Ab Initio Density Functional Theory for Open-Shell Systems, Excited States and Response Properties

Material Information

Ab Initio Density Functional Theory for Open-Shell Systems, Excited States and Response Properties
Bokhan, Denis
Place of Publication:
[Gainesville, Fla.]
University of Florida
Publication Date:
Physical Description:
1 online resource (135 p.)

Thesis/Dissertation Information

Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Committee Chair:
Bartlett, Rodney J.
Committee Members:
Monkhorst, Hendrik J.
Hirata, So
Ohrn, Nils Y.
Hershfield, Selman P.
Graduation Date:


Subjects / Keywords:
Approximation ( jstor )
Atoms ( jstor )
Density functional theory ( jstor )
Diagrams ( jstor )
Electrons ( jstor )
Energy ( jstor )
Flux density ( jstor )
Orbitals ( jstor )
Perturbation theory ( jstor )
Wave functions ( jstor )
Chemistry -- Dissertations, Academic -- UF
density, effective, excited, functional, hyperpolarizabilities, open, optimized, potential, properties, response, shell, states, system, theory
Electronic Thesis or Dissertation
born-digital ( sobekcm )
Chemistry thesis, Ph.D.


Ab initio density functional theory (DFT) based on the optimized effective potential (OEP) method is a new approach to study the electronic structure of atomic, molecular and^M solid state systems. It contains elements of both wave function and density functional theories and is free from limitations of conventional DFT because of using orbital-dependent functionals derived from systematic approximations of the wave function theory. Ab initio DFT methods with exchange-correlation functionals based on many-body perturbation theory (MBPT) have been derived and implemented recently. The exchange-correlation potentials derived from MBPT have a complicated structure and their derivation in higher-order of MBPT by the use of the chain-rule for functional differentiation requires significant effort. To facilitate such derivations, I developed a special diagrammatic formalism for taking functional derivatives. An alternative way to construct OEP MBPT exchange-correlation potentials is to use the density condition. It makes it possible to obtain potentials for different partitionings of the full molecular Hamiltonian with MBPT. Using the diagrammatic formalism developed for taking functional derivatives, we show an order-by-order equivalence between the functional derivative and the density condition approaches to OEP MBPT for the case of the Kohn-Sham partitioning of the molecular Hamiltonian. For any other partitionings, different exchange-correlation potentials are produced by the functional derivative and density condition approaches. The time-dependent extension of OEP in the adiabatic approximation with exchange-only potentials and kernels was recently implemented and applied to some molecular systems. The corresponding excitation energies and polarizabilities are in good agreement with time-dependent Hartree-Fock results. However, such results cannot be used for comparison with experimental values due to lack of a description of electron correlation effects. In order to account for correlation effects, a correlation kernel corresponding to the MBPT(2) potentials has been derived. Its structure and properties are described in detail. The first numerical results for excitation energies with pure ab initio potentials and kernels are presented. The recently implemented OEP MBPT(2) method has been extended to the case of spin-polarized open-shell systems. The total energies obtained for several open-shell systems are very close to the corresponding values obtained with the highly-correlated coupled cluster singles and doubles with perturbative triples (CCSD(T)) method. Comparison with results obtained with the OEP MBPT(2) exchange-correlation potentials and some density functionals has shown a qualitatively incorrect shape for some widely used exchange-correlation potentials. Higher-order response properties, such as hyperpolarizabilities, are described very poorly with conventional functionals. The typical relative errors described in the literature are about 100 percent. The reason for the poor description is an incorrect long-range asymptotic behavior and an incomplete cancellation of the coulombic self-interaction. The OEP method is free from those drawbacks and calculated values of static hyperpolarizabilities with the exchange-only potential are close to those derived from Hartree-Fock theory. The second kernel required for calculations of hyperpolarizabilities within the DFT framework has been derived for OEP potentials by using the developed diagrammatic technique. The structure and properties of the second OEP kernel are discussed. ( en )
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis (Ph.D.)--University of Florida, 2007.
Adviser: Bartlett, Rodney J.
Statement of Responsibility:
by Denis Bokhan.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright Bokhan, Denis. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Resource Identifier:
670421880 ( OCLC )
LD1780 2007 ( lcc )


This item has the following downloads:










































































































































Full Text

Equations (1-55) and (1-56) are the basis of linear response theory. Since equation (1

55) is not linear with respect to pi the solution should be obtained by some iterative


However, for all practical purposes a direct use of equation (1-55) is not convenient.

To get a more convenient form, the following matrix element should be considered

< plpi(w)l\pi >=< plvl(wi)lpi > + dr~cp(r)( I dr' r) >c~(r) +

+ J dr r)(J dr/'f,,(r, r', U)pIWr', U)) r) (1-59)

On the other hand, < o, lpi(r, w) | > can be expressed in terms of a response function
P,(r, ) = Pa( api + ai(W)ppi (1-60)
i a

Pia() =, i( ( =7 (1 61)
L + (Ea ~- i) (a C- i)
Substituting equation (1-61) into equation (1-59) we have
Pia(W) ( + Fa i <) =< aVI(W) \i > + ([< ab > +
j b
< f f (w) PaPb b + [< OiObjOa >j >+ < I fxc(w) >]Pbj)
-Pai(W)(Fa )i +i b ([< OiPbOa j >+
j b
< I -| fGc(w) \PaPj >]Pjb + [< Pab iPj > < fc(') >]Pbj) (1-62)

Introducing the notations

Aai,bj = 6ij6ab(Ea Ei)+ < 0aPb > + < f ;.fc(,) PalPb >

Bai,bj < K 1ipObjpaOj > + < f( c-l) (cPapj > (1-63)

When functional derivatives are taken from the diagram (VC4), diagrams (VC4-1)-(VC4-

14) (3-23) will appear.

2 i = b2

k d

c a
lc iJ c a c

(VC4-1) (VC4-2) (VC4-3) (VC4-4) (VC4-5) (VC4-6) (VC4-7)

_Ljd dTTT dF^T a ^
S 4-8 a a a5
=C 1 1C4-9 CC-
(VC4-8) (VC4-9) (VC4-10) (VC4-11)

(VC4-12) (VC4-13)

Diagram (VC5) produces diagrams (VC5-1)-(VC5-16) (3 24) after differentiation.

c c d
1 d c c ib 7ib c b

bj c b 1 ab a
j b 2i b 2 b k
k (k k 2
(VC5-1) (VC5-2) (VC5-3) (VC5-4) (VC5-5) (VC5-6) (VC5-7) (VC5-8)

S c 2 C 1 2 d 2

S bj b jk ab jkab ja b a b

(V 5-9) (VC5-10) (VC5-) (VC5-12) (VC5-13) (VC5-14)
(VC5-9) (VC5-10) (VC5-11) (VC5-12) (VC5-13) (VC5-14)


(VC5-16) (3-24)

Differentiation of diagram (VC6) generate diagrams (VC6-1)-(VC6-16) (3-25),

a b 2c bca ba ba ( b

(VC6-10) (VC6-1 1) (VC6-12) (VC6-13) (VC6-14)

bi\\ 2 k

aic b b

(VC6-15) (VC6-16) (3 25)

k b

a Fb



a Mbh







construction of approximate functionals is the Local Density Approximation (LDA), where

the energy is given through an integral of a local functional of the density. The second

level is Generalized Gradient Approximation (GGA) functionals, which improve upon

LDA by the addition of the dependence on the gradient of the density. The third level,

known as meta-GGA, contains also a dependence upon the gradients of the Kohn-Sham

orbitals. For the GGA and the meta-GGA there is the possibility to choose a form of a

functional. A number of different forms have been ir.-. -1. I A standard way of choosing

the form of GGA and meta-GGA functionals is to satisfy a set of theorems, which are

known to be satisfied by the exact functional. Some functionals contain a set of parame-

ters to reproduce experimental data. Such functionals are called semiempirical. There are

also functionals which do not contain parameters but satisfy an extended set of theorems

(non-empirical functionals).

Kohn-Sham DFT with conventional functionals can produce very accurate results for

some properties. Values of the total energy, obtained with GGA are usually comparable

with some ab initio correlated results, like MBPT(2) or CCSD. This is a great advantage

of conventional DFT. However, the accuracy of GGA functionals is restricted by its

analytical form. Generalized gradient approximation exchange functionals cannot provide

the exact elimination of coulombic self-interaction of the Hartree energy. Since the

exchange energy is much larger than the correlation energy, an incomplete elimination of

self-interaction can significantly reduce the accuracy of the total energy. Particularly, the

impossibility to describe weakly-interacting systems, bounded mostly by dispersion forces,

arises from an incomplete elimination of self-interaction and the wrong behavior of the

exchange-correlation functionals with the separation of the system onto several fragments.

Although the exchange-correlation energies are quite accurate with GGA functionals,

the corresponding potentials are not nearly accurate, especially in the inter-shell and

.I-i~,nIl ''I ic regions. As a consequence, the accuracy of the density is much lower than the

accuracy for the energy. Furthermore, the qualitatively incorrect potentials reduce the

Tr6) (r6)

S8(r7 )5 (r7 )- J(
(FX7) 7(FX8) (6-18)

Functional derivatives from the set (V1)-(V6) can be presented by two types of diagrams.

The first type contains the exchange kernel, diagrams (F13)-(F18) of (6-19)

(r6) 1(r7) 6(r6) (r7) a6(r6) (r6)
S(r7 ) (r6) 1 (r7) ( ) f(r) f(r3)
f(r) (r) (f(r3) b fx(r3) ) b (r b) (r,)
fxrr)x3rx) xr7
(F13) (F14) (F15) (F16) (F17) (F18) (6-19)

The second type contains the exchange potential, giving diagrams (VX1)-(VX8) of (6-20)

/ r4\ r 6 r ) 6r4 r5\ 6 6r4\ r5\ r6) (r\ rT \ 6 )
(r4\ r5\ r6 )a (r4\ r5\ r6 ) (r4\ r \5 r6) 6(r4\5 \ r5 6)
(r4 r5 r6 4 r5 r6 6r5(r4 r \ 6 ) 64r\ r )r6
(VX1) (VX2) (VX3) (VX4)
/ 6(r rs rr ) ( r4 r ) 6(r4\ r r6) (r4\ r\ r6 )

Vx ax x x
6(r4\ r5\ r6) a6r\ ( 5 r6) lrrr5\ r6 6 ) ar 64r 54 r 6 )
6(r4\ rs*5 x J ) b'4 s r54\ 17\ r6 ) V ^(4 rs\ r6 )
4 6456 456
(VX5) (VX6) (VX7) (VX8) (6-20)

The notation r4\r5\r6 means that for each particular diagram all six permutations of r4, r5

and r6 should be taken. The functional derivatives from diagrams (FX1)-(FX8) lead to the

set (SX1)-(SX14) of (6-21) and (6-22).

8/ 4 \ r \ r6) 1 (r4\ 5 6) \A(r4\ r\ r6 ) 6 (r \ r )
S5 6 4 5 6 ) 5r4 6 ( \7 (r5\ r \ r6
S8(r 54\ r5\ r6 4 5 6 )r (\ r55\ r6
k c a b 61r4 175 r
8(r4\ r5 6 r ) 8(r4\ r5\ r6 ra
(SX1) (SX2) (SX3) (SX4) (SX5)

A^ ^ ^ ) ar4 6\ r5\ r6
(r4\ r5\ r6 (r\ r5\ r6
k(r\r5\ r6 )\(r\ r5\ r6) -6(r,\ r5\ r6 ) ( rs r 6
6 )
S (r4 5 r 6 a 6a b k
b k 64) 175 6 k4 (5 ) 4 5 6(
(SX6) (SX7) (SX8) (SX9) (6-21)

After taking the functional derivative of diagram (VC14), we obtain diagrams (VC14-1)-

(VC14-15) (3-33)

b i Fb 6Z -a b a
b a j a ^.
(VC14-1) (VC14-2) (VC14-3)

(F F(F
a ba b k b

(VC14-7) (VC14-8) (VC14-9)

- 82 b a i j b

(VC14-4) (VC14-5)

a iai

C14-10) (VC14-
(VC14-10) (VC14-11)

(VC14-13) (VC14-14) (VC14-15)

The last set of diagrams (VC15-1)-(VC15-15) can be produced by

diagram (VC15) (3-34)

162 (1 62" kF F 53 C F k & F
(F k F F t k

1 1 a a & 1
1i a 1 ai %a61 -6, a
(VC15-1) (VC15-2) ( 15-3) (VC15-4) (VC1
(VC15-1) (VC15-2) (VC15-3) (VC15-4) (VC15-5) (

b k



F 1

b 'k c
(VC15-8) (VC15-9)
2 6 61


differentiation of

P a ,



F)8 b, F) bk UF k

b( (VC15-14)Li (C515 b

(VC15-12) (VC15-13) (VC15-14) (VC15-15)
(VC15-12) (VC15 13) (VC15 14) (VC15 15)

Together with diagrams (V1)-(V6), all diagrams presented in this section form a set, which

is necessary to construct the correlation kernel. The interpretation of all diagrams is given

in Appendix A.





------- OEP-semi
...... MBPT(2)
------- CCSD

0.5 1.0 1.5 2.0


Figure 4-4. OH potential energy curve.



-74.7 -










-7C C

where Nom' is the numerator, with interchanged indexes k and 1.These expressions can be

represented by the following two diagrams (2-46)

1 1

m* m

<+ <(2-46)

Namely these two diagrams will appear after taking functional derivatives using dia-

grammatic rules for taking functional derivatives. Proof for the case where functional

derivatives are taken with respect to unoccupied orbitals is the complete analog of this

one. Hence, we can formulate a second statement.

Statement 2: When functional derivatives from some vertex of a diagram are

taken with respect to occupied (unoccupied) orbitals, and a line which corresponds to

occupied (unoccupied) orbital inserted, together with diagrams which arise from taking

functional derivatives from denominators; we ahv-- have diagrammatic expressions, which

corresponds to part of EY )((k)+( C-k))s.

This statement is a direct corollary of the above proved statement about "bracket-

type" denominators and statement 1. Since we have correspondence in the second and

n-th order,using the method of mathematical induction, it is possible to prove, that the

correspondence taks place in all orders (finite or infinite).

2.3.4 Interconnection in Infinite Order

It should be noted that summation of perturbation corrections to the energy and the

wave functions up to infinite order provide correspondence with the full CI:

E n E) E(FCI)

n O (n) (FCI) (247)

first member of the right-hand side corresponds to 60(3), while the second corresponds to

56(1)+Q (2).

To finally establish the interconnection between the two approaches, we need to

show that it is possible to transform diagrams which contains "bracket-type denominators

to standard diagrams. In a previous subsection this procedure was described for the

case where one horizontal line crosses the contour in the diagram, which corresponds

to the energy in second order MBPT. We need to prove, that energy diagrams, which

contain contours crossed by m lines, after taking the functional derivative which respect to

unoccupied (occupied) orbitals and inserting lines of unoccupied (occupied) orbitals; can

be transformed to linear combination of m diagrams which do not contain "bracket-type"


Consider the following diagram (2-41):

< \ 1

I m

Numbers on lines show the conditional number of lines which are entered for simplicity of

further manipulations. The desirable sum of diagrams has the form (2-42)

2 2 1
k k

m m / / m
S+ + ......... + L k (2-42)

Together with diagrams which correspond to functional derivatives from denominators,

this set of diagrams forms products like Q(k)+S6(1), where k+l equals the order of MBPT.

In the same way diagram (VC12) generates diagrams (VC12-1)-(VC12-16) of set (3-31)


b i

2 j

2 6

b i

S b

1 b



F 1 F

(VC12-7) (VC12-8)

aj 6


U ij2

i a a

l J
,J a 8 CC

(VC12-10) (VC12-11) (VC12-12) (VC12-13)
6 b b

1 Ob
Fb Ca (V7a
(VC12-14) (VC12-15) (VC12-16)

Differentiation of diagram (VC13) gives us diagrams (VC13-1)-(VC13-15) (3-32)

1 2 1 2
b k b k j c j c
82 82 8
j j b b
1 1 a a
a a i a 13

(VC13-1) (VC13-2) (VC13-3) (VC13-4)

b l
2 k

a F

j 6
2 C b


(Cb -

a F i F

(VC13-5) (VC13-6)

0 V1 0i 2 2
b j k i C
b F F
c k
i 82 a 62
a b b
F F 61

(VC13-7) (VC13-8) (VC13-9) (VC13-10) (VC13-11) (VC13-12)

2b1 a -
S61Z 62 b J

b a i b
a(VC13-13) (Vc k a(VC13-15)
082 Nk c
(VC13-13) (VC13-14) (VC13-15)



j 6 2 2



Figure page

4-1 Exchange and correlation potentials of Li atom (radial part). A) Exchange po-
tential. B) Correlation potential ............... ... 73

4-2 Exchange and correlation potentials of 02 molecule across the molecular axis.
A) Exchange potential. B) Correlation potential .... . ... 73

4-3 LiH potential energy curve. ............... ......... 77

4-4 OH potential energy curve ............... ......... .. 78

4-5 HF potential energy curve. ............... .......... 79

5-1 Exchange potentials of Ne atom, obtained in different basis sets . ... 88

5-2 A 1II charge-transfered excited state of He ... Be ... . .... 96

5-3 LUMO-HOMO orbital energy difference ................ . 96

[64] S. J. A. van Gisbergen, P. R. T. Schipper, O. V. Gritsenko, E. J. Baerends, B.
C'!l 1,p1. ;iw., and B. Kirtman, Phys. Rev. Lett. 83 697 (1999).

[65] F. A. Bulat, A. Toro-Labbe, B. C'!i ip ,1. i B. Kirtman, W. Yang, J. C'!. in Phys.
123 014319 (2005).

[66] M. Kamiya, H. Sekino, T. Tsuneda and K. Hirao, J. C'!. in Phys 122 234111 (2005).

[67] R. J. Bartlett, D. M. Silver, Phys. Rev. A 10 1927 (1974).

[68] H. P. Kelly, Adv. C'., in Phys 12 314 (1963).

[69] R. J. Bartlett, H. Sekino, ACS Symposium Series 628: Nonlinear Optical Materials
23 57 (1996).

[70] A. M. Lee and S. M. Colwell, J. Chem. Phys. 101 9704 (1994).

[71] S. P. Karna and M. Dupuis, J. Comp. ('!I, ii 12 487 (1991).

[72] H. Sekino and R. J. Bartlett, J. ('!,. ii Phys 85 976 (1986).

[73] H. Sekino and R. J. Bartlett, J. ('!,. ii Phys 98 3022 (1993).

[74] D. Bokhan, R. J. Bartlett, J. ('C!. ii Phys. (submitted).

The condition (6-25) might be used as a hint for developing new density functionals, as

none of the existing functionals exhibit such behavior.

6.2 Numerical Results

As a numerical test the static hyperpolarizabilities of LiH, H20, CO and CN are

calculated. In all cases uncontracted ROOS-ADZP basis sets[27] are used and equlibrium

geometries are taken from ref[36]. The results calculated with different methods are

presented in Table 6-1. In the first four columns are static hyperpolarizabilities obtained

Table 6-1. Hyperpolarizabilities of several molecules (in a. u.)
3 312.130 314.910 312.126 621.093 691.406
f3. 201.150 207.868 201.641 404.297 204.171
,,, -31.016 -24.864 -29.455 -66.406 27.343
0.. -3.073 -4.660 -3.118 -13.203 5.859
0.. -1.299 0.211 -3.906 -1.953 164.02
A,, -17.651 -15.646 -18.121 -62.500 68.359
03. -6.713 -6.239 -7.812 -13.672 -5.859
j3. -0.497 -0.604 -1.215 -6.138 -3.870
3yy, -10.866 -10.32 -11.847 -23.437 -7.645

with Hartre-Fock non-local exchange and with the two local exchange operators: OEP-HF,

which means equation (6-5) without the second kernel term, and OEPx, which means

all terms in equation (6-5). The hyperpolarizability values calculated with the OEPx

method are in good agreement with the corresponding Hartree-Fock values, and generally

no better. This good agreement can be explained by the fact that OEPx method is free

from the self-interaction error, has the correct long-range ., -ii1! i .I ic behavior, and the

correct HOMO values. As hyperpolarizabilities are response properties of third order their

calculation within the DFT method requires very precise exchange-correlation potentials.

Only potentials which simultaneously satisfy all theorems and conditions pertaining

to the DFT exchange potentials can be expected to reproduce even the Hartree-Fock

Using the effective operator approach it is possible to define VC7) in all possible orders.

This procedure for an infinite sum of orders gives Vxc. Inserting this potential into the

Kohn-Sham equations, we can find p("),which corresponds to the one-particle density in

the full CI method.The same density could be used in a method like Zhao-Morrison-Parr

(ZMP)[19] to extract the corresponding exchange-correlation potential. Since the full CI

energy does not depend upon the choice of orbital basis set, the OEP procedure[13, 15]

cannot be used directly for this case. The infinite sum of all energy corrections does not

depend upon choice of orbital basis set, but each term of this sum does depend upon

the choice of orbitals. This fact enables us to consider all term of the infinite sum of

energy corrections as orbital-dependent functionals. After the construction of the set

Vx1) .... t) and summation up to infinity, we will have the exchange-correlation potential

which corresponds to the full CI. Using the equivalence of the functional derivative and

the effective operator approach, it is possible to conclude that after summation of this

set of potentials, we again will have the same result as in the effective operator approach.

Redefined in such a way, the OEP procedure[13, 15] for the full CI energy produces the

same density as the ZMP method[19] would from full CI.


For the calculation of last term in equation (6-5)an auxiliary basis set decomposition

is used

< 'Oaa'(ha Oc Iga r- OkO >= < OaOiXA >< 1 TiXp >< 'Pco'CkuXv > g\Pv (B-1)

where X, are the auxiliary functions. The expression for g\,, can be written in the form


gA/\ = Z(X )A(X ),(X )"r + Z(X1)A(X-1), +
Ki7,rl K,r
(X- 1)(X-1),)A( + (X- 1)A(X- ),,o (B 2)


X, = 2 < ajA>< a (B-3)
Ei Fa

Two separate g\, should be constructed for each spin component. Expressions for 0,,,,

OXA, and 0,,, can be written in terms of the diagrams (Fl) (F18).

,-,, 2_ f ia,(< 'PaPjK >< ', .'rlI > + < Pa pjrI >< ', -.'.. >)
ija Ea)(Ej Ea)
-2 fiav(< ibbK >< Pa brl > + < bPil >< aIbK >)
a(Eb Ea)(Ei Eb)
+2C fijv < OaOiK >< Oa)jrl > 2 fabv < 2i(4bK >< )Pia > (B4)
( a)( a) ab ( Ea)(E Eb)

where the integrals involving the exchange kernel can be calculated using the formula

fpqv >= < q > (X- 1)AQpX- 1)P/ (B-5)

When rules for taking functional derivatives diagrammatically are applied to this set of

diagrams, we will have the next set of diagrams (2-37)

2 + 2 +2 F + 2 ((2-37)

The first two diagrams are equal to diagram 16; the third and fourth equal to diagrams

17 and 16 respectively. Since the left-hand site of eq.12 can be represented by diagram

3, we have an exact equivalence between the functional derivative and the effective

operator approaches in the first and second order for the "Kohn-'lI ii partitioning of the


2.3.3 Interconnection in Higher Orders

Equations for the effective-operator approach in the n-th order of MBPT have the

form (2-38)

< (KS (1)+.... (n) (+)Ap(r)((1) .... (n))IK > 6(6Q(k)+Q(n-k)) = 0 (2-38)
This equation tells as that we need to collect and equate to zero the sum of all possible

combinations of wave operators and density correction operators, and these combinations

must be of n-th order. The same result should correspond to taking functional derivatives

from all energy diagrams of n-th order.

Statement 1 : When functional derivatives from some vertex of a diagram are

taken with respect to an occupied (unoccupied) orbital, and a line, which corresponds to

an unoccupied (occupied) orbital inserted, we .i.-- ,i- will have diagrammatic expressions

which correspond to part of k 0(Q(k)+2Q(n-k))sc.

Proof: When functional derivatives are taken with respect to an occupied (un-

occupied) orbital, and a line which corresponds to an unoccupied (occupied) orbital is

inserted, we alv--,v- have a contraction of the delta function with some fully connected


[1] A. Szabo and N. S. Ostlund, Modern Quantum C'il iii,,.ry, Introduction to Advanced
Electronic Structure Ti, ,.., (\lcGrow-Hill, Inc., New York, 1990).

[2] R. J. Bartlett, in Modern Electronic Structure Th(.. ,; Part I, edited by D. R.
Yarkony (World Scientific Publishing Co., Singapore, 1995).

[3] F. E. Harris, H. J. Monkhorst, and D. L. Freeman, Algebraic and Diagrammatic
Methods in M ri,;-Fermion Theory (Oxford University Press, New York, 1992).

[4] P.-O. L6wdin, J. C'!. in Phys. 19, 1936 (1951).
[5] P. Hohenberg and W. Kohn, Phys. Rev. 136 B864 (1964).

[6] W. Kohn and L. J. Sham, Phys. Rev. A 140 A1133 (1965).

[7] E. Runge and E. K. U. Gross, Phys. Rev. Lett, 52, 997 (1984).
[8] E. K. U. Gross and W. Kohn, Adv. Quantum. ('C!. ii, 21,255 (1990).

[9] M. E. Casida, in Recent Advances in D. -./il; Functional Methods,Part 1, edited by
D. P. C('!..i(World Scientific, Singapore, 1995).

[10] M. Petersilka, U. J. Gossmann, and E. K. U. Gross, Phys. Rev. Lett 76, 1212

[11] J. Lindenberg, Y. Ohrn, Propagators in Quantum C'l ii,,..iry (John Wiley & Sons,
Inc, New Jersey, 2004).

[12] S. Ivanov and R. J. Bartlett, J. ('C!. ii Phys. 114, 1952 (2001).

[13] I. Grabowski, S. Hirata, S. Ivanov and R. J. Bartlett, J. Chem. Phys. 116,4415

[14] A. G6rling and M. Levy, Phys. Rev. A 50,196 (1994).

