⎥
⎢ ∑ ∑ ∫ χ k ( x1 ) χ j ( x2 ) ⎜ ⎟ χ j ( x1 ) χ k ( x2 )dx2 ⎥
* *

⎢ − k =1 j =1 ⎝ r12 ⎠ ⎥
⎢ N ⎥
⎢
⎢⎣
∑
k =1
χ k ( x1 ) χ k ( x1 )
*
⎥
⎥⎦
Slater (1951) expressed the second term inside square brackets,
the common average exchange density for all orbitals, by substituting
the density, only in this term, as that of the homogeneous electron gas.
3 1/ 3 2 / 3 −1/ 3 Slater
E Slater
X [ ρ ] = CX ∫ ρ 4/3
( r ) dr, C X = 3 4 π ; VX (r ) α ρ 1/ 3 ( r )
8
 Local Exchange Approximation to HF   Orbital Equations
J. C.  Slater           Physical Review  81, 385‐390  (1951)
h1 χ i ( x1 ) +
⎡N ⎛ 1 ⎞ ⎤
⎢ ∑ ∫ χ j ( x2 ) ⎜ ⎟ χ j ( x2 )dx2 ⎥
*

⎢ k =1 ⎝ r12 ⎠ ⎥
⎢ 1 ⎥ χ i ( x1 ) = ε i χ i ( x1 )
⎢ 3 ⎧⎛ 3 ⎞ N * ⎫3 ⎥
⎢ − 4 ⎨⎜ 8π ⎟ ∑ χ k ( x1 ) χ k ( x1 ) ⎬ ⎥
⎣ ⎩⎝ ⎠ k =1 ⎭ ⎦
In the spherical volume at the location of electron "1"
4π 3
r0 ρ (r ) = 1, where r0 is the radius of the exchange hole.
3
1
∴ The inverse of the radius of exchange hole , → ( ρ (r ) )
1/ 3

r0
Contrast  This  With ORIGINAL HF   Orbital Equations

h1 χ i ( x1 ) +
⎡N ⎛ 1 ⎞ ⎤
⎢ ∑ ∫ χ j ( x2 ) ⎜ ⎟ χ j ( x2 )dx2 ⎥ χ i ( x1 )
*

⎣ j =1 ⎝ r12 ⎠ ⎦
⎡N ⎛ 1 ⎞ ⎤
− ⎢ ∑ ∫ χ j ( x2 ) ⎜ ⎟ χ i ( x2 )dx2 ⎥ χ j ( x1 )
*

⎣ j =1 ⎝ r12 ⎠ ⎦
= ε i χ i ( x1 ) While approximating the third term on LHS
the self interaction part in the Coulomb potential
(2 Term) is not fully cancelled by the approximate
Slater exchange potential. This is the main drawback
of LDA; it is not self-interaction free. Due to this,
the band gap is underestimated by LDA.
 LDA   Functionals
Exc [ ρ ] = Ex [ ρ ] + Ec [ ρ ]
• Slater form (exchange) – 1 parameter;
• Kohn‐Sham  Exchange ‐ Variational α = 2/3

3 1/ 3 2 / 3 −1/ 3
E S
X [ ρ ] = α CX ∫ ρ 4/3
( r ) dr , C X = 3 4 π
8
• Wigner correlation – 2 parameters
• Parameters Fixed by reproducing  Post HF  He 
Result on E

ρ (r )
E [ρ ] = a ∫
W
dr
1 + bρ (r )
−1 / 3
C
 All about Exchange-Correlation
Energy Density Functional

(
QXC [ρ ] = ∫ f r, ρ (r ), ∇ 2 ρ (r ), " dr )
ƒ LDA – f(r) is a function of
ρ(r) only
ƒ GGA – f(r) is a function of
ρ(r) and |∇ρ(r)|
ƒ Meta-GGA – f(r) is also a
function of ts(r), kinetic
energy density
ƒ Hybrid – f(r) is GGA
functional with extra Jacob's ladder for the five generation of DFT functionals,

contribution from Hartree- according to the vision of John Perdew with indication of
some of the most common DFT functionals within each rung.
Fock exchange energy

its.unc.edu 14
 Popular Functional: BLYP/B3LYP
Two most well-known functionals are the Becke exchange functional
Ex[ρ] with 2 extra parameters β & γ
2
χ2
∇ρ
E XB [ρ ] = E XLDA − β ∫ ρ 4 / 3 , χ =
1 + γχ 2 ρ 4/3
The Lee-Yang-Parr correlation functional Ec[ρ] with 4 parameters a-d
1 ⎧ −2 / 3 ⎡ ⎛1 1 2 ⎞⎤ −cρ −1 / 3 ⎫
E LYP
[ρ ] = −a ∫ ⎨ ρ + bρ ⎢C F ρ − 2tW + ⎜ tW + ∇ ρ ⎟⎥ e
5/3
⎬dr
1 + dρ −1/ 3 ⎩
c
⎣ ⎝ 9 18 ⎠⎦ ⎭
Together, they constitute the BLYP functional:

ExcBLYP = ExB + EcLYP = ∫ ρ (r )exB (ρ , ∇ρ )dr + ∫ ρ (r )ecLYP (ρ , ∇ρ )dr

The B3LYP functional is augmented with 20% of Hartree-Fock
exchange:
N N
E xcB3LYP = a E xB + E cLYP + b ∑P ∑P
m , n =1
mn
k , l =1
kl km nl

its.unc.edu 16
 Density Functionals

VWN5
LDA

Increasing quality and computational cost

local density Better scaling with system
BLYP size
HCTH
BP86 GGA Allow density fitting for even
gradient corrected better scaling

Meta-GGA Meta-GGA is “bleeding

TPSS edge” and therefore largely
M06-L kinetic energy density
included untested (but better in
theory…)
B3LYP Hybrid Hybrid makes bigger
B97/2
MPW1K “exact” HF exchange difference in cost and
component accuracy

Hybrid-meta-GGA
MPWB1K
M06

its.unc.edu 17
 Problems with DFT

ƒ ground-state theory only

ƒ universal functional still unknown
ƒ even hydrogen atom a problem: self-interaction
correction
ƒ no systematic way to improve approximations
like LDA, GGA, etc.
ƒ extension to excited states, spin multiplets,
etc., though proven exact in theory, is not
trivial in implementation and still far from
being generally accessible thus far

its.unc.edu 18
 DFT Developments
ƒ Theoretical
• Extensions to excited states, etc.
• Better functionals (mega-GGA), etc
• Understanding functional properties, etc.
ƒ Conceptual
• More concepts proposed, like electrophilicity, philicity, spin-
philicity, surfaced-integrated Fukui fnc
• Dynamic behaviors, profiles, etc.
ƒ Computational
• Linear scaling methods
• QM/MM related issues
• Applications
its.unc.edu 19
 Examples DFT vs. HF

its.unc.edu Hydrogen molecules - using the LSDA (LDA) 20

 LDA Functionals
Exc [ ρ ] = Ex [ ρ ] + Ec [ ρ ]
ƒ Slater form (exchange) – 1 parameter;
ƒ Kohn-Sham Exchange - Variational α = 2/3

3 1/ 3 2 / 3 −1/ 3
E S
X [ ρ ] = α CX ∫ ρ 4/3
( r ) dr , C X = 3 4 π
8

ƒ Wigner correlation – 2 parameters

ƒ Parameters Fixed by reproducing Post HF
He Result on E
ρ (r )
E [ρ ] = a ∫
W
dr
1 + bρ (r )
−1 / 3
C
its.unc.edu 21
 Let us recall The ORIGINAL HF   Orbital Equations

h1 χ i ( x1 ) +
⎡N ⎛ 1 ⎞ ⎤
⎢ ∑ ∫ χ j ( x2 ) ⎜ ⎟ χ j ( x2 )dx2 ⎥ χ i ( x1 )
*

⎣ j =1 ⎝ r12 ⎠ ⎦
⎡N ⎛ 1 ⎞ ⎤
− ⎢ ∑ ∫ χ j ( x2 ) ⎜ ⎟ χ i ( x2 )dx2 ⎥ χ j ( x1 )
*

⎣ j =1 ⎝ r12 ⎠ ⎦
= ε i χ i ( x1 )
“Self Interaction Free” Hamiltonian
is a necessary condition to be met.

its.unc.edu 22
We will now consider two of the important
theorems of Quatum Mechanics

(1)Virial Theorem

(2) Hellmann-Feynman Theorem

Consider H atom Ground State Wave Function & 
Calculate expectation values of     V  and   T 
E1s= ‐0.5 a.u.
=  ‐1  a.u.

<E >=  <T>   +  <V>
Statement of the Virial Theorem:

For the n point particles bound together 
into a system the time average of the 
kinetic energy of the particles, Σ½mivi², 
plus one half of the time average of 
ΣFi∙ri is equal to zero.
Use F = −∇V
THE QUANTUM MECHANICAL VIRIAL THEOREM
2 < T > = < r.∇V >
< T > = <r.∇V>
HELLMANN‐FEYNMAN THEOREM 
Parameters in the Hamiltonian

N
⎛ = 2
⎞ 2
e 1
H = ∑⎜ − ∇i + V (ri ) ⎟ + ∑
2
;
i =1 ⎝ 2m ⎠ i < j 4πε 0 rij
2
e Zα ,i
V (ri ) = ∑
i ,α 4πε 0 rα i
N
= ∑ h(ri ) + ∑ v(ri , rj )
i =1 i< j
λ is any parameter which can be varied inside the expression of the Hamiltonian