You are on page 1of 225

Election Bensity Functional Theoiy Page 1

Roi Baei

Electron Density
Functional Theory
Lecture notes (rough draft)
October 2009


Roi Baer



Institute of Chemistry,
The Fritz Haber Center for Molecular Dynamics
The Hebrew University of Jerusalem,
Jerusalem, 91904 Israel

Roi Baer
Election Bensity Functional Theoiy Page 2
Roi Baei

Contents
I. INTRODUCTORY BACKGROUND TOPICS ............................................................. 7
A. THE SCHRDINGER EQUATION ....................................................................................... 7
B. BOSONS AND FERMIONS ................................................................................................... 7
I. GROUND-STATE WAVE FUNCTION OF BOSONS .................................................................... 7
II. GROUND-STATE WAVE FUNCTION OF FERMIONS ............................................................. 10
C. WHY ELECTRONIC STRUCTURE IS AN IMPORTANT DIFFICULT PROBLEM .................. 10
D. THE BORN-OPPENHEIMER THEORY ............................................................................. 12
I. THE ADIABATIC THEOREM ................................................................................................. 12
II. MOTIVATION FOR THE BORN-OPPENHEIMER APPROXIMATION: CLASSICAL NUCLEI ....... 16
III. THE BORN-OPPENHEIMER APPROXIMATION IN QUANTUM NUCLEAR CASE .................... 18
E. ELECTRON CORRELATION ............................................................................................. 21
I. THE ELECTRONIC WAVE FUNCTION OF TWO NON-INTERACTING ELECTRONS ................... 21
II. CORRELATION IN ACTION: A WAVE FUNCTION OF 2 INTERACTING PARTICLES IN AN
HARMONIC TRAP ..................................................................................................................... 24
F. THE ELECTRON DENSITY IS A MUCH SIMPLER OBJECT THAN THE WAVE FUNCTION . 31
G. THE VARIATIONAL PRINCIPLE ...................................................................................... 34
H. DILATION RELATIONS .................................................................................................... 36
I. THE CONCEPT OF THE VIRIAL IN CLASSICAL MECHANICS .................................................. 37
II. THE VIRIAL THEOREM IN QUANTUM MECHANICS ............................................................ 39
III. SLATERS THEORY OF THE CHEMICAL BOND USING THE VIRAL THEOREM ..................... 40
IV. OTHER DILATION FACTS ................................................................................................. 47
I. INTRODUCTION TO FUNCTIONALS ................................................................................. 48
I. DEFINITION OF A FUNCTIONAL AND SOME EXAMPLES ...................................................... 48
II. LINEAR FUNCTIONALS ...................................................................................................... 49
III. FUNCTIONAL DERIVATIVES ............................................................................................. 50
IV. INVERTIBLE FUNCTION FUNCTIONALS AND THEIR FUNCTIONAL DERIVATIVES .............. 53
V. CONVEX FUNCTIONALS ................................................................................................... 54
J. MINIMIZING FUNCTIONALS IS AN IMPORTANT PART OF DEVELOPING A DFT ............ 58
I. MINIMIZATION OF FUNCTIONS ........................................................................................... 58
II. CONSTRAINED MINIMIZATION: LAGRANGE MULTIPLIERS ................................................ 60
III. MINIMIZATION OF FUNCTIONALS .................................................................................... 64
Election Bensity Functional Theoiy Page S
Roi Baei

II. MY FIRST DENSITY FUNCTIONAL: THOMAS-FERMI THEORY .................. 68


A. BASIC CONCEPTS IN THE ELECTRON GAS AND THE THOMAS-FERMI THEORY .......... 68
B. MINIMIZATION OF THE THOMAS-FERMI ENERGY ....................................................... 73
C. THOMAS-FERMI DOES NOT ACCOUNT FOR MOLECULES ............................................. 75
D. THOMAS-FERMI SCREENING ......................................................................................... 76
E. VON WEIZSCKER KINETIC ENERGY ............................................................................ 80
III. MANY-ELECTRON WAVE FUNCTIONS ............................................................. 81
A. THE ELECTRON SPIN ...................................................................................................... 81
B. THE PAULI PRINCIPLE .................................................................................................... 82
C. THE EXCITED STATES OF THE HELIUM ATOM ............................................................. 83
D. THE SLATER WAVE FUNCTION IS THE BASIC ANTI-SYMMETRIC FUNCTION
DESCRIBING N ELECTRONS IN N ORBITALS .......................................................................... 83
E. WITHOUT LOSS OF GENERALITY, WE MAY ASSUME THE ORBITALS OF A SLATER
WAVE FUNCTION ARE ORTHOGONAL .................................................................................... 84
F. ANY ANTISYMMETRIC FUNCTION CAN BE EXPANDED AS A SUM OF BASIC SLATER
(DETERMINANTAL) FUNCTIONS ............................................................................................. 85
G. DETERMINANT EXPECTATION VALUES ......................................................................... 87
I. ONE-BODY OPERATORS ..................................................................................................... 87
II. TWO-BODY OPERATORS ................................................................................................... 88
IV. THE HARTREE-FOCK THEORY ........................................................................... 89
A. THE HARTREE-FOCK ENERGY AND EQUATIONS ......................................................... 89
B. RESTRICTED CLOSED-SHELL HARTREE-FOCK ............................................................ 95
C. ATOMIC ORBITALS AND GAUSSIAN BASIS SETS ........................................................... 98
D. VARIATIONAL-ALGEBRAIC APPROACH HARTREE-FOCK ........................................... 99
E. THE ALGEBRAIC DENSITY MATRIX AND CHARGE ANALYSIS................................... 105
F. SOLVING THE HARTREE-FOCK EQUATIONS ............................................................... 107
I. DIRECT INVERSION IN ITERATIVE SPACE (DIIS) .............................................................. 107
II. DIRECT MINIMIZATION .................................................................................................. 108
G. PERFORMANCE OF THE HARTREE-FOCK APPROXIMATION ..................................... 109
Election Bensity Functional Theoiy Page 4
Roi Baei

H. BEYOND HARTREE-FOCK ............................................................................................ 110


V. ADVANCED TOPICS IN HARTREE-FOCK THEORY ....................................... 112
A. LOW-LYING EXCITATIONS AND THE STABILITY OF THE HARTREE-FOCK GROUND
STATE .................................................................................................................................... 112
I. CI-SINGLES AND BRILLOUINS THEOREM ....................................................................... 112
II. HARTREE-FOCK STABILITY ............................................................................................ 114
B. KOOPMANS THEOREM ................................................................................................ 116
C. FRACTIONAL OCCUPATION NUMBERS, THE HF ORBITAL FUNCTIONAL AND THE
GENERALIZED KOOPMANS THEOREM ............................................................................... 119
D. HARTREE-FOCK FOR THE HOMOGENEOUS ELECTRON GAS ...................................... 121
I. HARTREE-FOCK ORBITALS AND ORBITAL ENERGIES OF HEG ......................................... 121
II. THE DENSITY OF STATES OF THE HEG ........................................................................... 127
III. STABILITY OF THE HEG IN HARTREE-FOCK THEORY ................................................... 129
VI. THE HOHENBERG-KOHN DENSITY THEORY ............................................... 131
A. THE FIRST HK THEOREM ............................................................................................ 131
B. THE HK FUNCTIONAL .................................................................................................. 133
C. MINIMUM PRINCIPLE FOR DENSITY FUNCTIONAL THEORY ...................................... 135
D. AN INTERESTING OBSERVATION ON THE VARIATIONAL PRINCIPLE OF NON-
INTERACTING ELECTRONS ................................................................................................... 137
E. THE SET OF V-REPRESENTABILE DENSITIES ............................................................... 138
I. V-REP DENSITIES CORRESPOND TO GROUND STATES WAVE FUNCTION OF SOME POTENTIAL
WELL ...................................................................................................................................... 138
II. SOME NON-V-REPRESENTABILIT ISSUES ......................................................................... 139
III. THE SET OF V-REP DENSITIES OF A GIVEN ELECTRON NUMBER IS NOT CONVEX ........... 140
F. LEVY-LIEB GENERALIZATION OF THE HK FUNCTIONAL .......................................... 143
G. THE DILATION INEQUALITY FOR THE HK FUNCTIONAL ........................................... 148
VII. THE KOHN-SHAM METHOD .............................................................................. 149
A. NON-INTERACTING ELECTRONS .................................................................................. 149
Election Bensity Functional Theoiy Page S
Roi Baei

B. ORBITALS FOR THE NON-INTERACTING ELECTRONS ................................................. 153


C. THE CORRELATION ENERGY FUNCTIONAL: DEFINITION AND SOME FORMAL
PROPERTIES .......................................................................................................................... 154
D. THE KOHN SHAM EQUATIONS ..................................................................................... 157
I. THE KOHN-SHAM EQUATION FROM A SYSTEM OF NON-INTERACTING PROBLEM ........... 157
II. SYSTEMS WITH PARTIALLY OCCUPIED ORBITALS ........................................................... 159
III. IS THE GROUND STATE WAVE FUNCTION OF NON-INTERACTING PARTICLES ALWAYS A
SLATER WAVE FUNCTION ...................................................................................................... 161
IV. J ANAKS THEOREM ....................................................................................................... 163
E. VIRIAL THEOREM RELATED IDENTITIES IN DFT ................................................... 164
F. GALILEAN INVARIANCE................................................................................................ 168
G. HOLES AND THE ADIABATIC CONNECTION ................................................................. 170
I. THE EXCHANGE AND CORRELATION HOLES .................................................................... 170
II. THE FERMI-COULOMB HOLE FOR HARMONIC ELECTRONS ............................................. 175
III. THE FERMI HOLE IN THE NON-INTERACTING SYSTEM ................................................... 178
IV. THE ADIABATIC CONNECTION ...................................................................................... 181
H. DERIVATIVE DISCONTINUITY IN THE EXCHANGE CORRELATION POTENTIAL
FUNCTIONAL ......................................................................................................................... 184
VIII. APPROXIMATE CORRELATION ENERGY FUNCTIONALS ..................... 185
A. THE LOCAL DENSITY APPROXIMATION (LDA) ........................................................... 185
I. THE EXCHANGE ENERGY PER ELECTRON IN THE HEG .................................................... 186
II. CORRELATION ENERGY OF THE HEG: THE HIGH DENSITY LIMIT ................................... 187
III. CORRELATION ENERGY OF THE HEG: THE LOW DENSITY LIMIT AND THE WIGNER
CRYSTAL ................................................................................................................................ 189
IV. MONTE-CARLO DETERMINATION OF THE CORRELATION ENERGY FOR THE HEG ........ 192
V. THE POLARIZED HEG; LOCAL SPIN-DENSITY APPROXIMATION (LSDA) ....................... 192
VI. SUCCESSES AND FAILURES OF LSDA ............................................................................ 193
VII. PLAUSIBLE REASONS FOR THE SUCCESS OF LSDA ...................................................... 193
B. SEMILOCAL FUNCTIONALS AND THE GENERALIZED GRADIENT APPROXIMATION .. 196
IX. GENERALIZED KOHN-SHAM APPROACHES ................................................. 196
Election Bensity Functional Theoiy Page 6
Roi Baei

A. THE GENERALIZED KOHN-SHAM FRAMEWORK ........................................................ 196


B. KOHN-SHAM FROM GENERALIZED KS ....................................................................... 199
C. THE HYBRID FUNCTIONAL OF BECKE ........................................................................ 199
D. ........................................................................................................................................... 200
E. LONG-RANGE SELF-REPULSION AND LACK OF DERIVATIVE DISCONTINUITY .......... 200
F. RANGE SEPARATED HYBRIDS ....................................................................................... 200
G. ORBITAL FUNCTIONALS AND OPTIMIZED EFFECTIVE POTENTIALS .......................... 200
H. APPROXIMATE CORRELATION FUNCTIONALS AND THE BORN-OPPENHEIMER FORCE
ON NUCLEI............................................................................................................................. 200
X. MORE ON THE DFT CORRELATION ENERGY ................................................ 203
A. APPROXIMATIONS TO E
X
ARE APPROXIMATION TO E
C
............................................. 203
B. THE NECESSITY OF LONG-RANGE PARAMETER TUNING ............................................ 209
I. THE SYMMETRIC RADICAL CATION ................................................................................. 209
II. AROMATIC DONOR -TCNE ACCEPTOR CHARGE TRANSFER EXCITATION ....................... 211
XI. TDDFT ........................................................................................................................ 214
A. TIME-DEPENDENT LINEAR RESPONSE THEORY ......................................................... 214
B. ALGEBRAIC APPROACH TO LINEAR RESPONSE .......................................................... 216
C. FREQUENCY-DOMAIN RESPONSE ................................................................................. 219
D. EXCITATION ENERGIES FROM THE ALGEBRAIC TREATMENT ................................... 220


Election Bensity Functional Theoiy Page 7
Roi Baei

I. Introductory background topics


A. The Schrdinger equation
XXX
B. Bosons and Fermions
i. Ground-state wave function of Bosons
ConsideraonedimensionalSchrdingerequationforthegroundstate,which
may be taken real. We will ask: is this wave function of the same sign
everywhere? What we mean is: is it all nonnegative or all non positive? Of
course,ifitisallnonpositivethenwecanmultiplyitby1andaskagain:isit
allnonnegative?Sayingitisofaconstantsignisthesameassayingithasno
nodes,i.e.itdoesnotgothroughthexaxisanywhere.
Exercise:Provethefollowingtheorem:
Theorem: The groundstate wave function of a 1D wave nondegenerate
Schrdingerequationhasnonodes.
Proof.Letusassume(x)isthe(real)normalizedgroundstate.Thenitisthe
wavefunctionminimizesthefunctional:
E|] = I|] +I|] =
l
2
2m
]
i
(x)
2
Jx

-
+] (x)
2
I(x)Jx

-
] (x)
2
Jx

-

(1.1.1)
Arobustexpressionforthekineticenergyexpectationvalue
ThekineticenergyisusuallydefinedasI|] = -
l
2
2m
] (x)
-

ii
(x)Jx

-
.Letus
callthisI
1
.Forcontinuousandtwicedifferentiablefunctionsthatgotozeroat
_ an equivalent definition, is I|] =
l
2
2m
]
|
i
(x)|
2
Jx

-
. Let us call this
Election Bensity Functional Theoiy Page 8
Roi Baei

definition I
2
. We now show that I
2
is more robust as it holds for wave
functions that are discontinuous. To show that I
1
fails, consider the wave
function
0
(x) = c
-|x|
. Notice, it is normalized (
0
|
0
) = 1. Let us also
assumeforthisexamplethat
l
2
2m
= 1.Now,forx = u
0
i
(x) = -c
-|x|
|x|
x
(there
is a discontinuity of
0
i
(x) at x = u) and
0
ii
(x) =
0
(x). Now I
1
=
-]
0
(x)
-

0
ii
(x)Jx

-
= -1 which is physically absurd! Kinetic energy must
bepositive!Ontheotherhand,I
2
= ]
|
0
i
(x)|
2
Jx

-
= 1.Toseethatthesecond
resultisproper,letussmooththecuspbydefining
u
(x) = c
-_
x
2
|x|+o
]
.Clearly,
wheno = uwehavearebackattheoriginalfunction
0
(x),butforpositiveo,
no matter how small, the first derivative
u
i
(x) = -
|x|(2u+|x|)
(|x|+u)
2
c
-_
x
2
|x|+o
]
is
everywhere continuous while
0
i
(x) has a discontinuity at x = u. By using
nonzerooweregularizethediscontinuity.Now,wecancalculatetheintegrals
numericallyandwe:
o = u.1 (
u
|
u
) = 1.1S2 I
1
|
u
] = u.912 I
2
|
u
] = u.912
o = u.u1 (
u
|
u
) = 1.u19 I
1
|
u
] = u.988 I
2
|
u
] = u.988
o = u.uu1 (
u
|
u
) = 1.uu2 I
1
|
u
] = u.999 I
2
|
u
] = u.999
BothI
1
andI
2
giveherethesameresultnowsince
u
i
iscontinuous.
From the table it is clear that lim
u-0
I
1
|
u
] = lim
u-0
I
2
|
u
] = I
2
|
0
] =
I
1
|
0
].Clearly,theexpressionI
1
isnotsuitedfordiscontinuitiesin
i
.
UsingtheI
2
expression,weinferthatI|] = I|||].Explainwhy.Isthistrue
if(x)hasanodeatx
0
?NowexplainwhyI|] = I|||].
Now there are two options. Either || = _. In this case there is no sign
changeandthereisnonode.Otherwisethereisanode.Assumeitisinx
0
.
Election Bensity Functional Theoiy Page 9
Roi Baei

Let us compute the energy change when one makes a small perturbation to
||:
oE|||] = oI|||] +2 _ |(x)|(I(x) -E|||])o|(x)|Jx

-
(1.1.2)
Verifythisandshowweexplicitlyassumedthatthenormofis1.
Assumethenodeisatx
0
= usothewavefunctionchangessignthere.

Figure I1:Theabsolutevalueofthewavefunctionandtheparabolainit.
We plot in Figure I1 a typical situation. We determine a large enough
parameteroandtheparabola:
p(x) =
1
2
o(x -x
0
i
)
2
+c
which has the properties that it is tangent at some x
1
and x
2
to |(x)| (and
wherex
1
< x
0
i
< x
2
).Thenonedefinesanewwavefunction:
_(x) = _
|(x)| x < x
1
or x > x
2
p(x) x
1
< x < x
2
(1.1.3)
By increasing o the parabola becomes narrow; adjusting x
0
i
accordingly we
cancausex
1
andx
2
touniformlyapproachuascloseasneeded.Underthese
conditions,itispossibletoshowthat
X1 X2
||
g(x)
Election Bensity Functional Theoiy Page 1u
Roi Baei

-x
1
= x
2
=
|
i
(u)|
o

c =
1
2

|
i
(u)|
2
o

Now, I|_] -I|] o
-2
. One power of o
-1
comes from integration over the
interval of length x
2
-x
1
where the two functions differ. The second power
comes from the difference p(x
0
) -(x
0
) = c o
-1
. The kinetic energy
differenceis negative:I|_] -I|] = -
l
2
2m
|
|
(0)|
3
u
.Aso - the total changein
energyisdominatedbythechangeinthekineticenergyandisthusnegative,
showingthatE|_] < E|||] = E|].Thisisacontradiction,sinceisassumed
thegroundstate.
The proof can be extended to any number of dimensions. It can be used to
provethatthegroundstateofamanybosonwavefunctionhasnonodes.
Prove an important immediate corollary from this theorem: the ground state
ofbosonsystemsisnondegenerate.
ii. Ground-state wave function of Fermions
XXX
C. Why electronic structure is an important
difficult problem
In this course, we will study methods to treat electronic structure. This
problem is often considerably more complicated than nuclear dynamics of
molecules. The most obvious reason is that usually there are many more
electrons than nuclei in a molecule. However, there are more reasons:
electrons in molecules have extreme quantum mechanical character, while
nuclei are more classical. This last statement requires probably some
Election Bensity Functional Theoiy Page 11
Roi Baei

explanation.Wewilldiscussinthiscoursewaysofmappingtheelectronsina
molecule onto a system of noninteracting particles. In such a picture, each
electron has its own orbital or wavefunction. Generally, the orbitals of
different electrons in molecules strongly overlap. Furthermore, most of the
valenceorbitalsarewidelyspreadoverspace,encapsulatingmanymolecular
nuclei. Nuclei on the other hand, are massive, so their wave functions are
strongly localized and hardly overlap once again, they behave much more
likeclassicalparticles.
Describing electronic structure is therefore a quantum mechanical problem.
Now, the electrons interact, via the attractive Coulomb force, not only with
thestationarynucleibutalsowitheachother:eachelectronrepelseachother
electronviatherepulsiveCoulombforce.Thismakestheelectronicstructure
problemamanybodyproblem.
Quantum manybody problems are very difficult to track down, much more
difficult than classical mechanical ones. The reason is that for
e
N
electrons
themanybodywavefunctionisthenafunctionof
3
e
N
variables,
( )
1
...
e
N
r r ,
where
i
r (i=1
e
N
) is 3 dimensional the position vector. Highly accurate
numerical representation of such functions is next to impossible for
1
e
N >
.
Thus, electronic structure methods invariably include approximate methods.
Typically, the wave function encapsulates much more information than we
care to know about. For example, ( )
2
3 3
1 1
,...,
e e
N N
d r d r r r gives the
probabilityforfindinganelectronat
1
r andanelectronat
2
r andanelectronat
3
r etc.Indeed,thetotalprobabilityis1:
_|(r
1
, , r
N
c
)|
2
J
3
r
1
J
3
r
N
c
= 1 (1.3.1)
Election Bensity Functional Theoiy Page 12
Roi Baei

D. The Born-Oppenheimer theory


i. The adiabatic theorem
Suppose a Hamiltonian is dependent parameterically on R = (R
1
, R
2
, ). We
writethisasE

|R].DenotetheeigenstatesofthisHamiltonianby
E

|R]
n
|R] = E
n
|R]
n
|R] (1.4.1)
Now,supposewechangeRwithtime,sothatwehaveatrajectory,R(t).The
Hamiltonian will become time dependent: E

(t) = E

|R(t)]. Suppose the


systemisplacedinitsgroundstateattimet = u,(t = u) =
0
|R(u)].Itwill
evolveaccordingtotheTDSE:
il

(t) = E

(t)(t) (1.4.2)
Now,theadiabatictheoremsaysthatifR(t)changesveryslowlythen:
(t) = c
0(t)

0
|R(t)] (1.4.3)
Namely, except for a purely TD phase, the evolving state stays the
instantaneousgroundstateoftheHamiltianE

(t).
Toshowthis,weusetheinstantaneouseigenstatestoexpand(t):
(t) = o
n
(t)c
-
i
l
] L
n
(:)d:
t
0

n
|R(t)]

n=0
(1.4.4)
WewanttoplugthisexpansionintotheSE.Soletuscalculate:
il

(t) = |ilo
n
+E
n
o
n
]c
-
i
l
] L
n
(:)d:
t
0

n
|R(t)]

n=0
+ o
n
c
-
i
l
] L
n
(:)d:
t
0
il

n
|R(t)]

n=0

(1.4.5)
And:
Election Bensity Functional Theoiy Page 1S
Roi Baei

(t) = o
n
c
-
i
l
] L
n
(:)d:
t
0
E
n

n
|R(t)]

n=0
(1.4.6)
Equatingthetwoexpressionsgives:]
u = ilo
n
c
-
i
l
] L
n
(:)d:
t
0

n
|R(t)]

n=0
+ o
n
c
-
i
l
] L
n
(:)d:
t
0
il

n
|R(t)]

n=0
(1.4.7)
Nowletusmultiplyby
m
andintegrate(rememberthat(
n
|
m
) = o
nm
):
u = ilo
m
c
-
i
l
] L
m
(:)d:
t
0
+ o
n
c
-
i
l
] L
n
(:)d:
t
0
il(
m
|

n
)

n=0
(1.4.8)
Now,wehave:
u =
J
Jt
(
m
|
n
) = (
m
|

n
) +(

m
|
n
)
(1.4.9)
Andsince(

m
|
n
) = (
n
|

m
)
-
,wefind:

mn
= (
m
|

n
) = -(
n
|

m
)
-
= -
nm
-
(1.4.10)
The matrix
mn
= (
m
|

n
) is the matrix of timedependent nonadiabatic
couplings (TDNACs) and it is an antiHermitean matrix. Furthermore, for
n = m
Rc(
n
|

n
) = u (1.4.11)
Thus:
o
m
= -
mm
o
m
- c
] o
mn
(t
|
)dt
|
t
0

mn
o
n

n=m
(1.4.12)
Wesee
mn
createsthenonadiabatictransitions,i.e.thetransitionsoutofthe
groudstate.
Letusplayabitwith.TakethetimederivativeoftheTISE:
Election Bensity Functional Theoiy Page 14
Roi Baei

(t)
n
(t) +E

(t)

n
(t) = E

n
(t)
n
(t) +E
n
(t)

n
(t)
(1.4.13)
Multiplyby(
m
|andget:
[
m
E


n
+(
m
|E

n
) = E

n
(t)(
m
|
n
) +E
n
(
m
|

n
). (1.4.14)
Since: (
m
|E

n
) = E
m
(
m
|

n
)wehave:
[
m
E


n
= E

n
(t)(
m
|
n
) +(E
n
-E
m
)(
m
|

n
). (1.4.15)
Ifn = mthen:
E

n
(t) = [
n
E


n
. (1.4.16)
ThisiscalledtheHellmannFeynmantheorem,showingthatthepoweristhe
expectationvalueoftherateofchangeoftheHamiltonian.Ifn = m:

mn
= (
m
|

n
) =
[
m
E

E
n
-E
m
.
(1.4.17)
This is called Epsteins theorem, giving an expression for the TDNACs.
Remember that E

|R(t)] depends on t through the positions of the nuclei.


Thus:
J
Jt
E

|R(t)] = R

N

R
N
E

|R(t)] (1.4.18)
Thus:

mn
=
[
m

0
0R
N
E

|R(t)]
n

E
n
|R] -E
m
|R]
R

N
(t)
N
=
mn
N
R

N
(t)
N

(1.4.19)
where:

mn
N
|R] = [
m
_
o
oR
N

n
=
[
m
|R]
0
0R
N
E

|R]
n
|R]
E
n
|R] -E
m
|R]

(1.4.20)
Election Bensity Functional Theoiy Page 1S
Roi Baei

We see that the TDNACs depend on the velocity of the nuclei. When nuclei
are small the couplings out of the ground state are small. The
nm
N
are called
thenonadiabaticcouplings(NACs).WefindthataslongasE
n
-E
m
arenot
zero on the trajectory, one can always find small enough R

that makes the


TDNACs as small as we wish. All we need for states n and m to stay
decoupledisforthefollowingconditionstobemet:
|
mn
| = |R

N
(t)
mn
N
(R
N
(t))| < 1 (1.4.21)
Thus we can make the NACs small. So all that is left in Eq. (1.4.12) is o
m
=
-
mm
o
m
.Definethenondynamicalphaseas:0
nd
(t) = i ]
mm
(t
i
) Jt
i
t
0
(note0
isreal)andthen:
o
m
(t) = c
0
nd
(t)
(slow proccsscs = oJiobotic proccss). (1.4.22)
Thetotalstateis
(t) = o
m
(t)c
-
i
l
] L
m
(:)d:
t
0

m
|R(t)] = c
(0
nd
(t)+0
d
(t))

m
|R(t)]
(1.4.23)
where 0
d
(t) = -
1
l
] E
m
()J
t
0
is called the dynamic phase. It is easy to prove
thatifR(t)traversesaclosedloop,thenondynamicalphasedependsonlyon
that loop and not on the way it is traversed. For example, if we traverse the
same loop using different velocities, the dynamic phase may change but the
geometricphasewillnot.TheclosedloopgeometricangleiscalledtheBerry
phase.Theindependenceofpathisaresultofthefactthatthenondynamical
phaseisalineintegral,andcanbemadewithnoreferencetotime:
Election Bensity Functional Theoiy Page 16
Roi Baei

0
nd
(t) = i _ (
m
(R|])|

m
(R|]))J
t
0
= i _ [
m
|R]_
o
oR
N

m
|R] R

N
N
J
t
0
= i _
mm
N
(R)
N
JR
N
R(t)
R(0)

(1.4.24)
Thedynamicalphaseforexampleisnotalineintegralanditsvaluedepends
not only on the path itself but also on the velocity taken along the way. This
observation makes the nondynamical phase a special quantity. Berry has
shownthatthisquantitycangiveusinformationonthewaytheHamiltonian
is dependent on its parameters. For a real Hamiltonian, for example c
0
nd

aroundaclosedpathalwaysequalseither1or1.Ifthereisanevennumber
ofdegeneraciesenclosedbythepathitis1andifanoddnumberitis1.

ii. Motivation for the Born-Oppenheimer approximation: classical nuclei


Inamolecule,wecanthinkofthenuclei(havingcoordinatesR)asheavyand
therefore classical particles which are slowly moving. The electrons are light
and therefore quantum. They have coordinates r (r includes spin) and they
R
1
R
2

A path in parameter
spaceR
1
-R
2

Election Bensity Functional Theoiy Page 17


Roi Baei

feeltheeffectofslowlymovingnuclei.Theelectrondynamicsiscontrolledby
theHamiltonian:
E

c
|R] = I
`

+I(r, R). (1.4.25)


The potential I(r, R) describes the Coulombic interactions between electrons
andnuclei(inatomicunits):
I(r, R) = -
Z
N
|R

N
-r
n
|
n,N
+
1
2

1
r
nm
n=m
+
1
2

Z
N
Z
M
R
NM
N=M
.
(1.4.26)
In general, one wants to assume that the total energy of the molecule in this
classicalapproximationis:
E = I
N
+(
c
|R]|E

c
|R(t)]|
c
|R]) (1.4.27)
Where
c
|R] is the ground state of the electrons at nuclear configuration R.
The adiabatic theorem states that this is reasonable as long as nuclei move
slowly.Thus,theadiabatictheoremallowsustowrite:
E = I
N
+I
c
(R) (1.4.28)
where I
c
(R) is the ground state energy of the electrons at nuclear
configuration R. This energy is a usual classical mechanical energy and the
Newtonequationsofmotionapply:
H
N
R

N
= F
N
= -
o
oR
N
I
c
(R) (1.4.29)
Wesee,thattheadiabatictheoremallowsustoconsiderthenucleiasmoving
inapotentialwellwhichisessentiallytheeigensvalueoftheelectrons.Thisin
essenceistheBOapproximationwhenthenucleiareclassical.
Election Bensity Functional Theoiy Page 18
Roi Baei

iii. The Born-Oppenheimer approximation in quantum nuclear case


Theclassicalapproachmotivatesaquantumtreatment.Weareexpectingthat
nuclei will not excite electrons very efficiently. That is the motivation for the
BOapproximation.
The Born and Oppenheimer development is similar to that of the adiabatic
theorem,althoughtherearenoexternalfields.Supposewehaveasystemof
fast particles, with coordinates r = (r
1
, r
2
, ) and slow particles, with
coordinatesR = (R
1
, R
2
, ).TheHamiltoniancanbewrittenas:
E

= I
`
R
+I
`

+I(r, R) (1.4.30)
TheSchrdingerequationis:
E

n
(r, R) = E
n

n
(r, R) (1.4.31)
Notethatwecanassumethesewavefunctionsareorthogonal:
(
n
|
m
) = __
n
-
(r, R)
m
(r, R)Jr JR = o
nm
(1.4.32)
Now,toproceed,letusfirstconsiderthefastrHamitonian
F
`
|R] = I
`

+I(r, R) (1.4.33)
In this fast Hamiltonian F
`
|R], the slow variables are simply parameters (it
contains no derivatives with respect to R). Thus, it depends on R
parametrically.Thefasteigenavalueproblemis:
F
`
|R]
k
(r; R) = w
k
(R)
k
(r; R) (1.4.34)
The eigenvalues are functions of the parameters R. They are called the
adiabatic (or BornOppenheimer) surfaces. The notation with the semicolon
betweenrandRisdesignedtoemphasizethat
k
arewavefunctionsinrbut
they depend only parametrically on R. This means, for example that the
Election Bensity Functional Theoiy Page 19
Roi Baei

overlap of the fast eigenstates (called adiabatic states) involves integration


onlyoverr,thedynamicalcoordinates,whileRisheldfixed,sinceitisonlya
parameter:
(
k
(R)|
]
(R)) = _
k
-
(r; R)
]
(r; R)Jr = o
k]
(1.4.35)
Wecannotinfactsayanythingreallymeaningfulabout(
k
(R)|
]
(R
i
))when
R = R
i
(except when R
i
-R is infinitesimal, but we will not go into this issue
here). Now, we can expand the real wavefunction as a linear combination
oftheadiabaticfunctions:

n
(r, R) =
k
(r; R)
kn
(R)
k

(1.4.36)
We can do this because for any give R
k
(r; R) span the space of wave
functionsdependentonr.Infact,theexpansioncoefficients,
nk
(R)aregiven
by:

kn
(R) = _
k
-
(r; R)
n
(r, R)Jr (1.4.37)
Now,letusplugEq.(1.4.36)intotheSE(1.4.31):
E
n

n
(r, R) = (I
`
R
+F
`
)
n
(r, R)
= I
`
R
|
k
(r; R)
kn
(R)]
k
+w
k
(R)
k
(r; R)
kn
(R)
k

(1.4.38)
Notethat,sinceI
`
R
=
P
`
N
2
2M
N
N
where:P
`
N
= -il

R
N
.Wehave:
I
`
R
|
k
(r; R)
kn
(R)]
=
kn
(R)I
`
R
|
k
(r; R)]
+
1
H
N
[P
`
N

k
(r; R) [P
`
N

kn
(R)
N
+
k
(r; R)I
`
R

kn
(R)
(1.4.39)
Election Bensity Functional Theoiy Page 2u
Roi Baei

MultiplyingEq.(1.4.38)by
]
(r; R)
-
andintegratingoverrgives:
(A
]k
+B
`
]k
N
)
kn
(R)
k
+[I
`
R
+w
]
(R)
]n
(R) = E
n

]n
(R)
(1.4.40)
Where:
A
]k
= _
]
(r; R)
-
|I
`
R

k
(r; R)]Jr = (
]
|I
`
R

k
)
B
`
]k
N
=
]
]
(r; R)
-
|P
`
N

k
(r; R)]Jr
H
N
P
`
N
N
=
(
]
|P
`
N

k
)
H
N
P
`
N
N
= -
l
2
H
N
[
]
_
o
oR
N

o
oR
N
N
= -
l
2
H
N

]k
N
o
oR
N
N

(1.4.41)
The matrices
]k
N
= [
]


R
N

k
are the nonadiabatic couplings in the fast
system.Theseareexactlythenonadiabaticcoefficientsintheadiabatictheory
(Eq.(1.4.20)).
Itispossibletoshowthat:
(A
]k
+B
`
]k
N
)
kn
(R)
k
+I
`
R

]n
(R)
= -
l
2
2H
N
_
o
oR
N
+
]k
N
]
2

kn
(R)
N k

(1.4.42)
ThusEq.(1.4.40)becomes:
-
l
2
2H
N
_
o
oR
N
+
]k
N
]
2

kn
(R)
N k
+w
]
(R)
]n
(R) = E
n

]n
(R) (1.4.43)
ThisisaSchrdingerlikeequationwhichdeterminesthecoefficients
]n
(R).
Thesearecalledthesloweigenfunctions.Oncetheyarecomputed onehas
an exact solution to the Schrdinger equation. However, we do not really
want to solve this infinite set of coupled differential equations. Thus we
assumethatthequantities
]k
N
for] = kcanbeneglected.Notethatherethere
Election Bensity Functional Theoiy Page 21
Roi Baei

is no R

N
which can be taken as small as needed to make the effect of
]k
N
as
minuteasweneed.Stillwecanjopethatthefact that we choseR tobeslow
degrees of freedom allow us to make just this approximation! This was the
ideaofBornandOppenheimerwhoneglected
]k
N
:
-
l
2
2H
N
o
2
oR
N
2

]n
(R)
N
+w
]
(R)
]n
(R) = E
n

]n
(R)
(1.4.44)
The resulting equation is a Schrdinger equation for the slow degrees of
freedom, which move under the force of electrons derived from a potential
w
]
(R). When applied to molecules the slow degrees of freedom are usually
thenucleiandthefasttheelectrons..Thereisaproblemwithneglecting
]]
N

because of their nondynamical effect. Taking them into account results in


treatingthemasamagneticfield:

l
2
2H
N
_i
o
oR
N
-
mm
]
2

]n
(R)
N
+w
]
(R)
]n
(R) = E
n

]n
(R) (1.4.45)
The BO approximation breaks the molecular SE into two consecutive tasks.
First, the electronic SE Eq. (1.4.34) must be solved for any relevant clamped
positionofnucleiR.Then,thenuclearequation(1.4.26).
Furtherreadingonthissubject: M. Baer, Beyond BornOppenheimer:
electronic nonadiabatic coupling terms and conical intersections (Wiley,
Hoboken,N.J.,2006).
E. Electron correlation
i. The electronic wave function of two non-interacting electrons
In order to appreciate the complexity of the electronic wave function, let us
firststudyasimplesystem,oftwononinteractingelectronsina1Datomic
Election Bensity Functional Theoiy Page 22
Roi Baei

well.Weconsideranatomicwellgivenbythepotential:
cxt
(x)andweplace
initanelectron.TheHamiltonianis:
b
`
= -
l
2
2m
c
v
2
+:
cxt
(x) (1.5.1)
Now consider a 2electron problem. Assume we have two electrons,
Fermions,whicharenoninteractinginthewell.TheHamiltonianis
E

= b
`
(1) +b
`
(2) (1.5.2)
Thenotationb
`
(i)meanstheHamiltonianofEq.(1.5.1)withthecoordinateof
theithelectron.
What are the eigenstates in this case? First, since each electron can have a
spin,wemustdecideonthespinofthestate.Fornow,letusassumethestate
is spinpolarized, i.e. that the total spine is 1, both electrons are in spinup
orbitals.Wetrythefollowingformasawavefunction:
(x
1
, x
2
) =
1
2
|
0
(x
1
)
1
(x
2
) -
0
(x
2
)
1
(x
1
)]o(1)o(2) (1.5.3)
Noticethatthespatialpartisantisymmetricwhilethespinpartissymmetric.
Thisrenderstheentirewavefunctionantisymmetric,inaccordancewiththe
Pauli principle. The notation o(1)o(2) means both electrons have spin
projectionup (o).Wedonotyetknowifandunderwhatconditionsthiswave
function can actually describe eigenstates of the two electrons the
Hamiltonian (1.5.2). We assume that
0
(x) and
1
(x) are orthonormal. This
makesthestatenormalized,since:
Election Bensity Functional Theoiy Page 2S
Roi Baei

_(x
1
, x
2
)
2
Jx
1
Jx
2
=
1
2
_(
0
(x
1
)
1
(x
2
) -
0
(x
2
)
1
(x
1
))
2
Jx
1
Jx
2
=
1
2
_j(
0
(x
1
)
1
(x
2
))
2
+(
0
(x
2
)
1
(x
1
))
2
-2
0
(x
1
)
1
(x
2
)
0
(x
2
)
1
(x
1
)]
Jx
1
Jx
2

(1.5.4)
Thefirstandsecondtermsarebothequalto:
_(
0
(x
1
)
1
(x
2
))
2
Jx
1
Jx
2
=_
0
(x
1
)
2
Jx
1
_
1
(x
2
)
2
Jx
2
=1 1
= 1
(1.5.5)
Thethirdtermin(1.5.4)is:
_
0
(x
1
)
1
(x
2
)
0
(x
2
)
1
(x
1
) Jx
1
Jx
2
=_
0
(x
1
)
1
(x
1
) Jx
1
_
0
(x
2
)
1
(x
2
) Jx
2
= u u = u
(1.5.6)
Indeed the entire wave function is orthonormal (thanks to the factor 12 in
(1.5.3)).
Now, let us see under what condition the wave function in Eq. (1.5.3) is an
eigenstateoftheHamiltonianin(1.5.2)
E

(x
1
, x
2
) = [b
`
(1) +b
`
(2) (x
1
, x
2
)
=
1
2
|
1
(x
2
)b
`
(1)
0
(x
1
) +
0
(x
1
)b
`
(2)
1
(x
2
)|
-
1
2
|
0
(x
2
)b
`
(1)
1
(x
1
) +
1
(x
1
)b
`
(2)
0
(x
2
)|
(1.5.7)
Thisshouldbeequatedto:
E(x
1
, x
2
) =
1
2
{E
0
(x
1
)
1
(x
2
) -E
0
(x
2
)
1
(x
1
)] (1.5.8)
If we choose the orbitals
0
(x) and
1
(x) to be eigenstates of

h (which are
orthogonalsothatiscompatiblewithourpreviousassumption):

Election Bensity Functional Theoiy Page 24


Roi Baei

b
`

(x) = e

(x), i = u,1, (1.5.9)


Thus:
E

(x
1
, x
2
) =
1
2
{e
0

1
(x
2
)
0
(x
1
) +e
1

0
(x
1
)
1
(x
2
)]
-
1
2
{e
1

0
(x
2
)
1
(x
1
) +e
0

1
(x
1
)
0
(x
2
)]
= (e
0
+e
1
)(x
1
, x
2
)
(1.5.10)
And we see that indeed (x
1
, x
2
) is an eigenstate with energy
E = e
0
+e
1
.
ii. Correlation in action: a wave function of 2 interacting particles in an
Harmonic trap
Wenowbuildasimpleelectronicstructuremodelthatwillallowustostudy
insomedetailthemostbasicconcepts.Forthis,wesupposethattheelectrons
areinaharmonicatom,thatisthepotentialwell:
:
cxt
(x) =
1
2
p
2
x
2

(1.6.1)
ThetwolowesteigenstatesofaHarmonicoscillatorare:

0
(x) = N
0
c
-
1
2
n
l
x
2

1
(x) = N
1
_
p
l
xc
-
1
2
n
l
x
2

(1.6.2)
Thenormalizationconstantsare:
N
0
= [
p
nl

14
, N
1
= _
4p
nl
]
14

(1.6.3)
Andtheeigenvaluesare:
e
n
= _n +
1
2
] l, n = u,1,
(1.6.4)
Election Bensity Functional Theoiy Page 2S
Roi Baei

The groundstate energy of the 2electron system in the triplet state will be
placingoneelectronin
0
(x)andanotherin
1
(x):
E = e
0
+e
1
= 2l (1.6.5)
As discussed above singlet and triplet twoelectron ground state wave
functions composed of two orbitals must be spacesymmetric or
antisymmetric, respectively. We consider below 3 wave functions. The first

S,00
is the ground state singlet where both electrons are in
0
. The second
and third are a singlet and a triplet made from one electron in
0
and the
otherin
1
:

S,00
(x
1
, x
2
) =
0
(x
1
)
0
(x
2
) = N
00
c
-
1
2
n
l
(x
1
2
+x
2
2
)

S,01
(x
1
, x
2
) =
1
2
|
0
(x
1
)
1
(x
2
) +
0
(x
2
)
1
(x
1
)]
= N
01
c
-
1
2
n
l
(x
1
2
+x
2
2
)
(x
2
+x
1
)

T,01
(x
1
, x
2
) =
1
2
|
0
(x
1
)
1
(x
2
) -
0
(x
2
)
1
(x
1
)]
= N
01
c
-
1
2
n
l
(x
1
2
+x
2
2
)
(x
2
-x
1
)
(1.6.6)
Thenormalizationfactorsare:
N
00
= N
0
2
= _
p
nl
N
01
=
1
2
_
p
l
N
0
N
1
=
p
nl
(1.6.7)
Eq. (1.6.6) describes the distribution of positions of both electrons in their
correspondingstates.Wenowwanttoask,howmucharetheelectronsinthis
state aware of each other? Do they correlate their motion in some way? One
way to measure correlation is to consider the quantity (x
1
x
2
) -(x
1
)(x
2
). If
electrons are completely unaware of each other, this quantity, called the
positionautocorrelation function is zero because then the average of the
productoftheirpositionmustdecomposetotheproductoftheaverage.Any
Election Bensity Functional Theoiy Page 26
Roi Baei

deviancefromzeroindicatessomedegreeofcorrelation.Forthetripletwave
functioninEq.(1.6.6)wefind:
(x
1
) = (
1,01
|x
1
|
1,01
) =
1
2
(
0

1
-
1

0
|x
1
|
0

1
-
1

0
)
=
1
2
((
0
|x
1
|
0
)(
1
|
1
) +(
1
|x
1
|
1
)(
0
|
0
)
-(
0
|x
1
|
1
)(
1
|
0
) -(
1
|x
1
|
0
)(
0
|
1
))
=
1
2
((
0
|x
1
|
0
) +(
1
|x
1
|
1
)) = u
(1.6.8)
Ofcourse,thesameresultwouldbeobtainedifwecalculate(x
2
)becausethe
electronsareequivalent.Furthermore:
(x
1
x
2
) = (
1,01
|x
1
x
2
|
1,01
) =
1
2
(
0

1
-
1

0
|x
1
x
2
|
0

1
-
1

0
)
=
1
2
((
0
|x
1
|
0
)(
1
|x
2
|
1
) +(
1
|x
2
|
1
)(
0
|x
2
|
0
)
-(
0
|x
1
|
1
)(
1
|x
2
|
0
) -(
1
|x
1
|
0
)(
0
|x
2
|
1
))
= -|(
0
|x|
1
)|
2
= -
1
2
l
p

(1.6.9)
This negative quantity is there because the Pauli principle pushes the
electrons to opposite sides (when one electron has positive x coordinate the
otherhasnegativeandviceversa).Letsseewhathappensinthesingletwave
function
S,00
.Heretoo(x
1
) = (x
2
) = u.Then:
(x
1
x
2
) = (
S,00
|x
1
x
2
|
S,00
) = (
0

0
|x
1
x
2
|
0

0
)
= (
0
|x
1
|
0
)(
0
|x
2
|
0
) = u
(1.6.10)
Thus, the singlet ground state shows no correlation. However, this does not
mean that all singlet wave functions of 2 electrons have no correlation.
Indeed,letusstudythesituationin
S,01
.Thedevelopmentisverysimilarto
Eq.(1.6.9),exceptthattheminussignisnowaplussignso:
Election Bensity Functional Theoiy Page 27
Roi Baei