[15] J. D. Talman and W. F. Shadwick, Phys. Rev. A 14 36 (1976).

[16] R. J. Bartlett, V. Lotrich and I. Schweigert, J. C'!. iin Phys 123, 062205 (2005).

[17] R. J. Bartlett, I. Grabowski, S. Hirata and S. Ivanov, J. Chem. Phys. 122, 034104

[18] L. M. Frantz and R. L. Mills, Nucl. Phys. 15,16 (1960).

[19] Q. Zhao, R. C. Morrison and R. G. Parr, Phys. Rev. A 50, 2138 (1994).
[20] A. G6rling, Phys. Rev. A 57 3433 (1998).

[21] S. Hirata, S. Ivanov, I. Grabowski and R. J. Bartlett, J. Chem. Phys. 116,6468

Using equation (2-13) it is possible to write down the functional derivatives from some

denominator in the general form:

De K1 K1 Ka Ka
4 (_ | Ii2 1|2 2 )1|jl 2 c" 12 > ia Y ja +.+
V, / I ( occ I uno I J ( occ unocc) ..... +
i=1 j=1 a,a#1 i=1 j=1
K/ K3 Ka Ka
... ...+ ( 2 _- E I2) E E (2-28)
i=l j=1 a,a43p i=l j=1

Now it is possible to write down a term which includes the functional derivative from the


Norm Den
(Den)2 6V,
KT(vi3 1<3 2 KB3 2

Socc j noocc j u(2n29)

We now have the possibility to formulate how to take functional derivatives from denom-

inators. To take functional derivatives from denominators, all diagrams, where one of the

lines is doubled and between these lines which arise from doubling, the corresponding

diagonal 6-function is inserted. This procedure must be provided for all horizontal lines on

the diagram and for all the contours the lines cross.

2.3.2 Diagrammatic Functional Derivatives in Second-Order Many-Body
Perturbation Theory

Direct interconnection in first order is clear from diagram's 2 and 3.The left-hand

side of equation (2-17) is equal to diagram 2, the right-hand site is equal to diagram 3

when the functional derivative has been taken from the expression for the exchange energy.

The diagrammatic expression for the second-order energy in the general case has the


E(i b a b (
E(2) F +2 + 2 (230)

BIOGRAPHICAL SKETCH ................... .......... 135

. -i ,! II dll ic behavior, achieve by the Colle-N. -bet algorithm, makes it possible to avoid an

underestimation of Rydberg state energies and overestimation of the polarizabilities. This

overestimation for polymers, particularly for hyperpolarizabilities, has attracted a great

deal of attention[64-66]

The problem of charge-transfer excited states in TDDFT relates to the inability of

the zeroth-order orbital energy approximations, as the HF virtual orbitals and TDOEP

excited orbitals show qualitatively different behavior. If the HF and OEP operators were

the same, a proper treatment using the whole space should be able to overcome this

limitation, even if the usual zeroth-order approximation offered by TDDFT would be

a poor one. However, this difference arises from the form of the operator, local versus

non-local, and, as such, remains a fundamental problem for the DFT method itself, as has

been observed [51]

In conclusion, for all the properties considered in the paper, except for charge-

transfer excited states, there seems to be no significant difference between the results of

TDHF and exchange-only TDOEP, as one might expect. Hence, the focus should remain

on the correlation potential as it is in ab initio dft[16, 17, 48, 62].

Table 3-1. Orbital energies and zero-order approximations to excitation energies
Exchange-only OEP-MBPT(2)
orbital IP[28] Orbital a, FHOMO Orbital Fa FHOMO Excitation
energies energies energies(exp)
2s 48.42 -46.19 -43.25
2p 21.56 -22.51 -20.23
3s -5.15 17.36 -7.58 12.65 16.85
3p -2.59 19.92 -3.72 16.51 18.70
4s -1.95 20.56 -2.66 17.57 18.73
3d -1.49 21.02 -2.05 18.18 18.97

[28] This compares to the OEPx value of 22.51 and OEP-MBPT(2) value of 20.23. The

remaining unoccupied, but negative energy 3s orbital is changed by over 2 eV due to the

MBPT(2) correlation. The OEP2(sc) ab into dft value changes this to 5.18 eV, attesting

to the poor convergence of the standard KS partitioning of the Hamiltonian. Once

OEP2(sc) calculations are done for the 3p, 4s, and 3d states, there is similar agreement

between the OEPx and OEP2(sc) results, contrary to those shown in Table 3-2 for the

standard (ks) choice.

Table 3-2. Excitation energies of Ne atom using OEP-MBPT(2) Kohn-Sham orbital
Term EOM-CCSD TDDFT-exchange-only TDDFT OEP-MBPT(2) Exp[29].
3P 16.5353 17.1748 15.2132
1p 16.7274 17.5209 11.4434 16.847
3 18.8303 19.8681 16.3320
1 18.8822 19.9181 16.9465 18.703
3P 18.9156 19.9134 17.5645
1p 18.9791 19.9134 17.7686 18.725
S 19.5118 19.8811 17.t. ;
S 19.9323 2', '- ; 17.3454 18.965
3P 20.3747 20.4905 18.3455
1p 20.5588 20.6090 18.8788

Despite the differences, at least both the exchange-only and the OEP-MBPT(2) give

qualitatively correct results for the Rydberg series. Helping to ensure this is the fact that

the exchange potential has the correct .,-ii iil '1 i c behavior and the exact cancellation

To make the transition from an infinite basis to a finite computational basis, one

introduces the matrix representation of the OEP equation. Eqn (5-18) then becomes,

w(Vx + K) 0



w = aiR1

Using weighted least squares

6Vx[) :,ia < ilVxa > +Kia22] 0


gives the OEP equation, Eqn (5-16).

As is customary in practice, the computational form for the OEP equation can be

obtained by twice applying the one-electron projector,

0(1) = nr] >< rn\rl >'< r]1

where | r) defines an auxiliary basis, respectively, for electron 1 and 2. Without restriction

the basis can be assumed to be orthonormal The so-called outer projection[53] of the

response function is then

X,, = (ai )(iav) /(ci ea)


Using this expression and insisting upon the linear independence of I ry) gives the OEP

equations in the form,

< l|X(1,2) >< l|Vx(2) + oK(2))






S (pai) (ijla)
S(i ja)


where the o indicates the product of X and K in Eqn (5-18) that results in the Y column



Variation of the energy can be presented in following form:


dr2 6~p(r2)


We can write down variations of the orbitals and the Kohn-Sham potential too:



Using equations (2-8) and (2-9) we can rewrite the variation of the energy in the following

6E ") r= dr2
S 4p(T)

dr6p, (r2)

V (rl)
dr 6p(r)

Now we can write down the chain-rule[13, 14] for the calculation of Vc) :

6E ")r 6E ")
6Jp(r) 6 6p(r2)

6rp,(r2) 6V,(r3)
6Vs(r3) 6p(r3)

S 6E( 6V(r3)
dj (V() ( p(rl)

Taking into account the fact that[14]:

6Vs(ri) C p ,C
p,q p



it is possible to write an explicit expression for V~) (r) [13]

V(")(r) 2( J
p,q p
Using the fact that

dr26E- () dr3 p (r3) q (r3)X- (r, r3)
Op,(r2) Jp Eq


dr2X(r1, r2)X- (r2, r3)



X(r, r2) 2 z ) i( ) c)a(rl)i(r2)>a(r2)




6p(2f d p2 6V/(r) (
6V,(r) dr 6 p(r) 6p(r)
j 6P(r)

S,(r) ,(r)

6(r1 r3)



ACKNOW LEDGMENTS ................................. 3

LIST OF TABLES ....................... ............. 7

LIST OF FIGURES .................................... 8

A BSTR A CT . . . . . . . . .. . 9


1 INTRODUCTION ...................... .......... 12

1.1 Ab Initio Wavefunction-Based Methods .................. 13
1.1.1 Hartree-Fock Method ......... .......... .... 13
1.1.2 Electron-Correlation Methods ....... ............ 16
1.2 Kohn-Sham Density Functional Theory ............. .. .. 19
1.2.1 Time-Dependent Density Functional Theory . . 21
1.2.2 Time-Dependent Density Functional Theory Linear Response Theory 23
1.2.3 Problems with Conventional Functionals . . ..... 26
1.2.4 Orbital-Dependent Functionals ................ .. .. 28
1.2.5 Ab Initio Density Functional Theory ................ .. 29

TIONAL THEORY ........... ..... ........... 31

2.1 Equations for the Exchange-Correlation Potential in the Functional Deriva-
tive Approach ........ .. ...... ............... .. .. 31
2.2 Equations for the Exchange-Correlation Potential in an Effective Operator
Approach ................... . . .... 34
2.3 Interconnection in Arbitrary Order ............ ...... 37
2.3.1 Diagrammatic Functional Derivatives . . . .... 37
2.3.2 Diagrammatic Functional Derivatives in Second-Order ,M ,iv-Body
Perturbation Theory .................. ....... .. 38
2.3.3 Interconnection in Higher Orders .................. .. 41
2.3.4 Interconnection in Infinite Order .... . .. 45

EFFECTIVE POTENTIAL .................. .......... .. 47

3.1 Diagrammatic Construction of the Exchange-Correlation Kernels . 48
3.1.1 Formalism ................ .... . . 48
3.1.2 An Example: Diagrammatic Derivation of Exchange-Only Kernel 52

4.2.3 Dissociation Energies

Dissociation energies, calculated with semi-canonical OEP and PBE are presented

in Table 4-3. Potential curves for the LiH, OH and HF molecules are shown on Fig 4-3,

4-4 and 4-5. For all curves the dissociation energies have approximately the same level of

accuracy as in the MBPT(2) case, but the semi-canonical OEP-MBPT(2) improves the

shape of the curves.

Table 4-3. Dissociation energies (in kJ/mol)
PBE OEP-semi Exp.
N2 1013.27 964.56 941.64
02 598.14 549.12 493.59
CO 1122.16 1122.00 1071.80
CN 821.151 794.51 745.01

4.2.4 Singlet-Triplet Separation in Methylene

Results for the extensively studied splitting of singlet and triplet states of methylene

[41, 42] are reported, using the uncontracted ROOS-ADZP basis sets: 13s9p3d for carbon

and 8s4pld for hydrogen. Equilibrium geometries for both states are taken from reference

[41].The energies for the two states, calculated with different methods are presented in

Table 4-4. There are significant differences in the energy separation compared to the

experimental value (8.998 kcal/mol) for all of the presented methods. This is a basis

set issue, but can also be explained by the fact that the singlet state of methylene has

a significant contribution from two-determinants[41], so MBPT(2) is a much poorer

underlying approximation for their difference than infinite-order CCSD or CCSD(T), and

especially the two-determinant CCSD (TD-CCSD) results[41]. Where the absolute values

of PBE energies are not too good for the two states of CH2, the difference is consistent

with MBPT(2) and its OEP-generated ab initio dft form.

4.3 Conclusions

Ab initio dft calculations with OEP-MBPT(2) semi-canonical potentials show

significantly improved results over OEP-MBPT(2) with the Kohn-Sham partitioning of

while diagram (VC7) gives diagrams (VC7-1)-(VC7-14) (3-26).

2 i 1a

(VC7-1) (VC7-2)

(VC7-3) (VC7-4) (VC7-5) (VC7-6) (VC7-7)


(VC7-8) (VC7-9) (VC7-10) (VC7-11) (VC7-12) (VC7-13) (VC7-14)

Finally, diagram (VC8) produces diagrams (VC8-1)-(VC8-14) (3-27).

(VC8-1) (VC8-2) (VC8-3) (VC8-4) (VC8-5) (VC8-6) (VC8-7)

(VC8-8) (VC8-9) (VC8-10) (VC8-11) (VC8-12) (VC8-13)



In the same way the rest of the diagrams should be differentiated. Since the Fock operator

depends upon occupied orbitals, F-rings on the diagrams (VC9)-(VC15) must also be

differentiated. Diagram (VC9), after taking functional derivatives, produces diagrams

(VC9-1)-(VC9-10) (3-28).

a 6 a
2 k2 az2

(VC9-1) (VC9-2)

k F 6

F F b
(VC9-6) (VC9-7)
(VC9-6) (VC9-7)


b 6




(VC9-4) (VC9-5)
b k

V 2 a J6
(VC9-4) (VC9-105)


of an exact mapping between densities and external potentials. In the ground state

formalism, the existence proof relies on the Rayleigh-Ritz minimum principle for the

energy. A straightforward extension to the time-dependent domain is not possible since

a minimum principle is not available in this case. The existence proof for the one-to-one

mapping between time-dependent potentials and time-dependent densities, was first given

by Runge and Gross[7].

We can start from the time-dependent Schr6dinger equation

at H( (t)= (t) (1 42)

evolving from a fixed initial many-particle state

((to) -= o (1-43)

under the influence of different external potentials v(r,t). For each fixed initial state To,

the formal solution of the Schr6dinger equation (1-42) defines a map

A: v(r,t) -+ T(t) (1-44)

between the external potential and the corresponding time-dependent r: ,_vi--particle wave

function and a second map

B : (t) p(r, t) =< T (t) &(, t) (t) > (1-45)

Thus, the following mapping can be established:

G: v(r, t) p(r, t) (146)

The Runge-Gross theorem establishes that the G mapping is invertible up to some

additive, time-dependent constant. In other words, two densities p(r, t) and p'(r, t)

evolving from the common ground state To under the influence of potentials v(r,t) and

However, the number of possible excited determinants grows rapidly with the

number of electrons and basis functions in the system. Therefore, the full CI method

is computationally intractable for any but very small systems. Among the approximate

electron-correlation approaches, the most common are the Coupled Cluster method[2]

and Many Body Perturbation TI. ii [ :]. Any truncated version of the CI method

has a qualitatively wrong behavior of the energies and wave functions while increasing

the number of particles in the system. Therefore, the CI methods with limited level of

excitations cannot be used for highly-correlated systems. The Coupled Cluster method

and Many Body Perturbation theories are free from this lack of extensivity failure and

are very common for the molecular computations. In some cases perturbation theory

can provide an accurate description of electron-correlation effects at a much lower cost

than necessary for the Coupled Cluster method. The second-order Rayleigh-Schr6dinger

perturbation theory is the simplest and least expensive ab initio method for taking into

account electron correlation effects.

In this perturbation theory the solution of Schr6dinger equation

Hf = E' (1-16)

can be found using the Slater determinant as a reference. Generally, such a determinant

may be constructed from the orbitals, generated by some one-electron operator

hop = (1V2 + u)yp = pp (1 17)

The first step of any perturbation theory is the partitioning of the Hamiltonian into a

zero-order Ho and perturbation

H = Ho + V (1-18)

Ho = -Eo = (hi)K = (- F)K (1-19)
i i

this zeroth-order difference should be a decent approximation to the excitation energy, and

is, as shown elsewhere [16] when accurate exchange-correlation potentials are used. In this

sense, the orbital energies in KS-DFT should have a certain meaning.

Furthermore, as pointed out in ref [16], we can also consider this equation to offer

a Koopmans-like approximation to the principal ionization potentials, since, barring

pathological behavior, when we allow an electron to be excited into the continuum, its

orbital a will then have no overlap with the bound orbital matrix elements of hKS and

the kernel. Consequently, we are left with nothing but -ci in the TDDFT equations. So in

this 'sudden', adiabatic approximation, the KS orbital energies should offer an estimate

for each of the principal Ip's, not just the highest-occupied (homo) one. When based upon

the relatively correct V,, obtained from ab initio dft[16, 17] this estimate is superior to

Koopmans' theorem for the homo and the first few valence Ip's, but is inferior for the core

or'.il '1-[21]. See also C('l.: et al [25]. However, the OEP2 semi-canonical (sc) ab initio

dft [16, 17] approximation has the distinct advantage that it uses a much better behaved

unperturbed Hamiltonian than the usual KS choice, Ho = : hKs(i).

To illustrate the evaluation of the kernel and the solution of TDDFT equations, we

consider the Ne atom. All the excited states in Ne correspond to Rydberg excited states.

To obtain reasonable values requires a quite extensive, diffuse basis set. We choose to

start with the ROOS-ATZP atomic natural orbital '. ,--[[27] consisting of (14s9p4dlf)

primitive gaussian functions contracted to a [5s4p3dlf] set. This underlying basis was

then augmented by a set of even-tempered diffuse functions [3s3p3d], with exponential

parameters = ,' a8 = 0.015, a = 0.013, ad = 0.012, b=1/3. The auxiliary basis is

chosen to be the same, but without the p and d diffuse functions, since for the description

of the exchange-correlation potential, diffuse functions are not necessary. The orbital

energy estimates are shown in Table 3-1.

The ionization potential equation of motion coupled-cluster (IP-EOM CC) result

in this basis for the Ne homo Ip is 21.3 eV with the experimental value being 21.5645


Ro (1-30)
Eo Ho
and Q 1 I >< K is the projector onto the complementary space of |P >. Since Ho is

diagonal in the basis of Slater determinants, it is possible to write

R0 > ti i <1 j y j (1 3 1 )
Ro = Eo E, (-31)
n 0

For the special case of the Hartree-Fock reference determinant

Eo + E(1) =< liHo01 > + < 1V1 >= EHF (1-32)

the second-order correction to the energy is
10CC uOCC .
E (2) 1 oc < zjlab > (< ijIab > < ijlba >)
2 E EY E. -Eb
i,j a,b

For the case of a more general reference determinant, constructed from orbitals of Equa-

tion (1-17), the second-order correction to the energy has the following structure
ocC unoc 12 occ unocc .
E(2) I < i h- fla > 1 1 < ijlab > (< ijab > < i(ba> 34)
i a i,j a,b

1.2 Kohn-Sham Density Functional Theory

Density Functional Theory is an alternative approach to the description of the

electronic structure of molecular and solid-state systems. This method uses electronic

density instead of the wave function as the basic object of theory. The mathematical basis

of DFT is provided by two theorems, introduced by Kohn, Hohenberg and Sham. The

first one, known as Hohenberg-Kohn theorem[5], establishes a one-to-one mapping between

the ground-state electronic density and the external potential. The external potential

defines a particular object (atom, molecule, etc) and, because of the one-to-one mapping,

the density contains all the information about the system. In particular, the ground-state

energy can be written as a functional of the density. To get the ground state energy in

Finally, there is no doubt that the standard equation of motion coupled-cluster

methods (EOM-CC)[31]) are less time-consuming than is the present calculation. As long

as the kernel for a rigorous orbital-dependent correlation potential is this complicated,

two-particle wave function theories like EOM-CC are both superior and easier to do. But

before we can further exploit the interface between DFT and wave function theory to the

benefit of both, it is requisite to have the orbital-dependent kernel defined. This, and its

initial evaluation, is what this chapter offers.

6 2 (< calV.cb > +Kab) F5
i,j,a,b ( E )(E- b)(j a)
F5 (< pacjK >< CcjT >< > < ai > + >< a >< i > > +

< Pa ( >< i pj T1 >< ib> + < (ij >< K,- -,C > +

< Pja(~C >< (i~(jK >< ibi ab)l > + < (~a(jri >< (jCi(O >< ()(bK >) (B-11)

D7 -2 (< Va V b > +Kab) F6
(,jki Ei EbEi E

< ><(b c i)(j- b)( a)
F6 7 (< kpip-,K >< jpbcl >< Xpipa,( > + < icK: >< bc, >< cica > +

< Pi Pc >< b (Pcr >< i~aK > + < b 7l >< b()O{cK >< r r ,C > +

< pi >< bO >< Ci Cal > + < bOcX >< bO >< ccaK >) (B-12)

( < oi| |o > +Kij) F2
D8 = -2 (B-13)
i Ek Ea) E Ea) (Ek Ea)

The second part is represented by the sum of diagrams (SX1) (SX6)

D9-2 < ^Oi|a^fcO > *F7
(,Kjkab K (Ek Eb )(Ej b)(E Ea)
F7 = (< Ob i.' >< OkTl >< a( > + < Wb i-. >< kcOck >< aiOrT > +

< P0bPk( >< (c(jTl >< iaK > + < (j(c >< (~Cc( >< Xia( > +

SD10= 2 < Pi Pa PjPb > F8
ijab,c Ec(- Eb)(Ei Ea)
F8 = (< ;. / >< cWbl >< Oa( > + < (r,-', I r i'< c X Wal > +

< OPjc >< OcC cl >< Oi OaK > + < i jscrl >< PcObK >< Oa( > +

< (PjPcc >< OPcObK >< Oal > + < ijOPcjl >< a rl >< Oi aK >) (B-15)

exchange potentials (Fig 4-1 and 4-2) is guaranteed by the use of the Colle-N. -1 et seed

potential algorithm[30]. In all our calculations a Slater potential[38] was used as a seed

potential, since the Colle-N. -1 i algorithm requires the seed potential to be as close as

possible to the actual potential. Two separate Slater potentials, one for alpha and the

other for beta, were used. For all the closed-shell systems from Table 4-1, the numerical

- OEP-semi alpha
- OEP-semi beta
- PBE alpha
- PBE beta

2 3
R,a. u

Figure 4-1. Exchange and correlation potentials of Li atom (radial part). A) Exchange
potential. B) Correlation potential

R, a. u.

-1 -




R, a. u.

-6 -4 -2 0 2 4
R, a. u.

Figure 4-2. Exchange and correlation potentials of 02 molecule across the molecular axis.
A) Exchange potential. B) Correlation potential

results obtained from OEP-MBPT(2) with the semi-canonical potential and from DFT

(VC8 13) + (VC8 14)
< ablik >< ijlba >< ckA >< cjp >
(Ei + k Ea Eb)(Ei + sj -a b)(Ej -


2 f faifkj(< ijA >< akp >
i,j,k,a ( a)A (

faifbj(< ijA

>< abpi >
- Ea) (Ej -

) (A-70)

1) + (VC11- 8)-
+ < akA >< ijp >)
,a)(Ek a)

2) + (VC10 3)
+ < abA >< ijp >)
a) (Ej Eb)



ijA >< abp >
(i F)(i-

3) + (VC10
+< abA ><
E Eb)

ij >) (A73)

(VC9 4) + (VC12 5)

2 Y faifba(< ijA >< bjp > + < bjA >< ijp >)
i,j,a,b (i Ea)(E E(E Eb)

(VC9 5) + (VC9
fai0fka(< ijA >< kjp > + < kjA ><
,j,,a (E- Fa)(j Fa)(Ek a)
. j, Ak, a

ijp >)

(VC9- 7)
Sfaifaj < jbp >
( ija) ( a) ( b

(VC9 8) + (VC15 3)=
4 fai < jblak > (< ijA >< kbp > + < kbA >< ijp >)
,j,k,a,b ( Ea)(E Ea)(k b)


f, fbj(<





we can rewrite equation (2-15) as:

Y o.a(ri)o(ri) < t 1Vnloa > > < E(>) I p(F3)tq(3) (2 17)
a i Fpa q p p -
a,i p,qlP

Using the fact that orbitals p and q can be occupied or unoccupied,we can separate the

right-hand site of equation (2-17) into four parts:

Pa(ri)(Pt() < iK(V a > E(n) 3)(> (r3)
a Fai 6o( Fi -- F

a t a Fi 6 o Fi Ft
a,i J 4,i
6E(") ((3)Yb(r3) (2
+ > < I> a b + c.c (2-18)
abea (pa Fa Fb

This form of equation ( 2-18) will be the basis for the formulation of the diagrammatic

rules for taking functional derivatives and for making the connection with an effective

operator approach to the OEP-MBPT exchange correlation potential.

2.2 Equations for the Exchange-Correlation Potential in an Effective
Operator Approach

Let us consider a one-particle density operator in second-quantized form:

p(r) = < v(r rl)Pq > a+aq (2-19)

Using Wick theorem,we rewrite this operator in normal form with the Kohn-Sham

determinant as the Fermi-vacuum:

p(r) = < Pp J(r- rl)Jl q > {a+aq}+ < LKs 6(r rl)\l KS >
< Kp 1 \(r ri)|Pq>< > IP)2+ a a(I)+ (220)
p,q i=1

The first member of the previous expression will be called the /. ,, .:-/ correction:

Ap(r) = < ppJ6(r rl)|, > {aqa,} (2-21)

6.3 Conclusions

We have presented the analytical expressions and implemented them for the calcula-

tions of static hyperpolarizabilities applicable to general molecular systems with the exact

exchange functional of Kohn-Sham DFT through the TDOEPx formalism, the first step in

the application of ab initio dft[16] It has been shown that the rigorous exchange treatment

in TDOEP and TDHF provide hyperpolarizabilities that closely agree with each other, as

one would expect, although this will not happen for charge-transfer excited states[51, 74].

This clear agreement with TDHF is in contrast to conventional exchange functionals

that severely overestimate hyperpolarizabilities. OEP potentials and kernels are not only

capable of describing excitation energies, but also higher-order properties. Since none of

the standard DFT functionals shows the correct analytic properties of the second kernel,

the OEP formulas and results can be used for the testing and calibration of new density


exchange-correlation potentials are produced by the functional derivative and density

condition approaches.

The time-dependent extension of OEP in the adiabatic approximation with

exchange-only potentials and kernels was recently implemented and applied to some

molecular systems. The corresponding excitation energies and polarizabilities are in good

agreement with time-dependent Hartree-Fock results. However, such results cannot be

used for comparison with experimental values due to lack of a description of electron

correlation effects. In order to account for correlation effects, a correlation kernel corre-

sponding to the MBPT(2) potentials has been derived. Its structure and properties are

described in detail. The first numerical results for excitation energies with pure ab initio

potentials and kernels are presented.

The recently implemented OEP MBPT(2) method has been extended to the case

of spin-polarized open-shell systems. The total energies obtained for several open-shell

systems are very close to the corresponding values obtained with the highly-correlated

coupled cluster singles and doubles with perturbative triples (CCSD(T)) method. Com-

parison with results obtained with the OEP MBPT(2) exchange-correlation potentials and

some density functionals has shown a qualitatively incorrect shape for some widely used

exchange-correlation potentials.

Higher-order response properties, such as hyperpolarizabilities, are described

very poorly with conventional functionals. The typical relative errors described in the

literature are about 100 percent. The reason for the poor description is an incorrect

long-range .,i-mptotic behavior and an incomplete cancellation of the coulombic self-

interaction. The OEP method is free from those drawbacks and calculated values of static

hyperpolarizabilities with the exchange-only potential are close to those derived from

Hartree-Fock theory. The second kernel required for calculations of hyperpolarizabilities

within the DFT framework has been derived for OEP potentials by using the developed


elec Nucl

>I: Iri RAI
i A

VH(r) -yC f *(Fi)4)(ri) dr p(ri) dr
) rll /r i-- ^1
p(r) = (T) i (r) (1-10)

The Hartree-Fock exchange operator nix is non-local, i. e. it cannot be presented as an

analytical function of spatial variables. However, it is possible to write its action on some

orbital pi

vnlxi(r) ji(/ --)P1 (r)dri (1 11)

Since the Fock operator f is hermitian and vH(r) and vnixi(r) are invariant with respect

to unitary transformations of orbitals, Equation (1-8) can be rewritten in the canonical


fPi = Eii (1-12)

This form of the Hartree-Fock method is most common for practical implementations.

The Hartree-Fock method is a system of integro-differential equations, which cannot

be solved analytically. Iterative methods can be used for the approximate solution of

the Hartree-Fock equations with a given accuracy. As a first step some guess of orbitals

should be assumed and substituted into the equations. During the second step the

system of equations is solved and a new guess for the orbitals is obtained. Then those

new orbitals should be substituted into the equations again, until the total energy and

density at successive iterations differs less than a required accuracy. This is known as the

self-consistent field (SCF) method.

For practical purposes molecular orbitals are usually presented as being decomposed

into Gaussian functions, centered on the nuclei. In this case the number of solutions is

much larger than the number of electrons. To decide which orbitals should be included

the X based OEP have no problem even for this situation.

As seen in Table 5-2, even the notoriously unstable second-order OEP procedure,

based on HOK has no particular difficulties, either. Since the direct minimization proce-

dure of Yang and Wu[46] is based on the same equation as the regular OEP method, all

of the problems, described in ref [49], are caused by using the BFGS algorithm instead

of solving the OEP Fredholm equation with the SVD procedure for the X. That might

appear to be surprising since Ch = l, would mean that the resolvent operator is manifestly

singular, clearly not that appropriate to a non-degenerate perturbation theory functional,

which has alv--,i been the intent of such orbital dependent expressions. However, it is also

apparent from the standard, chain-rule differentiation used by Rohr et al[49], namely

Vxc(1) Exc/6p(1) = d2d3[6Exc/5,p(2)][6p(2)/6v,(3)][6vs(3)/6p(1)] (5-32)

that the derivative would not be defined from the beginning as 5Exc/56p(2) has to be

singular when cHOMO = LUMO. For comparison, the OEP2sc result from ab initio dft is

also shown, compared to the normally, poorly converged OEP2 based on the HKS

Other simplifications in the OEP solution are potentially possible. The most obvious

is to make an Unsold, constant energy denominator approximation(CEDA), replacing i -

a C v,9.(This should be contrasted with that of Kreiger-Li-Iafrate (KLI)[44] who made
such an approximation in the Green's function where the denominator consists of Cp Cq

before simplifying the from to the occupied-virtual separation invoked here[16, 54, 57].)

See Gritsenko, et al [56]. Such a constant energy denominator is a relatively painless

approximation here, since Cav appears on both sides of eqn. (18), making it disappear

from the equation for Vx. Once that is done, a resolution of the identity can be invoked to

eliminate the virtual orbitals to give,

(ip la)(a|K + Vxi) ({(ip(K + Vx)i (iplj)(-|K + Vxi) (5-33)
i,a i ij

(VC14 13) + (VC15 15)
v-4 ^ < cjlka >< ablij > (< ibA >< cky > + < ckA >< ibp >)
*4 -,(A-118)
bcijk (E Eb)(EF Ea) (Ek Ec)

(VC14- 15)
2 < ablji >< kilac >< jbA >< ck > (A 9)
ab,c,i,j,k (i a)( Eb)( c)

(VC15 7) (VC15 8)
8 fa < ablij > (< bkA >< jkp > + < jkA >< bkp >) (A 20)
i,j,k,a,b ( i Ea)(Ei Eb)(E- k Eb)

(VC15 9) + (VC15 10) =
8 fai < ablij > (< cjA >< bcp > + < bcA >< cjp >) (A 2
aL,(Ei E,)(E Ec)(Ej Eb)

(VC15 11)
-8 Y fai < aklij >< jbA >< kb >(A 22)
-8^ -(A-122)
ijkab (Ei Ea)(Ej Eb)(Ek Eb)

(VC15- 12)
8 z fa < ablic >< bjA >< cj > (A 2)
abci,j ( a)( Eb)(EF Ec)

(VC15 13)
16 < ablij >< cika >< bjA >< ck > (A 24)
a,(Eib,c,i,j a)( b)(Ek c)

(V1) + (V2)
-2 < ilVcla > (< abA >< i > + < bA >< ab >) (A 25)
,b ( a)(- b)(A125)


I would like to thank professors Henk Monkhorst and So Hirata for helpful discus-

sions. I want to thank also Dr. Igor Schweigert and Dr. Norbert Flocke for the helping me

to write OEP code. My special thanks to Tatyana and Thomas Albert for help with the

preparation of the dissertation text.

However, there are some other pertinent considerations before doing so. The

condition that the KS single determinant gives the exact density through first order

in its correlation perturbation, obviously imposes the condition that p' 0. This

density condition is very general as it applies to exchange and correlation, and has

been used quite successfully in defining ab initio dft exchange and, in higher orders,

correlation potentials[13, 16, 17, 48], since it also transcends any particular separation of

the Hamiltonian in perturbation theory. In particular, it is not subject to HKS = hs(i),

and thereby alleviates the pathological behavior that gives no convergence when defining

such correlation potentials from MBPT(2), for example [13]. For the exchange only,

however, such effects are less important[16, 17]. Therefore this paper will use the sum of

Sham Hamiltonians, but the principle for derivation remains the same.

Hence, requiring that

0 PKS (KS ARV \KS) + (KKS VR0 I KS) (5-11)

A p(x, x- ) (5-12)
a \ + lab W-ab\
Rl C Ca) 1 + 2 )(CtI + C Ca Cb) ( +... (5-13)
i,a I
I|h)R(h|l + |h2)Rl(h2 +... (5-14)

V= -[ (i)+ (i) (5-15)
Single excitations in the resolvent operator are indicated by a) ,and collectively by the row

matrix, |hi),doubles by b), h2),etc. It is easy to show[17] that only the singles contribute

to the first-order density, and

Ps y i I )fia +c 2 k(1l)a(1)iK + x|a)/(i )a) 0 (5-16)
?,a i,a

Then this condition defines the Vx(1) operator up to a constant[52]. Note that this

is a pointwise condition as it should equal zero at all xi, but actually imposes the weaker


7013 P13
K(1) (1) 3) 3) 1) (3) r1tp(3)d3j(l1) (5-2)

where the {Qj}are the occupied Kohn-Sham spin orbitals, p=i,j,k,l.... The latter are the

solutions to the equations,

..,/( (1) CpOP(1) (5-3)

hsl') = h(1) + J(1) + Vx(l) (5-4)

Ji() E I (2) 1 (2)d2 (5-5)

p(t) d2 (5-6)
I 12
PKS () i () (5-7)

where the unoccupied orbitals are indicated by p=a,b,c,d....

The condition that the density, p(l), be the Hartree-Fock one, is that

l, (HFlh+ J- K aHF) 0 (5-8)

PHF(l) -I *( (1) (5 -9)

which, as is well-known, is correct through first order in correlation measured relative to

the sum of HF one-particle operators, Ho = Y f(i), due to the Moeller-Plesset theorem.

For KS orbitals, fia is not zero. Hence, if the objective were to maximize the similarity

in the KS and HF density, it would then require minimizing a quantity composed of KS

orbitals, related to

min |\(if- 1sa)|2 = min1 (iK + x a)2 (5-10)
i,a i,a

,---.I -:i.-; a kind of least-squares minimization as has been discussed previously[13, 16, 17,

48]. Clearly, this minimization would have the trivial solution that (i K a) = -(i Vx a).

a(r4 r s5 r6 6(r4 r 5 r 6)- 6(r4 r5 r 6)4 8 (r4\ r\ r) 6(r 4 r 6)-6b
r4 r r6(r 4 rs\ r 6) 4 r56 r 6)
k a 6(r4 rs5 r6)l 6 ) rrr) r5 r 66 )jr r 6 r4 r5 5 r 6
4 5\ r6 )k k k
(SX10) (SX11) (SX12) (SX13) (SX146 -22)

Interpretation of all diagrams which contributes into the second kernel and details of the

implementation are given in Appendix B.

6.1.3 Properties of the Second Exact-Exchange Kernel

All the diagrams contributing to the second exact-exchange kernel have only one

contour and because of this fact only aaa and 333 spin components are present, as

in Goldstone diagrams where Sz has to be conserved at each vertex. The algebraic

expressions for all the diagrams have to be symmetric with respect to permutations of

rl, r2, and r3 variables. In the case of the two-electron spin-unpolarized systems the OEP
exchange potential becomes equivalent to the Slater potential[38]

vSlater(r) -_ i()j() r l) l (623)
Sizj PA IJ r ri(

Taking into account the HOMO condition and considering the expression for the exchange

kernel, it is possible to write for the special case of two electrons

,f EP(p1, 2) (6-24)
ri r2

Diagrams (VX1)-(VX5) and (VX8) of (6-20) will cancel diagrams (SX7), (SX8) and

(SX11)-(SX14) of (6-21) and (6-22) because of the HOMO condition and the behavior of

the Hartree-Fock exchange for the case of two electrons. Diagrams (VX6) and (VX7) will

cancel diagrams (SX1) and (SX2) because of Eqn (6-23). Diagrams (SX5), (SX6), (SX9)

and (SX10) will cancel diagrams (F1)-(F4), (F7)-(F10) and (F13)-(F16) because of Eqn

(6-24). Finally the sum of the diagrams (SX3) and (SX4) cancel the sum of (F5), (F6),

(Fll), (F12), (F17) and (F18). Thus for the special case of two electrons

gOEP (12, 3) 0 (6-25)

The definition of the exchange part of E,, can take into account its definition in

wave-function theory,

E, =< |,IV, > -EH (138)

Kohn-Sham orbitals are defined by an effective local potential vs. Transforming the

variational condition on the energy functional into a condition for the constrained search

over the orbitals, it is possible to write

(r) (E[p] T,) 6(E.Et +EH + Exc)
vs(r) vext + VH + Vc (1-39)
6p 6p

where the exchange-correlation potential is defined as the functional derivative of the

exchange-correlation energy

c(r) 6 (1-40)
Once Exc is approximated and the Kohn-Sham equations are solved, the total energy can

be found from the following expression

E = E drldr2P(1)P(2) drprlxc() + E (1-41)
i= 1 2 Iri r2
The solution of the Kohn-Sham equations is completely analogous to the solution for

the Hartree-Fock case, and it is usually done by an SCF procedure. After self-consistency

is reached, the Kohn-Sham orbitals are guaranteed to reproduce the true density of the

]rn i,--electron system.

Virtually all modern implementations of DFT use the Kohn-Sham scheme. However,

the theory still has open questions as to how to construct the exchange-correlation

functional. Thus, the main challenge for the theoretical development of DFT remains the

construction of accurate exchange-correlation functionals.

1.2.1 Time-Dependent Density Functional Theory

The Kohn-Shame scheme provides the possibility to describe ground state energies

and densities. For the description of excitation energies a time-dependent generalization of

conventional DFT can be used. Ordinary time-dependent DFT is based on the existence

(VC2 9) + (VC2 10)
-4 < ablij >< ijlkb > (< ckA >< acp > + < acA >< ck/ >)
(E, E, Ea -+ Eb) (Ek E) (Ek Ea)

(VC2- 11) + (VC4- 7)
-4 < ablij >< ijlkb > (< ckA >< acp > + < acA >< ck/ >)
b,c,,j,k (cE + j a -- b)(i + Cj Eb Ec)(Ek Ec)

(VC2 12) + (VC3 8)
4 < ablij >< ljlkb > (< akA >< ili > + < ilA >< akp >)
,b,i,,, (i + j Ea b)(El + Cj Ea Eb)(Ek Ea)

(VC2 13) + (VC2 14)
4 < ablij >< ijlkb > (< alA >< lki > + < alp >< lkA >)
a,b,i,j,k,l ( + a b)(k a)( a)

(VC2 16)
< ablij >< ijlcb >< akA >< ckp >
a,b,c,i,j,k + a Eb)E E) k c)

< ablik >< ijlcb
(Ei + EF -

< ablik >< ijlcb
(Ei + k a -

(VC3 3) + (VC3 5)
> (< jkA >< ili > + < jkp >< ilA >)
*b)(Ei + k Ea Eb)(El + j E Eb)

(VC3 4) + (VC4 5) =
> (< jkA >< ac/ > + < acA >< jki >)
Eb)(Ei + -k a b)(Ei + j b Ec)

(VC3 6) + (VC4 3)
> (< jkA >< ac/ > + < acA >< jk/ >)
Eb) (i + Ek Ec Eb) (i + j Ec Eb)






S< ablik >< ljlab
a,b,i,j,k,l ( + -






In this way, there is no dependence on the virtual orbitals as is deemed computationally

or formally important in some DFT circles[59] Of course, using the usual finite basis set

computational tools of quantum chemistry, the distinction is less important. This formula

has recently been applied[58].

Because the OEP equation is a point-wise identity having to be zero formally for all

x in a finite basis set it corresponds to a :r, ii -I-.,-few mapping in the general case. The

dependence upon the delta function makes the OEP procedure sensitive to the auxiliary

basis set used in the calculation. However, the SVD procedure handles the null space

functions plus any potential linear dependency in the basis set in a fairly automatic

v ,i-[;n]. If the X matrix is removed, this is not the case, and some algorithms can result

in unphysical or trivial solutions under certain combinations of auxiliary and molecular

bases. C'!....- i to invoke the average energy denominator and the resolution of the

identity A'. f. the least-squares mminimization, will not give Eqn(24). However, once

we eliminate the delta function that makes the approximation point-wise, we reduce the

problem to the minimization of the variance

min Var min (i(K + Vx)2i)- j iK + Vxi)(jlK + Vxli) (5-34)
i i,j

that can be used to define a Vx but one that is not point-wise, and, consequently, does

not satisfy all the conditions above. This further simplification of the weighted least

squares approach in Eqn (5-26) has also been considered recently [58]. Ultimately, one

primary criteria for the best exchange-only potential should be the satisfaction of the

Janak theorem. As shown in Table 5-1, and pointed out previously[16, 48] this is difficult

to achieve in any normal basis set.

3.2 Kernel for the Second-Order Optimized Effective
Perturbation Theory Correlation Potential .
3.3 Properties of the Correlation Kernel .......
3.4 Numerical Testing .. ..............
3.5 Conclusions . . . . .

Potential Many-Body

SYSTE M S .... ................... ..............

4.1 Theory ................... ..............
4.2 Results and Discussion .. .....................
4.2.1 Total Energies . . . . . . .
4.2.2 Ionization Potentials .. ..................
4.2.3 Dissociation Energies .. ..................
4.2.4 Singlet-Triplet Separation in Methylene .............
4.3 C conclusions . . . . . . . .


5.1 Exact-Exchange Density Functional Theory .
5.2 Time-Dependent Optimized Effective Potential ..
5.2.1 Theory and Implementation ........
5.2.2 Numerical Results ..............
5.2.3 C(!i ige-Transfer Excited States ......
5.3 Conclusions . . . . . .




6.1 T heory . . . . . . . . .
6.1.1 Time-Dependent Density Functional Theory Response Properties
6.1.2 Diagrammatic Derivation of the Second Exact-Exchange Kernel
6.1.3 Properties of the Second Exact-Exchange Kernel .. .......
6.2 Num erical Results . . . . . . . .
6.3 C conclusions . . . . . . . . .


CORRELATION KERNEL .. ........................

K E R N E L . . . . . . . . . .

REFERENCES ......................................

with the PBE potential are numerically close, and close to the reference CCSD(T)

energies. This, of course, is part of the power of GGA and meta DFT, even when such

energies are obtained with highly erroneous potentials. Instead, PBE partly benefits from

the cancellation of errors in V, and V, (Fig 1 and 2). For most of the open-shell systems

the energies of the semi-canonical OEP-MBPT(2) are much closer to CCSD(T) then the

corresponding PBE energies, which is also a consequence of the wrong behavior of the

PBE potential.

4.2.2 Ionization Potentials

Orbital energies and the corresponding ionization potentials are much more sen-

sitive to the exchange-correlation potential than the total energies. Vertical ionization

potentials, calculated as energy differences between the neutral and its ionized system are

calculated using the same basis sets and geometries as in the previous section. Results

are presented in Table 4-2. In exact DFT, the energy of the highest occupied molecular

Table 4-2. Ionization potentials (in e. v.)
Ne 21.01 21.59 13.35 21.69 21.56
N2 16.89 15.03 10.27 15.41 15.58
CO 13.68 13.69 9.05 13.87 14.01
CN- 4.19 3.51 0.149 3.72 3.86
H20 12.37 12.73 7.08 12.46 12.62
N 15.25 14.52 8.30 14.73 14.53
Li 4.89 5.38 3.23 5.59 5.39

orbital (HOMO) corresponds to the negative of the exact vertical ionization potential. In

the case of the semi-canonical OEP the HOMO energy is close to the corresponding AE

values, what can be explained by the correct shape of the exchange-correlation potential

(Fig 4-1 and 4-2 and ref [16, 17]), while PBE does not exhibit correct behavior of potential

and thus, fails to reproduce the correct HOMO energy values. Failure to reproduce the

correct HOMO energy causes the incorrect integer discontinuity of the PBE (or any other

GGA) exchange-correlation potential[39, 40].

the exchange and correlation potentials as a functional derivative of E" + E(2) [13, 33].

However, this is a numerically hopeless procedure without any resummation of terms: (1)

it suffers from adding a large diagonal term into the perturbation; (2) MBPT(2) is not
bounded from below and any effort to use the variational condition 6E is likely to suffer

from numerical problems.

Our approach is to correct the above by changing the choice of Ho and focusing on

the expectation of the density operator instead of any variational functional derivation.

The partitioning of the hamiltonian is chosen to be the one-body part as the zeroth-order


a, 3
Ho0=^ fiI{aIa+, } + Y fbbatabh} + fij{aaa} + f abab aataa} (4-6)
a i b i'j a4b

However, to avoid dealing with non-diagonal Ho, a semi-canonical transformation will be

performed to obtain the more convenient zeroth-order hamiltonian,

Ho fpp{ ap} (4 7)
( p

Dropping the ~ for simplicity, with the semi-canonical partitioning of the hamiltonian the

OEP exchange and correlation potentials assume the following form

V(r) -2 dr1 < ijlja > i(rl)a(r)( l(, (4-8)
Sijc fiir faa 1(

V, f Jdr l(25 < 1,jTlabr > (< aUbTlc,jT > < aobrljTc, >)(Pi(ril)(Pco(rl)
i,j,abc( + fjj faac fbbr)(fia fc)
S i,j,a,b,3
_- K < ij|Ta UbT > (< ijlkbT > < jiklb >)(Pkc-(Fl)oPao(1)
-2y Y < > (< >
Si,j,k,a,b iia jjf r faaa fbbT)(fkkc faaa)

-, K < ij, iab > (< aUbIkj, > < braUIkj >)*kaC. (1)i(r1)
Sijk,,b fii ( + fjj faaa fbbr)(fkk + fjjr faaa fbbr)

quite stable, and does not suffer from the problems alluded to by Staroverov,, even

when the auxiliary basis has a dimension greater than the MO basis.

In another interesting example, Rohr, et al[49] observes the occurrence of a degen-

eracy in performing OEP calculations, and to investigate this, intentionally construct a

problem where the HOMO and LUMO orbitals of the KS problem are degenerate. Subject

to direct minimization, they proceed to report on the failure of the OEP procedure. On

the other hand, it is shown in this paper that the original X based approach handles this

problem, too.

The exchange-only time-dependent OEP has been considered by Goerling[20] and

implemented by Hirata et. al[21, 50]. However, all results to date are for closed-shell

systems. To further address the exact local (OEPx) exchange versus time-dependent

Hartree-Fock (TDHF), TDOEPx is generalized to treat excited states for open-shell

species. Results from adiabatic TDOEPx are in good agreement with TDHF for both

excitation energies and polarizabilities, however, it is shown that charge-transfer states

cannot be properly described in TDDFT. For standard DFT methods, gradient-corrected,

hybrid, etc. this has been noted, and improved upon[51], but standard methods, unlike

TDOEPx, still suffer from other limitations, like the self-interaction error and the incorrect

long-range behavior of the potentials and kernels. In an approach that gives the 'right

answer for the right reason' these exact conditions are requisite. In TDOEPx there are no

such errors. Hence, failures of TDOEPx compared to TDHF have to be exclusively due to

the local versus non-local exchange operator, which is demonstrated.

Finally, C6 coefficients are obtained for open-shell systems from the TDOEPx

freu iiv.- -dependent polarizability.

5.1 Exact-Exchange Density Functional Theory

The most satisfactory way to introduce exact exchange DFT is to insist that

the Kohn-Sham single determinant provide the exact density, but subject to a local,

multiplicative exchange operator, Vx(1), instead of the usual non-local Hartree-Fock

When we need to calculate polarizabilities at frequency u, the following system of

linear equations (5-40) should be solved

(A A B d"- h
h) (540)
B A + ul d`+ h


hpq = / dr(pp,(r)ipyq,(r) (5-41)

Once these equations are solved for d, the dynamic polarizability is readily evaluated from
a, 3
a-(w; +w) 1- Z (h jh d + h,,d) (5 42)
i,a a

For the calculation of the C6 van der Waals coefficients, the system of equations (5-40)

should be solved for the imaginary frequencies iu from which the coefficients can by

calculated by
3 r[-
C/0 \ dwa(iw)a(iw) (5-43)
T Jo
where a(iw) = ((awxx ) + ,yy(iw) + a (iL)). Integration of equation (5-43) was carried

out by Gauss-('l. li--!i. v quadrature.

To ensure the correct ..i- mptotic behavior of the exchange potential, the

Colle-N. -bet algorithm[30] was used. As a seed potential, we used the one proposed by

,i ,, i [: ]

V later(T) jo dr (5 44)
,j P, Ir rll

For the spin-polarized case two different Slater potentials are used for the corresponding

spin component of the local exchange.

5.2.2 Numerical Results

Excitation energies and polarizabilities are calculated for several open-shell systems.

Equilibrium geometries are taken from ref[36]. For all calculations the uncontracted


The use of the Kohn-Sham orbitals and orbital energies to construct an implicit

density-dependent energy functional using,e.g perturbation theory[12, 13] is a straightfor-

ward approach to the ab initio DFT. The corresponding Kohn-Sham potential can then be

obtained by taking the functional derivative of the finite-order energy via the chain rule,

that transforms the derivative with respect to the density into the derivative with respect

to orbitals and orbital energies[14].This leads to the Optimized Effective Potential[13, 15]

equations which can be complicated already in second order[13]

Alternatively, one can determine the first and second order Kohn-Sham potential

by requiring that the corresponding first and second order perturbative corrections to the

reference density vanish.Unlike the functional derivative approach such a condition on

the density can be described with standard rii ii:-body techniques.The recent work[17]

uses diagrams to derive the second-order OEP equation in a systematic and compact

fashion,while a second paper [16] does so algebraically.

Both the functional derivative and density(effective operator) approaches lead to

exactly the same equation in the first order[17].However,the functional derivative of the

second energy involves a certain type of denominators that are not present in the density

condition(effective operator approach).Still the terms involving such denominators can be

transformed to match the density-based equation exactly. This caveat raises the question

of whether the two approaches are equivalent in higher orders,or for different partitionings

of the hamiltonian.

2.1 Equations for the Exchange-Correlation Potential in the Functional
Derivative Approach

To derive the equations in both approaches we first split the full electronic hamilto-

nian into a perturbation and a zero-order hamiltonian:

H Ho + V (21)

[40] J. P. Perdew, R. G. Parr, M. Levy and J. L. Balduz, Jr., Phys. Rev. Lett. 49, 1691
[41] A. Balkova, R. J. Bartlett, J. ('C!. i, Phys. 102,7116 (1995).
[42] Ch. W. Bauschliher, Jr. and P. R. Taylor, J. ('C!. i, Phys. 86, 2844 (1987).

[43] J. Schirmer, A. Dreuw, Phys. Rev. A 75 022513 (2007).
[44] J. B. Krieger, Y. Li and J. Iafrate, Phys. Rev. A 45 101 (1992).

[45] V. N. Staroverov, G. Scuseria, E. R. Davidson, J. C'!. in Phys. 124 141103 (2006).
[46] W. Yang and Q. Wu, Phys. Rev. Lett, 89 143002 (2002).

[47] A. G6rling, Phys. Rev. Lett, 83, 5459 (1999).
[48] R. J. Bartlett, I. V. Schweigert and V. F. Lotrich, J. Mol. Struct: Theochem 771, 1

[49] D. Rohr, O. Gritsenko, E. J. Baerends, C'!, in Phys. Lett. 432 336 (2006).
[50] S. Hirata, S. Ivanov, R. J. Bartlett, I. Grabowski, Phys. Rev. A 71, 032507 (2005).
[51] A. Dreuw, J. L. Weisman and M. Head Gordon, J. ('!. in, Phys. 119,2943 (2003).
[52] S. Hirata, S. Ivanov, I. Grabowski, R. J. Bartlett, K. Burke, J. D. Talman, J. ('!, 1'1
Phys. 115, 1635 (2001).

[53] P. O. L6wdin, Phys. Rev. 139, A357 A372 (1965).
[54] S. Ivanov, M. Levy, J. ('C!. in Phys. 119, 7087 (2003).