(x
1
x
2
) = (
S,01
|x
1
x
2
|
S,01
) =
1
2
(
0

1
+
1

0
|x
1
x
2
|
0

1
+
1

0
)
= |(
0
|x|
1
)|
2
=
1
2
l
p

(1.6.11)
Herewefindpositiveautocorrelation,indicativeofthefactthatspinopposite
noninteracting pairs of electrons want to stick together (when one has
positive x the other wants to have as well and when one has negative the
otheronewantsnegativeaswell)likenoninteractingbosons.
Since there is no interaction between the electrons, the correlation in these
wave functions arises only from the Pauli principle, i.e. because we impose
the fact that electrons are Fermions. This is called Fermi correlation. Our
lessonisthis:
1) Wavefunctionsthataremereproductsofsingeparticleorbitalshavenocorrelation.
2) Iftheproductsaresymmetrizedlikeinthecaseoftheexcitedsingletthecorrelation
(x
1
x
2
)ispositiveindicatingthattheparticlesliketobetogetheri.e.bothonthe
rightorbothontheleftoftheorigin.
3) Iftheproductsareantisymmetrizedlikeinthecaseofthetripletthecorrelation
(x
1
x
2
)isnegativeindicatingthattheparticlesliketobeseperated.
Uptonow,weassumednoeeinteraction.Sonowletsincludeitandaddto
theHamiltoniananinteractionterm:
E

= b
`
(1) +b
`
(2) +I(x
1
, x
2
) (1.6.12)
Letustakeonecaseofcouplingwhichissimpleenoughtoyieldtoanalytical
analysis:
I(x
1
, x
2
) = py
2
x
1
x
2
(1.6.13)
With y
2
<
2
. This interaction seems strange at first site because it does not
depend on the distance between the particles, as we are used to from
electrostatics,yet,itdoesdescribearepulsion:sinceifx
1
andx
2
arebothlarge
and of the same sign this is energycostly; if they are both large and of
oppositesignthatlowersenergy.Inthiscase,theHamiltonianis:
Election Bensity Functional Theoiy Page 28
Roi Baei

=
p
1
2
2p
+
p
2
2
2p
+
1
2
p
2
x
1
2
+
1
2
p
2
x
2
2
+py
2
x
1
x
2
;
(1.6.14)
Exercise I1
FindtheeigenvaluesandeigenfunctionsofthisHamiltonian.
Solution: Define new coordinates X and x by:
X+x
2
= x
1
and

X-x
2
= x
2
. The
conjugatemomentaare:p
1
=
P+p
2
andp
2
=
P-p
2
(showthatthisisindeedsoby
calculating the commutation relations |P, X], |P, x] etc.). Then the new
Hamiltonianis:
E

=
(P +p)
2
4p
+
(P -p)
2
4p
+
1
2
p
2
_
X +x
2
]
2
+
1
2
p
2
_
X -x
2
]
2
+py
2
_
X +x
2
] _
X -x
2
]
(1.6.15)
Or,afterrearranging:
E

= _
P
2
2p
+
1
2
p(
2
+y
2
)X
2
_ +_
p
2
2p
+
1
2
p(
2
-y
2
)x
2
_ (1.6.16)
We see that X and x do not interact and each is a Harmonic oscillator on its
own. Let us define cos 0 = [
y
o

2
and
1
=
2
+y
2
= 2cos
0
2
and
2
=

2
-y
2
= 2sin
0
2
.Wefind:

nm
(X, x) =
n
(X;
1
, p)
m
(x;
2
, p),
E
nm
= _n +
1
2
] l
1
+_m+
1
2
] l
2

(1.6.17)
With
n
(z; , p) = N
n
c
-
q
2
2
E
n-1
(q), n = u,1, , q = _

l
z, E
n
(q) are Hermite
polynomialsofordernandN
n
areorthonormalizationconstants.
Election Bensity Functional Theoiy Page 29
Roi Baei

Exercise I2
Write the groundstate energy and wave function for the triplet state of the
system in the previous exercise. Determine the effect of interaction on the
energy by calculating r =
L
gs,y
L
gs,y=0
and on the correlation function c = (x
1
x
2
) -
(x
1
)(x
2
).
Solution:Weneedtoimposeuponthespatialwavefunctionof(1.6.17)tobe
antisymmetric. The two variables X =
x
1
+x
2
2
and x =
x
1
-x
2
2
are respectively
symmetric and antisymmetric combinations of the positions of the electrons.
SincexistheantisymmetriccombinationwerequiretheHermitepolynomial
inx tobeodd.Thelowestantisymmetricstateisthecombinationn = uforX
andm = 1forx:

01
(X, x) = N
01
c
-

1
X
2
2l
-

2
x
2
2l
x
E
gs
=
1
2
l
1
+
S
2
l
2
=
1
2
l 2 _cos

2
+Ssin

2
]
(1.6.18)
Now,letuswritethewavefunctionofEq.(1.6.18)intermsofx
1
andx
2
:

01
(x
1
, x
2
) = N
01
c
-

1
4l
(x
1
+x
2
)
2
-

2
4l
(x
1
-x
2
)
2 (x
1
-x
2
)
2
=
N
01
2
c
-
(
1
+
2
)
4l
(x
1
2
+x
2
2
)
c
-
(
1
-
2
)
2l
x
1
x
2
(x
1
-x
2
)
(1.6.19)
Onecancomparetheeffectoftheinteractionbylookingattheratiobetween
groundstateenergiesofthesystemwithandwithoutinteractionasafunction
of0,theinteractionstrength:
r =
E
gs,y
E
gs
=
1
2

1
+
3
2

2
2
=
cos

2
+Ssin

2
22

(1.6.20)
TheresultisshowninFigure I2(leftpanel).
Election Bensity Functional Theoiy Page Su
Roi Baei

Figure I2:(Leftpanel)Theratior = F
12,y
F
12
;(Rightpanel)thecorrelationratioc(6)
c [
a
2
vsinteractionstrengthsus 6 = [
y
m

2

Theinteractionlowerstheenergy,becausenowthewavefunctioncanacquire
astructurethatpromotestheelectronsbeingawayfromeachother.Thusone
ispushedtowardsthe+xdirectionandtheothertowardsthatof-xandthus
they acquire a large negative value of x
1
x
2
. To see this note that the
expectation values of x and X are both zero and therefore (x
1
) and (x
2
) are
zero as well. Furthermore note that X
2
-x
2
= 2x
1
x
2
and the autocorrelation
is:
c(0) = (x
1
x
2
) -(x
1
)(x
2
) = (x
1
x
2
) =
1
2
((X
2
) -(x
2
))
(1.6.21)
The expectation value of the square position in harmonic oscillator is easily
obtained using HellmanFeynman theorem (rederive Eq. (1.4.16) in terms of
anyparameterdependentHamiltonian,notnecessarilythetimet):
E = I +
1
2
p
2
y
2
- _n +
1
2
] l =
JE
n
J
= p ( y
2
)
n

(1.6.22)
So:(X
2
)
0
=
1
2
l

1
andsimilarly(x
2
)
1
=
3
2
l

2
.Thus:
0.0 0.1 0.2 0.3 0.4 0.5
0.0
0.2
0.4
0.6
0.8
1.0
q
p
r
0.0 0.1 0.2 0.3 0.4 0.5
0
5
10
15
q
p
c
@
q
D

c
@
p

2
D
Election Bensity Functional Theoiy Page S1
Roi Baei

c(0) = (x
1
x
2
) -(x
1
)(x
2
) = (x
1
x
2
) =
1
2
(X
2
-x
2
) =
1
2
l
p
_
1
2
1

1
-
S
2
1

2
]
=
l
42p
_
1
cos

2
-
S
sin

2
_
(1.6.23)
Fornointeraction(0 =
n
2
),thereisonlycorrelationduetothePauliprinciple:
sincebothelectronshavespinuptheycannotoccupythesamepointinspace)
Wesee,asderivedabove:
c [
n
2
= Feimi Coiielation = -
l
2p
(1.6.24)
The ratio between full correlations and Fermi correlation is shown in Figure
I2(rightpanel).Thecorrelationisnegativeandlarger(inabsolutevalue)than
themerePaulicorrelation.Indeed,theinteractionpusheselectronsawayfrom
eachother.
LookingatthewavefunctionofEq.(1.6.19),itisevidentthatbecause
1
=
2

there is no way to write this as an antisymmetrized sum of products of 1


electon functions. From these exercises with harmonic oscillator systems we
find that the issue of correlation can be quiet complicated. In realistic
electronic systems, when the interaction is Coulombic, not Harmonic, the
situationisevenmorecomplicatedbecauseofthelackofanalyticalsolutions.
F. The electron density is a much simpler object
than the wave function
The complexity of the wave function is overwhelming. It includes all the
informationwecanhaveonthemoleculeinacertainstate.However,allthese
intricaciesanddetailsareoftenuninterestingforus:inmanycases,wesimply
have no direct use for them. Take, for instance, the electronic energy our
primary target in the BornOppenheimer picture. It depends only on the
Election Bensity Functional Theoiy Page S2
Roi Baei

relative distance of pairs of particles. This is because the electronelectron


Coulombrepulsionisapairwiseinteraction.
One interesting quantity, besides energy, that can be extracted from the
electronicwavefunctionistheelectrondensity ( ) n r
.This3Dfunctiontellsus
the expectation value of the density of electrons. That is
( )
3
n d r r is the
expectation value of the number of electrons at a small volume
3
d r around
point r .Thus,wecanwrite:
n(r) = (|n(r)|) (1.7.1)
Weusethenotation:
(|) _
-
(r
1
, r
2
, , r
N
c
)(r
1
, r
2
, , r
N
c
)J
3
r
1
J
3
r
2
J
3
r
N
c
(1.7.2)
Heren(r)

istheoperatorcorrespondingtotheelectronnumberdensity.Since
electronsarepointparticles,andthepositionoperatorforthei
th
electronisr

thisoperatorisdefinedby:
n(r) = o(r

-r)
N
c
=1

(1.7.3)
Weusedthedefinitionofaofunction,accordingtowhich:
_o(r
1
-r)(r
1
)J
3
r
1
= (r) (1.7.4)
Thefunctiono(r

-r)isthedensityofelectroniatr.
Exercise I3
Calculatetheexpectationvalueoftheelectronnumberdensityandshowthat:
n(r) = N
c
_(r, r
2
, , r
N
c
) (r, r
2
, , r
N
c
)J
3
r
2
J
3
r
N
c
(1.7.5)
Solution:Plugging(1.7.3)into(1.7.1)gives:
Election Bensity Functional Theoiy Page SS
Roi Baei

n(r) = _(r
1
, r
2
, , r
N
c
) n(r)(r
1
, r
2
, , r
N
c
)J
3
r
1
J
3
r
2
J
3
r
N
c
= _(r
1
, r
2
, , r
N
c
) o(r

N
c
=1
-r)(r
1
, r
2
, , r
N
c
)J
3
r
1
J
3
r
N
c
=
= _(r
1
, r
2
, , r
N
c
) o(r
1
N
c
=1
-r)(r
1
, r
2
, , r
N
c
)J
3
r
1
J
3
r
N
c

(1.7.6)
The laststep stems from thePauliprinciple:allelectrons areidentical, sowe
canreplaceelectronr

byelectronr
1
.Thesumisnowoveridenticalnumbers
soitbecomesameremultiplicationasinEq.(1.7.5).
Looking at Eq. (1.7.5), we see that n(r)

involves integrating out a huge


amount of wavefunction details. It is as if only the data concerning the
density distribution of a single electron remains! This is multiplied by N
c
in
Eq. (1.7.5)so n(r) accounts forthe combined densityofallelectrons. Indeed,
integratingovertheentirespace,oneobtainsfrom(1.7.5):
_n(r)J
3
r = N
c
(1.7.7)
ExpressingthefactthatthetotalnumberofelectronsisN
c
.
Exercise I4
Calculatethe1DelectrondensityofthetripletgroundstatefromExercise I2.
Solution:IfthewavefunctionofEq.(1.6.6)istakenthen:
Election Bensity Functional Theoiy Page S4
Roi Baei

n(x) = N_ c
-
n[x
2
+x
2
2

l (x
2
-x)
2
Jx

-
= Nc
-
nx
2
l
__ c
-
nx
2
2
l
x
2
2
Jx
2

-
+x
2
_ c
-
nx
2
2
l
Jx
2

-
_
= N_
ln
p
_ c
-
nx
2
2
l
Jx
2

-
|(x
2
) +x
2
]
= Nc
-
nx
2
l
_
ln
p
_
l
2p
+x
2
_
(1.7.8)
WechooseNtoensurethat]n(r)J
3
r = 2:
n(x) = _
p
nl
c
-
nx
2
l
_1 +
2p
l
x
2
_
(1.7.9)
Definingtheaveragefrequency:
1

=
1
2
_
1

1
+
1

2
] =
1
2
cos
0
2
+sin
0
2
sin0

(1.7.10)
WefindthedensityofthestateinEq.(1.6.19)is:
n(x) = N_ c
-

1
(x+x2)
2
2
-

2
(x-x
2
)
2
2
(x -x
2
)
2
Jx
2

-
= _

3
n
1
2
c
-x
2
(1 +2(2
1
-)x
2
)
(1.7.11)
G. The variational principle
When we look for the groundstate energy of a complicated system, with
HamiltonianE

,thevariationalprincipleisoftenextremelyimportant.Itsays,
quite simply that the groundstate energy is a minimum of a functional of
wavefunctions.Thefunctionalis:
E|] =
(|E

|)
(|)

(1.8.1)
Election Bensity Functional Theoiy Page SS
Roi Baei

Andthevariationalprinciplestatesthat:
E
gs
E|
gs
] E|] (1.8.2)
Forall.Whatthisprincipleallowsusistodeterminewhichofthetwowave
functions
1

and
2
gives an energy closer to E
gs
, without knowing E
gs
.
Simpletheonewithlowerenergyiscloser.
We can build a parameterized family of functions
y
and determine the
optimizedparameterastheonethatgivesminimumtoE(y) E|
y
].
Exercise I5
Consider the quantum groundstate problem for a particle in a well :(x) =
1
4
kx
4
.Weconsiderthefamilyoffunctions:

c
(x) =
c
-
x
2
4o
2

2no

(1.8.3)
Here isapositiveparameter.Whatisthebestfunctionforrepresentingthe
groundstate?
Solution:Thefunctionsarenormalized.TheenergyisE

= I
`
+I
`
.Wehave:
E(o) = E|
c
] = (
c
|I
`
|
c
) +(
c
I
`

c
) =
l
2
8mo
2
+
S
4
ko
4

(1.8.4)
Whereweusedthefact:
(
c
|I
`
|
c
) = -
l
2
2m
c
_
c
(x)
c
ii
(x)Jx

-
=
l
2
8m
c
o
2
(1.8.5)
Thustheminimumenergyisobtainedfrom:
u = E
i
(o
ii
) = -
l
2
4m
2
o
ii3
+Sko
-
3
- o
-
= _
l
2
12km
_
16

(1.8.6)
Andso:
Election Bensity Functional Theoiy Page S6
Roi Baei

E
-
=
l
2
8m[
l
2
12km

13
+
S
4
k _
l
2
12km
_
23
=
Sl
2
16m
1
o
-2

(1.8.7)
H. Dilation relations
Considerthenormalized(x).Fromit,wecandefineafamilyofnormalized
wavefunctions:

y
(x) = y
12
(yx) (1.9.1)
Wecancheckthateachmemberisindeednormalized:
(
y
|
y
) = _|
y
(x)|
2
Jx = y _|(yx)|
2
Jx = _|(y)|
2
Jy (1.9.2)
This operation of stretching the argument of a function is called dilation.
Dilationaffectsfunctionals.Considerthekineticenergyfunctional:
I|] (|I
`
|) = _(x) _-
1
2
J
2
Jx
2
_(x)Jx (1.9.3)
Exercise I6
ProvethedilationrelationofI|
y
]andI|]:
I|
y
] = y
2
I|] (1.9.4)
Solution:Weshowthatthisisasimplerelation:
I|
y
] _
y
(x) _-
1
2
J
2
Jx
2
_
y
(x)Jx
= y _(yx) _-
1
2
J
2
Jx
2
_(yx)Jx
= y
2
_(y) _-
1
2
J
2
Jy
2
_(y)Jy = y
2
I|]
(1.9.5)
Thepotentialbetweenallparticlesiscalledhomogeneousofordernif:
I(yx
1
, , yx
N
) = y
n
I(x
1
, , x
N
) (1.9.6)
Election Bensity Functional Theoiy Page S7
Roi Baei

Examplesaren = 2isforHarmonicpotentialwellsandharmonicinteractions
andn = -1isfortheCoulombpotentialwellsandCoulombicinteractions.
Exercise I7
Provethedilationrelationforhomogeneouspotentialsofordern:
I|
y
] = y
-n
I|] (1.9.7)
Solution:Forsuchapotential,
I|
y
] = _y
N
|
y
(yx
1
, , yx
N
)|
2
I(x
1
, , x
N
)Jx
1
Jx
N
=
= _|
y
(y
1
, , y
N
)|
2
I _
y
1
y
, ,
y
1
y
] Jy
1
Jy
N
= y
-n
_|
y
(y
1
, , y
N
)|
2
I(y
1
, , y
N
)Jy
1
Jy
N
= y
-n
I|
1
].
(1.9.8)
WecombinetheresultsfromEqs.(1.9.4)and(1.9.7)andobtainaninteresting
propertyofthetotalenergyforsystemswithhomogeneousinteractions:
E|
y
] = I|
y
] +I|
y
] = y
2
I|] +y
-n
I|] (1.9.9)
For a molecule, the interaction between the particles (electrons and nuclei) is
the Coulomb potential I(r; R) I( r
1
, , r
N
c
; R
1
, , R
N
N
) which is
homogeneousofordern = -1onefindstheenergyofamoleculeobeys:
E|
y
] = I|
y
] +I|
y
] = y
2
I|] +yI|] (Coulomb) (1.9.10)
i. The concept of the virial in classical mechanics
First, let us define the virial. For a system with coordinate q
n
collectively
denoted as q the virial in classical mechanics is the time average of q F
whereFistheforcevector:
:iriol = (q F) =
1

_ q(t
i
) F(t
i
)Jt
i
t+

2
t-

Election Bensity Functional Theoiy Page S8


Roi Baei

Itcanbeshown,thatforboundsystems:[5]
:iriol = (q F) = -2(I)
Forconservativesystemstheforceisagradientofthepotential,F = -v:(q).
Theviralrelatestodilationofthecoordinatesthrough:
J
Jy
:
y
(q) = q v:(yq) = q v:
y
(q). (1.9.11)
Forhomogeneouspotentialswehave:y
d
dy
:
y
(q) = ny
n
:(q) = n:
y
(q),thus:
n:
y
(q) = yq v:
y
(q) (: bomogcncous orJcr n) (1.9.12)
Inparticular,fory = 1:
n:(q) = q v:(q) (: bomogcncous orJcr n) (1.9.13)
WewillespeciallybeinterestedinCoulombsystems,wheren = -1,then:
:(q) +q v:(q) = u (: bomogcncous orJcr n = -1) (1.9.14)
Exercise I8
Eq. (1.9.13) is known as Eulers equation, after its discoverer. In
Thermodynamics it is extremely useful. Thermodynamics stipulates that if
you know the energy of a system u(S, I, N

) as a function of its macroscopic


parameters S, I, N

then you have complete thermodynamics knowledge. The


firstandsecondlawsofthermodynamicsstatethat:
Ju = IJS -pJI +p JN

,
(1.9.15)
where I, p anu p are respectively the temperature, the pressure and the
chemicalpotentials.Asecondstipulationisthatuisahomogeneousfunction
oforder1.
Fromthis,showthatforallthermodynamicalsystems
1. u = IS - PI + p N

Election Bensity Functional Theoiy Page S9


Roi Baei

2. [
p
1

=
S
v
.
Forhomogeneouspotentialstheviraltheorembecomes:
:iriol = -n(:) = -2(I) (1.9.16)
ForCoulombsystemsn = -1:
(:) = -2(I) (1.9.17)
Fromwhich:
E = (:) +(I) = -(I) =
(:)
2

(1.9.18)
Wenowshowthatthisrelationholdsalsoinquantummechanics.
ii. The Virial Theorem in quantum mechanics
Now, if is the groundstate energy then E|
y=1
] obtains the minimum and
therefore:
J
Jy
E|
y
] = u (1.9.19)
Pluggingthisinto(1.9.10),oneobtains:
u = 2yI|] -ny
-n-1
I|] (1.9.20)
Sinceweknowthattheoptimalvalueofyis1,then(droppingthe|]symbol
and remembering that all following quantities are ground state expectation
value:
2I = nI
E = I +I = _1 +
2
n
] I = [
n
2
+1 I
(1.9.21)
These relations show that the virial theorem (Eq. (1.9.18) holds in quantum
mechanicsprovidedisthefullmoleculareigenstate.Formoleculesn = -1:
andonehas:
Election Bensity Functional Theoiy Page 4u
Roi Baei

2I = -I (Coulomb)
E = I +I = -I =
1
2
I
(1.9.22)
A subtle issue: all energies and potentials in the above expressions are
absolute. We usually give energy and potentials only to within an additive
constant. However, the fact that the potential is homogenous, it cannot
tolerate addition of a constant (x
2
is homogeneous of order 2 but x
2
+o is
not).
iii. Slaters theory of the chemical bond using the viral theorem
What happens when is the electronic eigenstate in the Born Oppenheimer
approximation? If we look only at the electronic wave function we do not
expect Eq. (1.9.21) to be valid. Indeed, using E

c
= I
`
+I(r; R) where I
`
=
[-
1
2
v

2
andI(r; R) =
-z
|r-R|
(withobvioussummationonnucleiandelectrons)
E

c
|R]|R](r) = E|R]|R](r) (1.9.23)
Upondilation
y
|R](r) = y
3N
c
2
|R](yr)andnotethatRisnotdilated.Then
wehavefortheKEandpotential:
I
y
|R] (
y
|I
`
|
y
) = _
y
|R](r) _-
1
2
v

2
]
y
|R](r)J
3N
c
r
= y
3N
c
_|R](yr) _-
1
2
v

2
] |R](yr)J
3N
c
r
= y
2
_|R](x) _-
1
2
v
s
2
] |R](x)J
3N
c
s = y
2
I|R]
(1.9.24)
Andwedefine:
Election Bensity Functional Theoiy Page 41
Roi Baei

w
y
|R, R
i
] (
y
|R]|I
`
|R
i
]|
y
|R]) = _
y
(r, R)
2
I(r; R
i
)J
3N
c
r
= y
3N
c
_(yr; R)
2
I(r, R
i
)J
3N
c
r
= _(x, R)
2
I(y
-1
x; R
i
)J
3N
c
s
= y _(x, R)
2
I(x; yR
i
)J
3N
c
s = yw|R, yR
i
]
(1.9.25)
We have defined w, a two R quantity. Of course, the physical interaction
potentialisI|R] = w|R, R].Thereasonwedefinedwthiswayisthatwecan
nowwrite:
E
y
|R] = y
2
I
c
|R] +yw|R, yR] (1.9.26)
NotethatweinsertedthesubscriptctotheKEsincenowitisonlytheKEof
theelectrons.
Onestillhasthevariationalprinciple,i.e.JE
y
|R]Jy|
y
-
=1
= u(Eq.(1.9.19)):
u = 2yI
c
|R] +w|R, yR] +y(R v
R
|w)|R, R
i
]
R
|
=yR
(1.9.27)
Here we used the fact that
d
dy
(yR) = |R v](yR). Putting in the
solutiony
-
= 1wefind:

u = 2I
c
|R] +I|R] +R v
R
|w|R, R
i
]
R=R
| (1.9.28)
From theHellmannFeynmantheorem (see sectionXXX)v
R
|(w|R, R
i
])
R
|
=R
=
v
R
E(R). Adding to the electronic energy E the nuclear potential energy
E
N
=
1
2

z
I
z
I
|
|R
I
-R
I
|
|
I=I
| gives the BO energy E
B0
= E +E
N
and since the nuclear
energyishomogeneousoforder1,wecanuseEq.(1.9.14)and:
u = E
N
+R
]
v
R
]
]
E
N
,
(1.9.29)
Pluggingallthisintoeq.(1.9.28)wefindSlatersformula[6]:
Election Bensity Functional Theoiy Page 42
Roi Baei

I
c
= -E
B0
-R v
R
E
B0
(1.9.30)
Fromthis,usingE
B0
(R) = I
c
+I
tot
(whereI
tot
= I +E
N
isthetotalPEofthe
molecule) wefindalso:
I
tot
= 2E
B0
+R v
R
E
B0
(1.9.31)
We point out an interesting property of Eq. (1.9.30) is that we can obtain
electronic properties such as kinetic energy I
c
or potential energy I directly
fromthepotentialsurfaceE
B0
.Thiswillbeimportantforanalyzingproperties
of DFT quantities. Note that at stationary points on the Born Oppenheimer
PESwherev
R
E
B0
=0theusualvirialrelationeq.(1.9.21)holds(withneglectof
nuclearkineticenergies).
Slater derived Eq. (1.9.30) in a different manner, not from variational or
dilation considerations. He used it for analyzing the general nature of
chemical bonding of a diatomic molecule. We follow in his footspteps here.
Consideradiatomicmolecule.SupposetheBOcurveisgiveasapotentialof
thefollowingform,intermsoftheinternucleardistanceR:
E
B0
(R) =
A
R
m
-
B
R
n
+E

(1.9.32)
WhereAandBarepositiveconstantsandtypicallym > n.Forexample,inthe
case the PE curve is given by the LennardJones potential then m = 12 and
n = 6. This potential describes longrange attraction and shortrange
repulsion.Ithasaminimumat:
R
Mn
= _
Bn
Am
]
1
m+n

(1.9.33)
Onethenhas:
Election Bensity Functional Theoiy Page 4S
Roi Baei

I
c
=
B(1 -n)
R
n
-
A(1 -m)
R
m
-E

(1.9.34)
When R - we have: I
c
= -
B(n-1)
R
n
-E

. Let us consider what happens to


the KE as the distance R is reduced. Clearly it initially descends, however, at
somedistanceR
1
itreachesaminimum,andbelowthisdistanceitbeginsto
increase.Since:
J
JR
I
c
= -
Bn(1 -n)
R
n+1
+
Am(1 -m)
R
m+1

(1.9.35)
We find the minimum kinetic energy is obtained at a distance somewhat
largerthanR
Mn
:
R
1
= _
Am(m-1)
Bn(n -1)
_
1
m-n
= _
m-1
n -1
]
1
m-n
R
Mn
> R
Mn

(1.9.36)
Forexample,form = 12,n = 6wefind: R
1
= 1.14R
Mn
.Thesumofelectronic
andnuclearpotentialenergyis:
I
tot
=
A(2 -m)
R
m
-
B(2 -n)
R
n
+2E

(1.9.37)
At large R, I +E
N
actually increases as the atoms approach. The potential
energy does not rise indefinitely. At some internuclear distance, larger than
the distance at which the BO potential obtains its minimum, it reaches a
maximumandthenstartsdecreasing.Wehave:
Am(2 - m)
R
m+1
=
Bn(2 -n)
R
n+1

R
v
= _
Am(m -2)
Bn(n -2)
_
1
m-n
= _
m-2
n -2
]
1
m-n
R
Mn
> R
Mn

(1.9.38)
For the Lennard Jones potential we have: R
v
= 1.16R
Mn
. For R < R
1
we find
that the kinetic energy grows sharply as R decreases, while the potential
Election Bensity Functional Theoiy Page 44
Roi Baei

energydecreases.Thissharpbehaviorcanbeinterpretedasacondensationof
theelectroncloudaroundeachnucleiassuchacondensationcausesincrease
inkineticenergyIanddecreaseinelectronicpotentialenergyI.ThetotalBO
energycontinuestodropaslongasR > R
Mn
andthenstartsrising.
As an application of this theory consider the X
+
system where X is a
neutral atom and
+
a distant atomic ion (can also be a molecular ion). The
energyatlargedistanceRisthatofEq.(1.9.32)withA = uandB =
u(X)
2
where
o(X) is the polarizability of X and n = 4. This form of the potential surface
resultsfromasimpleexerciseinelectrostatics.Quantummechanicsisneeded
onlyforcalculatingthevalueofthepolarizabilityofX,o(X).Slatersanalysis
statesthat:
I
c
= -
So(X)
2R
4
-E

I
tot
=
o(X)
R
4
+2E

(1.9.39)
Sotheforceduetothetotalpotentialenergyisrepulsivewhilethatduetothe
KE is attractive. The KE attraction outweighs the PE repulsion by a factor of
threehalves.Slatersanalysisshowsthattheinteractionhereisfundamentally
different from that of electrostatics, where the force attributed entirely to the
Coulombicattraction.
Anotherapplicationistothesimplestchemicalbond,thatoftheE
2
+
molecule.
Thetextbookanalysisofthismoleculeassumesthatthemolecularorbitalisa
bondinglinearcombinationofthetwoatomicorbitals(LCAO),bothhaving
the 1s shape:
1s
(r) = c
-
r
c
0
, but each centered on its corresponding nucleus.
Quantitativelythispictureisveryinaccurate.Whileitdoesshowbinding,its
predicted a bond length is almost 1.4 (while the true bond length is 1.06)
Election Bensity Functional Theoiy Page 4S
Roi Baei

anditsestimateofatomizationenergyis1.7eVwhiletheexperimentalvalue
is close to 2.7eV. There are no electron correlation effects here, so the only
sourceofinaccuracyisthelimitedefficacyofthesmallbasis.

Figure I3: Theoptimal dilation parameter


-
asa functionoftheinternuclear distanceR
in H
2
+
, where the wave function is the sum of 2 1stype orbitals (localized on each of the
twonuclei).
Onecanaddadilationfactortothewavefunction
1s

(r) = c
-
{r
c
0
andperform
a variational determination of . Then, one can obtain, for each value of the
internuclear distance the optimal , denoted
-
(R). At R - the dilation
parameter should be equal to 1, as then the 1s orbital is exact. At R - u the
moleculebecomestheHeionand
-
shouldapproachtheexactvalueof2.See
Figure I3. But the optimal value
-
is not monotonic! At large R it becomes
smaller than 1, indicating that the electron cloud expands, the kinetic energy
drops and the potential energy rises. Thus longrange attraction is driven by
the kinetic energy. At R = S.2o
0
the value of
-
is again 1 and then grows
monotonicallyasRisreduced.Nowtheorbitaliscontracting,kineticenergyis
growingandpotentialenergyisdecreasing.Inthisrange,chemicalbindingis
driven by the potential energies. At the distance minimizing the BO energy
the value of is:
-
= 1.2S, i.e. the orbitals are then considerably contracted
(relative to the free H atom). With such contracted orbitals, the calculated
bond length of the LCAO calculation is 1.06. The bonu eneigy, computeu by
Election Bensity Functional Theoiy Page 46
Roi Baei

subtiacting the calculateu B


2
+
energyfromthatofthehydrogenatomgivesthe
bondenergyof2.35eV.Thefirsttoreportthistypeofcalculationisinref[7],a
clear exposition of the calculation is given in refs [8]. These improved
theoretical estimates, with a twofunction basis, achieved with just a single
dilationparameter,showthatorbitalshapesarecrucialforagooddescription
ofchemicalbonding.
This result contradicts our intuition regarding the behavior of electrons in a
covalentbond.ItseemsthatamajorsourceofthechemicalbondenergyinB
2
+

isnotduetotheelectrostaticbenefitresultingfromtheplacementofelectrons
in the middle region separating the two nuclei. Rather, it is due to the
contraction of the electron wave function near each of the nuclei induced by
change in Coulomb potential near that nucleus due to the presence of the
other nuclei. Around one nucleus of B
2
+
the nuclear Coulombic potential
becomes deeper and distorted towards the second nucleus. As a result, a
contraction of the electron density around the nucleus is obtained, and the
chargeplacedbetweentheatomsisnotinthemidwayregionofthebond:it
is between them alright but very close to each nucleus. Each nuclei seems to
share the electron not by placing it midway between them, but rather by
havingtheelectronsmuchclosertoeachnucleus.
Combiningthiswiththeaboveresult,wefindaverygeneralconclusion:
Rule 1 of chemical attraction:As atoms approach from afar, (for R > R
1
) the
kineticenergydecreasesandthepotentialenergy increasesand thus itisthe
loweringofkineticenergywhichisresponsibleforthedistantattractionbetween
molecules.
Rule 2 of chemical bonding: For distances near but larger than the bonding
distance the attraction is more subtle. The electronic kinetic energy rises
Election Bensity Functional Theoiy Page 47
Roi Baei

sharply and the potential energy drops even more sharply. Thus, a major
source of energy is the contraction of orbitals around their respective nuclei,
inspired by the deepening of potential wells due to the presence of closeby
nuclei.TheelectronicPEisthusthemajorgluewhenatomsareclosetoeach
other: It has to offset the nuclear repulsion and the repulsion due to kinetic
energyofelectrons.
iv. Other Dilation Facts
ConsidertheSchrdingerequation:
-
l
2
2m

ii
(x) +:(x)(x) = E(x)
(1.9.40)
We now dilate: we assume
x
(x) = (zx) is an eigenstate and find the
correspondingHamiltonian.Notethat
x
ii
(x) = z
2

ii
(zx)thus:
-
l
2
2mz
2

x
ii
(x) +:(zx)
x
(x) = E
x
(x)
(1.9.41)
WeseethattherequiredHamiltonianinvolvesadilatedpotentialandascalled
mass by a factor z
2
.The energy is left intact. Alternatively, we can multiply
throughbythesquareofthedilationfactorz
2
andobtain:
-
l
2
2m

x
ii
(x) +z
2
:(zx)
x
(x) = z
2
E
x
(x)
(1.9.42)
Now the potential is dilated and scaled: :(x) - z
2
:(zx) when then the wave
function is dilated by (x) - (zx) and the total energy is also scaled by
E - z
2
E.ThisisgeneralforbasicallyanySchrdingerequationsandholdsfor
anynumberofspatialdimensions.
Nowconsiderhomogeneouspotentialsin3D::(zr) = z
n
:(r).Inthiscasethe
potential transform is indeed a multiplicative scaling: :(r) - z
2+n
:(r). In
Election Bensity Functional Theoiy Page 48
Roi Baei

particular, in Coulomb systems, when n = -1, we find that the potential


scalingissimplybyafactorz.Wecanturnaroundthediscussionandstate:
If the Coulomb coupling constant is scaled by a factor z then the total
energy is scaled by a factor z
2
and the wave function is dilated by a
factor z. In particular, the density is dilated and scaled as n(r) -
z
3
n(zr).
Accordingtothevirialtheorem,thepotentialenergyandthetotalenergyare
relatedby
v
2
= -I = E,thustheexpectationvalueofthepotentialandkinetic
energiesisscaledbyafactorz
2
aswell.
Exercise I9
Anapplicationoftheabovediscussionallowsaninterestingexactconclusion
concerningthegroundstateenergyperparticleofthehomogeneouselectron
gasofdensityn(seesectionXXXfordetails).Denotethisquantity,e
HLu
(n, c
2
)
andtheCoulombcouplingconstantc
2
4ne
0
.Showthat:
Sn
oe
HLu
on
+c
2
oe
HLu
oc
2
= 2e
HLu

(1.9.43)
Hint:Startfrome
HLu
(z
3
n, c
2
) =
2
e
HLu
(n, c
2
).Thentakethederivativewith
respectto z.Thensetz = 1.
I. Introduction to Functionals
i. Definition of a functional and some examples
A formula that accepts a function and maps it to a number is called a
functional. We denote the number a functional F maps to a function n(r) by
this by the symbol F|n]. Take for example the simplest functional: F|n] = u.
This functional maps any function n(r) to the number zero. A more
interestingfunctionalis:
Election Bensity Functional Theoiy Page 49
Roi Baei

F
r

|n] = n(r

)
(1.10.1)
This functional maps to each function n(r) its value at a certain point r
0
.
Next,considerthefunctional
I
v
|n] =
1
I
_n(r)J
3
r
v
, (1.10.2)
mappingeachfunctionitsaveragevalueinagivenvolumeI.
Another familiar example is the kinetic energy functional for a particle of
masspinsomewavefunction(r):
I|] = -
l
2
2p
_
-
(r)v
2
(r)J
3
r
v
, (1.10.3)
OrtheenergyforagivenpotentialI(r):
E|] = I|] +I|], (1.10.4)
Where:
I|] = _I(r)|(r)|
2
J
3
r, (1.10.5)
ii. Linear functionals
A functional is called linear if for any two functions n(r) and m(r) one has:
F|n(r) +m(r)] = F|n(r)] +F|m(r)] and if F|on(r)] = oF|n(r)] for any
number o. The functionals above F
r

|n] and I
v
|n] are examples of such
functionals,pleasecheck.Anylinearfunctionalcanbewrittenintheform:
F|n] = _
L
(r)n(r)J
3
r. (1.10.6)
Thefunction
L
(r)iscalledthelinearresponsekernelofF.Forexample,the
linearresponsekernelofF
r

is
L
(r) = o(r -r
0
).AndthatofI
v
is
L
(r) =
1
v
.
Election Bensity Functional Theoiy Page Su
Roi Baei

iii. Functional Derivatives


Let us now introduce the concept of a functional derivative. Consider the
functional F
r

|n] of Eq. (1.10.1). A functional derivative, denoted


6P
6n(r)

measures how F|n] changes when we make a small localized change of n(r)
atr.
Let us give a more detailed definition. Suppose we change the argument
functioninF|n]bymakinganarbitrarilysmallchangein(r) -n(r) +e(r).
We use an arbitrary function (r) multiplied by an infinitesimal number e
(thatis,anumbereweintendtomakevanishinglysmallatalaterstatge).In
principle, for any finite e the change
6P
6n(r)
in F has a very complicated
dependence on e. But, since e is assumed small, we can Taylor expand with
respect to e and then throw out all terms beyond the term linear in e.
This makes sense since we assume that e is a small as we want. Thus, the
linearpartofoFisafunctionalofeanditisalinearfunctional.Wedenote
the linear response kernel of this functional by the symbol
6P
6n(r)
, called the
functionalderivative.Thereforewecanwrite:
oF F|n +e] -F|n] = _
oF
on(r)
e(r)J
3
r +0(e
2
) (1.10.7)
Note, that this the linear response kernel
6P
6n(r)
is independent of (r) and e.
Onetechniquetoobtainthekernelistaking(r) = o
r
(r) = o(r -r).Tomake
thisnotionprecise,wedefine:
oF
on(r)
lim
c-0
F|n +eo
r
] -F|n]
e
(1.10.8)
Where o
r
= o(r -r) is a delta function, expressing the fact that we make a
localizedchangeinn(r)at r .
Election Bensity Functional Theoiy Page S1
Roi Baei

Examples
Apply definition (1.10.8) to F
r
0
|n] , and I
v
|n]. Find the functional derivatives
andcheckthatEq.(1.10.7)holds.
Solution:
oF
r

on(r)
lim
c-0
F
r

|n +eo
r
] -F
r

|n]
e
lim
c-0
(n(r

) +eo
r
(r

)) - n(r

)
e

= o
r
(r

) = o(r

-r)
(1.10.9)
The functional derivative comes out o
r
(r) = o(r -r), meaning that any
change in n(r) made at r will not affect F
r
0
|n], unless it is made at r = r
0
.
Indeed,ifwemakeanarbitrarysmallchangen - e,wehavefrom(1.10.7):
oF
r

F
r

|n +eo
r
] -F
r

|n] = (n(r

) +eo
r
(r

)) -n(r

) = o
r
(r

)
= o(r

-r)
(1.10.10)
Andthisisseentobeindeedthecasedirectlyfromthedefinitionof
0
F
r
inEq.
(1.10.1).
AsforI
v
|n] =
1
v
] n(r)J
3
r
v
:
oI
v
on(r)
lim
c-0
I
v
|n +eo
r
] -I
v
|n]
e
lim
c-0
I
-1
] (n(r) +eo
r
(r))J
3
r
v
-I
-1
] n(r)J
3
r
v
e

= I
-1
_o
r
(r) J
3
r = I
-1

(1.10.11)
AnotherExample
An important functional, we will use is the Hartree energy, the classical
repulsionenergyassociatedwithachargedistributionn(r):
E
H
|n] =
1
2

n(r)n(r
i
)
|r -r
i
|
J
3
rJ
3
r' (1.10.12)
Election Bensity Functional Theoiy Page S2
Roi Baei

Prove that the functionalderivativeof theHartree Energy,calledtheHartree


potentialistheelectrostaticpotentialofthatdensityofelectrons:
:
H
(r)
oE
H
on(r)
= lim
c-0
1
2
__
(n(r)+c6
r
(r))[n(r
|
)+c6
r
(ri)
|r-r
|
|
-
n(r)n(r
|
)
|r-r
|
|
_J
3
rJ
3
r'
e
= lim
c-0
1
2
_
[c6
r
(r)n(r
|
)+n(r)c6
r
(ri)
|r-r
|
|
J
3
rJ
3
r'
e

= _
n(r
i
)
|r -r
i
|
J
3
r'
(1.10.13)
Thus,wefindasrequestedthat:
H
(r)isineedtheelectrostaticpotentialofthe
densityn(r).
Two shortcuts: when functional integrating, one can often use regular
differentiationrules.Alloneneedstorememberisthefollowingshortcuts:
on(x)
on(x
i
)
- o(x -x
i
)
on'(x)
on(x
i
)
- o'(x -x
i
) (1.10.14)
And use them within chain rules:
6](n(x))
6n(x
|
)
-
i
(n(x))o(x -x
i
) (where

i
(n)
d
dn
(n)).Anexampleoftheuseofthisruleisthepreviousexercise:
oI
v
on(r)
I
-1
_
on(r)
on(r)
J
3
r = I
-1
_o(r -r) J
3
r = I
-1
(1.10.15)
Anotherexampleisthefollowingfunctional,relatedtothekineticenergy:
[|] = _
i
(x)
2
Jx

-
(1.10.16)
Thefunctionalderivativeis:
Election Bensity Functional Theoiy Page SS
Roi Baei

o[|]
o(x )
= _
o(
i
(x)
2
)
o(x )
Jx

-
= _ 2
i
(x)
o
i
(x)
o(x )
Jx

-
= _ 2
i
(x)o'(x -x )Jx

-
= -2
ii
(x )
(1.10.17)
Here we used the identity: ] g(x)o
i
(x -x )Jx = -g
i
(x ) and the shortcut
6]
|
(x)
6](x )
= o'(x -x ).
iv. Invertible function functionals and their functional derivatives
Consider a continuous set of functionals which is indexed by a
continuous variable r in some domain . More precisely, :|n](r) assigns a
functional to each r in . Let us suppose further that n is a function of r
defined on the same domain . An example is the relation between the
electrostaticpotentialandthechargedensity:
:
H
|n](r) = _
n(r
i
)
|r -r
i
|
J
3
r
i
(1.10.18)
For any charge distribution n(r
i
) the integral forms a the electrostatic
potential :
H
(r). The potential at a give point r depends in principle on the
chargedensityatallpointsinspace.Thisisaverynonlocaldependence.
What about the functional derivative? If one makes a small change in the
functionnatthepointr
i
,on(r
i
),howwillthataffectthefunctionalatr?We
writethisasadefinitionofthefunctionalderivative:
o:|n](r) = _
o:|n](r)
on(r
i
)
on(r
i
)J
3
r
i
(1.10.19)
Intheexampleabove,itisimmediatelynoticeablethat:
o:
H
|n](r)
on(r
i
)
=
1
|r -r
i
|
(1.10.20)
Election Bensity Functional Theoiy Page S4
Roi Baei

Aninterestingconceptcanarisewhenoneknowsinadvancethattherelation
:|n]isunique,i.e.
{:|n](r) = :|n
i
](r) foi all r] = {n(r) = n
i
(r) foi all r] (1.10.21)
Inthiscaseonecanalsoconsiderntobeafunctionalof:.Indeed:
on|:](r) = _
on|:](r)
o:(r
i
)
o:(r
i
)J
3
r
i
(1.10.22)
Oncecancombinethisequationwith(1.10.20)andseethat:
on(r) = _
on|:](r)
o:(r
i
)
_
o:|n](r
i
)
on(r
ii
)
on(r
ii
)J
3
r
ii
J
3
r
i
(1.10.23)
Thisshowsthat:
_
on(r)
o:(r
i
)
o:(r
i
)
on(r
ii
)
J
3
r
i
= o(r -r
ii
) (1.10.24)
Thus, there is an inverse relation between the functional derivatives. This is
similar to what happens in usual derivatives with inverse functions. When a
function g(x) is invertible we can speak of x(g). The change in g due to a
change in x is: Jg(x) =
dg
dx
Jx. Similarly: Jx =
dx
dg
Jg. Therefore: [
dg
dx

x
=
_[
dx(g)
dg

-1
]
g(x)
.
v. Convex functionals
A function (x) is said to be convex if for any pair of abscisas x
1
and x
2
the
curve (x, (x)) described by (x) in the interval x
1
x x
2
is always below
the straight line connecting (x
1
, (x
1
)) and (x
2
, (x
2
)). A more formal
definitionisthatforanyu < z < 1wehave:
(zx
2
+(1 -z)x
1
) z(x
2
) +(1 -z)(x
1
) (1.10.25)
Election Bensity Functional Theoiy Page SS
Roi Baei

A useful interpretation of the above definition for convexity is in terms of


averages. We can view zx
1
+(1 -z)x
2
as a weighted average of the two
points, with positive weights (such an average is usually called a convex
average).Inthissensewehaveconvexityobey:
((x)) ((x)) (1.10.26)
Thisrelationismuchmoregeneralthanjust2points.Wecaneasilyshowthat
for convex functions it holds for any averaging procedure (x) = c

x

and
((x)) = c

(x

withc

uand c

= 1.Eq.(1.10.26)issometimescalled
Jensensinequality.
Oneoftheimportantimplicationsofaconvexfunctionisthatitcannothavea
localminimum:eitherthereisnominimum(forexample(x) = c
x
) or there
isjustoneglobalminimum.Ifafunctionisknowntobeconvexandtohavea
minimum then any descent algorithm is surely to find the (global)
minimum.Thisisuseful.
WecanTaylorexpand(zx
2
+(1 - z)x
1
) = (z(x
2
-x
1
) +x
1
)withrespectto
z,assumingnowz < 1.Then:
z(x
2
) +(1 -z)(x
1
) (z(x
2
-x
1
) +x
1
)
= (x
1
) +v(x
1
) z(x
2
-x
1
) +0(z
2
)
(1.10.27)
Afterrearrangementanddivisionbyzwesetztozeroandobtain:
(x
2
) -(x
1
) v(x
1
) (x
2
-x
1
) (1.10.28)
Thisrelation,ifobeyedforallpairsofpoints,isequivalenttotheconvexityof
thefunction.Byexchangingx
1
andx
2
weobtainalso:
v(x
2
) (x
2
-x
1
) (x
2
) -(x
1
) v(x
1
) (x
2
-x
1
) (1.10.29)
From(1.10.28)wefurtherhave,forz - u:
Election Bensity Functional Theoiy Page S6
Roi Baei