[55] H. Jiang, E. Engel, J. ('C!. i, Phys. 125 184108 (2006).
[56] 0. V. Gritsenko, E. J. Baerends, Phys. Rev. A, 64 042506 (2001).

[57] S. Ivanov, S. Hirata, I. Grabowskii, R. J. Bartlett, J. ('!., in Phys. 118 461 (2003).
[58] A. F. Izmaylov, V. N. Staroverov, G. E. Scuseria, E. R. Davidson, G. Stoltz and E.
Cances, J. ('!., ii Phys. 126, 084107 (2007).

[59] S. Kiimmel and J. P. Perdew, Phys. Rev. Lett, 90 043004 (2003).
[60] I. Schweigert, V. F. Lotrich and R. J. Bartlett, J. ('C!. ii Phys, 125 104108 (2006); I.
Schweigert, Ph. D. thesis, University of Florida (2005).
[61] R. J. Bartlett and M. Musial, Rev. Mod. Phys, 79 291 (2007).
[62] D. Bokhan, R. J. Bartlett, Phys. Rev. A, 73 022502 (2006).
[63] D. Tozer, J. Chem. Phys. 119 12697 (2003).


The time-dependent OEP exchange-only method in the adiabatic approximation

has been previously derived [20] and implemented for molecules by Hirata et. al [21].

Numerical results in that work show a reasonable description of both valence and Rydberg

excited states, partly due to the correct .,-i-!!, ,l'tic behavior of the exchange potential,

but also due to the elimination of the self-interaction error. In particular, OEP-TDDFT

is superior to standard approaches like TDDFT based upon local density approximation

(LDA) or Becke-Lee-Yang-Parr (BLYP) functionals. Similarly, exchange-only OEP with

exact local exchange (EXX)[22, 23] has been shown to greatly improve band-gaps in

polymers [23]

Another advantage of OEP based methods is that since virtual orbitals in the

exchange-only DFT as well as occupied orbitals are generated by a local potential,

which corresponds to the N-particle system, the differences between orbital energies of

virtual and occupied orbitals offer a good zeroth-order approximation to the excitation

energies[16, 17].This is not possible in the case of Hartree-Fock theory, where occupied

orbitals are generated by an N-l particle potential and the energies of unoccupied orbitals

come from N-electron potential, and thereby, approximate electron affinities.

Once OEP correlation is added [13, 17], the essential new element in the time-

dependent DFT scheme is the exchange-correlation kernel, which in the adiabatic approx-

imation is defined as a functional derivative of the exchange-correlation potential with

respect to the d, i,-il [20]:

Hence, in this paper we will derive the kernel for the OEP-MBPT(2) correlation potential,

when we use the standard KS Ho, Ho = i hks(i). In prior papers[16, 17] this has

been called OEP Kohn-Sham (KS) to distinguish it from other choices for Ho. It is

The matrix elements Fq, and e0, can be calculated from the relations (6 6)

F\ = h (AP + BRPq) UrA
q, pq\aiu iq a

6.1.2 Diagrammatic Derivation of the Second Exact-Exchange Kernel
For the derivation of the second kernel it is convenient to use the form (6-8)

g9(ri, r2, r3)

6fxc(r, r2)

Which can be rewritten as

f(r,=r2) dr4dr5 6X-- (rr4) Qr 5)X (r5,r2)+

Sdr4dr5X- (rI r4) 6Q(4, r5) X- (r5, 2)

Jdr4dr5X l(rr) (4,r5) 6p(r3)

Taking into account the fact that

6X-\(r^_, r2)6X(rs, r4 _
(rl -r2) dr4drf X (r, r) ( 4)X '(r2, 4)
bp(r3) V,(rs)

and doing some simple transformations, it is possible to write

g(rl, r2, r3)

drdr7dr8sf(r2, rs8)X- (r3,6) X(r, r) X (r, r7) +
J 6Vs,(r6)
dr4dr5dr6X-(rl, r4)X- (r3, r6)Q(4,r5)X (r2,r5)
r ir (Vs(r6)
dr6dr7dr8f (ri, r7)X- (r3, r6) 6X(r, r) X-(r2, 8)
j 6bV,(r6)



Sdr' p() + V(r)


(6 7)




Vs(r) = ve.t(r) +


of self-interaction is assured. The correct .-i-'!.' ictic behavior is achieved using the

Colle-N. -bet algorithm[30].

When we solve the TDDFT equations with the kernel developed in this paper,

we obtain the results for excitation energies shown in Table 3-2. In general, just as we

observed from the orbital energy differences, the OEP-MBPT(2) results tend to fall on the

low side of experiment and the OEPx on the high side. In fact, except for the excitation

energy for the lowest 1P state, where the very low orbital energy of -7.58 biases the

results, an average of the two would seem to be about right. There are, however, still

difficiences in the basis set, as seen by the EOM-CCSD results compared to experiment,

where besides the additional correlation effects introduced by EOM-CCSDT[31], the

further extension of the diffuse functions would remove the 1 eV error in the highest lying

(1S) state. Of course, the dependence of TDDFT on the basis and that for a two-particle

theory like EOM-CC should be quite different.

The great sensitivity of the results to the orbital energies from the underlying KS-

DFT calculation can be further appreciated by simply taking the energies from the OEPx

results and using them in the evaluation of the kernel and the matrix elements in the

TDDFT equations. These results are shown in Table 3-3. The near coincidence of results

for OEPx and OEP-MBPT(2) is apparent, with both now being too high.

A similar experiment can be made where we use orbitals and orbital energies from

OEP2(sc) ab into dft results as described elsewhere[16], to obtain the results in Table

3-4. That is the TDDFT equations and the kernel are assumed to be the same, but we

use orbitals and orbital energies obtained from OEP2(sc). Here we also show the results

from standard TDDFT applications using the LDA and B3LYP functionals for comparison

purposes. Clearly, we have improved results at both the OEPx(sc) and OEP2(sc) levels,

with the same pattern of the former being too high, but less so; and the latter, too low,

but better than before. It is apparent that the proper way to achieve the benefits of

At t = to the perturbation is applied so that the total potential is given by

v(r,t) '= (r) + vi(r,t) (1-53)

where vi(r,t)=0 for t < to.
The objective is to calculate the linear density response pi(r, t) to the perturbation

vi(r, t). Conventionally, pi(r, t) is computed from the full linear response function X as

pi(r,t) dr dt'x(r, t,r',t')vi(r',t') (1-54)

Since the time-dependent Kohn-Sham equations (1-50) provide a formally exact way of
calculating the time-dependent density, it is possible to compute exact density response

pi(r, t) as the response of the non-interacting system

p (r, t) = dr' dtt'XKS (T,, ', r t1) (1-55)

where v~i) (r, t) is the effective time-dependent potential evaluated to first order in the
perturbing potential, i.e.,

v) (r, t) -= v, (r, t) + dr'p t')+ + dr! f dt'f,(r,(t,r', t')pl(r',, t') (1-56)

The exchange-correlation kernel fc, is given by the functional derivative of v,,

f(r, r', t') v[P]) (1-57)

at p = po.
While the full response function X is very hard to calculate, the non-interacting

XKS can be computed fairly easily. In terms of the static Kohn-Sham orbitals the Fourier
transform of XKS(, t, T t') with respect to (t t') can be expressed as
cKS ( -r) W W1 ()-58)
XKS(F, r', w) limg a w a (158)
St-( a -(i ,

Table 5-6. Static polarizabilities (in a. u.)


















aZ 6.24

Table 5-7. Isotropic (
CN ... CN 93.A
CH3 ... CH3 94.:
N ... N 19.:

6.23 6.25 5.94

76 coefficients (in a. u.)
37 92.06
27 93.40
28 19.21

ROOS-ADZP basis set[27] has been used. Results for excitation energies are presented in

Table 5-3.

Table 5-3. Excitation energies (V valence state, R Rydberg)
2I(V) 4.18 8.16 4.91 1.96 1.52 1.32
2Z+(V) 5.47 10.58 5.48 3.19 3.62 3.22
2II(V) 7.55 9.98 8.23 3.03 3.43 3.26
2Z+(V) 11.14 9.37 10.82 4.99 6.14 5.82
2A' (R) 6.52 6.53 6.42 5.00 5.88 5.72
2A'2 (R) 7.93 7.97 7.93 5.93 7.18 7.44
4P (V) 9.84 11.11 9.88 10.98 10.84 10.35
4P (R) 13.59 13.16 13.18 11.76 1 13.62

As follows from the first two columns of Table 5-3, TDHF and TDOEPx produce

results of approximately the same accuracy, as would be expected by the fact that the

local exchange in TDOEP is meant to be a kind of least-squares fit to the HF non-local

exchange potential. This certainly makes the occupied orbitals quite similar, but the

spectrum of unoccupied orbital eigenvalues is very different, as shown in Table 5-4, since

TDOEPx will generate a Rydberg type series instead of anything like the HF virtual.

As is well known, the latter are determined in a potential of n-electrons, making them

appropriate for electron attached states, while the occupied ones feel a potential of n-

1 electrons. To the contrary, the orbitals obtained in OEPx have the same potential

for an electron in the occupied and unoccupied orbitals, which is why the latter more

nearly simulate Rydberg states, as some of the unoccupied orbitals will have negative

orbital energies. Of course, in a finite basis as long as the space separately spanned by

the occupied and the unoccupied orbitals is the same, there would be no difference in the

results, though the diagonal values (but not the trace) would change. The significant

1 This Rydberg state was not obtained in the EOM CCSD Davidson diagonalization

(VC12- 10)
S fibfbj(< iaA >< jap >
(Eb Eb)(E~ Ea)(E Ea)

(VC12- 11) + (VC15- 6)
4 fib < aclbj > (< iaA >< cjp > + < cjA >< iap >)
ab,c,i,j (Ei Eb)(EFi- Ea) (E Ec)

(VC12 12) + (VC15 1)
4 E fab < bclij > (< iaA >< cjp > + < cjA >< iap >)
a(Eij (- ~b) (Ei Fa) (E' Ec)

2 fab < jb\iC

2 fab < cblij

-2 fia < b a

Sb fia < bcja

(VC12 13) + (VC13 -
iaA >< cjpi > + < cjA ><
- Eb)(Ei Ea)(Ej Ec)

(VC12 14) + (VC14
iaA >< cjpi > + < cjA ><
- Eb) (Ei Ea)(j Ec)

(VC12 15) + (VC14 -
ibA >< cji > + < cjA ><
- Eb)(Ei Ea)(Ej Ec)

(VC12 16) + (VC13
ibA >< cjp > + < cjA ><
- Eb)(Ei Ea)(Ej Ec)

iap >)

iap >)

ibp >)

ibp >)

(VC13 1) + (VC13 2) =
S fai < bilaj > (< bkA >< kjp > + < kjA >< bkp >)
i,j,k,a,b (i a)(E Eb)(E Eb)









> (<

> (<



and frequently diverges. Since the density condition approach is completely equivalent [37]

to the variational OEP-MBPT(2) for the KS partitioning it has the same problems.

However, in the case of the non-variational, semi-canonical density condition approach, the

unboundness from below is less of a problem. The semi-canonical choice of Ho provides

good approximations to the energy and wave function in MBPT(2) and the corresponding

total energies are much closer to highly-accurate CCSD(T) ones, at least in the chosen

basis. The computational cost of the OEP-MBPT(2) method is comparable with the cost

of MBPT(2), the scaling of both methods is N5, where N is the number of basis functions.

This is more expensive than conventional DFT (scales like N3), but less expensive then

CCSD(T), computational time of which is proportional to N7. The correlation potential,

Table 4-1. Total energies
N2 -109.449149 -109.7,11 11, -109.457779 -109.457071 -109.468618
N+ -108. :-. ;;, no conv -108.905444 -108.890773 -108.900556
N -54.544740 -54.593111 -54.545199 -54.535569 -54.5 !-,4
H20 -76.370003 -76.510744 -76.373092 -76.369991 -76 ;"7
H20+ -75.901155 -75.989525 -75.902958 -75.912146 -75.921765
CN -92.598196 no cony -92.651919 -92.646898 -92.658287
CN- -92.772714 no cony -92.780718 -92.783715 -92.796498
CO -113.228520 -113.511203 -113.237817 -113.239593 -113.251239
CO+ -112.702208 no conv -112.734813 -112.729947 -112.739892
02 -150.205776 no conv -150.221870 -150.255051 -150.221560
Ne -128.859598 -128.951437 -128.861175 -12.> .-.844 -1 '.- -,133
Ne+ -128.066796 no conv -128.067574 -128.068711 1- NI IIII'
Mean error, 58.22 26.74 36.16

calculated with the semi-canonical OEP for the Ne atom[16] is close to the corresponding

quantum Monte-Carlo potential, and we can expect that for other systems, where QMC is

not available, semi-canonical OEP-MBPT(2) potentials should be a good alternative. For

the open-shell case potentials produced by Kohn-Sham OEP show an overestimation of

the correlation energy, the same situation previously reported for the closed-shell case[16].

(Fig 4-1 and 4-2). The correct -1 long-range ..-imptotic behavior of the OEP-MBPT(2)

The Hartree-Fock exchange functional provides an exchange part of orbital-

dependent exchange-correlation potentials. Kohn-Sham DFT with OEP exchange

potentials provides a superior results to all known density-dependent exchange func-

tionals. To make OEP potentials useful for practical calculations it is necessary to have

the correlation potential, which can be combined with it. Conventional GGA correlation

functionals are developed together with the corresponding exchange potentials, which are

very inaccurate. Because of this correlation potential of GGA compensates errors in the

exchange part. In most cases GGA correlation potentials have an opposite sign to a nearly

exact QMC potentials. The correcting terms incorporated into correlation potentials,

so it is really difficult to extract the "real correlation p oii So far, it is not surprising

that combining GGA correlation potentials with OEP exchange potentials the associated

results are inferior to exchange-only cases. Since conventional correlation functionals are

not compatible with OEP, the development of orbital-dependent correlation functionals is


Ab initio DFT uses orbital-dependent correlation energy functionals from the

rigorous wave-function methods. Since corresponding correlation potentials are derived

from ab initio methods, they are compatible with OEP exchange potentials. Unlike

conventional ones, ab initio functionals are systematically improvable, since one can ah-i--,

use a higher-level approximation to obtain a more accurate functional. They also have a

well-defined limit represented by the FCI method.

1.2.5 Ab Initio Density Functional Theory

The main advantage of orbital-dependent energy functionals is the fact, that

such functionals are known in analytical form. Particularly, the exchange functional can be

written in following way

E < Vee >-EH < ;.;,; > (167)

The expression, which is suppose be differentiated with respect to the potential on the

right-hand side of equation (3-10) can be represented by diagram (V) (3-11)

i a

V (n)

(V) (3-11)

The functional derivatives from diagram (V) produce the set of diagrams (V1)-(V6) of set


16 6 2 6 1 6 2 6
b 1 b 2 1 2b
I a (n
S 2 1 a 2 V a V_ I
V a V (n) a V (n) I V (n
xc xc xc xc
(VI) (V2) (V3) (V4) (V5) (V6) (3-12)

During the derivation of diagrams (V1)-(V4), diagrams containing "bracket-type" denom-

inators will appear, because one of the steps in the diagrammatic rules of differentiation

requires detaching the unoccupied line from diagram V and inserting the last two parts of

the 6- function (2-26). Such diagrams can be transformed into a set of regular diagrams,

using the diagrammatic relation (313). This relation shows us that for the transformation

to regular diagrams it is necessary to double each of the denominator lines and insert the

last two members of the 6-function (2-26), subject to the restricted summation, I / k.We

use the same procedure for the case of occupied orbitals.

Sa > 1 1
2 2 2> 2
k 2 > k k
+ :k + + + (Dl)

m .

...... MBPT(2)
------- OEP-semi
------- CCSD
-- CCSD(T)

1.0 1.5 2.0

Figure 4-5. HF potential energy curve.

-99.3 -

-99.4 -

-99.5 -

-99.6 -

-99.7 -

-99.8 -

-99.9 -







two diagrams gives the diagrams of equation (2-34)

Si j J

2 +2 (2-34)

On the two diagrams we impose the restriction i / k. To make a direct correspondence

with diagrams 7 and 11 we need to add diagrams that arise from taking functional

derivatives from denominators (2-35)

k b j k a b /
2 2 (2-35)

After these diagrams are added to the previous two, we will have a direct correspondence

with diagrams 7 and 11. When functional derivatives are taken with respect to the

unoccupied orbitals and an unoccupied line inserted, the same procedure will give a direct

correspondence with diagrams 6 and 10 of equation (2-25).

Now consider the first diagram for the second-order energy expression. The pro-

cedure for taking functional derivatives from the right side of the vertex, which do not

contain a closed F-ring are absolutely the same as in the case of other second and third

diagrams in the energy expression. When such functional derivatives are taken, we have

four diagrams corresponding to diagrams 12-15. To take functional derivatives from the

F-rings, the first diagram in the energy expression must be represented with more detail,

taking into account that f = h + J K,in following the form (2-36)

CF X + X (2 3

F X+ )+ + 2 X +2 +2 (2-36)

) 2007 Denis Bokhan

tb2 b)< ik\j b > *F9
i,j,k,a,b F b) -
F9 = (< ,-,,-, i.> >< pbpjrl >< ia > + < ',, i.K >< pbpj >< Cjac] > +

< ,, > >< b~ >< > + < ',- r, q >< b(6jK >< (OiOa > +

< '- >< j>< b( >< (pi(p > + < 'r, r-,7 q >< ib Pj >< (ri(baK >) (B-16)

D12 --2Z (< Papc \bcj > *F10
D12 -2
(Ei Ec) E( Eb (Ei Ea)
i,j,a,b,c c)( b)(i a)
F10 (< >, ,v-. >< cpipr] >< Cicga( > + < ><, .. >< ci< c >< aGil > +

< .< >< OcrX >< aOiaK > + < >' < ,* K>< c >< a( i(a> +

< ,- ,- >< cj >< ( i(al > + < '- 17 >< ><< >( i(aK >) (B-17)

D13 -2 << =b-a > *F11
D %:(E Ec) E_ Eb_)(i_ E)
i,j,a,b,c ( i
F11 = (< cp >< >< bj( > + < cK ac( >< > bjl > +

< iPc( >< OaOc)l >< Vb(PjK> + < icl >< acK b>< Vj > +

< i< acK >< Xb~Pj1l > + < Ol >< ac( >< Vb(jK >) (B-18)

D14 2 > < ,- ,.'~ ba > *F12
ijkab k Ea) (E Eb)(Ei Ea)
F12 = (< ickKX >< cackl >< ,, .< >+< kK Ci~ca>< O c k >< ',.>+

< ik( Cl > < ',OaOk .i.' > + < OOkl >< OaOk > 'K ..S > +

< >ik( >< OaOkK >< ,-. > + < cick X >< Oa(Ok >< '- .- >) (B-19)

To get final expression for BK all terms D1 D14 should be summed up.

After the transformation of irregular diagrams and adding the functional derivatives from

the denominator of diagram (V), diagrams (V1)-(V4) will appear. Using diagrammatic

expressions for the first member of equation (3-9) and diagrams (V1)-(V6) it is possible to

build the exchange-correlation kernels to any order.

3.1.2 An Example: Diagrammatic Derivation of Exchange-Only Kernel

The kernel for the exchange-only case was initial derived by G6rling [20], and then

rederived and initial implemented by Hirata et. al [21]. Such a derivation requires much

effort. Here we offer a facile derivation with the diagrammatic formalism.

The exchange energy (Ex=-~E~ j < ij ji >) can be represented by diagram (3-14).

E, (EX) (3-14)

After taking the first functional derivative with respect to V we will have the diagram


E,/6V, = 2

(VX) (3 15)

After that the diagrammatic rules must be applied one more time to get the second

functional derivative, as is necessary according to equation (3-9). After taking the second

functional derivatives we will have diagrams (FX1)-(FX8) of the set (3-16)

i a 1 2 1 2 aa 1 1 2 i -
i a b 1 1 i b b 1 1 i a

62 62 4 _
V- k k kj j k J b j
(FX1) (FX2) (FX3) (FX4) (FX5) (FX6) (FX7) 2 (FX8) (3-16)

After the addition of diagrams (V1)-(V6) we will have all the diagrams necessary for

building the exchange-only kernel. During the interpretation of diagrams (V1)-(V6) the

(VC7 4) + (VC8 5)
< aclij >< kilab > (< bcA >< kjp > + < bcp >< kjA >) (A62)
a(E E+ Ea EF)(Ei E+ Ea Eb)(EF Ek Ea Eb)

< aclij >< kilab > (< bcA >< kjp > + < bc >< kjA >) (A63)
a,b,c,i,j,k F F F b)

(VC7 6) + (VC8 7)
< aclij >< jklab > (< bcA >< kip > + < bc >< kiA >) (A64)
,ci,j,k (i E, E- )(Ej + k Ea E,)(cj + k -a b)

(VC7 7) + (VC8 4) -
< aclij >< jklab > (< bcA >< kip > + < bc >< kiA >) (A65)
ab,,,j,k ( + aj a )( + j a b)( -j + Ek Ea Eb)

(VC7 ) + (VC7 1)
< aclij >< jilad > (< bcA >< bdp > + < bdA >< bcp >)66)
,b,,j Ea Ec)( ( + Ea Eb)(Ei E+ a Ed)

(VC7 13) + (VC7- 14)
< aclij >< jilab >< bkA >< ckp >
(Ec, i + j a Ec)(i + j Fa Fb)(Ek Ec)

(VC8 2) + (VC8 3)
< ablik >< ljdba > (< jkA >< ilp > + < ilA >< jki >)
(A 68)
,,,j,k, (i + k Ea Eb)(i + j -a Eb)(EF +E Ea sb)

(VC8- 1) + (VC8- 2)
< ablik >< ijlba > (< klA >< jli > + < jlA >< kl/ >)
S+ j ) ( + )( + k -(A 6))
a,b,,,i, ja )a

(VC3 9) + (VC 10)
2 < ablik >< ijlab > (< jlA >< Ik > + < jl >< IkA >) (A30)
b,i,k, (i + k EF -- b)(Ei +l Fa -- b)(Ei + j Ea -- b)

(VC3 11) + (VC4 12)
-2 < ablik >< ijlac > (< jkA >< bcl > + < bcA >< jkp >) (A31)
,b,ci,j,k (C k Ea Eb)(CF C -a Ec)(Eci + c- Fa Eb)

(VC3 12) + (VC4 11)
-2 < ablik >< ijlac > (< jkA >< bc > + < bcA >< jk >) (A32)
(E,b,c,,,k i + Ek Ea Eb)(Ei + k -a Ec)(Ei + Ea Ec)

(VC3- 13) + (VC3 14)
< ablik >< ijlab >< jcA >< kcp >
a, + Ek Ea Eb)(i E+ Ea Eb)(Ek Ec)

(VC4- 4) + (VC4 6)
< aciij >< ijtdb > (< cbA >< adp > + < adA >< cbi >) (A34)
ab,c,d,i,j + Ea Ec)(i Ea Eb)(Ei Ed Eb)

(VC4 9) + (VC4 10)
2 < aclij >< ijad > (< cb >< bd > + < cb >< bd >) (A35)
,,c,,, (E + Fa a c)(E + Fa Fb)( + a b ( a d)
a,b,c,d,i, j \ c b\z a )

(VC4 13) + (VC4 14)

(A 36)
-4 < ac ij >< ijjab >< bkA >< kci > (A36)
S (i e+ C a c)(Ei + j a -Eb)(E c)

(VC5- 1) + (VC6 -1)
2 << ablij >< cjlbk > (< icA >< ak> > + < icl >< akA >)
2 (+ )(- ) -(A)37)

Our preference in the development of such orbital dependent functionals and

potentials is to start from the density condition that is fundamental to KS theory, as the

single determinant must provide the correct density for the exchange-correlation problem.

Once this is enforced for a given functional, Vxc is uniquely determined up to a constant.

This is the origin of the perturbation method of Goerling and Levy[14], but essentially

modified by Bartlett et al [16, 17] to avoid the failures of a simple sum of KS one-particle

Hamiltonians as the unperturbed problem, to generate correlation potentials. This is the

cornerstone of ab initio dft. The conceptual difference is that the density condition does

not explicitly use the variational determination, 6Exc/6p(1) = Vxc(1). This difference is

of critical importance in generating correlation potentials from low-orders of perturbation

theory. Without that change, no low-order orbital dependent correlation functional like

that from MBPT2 will generally work, but with those changes, it does very well [16, 48].

Those modifications to the perturbation theory also pertain to the exchange-only case, the

subject of this paper, but for that problem the distinctions are less important[16].

Once following this approach, which starts with the KS choice of Ho, the distinctions

between applying the density condition and using direct functional differentiation is

more conceptual than essential, as there is a correspondence in any order of perturbation

theory[24, 37].

The direct optimization procedure advocated by some, built upon the above varia-

tional determination of the functional derivative, though formally equivalent, differs from

the original, X-based OEP in the details of implementation. However, as was pointed

out by Staroverov et. al [45] under certain combinations of molecular and auxiliary basis

sets, particularly when the latter's dimension is larger than that for the former, the direct

optimization method can give the Hartree-Fock energy and density. The latter can be

viewed as a trivial solution, as it can be shown to correspond to the solution of a weighted

least-squares expression. On the other hand, the X-based OEP with proper handling is

(VC18) (V
(VC1-8) (VC1-9)


(VC1-11) (VC1-12) (VC1-13)

(VC1-14) (VC1-15)



Differentiation of diagram (VC2) produce diagrams (VC2-1)-(VC2-16) of set (3-21). These

diagrams have the same skeleton structure, but different positions of indices.

k ba ja a

i(VC2-) (C2-2) (C2
(VC2-11 (VC2-21 (VC2-3)

k 01 0- 0 1

k ba i b
i( i 6 ( c 1 2 C
2 V 2 2
(VC2-4) (VC2-5) (VC2-6) (VC2-7) (VC2-8)

a(C b ja(V b

(VC2-9) (VC2-10)

(VC2-11) (VC2-12) (VC2-13) (VC2-14) (VC2-15)

(VC2-16) (3-21)

After taking the functional derivatives from diagram (VC3) diagrams (VC3-1) (VC3-14)

(3-22) will appear.