(x
1
) +zv(x
1
) u (x
1
+zu)
= (x
1
) +zv(x
1
) ox
1
+z
2
1
2
ox
1
vv(x
1
) ox
1
+0(z
3
|u|
3
)
(1.10.30)
Fromthis,aftercancellinganddividingbyz
2
wefind
1
2
u vv(x
1
) u ufor
arbitrary u which means the Hessian vv(x
1
) is positive definite: another
characteristicofconvexfunctions
Asimilardefinitionofconvexitycanbeappliedtofunctionals
F|zn
1
+(1 -z)n
2
] zF|n
1
] +(1 -z)F|n
2
] (1.10.31)
And they too have the equivalent differential property, which will be useful
below:
F|n
1
] -F|n
2
] __
F
on(r)
]
n
2
(n
1
(r) -n
2
(r))J
3
r
(1.10.32)
Anexampleforaconvexfunctionalusefulfordensityfunctionaltheoryisthe
socalledHartreeenergyfunctional:
E
H
|n] =
1
2

n(r)n(r
i
)
|r -r
i
|
J
3
rJ
3
r
i
(1.10.33)
The second functional derivative of this function is |r -r
i
|
-1
which is a
manifestly positive matrix (this is easily seen by Fourier transforming the
function:(r) =
1

,withresult:(q) =
4n
q
2
whichispositive).Anotherexampleis
thevonWeizsackerkineticenergyfunctional:
E
w
|n] = -
1
2
_n(r)v
2
n(r)J
3
r (1.10.34)
The physical meaning of this functional is that it gives N times the kinetic
energy of a singleparticle with ground state wave function (r) = _
n(r)
N

Election Bensity Functional Theoiy Page S7
Roi Baei

where N = ]n(r)J
3
r.Again, by showing the Hessian is positive definite we
canstraightforwardlyshowthisfunctionalisconvexFirst,rewriteitas:
E
w
|n] =
1
2
_[vn(r)
2
J
3
r =
1
8
_
(vn)
2
n
J
3
r
(1.10.35)
We drop the explicit r in the last expression, for brevity. Then, make an
arbitraybutsmallperturbatione(r):
E
w
|n +e] =
1
8
_
(vn +ev)
2
n +e
J
3
r
(1.10.36)
Andexpandthedenominatortosecondorderine:(n +e)
-1
-
1
n
_1 -e
4
n
+
e
2
[
q
n

2
+

. Expanding the numerator and performing the multiplication


orderbyorder onecan thenread offthesecond ordercontribution andafter
somemanipulationbringittotheform:
e
2
o
2
E
w
|n +e] =
e
2
8
_

2
n
_
vn
n
-
v

]
2
J
3
r
(1.10.37)
Clearly,thissecondordertermisabsolutelynonnegativeforallperturbations
and at all physical densities. This shows that the underlying functional
derivativeHessianispositivedefiniteandthusthefunctionalisconvex.
A similar condition exists for the second functional derivative. The Jensen
inequalityholdshereaswell:
F|(n)] (F|n]) (1.10.38)
This relation, in combination with the fact that (x) = c
x
is convex is useful
fordevelopingmeanfieldapproachesinStatisticalmechanics.Itwasusedby
Gibbsandlater,inaquantumcontextbyPeierls.[9]
Election Bensity Functional Theoiy Page S8
Roi Baei

J. Minimizing functionals is an important part of


developing a DFT
EquationSection(Next)Minimization of functions
We start with the problem of minimization of a simple 1D function. Given a
function(x),wewanttofindapointx
-
suchthatthefunctionisminimal.It
is clear that the slope of (x) at x = x
-

must be zero. If the slope is positive


thenwecangoleft(decreasex
-
)andreduce(x).Samelogic(buttoright)if
itwasnegative.Thus,anecessaryconditionforaminimumis:

i
(x
-
) = u (1.11.1)
LetusexpandinaTaylorsseriesthefunctionaroundthepointx
-
.Clearly,we
have:
(x) = (x
-
) +
i
(x
-
)(x -x
-
) +
1
2

ii
(x
-
)(x -x
-
)
2
+
(1.11.2)
However,takingEq.(1.11.1)intoaccount,
(x) = (x
-
) +
1
2

ii
(x
-
)(x -x
-
)
2
+
(1.11.3)
Whenxisextremelyclosetox
-
wemayneglecthighordertermsandthenwe
see that Eq. (1.11.3) is the equation of a parabola. In order that (x) be a
minimum,wemust have an ascentof(x) whenwemoveawayfromx
-
,no
matterwhichdirectionwetake(leftorright).Forsmalldisplacementsx
-
- x,
we see from Eq. (1.11.3) that the change in (x) is
1
2

ii
(x)(x -x
-
)
2
. Since
(x -x
-
)
2
is always positive, no matter the direction we move, we must
demand that
ii
(x
-
) > u

as well. Thus, our necessary conditions for a


minimumare:

i
(x
-
) = u onJ
ii
(x
-
) > u (1.11.4)
Election Bensity Functional Theoiy Page S9
Roi Baei

Now, let us consider the case of functions of two variables, (r) where
r = [
x
y
(
x y
)
1
. Notice that we use the transpose symbol I to turn a
columnvectorintoarowvector.LetusTaylorexpand(r)

aroundapointr
-
:
(r) = (r
-
) +v(r
-
)
1
(r -r
-
) +
1
2
(r -r
-
)
T
vv(r
-
)(r -r
-
) +
= (r
-
) +g
-
T
(r -r
-
) +
1
2
(r -r
-
)
T
E
-
(r -r
-
) +
(1.11.5)
Notethatweusethenotations:
g
-
= g(r
-
) = v(r
-
) =
`

o
ox
o
oy/

E
-
= E(r
-
) =
`

o
2

ox
2
o
2

oyox
o
2

oxoy
o
2

oy
2
/

r
-

(1.11.6)
The vector g(r) is called the gradient and the matrix E(r)

the Hessian at r.
NotethattheHessianisasymmetricmatrix,sincetheorderofdifferentiation
isimmaterial.Whenr
-
isaminimum,movinginfinitesimallyawayfromitin
anydirectionwillnotchangethefunction.Why?Wecanshowthisleadstoa
contradiction. Let d be an arbitrary direction. If the function decreases when
movingfromr
-
inthatdirection,nomatterhowsmallthestepsize,thenthis
contradicts that r
-
is a minimum. Now, if, d is an ascent direction, then -d
will be a descent direction again a contradiction to r
-
being a minimum.
Thusweconclude,thatcannotchange(tofirstorder)orinotherwords,the
gradientg
-
of(r)atr
-
mustbezero.ThenEq.(1.11.5)becomes:
(r) = (r
-
) +
1
2
(r -r
-
)
1
E
-
(r -r
-
) +
(1.11.7)
Election Bensity Functional Theoiy Page 6u
Roi Baei

Continuing, in order for this function to have a minimum at r


-
the second
term on the right hand side must always be positive. Thus Hessian matrix
mustbesuchthatforanyvector= :
1
H: > u.Amatrixhavingthisproperty
is called positive definite (PD). When a PD matrix is symmetric, then its
eigenvalues must be strictly positive. We can summarize the necessary
conditionsforaminimumat

r
-
:
g
-
= v(r
-
) = u onJ H
-
is Pos. c. (1.11.8)
Asanexample,letustakethefunction(x) = x
2
+y
2
.Thisfunctionisalways
nonnegative, so its minimum is at x = y = u, where it obtains the value
0
.
Thegradientisindeedzeroat(x, y) = (u,u):
_
o
x
]
(0,0)
= (2x)
(0,0)
= u
_
o
y
]
(0,0)
= (2y)
(0,0)
= u
(1.11.9)
And the Hessian is 2I
2
where I
2
is the 22 identity matrix, thus it is trivially
PD.
ii. Constrained minimization: Lagrange multipliers
Next we discuss constrained minimization. Suppose we want to minimize
(x, y), under a constraint that y = g(x). This is easy, since it can be
transformed into an unconstrained minimization. All we have to do is
minimizeF(x) = (x, g(x)).Thiswillgive:
F
i
(x) = _
o
ox
]
(x,g(x))
+_
o
oy
]
(x,g(x))
g
i
(x)
(1.11.10)
Election Bensity Functional Theoiy Page 61
Roi Baei

F
ii
(x) = _
o
2

ox
2
_
(x,g(x))
+2_
o
2

oxoy
_
(x,g(x))
g
i
(x) +_
o
oy
]
(x,g(x))
g''(x)
+_
o
2

oy
2
_
(x,g(x))
g
i
(x)
2

Andwecanthensearchforx
-
suchthatF
i
(x
-
) = uandF
ii
(x
-
) > u.
Butsometimesitisnotpossibletowritetheconstraintdirectlyasy = g(x).A
more general form is: b(x, y) = u. This is also more symmetric. Consider the
situation of minimizing (x, y) under the constraint b(x, y) = u. Suppose the
pointdepictedinthefollowingfigureisthisminimum.

This means that moving slightly along the isopleth of b = u from r


-
will not
change

:ifitdecreasesthisisnotaminimumandifitincreaseswewill
moveintheoppositedirectionandwilldecrease.
Thus at r
-
the gradient of , v(r
-
), must be normal to the line b = u.
Furthermore,bydefinitionthegradientofbisnormaltoitsisopleth.Thus,the
necessary condition for minimum is that both vectors point to the same
direction and so are proportional. We denote the proportionality constant by
z
-
:
h=1
h=0
h=1
v
v
vb
vb
Election Bensity Functional Theoiy Page 62
Roi Baei

v(r
-
) = z
-
vb(r
-
)
b(r
-
) = u
(1.11.11)
LagrangenoticedthatthisrelationandwroteaLagrangianfortheconstrained
problem:
I(r, z) = (r) -zb(r) (1.11.12)
ThenewvariableziscalledaLagrangemultiplier.Thenecessaryconditions
for a minimum in Eq. (1.11.11) translate into the condition of finding a
stationarypointofI
vI(r
-
, z
-
) = u (1.11.13)
Where v (o
x
, o

, o
z
, o
x
). The derivative with respect to z gives us back the
constraint. As an example, consider minimizing (x, y) = x +y under the
constraintb(x, y) = x
2
+y
2
-1(i.e.findthepointontheunitcircleforwhich
x +yisminimal).Wehave:
I(r, z) = (x +y) -z(x
2
+y
2
-1) (1.11.14)
Thenecessaryconditionis:
vI = [1 -2z
-
x
-
, 1 -2z
-
y
-
, -(x
-
2
+y
-
2
-1) = (u,u,u)
= x
-
= y
-
=
1
2z
-
, 2 _
1
2z
-
]
2
= 1
(1.11.15)
Thesolutionisthus:
z
-
= _
1
2
x
-
= y
-
= _
1
2
(1.11.16)
Since these are necessary conditions, we need to consider them further (they
might correspond to a maximum). The minimum corresponds only to the
negativesolution.
Election Bensity Functional Theoiy Page 6S
Roi Baei

The method of Lagrange multipliers is often more convenient to work with


than direct replacement. The problem is thus that of finding a minimum of
(r
1
, , r
N
)undertheH < Nconstraintsh(r
1
, , r
N
) = ,orb

(r
1
, , r
N
) = c

i = 1, , His:
finu r
-
such that: h(r
-
) = c foi which (r
-
) (r)
foi any r such that: h(r) = c
(1.11.17)
Tofacilitatesuchasearch,weformulatetheLagrangianfunction:
I(r, 2; c) = (r) -z

(b

(r) -c

)
M
=1
(1.11.18)
OurplanistofindthepositionofrwhichminimesIforanychoiceof2and
thenchangecuntilh(r) = c.At2
-
wehaveIassumingaminimumatapoint
r
-
.A necessary condition for the constrained minimum to be achieved at the
pointr
-
andwiththeLagrangemultipliers2
-
is:
v
r
I(r
-
, I
-
; c) = u
v
I
I(r
-
, 2
-
; c) = h(r
-
) -c = u
(1.11.19)
Notethat2
-
isnotnecessarilyaminimizerofI.Infacttheoppositeistue:2
-

is a maximizer of I and the search for constrained minimizations is a search


for saddle points (see ref. [10] for a method to solve such a problem on the
largescale).
It is interesting now to ask how (r) changes if we change the value of the
constraint c

. Indeed, when the constraints are changed, the optimized point


andLagrangemultipliercanchange,sotheLagrangianischanged:
oI(r
-
, 2
-
; c) = I(r
-
+or
-
, 2
-
+o2; c +oc) -I(r
-
, 2
-
; c) (1.11.20)
Now,becauseofEq.(1.11.19)wefind:
Election Bensity Functional Theoiy Page 64
Roi Baei

oI(r
-
, 2
-
; c) = I(r
-
+or
-
, 2
-
+o2
-
; c +oc) -I(r
-
, 2
-
; c)
= I(r
-
, 2
-
; c +oc) -I(r
-
, 2
-
; c)
(1.11.21)
Thus:
J
Jc

I(r
-
, 2
-
; c) =
o
oc

I(r
-
, 2
-
; c) = z

-
(1.11.22)
SinceI(r
-
, 2
-
; c) = (r
-
; c)wefind:
J
Jc

(r
-
; c) = z

-
or V

(r
-
; c) = 2
-
(1.11.23)
This equation reveals the meaning of the Lagrange multipliers z

-
at the
optimal point: they are equal to the rate at which the optimal value of the
minimized function changes when c

, the value of the i constraint, is


changed.Thisisanimportantresultwhichweusebelowwheneverwewant
togivephysicalsignificancetoLagrangemultipliers.
iii. Minimization of functionals
The same considerations for functions apply for functionals. Given a
functional ( ) I f ,anecessaryconditionforitsminimumis:
oI
o(r)
= u (1.11.24)

For example, consider a 1D classical particle of mass min a potential well


:(x).TheactionSisafunctionalofthetrajectoryofthisparticle:
S|x] = _ _
1
2
mx (t)
2
+:(x(t))_ Jt
t
]
t
0
(1.11.25)
For any trajectory x(t) between times t
0
and t
]
, I|x] returns a number.
Lagrange showed that finding the trajectory that makes I|x] stationary
Election Bensity Functional Theoiy Page 6S
Roi Baei

(although, not necessarily minimal), under the condition that x(t


0
) = x
0
and
x(t
]
) = x
]
are given and thus not varied, is equivalent to solving Newtons
equations under these same constraints. The functional differentiation of the
kineticenergyisperformedwith:
o
ox(t)
_ x (t
i
)
2
Jt
i
t
]
t
0
= 2_ x (t
i
)
ox (t
i
)
ox(t)
Jt
i
t
]
t
0
= 2_ x (t
i
)o

(t -t
i
)Jt
i
t
]
t
0
= -2 _ x (t
i
)o(t -t
i
)Jt
i
t
]
t
0
= -2x (t)
(1.11.26)
The condition for stationarity of the action under changes in the trajectory
whichleavetheedgesintactgive:
u =
oS
ox(t)
= -mx (t) -:
i
(x(t)) (1.11.27)
From which we obtain Newtons equation mx (t) = -:
i
(x(t)). This equation
mustbesolvedsubjecttothegivenconstraintsattheendpoints,i.e.x(t
0
) = x
0

andx(t
]
) = x
]
.
Minimizing functionals with constraints can again be done using Lagrange
multipliers.Thenonedefines:
I|, z] = I|] +zb|] (1.11.28)
Andthenecessaryconditionsare:
oI
o(r)
_

-
,x
-
= u
oI
oz
_

-
,x
-
= b|
-
] = u
(1.11.29)
Forexample,letussolvethefollowingproblem.Findtheshapeinaplaneofa
closed contour encircling maximal area under the constraint of a given
circumference l
0
.Weneedtofindthecontourgivenbyr(0) withr(0 +2n) =
r(0)whichgiveszerovariationto
Election Bensity Functional Theoiy Page 66
Roi Baei

I|r, z] = S|r] +z(I|r] -l


0
) (1.11.30)
Where S is the area and I is the circumference. We limit ourselves to curves
thatcan beuniquelydefined.A pointon thecurve, isdefinedbytheangle0
andthedistancefromtheoriginr(0).Thevectortoapointonthecurveand
itsderivatibeis:
r(0) = r(0)(cos 0 , sin0)
r
i
(0) = r
i
(0)(cos 0 , sin0) +r(0)(-sin0 , cos 0).
(1.11.31)
The area of a small arc from 0 to 0 +J0

is JS =
1
2
|r(0) r(0 +J0)| =
1
2
|r(0) r'(0)|J0.Noticethat|r r
i
| = r
2

andsoJS =
1
2
r(0)
2
J0.
The square circumference is: JI
2
= (r(0 +J0) -r(0))
2
= (r
i
(0))
2
J0
2
. Thus
JI = |r
i
(0)|J0 = r(0)
2
+r
i
(0)
2
J0.Thus,theareaisafunctionalis:
S|r] = _
1
2
r(0)
2
J0
2n
0

I|r] = _ r(0)
2
+r
i
(0)
2
J0
2n
0

(1.11.32)
ThefunctionalderivativeofSiseasy:
oS|r]
or(0)
= r(0) (1.11.33)
ThefunctionalderivativeofIiscomputedbyaddingasmallfunctione(0):

( ) ( )
2
2 2
0
2
2 2
0
2
2 2
2 2
0
2 2
2 2
1
l r r d l
r r r r d l
r r
r r d l
r r


= + + +

= + + +

+

= + +

(1.11.34)
Then,wecanuse
( )
2
1
1 1
2
O + = + + so:
Election Bensity Functional Theoiy Page 67
Roi Baei

2
2 2 0
r r
l d
r r

(1.11.35)
Integrating by parts the second term (remembering that the end terms drop
because0 = 2nisthesamepointas0 = u:

ol = _
re
r
2
+r
i2
J0
i
2n
0
-_ _
r
r
2
+r
i2
]
i
eJ0
i
2n
0
(1.11.36)
Wethusget:
ol
or(0)
=
r
r
2
+r
i2
-_
r
r
2
+r
i2
]
i
(1.11.37)
WedefinetheLagrangian:
I|r] = S|r] -z(l|r] -l
0
) (1.11.38)
Andgetthemaximumfrom:
u =
oI
or(0)
= r -z _
r
r
2
+r
i2
-_
r
r
2
+r
i2
]
i
] (1.11.39)
Clearly, a solution is r(0) = z, a constant, since then all derivatives are zero.
The curve is then circle of radius z and given the circumference, we have:
z = l
0
2n.
EXERCISESFORCHAPTER1
1) Considerasystemoftwoparticlesinaharmonicwell.Eachparticlerepelstheother.
ThepotentialenergyisI(x
1
, x
2
) =
1
2
kx
1
2
+
1
2
kx
2
2
-
1
2
q(x
1
-x
2
)
2
.Weassumethat
thesystemisbound(sou < 2q < k).(a)WritedowntheexactHamiltonianforthis
system,assumingthemassarem
1
andm
2
.(b)Makeatransformationseparating
theHamiltonianintotwouncoupledparticles:thecenterofmassX =
m
1
x
1
+m
2
x
2
m
1
+m
2

andthereducedmassx = x
2
- x
1
.WritetheHamiltonianofeachpartand
determinethemassofeachparticle.Whatarethemasseswhenm
1
> m
2
?(c)
Writedowntheenergylevelsofthesystemandtheeigenfunctions.(d)Nowmakea
BornOppenheimerapproximation.Assumingm
1
> m
2
,determinetheBorn
Election Bensity Functional Theoiy Page 68
Roi Baei

Oppenheimerpotentialsurface(obtainedwhenx
1
isconsideredaparameter).
Solution:TheHamiltonianis:
E = -
l
2
2m
1
o
1
2
-
l
2
2m
1
o
2
2
+
1
2
kx
1
2
+
1
2
kx
2
2
-
1
2
q(x
1
-x
2
)
2
.

Inthenewcoordinates:
X =
(m
1
x
1
+m
2
x
2
)
H
, x = x
2
- x
1
, x
1
= X -
m
2
H
x, x
2
= X +
m
1
H
x.
If(x
1
, x
2
)isanyfunctionthanwecandefinethesamescalarfunctionas
F(x, X) = [X -
m
2
M
x, X +
m
1
M
x.
2) Calculatethecorrelationinthecaseofasinglet,wherebothelectronsareinthe
groundstateorbitalandwhenoneisinthegroundstateorbitalandtheotherinthe
firstexcitedorbital.

II. My first density functional:


Thomas-Fermi Theory
A. Basic concepts in the electron gas and the
Thomas-Fermi Theory
ThomasandFermiassumedthattheenergyofanatomoramoleculecanbe
writtenasafunctionalofthe1particledensityasfollows:
E
1P
|n] = I
1P
|n] +] :
cxt
(r)n(r)J
3
r +
1
2

n(r)n(r
i
)
|r -r
i
|
J
3
rJ
3
r
i
(2.1.1)
(Note, for use via the BornOppenheimer approximation, to this energy we
need to add the nuclearnuclear repulsion energy.) They then assumed that
the density that characterizes the groundstate minimizes this functional
undertheconstraint:
_n(r)J
3
r = N
c
(2.1.2)
The first question, beyond the rigor of this approach is, what is the kinetic
energy functional? In order to take into account the Fermi nature and the
Election Bensity Functional Theoiy Page 69
Roi Baei

quantum nature of the electrons, this functional must include both these
considerations.TheThomasFermisolutionistoassume:
I|n] = ] t
S
(n(r))n(r)J
3
r (2.1.3)
Whatshallwetakefort
S
(n)?Considerfirstasimplecase:ahomogeneousgas
ofdensityn(i.e.n(r)isindependentofr).Furthermore,letusassumethatthe
electrons are noninteracting. This is a simple enough system to enable the
analytic calculation of the kinetic energy functional. From the form of (2.1.3)
we see that the total kinetic energy is the sum of contributions of various
infinitesimal cells inspace.Each cellcontainsn(r)J
3
relectronsandso,ifwe
interprett(n)asthekineticenergyperelectronofahomogeneousgasofnon
interactingelectronsthenthissumisyieldsexactlythetotalkineticenergyfor
thishomogeneousgas.TheThomasFermiapproximationthenusesthissame
t(n)alsofortheinhomogeneousinteractingcase.
Letusnowcomputet(n).ConsiderahomogeneousgasofNunchargedelectrons.
Theyarenoninteracting.TheseelectronsareputinacubiccelloflengthI.The
electrondensityiseverywherethesamen =
N
c
v
=
N
c
L
3
.
Weassumethewavefunctionsareperiodicinthebox.AccordingtoFouriers
theorem,wecanwriteanyperiodicwavefunctionasalinearcombinationof
planewaves,asfollows:
(r) =
o
n
c

2n
L
nr
I
n

(2.1.4)
Where:
n = (l
x
, l

, l
z
). (2.1.5)
Election Bensity Functional Theoiy Page 7u
Roi Baei

and l
xz
are integers. Fouriers theorem is based on the orthonormality of
theplanewaves
(
h
|
h
|) =
1
I
c
(h-h
|
)r
J
3
r
v
= o
h,h
|
(2.1.6)
Wherewedefined

h
(r) =
c
hr
I
h =
2n
I
n
(2.1.7)
We imagine 3dimensional kspace divided into an array of small
compartments, indexed by a set of integers n = (l
x
, l

, l
z
) or by the vector k.
Eachcompartmentisofklengthk =
2n
L
anditskvolumeisk
3
=
(2n)
3
v
.For
large rspace boxes the kspace compartment is extremely small since k
3
is
proportionaltotheinverseboxvolume.Sinceweareinterestedeventuallyin
thelimitI - ,wemayassumeapproximatesumsofanyfunction(h)over
thediscretevaluesofh =
2n
L
nbyintegrals:
(h)
h
-
I
(2n)
3
] (h)J
3
k (v - )
(2.1.8)
LetsshowthatplanewavesareeigenstatesofkineticenergyoperatorI
`
1
:
I
`
1

h
(r) = -
l
2
2p
1
I
v
2
c
hr
=
l
2
k
2
2m
c

h
(r) (2.1.9)
Now, consider the wavefunction of the N
c
noninteracting electrons in their
groundstate. Since they are noninteracting, this wavefunction is a product
ofsingleelectronwavefunctions:
=
h
1
(r
1
)

h
1
(r
2
)
h
N
c
2
(r
N
c
-1
)

h
N
c
2
(r
N
c
) (2.1.10)
Here
h
(r)isthestateofaspinupelectronwithwavevectork.while

h
(r)is
the state of a spin down electron with wave vector k. Anticipating the
Election Bensity Functional Theoiy Page 71
Roi Baei

antisymmetry,webuildthiswavefunctionbyplacing2electronsinthesame
spatial orbital (once with spin up and the other with spin down). Since non
interacting electrons have only one type of energy, i.e. kinetic energy:
E

= [-
l
2
2m
c
v
n
2

N
c
n=1
, we can easily show that (2.1.10) is an eigenstate of the
Hamiltonian:
E

= _-
l
2
2m
c
v
n
2
_
N
c
n=1

h
1
(r
1
)

h
1
(r
2
)
h
N
c
2
(r
N
c
-1
)

h
N
c
2
(r
N
c
)
= 2 _
l
2
k
n
2
2m
c
_
N
c
2
n=1

h
1
(r
1
)

h
1
(r
2
)
h
N
c
2
(r
N
c
-1
)

h
N
c
2
(r
N
c
)
= 2 _
l
2
k
n
2
2m
c
_
N
c
2
n=1

(2.1.11)
Oneseesthattheenergyisjustthesumofkineticenergy 2[
l
2
k
n
2
2m
c

N
c
2
n=1
ineach
spinorbital of the product wave function. Let us now antisymmetrized this
productwavefunction.Wedothisbyaddingallproductsresultingfromeven
permutations of the electrons and subtracting all odd permutation products.
Oneconvenientwaytorepresentsuchasumisusingadeterminant,calleda
Slaterwavefunction:
=
1
N!
uet _

h
1
(r
1
)

h
N
c
(r
1
)
. .

h
1
_rN
c
2
]

h
N
c
_rN
c
2
]
_. (2.1.12)
For this wave function to be minimal energy must fill 2 electrons per level
starting from the lowest kinetic energy and going up until electrons are
exhausted.Denotethehighestfilledlevelbyh
P
.Then:
N
]IIcd
=
I
(2n)
3
0(k
P
-k)J
3
k. (2.1.13)
Election Bensity Functional Theoiy Page 72
Roi Baei

Where 0(x) is0 if x is negative and 1 otherwise. This is called the Heaviside
function.Wenowperformtheintegralusingsphericalcoordinates:
N
]IIcd
=
I
(2n)
3
_ 4nk
2
Jk
k
F
0
=
I
(2n)
3
4n
S
k
P
3
=
I
2n
2
k
P
3
S
. (2.1.14)
ThenumberoffilledorbitalsistheproductoftherealsapcevolumIandthe
kspace occupied state volume, divided by (2n)
3
. Since N
c
= 2N
]IIcd
and the
densityisn =
N
c
v
wehave:
n =
2N
]IIcd
I
=
k
P
3
Sn
2
.
(2.1.15)
Theelectrondensitydetermineddirectlythehighestfilledmomentumstate:
I
S
|n] =
2v
(2)
3
_ _
l
2
k
2
2m
c
_4nk
2
Jk
k
F
0
=
2v
(2)
3
4n
l
2
2m
c
k
P
5
S
=
2I
2n
2
l
2
2m
c
k
P
5
S
. (2.1.16)
Theenergyperparticleis:
t
S
(n) =
I
S
N
=
1
n
2
n
l
2
2m
c
k
P
5
S
=
S
S
l
2
k
P
2
2m
c
=
S
S
l
2
2m
c
(Sn
2
n)
2
3
= C
1P
n
23
,
(2.1.17)
where:
C
1P
=
S
S
l
2
2m
c
(S
2
)
2
3
=
S
1u
(S
2
)
2
3
au = 2.871 au . (2.1.18)
Plugging into Eq. (2.1.3), the ThomasFermi kinetic energy functional is
obtainedtobeusedinEq.(2.1.1):
I
S
|n] = C
1P
] n(r)
S
3
J
3
r.
(2.1.19)
============================================
Exercise:TheThomasFermifunctionalforthehydrogenatom.
a. RepeatthecalculationabovebutnowforaspinpolarizedHEG.Thatis,donot
assumethatthereare2electronsineachkstate(thespinunpolarizedcase)but
instead,thatallspinsareupandsothereisonlyoneelectronperkstate.
Election Bensity Functional Theoiy Page 7S
Roi Baei

b. Sincetheelectroninahydrogenlikeatomisspinpolarized,usetheThomasFermi
KEfunctionalderivedin(a)andcompareitsestimationofthekineticenergyofthe
electroninahydrogenlikeatomtotheexactvalue.Usingtheexactkineticenergyin
thehydrogenatom(youcanfinditusingthevirialtheorem),assessthequalityof
theresultasafunctionofthenucleuschargeZ.
============================================
B. Minimization of the Thomas-Fermi energy
Now,accordingtotheTFtheory,thetruegroundstateelectrondensityisthe
one that minimizes E
1P
|n]. But the electron density must also account for the
required number of electrons of N
c
, so there is a constraint for the
minimization:
_n(r)J
3
r = N
c
(2.1.20)
Thus,wemustbuildaLagrangiantobeminimizedas:
I|n, z] = E
1P
|n] -p __n(r)J
3
r -N
c
_ (2.1.21)
MinimizingitgivestheThomasFermiequation:
u =
oI
on(r)
=
oE
1P
on(r)
-p (2.1.22)
WeseethattheLagrangeconstantpisthechemicalpotential,sinceitisequal
to the change in energy when we perturb the density and this change is
everywhereconstant.Wemustnowcomputethefunctionalderivatives:
o
on(r)
_C
1P
n(r
i
)
53
J
3
r' =
S
S
C
1P
n(r)
23

o
on(r)
__:(r
i
)n(r
i
)J
3
r
i
] = :(r)
o
on(r)
1
2

n(r
ii
)n(r
i
)
|r
ii
-r
i
|
J
3
r
ii
J
3
r
i
= _
n(r
i
)
|r -r
i
|
J
3
r
i

(2.1.23)
PluggingintoEq.(2.1.22),oneobtainstheThomasFermiequationforanatom:
Election Bensity Functional Theoiy Page 74
Roi Baei

p =
S
S
C
1P
n(r)
23
+:(r) +_
n(r
i
)
|r -r
i
|
J
3
r
i
(2.1.24)
This is an integral equation for n(r). It is called the integral ThomasFermi
equation.Thepotential:(r)isduetothepositivecharge,hencewecanwrite:
:(r) = -]
n
+
(r
|
)
|r-r
|
|
J
3
r
i
.Nowwecandefineatotalpotential
(r) = p +_
n
+
(r
i
) -n(r
i
)
|r -r
i
|
J
3
r
i
, (2.1.25)
asthesumofthetotalelectrostaticpotentialandthechemicalpotential.Since
v
2
1

= 4no(r),wehave:
-v
2
(r) = 4n(n
+
(r) -n(r)), (2.1.26)
leadingtoanequationforthetotalpotential:
1
4n
v
2
= _
S
SC
1P
(r)_
32
-n
+
(r)
(2.1.27)
This is called the differential Thomas Fermi equation. The constant p is
buried in but it did not disappear: it must be chosen so that Eq. (2.1.20) is
obeyed. Furthermore, it is clear that the potential (r) is manifestly non
negativeinTFtheory(otherwisewecouldnottakeitssquareroot).
As we showed (see Eq. (1.11.23)), p can be shown to act as a chemical
potential, i.e.
L
TF
(N)
N
= p(N)
.
We will not solve this equation for atoms or
molecules. We just comment that it gives a smoothed value for the atomic
density, not showing the shell structure. We can see this in the following
figure,whereweplottheradialdensityascomputedbyarelativelyaccurate
theory,suchasHartreeFockandtheTFD(ThomasFermiDirac)theory.
Elec
R

The
fun
que
For
and
has
C.
Con
]on
oq
(so
of t
elec
ction Bens
Roi Baei
ere is a qu
nctional co
estionhasb
ru < z < 1
d
N
z
is held
stheprope
Thomas-
nsider an
n
+
(r)J
3
r,w
so as to pr
that oq =
the change
ctrostaticp
ity Functio
uestion of
ompare w
beenexam
l
z
1

(i.e.then
while Z -
ertythat:
E
uto
1P
-Fermi
existing s
whereon
+
reserve neu
]on(r)J
3
e in electro
positivecha
onal Theoiy
how does
ith the ac
mined.Itwa
im
-
E
1P
(N
E
cxuct
(N
numberof
- ). Note
om
(z, Z) =
does no
system, an
+
(r) uev
utrality. Th
r). The ch
onic energy
argeenerg
y
s the minim
ccurate qu
asfoundth
N = zZ)
N = zZ)
=
electrons
e that the T
Z
73
E
utom
1P
(
ot acco
nd now ad
verywhere.A
he electron
hange in th
y, poq, and
gy:
mal energ
uantum m
hatforatom
1
issmaller
Thomas Fe
(z, 1)
ount fo
dd a bit o
Also,incre
n distributi
he total ene
d the corre
gy of the T
mechanical
mswithZ
thanthat
ermi energ
or molec
of positive
asetheele
ion is chan
ergy oE|n]
esponding
Pag

Thomas Fe
energy. T
- weha
(2.1
oftheprot
gy for an a
(2.1
cules
e charge o
ctroniccha
nged by on
] is compo
change in
e 7S
ermi
This
ave:
1.28)
tons
atom
1.29)
oq =
arge
n(r)
osed
n the
Election Bensity Functional Theoiy Page 76
Roi Baei

oE|n] = oE
1P
|n] +oE
NN
|n]
= poq +
(n
+
(r) -n(r))on
+
(r
i
)
|r -r
i
|
J
3
rJ
3
r
i

(2.1.30)
UsingthedefinitionofthetotalpotentialinEq.(2.1.25)wefind:
oE|n] = _(r)on
+
(r)J
3
r (2.1.31)
Since everywhere we add positive charge, i.e. on
+
(r) u and since (r) is non
negative everywhere, this process increases the total energy! In TF theory
addition of infinitesimal positive charge followed by addition of a
compensating electronic charge causes an increase in total energy of the
system.InamoreelaboratetreatmentTellershowed[11]thatthetotalenergy
of any diatomic molecule is higher than the sum of energy of its constituent
atoms, i.e. that TF theory cannot account for stable molecules! He concluded
thatThomasFermitheoryisnotveryusefulforchemistry.
D. Thomas-Fermi Screening
When a point impurity Zc is inserted into an electronic system, it pulls (Z
positive)orrepels(Znegative)electronstowardsit.Thishasaneffectthatthe
impurityispartiallyscreenedbyoppositechargeandsoithasasmallereffect
on distant charges. Let us study this phenomenon in the electron gas, using
ThomasFermitheory.Thehomogeneousgasofelectronsisamodelforideal
metals, so the screening effect we address here is relevant for many metallic
systems. Macroscopically, the free metal electrons completely screen the
charged impurity. However microscopically, perfect screening is not possible
becauseelectronshavekineticenergyevenatzerotemperatureandashort
ranged electric field develops around the impurity. Thomas Fermi theory
Election Bensity Functional Theoiy Page 77
Roi Baei

takeskineticenergyeffectsintoaccountandcanbeusedtoestimatetheform
ofthelocalelectricfield,specificallyitssizeorlengthscale.
LetusstudyanunperturbedhomogeneouselectrongasusingThomasFermi
theory. Such a gas has no structure and it is characterized by only one
parameter: its density n
0
. In order to neutralize it and support the electron
homogeneity, we add positive smeared homogeneous charge density +cn
0
.
AlltheCoulombenergies(ee,eNandNN)cancelexactlysotheonlyenergy
leftistheelectronickineticenergy:
E
1P
|n
0
] = _C
1P
n
0
(r)
53
J
3
r (2.2.1)
Theconstraintminimizationofthisfunctionalyieldsthefollowingcondition,
relatingthedensitytothechemicalpotential:
p =
oE
1P
on
0
(r)
=
S
S
C
1P
n
0
(r)
23
(2.2.2)
Comparing with Eq. (2.1.15), and using Eq. (2.1.18) we find for the chemical
potential:
p =
oE
1P
on
0
(r)
=
S
S
C
1P
_
k
P
3
Sn
2
_
23
=
l
2
k
P
2
2m
c

(2.2.3)
Thus we see that indeed the electron density is constant and the chemical
potential is equal to the kinetic energy corresponding to the maximal
occupiedmomentumlk
P
.
NowweintroduceapositivechargeZc.Thedensityofelectronsischanged:
n(r) = n
0
+n
1
(r) (2.2.4)
It is physically clear that n
1
(r) is localized around the impurity (assumed at
theorigin).Wethereforehaveforthetotalenergyofthesystemintermsofn
1
:
Election Bensity Functional Theoiy Page 78
Roi Baei

E
1P
|n
1
] = _C
1P
(n
0
+n
1
(r))
53
J
3
r -_
Zc
2
n
1
(r)
r
J
3
r
+
c
2
2

n
1
(r)n
1
(r
i
)
|r -r
i
|
J
3
rJ
3
r
i

(2.2.5)
ThecorrespondingTFequationcomesfromminimizing:
p =
oE
1P
on
1
(r)
=
S
S
C
1P
(n
0
+n
1
(r))
23
-
Zc
2
r
+c
2
_
n
1
(r
i
)
|r -r
i
|
J
3
r
i

(2.2.6)
Wewrite:(r) = c ]
n
1
(r
|
)
|r-r
|
|
J
3
r
i
andso:
p =
oE
1P
on
1
(r)
=
S
S
C
1P
(n
0
+n
1
(r))
23
-
Zc
2
r
+c(r) (2.2.7)
Uponlinearizing,assumingn
1
< n
0
:
S
S
C
1P
n
0
23
_1 +
2
S
n
1
n
0
] -
Zc
2
r
+c(r) = p (2.2.8)
Wecanwrite:
5
3
C
1P
n
0
23
= p
0
andso:
1u
9
C
1P
n
0
-13
n
1
-
Zc
2
r
+c(r) = p -p
0

(2.2.9)
Finallysincev
2
= -4ncn
1
wehave:
-
1u
9 4nc
2
n
0
-13
C
1P
v
2
-
Zc
r
+(r) =
p -p
0
c

(2.2.10)
We have from Eq. (2.1.15) k
P
= (Sn
2
n
0
)
1
3
and we use the definition of the
Bohrradiuso
0
= -
l
2
m
c
c
2
,definingtheThomasFermiscreeningparameterk
1P
:
1u
9 4nc
2
n
0
-13
C
1P
=
no
0
4k
P

1
k
1P
2
(2.2.11)
Withthiswehavetheequation:
v
2
= k
1P
2
_(r) -
Zc
r
-
p -p
0
c
]
(2.2.12)
Election Bensity Functional Theoiy Page 79
Roi Baei

Passingtosphericalcoordinateswefind:
1
r
(r(r))
ii
= k
1P
2
_(r) -
Zc
r
-p ].
(2.2.13)
Defining_ = rwefind:
_
ii
= k
1P
2
(_ -Zc -pr) (2.2.14)
The homogeneous equation is _
ii
= k
1P
2
_ which has the solution _
H
=
Ac
-k
TF

+Bc
k
TF

. Clearly, for a localized potential solution we must take


B = u. To this we need add any solution of the inhomogeneous equation
whichclearlyis_
IH
= Zc +pr.Thus:
_ = Ac
-k
TF

+Zc +pr (2.2.15)


Thisleadsto:
=
Ac
-k
TF

r
+
Zc
r
+p
(2.2.16)
Inthelimitthatr - uwemusthaver(r) - usincetheelectronicchargen
1

hasnocusps.ThusA = -Zc .Thetotalelectrostaticpotentialis

tot
(r) = (r) -
Zc
r
=
-Zc
r
c
-k
TF

+p
(2.2.17)
Asidefromtheconstantp,farfromtheimpuritythesurfaceintegralofv
tot

evaluatestozeroandbyGaussstheoremalargespherearoundtheimpurity
includeszerochargeinit,meaningthatthetotalamountofelectroniccharge
pulledintothesphereisexactlyequaltothatoftheimpurity(Z).
It is interesting that the screening length is proportional to k
P
-12
or to n
0
-16
.
The higher the density the smaller the length, i.e. the more efficient is the
screening, however, the dependence on n
0
is mild because of the small
exponent.Itisalsointerestingtonotethatk
1P
isindependentofZ.However,
this latter results holds only in so far as our linearization is valid. For strong
Election Bensity Functional Theoiy Page 8u
Roi Baei

impurities the nonlinear equation will give a different result and the
screeningwilldependonZ.
E. Von Weizscker kinetic energy
The Thomas Fermi kinetic energy density functional is exact in the limit of
noninteracting homogeneous gas of electrons in an infinite box. We would
like to mention here another density functional which is exact in a certain
limit, i.e the limit of a single electron. In this case the kinetic energy is:
I = ] (r) [-
l
2
2
c
v
2
(r)J
3
r . For wave functions that decay to zero at r -
, one can integrate by parts and obtain I =
l
2
2
c
] (v(r))
2
J
3
r, stressing the
absolute positivity of kinetic energy (it cannot be zero). Finally, if (r) is a
nondegenerate groundstate it can be written as (r) = n(r) and so we
obtainthekineticenergyfunctionalofvonWeizscker:
I
w
|n] =
l
2
2p
c
_[vn(r)
2
J
3
r (2.2.18)
Whichcanbewrittenasfollows,usinglocalwavevector:
k(r)
1
2
vn(r)
n(r)
(2.2.19)
So:
I
w
|n] = _
l
2
k(r)
2
2p
c
n(r)J
3
r (2.2.20)
This functional is now used for any density, even a many electron one. The
variationis:
oI
w
=
l
2
8p
c
__
[v(n(r
i
) +e(r'))
2
n(r
i
) +e(r')
-
[v(n(r
i
))
2
n(r
i
)
_J
3
r'
(2.2.21)
Election Bensity Functional Theoiy Page 81
Roi Baei

Workingthisouttolineartermsine,using:(n(r
i
) +e(r'))
-1
= n(r
i
)
-1
[1 -
er'nr'weobtain:
oI
w
=
l
2
8p
c
e __
2vn(r
i
) v(r')
n(r
i
)
-[v(n(r
i
))
2
(r')
n(r
i
)
2
_J
3
r' (2.2.22)
Whichafterintegrationbypartsofthefirsttermfinallygives:
oI
w
=
l
2
8p
c
e __-2v _
vn(r
i
)
n(r
i
)
_ -_
v(n(r
i
))
n(r
i
)
_
2
_(r')J
3
r' (2.2.23)
ThusthevonWeizsckerpotentialis:
:
w
(r) = -
l
2
8p
c
_2v _
vn(r)
n(r)
_ +_
v(n(r))
n(r)
_
2
_ (2.2.24)
Whichcanbewrittenmorecompactlyas:
:
w
(r) = -
l
2
2p
c
|v k(r) +k(r)
2
] (2.2.25)
Exercise:For1electronsystem,discusstheclaims:1)Thewavevectork(r)is
the gradient of the log of the of the wavefunction: k(r) = vlog (r) (2) the
von Weizscker potential is the potential for which n(r) is the ground state
density.
III. Many-electron wave functions
A. The electron spin
Zeeman has shown that a small magnetic field causes the splitting of energy
levels in atoms. Each atomic level is split into a doublet. The amount of
splitting is proportional to the field. At zero field these doublets are
degenerate. The conclusion is that the electron has an intrinsic magnetic
moment which can take two values. The states of the internal magnetic
Election Bensity Functional Theoiy Page 82
Roi Baei

momentoftheelectronareassumedtobeproportionaltoaninternalangular
momentumcalledspin.Thespinofanelectroncanisassumedtohavetwo
values_
l
2
.Thisisanadditionaldegreeoffreedom.Itisnotcontinuous,but
is nevertheless it is degree of freedom. We denote a spinorbital
n
(x) =

n
(r, s)whererisapointin3Dspaceandsisaspinvariable,whichallows
us to perform a inner product of spin as explained now. There are two
possible spin functions for an electron, o(s) denotes spin up and [(s) spin
down.Thesetwostatesarecompleteandorthonormal:
_o(s)
-
o(s)Js = (o|o) = 1
_[(s)
-
o(s)Js = ([|o) = u
_o(s)
-
[(s)Js = (o|[) = u
_[(s)
-
[(s)Js = ([|[) = 1
Thevariablesisjustamneumonic.Withnewnotation,wehave:
(|) = _(x)
-
(x)Jx = __(r, s)
-
(r, s)Js J
3
r (3.1.1)
B. The Pauli principle
The electronic wavefunctions are functions of N
c
electronic coordinates and
spins (x
1
, , x
N
c
). Here x
]
(r
]
, s
]
). The Pauli principle states that this
wavefunction must be antisymmetric with respect to interchange of two
electrons:
(x

, , x
]
, ) -(x
]
, , x

, ) (3.2.1)
Election Bensity Functional Theoiy Page 8S
Roi Baei

This is a boundary condition we impose while solving for any electronic


wavefunction.
C. The Excited states of the Helium atom
Howshouldwerepresentthe,inanapproximateform,thelowlyingexcited
states of the Helium atom. He
+
has two low lying orbitals 1s and 2s (the 2p
orbitalsaredegeneratewiththe2s,butwewillnotconsiderthembecausein
the Helium atom they are of much higher energy. We can form a 2electron
wavefunctionby:(x
1
, x
2
) =
1s
(r
1
)
1s
(r
2
)|o(s
1
)[(s
2
) -o(s
2
)[(s
1
)].
The excited states will involve excitation of an electron to the 2s orbital. We
canthenwrite:

2
(x
1
, x
2
) = |
1s
(r
1
)
2s
(r
2
) -
1s
(r
2
)
2s
(r
1
)]o(s
1
)[(s
2
)

3
(x
1
, x
2
) = |
1s
(r
1
)
2s
(r
2
) -
1s
(r
2
)
2s
(r
1
)]|o(s
1
)[(s
2
)
+o(s
2
)[(s
1
)]

4
(x
1
, x
2
) = |
1s
(r
1
)
2s
(r
2
) -
1s
(r
2
)
2s
(r
1
)]|[(s
1
)[(s
2
)]

5
(x
1
, x
2
) = |
1s
(r
1
)
2s
(r
2
) +
1s
(r
2
)
2s
(r
1
)]|o(s
1
)[(s
2
)
-o(s
2
)[(s
1
)]
(3.3.1)
The first 3 states form a triplet the total spin is 1. The last is again a singlet
(likethegroundstate).
D. The Slater wave function is the basic anti-
symmetric function describing N electrons in
N orbitals
The previous example is difficult to generalize. In order to develop a way to
easily represent antisymmetric functions of all types, we consider the
following2electronfunction,composedof21electronspinorbitals:
Election Bensity Functional Theoiy Page 84
Roi Baei

(x
1
, x
2
) =
1
2
|
1
(x
1
)
2
(x
2
) -
1
(x
2
)
2
(x
1
)]
=
1
2
_

1
(x
1
)
1
(x
2
)

2
(x
1
)
2
(x
2
)
_
(3.4.1)
Ifwechoosetheorbitalstobeorthonormal,(

|
]
) = o
]

then:
(|) = _(x
1
, x
2
)
-
(x
1
, x
2
)Jx
1
Jx
2
=
1
2
|
1
(x
1
)
2
(x
2
) -
1
(x
2
)
2
(x
1
)]
2
Jx
1
Jx
2
=
1
2
|
1
(x
1
)
2

2
(x
2
)
2
+
1
(x
2
)
2

2
(x
1
)
2
+2
1
(x
2
)
2
(x
1
)
1
(x
2
)
2
(x
1
)]Jx
1
Jx
2
= 1
(3.4.2)
E. Without loss of generality, we may assume the
orbitals of a Slater wave function are
orthogonal
Butwhathappensiftheorbitalsarenotorthogonal?Supposethattheorbitals
werenotorthonormal:
(

|
]
) = S
]
(3.5.1)
It is then possible to orthonormalize them. i.e define two linear
combinationswhichareorthonormal.Define:

n
=
m
A
mn
m

(3.5.2)
anddemand:(

|
]
) = o
]
.Then:
o
]
(

|
]
) = A
m
-
m,n
(

|
]
)A
n]
= (A
|
SA)
]

(3.5.3)
Thus: A
|
SA = I or AA
|
= S
-1
. Note also that uet A = (uet S)
-12
.There are
many solutions to this equation. (For example, if A
0
is a solution then so is
A
0
uwhereuisanyunitarymatrix.)Eachsolutionwillgiveusadifferentset
Election Bensity Functional Theoiy Page 8S
Roi Baei

of orthonormal orbitals. The Slater wave function made out from these new
orbitalsis:
uet|

(x
]
)| = uet|(A
1
)
n

(x
]
)| = uet A uet|

(x
]
)|
=
1
uet S
uet|

(x
]
)|
(3.5.4)
Thus, the new wavefunction is the same as the old one, up to multiplication
by constant! Yet, it is always more convenient to work with normalized
orbitals, so we can assume the orbitals are orthonormal without any loss of
generality.ThisdevelopmentalsoshowsthatgivenanysetofNorbitalsfrom
which the Slater wave function has been constructed, we can take N linear
combinations of the orbitals to obtain new orbitals that give the same Slater
wavefunctionuptoaconstantfactor.
F. Any antisymmetric function can be expanded as a
sum of basic Slater (determinantal) functions
For orthonormal orbitals, the normalization is easy to compute. We write
explicitlythedeterminantas:
uet|
1

N
] (x
1
, , x
N
) = (-)
P
i
1
i
N
c
_

k
(x
k
)
N
k=1
1

N
(3.6.1)
Where (i
1
. . i
N
) is a permutation of the numbers 1 N (there are N! such
permutations). Each permutation can be obtained from a series of pair
swappingoperations.Forexample:(1S2)isobtainedfrom(12S)byswitching
the pair of numbers in position 2 and 3. We write this as: (1S2) = S
23
(12S).
(241S)isobtainedfrom(12S4)bythreeoperations:
S
23
S
34
S
12
(12S4) = S
23
S
34
(21S4) = S
23
(214S) = (241S) (3.6.2)
Election Bensity Functional Theoiy Page 86
Roi Baei

IfthenumberofswitchesisoddthepermutationisoddandP

N
= 1;ifthe
number of switches is even, the permutation is even and P

N
= u. The
normalization of a determinantal wave function composed of orthonormal
orbitalsis:

_|uet|
1

N
] (x
1
, , x
N
)|
2
Jx
1
Jx
N
= (-)
P
i
1
i
N
c (-)
P
]
1
]
N
c
__

k
(x
k
)
N
k=1
_
]
q
(x
q
)
N
q=1
Jx
1
Jx
N

N
]
1
]
N
= (-)
P
i
1
i
N
c (-)
P
]
1
]
N
c
_(

k
|
]
k
)
N
k=1
1

N
]
1
]
N

(3.6.3)
Becauseoforthonormalitytheorbitalintegral(

k
|
]
k
)iszerounless
k k
i j = .
These integrals appear in products so the product is nonzero only if
k k
i j =
for all k = 1N. The only conclusion is, that the two permutations must be
identicaland:
_|uet|
1

N
] (x
1
, , x
N
)|
2
Jx
1
Jx
N
= 1

N
]
1
]
N
P
= N!
(3.6.4)
Weconcludethatthenormalizationfactorofadeterminantalwavefunctionof
orthonormalorbitalsis
1
N!
andwrite:
1
N!
uet|

N
] |i
1
i
N
) (3.6.5)
Given a set of H > N orthonormal singleelectron spinorbitals
n
(x) =

n
(r , s)n = 1,2, , H,

wecanconsiderthespaceofalllinearcombinationsof
all Nparticledeterminants thatcan bemade.Thereare[
H
N
=
M!
N!(M-N)!
ways
to select determinants so this is the dimension of the space. The dimension
grows factorially with H. A typical antisymmetric wave function can e
approachedbylinearcombinationsofthesedeterminants:
Election Bensity Functional Theoiy Page 87
Roi Baei

(x
1
, , x
N
) =
1
N!
C

1
,,
N
{
1
,,
N
]
uet|

1
(x
1
)

N
(x
N
) ]
(3.6.6)
ThesumisoverallselectionsofNintegers,whereeachselectionisorderedso
that i
1
< i
2
< . If the orbitals are orthogonal, the constants C

1
,,
N
are
obtainedfrom:
C

1
,,
N
=
1
N!
_uet|

1
(x
1
)

N
(x
N
) ] (x
1
, , x
N
)Jx
1
Jx
N
(3.6.7)
G. Determinant expectation values
In this section we discuss the calculation of expectation values of many
electron operators for N electrons within a given Slater wave function

S
(x
1
, , x
N
) =
1
N!
uet|
1
(x
1
)
N
(x
N
) ]. We assume the orbitals

(x) are
orthonormal:(

|
]
) = o
]
.
i. One-body operators
Consider an operator o
1
(x) which operates on an electron with spin
coordinatesx.ForNelectronswedefinethesumofo
1
foreachelectron
0
`
= o
1
(x
n
)
N
n= 1
, (3.6.8)
Examples: when electrons are in a potential well :(r), the total potential
energy operator is I
`
= :(r
n
)
N
n=1
; the total kinetic energy is:
I
`
= I
1

(n)
N
n=1
= [-
l
2
2
c

2
x
n
2

N
n=1
.
When the system of N electrons is in a given Slater wave function

S
(x
1
, , x
N
) =
1
N!
uet|
1
(x
1
)
N
(x
N
) ], then using the notation of (3.6.3),
wehave:
Election Bensity Functional Theoiy Page 88
Roi Baei

(
S
|0
`
|
S
)
=
1
N!
_uet|
1
(x
1
)
N
(x
N
)] 0
`
uet|
1
(x
1
)
N
(x
N
)] Jx
1
Jx
N
=
1
N!
(-)
P
i
+P
]
_(

k
|
]
k
)
N
k=1
k=n
(

n
|o
1
|
]
n
)

N
]
1
]
N
N
n=1

(3.6.9)
Once again, a massive cancellation of terms happens in the first integral.
Inspection shows that both permutations, i and ], must be equal otherwise
there is always an orbital integral for which the integrals (

k
|
]
k
) is zero.
Whenthepermutationsareidenticalwehave:
(
S
|0
`
|
S
) =
1
N!
(

n
|o
1
|
]
n
)

N
N
n=1
= (
]
|o
1
|
n
)
N
n=1
(3.6.10)
Asanexample,letustaketheelectrondensityn(r) = o(r -r
n
)
N
n=1
.Thus:
n(r) = |
n
(r, s)|
2
N
n=1
(3.6.11)
Conclusion: The matrix element of a 1particle operator is the sum of its
singleelectronmatrixelements.
ii. Two-body operators
Consider an operator o
12
(x
1
, x
2
) which operates on two electrons with spin
coordinatesx
1
andx
2
.ForNelectronswedefinethesumofo
12
onallpairsof
electrons
0
`
= o
12
(x
n
, x
m
)
N
n<m= 1
=
1
2
o
12
(x
n
, x
m
)
N
n=m= 1
, (3.6.12)
Examples:The2electroninteractionpotentialisu
12
(r
1
, r
2
) =
kc
2
|r
1
-r
2
|
.Thetotal
interactionenergyoperatoris:u

=
1
2
u
12
(r

, r
]
)
N
]=1
]=
N
=1
.
Election Bensity Functional Theoiy Page 89
Roi Baei

Wecomputetheexpectationvalue
(
S
|u

|
S
)
=
1
N!
_uet|
1
(x
1
)
N
(x
N
)] u

uet|
1
(x
1
)
N
(x
N
)] Jx
1
Jx
N
=
1
N!
1
2
(-)
P
i
+P
]
_(

k
|
]
k
)(

]
m
|u
12
|

]
n
)
N
k=n
k=m

N
]
1
]
N
N
n,m=1
n=m

(3.6.13)
Whereweusedthenotation:
(

]
|u
12
|

|
]
|) _

(x)
]
(x')u
12
(r, r
i
)

|(x)
]
|(x
i
)JxJx
i
(3.6.14)
Thefollowingsymmetrypropertiesholdfromtheabovedefinition:
(

]
|u
12
|

|
]
|) = (

|
]
|u
12
|

]
|) = (

]
||u
12
|

|
]
)
= (

|
]
||u
12
|


]
) = ctc
(3.6.15)
For a pair of permutations to contribute to the integral in Eq. (3.6.13), the
permutations must either be identical or involve the permutation of a single
pairoforbitals.Thus:
(
S
|u

|
S
) =
1
2
|(
n

m
|u
12
|
n

m
) -(
n

m
|u
12
|
m

n
)]
N
n,m=1
n=m

(3.6.16)
IV. The Hartree-Fock Theory
A. The Hartree-Fock Energy and Equations
The variational principle says that the lowest expectation value of the
electronicHamiltonianattainedbythegroundstate.ThisHamiltonian,inthe
nonrelativisticapproximation,forNelectronsisgivenby:
Election Bensity Functional Theoiy Page 9u
Roi Baei

= b
`
+u

= b
`
n
N
n=1
+
c
2
2

1
r
nm
N
n=m=1
(4.1.1)
Wheretheonebodyoperatoris
b
`
= b
`
m
N
m=1
= -
l
2
2m
c
v
m
2
+:(r
m
)
N
m=1
(4.1.2)
AndthetwobodyCoulombrepulsionoperatoris:
u

=
c
2
2

1
|r
n
-r
m
|
N
n=m=1
(4.1.3)
Givenafamilyofwavefunctionswecanchoosethebestofthembyfinding
that which minimizes the expectation value of E

. For the Slater wave


functions
S
= |
1

N
|,theenergytobeminimizedis:
E|V
S
] = (
S
|E

|
S
) E
HP
|
1

N
] (4.1.4)
Becausewewanttheorbitalstobeorthonormal,wewriteaLagrangian:
I|V
S
] = I
HP
|
1

N
] = E
HP
|
1

N
] - p
]
|(

|
]
) -o
]
]
N
,]=1
(4.1.5)
Theminimumisattainedby:
u =
oI
HP
|
1

N
]
o

(x)
=
oE
HP
|
1

N
]
o

(x)
-(p
]
+p
]
)
]
(x)
N
]=1
(4.1.6)
Letustryasolutionwith
p
]
=
1
2
e

o
]

(4.1.7)
i.e.:
oE
HP
|
1

N
]
o

(x)
= e

(x) (4.1.8)
Election Bensity Functional Theoiy Page 91
Roi Baei

If we can solve this equation, and if the solutions are naturally orthogonal,
then we have obtained the necessary conditions for a minimum. Now, we
only need to estimate the left hand side of this equation. From the previous
work,weknow:
E
HF
|
1

N
] = (
S
|b
`
|
S
) +(
S
|u

|
S
) (4.1.9)
where
(
S
|b
`
|
S
) b|
1

N
c
] = (
m
|b
`
1
|
m
)
N
m=1
(4.1.10)
Is the one body energy defined the as the sum of kinetic energy and
externalenergy(i.e.energyduetothefrozennuclei):
b
`
1
= -
l
2
2p
c
v
2
+:(r) (4.1.11)
Furthermore,the2electronenergycanbewrittenas:
(
S
|u

|
S
) =
c
2
2
[
n

m
_
1
r
12
_
n

m
-[
n

m
_
1
r
12
_
m

N
n,m=1
n=m
[|
1

N
] +K|
1

N
]
(4.1.12)
WherethedirectorHartreeenergyis:
[|
1

N
] =
c
2
2
[
n

m
_
1
r
12
_
n

N
n,m=1
, (4.1.13)
andtheenergyexchangeis
K|
1

N
] = -
c
2
2
[
n

m
_
1
r
12
_
m

N
n,m=1
. (4.1.14)
Election Bensity Functional Theoiy Page 92
Roi Baei

ThedirectenergyisnumericallyequaltotheHartreeenergyE
H
|n]whichisa
functional of the Slater wavefunction electron density n(r) = (
S
|n(r)|
S
) =
|
m
(r, s)|
2
N
c
m=1
:
E
H
|n] =
c
2
2

n(r)n(r
i
)
|r -r
i
|
J
3
rJ
3
r
i

(4.1.15)
The exchange energy is numerically equal to the exchange energy E
X
|p]
whichisafunctionalofthedensitymatrixdefinedby
p(x, x
i
) =
m
(x)
m
(x
i
)
N
m=1
(4.1.16)
As:
E
X
|p] = -
c
2
2

|p(x, x
i
)|
2
|r -r
i
|
JxJx
i

(4.1.17)
Noticethat:]p(x, x)Js = n(r).Thedensitymatrixisidempotent:
_p(x, x
ii
)p(x
ii
, x
i
)J
3
x
ii
= p(x, x
i
) (4.1.18)
This result shows that the DM is a projection operator, projecting onto the
spaceoforbitalswhichdefinestheSlaterwavefunction.
With the direct and exchange energies we also define their functional
derivatives:
Election Bensity Functional Theoiy Page 9S
Roi Baei

o[
o

(x)
=
1
2

o
o

(x)
_
n

m
_
c
2
r
12
_
n

m
_
n,m
=
c
2
2

o
o

(x)

n
(x
1
)
2

m
(x
2
)
2
|r
1
-r
2
|
Jx
1
Jx
2
n,m
=
c
2
2
_2o
n

n
(x) _

m
(x
2
)
2
|r -r
2
|
Jx
2
n,m
+2o
m

m
(x) _

n
(x
1
)
2
|r
1
-r|
Jx
1
_
= 2c
2
__
n(r
2
)
|r -r
2
|
J
3
r
2
_

(x) 2:
H
(r)

(x)
2:
H

(x)
(4.1.19)
WhereinthelastlinewedefinedtheHartreepotential:
:
H
(r) = c
2
_
|
m
(x)|
2
|r -r
i
|
J
3
r
i
N
m=1
= c
2
_
n(r
i
)
|r -r
i
|
J
3
r
i
(4.1.20)
Thenthedirectenergycanalsobewrittenasafunctionaloftheorbitals:
[|
1

N
] =
1
2
(
m
|:
H
|
m
)
N
m=1
=
1
2
_:
H
(r)n(r)J
3
r =
1
2
(
S
|I
`
H
|
S
) (4.1.21)
where I
`
H
= :
H
(r
n
)
N
n=1
is the total Hartree potential. A similar treatment
existsfortheexchangeenergyfunctionalderivative:
oK
o

(x)
= -
c
2
2

o
o

(x)
[
n

m
_
1
r
12
_
m

n,m
= -
c
2
2
_2o
n

m
(x) _

m
(x
2
)
n
(x
2
)
|r -r
2
|
Jx
2
n,m
+2o
m

n
(x) _

n
(x
2
)
m
(x
2
)
|r -r
2
|
Jx
2
] =
= -2c
2
_

n
(x
2
)

(x
2
)
|r -r
2
|
Jr
2
n

n
(x) = 2K

(x)
(4.1.22)
Election Bensity Functional Theoiy Page 94
Roi Baei

Where the last equality is based on a definition of a oneparticle exchange


operator:
K

1
(x) -c
2

m
(x) _

m
(x
i
)(x
i
)
|r -r
i
|
Jx
i
N
m=1
= -c
2
_
p(x, x
i
)(x
i
)
|r -r
i
|
Jx
i

(4.1.23)
Thentheexchangeenergyiswrittenasafunctionaloftheorbitals:
K|
1

N
] =
1
2
(
n
|K

1
|
n
)
N
n=1
=
1
2
(
S
|
`
H
|
S
) (4.1.24)
where
`
H
= K

n
N
n=1
is the total exchange operator. The other functional
derivativesneededare:
o
o

(x)
_
n
(x')v
2

n
(x')J
3
x' = 2v
2

n
(x)o
n
(4.1.25)
and
o
o

(x)
_
n
(x
i
)
2
:(r
i
)Jx' = 2:(r)
n
(x)o
n
(4.1.26)
Thus:
o
o

(x)
_
n
(x
i
)b
`
(r
i
)
n
(x
i
)Jx
i
= 2b
`

n
(x)o
n
(4.1.27)
Plugging all these terms into Eq. (4.1.7), we obtain the HartreeFock
equations:
F
`

(x) = e

(x) (4.1.28)
where:
F
`
= b
`
+:
H
(r) +K

1
(4.1.29)
Election Bensity Functional Theoiy Page 9S
Roi Baei

Equations(4.1.28)seemverymuchlike1electroneigenvalueequationsofthe
Schrdingerequation,exceptthatinsteadofaregularHamiltonian,wehavea
Fock operator including the nonlocal exchange. We showed in the exercise
that K

is Hermitean and thus so is F


`
, i.e. (|F
`
|) = (|F
`
|)
-
. We can thus
choose the orbital solutions of the eigenvalue equation (4.1.28) orthonormal.
ThisshowsthatthechoiceEq.(4.1.7)isindeedacceptable.
Because :
H
(r) and K

themselves depend on
m
, the HartreeFock equations
arefundamentallydifferentfromtheSchrdingerEquation:theyarenonlinear
equations.
Nowthatthesumoforbitalenergiesis:
e
m
N
m=1
= (
m
|F
`
|
m
)
N
m=1
= |(
m
|b
`
1
|
m
) +(
m
|:
H
+K

1
|
m
)]
N
m=1
= b|
1
, ,
N
] +2[|
1
, ,
N
] +2K|
1
, ,
N
]
(4.1.30)
This shows that the orbital sum is not equal to the energy of the wave
function, since it involves double counting of the direct and exchange
energies.TheHFenergyisthus:
E
HP
|
1
, ,
N
] = e
m
N
m=1
-([|
1
, ,
N
] +K|
1
, ,
N
]) (4.1.31)
B. Restricted closed-shell Hartree-Fock
FormoleculeswithevennumberNofelectronsinaspinsingletstate,wecan
impose the following structure on the Slater wave function. We can assume
that the 2N spinorbitals come in pairs:
2]-1
(x) =
]
(r)o() and
2]
(x) =

]
(r)[(). Thus each orbital
]
, ] = 1, , N2 is doubly occupied by
electronsofbothspins.Byimposingthisconstraintweobtaintherestricted
HartreeFockgroundstate.Itwillsometimebeofhigherenergythanthefully
Election Bensity Functional Theoiy Page 96
Roi Baei

unrestrictedcase.However,thewavefunctionhasawelldefinedspinwhich
maybeadvantageousinsomeapplications.
The Restricted HartreeFock (RHF) energy in the closed shell case remain
essentially the same except for counting business. We can formulate all
expressions using only the spatial orbitals. Indeed, the RHF energy is given
by:
E
RHP
|
1
, ,
N2
]
= 2 (
m
|b
`
|
m
)
N2
m=1
+4[|
1
, ,
N2
] +2K|
1
, ,
N2
]
(4.2.1)
Where[|
1
, ,
N2
]andK|
1
, ,
N2
]aretheorbitalfunctionalsdefinedin
Eqs. (4.1.13) and (4.1.14) respectively. The reason for multiplying the one
body part by two is evident: each orbitals is double occupied so has double
contribution.Thedirectpartismultipliedby4sincethedensityismultiplied
bytwoandthedirectpartdependsonthedensitymultipliedbyitself.Finally,
the exchange part is multiplied by 2 and not 4 since only o -o and [ -[
contribute, while o -[ and [ -o do not (so only half the contribution of
direct).
TheRHFequationsarethen:
_-
l
2
2p
c
v
2
+:(r) +:
H
(r)_
m
(r) +K

m
(r) = e
m

m
(r) (4.2.2)
Where:
:
H
(r) = 2c
2
_
|
m
(r)|
2
|r -r
i
|
J
3
r
i
N2
m=1
(REF)
(4.2.3)
and
Election Bensity Functional Theoiy Page 97
Roi Baei

1
(r) -c
2

m
(r) _

m
(r
i
)(r
i
)
|r -r
i
|
Jr
i
N2
m=1
(REF)
(4.2.4)
Note that the Hartree interaction is between each electron and all other
electronsregardlessoftheirspinwhiletheexchangeinteractioninvolveseach
electronswithallotherelectronsofthesamespin.
Example:TheH2molecule
WeapplytheRHFtheoryfortheH2,havingapairofelectrons.The2electron
wave function includes just one spatial orbital populated by spinpaired
electrons:
(x
1
, x
2
) = _
(r
1
) (r
2
)

(r
1
)

(r
2
)
_ = (r
1
)(r
2
)|o(1)[(2) -o(2)[(1)] (4.2.5)
SincethereisonlyoneorbitaltheexchangeKanddirectJinvolvejustoneand
the same integral. Thus the RHF energy is in this case:
E
RHP
|] = 2(
m
|b
`
|
m
) + 2[|]andtheRHFequationis
_-
l
2
2p
c
v
2
+:(r) +
1
2
:
H
(r)_(r) = e(r) (4.2.6)
where :(r) = -
c
2
|r-R
A
|
-
c
2
|r-R
B
|
and :
H
(r) = 2c
2
]
|(r)|
2
|r-r
|
|
J
3
r
i
. The effect of
exchange here is to annihilate the Coulomb repulsion of the o ([) electron
withitself,leavingonlytheinteractionoftheo -[electrons.
This RHF approach works nicely for the case that the distance between the
nuclei |R
A
-R
B
| is close to the typical bond length of B
2
(which is close to
1.4o
0
) . The energy of the molecule at this configuration can be calculated
numerically and results in E
RHP
= -1.1S4E
h
. Compared to the energy of 2 H
atoms (-1E
h
) this results indicates that the atomization energy of B
2
is
u.1S4E
h
= S.65eV. The atomization energy based on expremintal results is
Election Bensity Functional Theoiy Page 98
Roi Baei

about 4.75eV. This shows that the RHF approximation does not give high
quality atomization energies, since a deficit of 1 eV is very substantial in
Chemicalenergyterms.
A more severe problem arises when we place the two nuclei far from each
other. We expect the resulting energy and wave function to resemble that of
two separated H atoms. I.e the exact wavefunction should approach
something like = |
1s
(r
1
-R
A
)

1s
(r
2
-R
B
)| -|
1s
(r
2
-R
A
)

1s
(r
1
-
R
B
)
|. This form however is not supported by the RHF ansatz of Eq. (4.2.5).
Indeed, if we think of the solution of the RHF equation as being
approximately given by =
A
(r) +
B
(r), where
X
(r) = (r -R
X
),
X = A, B,thentheRHFwavefunctionis:
(x
1
, x
2
) = (
A
(r
1
) +
B
(r
1
)) [

A
(r
2
) +

B
(r
2
)
= ||
A
(r
1
)

A
(r
2
)| +|
B
(r
1
)

B
(r
2
)|]
+||
A
(r
1
)

B
(r
2
)| +|
B
(r
1
)

A
(r
2
)|]
(4.2.7)
The first term is an ionic term, where both electrons are on the same atom
(either A or B) while the second term places one electron on each atom
neutral term. The problem of RHF theory is the ionic term. It may be
importantwhentheatomsareclosebutitshouldgotozerowhentheyarefar.
C. Atomic Orbitals and Gaussian Basis sets
Wheredowegetgoodbasisfunctions?Whatisgood?
We want a small basis that can still describe the electrons. On natural source
are the atomic orbitals of the atoms. These are of the form resembling
exponentialstimepolynomials.Thus,onechoiceis:
_
Im
A
(r) =
Im

A
(r -R
A
)
(4.2.8)
Election Bensity Functional Theoiy Page 99
Roi Baei

Where:

Im

(r) = r
I
c
-

Im
(0, )
(4.2.9)
Where
Im
are the spherical harmonics. One can also take appropriate
combinationsofthesefunctionstomakethemallreal.Thesefunctionshavea
desired analytical property: their ls derivative exhibits a cusp of the correct
order and structure at r = R
A
. There exist analytical formula for doing the
overlap and onebody integrals. But there are no convenient formulae for
the 2body integrals, although some progress was made in recent years (see
articlesbyHandy).
A more convenient, although less natural choice (no cusp). Is the use of
Gaussianfunctions,forexample:

Im

(r) = r
I
_o
,I

c
-[

2
_
Im
(0, ) (4.2.10)
Where o
I,
are called contraction coefficients. These are chosen so that
( o
,I
c
-[

2

) resembles

c
-
. With Gaussian functions very effective and
rapidalgorithmswerepublishedallowingextremelyfast2electronintegrals.
D. Variational-Algebraic approach Hartree-Fock
WehaveseenthattheHartreeFockequationscanbederivedbysearchingfor
that the most general Slater wave function that minimizes the HartreeFock
functional. However implementing a solution to such equations is usually
very difficult, if not impossible in practice. A more practical approach, that
keepsthespiritoftheHartree_FockapproachwasdevelopedbyRoothanand
Hall. In this approach we find the optimal Slater wave function of orbitals
constrained to lie in a finite dimensional vector space spanned by basis
Election Bensity Functional Theoiy Page 1uu
Roi Baei

functions,usuallycalledatomicorbitals(although,uptoapoint,weneednot
assume this) _
c
(x), o = 1, , H. Thus, a set of N molecular orbitals
n
(x)
(n = 1, N) in a determinant
S
|C]of this form must all be of the following
form:

n
(x) = _
c
(x)C
cn
M
c=1
(4.3.1)
The C
cn
coeficients form an H N matrix, called the MO coefficient matrix
forthedeterminant.NotethatforthistomakesensewemustdemandH > N.
The HF energy functional now becomes a function of these coefficients C
cn
.
TheconstrainedthattheMOsareorthonormal,(
n
|
m
) = o
nm
,becomes:
o
nm
= __
c
C
cn
M
c=1
_ _
c
|C
c
|
m
M
c
|
=1
_ = (C
1
SC)
nm
(4.3.2)
WhereweusematrixalgebranotationandtheH HmatrixSisdefinedby:
S
cc
| = (_
c
|_
c
|) (4.3.3)
Thus,theorthonormalityconditionis:
C
1
SC = I
N
(4.3.4)
Where I
N
is the N N unit matrix. Let us now derive an expression for the
expectation value of a onebody operator in a Slater wave function of these
MOs,byEq.(3.6.10):
(
S
|0
`
|
S
) = (
n
|0
`
1
|
n
)
N
n=1
= Ir|C
1
0C] (4.3.5)
Where0istheH HmatrixintheAObasis:
0
cc
| = (_
c
|0
`
1
|_
c
|) (4.3.6)
ItiscustomarytodefinetheH Hdensitymatrix:
Election Bensity Functional Theoiy Page 1u1
Roi Baei

P = CC
1
(4.3.7)
Andwithitwecanwrite:
(
S
|0
`
|
S
) = Ir|C
1
0C] = Ir|P0] (4.3.8)
Inthelaststepweusedthefactthatthetraceoftheproductoftwomatricesis
invarianttotheirorderofmultiplication:
Ir|AB] = (AB)
cc
M
c=1
= A
cn
B
nc
N
n=1
M
c=1
= B
nc
A
cn
M
c=1
N
n=1
= Ir|BA] (4.3.9)
NoticethattheDMhasthegeneralizedidempotencyproperty:
PSP = CC
1
SCC
1
= CC
1
= P (4.3.10)
One can see that P is a symmetric matrix. Furthermore, one can see that it is
positivesemidefinite,i.e.foranyvector::
1
P: = :
1
CC
1
: = (C
1
:)
1
(C
1
:) u.
Furthermore:
Ir|PS] = Ir|CC
1
S] = Ir|C
1
SC] = N
This last step is a result of Eq. (4.3.4). Finally, the 2body operator, by Eq.
(4.1.12),weneedthedirectandexchange.Weuse:
|kl|mn] = C
kc
1
C
c
|
I
|oo
i
|
i
]C
m:
1
C
:in
N
c,c
|
,:,:
|
=1
, (4.3.11)
Then:
[|C] =
c
2
2
C
nc
1
C
c
|
n
|oo
i
|
i
]C
m:
1
C
:im

K|C] = -
c
2
2
C
nc
1
C
c
|
m
|oo
i
|
i
]C
m:
1
C
:in
,
(4.3.12)
Election Bensity Functional Theoiy Page 1u2
Roi Baei

Whereweusetheconventionthatallrepeatedindicesaresummedover:latin
indicesaresummedbetween1andNandgreekindicesbetween1andH.The
summationsonn, mcanbedonefirstandweobtain:
[|C] =
c
2
2
P
c
|
c
|oo
i
|
i
]P
::
|
K|C] = -
c
2
2
P
c:
||oo
i
|
i
]P
c
|
:

(4.3.13)
Thelastexpressioncanbereindexed(assumingPisasymmetricmatrix)as:
K|C] = -
c
2
2
P
c
|
c
|o
i
|o
i
]P
::
|
(4.3.14)
Thus:
[|C] +K|C] =
c
2
2
P
c
|
c
(|oo
i
|
i
] -|o
i
|o
i
])P
::
|
(4.3.15)
Finally,defining:
I
I]
I
cc
|
,::
| = |oo
i
|
i
] -|o
i
|o
i
] = |oo
i
|
i
] -|o
i
|o
i
] (4.3.16)
Using the double indexing I (oo
ii
) and [ = (o
i
o
iii
). Note that I
I]
= I
]I
as
canbeseenfrom:
I
I]
= I
cc
|
,::
| = |oo
i
|
i
] -|o
i
|o
i
] = |
i
|oo
i
] -|o
i
|
i
o] = I
::
|
cc
|
= I
]I

(4.3.17)
Wemaythuswrite:
[|C] +K|C] =
c
2
2
P
t
IP
(4.3.18)

Where now, we consider P not as a H H matrix, but as a column vector of


H
2
elements.P
t
isthecorrespondingrowvector.Similarly,IisnotaH H
H HtensorbutasH
2
H
2
matrix.
Election Bensity Functional Theoiy Page 1uS
Roi Baei

TheHFenergyisthuscompactlywrittenas:
E
HP
|P] = P
t
b +
1
2
P
t
IP
(4.3.19)
Note,thatourunknownvariableisnowtheDMP.WewanttominimizeE
HP

withrespecttoP,however,weneedtoimpose2typesofconstraints.First,we
needtospecifythesubjecttotheconstraints:
Ir|PS] = N
0 = PSPS -PS = u
(4.3.20)
InordertominimizetheenergyweintroducetheLagrangian:
I|P] = E
HP
|P] -p(P
t
S -N) -
t
0 (4.3.21)
The number p and the H H matrix are Lagrange multipliers. The
algebraicHartreeFockequationsarenow
6L
6P
o
= u.Inordertoobtainworking
expressionswederive:
oE
HP
oP
I
= b
I
+
1
2
(P
]
I
]I
+I
I]
P
]
) = b
I
+I
I]
P
]
(4.3.22)
Where again, we use the convention that when a super index appears twice
wesumoverit.Thiscanbewrittenmorecompactlyas:
oE
HP
oP
= b +IP F
(4.3.23)
ThisgradientiswhatwecalltheFockmatrixF.InourpresentnotationFis
aH
2
vectorF
I
withindexI.Butsoonwewillconsideritasamatrixwithtwo
indicesF
cc
|.Theconstraintscaneasilybederivedinasimilarway,leadingto
thefollowingLagrangiangradient:
oI
oP
= F -pS - (SPS +SPS -S)
(4.3.24)
ItiseasytoconvinveonesselfthatFisalwaysasymmetricmatrix,foranyP.
Election Bensity Functional Theoiy Page 1u4
Roi Baei

Theconditionforminimumis:
F -pS - (SPS +SPS -S) = u (4.3.25)
MultplyingbySPfromtheleftwefind:
SPF -pSPS -SPSPS = u (4.3.26)
MultplyingbyPSfromtherightwefind:
FPS -pSPS -SPSPS = u (4.3.27)
Subtracting,weobtain:
SPF -FPS = u (4.3.28)
Thesetofequationsthatneedtobesolvedsimultaneouslyis:
SPF -FPS = u F = b +IP
PSP = P Ir|PS] = N
(4.3.29)
OnepracticalwayofdoingthisistogobacktothematrixC.Intermsofthese,
theequationsbecome:
SCC
1
F -FCC
1
S = u F = b +ICC
1
C
1
SC = I
N
(4.3.30)
Theseequationscanallbemetifwedemandthat:
FC = SCE F = b +ICC
1
(4.3.31)
WhereEisaH H diagonalmatrix.Indeed,fromthisequationwealsohave,
from the symmetry of F and S: C
1
F = EC
1
S. Leftmultiplying by SC we find
SCC
1
F = SCEC
1
S and using the first equation in (4.3.31) on the right hand
sideweobtainthefirstequationin(4.3.30).Furthermore,multiplyingthefirst
equationin(4.3.31)byC
1
wefind:C
1
FC = C
1
SCE.OntheleftwereplaceC
1
F
byEC
1
Sandobtain:
|E, C
1
SC] = u
Election Bensity Functional Theoiy Page 1uS
Roi Baei

WefindthatC
1
SCiscommutativewithadiagoinalmatrix.Ifnotwoelements
on the diagonal of E are equal then C
1
SC is diagonal. We know that the
diagonalentriesmustbepositivesinceSispositivedefinite.Furthermore,we
canchoosethenormofthecolumnsofCsothatalldiagonalelementsofC
1
SC
areequalto1.InthiscasethenC
1
SC = I.Whenthereareseveralelementson
the diagonal of E which are exactly equal, then one can take linear
combinations of the corresponding columns of the Cmatrix, without
disturbing their eigenstatishness. Once can then always create a situation
whichagainallowsC
1
SC = I.Wethusfindthattheprocedureoffindingthe
generalizedeigenstatesandeigenvaluesofFisindeedaprocedureforfinding
theminimum.
ThusEq.(4.3.31)isthealgebraicHartreeFockequation.Inactualcalculations,
it is very common that programs solve self consistently the algebraic HF
equation. This procedure is appropriate for small to medium sized systems.
But for larger system it may be beneficial to directly the minimize of the
Lagrangian, using the gradient in Eq. (4.3.24). Of course, iterations are still
neededbecauseasearchmustbemadefortheLagrangemultiplier.
E. The Algebraic Density Matrix and Charge
Analysis
WehaveseenthatthedensitymatrixisdefinedbytherelationP = CC
1
where
C
un
isthecoefficientoftheAO_
u
(x)in theexpansionofthe MO
n
(x).The
relation of P the the real space density matrix defined in XXX is (we use the
convention that repeated roman indices are summed from 1 to N and
repeatedgreekindicesaresummedfrom1toH):
p(x, x
i
) =
n
(x)
n
(x') = C
un
_
u
(x)_
[
(x')C
[n
= P
u[
_
u
(x)_
[
(x')
Election Bensity Functional Theoiy Page 1u6
Roi Baei

Thus we see that P


u[
determines p. Hence the common name for these two
quantities. Note also that P
u[
determines the density n(r) since that is easily
obtainedbyplacex = x
i
intheDMandintegratingoverspin:
n(r) = lim
q-0
+
P
u[
S
u[
_
u
(r)_
[
(r)
S
u[
+ip
= q
u
n
u
(r) + q
u[
n
u[
(r)
u=[

Where:
n
u
(r) =
_
u
(r)
2
S
uu

Aretheatomicelectronnumberdensities(eachintegratesto1)and
n
u[
(r) =
_
u
(r)_
[
(r)
S
u[
+ip

Is the bond electron number density (again, integrating to 1, or zero). The
atomicchargesarethenq
u
= P
uu
S
uu
andthebondchargesareq
u[
= P
u[
S
u[
.
This form of charge analysis is very popular and allows to obtainintuitive
picturesforthechargedistributioninthemolecule.Whileusefultomany,the
usershouldbewarnedthatthisanalysisisbasisformdependent.Whatwe
mean by this is that if we take different linear combitaions of the same set of
basis functions (i.e. we stay in the same Hilbert space), our charge analysis
will yield totally different results. This is because when we take linear
combinations:
i
= TthenthedensitymatrixchangesbyP
i
= T
1
PT.Thusin
generalthechargesoneachatomcanchangebythisprocedure.
Note that when one integrates over r , on the left hand one gets N. On the
righthandthefirstgives(assumingthebasisfunctionsarenormalizedP
uu

Election Bensity Functional Theoiy Page 1u7


Roi Baei

F. Solving the Hartree-Fock Equations


AplausiblealgorithmforsolvingtheHartreeFockequationsisasfollows:
1. Guess
m
(r),m = 1, , N
c
.
2. Build :
H
(r) (Eq.(4.1.20)), K

(Eq.(4.1.23)) thus determining F


`
(eq.
(4.1.29)).
3. Solvetheeigenvalueequations(Eq.(4.1.28)togetanewsetoforbitals
correspondingtothelowestenergyorbitals.
4. Redofromstep2usingtheneworbitals,untilyouconvergei.e.until
theorbitalschangenomore.
While this algorithm seems reasonable, in practice it rarely converges. There
areseveralwaystomakeanalgorithmpractical.
i. Direct inversion in iterative space (DIIS)
This method, devised by Pulay (P. Pulay, Chem. Phys. Lett. 73, 393 (1980)) is
designed to speed up the convergence. Suppose the iterative process has
producedHiterants:
m
m = 1, , H (Fockians,densitymatricesorsetsofN
c

orbitals).Wecandefineresidualsby:
o:
m
= :
m
-:
m-1
(4.4.1)
Wewanttoproduceanewiterantbyinterpolation:
: = w
m
:
m
M
m=1
(4.4.2)
wherew
m
aretheweightsandtheysumtounity:
w
m
M
m=1
= 1 (4.4.3)
Theseweightsareobtainedbyminimizingtheresidual,assuminglinearity:
o: = w
m
o:
m
M
m=1
(4.4.4)
Election Bensity Functional Theoiy Page 1u8
Roi Baei

Thefunctiontobeminimizedis:
[|w] =
1
2
o:
1
o: -z _ w
m
M
m=1
-1_
=
1
2
w
m
B
m,m
|w
m
|
M
m,m
|
=1
-z _ w
m
M
m=1
-1_
(4.4.5)
Where:
B
m,m
| = o:
m
1
o:
m
| (4.4.6)
Differentiatingwithrespectto
k
w gives:
u =
o[
ow
k
= B
km
w
m
-zw
k
(4.4.7)
Thesolutionoftheseequations,togetherwiththeconstraintsEq.(4.4.3)gives:

11 12 1
1
2
21 22 2
1 2
1 0
0 1
0 1
1 1 1 1 0
M
M
M
M M MM
B B B
w
w
B B B
w
B B B

(4.4.8)
The solution of this equation gives the desired weights. The use of this
algorithmcanbedoneinthefollowingway:
1. Getnew:asoutputfromtheiterativeprocedure.Addittothelisti.e.
designateitas:
M

2. Findweightsfromwhichgetinterpolant:
i
= w
m
:
m
M
m=1
.
3. Use:
i
asinputtotheiterativeprocedureandredofrom1.
ii. Direct Minimization
Sometimes the DIIS procedure is not effective and other methods are tried.
Oneofthemostusefulmethodsistousenumericalminimizationtechniques,
such as the conjugate gradients algorithm to directly minimize the energy of
Election Bensity Functional Theoiy Page 1u9
Roi Baei

the Slater wave function under the constraints. This methods is especially
useful when the molecule being studied is very large. Special tricks are used
to formulate the minimization problem is an unconstraint minimization
(seeforexample,Nunesetal,Phys.Rev.B17611,50(1994)).
G. Performance of the Hartree-Fock approximation
WeexaminetheperformanceofHartreeFockapproximationon,forexample
Formaldehyde.Thereare2sourcesoferror.Oneisintheapplication,sincewe
usefinitebasissets.Thenthereistheintrinsicerror.
In the table below, we see the prediction of various properties of
formaldehyde,calculatedwithincreasingqualityofbasissetandcomparedto
experiment.
Basis R(CO) R(CH) A(O-C-H) Energy(au)
sto-3g 1.2169 1.1014 122.73 -112.35435
3-21g 1.2071 1.0833 122.51 -113.22182
sto-6g 1.2163 1.0981 122.61 -113.44078
6-31g 1.2103 1.0816 121.69 -113.80837
D95 1.2170 1.0843 121.57 -113.83071
D95v* 1.1887 1.0935 121.96 -113.89173
6-311g** 1.1787 1.0949 122.09 -113.89915
6-311++g** 1.1797 1.0943 121.97 -113.90287
apvtz 1.1786 1.0927 121.94 -113.91534
experimental 1.210 1.1020 121.1
WeseeHartreeFockconvergeswhenbasissetqualityincreases.Howeverthe
convergedquantitydeviatessomewhatfromexperimentalvalues.
This deviance exists because HartreeFock theory is only an approximation.
What it assumes is that the electrons act as if they are independent particles
(since it imposes a single determinant). The real groundstate is composed
fromahugeseriesofdeterminants.Theindependentparticlesinteractwith
Election Bensity Functional Theoiy Page 11u
Roi Baei

the mean field of all other particles, while in essence each electrons has to
interact with each other electron, trying as much as possible to avoid it,
withoutpayingtoomuchinkineticorelectronnuclearpotentialenergy.
H. Beyond Hartree-Fock
The HartreeFock method is very successful, since it typically accounts for
over 99% of the electronic energy of molecules. Yet it is not accurate enough
for most applications in chemistry. The reason is that most quantities of
chemical significance are energy differences not absolute energies. When
differences are considered the errors in the HartreeFock approach are not
small.
One way to improve the situation is to approximate the groundstate wave
functionbyaseriesofdeterminants:

gs
= c
|n
1
,,n
N
c
]

n
1

n
N
c

|n
1
,,n
N
c
]

(4.6.1)
where
n
are an infinite orthonormal set of orbitals. Such an expansion can
alwaysbemade,withanysuchset.Wecanthustaketheorbitalsproducedby
theHartreeFockprocess.Thishastheaddednicetythatthefirstdeterminant
isalreadyagoodapproximationtothegroundstate.
Inthiscasewecanclassifythedeterminantsinthefollowingway.Wedivide
theorbitalsintotwosets:oneisthesetof
e
N
HForbitals,calledtheoccupied
orbitals,therestoforbitalsnamedvirtuals.Wethenclassifythedeterminants
bythenumberofoccupiedsmissing.Thuswespeakofallsinglesubstitutions,
doublesubstitutionsetc.
Election Bensity Functional Theoiy Page 111
Roi Baei

A commonly employed method is the configurationinteraction (CI) method.


Ina CI onetakes a number ofdeterminants
u
o = 1, , H andusesthem to
minimizetheenergy:
E|c] = c
m
(
m
|E

|
n
)c
n
M
m,n=1
(4.6.2)
undertheconstraintofnormalization: c
m
(
m
|
n
)c
n
M
m,n=1
= 1.Onecommon
wayofchoosingthedeterminantsthatgointothisexpansionisbycollecting
all single, double, triple etc excitations. A determinant is singly excited if
whencomparedtotheHFdeterminantithasoneoccupiedmolecularorbital
replacedbysomevirtualorbital.Virtualorbitalsareexcitedeigenfunctionsof
the Fock operator. One can show that the singles alone do not allow a
reduction of energy. However, singles and doubles give sometimes good
results.SuchamethodiscalledsinglesdoublesCI(SDCI).Oneproblemwith
this theory is that it is not size consistent. For example, calculating the
energy of 2 distant Helium atoms will not give the twice the energy of one
Heliumatomunderthesameorderoftheory.
Another approach is manybody perturbation theory, called MollerPlesser
theory.Inthisapproach,onewritesthemanybodyHamiltonianas:
E

= F
`
n
M
n=1
+w

(4.6.3)
Wherew

= E

- F
`
n

N
n=1
isconsideredasmallperturbation.Thisquantityis
not small enough and high order MP theory does not converge. However,
secondorderMPtheory,calledMP2,issometimesausefulapproach.Itissize
consistent. However, it relies heavily on the quality of the HartreeFock
Election Bensity Functional Theoiy Page 112
Roi Baei

solution: HartreeFock must be a good reference n which to base a


perturbationtheory.
AllwavefunctionmethodsbeyondHartreeFocktheorybecomequicklyvery
expensive as system size grows. In fact, the numerical cost of good methods
typically scales as 0(N
7
) for the coupled cluster method, which is a size
consistent nonvariational method (variational methods are derived from the
variational theorem), not discussed above. Thus, every enlargement of
numberofelectronsbyafactoroftwomakesthecalculationafactor100more
expensive!
V. Advanced topics in Hartree-Fock
theory
InthischapterwewillcontinueourstudyoftheHartreeFockapproximation,
and look into some of the formal issues, like stability, excitations and
ionization and generalizations like fractional occupations numbers. We will
then discuss the homogeneous electron gas in the HartreeFock
approximationandshowthatitbreaksdownwhentreatingthissystem.
A. Low-lying excitations and the stability of the
Hartree-Fock ground state
i. CI-Singles and Brillouins theorem
For a given system, one can think of the simplest excited states as linear
combinations involving low lying energy determinants. For example, all
determinants where one occupied orbital in the HF determinant is replaced
by an orbitals which is a virtual eigenstate of the Fockian (we call those N
eigenstates of the Fockian which are part of the HF determinant occupied
Election Bensity Functional Theoiy Page 11S
Roi Baei

and those which are not we term virtual). In the algebraic form of HF
theory,ifthereareNelectronsandthesizeofthebasisisH > Nthenthereare
N ways to choose the occupied orbital to be replaced and (H -N) ways to
choose the virtual orbital and so there are N(H -N) such singly excited
determinant.TogetherwiththeHFdeterminantswecanformaN(H -N) +
1dimensionaldeterminantspaceanddiagonalizetheexactHamiltonianinit.
Thiswillgiveanapproximatedescriptionofthelowlyingexcitedstatesofthe
system. In fact, there is no need to include the HF determinant itself in this
schemesinceBrillouinshowedthatforansinglyexciteddeterminant
u
n
(this
notation is for a determinant that is obtained from the HF determinant by
replacingtheoccupiedoorbitalbythevirtualnorbital):
(
u
n
|E

|
0
) = u (5.1.1)
Thus,theHFdeterminant is decoupled fromthesinglyexciteddeterminants
andonecanjustdiagonalizetheHamiltonianinthesinglyexctedspace.This
approachisoftencalledconfigurationinteractionsingles(CIS)anditisa
standardmethodforcalculatingexcitationenergiesinHFtheory.Infact,one
cannotexpectCIStogiveagoodapproximationfortheexcitestates,sincethe
real excited states are intricate linear combinations of determinants with
multiexcited electrons. On the other hand, our approximate ground state
wave function is also just a single (HartreeFock) determinant and it too
missesallthisessentialmixingwithdoublyandtriplyetcexcitations.Onecan
hope however for a mutual cancellation of errors. Indeed, there are many
examples where the CIS method gives quite respectable excitation energies,
evenwhenthewavefunctionsareofquestionablequality.
Election Bensity Functional Theoiy Page 114
Roi Baei

ii. Hartree-Fock Stability


We now consider the question of stability of the HartreeFock solution.
Basicallytheissue is this:howdo we know that we have producedthetruly
lowest energy by occupying the N orbitals with the lowest orbital energy?
Maybeifweusedadifferentansatzwecouldhaveproducedalowerenergy.
In other words, how do make sure that all singly excited determinants are
higherinenergythantheHFdeterminant.
Suppose we have set up the HartreeFock equations and solved them to
obtain a Fockian, a set of orbitals and orbital energies, out of which N
c
are
occupied. Suppose the HartreeFock determinantal wave function is

0
= |
1
, ,
u
, ,
N
c
|. Let us consider the determinantal wave function

u
n
= |
1
, ,
n
, ,
N
c
|obtainedfromtheHartreeFockfunctionbyreplacing
an occupied spinorbital
u
by the unoccupied spinorbital
n
. This can be
viewedasanelectronexcitationprocess:aholeismadein
u
andanelectron
is formed in
n
. The excitation energy for this excited state is the difference
betweentheexpectationvalues:
e
u
n
= (
u
n
|E

|
u
n
) -(
0
|E

|
0
)
= (
n
|b
`
1
|
n
) -(
u
|b
`
1
|
u
)
+
m
(1 -o
mu
)((nm|nm) -(nm|mn))
m
-
m
((om|om) -(om|mo))
m

(5.1.2)
Here, we introduced, for convenience the orbital occupation
m
= 1 for
m N
c
and
m
= uotherwise.Rewriting:
Election Bensity Functional Theoiy Page 11S
Roi Baei

e
u
n
= _(
n
|b
`
1
|
n
) +
m
((nm|nm) -(nm|mn))
m
_
-_(
u
|b
`
1
|
u
) +
m
((om|om) -(om|mo))
m
_
-((no|no) -(no|on))
(5.1.3)
Using the definition of orbital energies in (4.1.30) we can thus write this
electronholeexcitationenergyas:
e
u
n
= e
n
-e
u
-
un
(5.1.4)
where:

un
= (on|on) -(on|no). (5.1.5)
One way to understand
un
is as an overcounting term. We thus see that
excitation energies as calculated as the difference between excited state
energiesinHartreeFocktheoryarenotsimplythedifferencesoftheHartree
Fock orbital energies. They must actually be corrected for overcounting by
subtracting a quantity
un
, which we show henceforth to be manifestly
positive. One conclusion is that orbital energy differences in HartreeFock
form an upperbound to the excitation energies, as determined from the
HartreeFock single excited determinants. We may think of -
un
as the
Coulombenergyofattractionbetweentheexcitedelectronandtheholeitleaves
behind.Indeed
un
iscomposedoftheelectrostaticinteractionoftheelectron

n
and the hole
u
, [
u

c
2

12

u

n
corrected by a corresponding exchange
-[
u

c
2

12

n

u
term.
Toprove
un
ispositive,notethatitissymmetrical:
un
=
nu
.Thus:
Election Bensity Functional Theoiy Page 116
Roi Baei

un
=
1
2
|
un
+
nu
]
=
1
2
__
u

n
_
c
2
r
12
_
u

n
-
n

u
_
+_
n

u
_
c
2
r
12
_
n

u
-
u

n
__
=
1
2
_
u

n
-
n

u
_
c
2
r
12
_
u

n
-
n

u
_
(5.1.6)
Thisshowsthat

un
=
c
2
2

(
u
(1)
n
(2) -
u
(2)
n
(1))
2
|r
2
-r
1
|
J
3
r
1
J
3
r
2
> u
(5.1.7)
Stabilityisobtainedwhene
u
n
> u,or,inotherwordswhene
n
-e
u
>
u
n
.Since

u
n
> u, wefindthatanecessaryconditionforstabilityis:
e
n
-e
u
> u (5.1.8)
Clearly this might not be sufficient, however, it is necessary. That is: for
stability to be possible, the orbital energy of all occupied orbitals must be
lower than that of all the unoccupied orbitals otherwise there will definitely
definitelybesinglyexciteddeterminantswithlowerenergy.
B. Koopmans Theorem
What is the physical meaning of the orbital energies e
m
in Eq. (4.1.28)? This
was first discussed by the DutchAmerican scientist Tjalling C. Koopmans
(Physica 1934, 1, 104.)inhisPhDthesis.(AfterhisPh.D.withHansKramers,
Koopmans began a scientific career in economics. He was awarded the 1975
Nobel Prize in economics for his contribution to the optimal allocation of
resources).
Election Bensity Functional Theoiy Page 117
Roi Baei

Let us consider the ionization energy of a molecule in the HartreeFock


approximation:
IP
HP
= E
HP
|

1
, ,

N-1
] -E
HP
|
1
, ,
N
] (5.2.1)
Note that we calculated the HF energy for two systems, the system N
electronsandtheionizedsystemofN -1electrons.HFapproximationwillin
generalgivedifferentorbitalstothetwosystems,hencethenotation:
n
forthe
N electron system and

n
for the ionized system. Now Koopmans assumed
that the two sets of orbitals are identical. Of course they are not, but it is
known that often they are similar so we neglect their difference. Actually,
when the system is very large and is homogeneous (repeats itself), like an
infinite crystalline solid, this assumption is expected to be exact because the
orbitalsarespreadoutontheentiresystemandthereforeremovalofjustone
out of an infinite number cannot make a difference. For molecules this
assumption is a severe approximation. Nevertheless, under this
approximationweseethatalltheonebodytermscancelexceptthelastanda
largecancelationoftwobodytermstakesplaceaswell.Onlythetermswhich
involvetheremovedorbitalstay,theseincludedirectandexchangeterms:
-IP
HP
= (
N
c
|b
`
|
N
c
) +
1
2
((Nm|Nm) -(Nm|mN))
N
m=1
+
1
2
((mN|mN) -(mN|Nm))
N
m=1

(5.2.2)
Woringouttheexpressiongives:
-IP
HP
= (
N
|b
`
|
N
) + ((Nm|Nm) -(Nm|mN))
N
m=1
= e
N
(5.2.3)
Election Bensity Functional Theoiy Page 118
Roi Baei

The conclusion: e
N
c
is the HF approximation to the ionization energy.
Similarly,e
N
c
-1
approximatesthenextionizationenergyetc.
The flaw in this theorem is the neglect of orbital relaxation. In the next
section we will give a generalized formulation of Koopmans theorem which
is exact. In the section after that we discuss the homogeneous electron gas
whichisasystemforwhichorbitalrelaxationdoesnotexist.

Figure4:EstimationofionizationenergiesinHF,approximateandaccurateDFTmethods.
Twoapproachesaregivenforeachmethod,onebasedonKoopmansapproach,usingthe
orbital energy of the neutral. One can see that Koopmans HF orbital energies usually
overestimatetheIPsby~2eV.Takenfromref.[12].
ExamplesoftheperformanceofKoopmanstheoremwithinHartreeFockand
someDFTbrandsaregiveintheFigure I1.InthefigurethefirstIPisalways
the first left hand (brown) bar.Also shown are calculations for the IPs using
the SCF approach, where the IP is simply the difference between the HF
energiesofthecationandtheneutral.InHFtheorytheKoopmansapproach
for the first IP are off by ~1 eV SCF has errors tend to be a bit larger. The
7.00
6.00
5.00
4.00
3.00
2.00
1.00
0.00
1.00
2.00
BNL*
SCF/TD
BNL*
Koop.
HF
SCF/TD
HF
Koop.
LSDA
SCF/TD
LSDA
Koop.
B3LYP
SCF/TD
B3LYP
Koop.
SAOP KS
D
e
v
i
a
n
c
e

f
r
o
m

e
x
p
e
r
i
m
e
n
t

(
e
V
)
1b1( 12. 61eV)
3a1( 14. 74eV)
1b2( 18. 55eV)
2a1( 32. 20eV)
H
2
O,=0.56a
0
1
6.00
5.00
4.00
3.00
2.00
1.00
0.00
1.00
2.00
BNL*
SCF/TD
BNL*
Koop.
HF
SCF/TD
HF
Koop.
LSDA
SCF/TD
LSDA
Koop.
B3LYP
SCF/TD
B3LYP
Koop.
SAOP KS
D
e
v
i
a
n
c
e

f
r
o
m

e
x
p
e
r
i
m
e
n
t

(
e
V
)
3a1( 10. 88eV)
1e ( 16. 00eV)
2a1( 27. 00eV)
NH
3
,=0.50a
0
1
5.00
4.00
3.00
2.00
1.00
0.00
1.00
2.00
BNL*
SCF/TD
BNL*
Koop.
HF
SCF/TD
HF
Koop.
LSDA
SCF/TD
LSDA
Koop.
B3LYP
SCF/TD
B3LYP
Koop.
SAOP KS
D
e
v
i
a
n
c
e

f
r
o
m

e
x
p
e
r
i
m
e
n
t

(
e
V
)
10a' ( 11. 48eV)
2a' ' ( 12. 55eV)
9a' ( 14. 70eV)
1a' ' ( 15. 80eV)
8a' ( 17. 15eV)
7a' ( 17. 95eV)
6a' ( 21. 90eV)
HCOOH,
=0.46a
0
1
6.00
5.00
4.00
3.00
2.00
1.00
0.00
1.00
2.00
BNL*
SCF/TD
BNL*
Koop.
HF
SCF/TD
HF
Koop.
LSDA
SCF/TD
LSDA
Koop.
B3LYP
SCF/TD
B3LYP
Koop.
SAOP KS
D
e
v
i
a
n
c
e

f
r
o
m

e
x
p
e
r
i
m
e
n
t

(
e
V
)
2b2( 10. 88eV)
1b1( 14. 50eV)
5a1( 16. 10eV)
1b2( 17. 00eV)
4a1( 21. 40eV)
CH
2
O,
=0.49a
0
1
Election Bensity Functional Theoiy Page 119
Roi Baei

usualbrandsofDFT, suchasLSDA,B3LYPhavelargeKoopmanserrorsbut
smallSCFerrors.ThebrandcalledBNLhassmallerrorsinbothrespects.
C. Fractional occupation numbers, the HF orbital
functional and the generalized Koopmans
theorem
We now make a fundamental generalization of the HartreeFock energy.
Instead of viewing it as an expectation value of a determinantal wave
function, we view it as a new fundamental concept: an orbital functional. We
take Eq. (4.1.9) and write it as a functional of all orbitals and occupation
numbers:
E

HP
|
1
,
2
, ;
1
,
2
, ]
=
n
(
n
|b
`
1
|
n
)
n
+
1
2

m
((nm|nm) -(nm|mn))
nm

(5.2.4)
Here the functional depends on an (in principle) infinite set of orthonormal
orbitals and their corresponding occupation numbers. Since electrons are
fermionseachoccupationnumberislimitedtotheunitinterval,i.e.u

1
(sinceyoucannothavenegativeoccupationandyoucannothavemorethan1
electroninagivenorbitalbecauseofthePauliprinciple).Letusminimizethis
functional for N electrons. What we mean by this is that we minimize this
energy E

HP
with respect to the orbitals
n
(r) and the occupation numbers
n

under the constraints that the orbitals are orthonormal and the occupation
numbersarenonnegative,notgreaterthan1andthattheysumuptoN.Thus
theLagrangianforthisconstraintminimizationis
Election Bensity Functional Theoiy Page 12u
Roi Baei

HP
|
1
,
2
, ;
1
,
2
, ]
= E

HP
|
1
,
2
, ;
1
,
2
, ] -
n
e
n
|(
n
|
n
) -1]
n
-__
n
n
-N_
(5.2.5)
The derivative with respect to the orbitals give the analogous HF equations,
whicharesimilartoEqs.(4.1.28)withanimportantdifference:allsumsonN
orbitals are replaced by weighted sums: X
n
-
n
X
n n
N
n=1
. In essence, the
usual HF theory is the constant occupation numbers
n
= ]
1 n N
u n > N
. The
newequationsare:
b
`
1
|
n
) +:
`
H
|
n
) +K

H
|
n
) = e
n
|
n
)
(x|:
`
H
|) = c
2
__

m
|
m
(x
i
)|
2
|r' -r|
m
_Jx
i
(r)
[xK

= -c
2

m
(x) _

m
(x
i
)
-
(r
1
)
|r
i
-r|
Jx
i
n

(5.2.6)
Now,letsdiscusstheequationsobtainedbydemandingvariationalbehavior
with respect to
n
. Since we have a constraint that
n
e |u,1], we must
differentiatebetweenthreecases:the
n
= u,
n
= 1andfractional(u <
n
< 1)
cases. To derive with respect to
n
is allowed only when you are not at
constraint boundaries. So in the fractional case an arbitrary infinitesimal
changein
n
isindeedmeaningfulandthederivativeofI

withrespectto
n
is
equaltozeroatthevariationallyoptimalpoint.Thisgives:
_ =
oE

HP
o
n
= (
n
|b
`
1
|
n
) +
m
((nm|nm) -(nm|mn))
m
= e
n

(5.2.7)
This equation shows that all fractionally occupated orbitals must be of the
same energy _. We can divide all orbitals to be full (
n
= 1) or empty
Election Bensity Functional Theoiy Page 121
Roi Baei

(
n
= u) and partially occupied but these latter orbitals all have the same
orbital energy. We have seen that for the HartreeFock solution to be stable,
only the lowest orbitals can be occupied. Thus the picture that emerges is:
there are N -] fully occupied orbitals. Then there can be several degenerate
orbitals with fractional occupation that adds up to ] (assuming N is integer).
Therestoftheorbitalsareunoccied..
Eq. (5.2.7) is now an exact formulation of Koopmans theorem. Suppose we
have a system with slightly less than an integer number of electrons: N -p
wherepisasmallfraction.Then:
E

HP
|N] -E

HP
|N -p] = e
H
(N)p +0(p
2
) (5.2.8)
Where e
H
(N) is the energy of the highest (can be partially) occupied orbital
fortheNeletronssystem.Thusweseethatifweslightlyionizethesystem
Koopmanstheoremholdsexactly(inasense,thechangeissosmallthatthere
will be no orbital relxation). Of course, in real molecules there is no such
thingasafractionalelectron.Butstill,intermsoftheorbitalHFtheorythereis.
Thisnewconceptofanorbitalfunctionalhasallowedustoconsideranexact
andgeneralizedversionofKoopmanstheorem.Wewillseeinlaterchapters
that orbital functionals play an important role in advanced approaches to
DFT.
D. Hartree-Fock for the homogeneous electron gas
i. Hartree-Fock orbitals and orbital energies of HEG
LetusnowapplytheHartreeFocktheorytoanimportantsystemwhichisa
modelforthevalenceelectronsofasimplemetalsuchassodium.Thisisonce
againthehomogeneouselectron,gasofNelectronsinacubicboxofvolume
I under periodic boundary conditions. Note that we are imposing on our
Election Bensity Functional Theoiy Page 122
Roi Baei

electrongastobeuniform,althoughitisnotnecessarilyknownthatthisisthe
lowest energy solution. In HartreeFock theory for example there are non
homogeneous (symmetry broken) states which lead to lower energy. We will
not study these here however. The density is n = NI. For an interacting
systemonemusthavechargeneutrality,soastaticpositivechargeofthesame
density as the electrons is smeared in the box. This positive charge is called
Jellium. The ee direct term, the Jellium self energy (positivepositive
interaction)andtheelectronJelliumenergymustallcanceleachother:
[
cc
+] :nJ
3
r +]
pp
=
c
2
n
2
2

J
3
rJ
3
r
i
|r -r
i
|
-c
2
n
2

J
3
rJ
3
r
i
|r -r
i
|
+
c
2
n
2
2

J
3
rJ
3
r
i
|r -r
i
|
= u
(5.2.9)
SointheHartreeFockapproach,allthatisleftarethekineticenergyandthe
exchangeenergies:
e
HP
(n) = t(n) +e
X
(n) (5.2.10)
TheHFequationsare:
-
l
2
2m
c
7
2

k
(r) +K

k
(r) = e
k

k
(r) (5.2.11)
Thesolutionsmustbeplanewaves(sinceallpointsareequivalent).Thus:
e
k
=
l
2
k
2
2m
c
+X
k
-1
(5.2.12)
And:
K

k
= X
k
-1

k
(5.2.13)
Let us calculate the exchange eigenvalues X
k
-1
.Assume double occupancy of
allorbitals.Then:
Election Bensity Functional Theoiy Page 12S
Roi Baei

k
(r) = -
2c
2
I
32
c
qr
_
c
-qr
|
c
kr
|
|r -r
i
|
J
3
r'
|q|c
q
<c
F
|
= -
2c
2
I
32
c
qr
_
c
(k-q)r
|
|r -r
i
|
J
3
r
i
|q|c
q
<c
F
|
= -
2c
2
I
32
c
kr
_
c
(k-q)(r
|
-r)
|r
i
-r|
J
3
r
i
|q|c
q
<c
F
|

(5.2.14)
Here the summation symbol with |q|e
q
< e
P
| means that we sum over all
momentumstatesqwithe
q
< e
P
. Wenowdiscusstheevaluationof]
c
iwx
|x|
J
3
x.
Letsassumetheboxisa sphereofradiusR and adda damping factor p > u
(whichcanbesettozeroafterthecalculationisdone).Forafinitevolume
_
c
wx
x
c
-qx
J
3
x = _ c
-qx
_
c
wxcos 0
x
sin0 J0
n
0
R
0
2nx
2
Jx
= _
c
(w-q)x
-c
(-w-q)x

iwx
2
R
0
2nx
2
Jx
=
2n
iw
_
c
(w-q)R
-1
(iw -p)
+
c
(-w-q)R
-1
(iw +p)
_
=
2n
iw
_
(c
(w-q)R
-1)(iw +p) +(c
(-w-q)R
-1)(iw -p)
(iw -p)(iw +p)
_
=
4n
w
2
+p
2
j(1 -c
-qR
cos(wR)) -
p
w
c
-qR
sin(wR)[
TakethelimitR - andafterthatthelimitp - u.Thiswillleadtotheresult
4n
w
2
.ForafiniteR,takethep - ulimitfirst:
_
c
wx
x
J
3
x =
4n
w
2
]
1 -cos(wR) R finite
1 R -

The resultof inifinte R becomes undefined in the second case. So we use the
first.Thuswewrite:
Election Bensity Functional Theoiy Page 124
Roi Baei

k
(r) = -
2
I
32
c
kr
4nc
2
(k -q)
2
|q|c
q
<c
F
|
=
1
X
k

k
(r)
(5.2.15)
Where:
1
X
k
= -
2
I

4nc
2
(k -q)
2
|q|c
q
<c
F
|

(5.2.16)
Now, toproceed, wewant to assume that themomentum statesgetfilledup
in just the same way as they did in the ThomasFermi approach for a non
interacting gas. This will be valid if the orbital energies e
k
are ascending
functionsofk,whichaccordingtoEq.(5.2.12)isvalidwhen
1
X
k
isanincreasing
function of k as well. We can only know that however, after we evaluate the
summation in Eq. (5.2.16) What we can do, is work in the spirit of the self
consistent field approach. We will assume that e
k
increasing functions of k
then sum Eq. (5.2.16) and check, for self consistency, that
1
X
k
is an increasing
function ofk.Underthis plan,wereplacein Eq.(5.2.16) thesummationover
all q with e
q
< e
P
by a summation over all q with q < k
P
, which can be
approximatedbytheanintegral(Eq.(2.1.8)),wefind:
1
X
k
= -
2
(2n)
3
_
4n
(k -q)
2
J
3
q
k<k
F

(5.2.17)
Toevaluatethisintegralwepassovertosphericalcoordinates,k,0and:
1
X
k
= -
2
n
_ _
sin0J0
q
2
+k
2
-2kqcos0
q
2
Jq
n
0
k
F
0
(5.2.18)
The factor of 2n is a consequence of the integral over the angle (the
integrand isindependentof). Theintegralover0iscanbeperformedafter
changeofvariablesx = sin0,notingthat]
dx
q
2
+k
2
-2kqx
1
-1
= [
In(q
2
+k
2
-2kqx)
-2kq

-1
1
so:
Election Bensity Functional Theoiy Page 12S
Roi Baei

1
X
k
= -
2
kn
_ ln_
k +q
k -q
_ qJq
k
F
0
-
2k
P
n
F _
k
P
k
] (5.2.19)
Wechangedvariabletox =
q
k
andgavetheintegraloverxanewname:
F(x) =
1
x
_ ln_
1 +x
1 -x
_ xJx
x
0
= 1 +
1
2
_x -
1
x
] ln_
x +1
x -1
_ (5.2.20)
ThisfunctionisshowninFigure V5.Itismonotonicallyincreasingfromzero
reaching2whenx - .

Figure V5:ThefunctionF(x)definedinEq.(5.2.20)
TheHForbitalenergiesarenow
e
k
=
l
2
k
2
2m
c
-
c
2
4ne
0
2k
P
n
F _
k
P
k
] (5.2.21)
Clearly, as F(x) is increasing as a function of x, it is decreasing with k, yet it
comes into the orbital energy expression with a minus sign so overall e
k
is a
increasingfunctionofk.Aplotofe
k
,inatomicunitsisgiveninFigure V6.It
isseenthatindeede
k
isincreasingmonotonicallywithk.
Election Bensity Functional Theoiy Page 126
Roi Baei

Figure V6:TheHartreeFockorbitalenergiese
k
forvariousvaluesofk
F
(allinatomic
units).Foranygivenk
F
thefunctione
k
isincreasingwithk.
Exercise:DeterminetheHartreeFockdensitymatrixoftheHEG.
We have shown that the HartreeFock orbitals are plane waves with wave
vectorslessthank
P
.Thus:
p(r, r
i
) =
1
I
c
k(r-r
|
)
k<k
F

(5.2.22)
Denotingr -r
i
xwefindthatpdependsonlyonsandso:
p(s) = 2
1
(2n)
3
_ c
kx
J
3
k
k<k
F
=
2
(2n)
3
_ _ c
ksx
Jx
1
-1
2nk
2
Jk
k
F
0
=
2
(2n)
3
4n
s
3
_ sinx xJx
sk
F
0
=
1
(2n)
3
8n
s
3
(sinx -x cos x)
0
sk
F
=
k
P
3
n
2
_
sinsk
P
-sk
P
cos sk
P
(sk
P
)
3
]
(5.2.23)
Sincek
P
3
= Sn
2
nwefind:
p(s) = Sn _
sinsk
P
-sk
P
cos sk
P
(sk
P
)
3
] (5.2.24)
Election Bensity Functional Theoiy Page 127
Roi Baei

Weknowthatp(r, r) = p(u) = n.AplotoftheDMisplottedhere:

ii. The density of states of the HEG


An interestingquantity, determiningmanyof thepropertiesofthesubstance
isthedensityofsingleparticlestates(DOS)(e
P
)attheFermilevel.Onecan
calculateandobtain:
(e) =
I
(2n)
3
_ o(e -e
k
)4nk
2
Jk
k
F
0
=
I
(2n)
3
_ o(e -e
k
)4nk(e
k
)
2
Jk
Je
k
Je
k
c
F
0
=
I
2n
2
0(e
P
-e)k(e)
2
k'(e)
(5.2.25)
Hereweneedtoinverttherelatione
k
tok(e).Clearly,onecanwritethisas:
(e
k
) =
I
2n
2
0(e
P
-e
k
)k
2
(e
i
(k))
-1

(5.2.26)
Thus,toobtain theDOS we needtotaketheinverseof thederivative e
k
i
.Let
us look more closely into that. Note that for free electrons
k
2
c
|
(k)
= 2p
3
e thus
(e) e.Inparticular,theDOSattheFermilevelisfinite,atypicalsituation
for metals, which is what the HEG is. Now, let us look what happens
according to HF theory. First, let us look numericallygraphically at e
k
for
some arbitrary definite value k
P
= 1. This is shown in Figure V7 (left). It
seems perfectly OK. However, when we plot the derivative e
k
i
, one notices a
-20 -10 10 20
s k
F
0.2
0.4
0.6
0.8
1.0
rHsLn
Election Bensity Functional Theoiy Page 128
Roi Baei

divergentbehavioratk - k
P
.Indeed,takingthe derivativeofEq. (5.2.21)we
obtain,usingF
i
(x) =
1
x
[-1 +
1
2
[x +
1
x
ln
x+1
x-1
:

Figure V7:Thefunctionse
k
(topleft)ande
k
i
(topright)andtheDOSoftheHEG,
accordingtoHartreeFocktheory.
e
k
i
=
l
2
k
m
c
+
c
2
4ne
0
2
n
k
P
k
_-1 +
1
2
_
k
P
k
+
k
k
P
] ln _
k
P
+k
k
P
-k
_] (5.2.27)
Indeed, as k - k
P
there is a logarithmic divergence of this expression. Since
theDOSisproportionaltotheinverseofe
k
i
,thismeansthatate
P
theDOSis
zero! This prediction by HartreeFock theory can easily be checked
experimentally.Forexample,electronicconductionofmetalsishigh.Sincethe
conduction depends on the availability of electrons at the Fermi level we see
thatHFpredictssmallmetallicconductance,failingmiserably...
0.0 0.2 0.4 0.6 0.8 1.0
-1.2
-1.0
-0.8
-0.6
-0.4
-0.2
k
e
k
0.0 0.2 0.4 0.6 0.8 1.0
0
2
4
6
8
10
12
k
d
e
k
d
k
0.0 0.2 0.4 0.6 0.8 1.0
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
k Hfor kF=1L
D
O
S
H
a
r
b
u
n
i
t
s
L
Election Bensity Functional Theoiy Page 129
Roi Baei

iii. Stability of the HEG in Hartree-Fock theory


Thedensitymatrixassumingunpolarizedgas(allelectronsarespinpaired)is
given by Eq. (5.2.24). Now, we can calculate the total exchange energy. From
Eq.(4.1.17):
K = -
1
4

p(r, r
i
)
2
|r -r
i
|
J
3
rJ
3
r
i
= -
I
4
_ _
sink
P
R -k
P
R cos k
P
R
n
2
R
3
]
2
4nRJR
v
13
0
=
= -
Ik
P
4
n
3
_ _
sinx -x cos x
x
3
]
2
xJx
k
F
v
13
0

(5.2.28)
Notice that ] [
sInx-x cos x
x
3

2
xJx
x
0
=
cos 2x+2x sIn2x-1-2x
2
8x
4
+
1
4
. Thus for x -
whichisthelimitoflargevolume,wehave:
_ _
sinx -x cos x
x
3
]
2
xJx

0
=
1
4
(5.2.29)
Thusforlargevolume:
K = -
Ik
P
4
4n
3

(5.2.30)
From(2.1.15)n =
k
F
3
3n
2
so:
K = -
Sk
P
4n
N
(5.2.31)
Theexchangeenergyperparticleis:
e
x
= -
Sk
P
4n
= -C
X
n
13

(5.2.32)
WithC
X
=
3
4
[
3
n

13
= u.7S8S6.Letusrecallthekineticenergyperparticlefor
theHEG(Eq.(2.1.18)):
Election Bensity Functional Theoiy Page 1Su
Roi Baei

e
HP
= t +e
x
= C
1P
n
23
-C
X
n
13
(5.2.33)
WithC
1P
= 2.871.Weseethatforhighdensitytheenergyisprimarilykinetic
and rises with density. For low density the energy drops as the density is
increased.Thusthereisanequilibriumpoint.ThiscanbeseeninFigure V8.

Figure V8:Thetotalenergy,accordingtotheHartreeFockapproximation,ofJellium,asa
functionofthedensityn.Allquantitiesinatomicunits.
WeseethatJelliumismoststableatdensityn
-
calculatedfrom:
e
HP
i
=
2
S
C
1P
n
-13
-
1
S
C
X
n
-23
= u - n
-
= _
C
X
2C
1P
]
3

(5.2.34)
Thisgives:n
-
= u.uu21o
0
-3
ande
-
= -u.u48E
h
= -1.29 ev.
Forcomparison,sodium,whichisamonovalentmetalwithpropertiessimilar
to Jellium has a total energy per atom (valence electron) of: e
Nu
= -1.1S cI
andn
Nu
= u.uuS8 o
0
-3
.
0.000 0.005 0.010 0.015 0.020 0.025
-0.04
-0.02
0.00
0.02
n
e
H
F
Election Bensity Functional Theoiy Page 1S1
Roi Baei

VI. The Hohenberg-Kohn density theory


In view of the poor predictions of chemical bonds and molecular properties
affordedbyHFapproximationandthehighnumericalpriceofwavefunction
approaches, it is beneficial to seek out methods that circumvent the need to
represent the manybody electronic wavefunction. We studied in detail two
theories. One was based on the density, but had no real rigorous basis. The
otherwasamethodthatassumedtheelectronicwavefunctionisoftheform
applicable only for noninteracting electrons. We now want to describe a
rigorousmethodthatcombinesideasofthistypeinanewwaywhichisboth
rigorousandleadstoveryaccurateapproximations.
A. The first HK theorem
InelectronicstructuretheorytheHamiltonianisgivenas:
E

= I
`
+u

+_:(r)n(r)J
3
r (6.1.1)
Where all symbols have been defined in XXX. The identity of the molecular
system is captured in the external potential :(r). The other terms are
universali.e.thesameforallmolecules.InDFTtheyhaveaspecialsymbol:
F
`
= I
`
+u

. (6.1.2)
DifferentHamiltoniansdifferonlybytheirexternalpotentialdensityterm:
E

-E

i
= _(:(r) -:
i
(r))n(r)J
3
r (6.1.3)
This observation has a fundamental implication. Suppose we have two
electronic wave functions
1
and
2
of N
c
electrons which have the same
density,i.e.forallr:
(
1
|n(r)|
1
) = (
2
|n(r)|
2
) n(r). (6.1.4)
Now,if
Election Bensity Functional Theoiy Page 1S2
Roi Baei

(
1
|E

|
1
) < (
2
|E

|
2
) (6.1.5)
Then:
(
1
|F
`
|
1
) < (
2
|F
`
|
2
) (6.1.6)
Thustheinequalityisindependentofthepositionofthenuclei:onlythewave
functions affect it through the universal operator F
`
. The external potential
termdropsoutbecausebothwavefunctionshavethesamedensity.
An interesting natural conclusions is that if Eq. (6.1.5) holds for one
HamiltoniantheitholdsforallHamiltonians::
(
1
|E

'|
1
) < (
2
|E

'|
2
) (6.1.7)
This fact will now be used to prove the first theorem of DFT, due to
HohenbergandKohn:
Theorem (HohenbergKohn): When two Hamiltonians differing only by a
single particle potential term E

-E

i
]
|:(r) -:
i
(r)]n(r)J
3
r have non
degenerate ground states which integrate to the same density then these
Hamiltoniansareidenticaluptoaconstant(i.e.:(r) = :
i
(r) +const).
ProofoftheHKtheorem:Assumeotherwise:istheGSofE

and
i
thatof
E

i
, both wave functions assumed real and have the same expectation values
for the density at all points in space. The variational principle for dictates
(|E

|) < ('|E

|')which,asdiscussedinEq.(6.1.7)meansthat(|E

'|) <
('|E

'|') holds as well. But this latter inequality contradicts the variational
principlefor
i
asthegroundstateofE

',unlessitdiffersfromE

byatmosta
constant.
Thistheoremallowsonetothinkofthepotentialasafunctionaldependenton
thedensity.Thus,inadditiontotheusual
: - E

- - n (6.1.8)
Election Bensity Functional Theoiy Page 1SS
Roi Baei

Wenowhave:
n - : - E

- (6.1.9)
Thuseverythingaboutthemolecule(allitsproperties)isintheabovesense
afunctionalofthegroundstatedensity.
Ageneralizationofthetheorem,givinganinequalityis:
Theorem:Ifn

(r)isthedensityofthenondegenerateNparticlegroundstate
of E

= F
`
+]:

(r)n(r)J
3
r, where i = 1,2. Then, denoting X = X
1
-X
2
, we
have:
n = u = _n(r):(r)J
3
r < u (6.1.10)
Proof: Suppose n = u. Then
1
=
2
and because of nondegeneracy the
followinginequalityisstrict:
(
1
|E

1
|
1
) < (
2
|E

1
|
2
) = (
2
|E

2
|
2
) +_|:
1
(r) -:
2
(r)]n
2
(r)J
3
r (6.1.11)
DenotingE

= (
I
|E

|
I
),wefind:
E
1
< E
2
+_:(r)n
2
(r)J
3
r (6.1.12)
Andexchangingtheindices1 - 2:
E
2
< E
1
-_:(r)n
1
(r)J
3
r (6.1.13)
AddingthetwoinequalitiesandcleaningupgivesEq.(6.1.10),QED
B. The HK functional
Since everything is a functional of the density, we can assert that the
groundstate kinetic energy is a functional I|n] of the density and so is the
Election Bensity Functional Theoiy Page 1S4
Roi Baei

electron repulsion energy u|n]. We can thus define the HohenbergKohn


functionalofvrepdensitiesasfollows:
F
HK
|n] = I|n] +u|n] = (
gs
|n]|I
`
+u

|
gs
|n]) (6.2.1)
Where
gs
|n] is a groundstate wave function with density n. F
HK
|n] is a
universalfunctional,itisnotlimitedtoanyparticularmolecularsystem.Itis
validforallsystems.WehaveofcoursenopracticalwaytocalculateF
HK
|n]in
general.
Even the domain of definition of F
HK
is difficult to characterize. In fact, and
perhapsunexpectedly,thisdomainisnotevenconvex,asdiscoveredbyLevy
andPerdew.

Figure VI1:AconvexsetisasetofpointssuchthatifAandBareinthesetthenanypoint
ConthestriaightlinejoiningAandBisinthesetaswell.Theleftsetisconvexwhilethe
rightsetisnot.
Thismeans,thatifn
0
andn
n2
aretwodensitieswhichareinthedomain,itis
not guaranteed that the convex sum n
0
(cos
2
0)n
0
+(sin
2
0)n
n2
(where 0
is some parameter in the range ju,
n
2
[) is in it. Levy and Perdew showed an
example where convexity fails by considering the case where both are
densities are densities of two degenerate eigenstates of the same system (we
discuss this in more detail below). Technically, we say that the domain of
definition is not convex. However, suppose we do have three densities n
0
,
n
n2
and n
0
, where the latter is the convex sum of the two former densities.

Election Bensity Functional Theoiy Page 1SS


Roi Baei

SupposefurtherthatallthreebelongtothedomainofdefinitionofF
HK
.Then
F
HK
|n
0
] (cos
2
0)F
HK
|n
0
] +(sin
2
0)F
HK
|n
n2
], i.e. F
HK
is a convex functional
of the density (we discussed this concept in Chapter XXX). This is seen as a
result of the variational theorem. Suppose
0
is the groundstate wave
function, :
0
the potential and E

0
= I
`
+u

+I
`
0
the Hamiltonian
correspondington
0
then:
(

|E

) (
0
|E

|
0
)
(

|E

) (
2
|E

|
2
)
(6.2.2)
Fromthiswehave:
F
HK
|n
0
] = (

|E

) -_n
0
(r):
0
(r)J
3
r
(cos
2
0) _(
0
|E

|
0
) -_n
0
(r):
0
(r)J
3
r_
+(sin
2
0) _(
2
|E

|
2
) -_n
n2
(r):
0
(r)J
3
r_
= (cos
2
0)F
HK
|n
0
] +(sin
2
0)F
HK
|n
n2
]
(6.2.3)
We will see that the convextity property of F
HK
|n] has an important
implication for the variational property of density functional theory, that we
discussinthenextsection.
C. Minimum principle for density functional theory
The second HK theorem establishes a minimum principle involving the
density,anditcanbeusedtofindthedensitywithoutdirectreferencetothe
conceptofawavefunction.
Givenapotential:(r),considerthefollowingfunctionalforNelectrons:
E

|n] = F
HK
|n] +_:(r)n(r)J
3
r (6.3.1)
Election Bensity Functional Theoiy Page 1S6
Roi Baei

TheHKtheoremIIstatesthatthedensityn

whichisthegroundstateof:(r)
minimizesthisfunctional.Forn

weknowofcoursethat:
E

|n

] = (
gs
|n

]|E

|
gs
|n

]) = E
gs
(6.3.2)
WhereE

= I
`
+I
`
+u

.HKtheoremIIstatesthatforanyothervrepdensityn'
of

Nelectrons:
E

|n

] < E

|n
i
] (6.3.3)
The proof is an immediate consequence of tha quantum mechanical
variationalprinciple:
E

|n
i
] = (
gs
|n']|E

|
gs
|n']) > (
gs
|n

]|E

|
gs
|n

]) = E
gs
(6.3.4)
This theorem allows one to speak of a minimumprinciple concerning the
density.IfwehaveanapproximateE

|n]wecanfindanapproximationtothe
groundstatedensityn

,simplybyfindingtheminimum.
ThefactthatF
HK
|n]isconvex(inthelimitedsense,sincethereisstilltheissue
of domain of definition discussed below), as proved in the previous section,
implies that E

|n] is convex as well, since by adding a linear term to a


functiononecannotchangeitsconvex/nonconvexcharacter.Theconvexityis
desirable since it assures that the minimum is not only global, but that there
arenolocalminimaaswell,sinceaconvexfunction(andfunctional)haveno
trulylocalminima.Thereisstilltheproblemoftheconvexityofthedomainand
this spoils this useful conclusion. Below we show how to construct a
functionalwithaconvexdomain.Butthen,welosetheconvexpropertyofthe
functional.
We now derive the basic equation of DFT (ignoring for the time being the
problem of convexity of the domain). We need to minimize E

|n] under the


constraintthat]n(r)J
3
r = N
c
.ThusweneedtominimizetheLagrangian:
Election Bensity Functional Theoiy Page 1S7
Roi Baei

I
,N
c
|n] = E

|n] -p __n(r)J
3
r -N
c
_ (6.3.5)
Where the value of p is varied until the constraint is respected. The final
minimizingdensityobeys:
u =
oI
,N
c
|n]
on(r)
=
oE

|n]
on(r)
-p =
oF
HK
|n]
on(r)
+:(r) -p
(6.3.6)
This is the basic equation of DFT. It has no direct mention of the wave
function. Once we find an approximation for E

|n] we can get an


approximationforn

(r)fromthisequation.
The two theorems of HK put some rigor into the ThomasFermi
approximation.InthisKHtheoryE
1P
|n]

isanapproximationtoE

|n]andthe
TFequationsareanapplicationof(6.3.5).Still,weknowthatTFtheoryisvery
poor for chemistry. This means that despite the added rigor, the TF
approximationistoocumbersomeforquantumchemistry.
D. An interesting observation on the variational
principle of non-interacting electrons
ConsiderasystemofNnoninteractingparticlesinapotential:(r).Usually
wemayassumethatthegroundstateofthissystemisaSlaterwavefunction
= uet|
1

N
]. The variational theorem states that their ground state
energyisgivenbyminimizing:
E|:] = (
n
|I
`
1
+:|
n
)
N
n=1
- e
n
((
n
|
n
) -1)
N
n=1
(6.4.1)
Where I
`
1
= -
1
2
v
2
, e
n
are Lagrange multipliers imposing the unit norm of
eachorbital.Minimizingleadstotheequations:
(I
`
1
+:)
n
= e
n

n
(6.4.2)
Election Bensity Functional Theoiy Page 1S8
Roi Baei

AftersolvingfortheNlowestenergyeigenstatestheenergyisE|:] = e
n
N
n=1
,
the kinetic energy is I|:] = (
n
|I
`
1
|
n
)
N
n=1
the density is
n(r) = |
n
(r)|
2 N
n=1
.
Now consider a different problem. Suppose the density is given and one is
required to find the system with this density that has the minimum kinetic
energyI.ThisleadstotheLagrangian:
I|n] = (
n
|I
`
1
|
n
)
N
n=1
+_:(r) _|
n
(r)|
2
N
n=1
-n(r)_
- e
n
((
n
|
n
) -1)
N
n=1

(6.4.3)
Now:(r)areLagrangemultipliersandmustbesearchedforinorderthatthe
constraint n(r) = |
n
(r)|
2 N
n=1
be fulfilled. After minimizing we obtain the
equations:
(I
`
1
+:)
n
= e
n

n
(6.4.4)
This equation rises from the attempt to compute I|n] and is the same as Eq.
(6.4.2) which rose as when : was given and E|:] was calculated. This shows
that minimizing the kinetic energy under a given density invokes the
sameequations as minimizing the energy when the potential is given. This
factwillbeusedintheKohnShammethod.
E. The set of V-representabile densities
i. V-rep densities correspond to ground states wave function of some
potential well
The HK theorem shows that the ground state density of a system uniquely
determinestheonebodypotential.Thisisauniquenessstatement:thereisat
Election Bensity Functional Theoiy Page 1S9
Roi Baei

most one potential associated with a density. An interesting twist is the


reverse question: what are the conditions that a given density is the ground
state density (GSD) of some system? Of course there are some preliminary
conditions on the density: it must be nonnegative and it must integrate to a
positiveinteger:
n(r) u
_n(r)J
3
r = N
c

(6.5.1)
But in general, we have no good criterion for deciding whether a given
density n(r) is a GSD of some potential :(r). Densities which are GSDs of a
potentialarecalledvrepresentable.Inshortvrep.
When we say that everything is a functional of the density, we mean
everythingisafunctionalofavrepdensity.
ii. Some non-v-representabilit issues
We have seen in XXX that ground state wave functions of single particles is
nodeless.Acorollaryfromtheaboveanalysisisthatadensityofoneparticle
withanodeisnotvrepresentable.
However,thedensitydoesnotneedtoactuallydevelopazeroforthedensity
tobenonvrep.ConsidertheexamplebyEnglischandEnglisch:
n(x) (1 +(x
2
)
u
)
2
c
-x
2

1
4
o <
1
2
,
If(x)isawavefunctionthenthepotentialisgivenby:(x) =
1
2

||
(x)
(x)
.Using
theaboveformforo =
1
3
,wefindthatthepotentialisinfiniteattheorigin:
Election Bensity Functional Theoiy Page 14u
Roi Baei

Figure VI2:Examples,followingEnglischandEnglisch,ofanonvrepdensity(inpurple)
andthecorrespondingpotentialwhichissingular.Theleftpanelisa1Dexamplewhich
therightpanelistheradial3Dexample.
In 3 dimensions simply replace x by r and
ii
(x) by v
2
(r) =
1

2
(r
2
(r))
ii

thus: :(r) =
1
2
[
2
()
||

2
()
-
1

2
. The first term of the potential is seen to
dominatedbyatermgoingas
1

2
whenr - u.Thisisacentrifugalbarrierwith
= 1. Subtracting this barrier exposes the bare potential, shown in Figure
VI2.
iii. The set of v-rep densities of a given electron number is not convex
Degenerate Hamiltonians can generate nonvrep densities quiet easily. Thus,
nonvrepdensitiesaremuchmoreabundantthanonemaysuspect.Letussee
why there is a problem, in the following analysis due to Levy and Lieb
(developedseperatelyatthesametimemoreorless).
SupposeE

= I
`
+]:(r)n(r)J
3
r +u

isaHamiltonianwithdegenerateground
states, of energy E and full degeneracy . Thus,

,i = 1, , are Q ground
state wave functions with (

|
]
) = o
]
. There are infinitely many ways to
define the

. We select one arbitrarily. Denote: n

(r) = (

|n(r)|

). Then,
considerthedensitybuiltasaconvexsumofthesedegeneratestatedensities:
Election Bensity Functional Theoiy Page 141
Roi Baei

n(r) = c

(r)

=1
c

> u c

=1
= 1
(6.5.2)
Wenowshowthatthisdensityitisusuallynonvrepresentable.Ofcourse,it
may happen (in rare cases) that a linear combination of the degenerate wave
functions = b

=1
yields n(r) = (|n(r)|). This case however is not
ordinary and we consider the cases where this does not happen. We thus
proceed to show that n(r) cannot be the groundstate density of any other
Hamiltonian as well. We do this by contradiction and assume existence of a
wavefunction
i
which is the GS of some E

' and such that (


i
|n(r)|
i
) =
n(r). The variational principle states (
I
|E

'|
I
) > (
i
|E

'|
i
) for each i.
Multiply by the positive c

(so the inequality is not spoiled) and sum over i,


using c

= 1, and obtain c

(
I
|E

'|
I
)

> (
i
|E

'|
i
). Now use the same
reasoning that led to Eq. (6.1.7) and replace E

' by E

, obtaining:
c

(
I
|E

|
I
)

> (
i
|E

|
i
). However, (
I
|E

|
I
) = E and so this leads to
(
i
|E

|
i
) < E which condtradicts of the variational principle for

. Hence,
'cannotbethegroundstateofanyHamiltoian.