2 2

2 a a bj a
2 i j i
a k 1b1 i1 a
k -tik kk

i a ia
b 6 b 1 ) b\ 2a
k k a k

(VC3-1) (VC3-2) (VC3-3) (VC3-4) (VC3-5) (VC3-6) (VC3-7)
1 a a 1 na a 5 a
2 ia i' b b Jb b t b b/ b
k i k i k i V k i k iV k k 2

(VC3-8) (VC3-9) (VC3-10) (VC3-11) (VC3-12) (VC3-13)

2 k

G1c k

cib k

&-k c

(VC3-14) (3-22)

TDHF ones. For both excitation energies and polarizabilities this closeness arises from the

fact that the OEP potential derived from the HF energy functional and its time-dependent

response properties are similar. Systematic overestimation of polarizabilities is also caused

by the lack of the -1 ..i-mptotic dependence. In most cases this causes an even larger

error than that coming from the absence of correlation effects in TDHF and TDOEP. The

anisotropy of the polarizability, though, is usually reproduced much better by SVWN.

This fact must be due to cancellation of errors, coming from the incorrect .--i,.iinl. l ic and

incomplete cancellation of self-interaction error, plus a proper accounting of correlation


5.2.3 Charge-Transfer Excited States

Within the framework of the standard TDDFT description of excited states the

transfer of charge between two separated fragments has severe difficulties. Detailed

analysis of this problem has been presented by Tozer[63] and Dreuw et al[51]. According

to the latter analysis [51] even if the exact exchange-correlation functional were known, it

would still be impossible to get a proper description of charge-transfer excitations within

TDDFT. Consider the example of He ... Be. The charge-transfer 'I state calculated with

TDHF and TDOEP, taken with the Tamm-Dancoff (mono-excited CI) approximation for

simplicity. For both atoms the uncontracted ROOS-ADZP basis set was used. Potential

curves of the 'I charge-transfer excited state are presented on Fig 5-2. The potential

curve, calculated with the configuration interaction singles method exhibits the correct

long-range ..i-mptotic behavior, while the TDOEP results have a qualitatively wrong

.i',i!1 ill. ic behavior. This phenomena is caused by the fact that for charge-transfer

excitations all integrals in Eqn. (5-36) are equal to zero and the excitation frequency is

equal to the difference of the LUMO orbital energy of the Be atom and the HOMO energy

of the He atom.

What was not addressed numerically in Dreuw et al[51] was the effect of having

a proper self-interaction and .i-', i-i, ll ic behavior in DFT, which can only be achieved

potential vertex should be the matrix element < pp\ Vlpq > and a factor -2 must be added

according to equation (3-10). Interpretation of diagrams (FX1)-(FX8) and (V1)-(V6)

gives us the expression (3-17)

h(r3, 6 -2

< ialjb > (i(3)(ao(r3)(j (r6)L(b(r6)
(Ei Ea)(E. Eb)

2 > l< ij ba > i(r3) oa(r3) j (r6) b(r6)
ij, a, -F)(E b)
+2 y < jk|ka > [pj(r3) i(r3) i(r6) a(r6) + j (r6)(P(6)(i(r3) o(r3)]
ij,k,a (- Fa)(- Fa)
2 y K ai|j *I > [ao(r3)b(r3)-Eb(r6)aj(6) + ( Pa(r6)b(rF6)(cb(r3)i0(3)1
i,j,a,b ( i a
< iklkj > (PF3)La(r3)3j (r6)ao(r6)
,j,k,a (- )(- a)
-2 y < aj jb > Pi(F3)a(r3) (i 6)b(r6)
(Ei a)(Ei b)
i, j, a,b
-2 C < iy ll > i((3 a 3 j (6 + ( ( 6

S< aV lb > y4(r3) a(r3)Pi(r6)ob(r6)
(Ei Ea)(i b)

S [p(r3)fcj(r3)cj (r6)foa(r6) + 6 i6)6j(r 6)(Oj (r3)yOa(F3)1

< i\ [Va > [o(r3)b(r3) b(Tr6)(6T) + Yoa(r6)Yb(r6) b b(r3)3i(r3)1
(E- Ea)( Eb)


Expression (3-17) for h(r3, r6) exactly corresponds to the expression, obtained by Hirata

et. al, but the diagrammatic derivation requires far less effort and is unambiguous in

terms of signs and numerical factors. Since all the diagrams contain only one contour, it

is possible to make summations in equation (3-17) only over spatial orbitals, and, as a

result f,,(rl, r2) and faa(rl, r2) will appear separately. The exchange-only kernel does

not contain the f,3(rl, r2) part, which is a critical difference between the exchange and

correlation kernels.

Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy



Denis Bokhan

August 2007

('C! i: Rodney J. Bartlett
Major: Chemistry

Ab initio density functional theory (DFT) based on the optimized effective potential

(OEP) method is a new approach to study the electronic structure of atomic, molecular

and solid state systems. It contains elements of both wave function and density functional

theories and is free from limitations of conventional DFT because of using orbital-

dependent functionals derived from systematic approximations of the wave function


Ab initio DFT methods with exchange-correlation functionals based on many -

body perturbation theory (MI IPT) have been derived and implemented recently. The

exchange-correlation potentials derived from MBPT have a complicated structure and

their derivation in higher-order of MBPT by the use of the chain-rule for functional

differentiation requires significant effort. To facilitate such derivations, I developed a

special diagrammatic formalism for taking functional derivatives. An alternative way to

construct OEP MBPT exchange-correlation potentials is to use the density condition.

It makes it possible to obtain potentials for different partitionings of the full molecular

Hamiltonian with MBPT. Using the diagrammatic formalism developed for taking

functional derivatives, we show an order-by-order equivalence between the functional

derivative and the density condition approaches to OEP MBPT for the case of the Kohn-

Sham partitioning of the molecular Hamiltonian. For any other partitionings, different

condition to the density condition that would naturally lead to the analog of the kernel

is necessary to correctly use the OEP2(sc) method. Of course, in infinite order methods,

as opposed to perturbative approximations, there can be no difference between the two


3.5 Conclusions

The TDDFT kernel for an OEP method that includes a second-order MBPT(2)

orbital-dependent correlation functional is derived. This derivation is made possible by a

diagrammatic technique familiar from many-body theory. The complexity of the kernel

for MBPT(2) is excessive, resulting in more than 200 diagrams that have to be evaluated.

This is the price paid to retain the one-particle structure of DFT and TDDFT yet

introduce rigorous, orbital-dependent correlation functionals even at second order.

Yet, despite the kernel's complexity it has been evaluated in a large basis set and

illustrated for Ne. As only the KS choice of Ho leads naturally to the kernel as the second

functional derivative of the orbital-dependent functional, ELc, the numerical result can

suffer from the fact that the functional E(0) + E(1) + E(2) is not bounded, and without

modification of Ho as used in [16, 17, 26], the radius of convergence is poor, causing poor

convergence[16] in determining the correlation potential. Nevertheless, the first step in

the underlying framework has now been defined to apply TDDFT with OEP correlation


However, the complexity is great, though imposition of additional rigorous conditions

might result in simplification of the kernel. Lacking such simplifications, the question

arises as whether this is a case of diminishing returns to evaluate some 203 diagrams to

retain the attractive, one-particle, correlated structure of DFT? First, the answers for Ne

are not very good at the OEP-MBPT(2) level. Furthermore, the basis set dependence

of OEP methods when done in gaussian basis is severe to even get the V,, right[16].

The failure of most such OEP calculations to satisfy the exact HOMO condition <

HOMOIKIHOMO > < HOMOIVHOMO > is a case in point[16, 26].


EHF = min < THFIHI|HF > (14)

subject to the condition of orthonormality on the orbitals < ,; i; >= 6ij. Substituting

the explicit form of the wave function from Equation (1-3) into an expectation value, it is

possible to write the Hartree-Fock energy in terms of orbitals

elec Nucl elec
EHF i 2 ZA j > 1 >)

To derive the Hartree-Fock equations it is convenient to minimize the following


I = EHF : ij < > (1-6)
where Eij are Lagrange multipliers. Minimization of the functional from Equation (1-6)

requires that the functional derivatives of I with respect to the orbitals vanish

61 61
S0, 0 (17)
6oi 6(p,

Taking into account Equation (1-3) condition (1-7) can be presented as

fP=pi, F(1-8)

The operator f has the structure

f V2 + =Vt+ VH + nl (1-9)

Let us first consider the choice of Ho as a sum of Kohn-Sham orbital energies:
Ho0 = ,{(aa,} (2-2)

where K is the number of orbitals.The brackets means that the product of second

quantization operators are written in normal order. In the functional derivative method we

define the exchange-correlation potential of the n-th order as:

V (') (r) = (2-3)
v 6p(r)

where E(') is the energy of the n-th order of many-body perturbation theory. In single-

reference many-body perturbation theory we have the energy of the nth order as an

explicit functional of the orbitals.In the G6rling-Levy approach this functional is consid-

ered to be a functional for the orbital energies also [14] :

E(-) E(-)(v c..... p, 1, .....^ )

Another possible consideration[13],when orbital energies are considered to be a functionals

over corresponding orbitals

E ) = E (')(o .....o C ol [ ]..... ,c[.k ]) (2-4)

is equivalent to the G6rling-Levy[14] approach.

To construct the chain rule we should take into account the fact that the Kohn-Sham

orbitals are functionals of V,:

E(') = E(')(,(oV,)..... n(V)) (2-5)

Then the Kohn-Sham potential can be considered to be a functional of p(r):

E() = E (')[I 1(V(p(r))) ..... ,n(V8(p(r)))] (2-6)

v'(r,t) are alv--, different, if

v(r,t) v'(r, t) + c(t) (1-47)

After invertibility for the G mapping is established, the action functional can be

A[p] = dt < W[p](t)i HIW[p](t) > (1-48)

and the variational condition can be applied,

A[p] 0 (149)

Using the same manipulations as for the ground state a set of one-particle equations can
be derived
a (1
p(r, t) (- V2 + ,[p(r, t)])i(r,t) (1-50)
at 2
vs [p(r, t)] =vt(r,t) + Jdr' p(r +. [,1 (r,t) (1-51)
f r -( )

The great advantage of the time-dependent Kohn-Sham scheme lies in its compu-
tational simplicity compared to other electron-correlation methods. The time-dependent
Kohn-Sham scheme with explicit time-dependence of the density and the potential can
be applied for any type of external, time-dependent potentials. However, when the time-
dependent external potential is small it can be treated with time-dependent perturbation
theory. If the applied perturbation is a periodic electromagnetic field, it is more convenient
to use response theory.
1.2.2 Time-Dependent Density Functional Theory Linear Response Theory
Consider the N-electron system being initially, i. e. at t < to in its ground state. In
this case the initial density po can be calculated from the ordinary ground-state Kohn-
Sham equation

(-1 2 + vo(r) d' 0 0(plr)) 0)) +0) ) (r) (1-52)
( 2 | r' I r/ '| 0 O i i )(

The perturbation operator is usually defined as the difference between the full and the
zero-order Hamiltonians
elec elec elec
V = H Ho = t(r) ) + (1-20)
i i 'i

Introducing A as the small perturbation parameter, the Hamiltonian and the wave
function can be written as

H = Ho + XV (1-21)

S= + A(1) + A2(2) + ... (1-22)

E = Eo + + A2AEE + 2E2 ... (1-23)

These order-by-order corrections can be found by substituting expansions (1-21),

(1-22) and (1-23) into (1-16) and collecting terms with the corresponding order in A

(Eo -H fo)() >= (V E ())) > (1-24)

(Eo Ho)(2) >= (V E(1))(1) > -E(2) > (125)

C'!....-ig the perturbative corrections to be orthogonal to the reference determinant
< T(") I) > 0 (so called intermediate normalization) it is easy to get the expressions for

perturbative corrections at any order. Projecting the equations (1-24) and (1-25) onto the

reference determinant expressions for energy corrections are obtained

E(1) =< |IV| > (1-26)

E(2) =< |IVI|(1) > (1-27)

Expressions for the order-by-order expansion of wave functions can be written using

the resolvent operator[4]

|I(1) >= RoVI) > (1-28)

(2) > 0(V E(1)) () > Ro(V E 1))RoVI) > (1-29)

where the c.c means the complex-conjugate. Inserting equation (3-3) into (3-2) we obtain

: S Jdrz4 (J
p,q p r,sr

d 6E(")
dr.5 s (r5) x

x dr6 r X -(r6, r6)+ C.)(r4) x
Sr Es
x dr3 Op(r3 )q- 3X (r2,3) + C.
Sp Eq

This may be rewritten as (3-5)

SJdr3 6 (J dr5 ) (P)(r5) X
p,q4p (r3) Ep(r5)

xJdr6 p (r6)q (r6)X- (rl,r6) + .c)Xl(r2, r3) + c.c
J p c

Equation (3-5) can be made more explicit,

P J 3X (r2,r3
p,q p 6V,(r3


x dr6 P(6)q(76)-]X- (ri, r6) +c.c) + > J
p p,qp

x f E dr )q (r5)
j JOp(r5)

5 q(r5) x
3- p(r5)
dr3X -I(r2, r3) X

d p(r6) q (6) X-1(r, r6)
dF6 + c(r)
Ep Eq 6Vs(r3)

Using the fact that [14]


dr drX rr5) X(5, r6)X-1(r6,2)
Jdr 6Vs(r3)

it is possible to rewrite equation (3-6) in the following form (3-8)

dr3 dr6X-(ri, r3)h(3,6)X-(r6,2)

Taking into account equation(3-3) we have the explicit expression (3-9) for h(r3, r6)

h(r3, r6) V(r3)Y J
p,q p

Edr() p (r6)q (r6)
d (r5 ) r5) q
6()p (r5) Ep Eq

V) ((r5)6X(r5, r6)
dr5 0 6V(rs)

+ c.c




(3 7)

(3 8)

f ()( r2)

f/(n 1, r2)

/f()(rl, 2)



(VC13 3) + (VC13 4)
-2 y fa, < bihaj > (< cjA >< bcp > + < bcA >< cjp >) (A110)
(Ei ,a)(Ej Eb)(Ej Ec)

(VC13 5) + (VC14 9)
2 fai < kilaj > (< bjA >< bkp > + < bkA >< bjp >) (A
i,j,k,a,b (E -a)(j b)(k b)

(VC13 6) + (VC14 6)
-2 fai < bilac > (< cjA >< bjp > + < bjA >< cjp >) (A 112)
(Ei Ea)(E'j Eb) (E- Ec)
a,b,c,i,j ( )()(

(VC13 13) + (VC15 14) =
4 z < aclik >< bilaj > (< bjA >< ckp > + < ckA >< bjp >) (A
a,,(i E,)(Ej Eb)(Ek Ec)

(VC13 14)
S < aklci >< bilaj >< bjA >< ckp > (A )
2 -(A-114)
ab,c,i,j,k (Ei a)( Eb)(E Ec)

(VC13 15) + (VC14 14)
2 < aclki >< bilaj > (< bjA >< ckp > + < ckA >< bjp >(115)
2 ------~ --- i --- i -- ^------ (A 1 1 5 )
ab,c,i,j,k (C E,)(Ej b)(Ek c)

(VC14 7) + (VC14 8) =
2 faj < ablij > (< bkA >< ikp > + < ikA >< bkp >) (A 6)
i,j,,,b ( a) a(i b)(Ek b)

(VC14 11) + (VC14 12) =
-2 faj < ablij > (< icA >< bcp > + < bcA >< icp >) (A117)
b,c,i,j (Ej Ea) (Ei Eb) (Ei Ec)

(VC10 10) + (VC13 9)
-2 fib < iclja > (< abA >< cjp > + < cjA >< abi >)
-2 (A-86)
(Ei- E,)(E Eb)(Ej ,c)

(VC11 1) + (VC11 2) -
2 z fijfaj(< akA >< iki > + < ikA >< aki >) (A 87)
,j,k,a (-i a)(j a)(k a)

(VC11 3) + (VC11 4)-
-2 z fifaj(< abA >< ib. > + < ibA >< ab. >) (A-88)
S(Ej a) (Ei Ea) (Ei Eb)
i,j,a,b b a)

(VC11- 6) =
2 fjkfki(< ajA >< aip > + < aiA >< ajp >) (A89)
,j,, (E Fa)(k ,a)(Ej Ea)

(VC11 7) + (VC12 8)
-2 fijfab(< bjA >< aip> + < aiA >< bjp >) (A90)
i,j,a,b (i F)(Ei -b)(E Eb)

(VC11- 9) + (VC12- 9)
2 faifjb(< ajA >< bip > + < biA >< ajp >) (A 91)
,j,a,b (i- Fa)(j a)( b)

(VC11- o10)
-2 faifib < ajA >< bjp > (A92)
(i, i E)( )(j E)(Ej Eb)

(VC11 11) + (VC15 5) =
4 fi < ibljk > (< ajA >< kbp > + < kbA >< ajp >) (93)
,j,k,a,b (i a)(j E)(E Eb)



j r r/1

To address second-order properties requires additional effects of the density change given

as the second and higher terms of Taylor's expansion of Ex,. The kernel

f(ri, r2) -

is used in [21, 50].However, once our objective is third-order molecular property, the

essential new element in TDDFT is the second exchange-correlation kernel, which in an

adiabatic approximation is defined as the second functional derivative of the exchange-

correlation potential with respect to the density

g(ri, 2, r3) =-p (6-1)

Hence, in this paper we will derive the second kernel for the exact-exchange DFT, based

on the OEPx approach. This is a complicated quantity. The traditional way of deriving

kernels and potentials is to tediously derive all terms with the use of the chain-rule for

functional differentiation. Yet, even the exchange- only kernel, f, has a complicated

structure in OEP, and its further extension to a second kernel is very difficult. To avoid

the use of traditional methods, an effective diagrammatic formalism for taking functional

derivatives was recently developed[37, 62]. In this paper we use that formalism and

apply it to the exchange second kernel for OEPx based time-dependent DFT. After

discussing some properties of the second kernel, we report numerical results to obtain

hyperpolarizabilities, compared to those from Hartree-Fock and coupled cluster singles and

doubles (CCSD).

There is a long history of the treatment of hyperpolarizabilities and associated non-

linear optical (NLO) properties. See [69] for a review. In particular, issues of the incorrect

3.3 Properties of the Correlation Kernel

The second-order correlation kernel has different spin components, which can be

represented by the formula

fn(ri, r2) dr3 dr6Xaa(rl,r3)haar3, r6)Xa (r6,r2)

/ (n r2) f dr df6Xa-(ri, r3) 3, 6)X (r6, r2) (3 35)

Diagrams, which contain only one contour cannot produce ha3 components after differenti-

ation. Three of the diagrams (VC5)-(VC8) after differentiation can produce only diagrams

contributing to the haa or h3p parts. The diagrams (VC9-1)-(VC9-7), (VC9-9), (VC9-10),

(VC10-1)-(VC10-7), (VC10-9), (VC10-10), (VC11-1)-(VC11-10), (VC11-13)-(VC11-16),

(VC12-1)-(VC12-10), (VC12-13)-(VC12-16), (VC13-1)-(VC13-12), (VC13-14), (VC13-15),

(VC14-1)-(VC14-12), (VC14-14), (VC14-15) have the same property. Diagrams containing

2 or more contours can produce all spin components. Sets of diagrams (VC1-7)-(VC1-16),

(VC2-7)-(VC2-16), (VC3-7)-(VC3-14), (VC4-7)-(VC414), (VC15-7)-(VC15-12) have two

contours, but both 6-functions present in one of the contours means that these diagrams

make a contribution to haa or hp parts, but have an extra factor of 2. That factor

appears after summation over all spin-orbitals of the second contour. Diagrams (VC1-

1)-(VC1-6), (VC2-1)-(VC2-6), (VC3-1)-(VC3-6), (VC4-1)-(VC4-6), (VC9-8), (VC10-8),

(VC11-11), (VC11-12), (VC12-11), (VC12-12), (VC13-13), (VC14-13), (VC15-1)-(VC15-6),

(VC15-4)-(VC15-15) have 6 functions on different contours, so they contribute to both haa

and h3p parts. Diagram (VC15-13) have an additional factor of 2 and also contribute into

both spin parts. To build all spin parts of the correlation kernel we need to substitute the

above diagrams into equation (3-35)

The next essential property of the exchange and correlation kernels is the symmetry

with respect to permutation of its arguments

S(r1, 2) f(r2, r1) (3-36)

potentials are derived from orbital-dependent energy functionals, taken from wave function


The main advantage of wave-function based methods is the possibility to obtain

systematically improvable results. This means that with the extension of the basis set, it

will be possible to obtain more accurate energies and wave functions, and in the complete

set it is possible to get the exact solution of the Schr6dinger equation. On the other

hand, rigorous ab initio methods using wave function methods are usually very costly for

computations on large molecules. Ab initio wave function methods can usually be applied

only for systems with ~20-30 atoms in reasonable basis sets. The main DFT advantage

is a small computational cost; it can handle systems with several hundreds of atoms.

However, most of conventional density functionals do not have the capacity to produce

systematically improvable results.

Ab initio DFT is capable of producing systematically improvable results, but it

is computationally more costly than DFT with conventional functionals. Despite its

computational cost, ab initio DFT can be used as a method for the calibration of density-

dependent functionals. An alternative way of providing such information is Quantum

Monte-Carlo (QMC) method, however, QMC results are not available for molecules and

open-shell systems.

1.1 Ab Initio Wavefunction-Based Methods

1.1.1 Hartree-Fock Method

In the Hartree-Fock method, a wave function is considered as a Slater determinant[1]

0 (ri) ...* ci(rN)
HF= : (1-3)

(PN(r1) ... (PN(rN)

The single-electron wave functions 4p(r) (or orbitals) are determined by the condition

that the corresponding determinant minimizes the expectation value of the electronic


Table page

3-1 Orbital energies and zero-order approximations to excitation energies . 64

3-2 Excitation energies of Ne atom using OEP-MBPT(2) Kohn-Sham orbital energies 64

3-3 Excitation energies of Ne atom using exchange-only orbital energies ...... ..66

3-4 Excitation energies of Ne atom using orbital energies and orbitals from OEP2(sc).
All equations for TDDFT are the same .................. ...... 66

4-1 Total energies .................. ................. .. 72

4-2 Ionization potentials (in e. v.) .................. ........ .. 74

4-3 Dissociation energies (in kJ/mol) .................. ....... .. 75

4-4 Singlet and triplet energies of methylene .................. ..... 76

5-1 Total (in a. u.) and orbital (in e. v.) energies of Ne atom . ..... 88

5-2 Total and orbital energies of He atom .................. ...... 88

5-3 Excitation energies (V valence state, R Rydberg) .............. ..93

5-4 Orbital energies (in e. v.) of Ne atom .................. ...... 94

5-5 Ionization energies (in e. v.) .................. ......... .. 94

5-6 Static polarizabilities (in a. u.) .................. ........ .. 99

5-7 Isotropic C6 coefficients (in a. u.) .................. ..... .. 99

6-1 Hyperpolarizabilities of several molecules (in a. u.) ............... ..107


I was born in Avdeevka, a small town located in the southeastern part of the

Ukraine in the Donetsk region. I grew up in the city of Gorlovka, a working-class mining

community. When I was in ninth grade, I acquired an interest in chemistry and decided

that I would become a chemist. During my last year of high school, I was one of the win-

ners of the Ukrainian ('!, iii-I ry Olympiad. As a result, I was admitted to the prestigious

Moscow State University without the otherwise necessary entrance examinations. During

my sophomore year of college, I decided to specialize in quantum chemistry. After I de-

fended my undergraduate thesis and obtained my degree in 2002, I began to work toward

joining Dr. Rodney Bartlett's research group at the University of Florida. After working

as a system administrator for several companies in Moscow for 16 months, I obtained the

necessary funds to l i, for transportation to the United States as well as for the necessary

exams, which included the Graduate Record Examination (GRE) and the Test of English

as a Foreign Language (TOEFL). Finally, I joined Dr. Bartlett's group at the University

of Florida on January 21st, 2004.


The time-dependent OEP exchange-only method (OEPx) has been previously

derived by G6rling[20] and implemented for molecules by Hirata et. al [21, 50]. Numerical

results in the latter papers show a reasonable description of both valence and Rydberg

excited states. Reasonable static and dynamic polarizabilities are also obtained. Good

results of TDOEPx for excited states and properties are obtained partly due to the correct

.i-vinl Il' '1 ic behavior of the exchange potential, but also due to the elimination of the

self-interaction error. In particular, OEP-TDDFTx tends to be superior to standard

approaches like TDDFT based upon the local density approximation (LDA) or Becke-Lee-

Yang-Parr (BLYP) functionals. Similarly, exchange-only OEP with exact local exchange

(EXX) [22, 23] has been shown to greatly improve band-gaps in polymers [23]

Another advantage of OEP based methods is that since virtual orbitals in the

exchange-only DFT as well as the occupied orbitals are generated by a local potential,

which corresponds to the N-l particle system due to the satisfaction of the self-interaction

cancellation, the differences between orbital energies of virtual and occupied orbitals

offer a good zeroth-order approximation to the excitation energies[16, 17]. This is not

possible in the case of Hartree-Fock theory, without adding a VN-1 potential[67, 68],

since occupied orbitals are generated by an N-l particle potential and the energies of

unoccupied orbitals come from N-electron potential, and thereby, approximate electron


In Kohn-Sham DFT

hKS = T + Vezt + J + V1e

2.3 Interconnection in Arbitrary Order

2.3.1 Diagrammatic Functional Derivatives

For formulation of the rules we will use equation (2-18). The 6 function needs to be

introduced according to equation (2-26):


6 (rqP(r) + + i+ a (2-26)
p,qPp q

Brackets on the last two diagrams denote denominators. Expressions for the energy

in MBPT consist of linear combinations of terms, which have products of molecular

integrals in the numerators and product of differences of one-particle energies in the

denominators, so it is easy to make a diagrammatic representation for such expressions.To

take the functional derivatives from numerators, according to equation (2-18), all lines

connected to some vertex must be disconnected from the corresponding place of linking,

and when this is done, a new line (corresponding to occupied or unoccupied orbitals) must

be inserted. In the final step contraction with the corresponding 6 function must be

provided. All lines corresponding to denominators are still unchanged. This procedure

must be provided for all vertices,because when we take functional derivatives from

products of functions, we have a sum of products, according to rules for taking derivatives.