Example:Theabovetheoremisgeneralandholdsforanysystem
of particles. We consider noninteracting electrons. Consider the
density of noninteracting electrons in the potential well created
by a a Lithium nucleus. The ground state is 4fold degenerate

1
= |1s1s

2s|,
2
= |1s1s

2p
x
|,
3
= |1s1s

2p

|,
4
= |1s1s

2p
z
|
with
1s
=
c
-r
n
,
2s
=
c
-
r
2(-2)
32n
,
2p
x
=
c
-
r
2
x
32n
,
2p
j
=
c
-
r
2

32n
,
2p
z
=
c
-
r
2
z
32n
. The 4 densities are: n
1
= 2n
1s
+n
2s
, n
2
= 2n
1s
+n
2p
x
,
n
3
= 2n
1s
+n
2p
j
,n
4
= 2n
1s
+n
2p
z
.Theaverageisgivenby:
n(r) =
1
4
n

(r)
4
=1
= 2
c
-2
n
+
c
-
128n
((r -2)
2
+r
2
)
Election Bensity Functional Theoiy Page 142
Roi Baei

This density is spherically symmetric and is plotted below,


togetherwiththedensityn
1
(r)forcomparison:

Figure VI3:Leftpanel:Theaveragedensityofn(r)plottedtogetherwith
thealmostidenticaldensityn
1
(r)orbital.Bothdensitiesseemalmost
identicalyetthelatterisvrepwhiletheformernot.RightPanel:The
differencebetweenthetwodensities
This density is very nave and there is no visible indications
suggestingthisisnot(noninteracting)vrepresentable.
We ave found that such convex sum densities are not vrepresentable in the
sensethatthereisnoHamiltonianforwhichagroundstateyieldsthedensity.
However, clearly these densities are associated with the potential v(i) which
createdthedegeneratestates.OnecanextendtheHohenbergKohnfunctional
definitiontodealwiththeseconvexsumdensitiesinthefollowingway.Given
a density which is not vrepresentable in the usual way, we now call it non
purestatevrepresentable.Thenwewillsaythatsuchadensityisensemble
vrepresentableandassumeitisassociatedwiththepotentialv(i)fromwhich
thedegenerategroundstates

wereformed.ThenforsuchadensitytheHK
functionaliswrittenas:
F
HK
|n] = c

|I
`
+u

=1
n(r) = c

|n(r)|

=1

(6.5.3)
Here, we implicitly made the identification n - : - c

. The HK uniqueness
theorem guaranteeing that only one potential : can form a given n can be
extended also to this case (i.e. that n - : is meaningful) still holds.
Election Bensity Functional Theoiy Page 14S
Roi Baei

Furthermore,theminimumprinciple,i.e.thatE

|n]withthemoregeneralF
HK

obtainsitsminimumatthedensitycorrespondingto:(r)remainstrue.

F. Levy-Lieb generalization of the HK functional


TheminimumprincipleofHKisofcrucialimportancefordensityfunctional
theory.Yet,thebasicequation(Eq.(6.3.6))whichisthebasisfordevelopment
of DFT into a practical approach is somewhat problematic from the
mathematical point of view. In order to define the functional derivative, one
mustensurethatthereisanopenneighborhoodofthegroundstatedensity
n

(r) for which F


HK
|n] can be defined for any density.As we saw, there is a
fundamental problem is that of vrepresentability. The ensemble v
representabilitysolvedonekindofproblemsbutitisnotcleariftheremaybe
additional classes of nonvrepresentability. It is not possible to assume that
any density that is positive and integrates to an integer is a density of a
ground state wave function of a Hamiltonian. Furthermore, we cannot even
assume that around a vrepresentable density there is an open
neighborhoodofvrepresentabiledensities.
Ingeneralonecannotwaivethepossibilitythattheremaybedensitieswhich
arenotthegroundstatedensitiesofanysystemandstillarearbitrarilyclose
to any n

(r) of interest. In such a case the functional derivative of F


HK
is
formallyundefined.
A way around these problems, developed separately by Levy and by Lieb is
to formulateafunctional F
LL
|n]defined foranydensityontheonehandand
equal to F
HK
|n] for vrepresentable densities on the other which still allows
for the same type of minimum principle as does the HK functional. We first
Election Bensity Functional Theoiy Page 144
Roi Baei

note an important property of F


HK
|n]. We say that the wave function
realizesthedensityn(r)(andsymbolizethisas - n)if:
(|n(r)|) = n(r) = - n
(6.5.4)
Wewillnowshowthatforanywavefunctionwhichrealizesn(r)
- n = F
HK
|n] (|I
`
+u

|) (6.5.5)
This is a direct result of the variational principle: F
HK
|n] = E
gs
|n] -
]:(r)n(r)J
3
r (|E

|) - ]:(r)n(r)J
3
r = (|E

-I
`
|) = (|I
`
+u

|).
Fromthisdevelopmentweseealsothatthewavefunctionwhichrealizesthis
minimum is the groundstate wave function and is denoted
gs
|n]. One has
therefore:
F
HK
|n] = min
+-n
(|I
`
+u

|)
(6.5.6)
LevyandLiebdecidedtousethisrelation,validonlyforvrepdensitiesasthe
definitionoftheirfunctional,validforanydensity:
F
LL
|n] min
+-n
(|I
`
+u

|)
(6.5.7)

Exercise: Show that for any density there is at least one wave
functionwhichrealizesit
Solution:Foraoneelectroncasethisistrivial,sincethedensityis
nonnegative we can take the wave function as (r) = n(r).
For N electrons, just slice n(r) to N nonoverlapping parts
each:positive,integratingto1:(1)n(r) = n
k
(r)
N
k=1
, (2) n
k
(r) >
= u, (3)

(r)
]
(r) = o
]
n
]
(r) where
k
(r) = n
k
(r), and (4)
]n
k
(r) = 1. This can be done in countless ways (with a good
sharp knife). Then the determinant = |
1

N
c
|, is a
wavefunctionwhichcreatesn(r).

Election Bensity Functional Theoiy Page 14S


Roi Baei

The above exercise shows, that the definition in Eq. (6.5.7) makes sense for
practicallyalldensities.OnecannowuseF
LL
|n]insteadofF
HK
|n]inEq.(6.3.1)
and convert the HohenbergKohn minimum principle to a variational
minimum principle. The search for the constrained minimum can be done
usingaLagrangemultiplierapproach.WeformulatetheLagrangian:
F

LL
|, n, :] (|I
`
+u

|) +_:(r)|(|n(r)|) -n(r)]J
3
r (6.5.8)
Givenn(r),onenowfindsuitableLagrangemultipliers:(r)suchthatwhen
minimizingF
LL
withrespecttoanantisymmetricwavefunctionleadstothe
constraint(|n(r)|) = n(r)atanyr.Thesearchforaminimizingwave
functionwillceasewhenthegradientiszero,soanecessaryconditionis:
o
o
F

LL
|, n, :]|
+=+
LL
,
=
LL
= u = _I
`
+u

+_:
LL
(r)n(r)J
3
r]
LL
= E

LL

LL

(6.5.9)
Thus,whenn(r)isgiven,theLagrangemultiplierfunction:
LL
(r)that
imposestheconditiondefinesaHamiltonianE

LL
= I
`
+]:
LL
(r)n(r)J
3
r +u

.
Notethat:
LL
(r)isdeterminedbytheprocedureuptoaconstant.Ifn(r)isv
repthennecessarily,byKohnstheorem,F
LL
|n] = F
HK
|n]and
LL
isthe
groundstatewavefunction.However,wearenotingeneralassuredthat
LL

isalwaysthegroundstate,sincen(r)isnotnecessarilyvrep.Indeed,from
(6.5.9),allwecansayisthat
LL
mustbesomeeigenstateofE

LL
.
Oncetheminimumisattained,onecantakethefunctionalderivativeofF
LL
by
n(r).Aswithanyconstrainedminimization,whenthederivativeofthe
optimizedsolutionistakenwithrespecttotheconstraints,oneobtainsthe
Lagrangemultiplied(see(1.11.23)):
Election Bensity Functional Theoiy Page 146
Roi Baei

o
on(r)
F

LL
|, n, :] = -:
LL
(r) (6.5.10)
Now,oncetheconstrainedminimumF

hasbeenfound,onecanplugF

into
theHKschemeandconsidertheLagrangiancorrespondingtoE

= F

LL
+
]:(r)n(r)J
3
r:
I

|n] = F

LL
+_:(r)n(r)J
3
r -p __n(r)J
3
r -N_ (6.5.11)
Ifnowonesearchesforthedensitythatminimesthisfunctionalone
immediatelyobtains:
u =
oI

on(r)
= -:
LL
(r) +:(r) -p (6.5.12)
ShowingthattheminimumofE

|n]isobtainedwhenthedensityfound
admits,asLagrangemultiplier:
LL
(r),thesamepotentialaswasgivetostart
with:(r),uptoaconstant.
EvenwiththeLLfunctionaloneproblemisstillnotsolved:convexity.While
thedomainofdefinitionisnowfinallyconvex,thefunctionalhaslost
convexity.Acounterexampleforconvexitycanbegiven,usingthedegenerate
systemusedabovetoprovethatthedomainofdefinitionoftheHKfunctional
isnonconvex.Indeed,forthatsystem,whichhadapotential:(r)and
degenerateeigensstates

(i = 1, , )allofenergyE:
Election Bensity Functional Theoiy Page 147
Roi Baei

F
LL
_c

_ = (V
LL
|I
`
+u

|V
LL
)
= (V
LL
|E

|V
LL
) -_c

(r):(r)J
3
r

c
I
E -_c

(r):(r)J
3
r

= c

_(
I
|T

+0

+v|
I
) -_n

(r):(r)J
3
r_

= c

F
LL
|n


(6.5.13)
Showingthatforthesedensitiesthefunctionalisconcave(whileforthecase
wheren

and c

n

areallvrepthefunctionalisconvex).Thereforeoverall
thefunctionalisnotconvex.
Onceagain,itispossibletocircumventthisproblembyextendingthe
definitiontoensembledensities.

Exercise:DiscusstheLevyapproachforagivendensityn(r)ofa
systemofnoninteractingelectrons.
In this case the Hamiltonian is E

= I
`
+]:(r)n(r)J
3
r i.e. a 1
body Hamiltonian. Suppose
m
, m = 1,2, are the 1particle
stateswhichareeigenstatesofE

1
= I
`
1
+]:(r)n(r)J
3
r:
-
l
2
2m
c
v
2

m
(r) +:(r)
m
(r)
= e
m

m
(r)
(6.5.14)
Let us suppose orbital energy ordering, so that e
1
e
2
e
3
If
the density isnoninteracting vrepthen it must bedue tothe
Slater wave function
gs
= |
1

N
c
| of the lowest energy N
c

orbitals.Inthiscase,thekineticenergyisgivenby:
I
LL
|n] = (
m
|I
`
|
m
)
N
c
m=1

(6.5.15)
Andthefunctionalderivativeis:
Election Bensity Functional Theoiy Page 148
Roi Baei

o
on(r)
I
LL
|n] = -:(r) (6.5.16)
If we are given a noninteracting nonvrep density then we
might find not a ground state but an excited determinant, for
example:
cx
= |
1

N
c
-1

N
c
+1
|. This leads to holes in the
noninteractingsystem.

G. The dilation inequality for the HK functional


Wehave already seen the theHFfunctionalis convex.Moreexactproperties
aredesirable,sothatwhenonederivesapproximationstothisfunctional,they
can perhaps beforcedto obeytheknownexactrelations.One approach isto
usedilationconsiderations.Thiswillleadtoaninterestinginequlaity.
Wewillusethefollowingsymbolconvention:

y
(r
1
, r
2
, r
N
c
) y
3N
c
2
(yr
1
, yr
2
, yr
N
c
)
n
y
(r) y
3
n(yr)
(6.6.1)
Itisquitestraightforwardtoshowthefollowingrelations:
Exercise
Derivethefollowingrelation
(
y
|n(r)|
y
) = y
3
(|n(yr)|)
(
y
|I
`
|
y
) = y
2
(|I
`
|)
(
y
|u

|
y
) = y(|u

|)
(6.6.2)
Fromthefirstrelationinthisequation,onecandeducethatif - n(r)(i.e.
isamanyelectronwavefunctionexhibitingthespatialdensityn(r))then:
(
y
|n(r)|
y
) = y
3
(|n(yr)|) = y
3
n(yr) = n
y
(r)
(6.6.3)
One can now also use this relation to get a dilation inequality. Indeed, since
|n]
y
realizes the density n
y
(r) (see the first of Eqs. (6.6.2)) we have
Election Bensity Functional Theoiy Page 149
Roi Baei

F
HK
|n
y
] (|n]
y
|I
`
+u

||n]
y
) = y
2
(|n]|I
`
||n]) +y(|n]|u

||n]), thus we
find:
F
HK
|n
y
] y
2
I|n] +yu|n] (6.6.4)
VII. The Kohn-Sham method
Kohn and Sham noticed that the HK theory is valid for both interacting and
noninteracting electrons. Now, they ask, what happens if for any system of
interacting electrons, with density n there is a noninteracting system of the
same density? It is clear that if such two systems exist they are unique. The
noninteractingsystemhasoneadvantageovertheinteractingsystem:wecan
find its groundstate rather easily, since the manybody wavefunction is a
Slaterwavefunction.So,theproblemis:howtoperformsuchamapping.
A. Non-interacting electrons
If noninteracting electrons are tractable, let then study their density
functionals. First, if we are given n(r) we assume it is noninteracting v
representable. That is we assume there exists a potentials :
s
(r) such that the
groundstate of theresultingnoninteractingHamiltonianE

s
= I
`
+I
`
s
admits
a groundstate having n(r) as density. We can later use the LevyLieb
approachtopatchupthevrepresentabilityrequirement.
Letusdenotethegroundstatewavefunctionofnoninteractingelectronsthat
have the density n(r) by
gs
|n]. The HohenbergKohn functional for non
interactingelectronsisreducedtojustthekineticenergy,i.e.wedefine:
I
S
|n] = (
gs
|n]|I
`
|
gs
|n]) (7.1.1)
Election Bensity Functional Theoiy Page 1Su
Roi Baei

An interesting question is can we devise a method to compute I


S
|n]. For
most densities, the noninteracting wave function is simply a Slater wave
function.TheminimumprincipleforI
S
|n](derivedfromEq.(6.5.5))is:
- n = I
S
|n] (|I
`
|) (7.1.2)
One should not treat this equation lightly and realize its nontrivial
consequences:thatthegroundstatewavefunctionofnoninteractingparticles
associated with n(r) minimizes the kinetic energy! Let us see some
consequencethisminimumprinciple.
Let
gs
|n]bethegroundstatewavefunctionofthesystemofnoninteracting
electronsrealizing thedensity n(r).FromthefirstequationinEq.(6.6.2)one
sees that
gs
|n]
y
realizes n
y
(r),thusonecanplugit intotheright handside
ofEq.(7.1.2),withn
y
(r)pluggedintoI
S
:
I
S
|n
y
] = (
gs
|n
y
]|I
`
|
gs
|n
y
]) (
gs
|n]
y
|I
`
|
gs
|n]
y
) (7.1.3)
Sincenowtherighthandsideisanexpressionofascaledwavefunction,one
canuseEq.(6.6.2)andobtain:
I
S
|n
y
] (
gs
|n]
y
|I
`
|
gs
|n]
y
) = y
2
(
gs
|n]|I
`
|
gs
|n]) = y
2
I
S
|n] (7.1.4)
AndsoI
S
|n
y
] y
2
I
S
|n].Butwecanalsousethisequationtoscalen
y
backto
n by dilating by 1y, since: (n
y
)
1y
= n. Then, the same rule applies and we
get:I
S
|n] y
-2
I
S
|n
y
],andsoI
S
|n
y
] y
2
I
S
|n].Weobtainedtwocontradicting
equationswhichcanagreeonlyifbotharereducedtoequality.Thus:
I
S
|n
y
] = y
2
I
S
|n]. (7.1.5)
ThisshouldbecomparedwiththeanalogousresultofEq.(6.6.4)whichisan
inequality.Thedilationeffectsfornoninteractingelectronsisobviouslymuch
simpler. One important corollary of (7.1.5) is that the noninteracting wave
Election Bensity Functional Theoiy Page 1S1
Roi Baei

functions scale with the density. This can be seen if one remembers that

gs
|n] minimizes the kinetic energy. Therefore any function |n] for which
(|n]|I
`
||n]) equals I
S
|n] must be the ground state of the noninteracting
systemwithdensityn.Wesawthat
gs
|n]
y
givesthecorrectkineticenergyof
forthesystemwithdensityn
y
,I
S
|n
y
],andtherefore,necessarily:

gs
|n]
y
=
gs
|n
y
] . (non -intcrocting porticlcs) (7.1.6)
Exercise
1) We can define a functional called the Hartree energy, which is the
classical electrostatic energy associated with the charge distribution
n(r):
E
H
|n] =
n(r)n(r
i
)
|r -r
i
|
J
3
rJ
3
r
i
(7.1.7)
Provethefollowingdilationrelation:
E
H
|n
y
] = yE
H
|n] (7.1.8)
2) Now define the exchange energy functional (see also Eq. (4.1.17) for a
definitionbasedontheorbitals):
K|n] (
gs
|n]|u

|
gs
|n]) -E
H
|n] (7.1.9)
Use(7.1.6).Prove:
K|n
y
] = yK|n] (7.1.10)
What is the relation between the potential :
S
(r) for which n(r) is a non
interacting ground state density and :
S,y
(r) for which n
y
(r) is a non
interactinggroundstate?WecanusethebasicDFTequationtoanswerthis.
Fromthebasicdefinitionoffunctionalderivation:
Election Bensity Functional Theoiy Page 1S2
Roi Baei

y
-:
S,y
(r
i
) =
oI
S
on(r
i
)
_
n
y
= lim
q-0
(I
S
|n
y
(r) +po(r -r
i
)] -I
S
|n
y
(r)])
p
(7.1.11)
The 3D deltafunction has the density dilation structure: o(r -r
i
) =
z
3
o(zr -zr
i
),so:

y
-:
S,y
(r
i
) = lim
q-0
(I
S
|y
3
(n(yr) +po(yr -yr
i
))] -I
S
|y
3
n(yr)])
p

(7.1.12)
ThenusingEq.:

y
-:
S,y
(r
i
) = lim
q-0
y
2
(I
S
|n(r) +po(r -yr
i
)] -I
S
|n(r)])
p
= y
2
(p -:
S
(yr
i
))
(7.1.13)
Bydilatinginthereversedirectionwecaneasilyseethat:

y
= y
2
p
:
S,y
(r
i
) = y
2
:
S
(yr
i
)
(7.1.14)
We could have obtained this result directly from the Schrdinger equation.
Suppose({r])isamanybodyeigenfunctionofHamiltonianE

= I
`
+I({r]),
i.eI
`
= (E -I).Defineascaledwavefunction:

y
(r
1
, , r
N
) = y
-
3N
2
(yr
1
, , yr
N
)
(7.1.15)
Thenclearly:
I
`

y
({r]) = y
2
y
-
3N
2
(I
`
)({yr]) = y
2
(E -I({yr]))
y
({r])
(7.1.16)
Andso
y
obeys the S. E.I
`

y
= (E
y
-I
y
)
y
with eneigy anu potential given by:
I
y
({r]) = y
2
I({yr])
E
y
= y
2
E
(7.1.17)
Election Bensity Functional Theoiy Page 1SS
Roi Baei

For a ground state of noninteracting the first equation means that :


s,y
(r) =
y
2
:
s
(yr)istheoneparticlepotentialforthescaleddeterminant,andthusfor
thescaleddensity(sincethescaleddeterminantrealizesthescaleddensity).
B. Orbitals for the non-interacting electrons
Consider the ground state wave function for the noninteracting electrons

gs
|n]. Suppose the noninteracting electrons reside in the potential :
S
(the
subscript S is in honor of Slater) then (I
`
+I
`
S
)
gs
|n] = E
S

gs
|n]. In most
cases this wave function is a normalized Slater wave function. The only
exception might be the existence of a degeneracy and then there may be
several independent Slater wave functions and
gs
is a linear combination.
However, one can always assume that he is looking for one of the
determinants and not for a linear combination (this means that one must be
careful not to impose possible symmetries of the Hamiltonian). Once one
doesthat,wecanintroduceN
c

orthormal

spinorbitals
q
(x)q = 1N
c
,from
which
gs
|n]isbuilt.Theseorbitalsareexcitedstatesofthepotentialwellin
which the noninteracting electrons reside. Thus the orbitals must each obey
thesingleelectronSchrdingerequation:
-
l
2
2m
c
7
2

q
(x) +:
S
(r)
q
(x) = e
q

q
(x)
(7.1.18)
The orbitals correspond to the N
c
lowest eigenvalues e
q
. The fact that
gs

realizesthedensityn(r)isexpressedas:
n(r) = |
q
(x)|
2
N
c
q=1
(7.1.19)
ThenoninteractingkineticenergyI
S
|n]isthen:
Election Bensity Functional Theoiy Page 1S4
Roi Baei

I
S
|n] = _
q
(x) _-
l
2
2m
c
7
2
_
q
(x)Jx
N
c
q=1
(7.1.20)
When one wants to find the functional derivative of I
S
with respect to n(r)
one can turn to basic equation, Eq. (6.3.6) which in case of noninteracting
electronsbecomes:
oI
S
|n]
on(r)
+:
S
(r) = p (7.1.21)
We will see that this equation is important for the method known as the
KohnShammethod.
Notethatthediscussionofdilationintheprevioussubsectioncanbecarried
on to the orbitals. The only additional information to convey is that the
orbitals scaleasthedensity and eachof theorbitalenergiesscaleasthe total
energy:
e
q,y
= y
2
e
q
(7.1.22)
C. The correlation energy functional: definition
and some formal properties
The ground state energy of the an systemof electrons in density n(r) can be
written in terms of the noninteracting (Slater wave function) wave function

gs
|n]:
E|n] = (
gs
|n]|E

|
gs
|n]) +E
c
|n] (7.2.1)
This equation is actually a definition of a new density functional, the
correlation energy functional E
C
|n]. If we suppose for the time being that
E
C
|n] is known, this expression can be used to define a working procedure
for DFT known as the KohnSham method. From our studies of the Hartree
Election Bensity Functional Theoiy Page 1SS
Roi Baei

Fock theory, we know that the expectation value of E

within a determinant
canbewrittenas:
(
gs
|n]|E

|
gs
|n]) = E
H
|n] +K|n] u
S
|n] (7.2.2)
WhereE
H
istheHartreeenergy:
E
H
|n] =
c
2
2

n(r)n(r
i
)
|r -r
i
|
J
3
rJ
3
r
i

(7.2.3)
And K is given in terms of the orbitals from which
gs
|n] is composed (Eq.
(4.1.17)):
K|n] = -
c
2
2

|p(x, x
i
)|
2
|r -r
i
|
J
3
xJ
3
x
i

(7.2.4)
Basedonthis,werewriteF
HK
|n]as:
F
HK
|n] = (
gs
|n]|F
`
|
gs
|n]) = (
gs
|n]|F
`
|
gs
|n]) +E
C
|n] (7.2.5)
ThisallowsustowriteEq.(7.2.5)compactlyas:
F
HK
|n] = I
S
|n] +u
S
|n] +E
C
|n] (7.2.6)
Clearlywehavealsotheequivalentequation:
E
C
|n] = I|n] -I
s
|n] +u|n] -u
S
|n] (7.2.7)
Physical intuition concerning molecules and solids tells us that E
C
|n]

is a
smallquantitywhencomparedtoI
s
|n]oru
S
|n].Thus,itisreasonabletolook
for approximations to this quantity.Approximation to the correlation energy
functional is the most important issue in DFT. It is an open question, still
beingworkedupon.
We shall deal with approximations later. Meanwhile, let us ask what can be
safelysaidaboutthecorrelationfunctional.Weprovehereseveralimportant
Election Bensity Functional Theoiy Page 1S6
Roi Baei

inequalities.First,considerthedifferenceI
C
|n] I|n] -I
S
|n],thecorrelation
kineticenergy.
Exercise
Show I
C
|n] is a positive quantity. This is actually intuitively clear: the
interacting electrons must have much more complicated paths in the
interacting case because they want to avoid bumping into other electrons.
Anythingwithmoreswirlsmusthavehigherkineticenergy.
Solution
Usingthevariationaltheorem:
I
S
|n] +I
S
|n] = (
gs
|n]|I
`
+I
`
S
|
gs
|n]) (
gs
|n]|I
`
+I
`
S
|
gs
|n])
= I|n] +I
S
|n]
(7.2.8)
Where
gs
|n]istheinteractinggroundstatewavefunctiondeterminedbyn.
Comparingthetwosideswehave:
I
S
|n] I|n] (7.2.9)
Next, we can show that the exchange correlation energy is always negative.
This comes about from our experience with expectation values of
determinants:
F
HK
|n] +I|n] = E
gs
(
gs
|n]|E

|
gs
|n]) = I
S
|n] +I|n] +u
S
|n] (7.2.10)
UsingEq.(7.2.6)wefind:
E
C
|n] < u (7.2.11)
Furthermoreu
C
|n] u|n] -u
S
|n]isnegativeascanbeseenfromthefactthat
E
C
isnegativeandI
C
ispositive:
u
C
|n] = E
C
|n] -I
C
|n] u (7.2.12)
An additional property is the dilating relations. We have proved that
F
HK
|n
y
] y
2
I|n] +yu|n]andI
S
|n
y
] = y
2
I
S
|n],u
S
|n
y
] = yu
S
|n].Thus:
Election Bensity Functional Theoiy Page 1S7
Roi Baei

E
C
|n
y
] = I
C
|n
y
] +u
C
|n
y
] y
2
I
C
|n] +yu
C
|n] (7.2.13)
Onewaytoproceedistosubstituteu
C
withE
C
-I
C
.Theotheristosubstitute
I
C
withE
C
-u
C
.Weobtainfromeachpossibility:
E
C
|n
y
] y(y -1)I
C
|n] +yE
C
|n]
E
C
|n
y
] y
2
E
C
|n] +y(1 -y)u
C
|n]
(7.2.14)
And,sinceI
C
isalwayspositiveandu
C
negative,wefindthat:
E
C
|n
y
] yE
C
|n] y < 1
E
C
|n
y
] y
2
E
C
|n] y < 1
(7.2.15)
Obviously, the second relation is contained in the first (since E
C
< u) and so
only the first relation is important; it can be used to derive complementary
inequalities.Indeed,applyingthemfory - y
-1
wefind:
E
C
|n
1y
] y
-1
E
C
|n] y > 1 (7.2.16)
Thisholdsforanyn,sowestickin(7.2.16)n
y
insteadofnandusingthefact
that(n
y
)
1y
= ntoobtain:
E
C
|n
y
] yE
C
|n] y > 1 (7.2.17)
D. The Kohn Sham equations
i. The Kohn-Sham equation from a system of non-interacting problem
Let us now turn to the issue of determining

:
s
(r), required for the mapping
between the interacting noninteracting systems. Let us start form the basic
equationEq.(6.3.6)whichbecomes,usingEq.(7.2.6):
oI
S
|n]
on(r)
+:
HX
(r) +:
C
(r) +:(r) = p (7.3.1)
Where:
:
HX
(r) =
ou
S
|n]
on(r)
= :
H
(r) +:
X
(r) (7.3.2)
Election Bensity Functional Theoiy Page 1S8
Roi Baei

istheHartreeExchangepotential.Noticethat:
:
H
(r) =
oE
H
|n]
on(r)
= _
n(r
i
)
|r -r
i
|
J
3
r
i
(7.3.3)
and:
:
X
(r) =
oK|n]
on(r)
(7.3.4)
Thus,fromEq.(7.1.21)andEq.(7.3.1)(uptoaconstant):
:
S
(r) = :(r) +:
HX
(r) +:
C
(r) (7.3.5)
This equation gives us the potential of the noninteracting system. Thus, we
havemade a definite connection between theinteracting and noninteracting
systems.
Now, an important observation allows us to set up a simple method for
obtaining the groundstate of an interacting system of electrons. We need to
findadensitythatobeystwoconditions:
a) It is the density of the noninteracting electrons so it is the sum of square
orbitals (Eq. (7.1.19)) that obey the Schrdinger equations (7.1.18) with
potential:
S
(r).
b)Thepotential:
S
(r)

mustberelatedtotheinteractingsystembyEq.(7.3.5)
This leads to a simple SCF procedure, called The KohnSham method
reminiscentoftheHartreeFockalgorithm:
1. Guessn(r).
2. Build:
HX
(r),:
C
(r)fromn(r)(Eq.(7.3.3)and(7.3.4)).
3. Obtaintheorbitals
q
(x)fromthelowestN
c
eigenvaluesofEq.(7.1.18).
4. ComputethedensityfromEq.(7.1.19)
5. Redofromstep2usingthenewdensityuntilyouconvergei.e.until
thedensitychangesnomore.
Election Bensity Functional Theoiy Page 1S9
Roi Baei

Whentheprocessconvergence,wehavetheexactgroundstatedensityn

(r).
It can be used to compute the groundstate energy by plugging it into the
energyfunctionalofEq.(6.3.1):
E

|n] = I
s
|n] +_:(r)n(r)J
3
r +u
S
|n] +E
C
|n] (7.3.6)
ii. Systems with partially occupied orbitals
Partially occupied orbitals arise in the realm of ensembleDFT, i.e. when the
densitysearchedforisnotnoninteractingvrepresentable.Westillassume
thedensityisasumoforbitaldensitiesasfollows:
n(r) = |
q
(x)|
2
q
,
(7.3.7)
butunlikeEq.(7.1.19)wedonotimposeexactlyN
c
orbitals.Theorbitalsare
almost completely unconstrained, except for the Fermionic condition, that
theirnormmustbenotbegreatthan1:
(
q
|
q
) = n
q
1, (7.3.8)
butthetotalnumberofelectronsisstillN
c
:
_n(r)J
3
r = n
q
q
= N
c

(7.3.9)
AndthekineticenergyiswrittensimilarlytoEq.(7.1.20):
I
S
|n] = _
q
(x) _-
l
2
2m
c
7
2
_
q
(x)Jx
q

(7.3.10)
WegeneralizedEqs.(7.1.19)and(7.1.20)anynumberoforbitalsleadingtothe
followingorbitalLagrangian:
Election Bensity Functional Theoiy Page 16u
Roi Baei

I
ob
||
q
|, |n
q
|]
= E

_|
q
(x)|
2
q
_ -e
q
((
q
|
q
) -n
q
)
q
-p _n
q
q
-N
c
_
(7.3.11)
e
q
(q = 1,2, ) and p are Lagrange multipliers assuring orbital normality and
total number of particles. We minimize the Lagrangian with respect to the
orbitals
q
(r)andtheoccupationnumbers(constrainttobenonnegativeand
notgreaterthan1).Theconstraintminimumprocedurethusrequiresthat:
oI
ob
o
q
(x)
= u onJ
oI
ob
on
q
= u (7.3.12)
ThefirstequationleadstotheKSequations:
-
l
2
2m
c
7
2

q
(x) +:
S
(r)
q
(x) = e
q

q
(x) (7.3.13)
Forconvenience,weordertheindexingsothattheseriesoforbitalenergiesis
ascending: e
1
e
2
The orbitals are now eigenfunctions of a Hermitian
Hamiltonian,andsowecanassumetheyareorthogonal:
(
q
|
qi
) = n
q
o
qq
|, (7.3.14)
We define as a shorthand notation :
S
(r) :(r) +:
HX
(r) +:
C
(r) (see Eq.
(7.3.5)).BymultiplyingEq.(7.3.13)by
q
(x)integratingonxandsummingon
qwefind:
I
S
+_:
s
(r)n(r)J
3
r = e
q
n
q
q

(7.3.15)
The second condition for minimum of I
ob
holds only in cases that the
minimum is attained with noninteger electron number, n
q
< 1. The cases
Election Bensity Functional Theoiy Page 161
Roi Baei

n
q
= 1 or n
q
= u are the boundary of the constraints and the derivative need
notbezerothere.Fromthissecondpartoftheequation,wehave:
e
q
= p, u < n
q
< 1 (7.3.16)
This shows that all incompletely occupied orbitals have the same orbital
energy equal to the chemical potential p, the Lagrange multiplier imposing
thenumberofparticleconstraintoftheHohenbergKohntheory(Eq.(6.3.6)).
Typically, all orbitals with e
q
< p are fully occupied (n
q
= 1) and those with
e
q
> p are fully unoccupied (n
q
= u). The interacting electron energy is
obtainedusingEqs.(7.3.6)and(7.3.15)as
E

= e
q
n
q
q
+E
HX
|n] +E
C
|n] _n(r)(:
HX
(r) +:
C
(r))J
3
r
(7.3.17)
We should note that the development here assumed noninteracting v
representability. In cases where this is not valid other occupation rules may
apply.
iii. Is the ground state wave function of non-interacting particles always a
Slater wave function
ThestandardKSapproachtothenoninteractingkineticenergyisbydefining
the functional I
S
as a minimum principle on the manifold of single
determinants:
I
S
|n] = min
4-n(r)
(|I
`
|)
(7.3.18)
SettingupaLagrangianandsearchingfortheconstrainedminimumyieldsN
occupiedorbitals.Ifthedensityisnotvrepresentableoneormoreofthelow
energyorbitalsmayhave0occupationnumbers.
Election Bensity Functional Theoiy Page 162
Roi Baei

Another way to define a noninteracting kinetic energy functional is by an


extendedminimumsearchovermoregeneralwavefunctions:
I
NI
|n] = min
+-n(r)
(|I
`
|)
(7.3.19)
Usually this search ends with being a single Slater wave function and for
thesecaseI
NI
andI
S
givenarethesame.Yet,thismaynotalwaysbethecase.
Let us assume that we are searching through all = cos 0
A
+sin0
B

where
A
or
B
are Slater wave functions. In general, the kinetic energy is
then:
I
NI
|n] = min
+-n(r)
(
A
|I
`
|
A
) cos
2
0 +sin20 (
A
|I
`
|
B
)
+(
B
|I
`
|
B
) sin
2
0
(7.3.20)
Now,if
A
and
B
differbyonlyoneorbital(say
A
k
=
B
k
fork = 1, , N -1)
then is actually a Slater wave function: its orbitals are the N -1
A
k
s and
then a new orbital
N
ncw
= cos 0
A
N
+sin0
B
N
is added. Next, if
A
and
B

differ by two (or more) orbitals the cross term in (7.3.20) is zero and the
kinetic energy is simply the sum of orbitals kinetic energies with occupation
numbers given by cos
2
0 and sin
2
0. The orbitals shared by the two
determinants will have unity occupation number (since cos
2
0 +sin
2
0 = 1)
while orbitals in the A determinant but not in the B determinant will have
occupation number cos
2
0 and those in the B determinant but not in the A
determinant will have occupation number sin
2
0. The orbitals coming out
from the minimization will all solve a Schrdinger equation with the same
potential.The4oddorbitals(2fromeachdeterminant)willallhavethesame
orbitalenergyequaltop.Whentherearemorethan2Slaterwavefunctionsa
similartreatmentwillresultandevenmoreorbitalswillbedegenerateatthe
Election Bensity Functional Theoiy Page 16S
Roi Baei

chemicalpotential.Evenifnisvrepresentablethistypeofwavefunctioncan
arise.
Athirdwaytodefinethekineticenergyisasanensembleaverage.Insteadof
a wave function, one uses a mixed density matrix = c
n
|
n
)(
n
|
n
. The
constants c
n
are positive and sum to 1. The density is the convex sum
c
n
n
n
(r)
n
of densities from each participating determinant. A similar
expressionwillariseforthekineticenergy.Thisapproachisdesignedtosolve
the problems of noninteracting vrepresentability emanating from convex
sumsofdegeneratewavefunctions(seesectionXXX).
iv. Janaks Theorem
A very general theorem was noted by Janak [13], based on earlier work of
SlaterandWood[14]concerningthemeaningoforbitalenergies.Letusreturn
tothefunctionalofEq.(7.3.11)andassumenowthattheoccupationnumbers
n
q
aregivenandtheyareallnonnegativeandnotlargerthan1andthatthey
sum up to the number of electrons. Thus, for a given set of occupation
numbers we can search for the orbitals that minimize the following
functional:
I||
q
|] = E

_|
q
(x)|
2
q
_ -e
q
((
q
|
q
) -n
q
)
q
(7.3.21)
The equations that the orbitals must obey are still derived from Eq. (7.3.12)
leadingtothesameequationsasin(7.3.13),theKSequation.Now,letusask:
what happens to the energy when we change the occupation number of one
oftheorbitalsn
q
byaninfinitesimalamounton
q
. Whenwedothisthetotal
numberofelectronsN
c
changesbythisamountaswell.Thisisnotaphysical
change (since electrons cannot change by nonintereger amounts) but still
Election Bensity Functional Theoiy Page 164
Roi Baei

mathematically speaking the change can be studied. Since n


q
are the
constraintsande
q
theLagrangemultipliersinaminimizationproblemwecan
use the general result of Eq. (1.11.23) that the rate at which the minimized
function changes when the constraints change is equal to the Lagrange
multiplier:
oE

on
q
= e
q
(7.3.22)
This relation, giving some meaning to the orbital energies is called Janaks
theorem. This theorem is quite general but relies on some analytical
assumptions of the energy functional. For example, when the occupation
numberis1thechangecanonlybebyanegativeamountandwhenitiszero
only positive. For approximate functionals, that are analytical with respect
to any n(r) and
q
(r) this relation holds. Such is the case for the often used
local, semilocal and even most hybrid functionals, including HartreeFock
theorem.
IntheconextofHartreeFocktheorythisresultisarestatementofKoopmans
theorem,bywhich e
q
istheunrelaxedionizationenergyfromorbitalq.The
orbitalrelaxationisasecondordereffectandthusnegligiblewhenoccupation
numberschangeinfinitesimally.
E. Virial Theorem related identities in DFT
The following development is inspired by the virial theorem treatment. It
continues to consider dilation relations. Taking the derivative of n
y
(r)with
respect to y, we have
d
dy
n
y
(r) = y
3
r vn(yr) + Sy
2
n(yr) and since, v (r(r)) =
S(r) + r v(r),wefind
Election Bensity Functional Theoiy Page 16S
Roi Baei

J
Jy
n
y
(r) = y
2
|v
s
(xn(x))]
x=yr
(7.4.1)
For a general density functional, A|n], we have, using the chain rule for
derivatives,o|n](r) = oA|n]on(r)andEq.(7.4.1)then:
J
Jy
A|n
y
] = _o|n
y
](r)y
2
|v
s
(xn(x))]
x=yr
J
3
r = _y
-1
o|n
y
](y
-1
x)v
s
(xn(x))J
3
s
= -y
-1
_[r v
r
o|n
y
](r)
r=y
-1
x
n(x)J
3
s
From which a completely general virialdilation relation holds for any
functional:
JA|n
y
]
Jy
+ y
2
_[r v
r
o|n
y
](r) n(yr)J
3
r = u
(7.4.2)
Wewillbeespeciallyinterestedinthecaseofy = 1.Thusthebasicrelation:
_
JA|n
y
]
Jy
_
y=1
+ _(r vo|n](r))n(r)J
3
r = u (7.4.3)
Wefurtherfindthat
A|n] + _(r vo|n](r))n(r)J
3
r = _
J(yA|n] - A|n
y
])
Jy
_
y=1
(7.4.4)
LetusapplythisfortheHartreeenergy.FromEq.(7.1.8)wefindthefollowing
relation,validforanyy:yE
H
|n] -E
H
|n
y
] = u,so:
E
H
|n] +_r v:
H
|n](r) n(r)J
3
r = u (7.4.5)
A similar relation, namely yK
X
|n] -K
X
|n
y
] = u hold also for the exchange
energyK
X
|n](seeEq.(7.1.10)),andso:
K
X
|n] +_r v:
X
|n](r) n(r)J
S
r = u (7.4.6)
Where:
X
|n](r) = oK
X
|n]on(r)istheexchangepotential.
Election Bensity Functional Theoiy Page 166
Roi Baei

Now, what about the correlation energy? Let us consider the KS DFT
functional:
E

|n] = I
S
|n] +_:(r) n(r)J
3
r +u
S
|n] +E
C
|n] (7.4.7)
Supposen(r)isthedensityminimizingminimizerofE

|n]andnowpluginto
thelatterthescaleddensity.Thiswillgiveaydependentenergy:
E

(y) = E

|n
y
] (7.4.8)
Andclearly,theminimumisaty = 1so:
E

i
(1) = u (7.4.9)
Now,letusevaluateE

i
(y) usingEq.(7.1.5):
E

i
(y) = 2yI
S
|n] +
J
Jy
__:(r)n
y
(r)J
3
r] +u
S
|n] +
J
Jy
E
C
|n
y
], (7.4.10)
andpluginy = 1.Weobtain:
J
Jy
_E
C
|n
y
] +_:(r)n
y
(r)J
3
r_
y=1
= -2I
S
|n] -u
S
|n]. (7.4.11)
or,usingEq.(7.4.3):
_[r v
r
(:
C
|n](r) + :(r)) n(r)J
3
r = -2I
S
|n] -u
S
|n] (7.4.12)
The second term in the parenthesis can be related to the interacting system.
Indeedwehave,usingEqs.(6.6.2):
E|
y
] = I|
y
] +u|
y
] +_:(r) n
y
(r)J
3
r
= y
2
I|] +yu|] +_:(r) n
y
(r)J
3
r
(7.4.13)
Taking the derivative with respect to y, remembering that E
i
|
y
]
y=1
= u we
find,fory = 1:
Election Bensity Functional Theoiy Page 167
Roi Baei

u = 2I|] +u|] +
J
Jy
__:(r) n
y
(r)J
3
r]
y=1
(7.4.14)
(As a sidenote, you can see that substituting this relation in Eq. (7.4.1) gives
after trivial manipulation the virial theorem 2I|] +u|] = ](r
v:(r))n(r)J
3
r). Continuing the above development, using Eqs. (7.4.11) and
(7.4.14)wefinallyfind:
J
Jy
E
C
|n
y
]|
y=1
= 2I
C
+u
C
, (7.4.15)
whichcanbewrittenequivalentlyas:
E
C
|n] -
J
Jy
E
C
|n
y
]|
y=1
= -I
C
|n], (7.4.16)
or,usingEq.(7.4.3):
E
C
|n] +_(r v
r
:
C
|n](r))n(r)J
S
r = -I
C
|n] u. (7.4.17)
The Hartree and exchange energies give zero while the correlation energy
givesanegativequantityequalexactlyto-I
C
|n].
This result shows that the correlation energy functional and the correlation
potentialareenoughfordeterminingthecorrelationkineticenergy(andfrom
it,byI = I
S
+I
C
)thekineticenergyitself.
This latter result is related to the virial theorem of Slater, which shows that
onecanderivethekineticenergyoftheelectronsfromtheBornOppneheimer
potential surface itself (Eq. (1.9.30)). Since DFT gives, in principle the Born
Oppenheimer potential surface, one can access the kinetic energy (and the
potentialenergy)fromtheDFTcalculation.
Election Bensity Functional Theoiy Page 168
Roi Baei

F. Galilean invariance
A basic property of the electronelectron interaction is that if the coordinates
ofallelectronsareshiftedbyaconstantL:
r
n
i
= r
n
-L (7.5.1)
theeeinteractionenegydoesnotchange
u
i
=
1
2

1
|r
n
i
-r
m
i
|
n=m
=
1
2

1
|r
n
-r
m
|
n=m
= u
(7.5.2)
Thispropertyisshowsthattheeeinteractionistranslationallyinvariant.This
propertyisalsocalledGalileaninvariance.Thesamepropertyholdswhenthe
coordinatesofeachelectronarerotatedaroundsomeaxis.Thisroatationcan
bedescribedbya33orthogonalmatrix0,where0
1
0 = 00
1
= I:
r
n
ii
= 0r
n
(7.5.3)
Thelengthsofvectorsarepreservedunderarotation:
|r
n
ii
|
2
= r
n
ii
1
r
n
ii
= (0r
n
)
1
0r
n
= r
n
1
0
1
0r
n
= r
n
1
r
n
= |r
n
|
2

(7.5.4)
Thustheeeinteractionenegydoesnotchange
u
ii
=
1
2

1
|r
n
ii
-r
m
ii
|
n=m
=
1
2

1
|0r
n
-0r
m
|
n=m
=
1
2

1
|0(r
n
-r
m
)|
n=m
=
1
2

1
|r
n
-r
m
|
n=m
= u
(7.5.5)
Thustheeeenergyisrotationalinvariant.
These relations indeed hold for the density functional E
H
|n] since it is a
reflectionoftheeefunctional.Indeed,definetheshifteddensity:
n
i
(r
i
) = n(r) = n(r
i
+L) (7.5.6)
Then:
Election Bensity Functional Theoiy Page 169
Roi Baei

E
H
|n
i
] =
1
2

n
i
(r
1
i
)n
i
(r
2
i
)
|r
1
i
-r
2
i
|
J
3
r
1
i
J
3
r
2
i
=
1
2

n(r
1
i
+L)n(r
2
i
+L)
|r
1
i
-r
2
i
|
J
3
r
1
i
J
3
r
2
i

(7.5.7)
Wenowmakeavariablereplacement:r
n
i
- r
n
-Landobtain:
E
H
|n
i
] =
1
2

n(r
1
)n(r
2
)
|(r
1
-L) -(r
2
-L)|
J
3
r
1
J
3
r
2
=
1
2

n(r
1
)n(r
2
)
|r
1
-r
2
|
J
3
r
1
J
3
r
2
= E
H
|n]
(7.5.8)
This condition, that E
H
|n] = E
H
|n
i
] is called Galilean invariance. It is easy to
showthatE
H
|n]functionalisalsorotationalinvariant.
The exchange energy is also Galilean invariant, since the translation of the
densitywillcauseatranslationofthedensitymatrix:
p|n
i
](r
1
i
, r
2
i
) = p|n](r
1
, r
2
) (7.5.9)
ItiseasybasedonthistoshowthatE
X
|n] = E
X
|n
i
].
Finally, the same will hold for the kinetic energies I|n], I
S
|n].All this shows
thatwemustdemandthisinvarianceofthecorrelationenergy:
E
C
|n
ii
] = E
C
|n
i
] = E
C
|n] (7.5.10)
One consequence is the property of Galilean covariance of the potentials for
eachoftheaboveenergyfunctionals.Forexample,forthecorrelationenergy
we have the following result. Suppose we shift the density by a small
displacement:
n
i
(r) = n(r
i
+oL) = n(r
i
) +oL Vn(r
i
)
Thus:
Election Bensity Functional Theoiy Page 17u
Roi Baei

E
C
|n
i
] = E
C
i
|n +oL Vn] = E
C
|n] +oL _
oE
C
on(r
i
)
Vn(r
i
)J
3
r
i
(7.5.11)
SincewedemandGalileaninvarianceE
C
|n
i
] = E
C
|n]wefind:
oL _
oE
C
on(r
i
)
Vn(r
i
)J
3
r
i
= u (7.5.12)
Since
6L
C
6n(r
|
)
= :
C
|n](r)andoLisarbitrary:
_:
C
|n](r) Vn(r)J
3
r = u (7.5.13)
Forfinitesystemswherethedensitydropstozeroatinfinitywecanmovethe
nablasigntothepotential:
_-V:
C
|n](r) n(r)J
3
r = u (7.5.14)
-V:
C
|n](r)istheforcederivedfromthecorrelationpotential.Thisshowsthat
thetotalcorrelationforceisalwayszero.AnotherconsequencefromGalilean
invarianceofthecorrelationenergyistheGalileancovariance:
:
C
|n
i
](r
i
) = :
C
|n](r) = :
C
|n](r
i
+L) (7.5.15)
Similarconditionscanbeprovedfromrotationalinvariance.Forexamplethe
torque:
_r V:
C
|n](r) n(r)J
3
r = u (7.5.16)
G. Holes and the adiabatic connection
i. The exchange and correlation holes
Letusnowtakeastepbackandreturntowavefunctiontheory.Weexamine
theelectronelectroninteractionenergy
Election Bensity Functional Theoiy Page 171
Roi Baei

u = (
gs
|u

|
gs
) (7.6.1)
Whichwewriteusingthefollowingoperator:
u

=
c
2
2

1
r
]
=]
=
c
2
2

n
2
(r, r
i
)
|r -r
i
|
J
3
rJ
3
r
i

(7.6.2)
In the first term, we sum over the pairs of rvectors of each of the N
coordiantes(indicesiand]).Inthesecondtermweusethedefinition:
n
2
(r, r
i
) = o(r -r

)o(r -r
]
)
=]
,
(7.6.3)
whichisthepairdensityoperator.Notetherelationbetweentheoneandtwo
densties:
n
2
(r, r
i
) = o(r -r

)o(r' -r
]
)
,]
-o(r -r

)o(r' -r

= n(r)n(r') -o(r -r
i
)n(r) = n(r)|n(r
i
) -o(r -r
i
)].
(7.6.4)
Withn(r) = o(r -r

)
N
=0
.Withthisdefinition,wehave:
_n(r)J
3
r = N
_n
2
(r, r
i
)J
3
r
i
= (N -1)n(r)
_n
2
(r, r
i
)J
3
r
i
J
3
r = N(N -1)n(r)
(7.6.5)
Theexpectationvalueofn(r, r
i
)isthetwoelectrondensityfunction:
(r, r
i
) = (
gs
|n
2
(r, r
i
)|
gs
) (7.6.6)
Withthetwoelectrondensityfunction,theinteractionenergyis:
u =
c
2
2

(r, r
i
)
|r -r
i
|
J
3
rJ
3
r
i

(7.6.7)
Election Bensity Functional Theoiy Page 172
Roi Baei

This pair density function has the symmetry, positivity and normalization
propertiesgivenby:
(r, r
i
) = (r
i
, r)
(r, r
i
) u
_(r, r
i
)J
3
r
i
= (N -1)n(r)
_(r, r
i
)J
3
r = (N -1)n(r
i
)
(r, r
i
)J
3
rJ
3
r
i
= (N -1)N
(7.6.8)
Thenormalizationallowsinterpretationof
I(r.r
|
)
N(N-1)
astheprobabilitydensityto
findanelectronatrandanotherelectronatr
i
.Onepropertythatisintuitively
expected of (r, r
i
) is as the limit |r -r
i
| - is approached electrons will
graduallyuncorrelateand(r, r
i
)collapsetothedensityproductn(r
1
)n(r
2
).
Indeed,thisisinsightbearsoutinmostcases:
(r, r
i
) - n(r)n(r
i
) (|r -r
i
| - , in most cases), (7.6.9)
but not always. For example, the ground state wavefunction of the Carbon
atom
C
(r
1
, r
2
, ) (we are neglecting to write spin indices for sake of
notational simplicity) in the large r
1
limit: for minimal energy reasons the
remaining electrons will lower their energy to a maximal extent thus, the
wave function should obey:
C
(r
1
, r
2
, ) - _
n(r
1
)
N

C
+
,r
1
(r
2
, ) where

C
+
,r
1
(r
2
, ) is the ground state wave function of the cation C
+
and _
n(r
1
)
N
is
therootofthepropabilitytofindanelectronfaratr
1
.Notehoweverthatthis
cation has a 3fold degeneracy in its groundstate energy and thus for any
finite r
1
, no matter how large, the
C
+
,r
1
(r
2
, ) wave function is that
Election Bensity Functional Theoiy Page 17S
Roi Baei

degeneratewave function distortsina certain fashionin correlation withthe


direction r
1
r
1
(for more details see Phys. Rev. A 49, 809 (1994) or J. Chem.
Phys.105,2798(1996)).
We can also look at the conditional probability density to find an electron at r
giventhatthereisoneatr
i
(thislatterprobabilityisn(r
i
)N
c
),givenby
P(r|r') =
(r, r
i
)
(N
c
-1)n(r
i
)
(7.6.10)
Obviously,ifoneintegratesonronegetsunity.Wecanthusview:
n
cond
(r|r
i
) = (N
c
-1)P(r|r
i
) =
(r, r
i
)
n(r
i
)
(7.6.11)
as a conditional density, the density at r of N
c
-1 electrons: all other
electrons except that one electron is known to be at r
i
. Indeed, upon
integrationoverr,weget,irrespectiveofr
i
:
_n
cond
(r|r
i
)J
3
r = (N
c
-1) (7.6.12)
Furthermore,wehavefor|r -r
i
| - :
n
cond
(r|r
i
) - n(r) (|r -r
i
| - , in most cases), (7.6.13)
This shows that the density far from the localized electron is unperturbed.
Now,letussubtractfromthisconditionaldensitythetotalN
c
electrondensity
n(r)andobtaintheFermiCoulombholefunction:
n
PC
(r|r
i
) n
cond
(r|r
i
) -n(r) (7.6.14)
Sincewelocalizedanelectronatr
i
therestoftheelectronswillrearrangeso
as to be repelled from the stationary source. This will give us a missing
densityorholedensity,i.e.thechargedensityatrexpelledbyanelectron
Election Bensity Functional Theoiy Page 174
Roi Baei

atr
i
.Weexpectthetotalchargeoftheholeis-1.IndeedpluggingEq.(7.6.14)
into(7.6.12)wefind:
_n
PC
(r|r
i
)J
3
r = -1. (7.6.15)
FurthermorefarfromtheholecenterwehavefromEqs.(7.6.13)and(7.6.14):
n
PC
(r|r
i
)
n(r)
- u (|r -r
i
| - , in most cases), (7.6.16)
This shows that the FC hole decays faster than the density far from the
system: for localized systems it is, in most cases, a highly localized overall
singlychargeddistribution.
Now, because (r, r
i
) = n(r
i
)n
cond
(r|r
i
) = n(r
i
)(n(r) +n
PC
(r|r
i
)), the
Coulombinteractionenergycanbewrittenas:
u =
c
2
2

n(r
i
)(n(r) +n
PC
(r|r
i
))
|r -r
i
|
J
3
rJ
3
r
i

(7.6.17)
AndintermsoftheFChole:
u = E
H
+
c
2
2

n(r
i
)n
PC
(r|r
i
)
|r -r
i
|
J
3
rJ
3
r
i

(7.6.18)
ThusthepartoftheinteractionenergybeyondtheHartreeenergyisthesum
of all interaction energies e
PC
(r
i
) =
c
2
2
]
n
FC
(r|r
|
)
|r-r
|
|
J
3
r between an electron at r
i

anditsFermiCoulombholen
PC
(r|r
i
).Wewillshortlyseethatthecorrelation
energyadimitsasimilaranalysisonlywithslightlymodifiedquantities.
By using noninteracting electrons we can also pull out of the integral the
exchangeenergyandwrite:
u = u
S
+
c
2
2

n(r
i
)n
C
(r|r
i
)
|r -r
i
|
J
3
rJ
3
r
i

(7.6.19)
Election Bensity Functional Theoiy Page 17S
Roi Baei

This leaves a Coulomb hole which is overall neutral. It too is localized. We


discussthisinthesectionafternext.
ii. The Fermi-Coulomb hole for harmonic electrons
Letuscalculatethesequantities forour2harmonicelectronsin theirground
statetriplet(sowehavebothexchangeandcorrelation).Thepairdensityand
densityforthewavefunctionin(1.6.18)is:
(x
1
, x
2
) =
2
n
_
2
3

1
c
-

1
(x
1
+x
2
)
2
+
2
(x
1
-x
2
)
2
2 (x
1
-x
2
)
2

(7.6.20)
Thedensitycanbeobtainedbyintegrating:
n(x) =
_
8
n

2
3

1
(
1
+
2
)
52
c
-2

1

2

1
+
2
x
2
( +
2
+4x
2

2
_)
(7.6.21)
The density is plotted for several values of the correlation constant 0,
cos 0 = [
y
o

2
.Thiscalculationisfor = 1:
Election Bensity Functional Theoiy Page 176
Roi Baei

Figure VII1: The 1particle density, for a system of two harmonic fermions placed in a
harmonic well in their triplet ground state for various interaction strengths. When 0 =
a
2

thereisnointeractionandthedipinx = isduetothePaulirepulsion.Asinteraction
growsthedipbecomesdeeperandbroader.

Weplottheconditionalprobabilityp(x|x
1
)forthissysteminFigure VII2.
Election Bensity Functional Theoiy Page 177
Roi Baei

Figure VII2: Contour plots for the conditional probability distribution p(x|x
1
) for a
system of two Fermions in their triplet ground state for various interaction strengths.
When 0 =
a
2
there is no interaction and the only correlation is due to the Pauli principle.
Asinteractiongrowstheprobabilitydistributionrotatesby45
u
.
TheXCholeisplottednextinFigure VII3.
Election Bensity Functional Theoiy Page 178
Roi Baei

Figure VII3:ContourplotsfortheFCholen
XC
(x, x
1
)forasystemoftwoFermionsintheir
triplet ground state for various interaction strengths. When 0 =
a
2
there is no interaction
andtheonlycorrelationisduetothePauliprinciple.
iii. The Fermi hole in the non-interacting system
LetusconsidernowtheFCholeinthenoninteractingsystem.Sincethereis
no correlation in absence of interactions, we attribute the hole only to the
exchange (Fermi) effects. A noninteracting system having the density n(r)
that has a closed shell KohnSham determinant, composed of orthonormal
orbitals
u
(r),where1indicatesspinupand1spindown.
Election Bensity Functional Theoiy Page 179
Roi Baei

S
(r, r
i
) = (|n]
gs
|n(r, r
i
)||n]
gs
)
= |
u
(r)
2

b
(r
i
)
2
-
u
(r)
b
(r)
b
(r
i
)
u
(r
i
)]
u=b
= |
u
(r)
2

b
(r
i
)
2
-
u
(r)
b
(r)
b
(r
i
)
u
(r
i
)]
u,b
= n(r)n(r
i
) -P(r, r
i
)
2

(7.6.22)
Where the sum is over the orbitals in
gs
(the occupied KS orbitals) and we
definedthedensitynoninteractingmatrixP(r, r
i
) =
u
(r)
u
(r
i
)
u
.
Exercise:Provethat
_P(r, r
i
)
2
J
3
r = n(r
i
) (7.6.23)
Exercise:Asacheck,integrate
S
(r, r
i
)overrandfind:
_
S
(r, r
i
)J
3
r = n(r
i
)|N
c
-1] (7.6.24)
TheFermiconditionaldensityis:
n(r|r
i
) = n(r) -
P(r, r
i
)
2
n(r')

(7.6.25)
AndtheFermiholeis:
n
P
(r|r
i
) = -
P(r, r
i
)
2
n(r
i
)

(7.6.26)
It can be shown[15] that in most cases the density matrix P(r, r
i
) decays
exponentiallyas|r -r
i
| - ,althoughthiscouldbemuchslowerthann(r).
Thuswemaysay:
n
P
(r|r
i
) - u (|r -r
i
| - , in most cases), (7.6.27)
This is weaker than Eq. (7.6.16) for the total FC hole. This shows that n
C
=
n
PC
-n
P
decaystozeroinasimilarbutoppositewaythann
P
(r|r
i
):
Election Bensity Functional Theoiy Page 18u
Roi Baei

n
P
(r|r
i
) - -n
C
(r|r
i
) (|r -r
i
| - , in most cases), (7.6.28)
BasedonEq.(7.6.23)theFermiholecarriesallthechargeoftheFChole:
_n
P
(r|r
i
)J
3
r = -1 (7.6.29)
This allows one to say that it is the Fermi or exchangehole in the non
interacting system that carries the charge of the exchange correlationhole
intheinteractingsystem.Oncetheinteractingsystemhasbeenmappedonto
the noninteracting system the Fermihole is easily calculated. This can be
usedtodefinetheCoulombholeby:
n
C
(r|r
i
) = n
PC
(r|r
i
) -n
P
(r|r
i
) (7.6.30)
Ithasnototalcharge:
_n
C
(r|r
i
)J
3
r = u (7.6.31)
Theinteractionenergycanbewrittennowas:
u = u
S
+
1
2

n(r
i
)n
C
(r|r
i
)
|r -r
i
|
J
3
rJ
3
r
i
(7.6.32)
Exercise:ComputetheFermiholefunctionofthehomogeneouselectrongas
Solution:Wealreadydeterminedthedensitymatrix(seeEq.(5.2.24))
P(s) = Sn
]
1
(sk
P
)
sk
P
(7.6.33)
wheres = |r -r
i
|and]
1
(x) =
sInx-x cos x
x
2
.Thexholeisofcourseindependent
ofr:
n
P
HLu
(r; r +x) = -
P(s)
2
n
= -9n
]
1
(sk
P
)
2
(sk
P
)
2
(7.6.34)
PlottingshowstheformoftheHEGxholefunction:
Election Bensity Functional Theoiy Page 181
Roi Baei

Herek
P
3
= Sn
2
n(unpolarizedgas).Since4n ]
]
1
(x)
2
x
Jx

0
= n,wefindHEG
exchangeenergyis,:
u
P
N
=
1
N
1
2

n(r)n
P
HLu
(r; r
i
)
|r -r
i
|
J
3
rJ
3
r' =
1
2
4n _
n
P
HLu
(s)
s
s
2
Js

0
= -
9n
2k
P
2
4n _
]
1
(x)
2
x
Jx

0
= -
S
2n
k
P
= -S _
S
8n
n]
13

(7.6.35)
ThisisindeedtheLDAexchangeenergyperparticle.
iv. The Adiabatic Connection
Having written down the relation between the interaction energy u and the
XChole,westillhavenosuchrelationforthecorrelationenergyE
C
.Wenow
derivesucharelation.Rememberthecorrelationenergyisdefinedas:
E
C
|n] = I|n] -I
S
|n] +(u|n] -u
S
|n]) = I
C
|n] +u
C
|n] (7.6.36)
Givenagroundstateelectrondensityn(r),considerafamilyofN
c
electron
systems,withparameteru z 1,where:
E

x
= I
`
+_:
x
(r)n(r)J
3
r +zu

(7.6.37)
The groundstate is denoted
x
. The potential :
x
(r) is chosen in such a
mannerthatthedensityatthegroundstatewavefunctionis

n(r),i.e.:
(
x
|n(r)|
x
) = n(r) (7.6.38)
-20 -10 10 20
s k
F
-1.0
-0.8
-0.6
-0.4
-0.2
r
X
HsLn
Election Bensity Functional Theoiy Page 182
Roi Baei

This is a generalization of the idea by Kohn and Sham, that the interacting
electron system is mapped onto a noninteracting electron system with the
same density. Except that now we map our system to a system of electrons
with interaction zu

. When z = u we have the noninteracting system and


:
x=0
(r)istheKohnShampotential:
s
(r)andwehave:
E
0
|n] = I
S
|n] +_:
0
(r)n(r)J
3
r (7.6.39)
whereI
S
|n]isthekineticenergyofnoninteractingelectrons.Whenz = 1we
have the fully interacting system and :
x=1
(r) = :(r) is the actual external
potentialontheelectronsystemandtheenergyis:
E
1
|n] = E

|n] = I|n] +_:


1
(r)n(r)J
3
r +u|n] (7.6.40)
Wecanalsodefinetheobviousquantities:
E
x
|n] = I
x
|n] +_:
x
(r)n(r)J
3
r +u
x
|n] (7.6.41)
FromEq.(7.6.36):
E
1
-E
0
= _|:
1
(r) -:
0
(r)]n(r)J
3
r +u
S
|n] +E
C
|n]
E
x
-E
0
= _|:
x
(r) -:
0
(r)]n(r)J
3
r +zu
S
|n] +E
C,x
|n]
(7.6.42)
With:
E
C,x
|n] = (
x
|I
`
+zu

|
x
) -(
0
|I
`
+zu

|
0
)
Now, the groundstate energy of the intermediately interacting electrons
obeys,byHellmannFeynmanstheorem:
J
Jz
E
x
= [
x
_
J
Jz
E

x
_
x
= __
J
Jz
:
x
(r)_ n(r)J
3
r +(
x
|u

|
x
) (7.6.43)
Fromthesecondequationthen:
Election Bensity Functional Theoiy Page 18S
Roi Baei

J
Jz
E
C,x
|n] = (
x
|u

|
x
) -u
S
|n] = u
C,x
|n]
(7.6.44)
This expression is the differential form of the adiabatic connection. If we
integrateitwithrespecttozfrom0to1,wefind:
E
C
|n] = _ (
x
||u

|
x
|)Jz
i
1
0
-u
S
|n]. (7.6.45)
This formula is called the adiabatic connection formula for the XC energy
[16]. We may write: E
C,x
= I
C,x
+zu
C,x
. Then
d
dx
E
C,x
|n] =
d
dx
I
C,x
+u
C,x
+
z
d
dx
u
C,x
andso
JI
C,x
|n]
Jz
+z
Ju
C,x
|n]
Jz
= u.
(7.6.46)
Wecanrewriteu
C,x
intermsofthecorrelationhole.Indeed,ifn
C
x
(r|r
i
),isthe
correlationholeforthezsystemthenusing(7.6.32):
u
C,x
|n] =
1
2

n(r
i
)n
C
x
(r|r
i
)
|r -r
i
|
J
3
rJ
3
r
i

(7.6.47)
Fromwhich:
E
C
|n] =
1
2

n(r
i
)n
C
(r|r
i
)
|r -r
i
|
J
3
r
i
J
3
r (7.6.48)
And we see that the correlation energy can be obtained from the the z
averagedCoulombhole,calledthecorrelationhole(sinceitisassociatedwith
thecorrelationenergy):
n
C
(r|r
i
) = _ n
C
x
(r|r
i
)Jz
1
0
(7.6.49)
Notethatbecause]n
C
x
(r, r
i
)J
3
r = u,wehavealso:
Election Bensity Functional Theoiy Page 184
Roi Baei

_n
C
(r|r
i
)J
3
r = u (7.6.50)
It is interesting that the correlation energy, like to Coulomb energy, can be
represented as a Coulomb interaction of the density and a hole as in Eq.
(7.6.48). Note however that the relevant hole as a couplingconstant (z)
averagedcorrelationholeandnottheCoulombholeitself.
Let us discuss one of the important consequences of Eq. (7.6.50) i.e. that the
totalchargeofthecorrelationholeiszeroforlocalizedchargesystems.Ifwe
rewritethecorrelationenergyas:
E
C
|n] =
1
2

n(r
i
) n
C
(r|r
i
)
|r -r
i
|
J
3
r
i
J
3
r (7.6.51)
Weseethatthecorrelationenergycanbewrittenas
E
C
|n] = _e
C
(r')n(r')J
3
r' (7.6.52)
where:
e
C
(r') =
c
2
2
_
n
C
(r|r
i
)
|r -r
i
|
J
3
r
(7.6.53)
(Notethatthisisjustasuggestionsinceaddingtoe
C
(r
i
)anyfunctione
C
(r
i
)
for which ]e
C
(r')n(r')J
3
r' = u will give the same correlation energy).
Because for a fix r' n
C
(r|r
i
) is an oveall neutral charge density in r space its
Coulombic potential
c
C
(r
|
)
c
is expected to decay relatively fast for r(faster
than1r
i
).
H. Derivative Discontinuity in the exchange
correlation potential functional

Election Bensity Functional Theoiy Page 18S


Roi Baei

VIII. Approximate correlation energy


functionals
While the correlation energy in atoms and molecules is only a small fraction
of the total electronic energy, it is found that it is in fact a very large
percentage when one computes energy differences, such as energy of
atomization(i.e.thedifferencebetweentheenergyoftheatomicconstituents
totheenergyofthemolecule)orrelativeenergiesofdifferentconformations,
asthosedeterminingtheshapeoftheBornOppenheimerpotentialsurface.In
essence,theexchangecorrelationenergyisthechemicalbondingenergy.Itis
therefore crucialto modelthisenergyaccurately.We describebelowsome of
thebasicapproximationsfordensityfunctionaltheory.
A. The local density approximation (LDA)
The mapping of the interacting electron system onto the noninteracting
system,encapsulatedinEq.XX,isofformalinterestonly,unlesswedevisea
way to approximate the correlation potential. One way is to consider the
correlation energy per electron e
C
(n) in the homogeneous electron gas of
density n. This energy can be computed with relatively high precision using
MonteCarlomethods.Underthisapproximationwecanwritethecorrelation
energy as ]e
C
(n(r)) n(r)J
3
r. However, this does not yield in practice good
enoughresultsandisthusseldomused.Amoresuccessfulwayswasdevised
by Kohn and Sham. They considered both the exchange and correlation per
anelectroninthehomogeneousgas,e
XC
(n).Inthiscasethecorrelationenergy
comesout:
E
XC
LA
|n] = _e
XC
(n(r))n(r)J
3
r -E
X
|n] (8.1.1)
Election Bensity Functional Theoiy Page 186
Roi Baei

This approximation is called the local density approximation (LDA)[17]. The


functional E
XC
LA
|n]. It leads to the following LDA approximation for the
energyfunctional:
E

LA
|n] = I
S
|n] +_:(r)n(r)J
3
r +E
H
|n] +_e
XC
(n(r))n(r)J
3
r (8.1.2)
Theminimizationofthisfunctional,bytheKohnShamapproachleadstothe
LDA approximation of DFT. This approach is highly successful and is
considered in DFT as the basis for most of the developments of other
functionals.
Note however that in LDA the correlation energy is extremely awkward
lookingbecauseoftheterm E
X
|n].Thepresenceofthistermhasdetrimental
effectswhichharmsomeofthepredictionsofDFT.
i. The exchange energy per electron in the HEG
In section XXX we discussed in some detail the HartreeFock theory of the
homogeneous electron gas. We defined a Jellium as a smeared positive
background of Volume I at density n =
N
c
v
together with N
c
electrons. We
showedthattheJelliumselfenergy,theJelliumelectronattractionenergyand
the electron Hartree energy all cancel exactly in the HEG. Thus, the energy
perparticleisgivenby:
e(n) = t
S
(n) +e
XC
(n) = t
S
(n) +e
X
(n) +e
C
(n) (8.1.3)
We already calculate exchange energy, using a HartreeFock treatment of the
HEG and wesaw that e
X
= -C
X
n
13
(see Eq. (5.2.32)) with C
X
=
3
4
[
3
n

13
E
h
o
0
.
In terms of the WignerSietz radius, which is a dimensionless quantity given
by:
Election Bensity Functional Theoiy Page 187
Roi Baei

r
s
= _
S
4nn
]
13
o
0

(8.1.4)
istheradiusofaJelliumspherecontainingthechargeofanelectron.Thus:
e
X
= -_
24S
2S6n
2
]
13
E
h
r
s
= -
u.4S817
r
s
E
h

(8.1.5)
If we have a way of computing e(n) and t(n), we can then find e
XC
(n) from
Eq.(8.1.3).Wecanthenalsocomputee
C
(n)fromEqs.(8.1.5).
ii. Correlation energy of the HEG: the high density limit
The calculation of e
C
can presently be done analytically in two limits. One is
the high density limit r
s
- u where the kinetic energy dominates and the
Coulombinteractioncanbetreatedasaperturbation.Inthislimitthekinetic
energy isthatof noninteracting particles. Thusinthe perturbativeapproach
onecantakeE

0
= -
l
2
2m
c
v
n
2
N
c
n=1
.TheunperturbedwavefunctionistheSlater
wave functional wave function
0
composed of the plane wave orbitals
c
ikr
v

with k taking the N
c
lowest momentum vectors k < k
P
. The energy of the
unperturbedstateisE
0
=
l
2
k
2
2m
c
k<k
F
.Thefirstordercontributionofthetheee
Coulomb repulsions is (
0
|u

|
0
). This quantity is equal to the direct and
exchange contribution (see Eq. (4.1.12)). The direct part is nullified by the
other electrostatic interactions, so the 1
st
order contribution is essentially the
exchange energy of the HEG which we already included e (see Eq. (8.1.3).
Thus to continue and determine the interaction energy beyond exchange, i.e.
the correlation energy, we must move to at least second order perturbation
theory. When one does this, one finds that the usual second order
perturbationtheoryyieldsinfiniteterms.Theseareassociatedwithlowwave
lengthexcitationswhereapairofelectronshavingthemomentumstatse|k
1
)
Election Bensity Functional Theoiy Page 188
Roi Baei

and|k
2
),areexcitedbytheCoulombinteractionI
q
= 4nq
2
tostates|k
1
+q)
and |k
2
-q) (such that
1
2
(k
1
+q)
2
> e
P
and
1
2
(k
2
-q)
2
> e
P
). One can show
thatforsmallqthisprocessgivesatermproportionaltolnqintheexpression
forthe2
nd
orderperturbationenergy.Thistermissingularatlowq.Amethod
of performing perturbation theory which is nonsingular and goes beyond
second order, was devised in 1957[18]. This theory is essentially exact at the
highdensitylimitandleadstothefollowingrelation:
e
C
= Aln r
s
+C +0(r
s
) (8.1.6)
Wherer
s
=
1
u
0
[
3
4nn

-3
istheWignerSeitzradius,namelyo
0
r
s
istheradiusofa
sphereintheJeliumwhichscoopsanamountofchargeequalto1c,wherecis
the elecmentary quantum of charge (the electron charge). A = u.uS11E
h
and
C = -u.u48 _u.uu1E
h
. Later work refined these constants: A = u.uS1u91E
h

andC = -u.u46644E
h
.
Exercise VIII1
UsingEq.(1.9.43)andtheHellmanFeynmantheoremprovethat
t
C
t -t
S
= S:
XC
HLu
-4e
XC
HLu
= S:
C
HLu
-4e
C
HLu
(8.1.7)
Hints:
(a)Showthatc
2
c
HEG
c
2
= u
cc
-u
H
whereu
cc
andu
H
are,respectively,thetotal
electronicrepulsionenergyperelectronandtheHartreeenergyperelectron.
(b)Fromthethefactthate
X
n
13
showthatthelastequalityiscorrect.

Election Bensity Functional Theoiy Page 189


Roi Baei

iii. Correlation energy of the HEG: the low density limit and the Wigner
crystal
Thesecondlimitisthatoflowdensity,wheretheelectronsformacrystal.We
give the development of this limit, originally proposed by Wigner[19] when
hedevisedatheoryfortheelectrondensityinmetallicsodium.
Wigner assumed that at low energy the homogeneous electron gas forms a
crystal.Nowthatmaysoundstrange:howcanthedensitybeuniformandat
the same time the electrons form a crystal? Thanks to Quantum Mechanics
thisisactuallynotacontradiction,asthefollowingexampleshows.
Exercise: Show that for 2 particle in a 3D box of volume I with periodic
boundaryconditions,iftheHamiltonianis:
E

= -
l
2
2m
v
1
2
-
l
2
2m
v
2
2
+u(r
1
-r
2
)
(8.1.8)
Then
1) Theeigenstateshaveahomogeneous1particledensityn =
2
v

2) Thepaircorrelationfunctionhasstructure.(g(x
1
, x
2
) =
P(x
1
|x
2
)
P
clcss
(x
1
|x
2
)
=
I(x
1
,x
2
)
n(x
1
)n(x
2
)
)
In the low density regime the electron kinetic energy (per electron) can be
neglected since, as seen in XXX it is proportional to r
s
-2
while the repulsion
energy between the electrons per electron is proportional to r
s
-1
. At lkow
density the Pauli exclusion priniciple is nonoperative, since electrons do not
overlap.Thusthequantumnatureoftheelectronisgoneatthislimitandwe
can think of the electron as a classical particle that localizes. This is because
nonlocalization of particles in quantum mechanics arises only from their
need to reduce kinetic energy. The electrons will arrange themselves in the
lowestenergystatebyformingaclosepackedcrystal.Eachelectronisthenas
far as possible from each other electron, while still filling 3D space with
Election Bensity Functional Theoiy Page 19u
Roi Baei

averagedensity n .Letuscalculatetheenergyofsuchacrystal.Considerone
oftheelectronsinthecrystal.Weimagineittogetherwithacellcontaining1
unit of positive charge. This cell shape depends on the crystal symmetry.
Following Wigner, we neglect the crystal structure and assumeeach electron
is surrounded by a sphere of positive charge completely neutralizing it. Our
approximation then neglects the volume of the space between the spheres.
Theradiusofthesphereisr
s
anditisfilledwithsmearedpositivechargeand
with one negative charged pointelectron at its center. The spheres do not
interactsincetheyareneutralandhavenoelectricmoments.
The total energy per electron is the energy e
sp
(r
s
o
0
) to assemble the Jellium
sphereandtheenergye
cI
(r
s
o
0
)neededtobringtheelectronfrominfinityinto
thecenterofthesphere.
Lete
sp
(R)betheenergytoassembleasphereofchargedensitynandradius
R. Suppose we now enlarge it by adding a shell of radius JR. The electric
potentialatdistancer > Routsidethesphereisrwhere =
4n
3
cnR
3
isthe
chargeinthesphere.ThechargeintheshellisJq = n4nR
2
JRandbringingit
from infinity, where the potential is zero to its place on top of the existing
sphereinvolvestheenergyJe
sp
= Jq

R
=
(4n)
2
3
R
4
c
2
n
2
JR .Thus,byintegration
fromutoR,wefind:e
sp
(R) = [
4nn
3

2
3
5
R
5
.AndsoatR = r
s
o
0
:
e
sp
(r
s
) =
S
S
1
r
s
E
h
(8.1.9)
Next,wewanttocalculatetheenergytobringanelectronfrominfinitytothe
centerofthesphere.Thiswillbedoneintwostages,firstbringingtheelectron
frominfinitytotherimofthesphere,adistanceo
0
r
s
fromitscenterandthen
fromtherimtothecenter.Accordinglywritee
c
= e
m
+e
ccntc
.Thefirstpart
Election Bensity Functional Theoiy Page 191
Roi Baei

iseasysincewealreadyknowthepotential,anditisnegativesinceenergyis
releasedbythisprocess,so:
e
m
= -
1
r
s
E
h
(8.1.10)
Insidethesphere,atadistanceRfromthecenterthereexistsanelectricfield
dueto the Jellium,which according toGaussisF =
(R)R
R
3
=
4n
3
nR.Thisforce
isaHarmonicforce,withforceconstantk
H
=
4n
3
n =
1
(u
0

s
)
3
.Theworktomove
an electron in this field to the center is: e
ccntc
= ] F(R) JR
0

s
= -
1
2
s
. The
eneigy foi the seconu stage is theiefoie e
cI
(r
s
) = -
3
2
L
h

s
anu the total eneigy pei
election in the ciystal is:
e(r
s
) = -
9
1u
1
r
s
E
h
(8.1.11)
Thisthenistheexchangecorrelationenergyforlowdensity.Weneglectedthe
volume between the spheres. The exchange energy we already know from
(8.1.5), is e
X
(r
s
) = -
0.45817

s
E
h
. Thus, Wigners approximation for the
correlationenergyinthelowdensitylimitis:
e
C
(r
s
) = e(r
s
) -e
X
(r
s
) = -
u.4418S
r
s
E
h
(low Jcnsity, r
s
- ) (8.1.12)
Wigneralsoconsideredthecorrectionduetothefinitekineticenergywhenr
s

is finite. Since we saw that the electron inside the spherical Jellium drop is a
Harmonic potential, one can reduce the correlation energy by the 3D
Harmoniczeropointpotential,
3
2
k
H
=
3
2
s
32
.Thecorrelationenergyisthen:
e
C
(r
s
) = -
u.4418S
r
s
E
h
+
S
2r
s
32
(low Jcnsity, r
s
- )
(8.1.13)
Election Bensity Functional Theoiy Page 192
Roi Baei

iv. Monte-Carlo determination of the correlation energy for the HEG


Betweenthesehighandlowlimitsthereisnoanalyticaltheory,ingeneraland
a numerical computation can be made based on quantum Monte Carlo
methods. The results of the calculation are then fitted to an analytical form
whichrespectsthelimits
v. The polarized HEG; local spin-density approximation (LSDA)
Uptonowwehaveassumedthattheelectrongasisunpolarized,i.e.thetotal
z component of spin Sz per electron is zero. However, Sz is a good quantum
number and it can vary continuously from -
1
2
to
1
2
. The extreme case is the
fullypolarizedcase.Ingeneralonemaydefinethedensityofspinupelectron
n
l
(r)andthatofspindownn
l
(r).Then:
n(r) = n
l
(r) +n
l
(r)
(r) =
n
l
(r) -n
l
(r)
n
l
(r) +n
l
(r)

(8.1.14)
For a fully polarized gas = 1 and the difference is first of all in the Fermi
energy. For the HEG, since every momentum state can populate only one
electron,wefindbyasimilaranalysisasintheunpolarizedcase:
k
Pll
= k
P
(1 _)
13
e
Pll
=
l
2
k
Pll
2
2m
c
= e
P
(1 _)
23

t
ll
=
I
ll
N
ll
=
S
S
e
Pll

(8.1.15)
The total kinetic energy us the sum of up and down contributions: t =
n
l
t
l
+n
l
t
l
n
,whichisevaluatedtobet =
3
5
c
Fl
n
l
+c
Fl
n
l
n
usingtheexpressionsforn
ll

weobtain
Election Bensity Functional Theoiy Page 19S
Roi Baei

t =
S
S
e
P
(1 +)
53
+(1 -)
53
2

(8.1.16)
As for exchange energy, since exchange interaction occurs only between like
spins,weXXXXX

vi. Successes and failures of LSDA


vii. Plausible reasons for the success of LSDA
OneoftheusesofEq.(7.6.48)expressionistoexplainthesuccessofasimple
theory such as LDA[20]. TO see this, let us expand the XC hole in terms of
momentsaroundr:
n
XC
x
(r, r +R) = n
Im
x
(r; R)
Im
(0, )
I
m=-I

I=0
(8.1.17)
where:R = Ru (R sin0 cos , R sin0 sin, R cos 0)and
n
Im
x
(r; R) = _ _ n
XC
x
(r, r +Ru)
2n
0

Im
(0, ) sin0 J0
n
0
J (8.1.18)
ThenconsidertheXCenergy,itcanbewrittenas:
E
XC
|n] =
1
2

n(r)n
XC
(r, r +R)
R
J
3
RJ
3
r
=
1
24n

n(r)n
00
(r, R)
R
J
3
rR
2
JR
(8.1.19)
Were:
n
00
(r, R) = _ n
00
x
(r, R)Jz
1
0
(8.1.20)
And:
Election Bensity Functional Theoiy Page 194
Roi Baei

4n _ n
00
(r, R)R
2
JR

0
= -1 (8.1.21)
Thus,onlythe00momentofthezdependentXCholesenterstheexpression.
Therefore,inasensetheangularshapeoftheaverageXCholegetsaveraged
over and only the radial dependence affects the XC energy. This is used to
explainsomeofthesuccessofLDA.
WeseethattheXCanisotropyoftheXCholearound r isaveragedover.Only
thisaverageenterstheXCenergyformula.InLDAweusethehomogeneous
electrongastocomputetheHEG.OfcoursethisleadstoanisotropicXChole.
Yet,sinceonlythesphericalaverageoftheholeentersintotheXCenergy,this
drasticapproximationgivesareasonablygoodXCenergy.
Exercise:CalculatethesphericallyaveragedXholefora1electronsystem(H
atomforexample)
Solution:Theorbitalis(r),thedensityisn(r) = (r)
2
andtheDMis
P(r, r
i
) = (r)(r')thus:
n
X
(r; r
i
) = -n(r
i
) (8.1.22)
Theholeisindependentofthereferencepointr.
Oneoftheimportantresultsshownbelowisthatonlythespherically
averagedholeenterstheXCenergy.Thusweonlyneedthespherically
averagedhole:
n
X
SA
(r; s) =
1
4n
_n
X
(r; r +x)J
s
(8.1.23)
Whichbecomes:
n
X
SA,1cI
(r; s) =
1
4n
_n(r +x)J
s
(8.1.24)
Election Bensity Functional Theoiy Page 19S
Roi Baei

FortheHatomn(r) = A
2
c
-u
,defining[ = or
2
+s
2
y =
2s
(
2
+s
2
)
therefore
[1 _y = o|r _s|andy[
2
= o
2
2rs:
n
X
SA,1cI
(r; s) =
1
4n
A
2
_c
-u|r+x|
J
s
=
1
2
A
2
_ c
-u
2
+s
2
+2scos0
J cos
n
0
=
A
2
2y
_ c
-[1+
Jy
y
-y
=
A
2
o
2
2rs
|(1 +o|r -s|)c
-u|-s|
-(1 +o|r +s|)c
-u|+s|
]

(8.1.25)
Theformofthissphericallyaveragedholefunctionasfunctionofsandris
shownhere:

Theholehasacuspatthersorigin.