If a new diagram changes its sign, a minus sign must be assigned to this diagram.

When functional derivatives from denominators are taken, it is more suitable to use

equation (2-13) The most general form for the denominator can be represented by the

following formula:
Ka Ka
Den= c( fia -V E )) (2-27)
a i=l j=1

(VC6- 5) + (VC8- 9)
2 z < ablij >< likb > (< akA >< jl/ > + < jlA >< akp >)54)
,b,, (E + j j a Eb)( + Fl Fa Eb)(k Ea)

(VC6 7)
-2 < allij >< jilkb >< akA >< lbp > (A55)
a,b,i,j,k, + E Ea b) (E a)(E- b)

(VC6 9)
-2,, < lblij >< ak >< lap >56)
,,k, (Ei + c- Ea- b)(Ek -a E)(El Ea)

(VC6 10) + (VC7 9)=
2 < ablij >< jilkb > (< ckA >< acp > + < ck >< acA >) (A57)
E,, (+ j -- -- Eb)(Ei + E, -- Eb)(Ek Ec)

(VC6- 11) + (VC6- 2)
2 < cb ij >< jilkb > (< akA >< acp > + < akp >< acA >) (A
2 +----b)( k k-) (A-58)
alb,c,i,j,k (Ei E+ Ec Eb)(EFk Ec)(Ek Ea)

(VC6- 13) + (VC6- 14)
-2 < ablij >< jikb > (< klA >< al > + < a >< kl >) (A59)
a,b,i,j,k,l ( E+ F Eb)(l a)(Ek a)

(VC6 16) =
< ablij >< jilcb >< akA >< kcp >
2 -(A60)
,,,. (Ei Fa Eb)(Ek a)(Ek c)

(VC7 2) + (VC7 3)
< aclij >< jildb > (< bcA >< adp > + < adA >< bcp >)
,c,d,i,j (Ei + F Ec)(Ei + j a Eb)(Ei j bd Eb)

Consider the second and third diagram in the energy expression. For the sake of simplicity

only non-equivalent diagrams will be given. Taking functional derivatives with respect to

occupied orbitals we will have equation (2-31)

i J bi + (231)

Inserting the lines of unoccupied orbitals and making contractions with the 6 function,

diagrams 4 and 8 of equation (2-25) will be obtained (2-32):


a a

2 4 +2 8 (2-32)

When functional derivatives are taken with respect to unoccupied orbitals and occupied

orbitals are inserted, diagrams 5 and 9 of equation (2-25) will be obtained.When this

procedure is done for the lower vertex, complex conjugate diagrams will be obtained

and we will have the same number of diagrams of this sort as in the effective operator


When functional derivatives are taken with respect to occupied orbitals for the

upper and lower vertexes and an occupied line is inserted, we have the diagrams, given by

equation (2 33)

k a 1i a k a
a b b jjk a b j k a b j
-22-2 -2 -2 (2-33)

"Bracket-type" denominator means the difference Ei Ek. Summation of the fist two

diagrams according to the Frantz-Mills theorem[18] and the same procedure for the second


The definition of a chemical reaction as a transformation of one substance to another

at the molecular level means a regrouping of nuclei and electrons. In the adiabatic

approximation, such nuclear regroupings can be presented as a motion in the field of

some potential, also known as a potential energy -, if',.. Potential energy surfaces can be

obtained by solving the eigenvalue problem with the so called electronic Hamiltonian

1 lec elec Nucl ZA elec 1
2 E Iri RA I ri rj
i=1 i=1 A=1 i
Whole nuclear coordinates are treated as a set of parameters. If there are more than two

particles in the system, the eigenvalue problem

H,i(R...RM)(rl... rN, R1...RM) = E(R ...RM)1(rl... rN, R... RM) (1-2)

can not be solved analytically. For the calculation of electronic energy levels, some

approximations have to be used.

Modern quantum chemistry includes three classes of methods for the solution

of the electronic problem. The first one, called wave function 'I' ,., '; uses different

approximations for the calculation of the wave function '(rl...rN, R1...RM) and the

corresponding energy E(R1...RM). The second class, known as /. ,:-.:/;/ functional theory

uses the density as a primary object. Within the DFT approach, the energy is written as

a functional of density, and the construction of the wave functions is not necessary. The

third class of methods uses a whole density matrix. The density matrix renolmalization

group (DMRG) and the density matrix functinal theory (DMFT) are typical methods of

the third class.

Ab initio DFT contains elements of the first two classes. It uses a local multiplicative

potential, typical for the DFT approach. However, the corresponding exchange-correlation

3.2 Kernel for the Second-Order Optimized Effective Potential Many-Body
Perturbation Theory Correlation Potential

For the derivation of the second-order exchange-correlation kernel we require

functional derivatives of the second-order correlation energy with respect to V,. These

functional derivatives can be represented by diagrams (VC1)-(VC15) of (3-18).

c ^ k

a b j b j6 a

(VC1) (VC2) (VC3) (VC4)

b k

c b 1 i\ /a b
\7 \/

(VC5) (VC6) (VC7) (VC8)
b i F
j c
F a F a b
a a
a a i a i b j
(VC9) (VC10) (VC11) (VC12)(vC13) (VC14) (VC15) (3-18)

Diagrams (VC1), (VC2), (VC5), (VC6) and (VC15) of (3-18) have an external factor 4,

diagrams (VC9) and (VC10) have no factor, while the rest of the diagrams have a factor

of 2. To get the correlation kernel diagrams we need to take functional derivatives with

respect to the potential from diagrams (V1)-(V15). After applying the diagrammatic

rules to diagram (VC1) and the use of relation (3-13), diagrams (VC1-1)-(VC1-16) of sets

(3-19) and (3-20) will appear.

~l _2 /01 2 f1 k1 C1 C1 A0,
c k cd ci
b i Ji j- i b b i i b j

(VC1-1) (VC1-2) (VC1-3) (VC1-4) (VC1-5) (VC1-6) (VC1-7) (3-19)

behavior of TDDFT hyperpolarizabilities for NLO design for polymers[64, 65] points to

the necessity of an improved theory. A rigorous (ab initio) DFT analogue starts with

exchange-only, TDOEPx.

6.1 Theory

6.1.1 Time-Dependent Density Functional Theory Response Properties

When an external time-dependent electric field is applied, the perturbation can be

written in the form (6-2)

V(t) =- Z- r E(t) (6-2)

For the case of periodic perturbations, the virtual-occupied block of the linear response of

the density can be found from the following system of linear equations

(A + B)U = -h (6-3)

The matrices A, B have the structure, given by Eqn (5-36) and h is

haa =< a, r i, > (6-4)

The calculation of the components of the first, static hyperpolarizability tensor

requires [70]

S=2 (U F U UF U +UF U +

!3A~w 2Z(U> F qJ zpU(;Jqa pq?,qia i~pa pqcT qicI
UFP U FU U UP, F, U) +

-2 (U\. c UJ + UX U + U +
U + U" U U U ) +

8S< baubc gj i Pib\ j9T\ k > lUa Utjvk (6-5)
a ab,c,i,j,k

(VC1 4) + (VC4 1)
< adlij >< cj|ab > (< icA >< db/ > + < ic/ >< dbA >)
(i + Ej Ea Eb)(i + Ej a Ed)(Ei Ec)

(VC1 5)=

2 < ablid >< cj ab >< icA >< dji >
a,b,c,d,i,j (i + j b)(i c)(j d)

(VC1 6) + (VC2 5)
< ak ij >< cj|ab > (< icA >< kbp > + < kbA >< icp >)
S(i + Ej a Eb)(Ei Ec)( k Eb)

(VC1 7)
< abldj >< cjlab >< icA >< idi >
a,b,c, Ea Eb)(Ei Ec)(Ei Ed)

< kblij >< cjlab >
(Ei + Ej

< abiij >< cj|ab >
(Ei + Ej

< abiij >< cj|ab >
(Ei + Ej a

(VC1 8) + (VC2 8)
(< icA >< ka/ > + < kaA >< icp >)
- Eb)(Ei Ec)(k Ea)

(VC1 9) + (VC1 10)
(< ikA >< kci > + < kcA >< iki >)
- Eb)(i ( c)(k c)

(VC 11) + (VC3 7)
(< ikA >< kci > + < kcA >< iki >)
- Eb)(Ek + Ej a Eb)(Ek Ec)

(VC1 12) + (VC4 8)
< ablij >< cjldb > (< icA >< dap > + < ic/ >< daA >)
(Ei + Ej a b)(Ei + Ej d Eb)(Ei Ec)






2 Z





-4 a

-4 E

-4 a


terms of the density, the Kohn-Sham theorem[6] can be used. This theorem states, that

the ground-state energy as a functional of density has a minimum, if the density is exact.

Therefore, given the energy functional, one can obtain the ground-state density and energy

by variational minimization of the functional. The formal definition of DFT does not

tell how to construct such a functional. Several approximate forms have been -ir-i-. -1. I

however, whose accuracy varies greatly for different properties. The kinetic energy of

electronic motion is particularly difficult to approximate as a density functional.

The basic idea of Kohn and Sham was to transform the variational search over the

density into a search over the orbitals that integrate to a given trial density

(- V2 + v(r))P'(r) -= p,(r) (1-35)

Such a transformation does not restrict the variational space, provided that every phys-

ically meaningful density corresponds to a unique set of orbitals (the v-representability

condition). The use of the Kohn-Sham SCF model ensures not only that the variational

density is a fermionic density, but it also provides a good approximation for the kinetic

energy. If orbitals integrate to the true density it is natural to expect that
T, < |V21> (136)

accounts for a large part of real kinetic energy.

The rest of the unknown terms are grouped into the exchange-correlation functional

Ec[p] = E[p] T, Eet EH (137)

where EH is a Hartree energy and can be easily be calculated from a given set of Kohn-

Sham orbitals. The non-interacting kinetic energy T, is supposed to reproduce a large

part of the exact kinetic energy T. Therefore, Ec is easier to approximate as a density

functional than the total energy E.


(VC5 2)
< ablij >< cdlba >< icA >< djp >
(E + Ej a Eb)(i E,) Ej Ed)

(VC5 3) + (VC8 1)

2 < ablik >< cjba > (< icA >< jk/ > + < jkA>)
ab,c,i,j,k (c( + F- i a -Eb)(Ei+ Ek Ea Eb)(Ei Ec)

(VC5 4) + (VC7 12)
2 < dbjij >< cj|ba > (< icA >< dap > + < daA >< icp >)
~,b,cd(i +C d Ib)( -+ Ea b)(Ei Ec)

(VC5 5) + (VC7 8) =

-2 < abiij >< cj|da > (< icA >< dbi > + < dbA >< icp >)
,c,d,i, (E Ej Ea b)(E Ej -Fa E- Ed) (E Ec)

Z < aklij >< cjlba >
a,b,c,i,j,k ( -

2 Z

< ablid
(Ei + Fj

2 < kblij >< cjlba >
b,,i,j,k (Ei + Ej

(VC5 6) + (VC6 -8)
(< icA >< kb/ > + < ic/ >< kbA >)
- Ea Eb)(i Ec)(Ek Eb)

(VC5 7)
>< cjbba >< icA >< jbp >
- a Eb)(E Ec)(E -? d)

(VC5 8) + (VC6 -6)
(< icA >< kai > + < icp >< kaA >)
- a Eb)(Ei ,c)(k a)

(VC5 9)
>< cba >< ic >< idi >
- E Eb)(i Ec)(Ei d)








< abldj
(Ei + Ej



Full Text






IwouldliketothankprofessorsHenkMonkhorstandSoHirataforhelpfulldiscus-sions.IwanttothankalsoDr.IgorSchweigertandDr.NorbertFlockeforthehelpingmetowriteOEPcode.MyspecialthankstoTatyanaandThomasAlbertforhelpwiththepreparationofthedissertationtext. 3


page ACKNOWLEDGMENTS ................................. 3 LISTOFTABLES ..................................... 7 LISTOFFIGURES .................................... 8 ABSTRACT ........................................ 9 CHAPTER 1INTRODUCTION .................................. 12 1.1AbInitioWavefunction-BasedMethods .................... 13 1.1.1Hartree-FockMethod .......................... 13 1.1.2Electron-CorrelationMethods ..................... 16 1.2Kohn-ShamDensityFunctionalTheory .................... 19 1.2.1Time-DependentDensityFunctionalTheory ............. 21 1.2.2Time-DependentDensityFunctionalTheoryLinearResponseTheory 23 1.2.3ProblemswithConventionalFunctionals ............... 26 1.2.4Orbital-DependentFunctionals ..................... 28 1.2.5AbInitioDensityFunctionalTheory ................. 29 2INTERCONNECTIONBETWEENFUNCTIONALDERIVATIVEANDEF-FECTIVEOPERATORAPPROACHESTOABINITIODENSITYFUNC-TIONALTHEORY .................................. 31 2.1EquationsfortheExchange-CorrelationPotentialintheFunctionalDeriva-tiveApproach .................................. 31 2.2EquationsfortheExchange-CorrelationPotentialinanEectiveOperatorApproach .................................... 34 2.3InterconnectioninArbitraryOrder ...................... 37 2.3.1DiagrammaticFunctionalDerivatives ................. 37 2.3.2DiagrammaticFunctionalDerivativesinSecond-OrderMany-BodyPerturbationTheory .......................... 38 2.3.3InterconnectioninHigherOrders ................... 41 2.3.4InterconnectioninInniteOrder .................... 45 3ABINITIOTIME-DEPENDENTDENSITYFUNCTIONALTHEORYEM-PLOYINGSECOND-ORDERMANY-BODYPERTURBATIONOPTIMIZEDEFFECTIVEPOTENTIAL ............................. 47 3.1DiagrammaticConstructionoftheExchange-CorrelationKernels ...... 48 3.1.1Formalism ................................ 48 3.1.2AnExample:DiagrammaticDerivationofExchange-OnlyKernel .. 52 4


.................. 54 3.3PropertiesoftheCorrelationKernel ...................... 61 3.4NumericalTesting ................................ 62 3.5Conclusions ................................... 67 4ABINITIODENSITYFUNCTIONALTHEORYFORSPIN-POLARIZEDSYSTEMS ....................................... 69 4.1Theory ...................................... 69 4.2ResultsandDiscussion ............................. 71 4.2.1TotalEnergies .............................. 71 4.2.2IonizationPotentials .......................... 74 4.2.3DissociationEnergies .......................... 75 4.2.4Singlet-TripletSeparationinMethylene ................ 75 4.3Conclusions ................................... 75 5EXACT-EXCHANGETIME-DEPENDENTDENSITYFUNCTIONALTHE-ORYFOROPEN-SHELLSYSTEMS ........................ 80 5.1Exact-ExchangeDensityFunctionalTheory ................. 82 5.2Time-DependentOptimizedEectivePotential ................ 91 5.2.1TheoryandImplementation ...................... 91 5.2.2NumericalResults ............................ 92 5.2.3Charge-TransferExcitedStates .................... 95 5.3Conclusions ................................... 97 6EXACTEXCHANGETIME-DEPENDENTDENSITYFUNCTIONALTHE-ORYFORHYPERPOLARIZABILITIES ..................... 100 6.1Theory ...................................... 102 6.1.1Time-DependentDensityFunctionalTheoryResponseProperties .. 102 6.1.2DiagrammaticDerivationoftheSecondExact-ExchangeKernel .. 103 6.1.3PropertiesoftheSecondExact-ExchangeKernel ........... 106 6.2NumericalResults ................................ 107 6.3Conclusions ................................... 109 APPENDIX AINTERPRETATIONOFDIAGRAMSOFTHESECOND-ORDERMANY-BODYPERTURBATIONTHEORYOPTIMIZEDEFFECTIVEPOTENTIALCORRELATIONKERNEL ............................. 110 BINTERPRETATIONOFDIAGRAMSOFEXACT-EXCHANGESECONDKERNEL ....................................... 127 REFERENCES ....................................... 131 5


................................ 135 6


Table page 3-1Orbitalenergiesandzero-orderapproximationstoexcitationenergies ...... 64 3-2ExcitationenergiesofNeatomusingOEP-MBPT(2)Kohn-Shamorbitalenergies 64 3-3ExcitationenergiesofNeatomusingexchange-onlyorbitalenergies ....... 66 3-4ExcitationenergiesofNeatomusingorbitalenergiesandorbitalsfromOEP2(sc).AllequationsforTDDFTarethesame ....................... 66 4-1Totalenergies .................................... 72 4-2Ionizationpotentials(ine.v.) ........................... 74 4-3Dissociationenergies(inkJ/mol) .......................... 75 4-4Singletandtripletenergiesofmethylene ...................... 76 5-1Total(ina.u.)andorbital(ine.v.)energiesofNeatom ............ 88 5-2TotalandorbitalenergiesofHeatom ....................... 88 5-3Excitationenergies(V-valencestate,R-Rydberg) ............... 93 5-4Orbitalenergies(ine.v.)ofNeatom ....................... 94 5-5Ionizationenergies(ine.v.) ............................ 94 5-6Staticpolarizabilities(ina.u.) ........................... 99 5-7IsotropicC6coecients(ina.u.) ......................... 99 6-1Hyperpolarizabilitiesofseveralmolecules(ina.u.) ................ 107 7


Figure page 4-1ExchangeandcorrelationpotentialsofLiatom(radialpart).A)Exchangepo-tential.B)Correlationpotential ........................... 73 4-2ExchangeandcorrelationpotentialsofO2moleculeacrossthemolecularaxis.A)Exchangepotential.B)Correlationpotential .................. 73 4-3LiHpotentialenergycurve. ............................. 77 4-4OHpotentialenergycurve. .............................. 78 4-5HFpotentialenergycurve. .............................. 79 5-1ExchangepotentialsofNeatom,obtainedindierendbasissets ......... 88 5-2A1charge-transferedexcitedstateofHe...Be ................. 96 5-3LUMO-HOMOorbitalenergydierence ...................... 96 8








^Hel=1 2elecXi=1r2ielecXi=1NuclXA=1ZA ^Hel(R1:::RM)(r1:::rN;R1:::RM)=E(R1:::RM)(r1:::rN;R1:::RM)(1{2)cannotbesolvedanalytically.Forthecalculationofelectronicenergylevels,someapproximationshavetobeused.Modernquantumchemistryincludesthreeclassesofmethodsforthesolutionoftheelectronicproblem.Therstone,calledwavefunctiontheory,usesdierentapproximationsforthecalculationofthewavefunction(r1:::rN;R1:::RM)andthecorrespondingenergyE(R1:::RM).Thesecondclass,knownasdensityfunctionaltheoryusesthedensityasaprimaryobject.WithintheDFTapproach,theenergyiswrittenasafunctionalofdensity,andtheconstructionofthewavefunctionsisnotnecessary.Thethirdclassofmethodsusesawholedensitymatrix.Thedensitymatrixrenolmalizationgroup(DMRG)andthedensitymatrixfunctinaltheory(DMFT)aretypicalmethodsofthethirdclass.AbinitioDFTcontainselementsofthersttwoclasses.Itusesalocalmultiplicativepotential,typicalfortheDFTapproach.However,thecorrespondingexchange-correlation 12


1.1.1Hartree-FockMethodIntheHartree-Fockmethod,awavefunctionisconsideredasaSlaterdeterminant[ 1 ] HF='1(r1)'1(rN)......'N(r1)'N(rN)(1{3)Thesingle-electronwavefunctions'i(r)(ororbitals)aredeterminedbytheconditionthatthecorrespondingdeterminantminimizestheexpectationvalueoftheelectronic 13


1{3 )intoanexpectationvalue,itispossibletowritetheHartree-Fockenergyintermsoforbitals 2r2NuclXAZA 2elecXi;j(<'i'jj'i'j><'i'jj'j'i>)(1{5)ToderivetheHartree-Fockequationsitisconvenienttominimizethefollowingfunctional 1{6 )requiresthatthefunctionalderivativesofIwithrespecttotheorbitalsvanish 'i=0;I 'i=0(1{7)TakingintoaccountEquation( 1{3 )condition( 1{7 )canbepresentedas 2r2+^vext+^vH+^vnlx(1{9) 14


^vext(r)=elecXiNuclXAZA (1{10) TheHartree-Fockexchangeoperator^vnlxisnon-local,i.e.itcannotbepresentedasananalyticalfunctionofspatialvariables.However,itispossibletowriteitsactiononsomeorbital'i 1{8 )canberewritteninthecanonicalform 15


FCI=HF+occXivirtXaCaiai+occXi6=jvirtXa6=bCabijabij+:::(1{14)whereai,abij,etcareSlaterdeterminants,formedbysubstitutionofoccupiedorbitalsi,j...byvirtualorbitalsa,b...withthecorrespondingreorderingofrows.TheexpansioncoecientsarefoundfromthevariationalconditionontheexpectationvalueoftheHamiltonian (1{15) 16


2 ]andMany-BodyPerturbationTheory[ 3 ].AnytruncatedversionoftheCImethodhasaqualitativelywrongbehavioroftheenergiesandwavefunctionswhileincreasingthenumberofparticlesinthesystem.Therefore,theCImethodswithlimitedlevelofexcitationscannotbeusedforhighly-correlatedsystems.TheCoupledClustermethodandMany-BodyPerturbationtheoriesarefreefromthislackofextensivityfailureandareverycommonforthemolecularcomputations.Insomecasesperturbationtheorycanprovideanaccuratedescriptionofelectron-correlationeectsatamuchlowercostthannecessaryfortheCoupledClustermethod.Thesecond-orderRayleigh-Schrodingerperturbationtheoryisthesimplestandleastexpensiveabinitiomethodfortakingintoaccountelectroncorrelationeects.InthisperturbationtheorythesolutionofSchrodingerequation ^H=E(1{16)canbefoundusingtheSlaterdeterminantasareference.Generally,suchadeterminantmaybeconstructedfromtheorbitals,generatedbysomeone-electronoperator ^h'p=(1 2r2+^u)'p="p'p(1{17)TherststepofanyperturbationtheoryisthepartitioningoftheHamiltonianintoazero-orderH0andperturbation ^H=^H0+^V(1{18)where ^H0=E0=Xi(hi)=(elecXi"i)(1{19) 17


^V=^H^H0=elecXi^vext(ri)elecXi^u(ri)+elecXi6=j1 ^H=^H0+^V =+(1)+2(2)+::: Theseorder-by-ordercorrectionscanbefoundbysubstitutingexpansions( 1{21 ),( 1{22 )and( 1{23 )into( 1{16 )andcollectingtermswiththecorrespondingorderin (E0^H0)j(2)>=(^VE(1))j(1)>E(2)j> Choosingtheperturbativecorrectionstobeorthogonaltothereferencedeterminant<(n)j>=0(socalledintermediatenormalization)itiseasytogettheexpressionsforperturbativecorrectionsatanyorder.Projectingtheequations( 1{24 )and( 1{25 )ontothereferencedeterminantexpressionsforenergycorrectionsareobtained Expressionsfortheorder-by-orderexpansionofwavefunctionscanbewrittenusingtheresolventoperator[ 4 ] 18


^R0=^Q E0^H0(1{30)and^Q=1-j>.Since^H0isdiagonalinthebasisofSlaterdeterminants,itispossibletowrite ^R0=Xn6=0jn>() 1{17 ),thesecond-ordercorrectiontotheenergyhasthefollowingstructure 2occXi;junoccXa;b() 5 ],establishesaone-to-onemappingbetweentheground-stateelectronicdensityandtheexternalpotential.Theexternalpotentialdenesaparticularobject(atom,molecule,etc)and,becauseoftheone-to-onemapping,thedensitycontainsalltheinformationaboutthesystem.Inparticular,theground-stateenergycanbewrittenasafunctionalofthedensity.Togetthegroundstateenergyin 19


6 ]canbeused.Thistheoremstates,thattheground-stateenergyasafunctionalofdensityhasaminimum,ifthedensityisexact.Therefore,giventheenergyfunctional,onecanobtaintheground-statedensityandenergybyvariationalminimizationofthefunctional.TheformaldenitionofDFTdoesnottellhowtoconstructsuchafunctional.Severalapproximateformshavebeensuggested,however,whoseaccuracyvariesgreatlyfordierendproperties.Thekineticenergyofelectronicmotionisparticularlydiculttoapproximateasadensityfunctional.ThebasicideaofKohnandShamwastotransformthevariationalsearchoverthedensityintoasearchovertheorbitalsthatintegratetoagiventrialdensity (1 2r2+vs(r))'p(r)="p'p(r)(1{35)Suchatransformationdoesnotrestrictthevariationalspace,providedthateveryphys-icallymeaningfuldensitycorrespondstoauniquesetoforbitals(thev-representabilitycondition).TheuseoftheKohn-ShamSCFmodelensuresnotonlythatthevariationaldensityisafermionicdensity,butitalsoprovidesagoodapproximationforthekineticenergy.Iforbitalsintegratetothetruedensityitisnaturalytoexpectthat 2elecXi<'ijr2j'i>(1{36)accountsforalargepartofrealkineticenergy.Therestoftheunknowntermsaregroupedintotheexchange-correlationfunctional 20


2Zdr1dr2(r1)(r2) 21


7 ].Wecanstartfromthetime-dependentSchrodingerequation (t0)=0(1{43)undertheinuenceofdierentexternalpotentialsv(r,t).Foreachxedinitialstate0,theformalsolutionoftheSchrodingerequation( 1{42 )denesamap 22


@t^Hj[](t)>(1{48)andthevariationalconditioncanbeapplied, @t'i(r;t)=(1 2r2+vs[(r;t)])'i(r;t)(1{50)where (1 2r2+v0(r)+Zdr00(r0) 23


1{50 )provideaformallyexactwayofcalculatingthetime-dependentdensity,itispossibletocomputeexactdensityresponse1(r;t)astheresponseofthenon-interactingsystem 24


1{55 )and( 1{56 )arethebasisoflinearresponsetheory.Sinceequation( 1{55 )isnotlinearwithrespectto1thesolutionshouldbeobtainedbysomeiterativeprocedure.However,forallpracticalpurposesadirectuseofequation( 1{55 )isnotconvenient.Togetamoreconvenientform,thefollowingmatrixelementshouldbeconsidered (1{59) Ontheotherhand,<'aj1(r;!)j'i>canbeexpressedintermsofaresponsefunction !+("a"i);Pai(!)=<'ijv1(!)j'a> 1{61 )intoequation( 1{59 )wehave (1{62) Introducingthenotations 25


1{62 )canberewritteninthefollowingmatrixform 1{64 )canberewrittenas[ 8 { 11 ] 1{65 )iscommonforthecalculationofexcitationenergieswithtime-dependentdensityfunctionaltheory(TDDFT).MostmodernimplementationsofTDDFTusethefrequency-independentexchange-correlationkernels,denedasasimplefunctionalderivative 26






2Xi;j<'i'jj'j'i>(1{67) 29




12 13 ]isastraightfor-wardapproachtotheabinitioDFT.ThecorrespondingKohn-Shampotentialcanthenbeobtainedbytakingthefunctionalderivativeofthenite-orderenergyviathechainrule,thattransformsthederivativewithrespecttothedensityintothederivativewithrespecttoorbitalsandorbitalenergies[ 14 ].ThisleadstotheOptimizedEectivePotential[ 13 15 ]equationswhichcanbecomplicatedalreadyinsecondorder[ 13 ]Alternatively,onecandeterminetherstandsecondorderKohn-Shampotentialbyrequiringthatthecorrespondingrstandsecondorderperturbativecorrectionstothereferencedensityvanish.Unlikethefunctionalderivativeapproachsuchaconditiononthedensitycanbedescribedwithstandardmany-bodytechniques.Therecentwork[ 17 ]usesdiagramstoderivethesecond-orderOEPequationinasystematicandcompactfashion,whileasecondpaper[ 16 ]doessoalgebraically.Boththefunctionalderivativeanddensity(eectiveoperator)approachesleadtoexactlythesameequationintherstorder[ 17 ].However,thefunctionalderivativeofthesecondenergyinvolvesacertaintypeofdenominatorsthatarenotpresentinthedensitycondition(eectiveoperatorapproach).Stillthetermsinvolvingsuchdenominatorscanbetransformedtomatchthedensity-basedequationexactly.Thiscaveatraisesthequestionofwhetherthetwoapproachesareequivalentinhigherorders,orfordierentpartitioningsofthehamiltonian. 31


14 ]:E(n)=E(n)('1:::::'n;1:::::n)Anotherpossibleconsideration[ 13 ],whenorbitalenergiesareconsideredtobeafunctionalsovercorrespondingorbitals 14 ]approach.ToconstructthechainruleweshouldtakeintoaccountthefactthattheKohn-ShamorbitalsarefunctionalsofVs: 32


(2{8) (2{9) Usingequations( 2{8 )and( 2{9 )wecanrewritethevariationoftheenergyinthefollowingform: 13 14 ]forthecalculationofV(n)xc: 14 ]: 13 ] 33


2{15 )as: "i"a=Xp;q6=p "i"a=Xa;ifa+paqg(2{21) 34


17 ].Thismeans,thatifweconstructaneectiveoperatorofthedensityusingMBPT,thecorrectiontothedensitymustvanishinanyorder.Thisisthemainideaoftheeectiveoperatorapproach.Fortheexchange-correlationpotentialofrstorderwewillhave: 2{23 ). (1)(2{23)Usingthefactthatfpq="ppq+<'pjVHFxVx()Vcj'q>,wecanextractourdesirableexchangepotential.Thecorrelationpartisexcludedtomaintainrstorder,asthecorrelationpotentialcontainsexpressionsofsecondandhigherorders.ThentheequationforVxcanbediagrammaticallyrepresentedbydiagrams2and3. Forthesecond-ordereectiveoperatorforthedensitycorrectionwewillusetheexpression( 2{24 ) (2{24) 35


2{25 ). 2 =4 +4 ++2 +2 ++4 +4 +2 +2 + + +2 +2 ++4 +2 +2 (2{25) Diagrammaticrepresentationofpotential,derivedfromthedensityconditionwillbeusedfortheestablishinginterconnectionwiththefunctionalderivativeapproach. 36


2.3.1DiagrammaticFunctionalDerivativesForformulationoftheruleswewilluseequation( 2{18 ).The-functionneedstobeintroducedaccordingtoequation( 2{26 ): + + + (2{26)Bracketsonthelasttwodiagramsdenotedenominators.ExpressionsfortheenergyinMBPTconsistoflinearcombinationsofterms,whichhaveproductsofmolecularintegralsinthenumeratorsandproductofdierencesofone-particleenergiesinthedenominators,soitiseasytomakeadiagrammaticrepresentationforsuchexpressions.Totakethefunctionalderivativesfromnumerators,accordingtoequation( 2{18 ),alllinesconnectedtosomevertexmustbedisconnectedfromthecorrespondingplaceoflinking,andwhenthisisdone,anewline(correspondingtooccupiedorunoccupiedorbitals)mustbeinserted.Inthenalstepcontractionwiththecorresponding-functionmustbeprovided.Alllinescorrespondingtodenominatorsarestillunchanged.Thisproceduremustbeprovidedforallvertices,becausewhenwetakefunctionalderivativesfromproductsoffunctions,wehaveasumofproducts,accordingtorulesfortakingderivatives.Ifanewdiagramchangesitssign,aminussignmustbeassignedtothisdiagram.Whenfunctionalderivativesfromdenominatorsaretaken,itismoresuitabletouseequation( 2{13 ).Themostgeneralformforthedenominatorcanberepresentedbythefollowingformula: 37


2{13 )itispossibletowritedownthefunctionalderivativesfromsomedenominatorinthegeneralform: Vs=(K1Xi=1j'i1occj2K1Xj=1j'j1unoccj2)Y;6=1(KXi=1ioccKXj=1junocc)+:::::+:::::+(KXi=1j'ioccj2KXj=1j'junoccj2)Y;6=(KXi=1ioccKXj=1junocc) (2{28) Nowitispossibletowritedownatermwhichincludesthefunctionalderivativefromthedenominator: Vs==XNom(PKi=1j'ioccj2PKj=1j'junoccj2) (PKi=1ioccPKj=1junocc)2Q;6=(PKi=1ioccPKj=1junocc) (2{29) Wenowhavethepossibilitytoformulatehowtotakefunctionalderivativesfromdenom-inators.Totakefunctionalderivativesfromdenominators,alldiagrams,whereoneofthelinesisdoubledandbetweentheselineswhicharisefromdoubling,thecorrespondingdiagonal-functionisinserted.Thisproceduremustbeprovidedforallhorizontallinesonthediagramandforallthecontoursthelinescross. 2{17 )isequaltodiagram2,theright-handsiteisequaltodiagram3whenthefunctionalderivativehasbeentakenfromtheexpressionfortheexchangeenergy.Thediagrammaticexpressionforthesecond-orderenergyinthegeneralcasehastheform[ 1 ]: +2 +2 (2{30) 38


2{31 ) + (2{31)Insertingthelinesofunoccupiedorbitalsandmakingcontractionswiththe-function,diagrams4and8ofequation( 2{25 )willbeobtained( 2{32 ): 2 +2 (2{32)Whenfunctionalderivativesaretakenwithrespecttounoccupiedorbitalsandoccupiedorbitalsareinserted,diagrams5and9ofequation( 2{25 )willbeobtained.Whenthisprocedureisdoneforthelowervertex,complexconjugatediagramswillbeobtainedandwewillhavethesamenumberofdiagramsofthissortasintheeectiveoperatorapproach.Whenfunctionalderivativesaretakenwithrespecttooccupiedorbitalsfortheupperandlowervertexesandanoccupiedlineisinserted,wehavethediagrams,givenbyequation( 2{33 ) (2{33)"Bracket-type"denominatormeansthedierence"i"k.SummationofthesttwodiagramsaccordingtotheFrantz-Millstheorem[ 18 ]andthesameprocedureforthesecond 39


2{34 ) 2 +2 (2{34)Onthetwodiagramsweimposetherestrictioni6=k.Tomakeadirectcorrespondencewithdiagrams7and11weneedtoadddiagramsthatarisefromtakingfunctionalderivativesfromdenominators( 2{35 ) 2 2 (2{35)Afterthesediagramsareaddedtotheprevioustwo,wewillhaveadirectcorrespondencewithdiagrams7and11.Whenfunctionalderivativesaretakenwithrespecttotheunoccupiedorbitalsandanunoccupiedlineinserted,thesameprocedurewillgiveadirectcorrespondencewithdiagrams6and10ofequation( 2{25 ).Nowconsidertherstdiagramforthesecond-orderenergyexpression.Thepro-cedurefortakingfunctionalderivativesfromtherightsideofthevertex,whichdonotcontainaclosedF-ringareabsolutelythesameasinthecaseofothersecondandthirddiagramsintheenergyexpression.Whensuchfunctionalderivativesaretaken,wehavefourdiagramscorrespondingtodiagrams12-15.TotakefunctionalderivativesfromtheF-rings,therstdiagramintheenergyexpressionmustberepresentedwithmoredetail,takingintoaccountthatf=h+J-K,infollowingtheform( 2{36 ) = + + +2 +2 +2 (2{36) 40


2{37 ) 2 +2 +2 +2 (2{37)Thersttwodiagramsareequaltodiagram16;thethirdandfourthequaltodiagrams17and16respectively.Sincetheleft-handsiteofeq.12canberepresentedbydiagram3,wehaveanexactequivalencebetweenthefunctionalderivativeandtheeectiveoperatorapproachesintherstandsecondorderforthe"Kohn-Sham"partitioningofthehamiltonian. 2{38 ) 41


18 ]factorizationtheorem.Sinceallinitialhorizontallinescrossalllines,whichgoupfromlowerlyingvertexes,itispossibletorepresentthediagrams,whichappearafterapplyingtheFrantz-Mills[ 18 ]theoremintheform( 2{39 ) = + (2{40)Thediagramontheleft-handsideistheresultoftakingfunctionalderivativesfromtypicalenergydiagramsofthirdorder.Applyingthefactorizationtheorem[ 18 ]totheright-handsidewewillhavethesameresultasontheleft-handsideofthisrelation.The 42


2{41 ): (2{41)Numbersonlinesshowtheconditionalnumberoflineswhichareenteredforsimplicityoffurthermanipulations.Thedesirablesumofdiagramshastheform( 2{42 ) + +:::::::::+ (2{42)Togetherwithdiagramswhichcorrespondtofunctionalderivativesfromdenominators,thissetofdiagramsformsproductslike(k)+(l),wherek+lequalstheorderofMBPT. 43


2{43 ) (2{43) IntheseexpressionsNommeansnumerator,Aimeanstherestoftheterms,whicharepresentindenominators.Aftermultiplicationanddivisionby(kl)wehave( 2{44 ) (kl)(l+A1)(k+A1)::::(k+Am)+XNom(kl+A2A2) (kl)(l+A1)(l+A2)(k+A2)::::(k+Am)+::::::::+XNom(kl+AmAm) (kl)(l+A1)::::(l+Am)(k+Am)==XNom (2{44) Aftercancellationofallequivalentterms,wehavetheexpressions( 2{45 ) 44


2{46 ) + (2{46)Namelythesetwodiagramswillappearaftertakingfunctionalderivativesusingdia-grammaticrulesfortakingfunctionalderivatives.Proofforthecasewherefunctionalderivativesaretakenwithrespecttounoccupiedorbitalsisthecompleteanalogofthisone.Hence,wecanformulateasecondstatement.Statement2:Whenfunctionalderivativesfromsomevertexofadiagramaretakenwithrespecttooccupied(unoccupied)orbitals,andalinewhichcorrespondstooccupied(unoccupied)orbitalinserted,togetherwithdiagramswhicharisefromtakingfunctionalderivativesfromdenominators;wealwayshavediagrammaticexpressions,whichcorrespondstopartofPnk=0((k)+(nk))Sc.Thisstatementisadirectcorollaryoftheaboveprovedstatementabout"bracket-type"denominatorsandstatement1.Sincewehavecorrespondenceinthesecondandn-thorder,usingthemethodofmathematicalinduction,itispossibletoprove,thatthecorrespondencetaksplaceinallorders(niteorinnite). 45


19 ]toexctractthecorrespondingexchange-correlationpotential.SincethefullCIenergydoesnotdependuponthechoiceoforbitalbasisset,theOEPprocedure[ 13 15 ]cannotbeuseddirectlyforthiscase.Theinnitesumofallenergycorrectionsdoesnotdependuponchoiceoforbitalbasisset,buteachtermofthissumdoesdependuponthechoiceoforbitals.Thisfactenablesustoconsideralltermoftheinnitesumofenergycorrectionsasorbital-dependentfunctionals.AftertheconstructionofthesetV(1)xc::::V(1)xcandsummationuptoinnity,wewillhavetheexchange-correlationpotentialwhichcorrespondstothefullCI.Usingtheequivalenceofthefunctionalderivativeandtheeectiveoperatorapproach,itispossibletoconcludethataftersummationofthissetofpotentials,weagainwillhavethesameresultasintheeectiveoperatorapproach.Redenedinsuchaway,theOEPprocedure[ 13 15 ]forthefullCIenergyproducesthesamedensityastheZMPmethod[ 19 ]wouldfromfullCI. 46


20 ][ 21 ].NumericalresultsinthatworkshowareasonabledescriptionofbothvalenceandRydbergexcitedstates,partlyduetothecorrectasymptoticbehavioroftheexchangepotential,butalsoduetotheeliminationoftheself-interactionerror.Inparticular,OEP-TDDFTissuperiortostandardapproacheslikeTDDFTbaseduponlocaldensityapproximation(LDA)orBecke-Lee-Yang-Parr(BLYP)functionals.Similarly,exchange-onlyOEPwithexactlocalexchange(EXX)[ 22 23 ]hasbeenshowntogreatlyimproveband-gapsinpolymers[ 23 ]AnotheradvantageofOEPbasedmethodsisthatsincevirtualorbitalsintheexchange-onlyDFTaswellasoccupiedorbitalsaregeneratedbyalocalpotential,whichcorrespondstotheN-particlesystem,thedierencesbetweenorbitalenergiesofvirtualandoccupiedorbitalsoeragoodzeroth-orderapproximationtotheexcitationenergies[ 16 17 ].ThisisnotpossibleinthecaseofHartree-Focktheory,whereoccupiedorbitalsaregeneratedbyanN-1particlepotentialandtheenergiesofunoccupiedorbitalscomefromN-electronpotential,andthereby,approximateelectronanities.OnceOEPcorrelationisadded[ 13 17 ],theessentialnewelementinthetime-dependentDFTschemeistheexchange-correlationkernel,whichintheadiabaticapprox-imationisdenedasafunctionalderivativeoftheexchange-correlationpotentialwithrespecttothedensity[ 20 ]: 16 17 ]thishasbeencalledOEPKohn-Sham(KS)todistinguishitfromotherchoicesforH0.Itis 47


14 24 ].OnlyforthischoiceisthereanimmediatecorrespondencebetweenthefunctionalderivativeandKSdensityconditionthatasingledeterminantprovidestheexactdensity[ 24 ].TheotherchoicesforH0leadtowhatiscalledabinitioDFT[ 16 17 ]andbecomesabetterseparationofhamiltonianintoH0+V0.Itsconvergencefororbital-dependentperturbationapproximationstothecorrelationenergyfunctionalsareconsiderablybetterbehavedcomparedtotheKSchoicethatfrequentlycausesdivergence.Nevertheless,inthisrstapplicationwewilladheretothestandardKSseparation,wherethepotentialandkernelarefunctionalderivatives.Thetraditionalwayofderivingkernelsandpotentialsistotediouslyderivealltermswiththeuseofthechain-ruleforfunctionaldierentiation.Yet,eventheexchange-onlykernelhasacomplicatedstructureinOEP,anditsfurtherextensiontoincludecorrelationwouldbealmostimpossible.Toavoidtheuseoftraditionalmethods,aneectivediagrammaticformalismfortakingfunctionalderivativeshasbeendevelopedandimplemented. 3.1.1FormalismTheadiabatic(frequency-independent)approximationtothekernelofthenthorderisdenedas( 3{2 ) 13 ]( 3{3 ) 48


3{3 )into( 3{2 )weobtain 'p(r4)(Zdr5E(n) Thismayberewrittenas( 3{5 ) Vs(r3)(Zdr5E(n) Equation( 3{5 )canbemademoreexplicit, Vs(r3)[Zdr5E(n) Usingthefactthat[ 14 ] 3{6 )inthefollowingform( 3{8 ) 3{3 )wehavetheexplicitexpression( 3{9 )forh(r3;r6) Vs(r3)[Xp;q6=pZdr5E(n) 49


3{9 )willbeusedasthebasisforthediagrammaticprocedurefortakingfunctionalderivativesfromtheOEP-MBPTexchange-correlationpotentials.Thersttermofequation( 3{9 )containsthefunctionalderivativewithrespecttothepotential,fromanexpressionwhichalreadyincludesthefunctionalderivativeoftheenergywithrespecttothesamepotential.Diagrammaticrulesfortakingfunctionalderivativesfromnumeratorsofdiagramsarethefollowing: 2{26 )togetthefullyconnecteddiagram 2{26 )shouldbeinsertedbetweendoubledlinesforalllineswhichareintersectedbythedenominatorline.Thisprocedureshouldbedoneforallpossibledenominatorlines.TherulesofinterpretationarethesameasfortheusualGoldstonediagrams,exceptthenumericalfactorshouldbetakenfromtheinitialdiagramwhenfunctionalderivativesaretaken.Aftertakingthefunctionalderivativesfromtheexchange-correlationwithrespecttothepotentialusingdiagrammaticrules,asetofdiagrams,containingthe^-functionvertexwillappear.Togettherstelementofequation( 3{9 ),itisnecessarytoapplydiagrammaticrulestothesetoffunctionalderivativediagramsonemoretime.Afterthatwewillhaveasetofdiagramscontainingtwo^-functions,whicharenecessaryfortheconstructionoftheexchange-correlationkernels.Takingintoaccountequation( 2{14 ),thesecondtermofequation( 3{9 )canberewritteninthefollowingway( 3{10 ) Vs(r3)['i(r5'a(r5)'i(r6)'a(r6) 50


3{10 )canberepresentedbydiagram(V)( 3{11 ) (3{11)Thefunctionalderivativesfromdiagram(V)producethesetofdiagrams(V1)-(V6)ofset( 3{12 ) (3{12)Duringthederivationofdiagrams(V1)-(V4),diagramscontaining"bracket-type"denom-inatorswillappear,becauseoneofthestepsinthediagrammaticrulesofdierentiationrequiresdetachingtheunoccupiedlinefromdiagramVandinsertingthelasttwopartsofthe^-function( 2{26 ).Suchdiagramscanbetransformedintoasetofregulardiagrams,usingthediagrammaticrelation( 3{13 ).Thisrelationshowsusthatforthetransformationtoregulardiagramsitisnecessarytodoubleeachofthedenominatorlinesandinsertthelasttwomembersofthe^-function( 2{26 ),subjecttotherestrictedsummation,l6=k.Weusethesameprocedureforthecaseofoccupiedorbitals. + = + +...+ (D1)(3{13) 51


3{9 )anddiagrams(V1)-(V6)itispossibletobuildtheexchange-correlationkernelstoanyorder. 20 ],[ 21 ].Suchaderivationrequiresmucheort.Hereweoerafacilederivationwiththediagrammaticformalism.Theexchangeenergy(Ex=-1 2Pi;j)canberepresentedbydiagram( 3{14 ). (EX)(3{14)AftertakingtherstfunctionalderivativewithrespecttoVswewillhavethediagram( 3{15 ). (3{15)Afterthatthediagrammaticrulesmustbeappliedonemoretimetogetthesecondfunctionalderivative,asisnecessaryaccordingtoequation( 3{9 ).Aftertakingthesecondfunctionalderivativeswewillhavediagrams(FX1)-(FX8)oftheset( 3{16 ) (3{16)Aftertheadditionofdiagrams(V1)-(V6)wewillhaveallthediagramsnecessaryforbuildingtheexchange-onlykernel.Duringtheinterpretationofdiagrams(V1)-(V6)the 52


3{10 ).Interpretationofdiagrams(FX1)-(FX8)and(V1)-(V6)givesustheexpression( 3{17 ) ("i"a)("j"b)2Xi;j;a;b'i(r3)'a(r3)'j(r6)'b(r6) ("i"a)("j"b)+2Xi;j;k;a['j(r3)'i(r3)'i(r6)'a(r6)+'j(r6)'i(r6)'i(r3)'a(r3)] ("i"a)("j"a)2Xi;j;a;b['a(r3)'b(r3)'b(r6)'i(r6)+'a(r6)'b(r6)'b(r3)'i(r3)] ("i"a)("i"b)+2Xi;j;k;a'i(r3)'a(r3)'j(r6)'a(r6) ("i"a)("j"a)2Xi;j;a;b'i(r3)'a(r3)'i(r6)'b(r6) ("i"a)("i"b)+2Xi;j;a'i(r3)'a(r3)'j(r6)'a(r6) ("i"a)("j"a)2Xa;b;i'i(r3)'a(r3)'i(r6)'b(r6) ("i"a)("i"b)+2Xi;j;a['i(r3)'j(r3)'j(r6)'a(r6)+'i(r6)'j(r6)'j(r3)'a(r3)] ("i"a)("j"a)2Xa;b;i['a(r3)'b(r3)'b(r6)'i(r6)+'a(r6)'b(r6)'b(r3)'i(r3)] ("i"a)("i"b) (3{17) Expression( 3{17 )forh(r3;r6)exactlycorrespondstotheexpression,,butthediagrammaticderivationrequiresfarlesseortandisunambiguousintermsofsignsandnumericalfactors.Sinceallthediagramscontainonlyonecontour,itispossibletomakesummationsinequation( 3{17 )onlyoverspatialorbitals,and,asaresultf(r1;r2)andf(r1;r2)willappearseparately.Theexchange-onlykerneldoesnotcontainthef(r1;r2)part,whichisacriticaldierencebetweentheexchangeandcorrelationkernels. 53


3{18 ). (3{18) Diagrams(VC1),(VC2),(VC5),(VC6)and(VC15)of( 3{18 )haveanexternalfactor4,diagrams(VC9)and(VC10)havenofactor,whiletherestofthediagramshaveafactorof2.Togetthecorrelationkerneldiagramsweneedtotakefunctionalderivativeswithrespecttothepotentialfromdiagrams(V1)-(V15).Afterapplyingthediagrammaticrulestodiagram(VC1)andtheuseofrelation( 3{13 ),diagrams(VC1-1)-(VC1-16)ofsets( 3{19 )and( 3{20 )willappear. (3{19) 54


Dierentiationofdiagram(VC2)producediagrams(VC2-1)-(VC2-16)ofset( 3{21 ).Thesediagramshavethesameskeletonstructure,butdierentpositionsofindices. (3{21) Aftertakingthefunctionalderivativesfromdiagram(VC3)diagrams(VC3-1)-(VC3-14)( 3{22 )willappear. (3{22) 55


3{23 )willappear. (3{23) Diagram(VC5)producesdiagrams(VC5-1)-(VC5-16)( 3{24 )afterdierentiation. (3{24) Dierentiationofdiagram(VC6)generatediagrams(VC6-1)-(VC6-16)( 3{25 ), (3{25) 56


3{26 ). (3{26) Finally,diagram(VC8)producesdiagrams(VC8-1)-(VC8-14)( 3{27 ). (3{27) Inthesamewaytherestofthediagramsshouldbedierentiated.SincetheFockoperatordependsuponoccupiedorbitals,F-ringsonthediagrams(VC9)-(VC15)mustalsobedierentiated.Diagram(VC9),aftertakingfunctionalderivatives,producesdiagrams(VC9-1)-(VC9-10)( 3{28 ). (3{28) 57


3{29 ). (3{29) Diagram(VC11),producesdiagrams(VC11-1)-(VC11-16)( 3{30 ). (3{30) 58


3{31 ) (3{31) Dierentiationofdiagram(VC13)givesusdiagrams(VC13-1)-(VC13-15)( 3{32 ) (3{32) 59


3{33 ) (3{33) Thelastsetofdiagrams(VC15-1)-(VC15-15)canbeproducedbydierentiationofdiagram(VC15)( 3{34 ) (3{34) Togetherwithdiagrams(V1)-(V6),alldiagramspresentedinthissectionformaset,whichisnesessarytoconstructthecorrelationkernel.TheinterpretationofalldiagramsisgiveninAppendix A 60


(3{35) Diagrams,whichcontainonlyonecontourcannotproducehcomponentsafterdierenti-ation.Threeofthediagrams(VC5)-(VC8)afterdierentiationcanproduceonlydiagramscontributingtothehorhparts.Thediagrams(VC9-1)-(VC9-7),(VC9-9),(VC9-10),(VC10-1)-(VC10-7),(VC10-9),(VC10-10),(VC11-1)-(VC11-10),(VC11-13)-(VC11-16),(VC12-1)-(VC12-10),(VC12-13)-(VC12-16),(VC13-1)-(VC13-12),(VC13-14),(VC13-15),(VC14-1)-(VC14-12),(VC14-14),(VC14-15)havethesameproperty.Diagramscontaining2ormorecontourscanproduceallspincomponents.Setsofdiagrams(VC1-7)-(VC1-16),(VC2-7)-(VC2-16),(VC3-7)-(VC3-14),(VC4-7)-(VC414),(VC15-7)-(VC15-12)havetwocontours,butboth^-functionspresentinoneofthecontoursmeansthatthesediagramsmakeacontributiontohorhparts,buthaveanextrafactorof2.Thatfactorappearsaftersummationoverallspin-orbitalsofthesecondcontour.Diagrams(VC1-1)-(VC1-6),(VC2-1)-(VC2-6),(VC3-1)-(VC3-6),(VC4-1)-(VC4-6),(VC9-8),(VC10-8),(VC11-11),(VC11-12),(VC12-11),(VC12-12),(VC13-13),(VC14-13),(VC15-1)-(VC15-6),(VC15-4)-(VC15-15)have^functionsondierentcontours,sotheycontributetobothhandhparts.Diagram(VC15-13)haveanadditionalfactorof2andalsocontributeintobothspinparts.Tobuildallspinpartsofthecorrelationkernelweneedtosubstitutetheabovediagramsintoequation( 3{35 )Thenextessentialpropertyoftheexchangeandcorrelationkernelsisthesymmetrywithrespecttopermutationofitsarguments 61