Election Bensity Functional Theoiy Page 196


Roi Baei

B. Semilocal functionals and the generalized


gradient approximation
IX. Generalized Kohn-Sham approaches
A. The generalized Kohn-Sham framework
The KohnSham approach is based on pure density functionals for the
exchangecorrelationenergy.ThisapproachisbasedontheHohenbergKohn
universaldensityfunctionalF
HK
|n]iswrittenas:
F
HK
|n] = I|n] +u|n] (9.1.1)
WenotedoneproblemwithF
HK
isthatitisnotdefinedforalldensities.This
problemcanbecuredbytheLevyLiebprocedure,wheretheF
HK
isreplaced
byafunctionalbasedonaconstrainedminimumprinciple:
F
LL
|n] = min
+-n
(|I
`
+u

|)
(9.1.2)
Similarly,forthenoninteractingsystem,onefinds:
I
S
|n] = min
4-n
(|I
`
|) = min
4-n
S|]
(9.1.3)
Where here the search is over all determinants. Here we defined here a new
determinantal functional S|] with the intent of generalizing the KS
approach, as we do now. Using the minimizing determinant
S
, the Kohn
Shamcorrelationenergy:
E
C
|n] = F
LL
-I
S
-u
S
(9.1.4)
IswrittenusingI
S
= (
S
|I
`
|
S
) andu
S
= (
S
|u

|
S
).
Since a Slater wave function is given in terms of orthonormal oneelectron
orbitals
n
(x),itisbesttoviewSasafunctionaloftheorbitalsandwewrite
S|{]].WecannowgeneralizeanddemandthatS|{]]notbejustthekinetic
Election Bensity Functional Theoiy Page 197
Roi Baei

energy but a more general functional of the Slater wave function. We denote
thefunctionalderivativeofSasfollows:
oS|]
o
n
(r)
= 0
`
S

n
(r) (9.1.5)
Where0
S
issomeconvenientoperator.
We now define the energy functional for a given potential :
S
(r) and
determinant:
E
S
|:
S
, ] = S|] +] :
S
(r)n|](r)J
3
r (9.1.6)
Where:
n|](r) = (|n(r)|) (9.1.7)
We then minimize with respect to the orbitals of
, keeping them noimalizeu. This constrained minimization results in the
orbitalequations:
[0
`
S
+:
S
(r)
n
= e
n

n
. (9.1.8)
Furthermore, let us define the density functional F
S
resulting from
minimizingSwithconstrainedofagivendensity:
F
S
|n] = min
4-n
S|].
(9.1.9)
FromthiswefurtherdefinethedensityfunctionalR
S
|n]:
F
LL
|n] = F
S
|n] +R
S
|n] (9.1.10)
InactualimplementationsthefunctionalR
S
mustofcoursebeapproximated.
Now we put all ingredients together. The HK DFT energy is developed as a
trainofequalitiesasfollows:
Election Bensity Functional Theoiy Page 198
Roi Baei

= min
n-N
_F
LL
|n] +_:(r)n(r)J
3
r_
= min
n-N
_F
S
|n] +R
S
|n] +_:(r)n(r)J
3
r_
= min
n-N
_min
4-n
S|] +R
S
|n] +_:(r)n(r)J
3
r_
= min
4-N
_S|] +R
S
|n|]] +_:(r)n|](r)J
3
r_
(9.1.11)
In the last step we changed the order of the minimum procedure, assuming
this is OK. Note that the final procedure is minimization with respect to
orbitals instead of to density. The minimization, under normality of the
orbitals
n
(x) from which the Slater wave function is built is gives the
generalizedKohnShamequation:
[0
`
S
+:
R
(r) +:(r)
n
= e
n

n
(9.1.12)
Where:
:
R
(r) =
oR
S
on(r)
(9.1.13)
Ifwewrite:
:
S
(r) = :
R
(r) +:(r) (9.1.14)
we find that the orbitals also obey equations (9.1.8). Thus, the density n(r)
obtainedfromtheorbitalsoftheGKSprocedureistheminimizingdensityof
thefollowingdensityfunctional
E
S
|:
S
] = min
n(r)
|F
S
|n] +] :
S
(r)n(r)J
3
r]
(9.1.15)
The strength of the GKS scheme is that it incorporates explicit orbital
functionals.ThisgreatlyenlargesthescopeofapproximationsforDFT.
Election Bensity Functional Theoiy Page 199
Roi Baei

B. Kohn-Sham from generalized KS


By choosing: 0
S
|] = (|I
`
+u

|) one immediately obtained the usual KS


theory,withR
S
|n] = E
C
|n].
C. The Hybrid functional of Becke
LetusreturntotheadiabaticconnectionformulaEq.(7.6.45):
E
C
|n] = _ (
x
|u

|
x
)Jz
1
0
-u
S
|n] (9.2.1)
Becke [21], suggested the following trapezoidal approximation for the
integral:
_ (
x
|u

|
\
)Jz
1
0
=
(
0
|u

|
0
) +(
1
|u

|
1
)
2
(9.2.2)
LeadingtotheBeckesHalf&Halfapproximation:
E
C
|n] =
1
2
u
C
1
-
1
2
u
S
=
1
2
(u
C
1
+E
H
) -
1
2
K
(9.2.3)
Thefunction(u
C
1
+E
H
)|n]isnextapproximatedasalocaldensityfunctional:
(u
C
1
+E
H
)|n] = _u
XC
(n(r))n(r)J
3
r (9.2.4)
Whereu
XC
(n) = e
XC
(n) -t(n) +t
S
(n).theinteractionenergyperelectronina
homogeneouselectrongasisu
XC
(n).Thisfunctioncanbecomputedusingthe
QuantumMonteCarloresultsconcerningtheXCenergy.
Becke showed that this half and half theory gives a big improvement over
LDA and in some cases also over GGA functionals. This is because he has
foundawaytoapproximateHartreeFocktheoryandLDA.
The Becke Half & Half orbital (BH&H) was the first example of a hybrid
theory, a theory which is a mixture of a HartreeFock functional of orbitals
Election Bensity Functional Theoiy Page 2uu
Roi Baei

and a DFT functional of density. Becke eventually wrote down a functional


whichcontains3parameters[22]:
E
B3LP
|
1
, ,
N
c
]
= E
XC
LSDA
|n] +o
0
(K -E
X
LSDA
|n]) +o
X
(E
X
B88
-E
X
LSDA
|n])
+o
C
(E
C
PW91
|n] -E
C
LSDA
|n])
(9.2.5)
WeseethattheB3LYPfunctionalstartsfromaLSDAfunctionalandaddstoit
3 corrections. One is a fraction of the explicit orbital exchange correction
(K -E
X
LSDA
|n]) then a correction (E
X
B88
-E
X
LSDA
|n]) to the LSDA exchange
given by a GGA type functional called Becke88 [23] and finally a GGA
correction for the correlation energy, given by the PW91 functional [24]. The
valuesofthe3parameterso
0
= u.2u,o
X
= u.72ando
C
= u.81werefoundby
optimizing the performance of the functional for 56 atomization energies, 42
ionizationpotentials(calculatedbySCFmethod)and8protonaffinities.
D.
E. Long-Range self-repulsion and lack of
derivative discontinuity
F. Range separated hybrids
G. Orbital functionals and optimized effective
potentials
H. Approximate correlation functionals and the
Born-Oppenheimer force on nuclei
The KohnSham density functional theory method (KSDFT), when applied
for electrons of molecules makes sense only for frozen nuclei, because the
external potential in DFT is assumed timeindependent. However, the nuclei
Election Bensity Functional Theoiy Page 2u1
Roi Baei

in molecules are not in general motionless. How do we KSDFT a useful


approachformolecules?HereweusetheBornOppneheimerapproximation,
which allows us to divide the electronnuclear problem into 2 stages: first
computing the electronic energy E({R]) for each nuclear configuration
{R] and then combining it with the nuclear repulsion energy I
cp
({R]) to
obtainaneffectivepotentialforthemotionofthenuclei.Thenuclearelectron
potentialisgivenby:
:
nucI
(r, {R]) = -
Z
I
c
2
|r -R
I
|
I

(9.3.1)
Inordertocomputetheelectronicenergywefirstwritedownthefunctional:
E
({R])
|n] = I
s
|n] +_:
nuc
(r, {R])n(r)J
3
r +E
HXC
|n] (9.3.2)
We then minimize it under the constraint for the number of electrons
] n(r)J
3
r = N.Thisgivesaminimizingdensityn
-
(r)andwehave:
E|{R]] = E
({R])
|n
-
] = I
s
|n
-
] +_:
nuc
(r, {R])n
-
(r)J
3
r +E
HXC
|n
-
] (9.3.3)
SinceinexactDFTE|{R]]istheexactelectronicenergy,weareassuredthat:
F
I
= _-v
I
:
nuc
(r, {R])n
-
(r)J
3
r (9.3.4)
Since thisrelationis true for theelectronic Hamiltonian(ascanbeseenfrom
HellmanFeynamntheorem).
However,whathappenswhenweuseanapproximationforE
HXC
,asdonein
all applications of KSDFT? Does this relation still hold? We now show that
indeeditdoes.
Thedensityn
-
isdeterminedby:
Election Bensity Functional Theoiy Page 2u2
Roi Baei

oE
({R])
|n]
on(r)
_
n
-
= p
(9.3.5)
Theelectronicforceonthenucleiisgivenby:
F
I
= -v
I
E = -v
I
E
({R])
|n
-
]
= _-v
I
:
nuc
(r, {R])n
-
(r)J
3
r
-_
oE
({R])
|n]
on(r)
_
n
-
v
I
n
-
(r)J
3
r
(9.3.6)
Thesecondintegraliszerobecause:
_
oE
({R])
|n]
on(r)
_
n
-
v
I
n
-
(r)J
3
r = _pv
I
n
-
(r)J
3
r = pv
I
_n
-
(r)J
3
r = u
(9.3.7)
Thuswefindthattherelation:
F
I
= _-v
I
:
nuc
(r, {R])n
-
(r)J
3
r (9.3.8)
still holds. Even if the approximate XC functional does not yield the exact
density,theformalrelationbetweentheforceandthedensityisstillvalid.

Election Bensity Functional Theoiy Page 2uS
Roi Baei

X. More on the DFT correlation energy


A. Approximations to E
X
are approximation to E
C

In KohnSham DFT the ground state energy of a system of electrons having
theparticledensityn(r)iswrittenasasumofdensityfunctionals[17]:
E = I
s
|n] +] :(r)n(r)J
3
r +E
H
|n] +E
XC
|n] (10.1.1)
Here,I
s
|n]isthekineticenergyofasystemofnoninteractingfermionshaving
the same mass and groundstate density as the electrons, :(r) the external
potential exerted on the electrons, E
H
=
1
2
_
n(r)n(r
|
)
|r-r
|
|
J
3
rJ
3
r
i
is the socalled
Hartree energy of the system and E
XC
|n] is the XC energy functional. The
explicit form of the latter functional is formally known but practically
inaccessible to computation. The basic challenge in DFT is to develop an
efficaciousapproximationtotheXCenergyfunctional.
The most widely used functionals of DFT fall into two broad categories.
Semilocalfunctionalswhicharebasedlocallyonthedensitytheorbitalsofthe
noninteracting system or on their derivatives, such as the local density
approximation(LDA),generalizedgradientsapproximation(GGA)andMeta
GGAs[23,25]Thesecondcategoryincludesnonlocalfunctionalssuchasthe
widely used B3LYP[22]. Recently a new brand of hybrid functionals was
developed,designedtoremovethelongrangeselfinteractioninDFT[2628].
For a given density one can use the wave function
S
of the noninteracting
system and define the exact exchange energy E
X
|n] = (
S
|u

|
S
) -E
H
,
whereu

=
1
2

nm
n=m
istherepulsionenergyoperator.Thefunctionaldefined
by E
C
|n] = E
XC
|n] -E
X
|n] is called the correlation energy functional. Kohn
Election Bensity Functional Theoiy Page 2u4
Roi Baei

and Sham[17] recognized that in practical calculations it is beneficial to


approximate the exchange energy than to use its exact form. Indeed it was
foundthat the errors introduced into E
XC
byusingsimple approximationsto
E
C
|n] are partially cancelled by the approximate treatment of the exchange
energy. This error cancellation is the one of the major reasons for success of
DFT.SinceE
X
isactuallyknownexactlyandeasytocalculate,itisevidentthat
usinganapproximateexchangefunctionalE
X
uppox
inplaceoftheexactE
X
isa
way to correct the approximate correlation energy. For example, the
correlation energy functional in the LDA is NOT the LDA of the correlation
energyE
C
HLu
= ]e
C
HLu
(n(r))n(r)J
3
r,butinstead:
E
C
LA
= E
C
HLu
+__e
X
HLu
(n(r))n(r)J
3
r -E
X
_ (10.1.2)
This view emphasizes the fact that LDA of local exchange is a way of
correcting the correlation energy for inhomogeneous systems. Similar
expressions are to be used in more advanced functionals (GGA etc). Any
correction to e
X
for example, using a GGA type expression
]
X
(n(r), |vn(r)|) J
3
r should be viewed as an essentially correlation
correction.
This way of thinking leads naturally to hybrid functionals. A hybrid
functional is a further improvement of LDA (or GGA) obtained simply by
multiplyingthecorrectionin(10.1.2)byafactorz:
E
C
HB,1
= _e
C
HLu
(n(r))n(r)J
3
r +z __e
X
HLu
(n(r))n(r)J
3
r -E
X
_ (10.1.3)
Or,basedonBLYP,oneoftheGGAfunctionalswecanwrite:
E
C
HB,2
= _
C
LP
(n(r), |vn(r)|)J
3
r +z __
X
B
(n(r), |vn(r)|)J
3
r -E
X
_ (10.1.4)
Election Bensity Functional Theoiy Page 2uS
Roi Baei

Becke [22] found by a series of calculations on many molecules that the


naturalassumptionz = 1isnotoptimalandzshouldbesomewhatsmaller,
around 0.8. This lead to development of the extremely successful B3LYP
functional. However, experience has shown that some systems require lower
values of z while others strive for higher values. An example is reaction
barriers where lower values of z (around 0.50.7) were needed. There is
presentlynogeneralsystematicwaytosetthevalueofz.
ThehybridDFTapproachesarewellestablishedelectronicstructuremethods
however in the past few years it has become increasingly clear that several
typesofcalculations,forexample,polarizability,electrontransferexcitations,
Rydbergstatesandchargeallocationinweaklycoupledsystemsareoftennot
well described by the above DFT/TDDFT methods[2931]. This deficiency,
which is not cured by the Becke approach to hybrid functionals, was
attributed to spurious self interaction[32] and missing derivative
discontinuities[33] two pervasive problems in density functional theory
(DFT)thatareintimatelyrelated.[34,35]
One way to mitigate the spurious self interaction and to retain a good
treatment of correlation is to deploy a rangeseparated hybrid functional[26
28, 34, 36] In this approach, the exchange term in the Kohn Sham energy
functional is split into longrange and shortrange terms, e.g., via r
-1
=
r
-1
cr (yr) +r
-1
crc (yr). The shortrange exchange is represented by a
local potential derived from a Semilocal functional. The longrange part is
treated via an explicit or exact exchange term. If one assumes that an
appropriate choice for is system independent, its value can be optimized
using a molecular training set for optimizing its value. Such semiempirical
approaches, typically with in the range of 0.30.5 a0
1
, were shown to
Election Bensity Functional Theoiy Page 2u6
Roi Baei

achieve impressive results for the ground state properties of some classes of
systems.[27, 37, 38] Furthermore, it was demonstrated on the benchmark
model of Dreuw and HeadGordon[39] the C2H4

C2F4 dimer at large


molecular distances that the rangeseparated hybrid corrects the principal
deficienciesofthechargetransferexcitationpredictionofTDDFT.[38,40]
One of the drawbacks of the semilocal DFT and Becke schemes is that the
correlation energy of Eqs. (10.1.3) (10.1.4) is much too nonlocal due to the
presence of the nonlocal exchange E
X
. This is unbalanced in view of the
presence of E
H
in the total energy. This is the root cause of longrange self
repulsion which exists in all these functionals. A way to improve the
situation is to include in the exchange functional a nonlocal longrange
exchange,whichleadstothefollowingformforthecorrelation[26,27,41]:
E
C
HB,d
= _
C
LP
(n(r), |vn(r)|)J
3
r
+_[
X
B
(n(r), |vn(r)|) -
X,y
(n(r), |vn(r)|) J
3
r
-(E
X
-E
X,y
)
(10.1.5)
ThefunctionalE
X,y
istheexchangeenergyofasystemofparticlesatdensityn
mutuallyinteractingthroughtherepulsivepotentialu
y
(r) =
crI(y)

whereyis
aparameter:
E
X,y
= -
1
2
|p(r, r
i
)|
2
u
y
(r)J
3
rJ
3
r'. (10.1.6)

X
-
X,y
is a semilocal density approximation to the complementary
interaction y
y
(r) =
1

-u
y
(r) =
crIc(y)

. The idea behind this appealing


approach is to eliminate in the correlation functional the long range
dependence on the density. y is treated as a universal parameter determined
by fitting the DFT predictions based on Eq. (10.1.5) to a large database of
Election Bensity Functional Theoiy Page 2u7
Roi Baei

experimental data. This forms the basis of several recent semiempirical


approaches all of which use a universal longrange parameter y determined
bysemiempiricalfittingtoknownexperimentalandsometimeshighquality
abinitio results [2628, 34, 3638]. These approaches have been quite
successfulforabroadvarietyofmolecules,sometimesbeyondthetypesused
inthefittingprocedures.
The approach of a universal y exchange functional with complementary
semilocal functionals cannot however claim universality. In some systems the
exactvalueofyiscriticalandnosemilocalDFTmethodcancorrectforit.We
show examples below.An alternative viewwas developed in ref. [28], which
provides a theory for the longrange parameter y based on the adiabatic
connection theorem, from which the following exact expression for the
correlationenergycanbededuced:
E
C
= (|
`
y
|) -(
S
|
`
y
|
S
) (10.1.7)
Where
`
y
=
1
2
y
y
(r
nm
)
n=m
. In ref. [28] arguments were given for the
existence of a parameter y for which Eq. (10.1.7) holds for practically all
densities. This equation describes the correlation energy as a difference
between potential energies of interaction in the interacting and in the non
interacting systems. Unlike the standard definition of correlation energy as
E
C
= (|I
`
|) -(
S
|I
`
|
S
) +(|u

|) -(
S
|u

|
S
) -E
X
, the there are no
explicit contributions of kinetic energy differences in Eq. (10.1.7). These are
absorbed by the interaction screening parameter y. In principle all quantities
in this equation, including y, are densitydependent. We stress that the
differencebetweenthisexacttheoryandtheusualexpositionofRSHsisthat
the latter do not derive (or even strive to derive) an exact expression for the
correlationenergy.Eq.(10.1.7)canbeappliedtotheHEGandtheexactvalue
Election Bensity Functional Theoiy Page 2u8
Roi Baei

ofy canbedeterminedasafunctionofdensity[38,42].Theresultisshownin
Figure X1anditisseenthatyisinfactstronglydensitydependent,certainly
not a universal constant. In particular, for molecular densities, where r
s
is
between 0.5 and 5 the longrange parameter y changes by an order of
magnitude.
When we add the exact exchange
energy to the exact correlation
energyofEq(10.1.7)weobtainfor
the exchangecorrelation the
followingexpression[28]:
E
XC
= (|
`
y
|)
-
1
2
_n(r)n(r
i
)y
y
(|r
-r
i
|
)J
3
rJ
3
r
i
+E
X,y
E
XC
y
+E
X,y

(10.1.8)
Thisexpressionisexact.Theright
hand equivalence defines a new functional, the complementary XC energy
functionalE
XC
y
.Itisthefocusforapproximationinourproposal.Becausethe
potential y
y
(r) is shortrange, we expect that a semilocal approximation,
depending on density and perhaps gradients, is of sufficient accuracy for
describingthecomplementaryXCenergy.
Except for the HEG, we have no practical way of calculating the expectation
valuesinEq.(10.1.8)soanapproximationmustbedevelopedfordetermining
the longrange parameter y and for determining the complementary XC
energy functional E
XC
y
. The procedure for determining the variable y will be

Figure X1:Theparameteryasafunctionof
the density paramter r
x
= a

[
3
4a
n
13
for
the fully Polarized and unpolarized
homogeneouselectrongas.
Election Bensity Functional Theoiy Page 2u9
Roi Baei

calledbelowlongrangeparametertuning.Wenowshowsomepreliminary
results which demonstrate the fact that one cannot rely on a universal long
rangeparametery.
B. The necessity of long-range parameter tuning
The requirement for a systemdependent parameter is not just a theory it
comesupinpracticalcalculations.Herewegivetwoexamplesofunpublished
results,oneinDFTandoneinTDDFT.
i. The symmetric radical cation
In DFT, a clear example of this issue is seen in the symmetric radical cation
systems of the type R
2
+
- R +R
+
, where R is any molecule or atom. Specific
examplesareR = B, Be and Ne.Itisknownthatforthesesystemsasemilocal
or even nonlocal (B3LYP) DFT gives qualitatively spurious potential surfaces
andthebigissueisselfrepulsion[1,43].Yet,evenwithastandardyfunctional
theresultsarenotsatisfactory.Takeforexamplethebonddissociationenergy
(BDE).Thiscanbeestimatedintwoways.Thefirstistheatomizationenergy.
Thesecondisthedepthofthemolecularwellrelativetotheasymptoteofthe
potential surface. From both numbers we subtract the zeropoint vibrational
energy to get the BDE. We can see in Table 1 that the BDE of R=He and Ne
based on atomization energy using a semiempirical y = u.S o
0
-1
(denoted
BNL functional[38]) gives BDE around 1.52 times the experimental value!
WhileintheasymptoticmethodthecalculatedBDEistoosmallby15%.Such
discrepancies cannot be waived by a better semilocal functional. They reflect
aninherentimbalancebetweenshortandlongrangeexchangeinflectedbya
wronglongrangeparametery.Indeed,bytuningthelongrangeparameterin
thefollowingwayweobtainavaluey
-
whichisspecificforeachsystem.The
Election Bensity Functional Theoiy Page 21u
Roi Baei

tuningprocedureisabinitiointhatitdoesnotuseanyexperimentaldata,
and results from the calculations themselves. It is describe in ref. [1] and is
equivalent to the
procedure depicted
in a more general
treatment in ref
[38].
The resulting
functional is
denoted BNL*. This
procedure chooses
y
-
in such a way as
to impose the
energy removal
theorem namely
that the energy to
remove an electron
from the interacting
system is equal exactly to -e
H0M0
(the HOMO energy in the noninteracting
system).[44] The parameter y
-
, as we compute it, obeys the equation
e
H0M0
y
-
= E
y
-
(N) -E
y
-
(N -1) where e
H0M0
y
is the HOMO energy and E
y
is
the calculated groundstate energy of the Nelectron system. This procedure
approximately imposes the energy removal theorem and defines BNL*. The
merit is that the HOMO and the SCF procedures are now wellbalanced.
Unlike the results of BNL, where y is not optimal, those of BNL* are very
satisfactoryforseveralobservableswhencomparedtoexperiment,asseenin
Table 1: (Taken from ref [1]) Data for R2
+
, R=H, He and Ne
calculatedintheccpVTZbasis[2,3].Allenergiesinkcal/mole.
r
eq
is the internuclear distance at which the F(r) is minimal.
m = _2
-1
F
ii
(r
eq
)

is the harmonic frequency and BDE is


F
atumx
-F(r
eq
) +
1
2
lm where F
atumx
is estimated either as the
sum of atomic energies (SUM) or by the asymptotic value of
the potential energy curve (ASS). u
R
is the calculated
polarizability of the atom R and u
e
= ltm
r-
1
2
F
i
(r)r
5
is the
polarizability as estimated from the form of the asymptotic
potential. Finally u
R
daug
is the polarizability of atom R
computedwithalargebasisset.
Property
R BLYP B3LYP HF BNL BNL* Exp.[4]
BDEby
( ) sum
atoms
E
H 66 65 60.9 60.9 60.9 61
He 82 75 43 74 59 55
Ne 75 60 2 59 34 32
BDEby
( ) ass
atoms
E
H NA NA 60.9 50 60.9 61
He NA NA 43 42 59 55
Ne NA NA 2 27 34 32
( )

eq
r
H 1.1 1.1 1.06 1.2 1.06 1.05
He 1.2 1.1 1.075 1.2 1.078 1.080
Ne 1.9 1.9 1.7 1.760 1.72 1.765
2
H 2.7 2.9 3.3 2.9 3.3 3.32
He 1.7 2.0 2.5 2.1 2.5 2.42
Ne 0.5 0.6 0.9 0.726 0.8 0.729
R eff

H NA NA 1 0.6 1 1
He NA NA 0.98 NA 0.98 1
Ne NA NA 1.01 NA 1.02 1
( )
3
0
daug
R
a a
H 5.3 5.6 4.51 5.8 4.51 4.50
He 1.6 1.5 1.34 1.8 1.41 1.38
Ne 3.1 2.9 2.4 3.2 2.70 2.66

Election Bensity Functional Theoiy Page 211


Roi Baei

Table 1. The comparison with regular functionals is convincing and the


improvement over BNL and HartreeFock is also large. Even the complete
basis set limit of the atomic polarizability as calculated by BNL* compares
accuratelywithexperiment.
ii. Aromatic donor -TCNE acceptor charge transfer excitation
The second example concerns chargetransfer excitations in TDDFT. The
results shown here are part of a manuscript recently submitted for
publication[45]. It is known that semilocal functionals cannot reasonably
describe such excitations[39], so a rangeseparated hybrid is a welcome cure.
We tested our approach on complexes formed by an aromatic donor
(Ar=benzene, toluene, oxylene and naphthalene) and the tetracyanoethylene
(TCNE) acceptor, for which optical absorption is available both in gas phase
and in solution.[46] All calculations were performed using QCHEM 3.1,[47]
modified to include the rangeseparated BNL functional,[38] using the cc
pVDZ basis set.[3] The internal structure of the molecules in the complex is
known to be littleperturbed by complex formation[30, 31, 48, 49] and the
equilibrium distance and relative orientation of the stacked donor and
acceptor determined from the conventional B3LYP[21] hybrid functional is
knowntocomparewellwithexperiment(whereavailable).[30,49]Therefore,
B3LYPoptimized geometry was used throughout. The TDB3LYP excitations
energies were much too low and the BNL much too high (see Table 1). In
molecular complexes, the lowest photon energy required to induce a CT
excitation, hCT,isgivenforasymptoticallylargedonoracceptordistancesby
theMullikenrule[50]
bv
C1
= IP() -EA(A) -1R (10.1.9)
Election Bensity Functional Theoiy Page 212
Roi Baei

Where IP() and EA(A) are the donor ionization potential and acceptor
electron affinity, respectively. The last term on the right hand side is the
Coulomb energy of attraction between the electronhole pair formed by the
charge transfer, where R is the intermolecular separation. For our TDDFT
calculationtoconformtotheMullikenrule,theionizationenergiescomputed
fromEq.(1)mustcorrespondtotheHOMOenergyoftheneutraldonor,but
alsototheHOMOenergyofacceptoranion.Thus,oneneedstogeneralizeEq.
(1)soastoyield,ascloselyaspossible,twolimits.Wethereforelookforythat
minimizesthefollowing[(y)function
[(y) = e
H0M0
y,
-[E
y
(N

-1) -E
y
(N

)
=
0
,A
-

(10.1.10)
Forcomplexeswherearangeparameterythatrenders[(y)verysmallcanbe
found (which is the case for all complexes we studied), we expect the range
separated hybrid to yield a quantitative description of CT excitations. An
exampleofthewaythistuningprocedureworksanditsabilitytoreplicatethe
MullikenlawforlargeseparationsRinsowninFigure X2

Figure X2: Tuning in the BenzeneTCNE complex. Left: J

of Eq.
Error! Reference source not found. as a function of . The optimal is 0.331 for which J is
very small. Right: The Mulliken rule, compared to TDDFT results obtained from the
optimal,asafunctionofinverseintermolecularseparation.
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
0.2 0.3 0.4 0.5
J
(

(
e
V
)
(a
0
1
)
5.2
5.4
5.6
5.8
6.0
6.2
6.4
0 0.01 0.02 0.03 0.04 0.05
h

C
T
(
e
V
)
1/R(a0)
TDBNL(g*)
Mullikenlaw
Election Bensity Functional Theoiy Page 21S
Roi Baei

Withtheasymptoticbehaviorenforced,weexpectproperbalancebetweenthe
semilocalandnonlocalexchangecomponents.InError!Referencesourcenot
found. we compare the calculated and experimental gasphase results for
various ArTCNE complexes. It is readily seen that the B3LYP results are
unacceptably low and predictive power is absent. Results of generalized
gradient(GGA)calculations(notshown)areevenlowerthantheB3LYPones.
With the rangeseparated BNL functional,[38] excitation energies determined
with an offtheshelf value of 0.5 a0
1
are much too high with respect to
experiment. But with * quantitative agreement is obtained to within 0.2eV.
For benzene and toluene, the theoretical oscillator strengths are also in good
agreement with experiment, but are too weak for xylene and naphthalene.
Thisislikelyabasissetissue,asoscillatorstrengthsaremuchmoresensitive
tothebasissetthantheexcitationenergies.

Election Bensity Functional Theoiy Page 214
Roi Baei

XI. TDDFT
A. Time-dependent Linear response theory
Thelinearresponseof
il
o
ot

k
(r, t) = [E

|p(t)] +:
1
(r, t)
k
(r, t)

k
(r, u) =
k
0
(r)
(10.2.1)
WheretheTDHamiltonian,intheadiabaticTDDFTapproximation,is:
E

|p] = I
`
+:(r) +:
H
|n(t)](r) +:
XC
y
|n(t)](r) +K

y
|p] (10.2.2)
AndtheTDdensitymatrixisdefinedintermsoftheTDorbitals:
p(r, r
i
, t) =
k
(r, t)
k
(r
i
, t)
-
N
k=1
(10.2.3)
TheinitialorbitalsareassumedthelowesteigenstatesofE

|p
0
]:
E

|p
0
]
k
0
(r) = e
k

k
0
(r) (10.2.4)
where:
p
0
(r, r
i
) =
k
0
(r)
k
0
(r
i
)
-
N
k=1
(10.2.5)
Weassumethattheperturbation:
1
(r, t)isweakandwewilltreatitasafirst
order quantity. Before we do this however, let us prepare the equation for
linearizationbydefining:

k
(r, t) = c
-c
k
tl
[
k
0
(r) +
k
1
(r, t) (10.2.6)
and:
Election Bensity Functional Theoiy Page 21S
Roi Baei

p(r, r
i
, t) = p
0
(r, r
i
) +p
1
(r, r
i
, t)
n(r, t) = n
0
(r) +n
1
(r, t)
(10.2.7)
Withthesedefinitionswehave:
il
o
ot
jc
-c
k
tl
[
k
0
(r) +
k
1
(r, t)[
= [E

|p
0
+p
1
(t)]
+:
1
(r, t)) jc
-c
k
tl
[
k
0
(r) +
k
1
(r, t)[
(10.2.8)
Fromthiswefind:
il
o
ot

k
1
(r, t) = [E

|p
0
+p
1
(t)] -e
k
+:
1
(r, t) [
k
0
(r) +
k
1
(r, t)
(10.2.9)
This equation is rigorously equivalent to the first equation. It is a non
homoegenouslinearequationfortheunknown
k
1
(t).
We now make the linearization step. We neglect all quantities beyond first
order.Asafirststepwehave:
il
o
ot

k
1
(r, t) = [E

1
+:
1
(r, t)
k
0
(r) +(E

|p
0
] -e
k
)
k
1
(r, t)
(10.2.10)
Here,thesymbol=meansequaltofirstorder.Weusethedefinition:
E

1
= E

|p
0
+p
1
(t)] -E

|p
0
] (10.2.11)
An efficient method to obtain linear response of a perturbation is as follows.
Select a small parameter e and solve the following time dependent equation
(usingastandardpropagator,suchasRungeKutta)[51]:
il
o
ot

k
1
(r, t) = _
E

|p
0
+ep
1
(t)] -e
k
e
+:
1
(r, t)_
k
0
(r)
+(E

|p
0
] -e
k
)
k
1
(r, t)
(10.2.12)
With:
Election Bensity Functional Theoiy Page 216
Roi Baei

p
1
(r, r
i
, t) =
k
0
(r)
k
1
(r
i
, t)
-
N
k=1
+
k
1
(r, t)
k
0
(r
i
)
N
k=1

n
1
(r, t) = 2Rc
k
0
(r)
k
1
(r, t)
N
k=1

The size of e must be so small that the orbital amplitudes are linear with e.
ThesolutioninvolvesrepeatedapplicationsoftheHamiltonian.Thiscouldbe
considerably more efficient than the algebraic method described below,
requiring the computation of all the eigenvectors of the Hamiltonian. Of
course, for small systems with a small basis the algebraic approach may be
extremelyeffectivethuswediscussitnow.
B. Algebraic Approach to Linear response
In order to derive an algebraic approach, we need expressions for all the
linearquantities.ForE

1
wefind:
E

1
= :
H
1
+:
XC
y1
+K
y1
(10.3.1)
And:
:
H
1
(r, t) = _
n
1
(r
i
, t)
|r -r
i
|
J
3
r
i

:
XC
y1
(r) = _
XC
y
(r, r')n
1
(r
i
, t)J
3
r
i

y1
(r) = -_p
1
(r, r
i
, t)u
y
(|r -r
i
|)(r
i
)J
3
r'
(10.3.2)
PluggingallthisintoEq.(10.2.10):
Election Bensity Functional Theoiy Page 217
Roi Baei

il
o
ot

k
1
(r, t) = ___
1
|r -r
i
|
+
XC
y
(r, r
i
)_ n
1
(r
i
, t)J
3
r'
+:
1
(r, t)_
k
0
(r) -_p
1
(r, r
i
, t)u
y
(|r -r
i
|)
k
0
(r
i
)J
3
r'
+(E

|p
0
] -e
k
)
k
1
(r, t)
(10.3.3)
Plugginginthe1
st
orderdefinitionsofn
1
andp
1
wefind:
il
o
ot

k
1
(r, t) = 2Rc _w
k]
(r, r
i
)
]
1
(r
i
, t)J
3
r'
N
]=1
+(E

|p
0
] -e
k
)
k
1
(r, t) +:
1
(r, t)
k
0
(r)
w
k]
(r, r
i
) _
1
|r -r
i
|

k
0
(r)
]
0
(r
i
) -u
y
(|r -r
i
|)
k
0
(r
i
)
]
0
(r)
+
XC
y
(r, r
i
)
k
0
(r)
]
0
(r
i
)_
(10.3.4)
We now take a crucial step to get rid of all the integrals and special
dependence by expanding in terms of the orthonormal complete
eigenfunction of the Hamiltonitan E|p
0
]. Because the evolution according to
the KohnSham equation preserves orthonormality. Thus, the response
functionsareorthogonaltoalloccupiedMOs.Thisallowsustowrite:

k
1
(r, t) = (c
ks
i
(t) +ic
ks
ii
(t))
s
0
(r)
M
s=N+1
(10.3.5)
Weusethefollowingindexnotations:
], k -occupicJ orbitols
q, s -unoccupicJ orbitols
p, n -oll orbitols
Thewoperatorcanbeturnedintoa4indexmatrix:
Election Bensity Functional Theoiy Page 218
Roi Baei

w
kp,]n
= w
k]
(r, r
i
)
p
0
(r)
n
0
(r
i
)J
3
rJ
3
r
i
= __
1
|r -r
i
|
+
XC
y
(r, r
i
)_
k
0
(r)
]
0
(r
i
)
-u
y
(|r -r
i
|)
k
0
(r
i
)
]
0
(r)_
p
0
(r)
n
0
(r
i
)J
3
rJ
3
r
i

- w
k]
(r, r
i
) = w
kp,]n

p
0
(r)
n
0
(r
i
)
pn

(10.3.6)
Usingtheseexpansionswefind:
il(c
ks
i
(t) +ic
ks
ii
(t))
s
0
(r)
s
= 2 c
]s
i
(t)w
kp,]s

p
0
(r)
s,],p
+(c
ks
i
(t) +ic
ks
ii
(t))
sk

s
0
(r)
s
+:
1
(r, t)
k
0
(r)
(10.3.7)
with:l
sk
= e
s
-e
k
.Nowmultiplyby
q
(r)andintegrateoverrandobtain
atimedependentalgebraicequation:
il[c
kq
i
(t) +ic
kq
ii
(t)
= 2w
kq,]s
c
]s
i
(t)
] s
+[c
kq
i
(t) +ic
kq
ii
(t) l
qk
+:
kq
1
(t)
(10.3.8)
Where:
:
qk
1
(t) = _
q
0
(r):
1
(r, t)
k
0
(r)J
3
r (10.3.9)
Wecanseparatetotworealequationsbytherealandimaginaryparts:
Election Bensity Functional Theoiy Page 219
Roi Baei

-lc
kq
ii
(t) = 2w
kq,]s
c
]s
i
(t)
] s
+c
kq
i
(t)l
qk
+:
kq
1
(t)
c
kq
i
(t) = c
kq
ii
(t)
qk

(10.3.10)
Thisequationcanbeusedtogetatimedependentresponse.Itwillbeuseful
fordescribingtheresponsetoacomplicatedweakpulse:
qk
1
(t).
C. Frequency-domain response
Insteadofviewingtheresponseintimewecanviewitinfrequency.Thiswill
allow us to address questions of response to sinusoidal fields, such as laser
absorption.WeFouriertransformthecoefficients:
J
kq
() = _ c
kq
(t)c
ot
Jt

0
(10.3.11)
Notethatc
kq
(u) = uandso:-iJ
kq
(v) = ] c
kq
(t)c
ot
Jt

0
.
NotefurtherthatJ
i
andJ
ii
arenotrealanymore.Infact,becausecsarereal
wenowhave:J
kq
()
-
= J
kq
(-).
ApplyingaFouriertransformtothetimedependentequationsgives:
ilJ
kq
ii
() = 2w
kq,]s
J
]s
i
()
]s
+J
kq
i
()l
qk
+:
kq
1
()
-iJ
kq
i
() = J
kq
ii
()
qk

(10.3.12)
Combining the equations we rid ourselves of J
i
and obtain an equation for
J
ii
:
|2l
-1

qk
w
kq,]s
+(
qk
2
-
2
)o
(qk)(s])
]J
kq
ii
()
]s
= il
-1
:
kq
1
(),
(10.3.13)
Thenumberofthesecoupledalgebraicequationsisinfinite.However,inany
practicalapplicationweuseafinite number ofstatesand sowehaveafinite
numberofequations.IfMisthetotalnumberofstatesusedwethenforany
Election Bensity Functional Theoiy Page 22u
Roi Baei

given perturbation one can get an approximate response by solving one of


these two equations for J
]s
i
(). This involves solving a linear NH NH setof
equations and obtain the response to the perturbation. We can write it in
matrixform:
(A -
2
I)J
ii
() = il
-1
:
1
(). (10.3.14)
With:
A
qk]s
= 2l
-1

qk
w
kq,]s
+
qk
2
o
(qk)(s])
(10.3.15)
Wenowprovethataslongas
qk
= uforallpossibleqandkthematrixAhas
only real eigenvalues. In fact,
qk
is always positive unless there is a
degeneracyattheHOMOandthelevelisnotfull.Theeigenvalueequationis:
(2l
-1

qk
w
kq,]s
+
qk
2
o
(qk)(s])
)y
s]
=
2
y
qk
(10.3.16)
where y
s]
are the eigenvector components. Under the assumption we can
bringthisequationtothefollowingequivalentform:
(2l
-1
w
kq,]s

qk

s]
+
qk
2
o
(qk)(s])
)x
s]
]s
=
2
x
qk

(10.3.17)
Where x
qk
=

qk

o
qk
. These new equations involve the eigenvalues of a
symmetricmatrix.Thus,theeigenvalues
2
areallreal.Eq.(10.3.17)iscalled
theCasidaequations.
D. Excitation energies from the algebraic
treatment
The algebraic equation for linear response derived in Eq. (10.3.14) allows
calculation of a specific response problem. One incredible feature of this
equationisthatthematrixAontheleftisindependentoftheperturbation:
1
.
We can thus analyze A to get general response properties of the system,
Election Bensity Functional Theoiy Page 221
Roi Baei

without reference to any specific perturbation. We can define a response


matrix:
R() = il
-1
(A -
2
I)
-1
. (10.3.18)
WhereAisdefinedin(10.3.15)Wecandefinerightandlefteigenvectors:
Ay
u
=
u
2
y
u
, A
1
z
u
=
u
2
z
u
(10.3.19)
ItiseasytoseefromEq.(10.3.16)that:

qk
(z
u
)
qk
= (y
u
)
qk
(10.3.20)
Wechoosethenormalizationssothat:
z
u
1
y
[
= o
u[
= (z
u
)
]s
(y
[
)
]s
]s
=
1

s]
(y
u
)
]s
(y
[
)
]s
]s

(10.3.21)
Collecting the vectors ina a matrix and the eigenvalues in a diagonal matrix

2
wecanwrite:
A = y
1

2
z (10.3.22)
And:
R() = y
1
il
-1

(
2
-
2
I)
z
(10.3.23)
Wecannowtakenearaneigenvalue
u
.Thishasaninfiniteresponse.Thus
wecanthenwrite:
R() = y
u
1
il
-1

(
u
2
-
2
)
z
u
-
u
(10.3.24)
A general technique to study the resonance is to insert an infinitesimal
imaginary part which we eventually take to zero: =
u
+ip. Then, after a
few algebraic manipulations (R() = y
u
1
o
(
o
+o)(
o
-o)
z
u
- R(
u
+ip) =
y
u
1
(
o
+q)
(2
o
+q)(-q)
z
u
):
Election Bensity Functional Theoiy Page 222
Roi Baei

R(
u
+ip) = l
-1
y
u
1

u
(2
u
+ip)
-p(4
u
2
+p
2
)
z
u
(10.3.25)
Thus:
lim
q-
+|R(
u
+ip)] =
lim
q-
,|R(
u
+ip)] = y
u
1
-1
4l
u
z
u

(10.3.26)
Theimaginarypartoftheresponseisfiniteatresonances.

Election Bensity Functional Theoiy Page 22S
Roi Baei

References
1
E.LivshitsandR.Baer,J.Phys.Chem.A112,12789(2008).
2
D.E.WoonandT.H.Dunning,J.Chem.Phys.100,2975(1994);R.M.
DreizlerandE.K.U.Gross,DensityFunctionalTheory:AnApproachto
theQuantumManyBodyProblem(Springer,Berlin,1990);R.G.Parrand
W.Yang,DensityFunctionalTheoryofAtomsandMolecules(Oxford
UniversityPress,Oxford,1989);W.KochandM.C.Holthausen,A
ChemistsGuidetoDensityFunctionalTheory(Wiley,Heidelberg,2001);
C.Fiolhais,F.Nogueira,andM.Marques,Aprimerindensityfunctional
theory(Springer,Berlin;NewYork,2003).
3
T.H.Dunning,J.Chem.Phys.90,1007(1989).
4
K.P.HuberandG.Herzberg,MolecularSpectraandMolecularStructure,
Vol4:ConstantsofDiatomicMolecules(PrentoceHall,NewYork,1979);
A.Dalgarno,Adv.Phys.11,281(1962);F.FurcheandR.Ahlrichs,J.
Chem.Phys.117,7433(2002);F.Neese,JournalofBiologicalInorganic
Chemistry11,702(2006);M.Marques,A.Rubio,C.A.Ullrich,etal.,in
Lecturenotesinphysics,706(Springer,Berlin,2006).
5
H.Goldstein,C.P.Poole,andJ.L.Safko,Classicalmechanics(Addison
Wesley,SanFrancisco,2002).
6
J.C.Slater,J.Chem.Phys.1,687(1933).
7
B.N.FinkelsteinandG.E.Horowitz,Z.Phys.48,118(1928).
8
I.N.Levine,Quantumchemistry(PrenticeHall,UpperSaddleRiver,
N.J.,2000);F.Rioux,TheChemicalEducator2,1(1997).
9
R.Peierls,Phys.Rev54,918(1938).
10
R.GranotandR.Baer,J.Chem.Phys.128,184111(2008).
11
E.Teller,Rev.Mod.Phys.34,627(1962).
12
U.SalznerandR.Baer,J.Chem.Phys.inpress(2010).
13
J.Janak,Phys.Rev.B18,7165(1978).
14
J.C.SlaterandJ.H.Wood,Int.J.Quant.Chem.S4,3(1971).
15
R.BaerandM.HeadGordon,Phys.Rev.Lett.79,3962(1997).
16
D.C.LangrethandJ.P.Perdew,Sol.Stat.Comm.17,1425(1975).
17
W.KohnandL.J.Sham,Phys.Rev140,A1133(1965).
18
M.GellmannandK.A.Brueckner,Phys.Rev106,364(1957).
19
E.Wigner,Phys.Rev46,1002(1934).
20
O.GunnarssonandB.I.Lundqvist,Phys.Rev.B13,4274(1976).
21
A.D.Becke,J.Chem.Phys.98,1372(1993).
22
A.D.Becke,J.Chem.Phys.98,5648(1993).
23
A.D.Becke,Phys.Rev.A38,3098(1988).
Election Bensity Functional Theoiy Page 224
Roi Baei

24
J.P.Perdew,inElectronicstructureofsolids91:proceedingsofthe75.WE
HeraeusSeminarand21stAnnualInternationalSymposiumonElectronic
StructureofSolidsheldinGaussig(Germany),March1115,1991,edited
byP.ZiescheandH.Eschrig(AkademieVerlag,Berlin,1991),Vol.17,
p.11;J.P.Perdew,J.A.Chevary,S.H.Vosko,etal.,Phys.Rev.B48,
4978(1993);J.P.Perdew,J.A.Chevary,S.H.Vosko,etal.,Phys.Rev.B
46,6671(1992).
25
P.HohenbergandW.Kohn,Phys.Rev136,B864(1964);S.H.Vosko
andL.Wilk,J.Phys.B16,3687(1983);J.P.PerdewandW.Yue,Phys.
Rev.B33,8800(1986);C.T.Lee,W.T.Yang,andR.G.Parr,Phys.Rev.B
37,785(1988);J.P.PerdewandY.Wang,Phys.Rev.B45,13244(1992);J.
P.Perdew,K.Burke,andM.Ernzerhof,Phys.Rev.Lett.77,3865(1996);
J.P.Perdew,S.Kurth,A.Zupan,etal.,Phys.Rev.Lett.82,2544(1999);J.
M.Tao,J.P.Perdew,V.N.Staroverov,etal.,Phys.Rev.Lett.91,146401
(2003).
26
T.Leininger,H.Stoll,H.J.Werner,etal.,Chem.Phys.Lett.275,151
(1997).
27
H.Iikura,T.Tsuneda,T.Yanai,etal.,J.Chem.Phys.115,3540(2001).
28
R.BaerandD.Neuhauser,Phys.Rev.Lett.94,043002(2005).
29
D.J.Tozer,R.D.Amos,N.C.Handy,etal.,Mol.Phys.97,859(1999);J.
Fabian,Th.Chem.Acc.106,199(2001).
30
M.S.Liao,Y.Lu,V.D.Parker,etal.,J.Phys.Chem.A107,8939(2003).
31
M.S.Liao,Y.Lu,andS.Scheiner,J.Comp.Chem.24,623(2003).
32
A.DreuwandM.HeadGordon,J.Am.Chem.Soc.126,4007(2004).
33
D.J.Tozer,J.Chem.Phys.119,12697(2003).
34
S.KuemmelandL.Kronik,Rev.Mod.Phys.80,3(2008).
35
A.J.Cohen,P.MoriSanchez,andW.T.Yang,Science321,792(2008);J.
P.Perdew,Adv.Chem.Phys.21,113(1990).
36
T.Yanai,D.P.Tew,andN.C.Handy,Chem.Phys.Lett.393,51(2004).
37
M.J.G.Peach,T.Helgaker,P.Salek,etal.,Phys.Chem.Chem.Phys.8,
558(2005);O.A.VydrovandG.E.Scuseria,J.Chem.Phys.125,234109
(2006);Y.ZhaoandD.G.Truhlar,J.Phys.Chem.A110,13126(2006);J.
D.ChaiandM.HeadGordon,J.Chem.Phys.128,084106(2008).
38
E.LivshitsandR.Baer,Phys.Chem.Chem.Phys.9,2932(2007).
39
A.Dreuw,J.L.Weisman,andM.HeadGordon,J.Chem.Phys.119,
2943(2003).
40
Y.Tawada,T.Tsuneda,S.Yanagisawa,etal.,J.Chem.Phys.120,8425
(2004).
41
A.Savin,inRecentAdvancesinDensityFunctionalMethodsPartI,edited
byD.P.Chong(WorldScientific,Singapore,1995),p.129.
Election Bensity Functional Theoiy Page 22S
Roi Baei

42
R.Baer,E.Livshits,andD.Neuhauser,Chem.Phys.329,266(2006).
43
R.Merkle,A.Savin,andH.Preuss,J.Chem.Phys.97,9216(1992);T.
BallyandG.N.Sastry,J.Phys.Chem.A101,7923(1997);M.Lundberg
andP.E.M.Siegbahn,J.Chem.Phys.122(2005);J.P.Perdew,A.
Ruzsinszky,G.I.Csonka,etal.,Phys.Rev.A76,040501(2007);O.A.
Vydrov,G.E.Scuseria,andJ.P.Perdew,J.Chem.Phys.126,154109
(2007);A.Ruzsinszky,J.P.Perdew,G.I.Csonka,etal.,J.Chem.Phys.
126,104102(2007);J.Grafenstein,E.Kraka,andD.Cremer,Phys.
Chem.Chem.Phys.6,1096(2004).
44
J.P.Perdew,R.G.Parr,M.Levy,etal.,Phys.Rev.Lett.49,1691(1982);
C.O.AlmbladhandU.vonBarth,Phys.Rev.B31,3231(1985).
45
T.Stein,R.Baer,andL.Kronik,submitted(2008).
46
I.Hanazaki,J.Phys.Chem.76,1982(1972).
47
Y.Shao,L.F.Molnar,Y.Jung,etal.,Phys.Chem.Chem.Phys.8,3172
(2006).
48
B.Uno,N.Okumura,andK.Seto,J.Phys.Chem.A104,3064(2000).
49
K.E.WiseandR.A.Wheeler,J.Phys.Chem.A103,8279(1999).
50
R.S.Mulliken,J.Am.Chem.Soc.72,600(1950).
51
R.BaerandD.Neuhauser,J.Chem.Phys.121,9803(2004);D.
NeuhauserandR.Baer,J.Chem.Phys.123,204105(2005).