3{35 )and 3{35 )weagainneedtoanalyzethestructureofthediagrams.Diagrams(VC1-2),(VC1-5),(VC1-7),(VC2-1),(VC2-6),(VC2-7),(VC5-2),(VC5-7),(VC5-9),(VC6-2),(VC6-7),(VC6-9),(VC11-6),(VC12-6),(VC13-14),(VC14-15),(VC15-13),(VC15-11),(VC15-12),(V5)and(V6)aresymmetricwithrespecttothepermutationofvariablesr1andr2.Afterconsiderationoftherestofthediagrams,itiseasytoseethatforeachdiagramthereisanotherdiagram,whichdiersfromtheinitialonebyonlytheinterchangedvariables,r1andr2.Suchpairsprovideinvariancewithrespecttopermutationofvariables.DuringtheconstructionofthecorrelationkernelKohn-Sham(KS)orbitalsandorbitalenergiesareused.Thisimpliesthefactthatourzero-orderhamiltonianischosentobetheKohn-Shamchoice 1{65 )shouldbesolved.IntheadiabaticapproximationithastheRPA-likeform.Thecriticalnewquantityistheexchange-correlationkernel,f(xc);whichcanbeseenfromtheforegoingisaquitecomplicatedquantityifweinsistuponobtainingitrigorouslyforOEP-MBPT(2).ThediagonaldependenceofAonaitellsusthatifwearetogetgoodexcitationenergies, 62


16 ]whenaccurateexchange-correlationpotentialsareused.Inthissense,theorbitalenergiesinKS-DFTshouldhaveacertainmeaning.Furthermore,aspointedoutinref[ 16 ],wecanalsoconsiderthisequationtooeraKoopmans-likeapproximationtotheprincipalionizationpotentials,since,barringpathologicalbehavior,whenweallowanelectrontobeexcitedintothecontinuum,itsorbitalawillthenhavenooverlapwiththeboundorbitalmatrixelementsofhKSandthekernel.Consequently,weareleftwithnothingbut-iintheTDDFTequations.Sointhis'sudden',adiabaticapproximation,theKSorbitalenergiesshouldoeranestimateforeachoftheprincipalIp's,notjustthehighest-occupied(homo)one.WhenbasedupontherelativelycorrectVxcobtainedfromabinitiodft[ 16 17 ]thisestimateissuperiortoKoopmans'theoremforthehomoandtherstfewvalenceIp's,butisinferiorforthecoreorbitals[ 26 ].SeealsoChong,etal[ 25 ].However,theOEP2semi-canonical(sc)abinitiodft[ 16 17 ]approximationhasthedistinctadvantagethatitusesamuchbetterbehavedunperturbedHamiltonianthantheusualKSchoice,H0=PihKS(i).ToillustratetheevaluationofthekernelandthesolutionofTDDFTequations,weconsidertheNeatom.AlltheexcitedstatesinNecorrespondtoRydbergexcitedstates.Toobtainreasonablevaluesrequiresaquiteextensive,diusebasisset.WechoosetostartwiththeROOS-ATZPatomicnaturalorbitalbasis[ 27 ]consistingof(14s9p4d1f)primitivegaussianfunctionscontractedtoa[5s4p3d1f]set.Thisunderlyingbasiswasthenaugmentedbyasetofeven-tempereddiusefunctions[3s3p3d],withexponentialparameters=abn;as=0:015;ap=0:013;ad=0:012;b=1/3.Theauxiliarybasisischosentobethesame,butwithoutthepandddiusefunctions,sinceforthedescriptionoftheexchange-correlationpotential,diusefunctionsarenotnecessary.TheorbitalenergyestimatesareshowninTable 3-1 .Theionizationpotentialequationofmotioncoupled-cluster(IP-EOMCC)resultinthisbasisfortheNehomoIpis21.3eVwiththeexperimentalvaluebeing21.5645 63


Orbitalenergiesandzero-orderapproximationstoexcitationenergies Exchange-onlyOEP-MBPT(2)orbitalIP[ 28 ]Orbital"a"HOMOOrbital"a"HOMOExcitationenergiesenergiesenergies(exp) 2s48.42-46.19-43.252p21.56-22.51-20.233s-5.1517.36-7.5812.6516.853p-2.5919.92-3.7216.5118.704s-1.9520.56-2.6617.5718.733d-1.4921.02-2.0518.1818.97 [ 28 ].ThiscomparestotheOEPxvalueof22.51andOEP-MBPT(2)valueof20.23.Theremainingunoccupied,butnegativeenergy3sorbitalischangedbyover2eVduetotheMBPT(2)correlation.TheOEP2(sc)abinitodftvaluechangesthisto5.18eV,attestingtothepoorconvergenceofthestandardKSpartitioningoftheHamiltonian.OnceOEP2(sc)calculationsaredoneforthe3p,4s,and3dstates,thereissimilaragreementbetweentheOEPxandOEP2(sc)results,contrarytothoseshowninTable 3-2 forthestandard(ks)choice. Table3-2. ExcitationenergiesofNeatomusingOEP-MBPT(2)Kohn-Shamorbitalenergies TermEOM-CCSDTDDFT-exchange-onlyTDDFTOEP-MBPT(2)Exp[ 29 ]. Despitethedierences,atleastboththeexchange-onlyandtheOEP-MBPT(2)givequalitativelycorrectresultsfortheRydbergseries.Helpingtoensurethisisthefactthattheexchangepotentialhasthecorrectasymptoticbehaviorandtheexactcancellation 64


30 ].WhenwesolvetheTDDFTequationswiththekerneldevelopedinthispaper,weobtaintheresultsforexcitationenergiesshowninTable 3-2 .Ingeneral,justasweobservedfromtheorbitalenergydierences,theOEP-MBPT(2)resultstendtofallonthelowsideofexperimentandtheOEPxonthehighside.Infact,exceptfortheexcitationenergyforthelowest1Pstate,wheretheveryloworbitalenergyof-7.58biasestheresults,anaverageofthetwowouldseemtobeaboutright.Thereare,however,stilldiciencesinthebasisset,asseenbytheEOM-CCSDresultscomparedtoexperiment,wherebesidestheadditionalcorrelationeectsintroducedbyEOM-CCSDT[ 31 ],thefurtherextensionofthediusefunctionswouldremovethe~1eVerrorinthehighestlying(1S)state.Ofcourse,thedependenceofTDDFTonthebasisandthatforatwo-particletheorylikeEOM-CCshouldbequitedierent.ThegreatsensitivityoftheresultstotheorbitalenergiesfromtheunderlyingKS-DFTcalculationcanbefurtherappreciatedbysimplytakingtheenergiesfromtheOEPxresultsandusingthemintheevaluationofthekernelandthematrixelementsintheTDDFTequations.TheseresultsareshowninTable 3-3 .ThenearcoincidenceofresultsforOEPxandOEP-MBPT(2)isapparent,withbothnowbeingtoohigh.AsimilarexperimentcanbemadewhereweuseorbitalsandorbitalenergiesfromOEP2(sc)abinitodftresultsasdescribedelsewhere[ 16 ],toobtaintheresultsinTable 3-4 .ThatistheTDDFTequationsandthekernelareassumedtobethesame,butweuseorbitalsandorbitalenergiesobtainedfromOEP2(sc).HerewealsoshowtheresultsfromstandardTDDFTapplicationsusingtheLDAandB3LYPfunctionalsforcomparisonpurposes.Clearly,wehaveimprovedresultsatboththeOEPx(sc)andOEP2(sc)levels,withthesamepatternoftheformerbeingtoohigh,butlessso;andthelatter,toolow,butbetterthanbefore.Itisapparentthattheproperwaytoachievethebenetsof 65


Table3-3. ExcitationenergiesofNeatomusingexchange-onlyorbitalenergies TermTDDFT-exchange-onlyTDDFTOEP-MBPT(2)Exp[ 29 ]. Table3-4. ExcitationenergiesofNeatomusingorbitalenergiesandorbitalsfromOEP2(sc).AllequationsforTDDFTarethesame TermEOM-CCSDB3LYPLDATDDFT-TDDFT-Exp[ 29 ].[ 32 ][ 32 ]exchange-OEPonly(sc)MBPT(2)(sc) ThecornerstoneofOEP(sc)forexchange-correlationpotentialsisthatwecanimposetheconditionthattheKSdeterminanthastogivetheexact,correlateddensity.ThenthisdensityconditionprovidesequationsthatdeneVxcforagivenfunctional[ 17 ].Sincenoenergyvariationalconditionorfunctionalderivativeisused,wecanchangethechoiceofH0fromtheKSonetothesemi-canonical(sc)onethatoersgreatlyenhancedconvergenceofperturbationtheory.ButthenthereisnoapparentimmediatecorrespondencetoafunctionalderivativeasthereisinthestandardKStheory.Asimilar 66


16 17 26 ],theradiusofconvergenceispoor,causingpoorconvergence[ 16 ]indeterminingthecorrelationpotential.Nevertheless,therststepintheunderlyingframeworkhasnowbeendenedtoapplyTDDFTwithOEPcorrelationpotentials.However,thecomplexityisgreat,thoughimpositionofadditionalrigorousconditionsmightresultinsimplicationofthekernel.Lackingsuchsimplications,thequestionarisesaswhetherthisisacaseofdiminishingreturnstoevaluatesome203diagramstoretaintheattractive,one-particle,correlatedstructureofDFT?First,theanswersforNearenotverygoodattheOEP-MBPT(2)level.Furthermore,thebasissetdependenceofOEPmethodswhendoneingaussianbasisisseveretoevengettheVxcright[ 16 ].ThefailureofmostsuchOEPcalculationstosatisfytheexactHOMOcondition=isacaseinpoint[ 16 26 ]. 67


31 ])arelesstime-consumingthanisthepresentcalculation.Aslongasthekernelforarigorousorbital-dependentcorrelationpotentialisthiscomplicated,two-particlewavefunctiontheorieslikeEOM-CCarebothsuperiorandeasiertodo.ButbeforewecanfurtherexploittheinterfacebetweenDFTandwavefunctiontheorytothebenetofboth,itisrequisitetohavetheorbital-dependentkerneldened.This,anditsinitialevaluation,iswhatthischapteroers. 68


^(r)=Xp;q<'pj(rr1)j'q>a+paq(4{1)UsingWicktheorem,werewritethisoperatorinnormalformwiththeKohn-ShamdeterminantastheFermi-vacuum: Thesecondtermisks,hence,thersttermwillbecalledthedensitycorrection: (r)=Xp;q<'pj(rr1)j'q>fa+paqg(4{3)SincetheconvergedKohn-Shamschemegivesanexactdensity,allcorrectionstothisdensitymustbeequaltozero.Hence,ifweconstructaneectiveoperatorofthedensityusingMBPT,thecorrectiontothedensitymustvanishinanyorder[ 17 24 ].TherstandsecondordersoftheMBPTdensityconditioncanbewritteninthefollowingway (4{5) UsingtheKSpartitioningofthehamiltonian[ 13 ]andtakingintoaccounttherelationfpq="ppq+<'pjVHFxVx()Vcj'q>,itispossibletoderivethesameexpressionsfor 69


13 33 ].However,thisisanumericallyhopelessprocedurewithoutanyresummationofterms:(1)itsuersfromaddingalargediagonaltermintotheperturbation;(2)MBPT(2)isnotboundedfrombelowandanyeorttousethevariationalconditionE(2)c However,toavoiddealingwithnon-diagonalH0,asemi-canonicaltransformationwillbeperformedtoobtainthemoreconvenientzeroth-orderhamiltonian, (fii+fjjfaafbb)(fifc)2;XXi;j;k;a;b()'k(r1)'a(r1) (fii+fjjfaafbb)(fkkfaa);XXi;j;k;a;b()'k(r1)'i(r1) (fii+fjjfaafbb)(fkk+fjjfaafbb) 70


(fii+fjjfaafbb)(fii+fjjfccfbb)+Xa;b;ifibfia'i(r1)'a(r1) (fiifaa)(fiifbb)Xi;j;afajfij'i(r1)'a(r1) (fiifaa)(fjjfaa)+Xa;b;ifiafib'a(r1)'b(r1) (fiifaa)(fiifbb)Xi;j;afaifaj'i(r1)'j(r1) (fiifaa)(fjjfaa)Xi;j;a;bfbj(+)'i(r1)'a(r1) (fiifaa)(fjjfbb))X1(r1;r) (4{9) where 13 17 ]. 4.2.1TotalEnergiesGround-stateenergiesarecalculatedforseveralsystemswithresultspresentedinTable 4-1 .TheuncontractedROOS-ATZPbasisisusedfortheNeatom,whiletherestoftheatomsandmoleculesarecalculatedusingtheuncontractedROOS-ADZPbasis[ 27 ].Allthemoleculesandtheirionsareconsideredtobeintheequilibriumgeometry[ 36 ]ofthecorrespondingneutralsystem,exceptwhenpotentialenergycurvesareconsidered.ForabinitioDFTcalculationsexchange-correlationpotentialsareusedfrom[ 13 ]andequations( 4{8 )-( 4{9 ).ForthecomparisonenergiesobtainedfromKSDFTwiththePerdew-Burke-Ernzerhof[ 34 ](PBE)exchange-correlationpotentialandcoupled-clusterwithsingle,doublesandperturbativetripleexcitation[ 35 ](CCSD(T))arealsoshowninTable 4-1 .Totalenergies,calculatedwithOEP-MBPT(2),basedonthedensityconditionapproachwiththesemi-canonicalpartitioningaremuchclosertoCCSD(T)thenOEP-MBPT(2)withtheKSpartitioning.Thelattergreatlyoverestimatescorrelationenergy 71


37 ]tothevariationalOEP-MBPT(2)fortheKSpartitioningithasthesameproblems.However,inthecaseofthenon-variational,semi-canonicaldensityconditionapproach,theunboundnessfrombelowislessofaproblem.Thesemi-canonicalchoiceofH0providesgoodapproximationstotheenergyandwavefunctioninMBPT(2)andthecorrespondingtotalenergiesaremuchclosertohighly-accurateCCSD(T)ones,atleastinthechosenbasis.ThecomputationalcostoftheOEP-MBPT(2)methodiscomparablewiththecostofMBPT(2),thescalingofbothmethodsisN5,whereNisthenumberofbasisfunctions.ThisismoreexpensivethanconventionalDFT(scaleslikeN3),butlessexpensivethenCCSD(T),computationaltimeofwhichisproportionaltoN7.Thecorrelationpotential, Table4-1. Totalenergies MP2OEP-KSOEP-semiPBECCSD(T) calculatedwiththesemi-canonicalOEPfortheNeatom[ 16 ]isclosetothecorrespondingquantumMonte-Carlopotential,andwecanexpectthatforothersystems,whereQMCisnotavailable,semi-canonicalOEP-MBPT(2)potentialsshouldbeagoodalternative.Fortheopen-shellcasepotentialsproducedbyKohn-ShamOEPshowanoverestimationofthecorrelationenergy,thesamesituationpreviouslyreportedfortheclosed-shellcase[ 16 ].(Fig 4-1 and 4-2 ).Thecorrect1 72


4-1 and 4-2 )isguaranteedbytheuseoftheColle-Nesbetseedpotentialalgorithm[ 30 ].InallourcalculationsaSlaterpotential[ 38 ]wasusedasaseedpotential,sincetheColle-Nesbetalgorithmrequirestheseedpotentialtobeascloseaspossibletotheactualpotential.TwoseparateSlaterpotentials,oneforalphaandtheotherforbeta,wereused.Foralltheclosed-shellsystemsfromTable 4-1 ,thenumerical Figure4-1. ExchangeandcorrelationpotentialsofLiatom(radialpart).A)Exchangepotential.B)Correlationpotential Figure4-2. ExchangeandcorrelationpotentialsofO2moleculeacrossthemolecularaxis.A)Exchangepotential.B)Correlationpotential resultsobtainedfromOEP-MBPT(2)withthesemi-canonicalpotentialandfromDFT 73


4-2 .InexactDFT,theenergyofthehighestoccupiedmolecular Table4-2. Ionizationpotentials(ine.v.) HOMO,OEP-semiE,OEP-semiHOMO,PBEE,PBEExp Ne21.0121.5913.3521.6921.56N216.8915.0310.2715.4115.58CO13.6813.699.0513.8714.01CN4.193.510.1493.723.86H2O12.3712.737.0812.4612.62N15.2514.528.3014.7314.53Li4.895.383.235.595.39 orbital(HOMO)correspondstothenegativeoftheexactverticalionizationpotential.Inthecaseofthesemi-canonicalOEPtheHOMOenergyisclosetothecorrespondingEvalues,whatcanbeexplainedbythecorrectshapeoftheexchange-correlationpotential(Fig 4-1 and 4-2 andref[ 16 17 ]),whilePBEdoesnotexhibitcorrectbehaviorofpotentialandthus,failstoreproducethecorrectHOMOenergyvalues.FailuretoreproducethecorrectHOMOenergycausestheincorrectintegerdiscontinuityofthePBE(oranyotherGGA)exchange-correlationpotential[ 39 40 ]. 74


4-3 .PotentialcurvesfortheLiH,OHandHFmoleculesareshownonFig 4-3 4-4 and 4-5 .ForallcurvesthedissociationenergieshaveapproximatelythesamelevelofaccuracyasintheMBPT(2)case,butthesemi-canonicalOEP-MBPT(2)improvestheshapeofthecurves. Table4-3. Dissociationenergies(inkJ/mol) PBEOEP-semiExp. 41 42 ]arereported,usingtheuncontractedROOS-ADZPbasissets:13s9p3dforcarbonand8s4p1dforhydrogen.Equilibriumgeometriesforbothstatesaretakenfromreference[ 41 ].Theenergiesforthetwostates,calculatedwithdierentmethodsarepresentedinTable 4-4 .Therearesignicantdierencesintheenergyseparationcomparedtotheexperimentalvalue(8.998kcal/mol)forallofthepresentedmethods.Thisisabasissetissue,butcanalsobeexplainedbythefactthatthesingletstateofmethylenehasasignicantcontributionfromtwo-determinants[ 41 ],soMBPT(2)isamuchpoorerunderlyingapproximationfortheirdierencethaninnite-orderCCSDorCCSD(T),andespeciallythetwo-determinantCCSD(TD-CCSD)results[ 41 ].WheretheabsolutevaluesofPBEenergiesarenottoogoodforthetwostatesofCH2,thedierenceisconsistentwithMBPT(2)anditsOEP-generatedabinitiodftform. 75


Singletandtripletenergiesofmethylene CCSD(T)-39.127004-39.10975110.826CCSD-39.123412-39.10456911.824OEP-semi-39.103994-39.07915215.588PBE-39.108321-39.08344815.610MP2-39.102565-39.07767415.619 76


LiHpotentialenergycurve. 77


OHpotentialenergycurve. 78


HFpotentialenergycurve. 79


5 6 ](DFT)isawidelyusedmethodforground-stateenergiesandotherproperties.Thetime-dependentextensionofKohn-ShamDFT,whoserigorousfoundationhasbeenquestioned[ 43 ],isaverypopularmethodforthedescriptionofexcitedstatesenergies,sinceitcanoftenprovideaccurateresultsbythediagonalizationofasingle-excitationdimensionalmatrixasintime-dependentHartree-Fockormono-excitedCI.HoweverKSDFTandTDDFThavetheirnaturallimitations.Standardfunctionalssuerfromanincompletecancellationofself-interactiontermsintheKSequationsandincorrectlong-rangeasymptoticbehaviorofthepotentials.Bothcanbeparticularlytroublesomeforexcited,particularlyRydbergstates,andionizedstates[ 16 21 48 ],makingitdiculttogetsystematicallyimprovableresultsfromstandardTDDFT.Exact-exchangedensityfunctionaltheoryisbasedontheoptimizedeectivepo-tentialmethodintroducedbyTalmanandShadwick[ 15 ].Inthismethodtheexchangepotentialisdenedasthefunctionalderivativeofthenon-localorbital-dependentex-changefunctionalfromtheHartree-Fockmethod(EXX),ieEX=(1)=VX(1):TheOEPmethodisfreefromtheself-interactionproblem,itexhibitsthecorrect1 44 ] 16 22 45 { 47 ],andthesealternativeapproachescansuerfromnumericalproblems.However,whenevaluatedwithapropertreatmentofthedensity-densityresponsematrixX,mostofthenumericalproblemsthathavebeenencounteredareresolved. 80


14 ],butessentiallymodiedbyBartlettetal[ 16 17 ]toavoidthefailuresofasimplesumofKSone-particleHamiltoniansastheunperturbedproblem,togeneratecorrelationpotentials.Thisisthecornerstoneofabinitiodft.Theconceptualdierenceisthatthedensityconditiondoesnotexplicitlyusethevariationaldetermination,EXC=(1)=VXC(1).Thisdierenceisofcriticalimportanceingeneratingcorrelationpotentialsfromlow-ordersofperturbationtheory.Withoutthatchange,nolow-orderorbitaldependentcorrelationfunctionallikethatfromMBPT2willgenerallywork,butwiththosechanges,itdoesverywell[ 16 48 ].Thosemodicationstotheperturbationtheoryalsopertaintotheexchange-onlycase,thesubjectofthispaper,butforthatproblemthedistinctionsarelessimportant[ 16 ].Oncefollowingthisapproach,whichstartswiththeKSchoiceofH0;thedistinctionsbetweenapplyingthedensityconditionandusingdirectfunctionaldierentiationismoreconceptualthanessential,asthereisacorrespondenceinanyorderofperturbationtheory[ 24 37 ].Thedirectoptimizationprocedureadvocatedbysome,builtupontheabovevaria-tionaldeterminationofthefunctionalderivative,thoughformallyequivalent,diersfromtheoriginal,X-basedOEPinthedetailsofimplementation.However,aswaspointedoutbyStaroverovet:al[ 45 ]undercertaincombinationsofmolecularandauxiliarybasissets,particularlywhenthelatter'sdimensionislargerthanthatfortheformer,thedirectoptimizationmethodcangivetheHartree-Fockenergyanddensity.Thelattercanbeviewedasatrivialsolution,asitcanbeshowntocorrespondtothesolutionofaweightedleast-squaresexpression.Ontheotherhand,theX-basedOEPwithproperhandlingis 81


49 ]observestheoccurrenceofadegen-eracyinperformingOEPcalculations,andtoinvestigatethis,intentionallyconstructaproblemwheretheHOMOandLUMOorbitalsoftheKSproblemaredegenerate.Subjecttodirectminimization,theyproceedtoreportonthefailureoftheOEPprocedure.Ontheotherhand,itisshowninthispaperthattheoriginalXbasedapproachhandlesthisproblem,too.Theexchange-onlytime-dependentOEPhasbeenconsideredbyGoerling[ 20 ][ 21 50 ].However,allresultstodateareforclosed-shellsystems.Tofurtheraddresstheexactlocal(OEPx)exchangeversustime-dependentHartree-Fock(TDHF),TDOEPxisgeneralizedtotreatexcitedstatesforopen-shellspecies.ResultsfromadiabaticTDOEPxareingoodagreementwithTDHFforbothexcitationenergiesandpolarizabilities,however,itisshownthatcharge-transferstatescannotbeproperlydescribedinTDDFT.ForstandardDFTmethods,gradient-corrected,hybrid,etc.thishasbeennoted,andimprovedupon[ 51 ],butstandardmethods,unlikeTDOEPx,stillsuerfromotherlimitations,liketheself-interactionerrorandtheincorrectlong-rangebehaviorofthepotentialsandkernels.Inanapproachthatgivesthe'rightanswerfortherightreason'theseexactconditionsarerequisite.InTDOEPxtherearenosucherrors.Hence,failuresofTDOEPxcomparedtoTDHFhavetobeexclusivelyduetothelocalversusnon-localexchangeoperator,whichisdemonstrated.Finally,C6coecientsareobtainedforopen-shellsystemsfromtheTDOEPxfrequency-dependentpolarizability. 82


(5{3)bhS(1)=bh(1)+bJ(1)+bVX(1) (5{4)bJ(1)=XjZj(2)1 (5{5)=Z(1) (5{6)KS(1)=Xjj(1)j(1) (5{7)wheretheunoccupiedorbitalsareindicatedbyp=a,b,c,d....Theconditionthatthedensity,(1);betheHartree-Fockone,isthatfia=hHFijbh+bJbKjHFai=0 (5{8)HF(1)=XjHFj(1)HFj(1) (5{9)which,asiswell-known,iscorrectthroughrstorderincorrelationmeasuredrelativetothesumofHFone-particleoperators,H0=Pif(i);duetotheMoeller-Plessettheorem.ForKSorbitals,fiaisnotzero.Hence,iftheobjectiveweretomaximizethesimilarityintheKSandHFdensity,itwouldthenrequireminimizingaquantitycomposedofKSorbitals,relatedto 13 16 17 48 ].Clearly,thisminimizationwouldhavethetrivialsolutionthathijKjai=hijVXjai. 83


13 16 17 48 ],sinceitalsotranscendsanyparticularseparationoftheHamiltonianinperturbationtheory.Inparticular,itisnotsubjecttoHKS0=PihS(i);andtherebyalleviatesthepathologicalbehaviorthatgivesnoconvergencewhendeningsuchcorrelationpotentialsfromMBPT(2),forexample[ 13 ].Fortheexchangeonly,however,sucheectsarelessimportant[ 16 17 ].ThereforethispaperwillusethesumofShamHamiltonians,buttheprincipleforderivationremainsthesame.Hence,requiringthat0=(1)KS=hKSjbR0bVjKSi+hKSjbVR0bjKSi (5{12)R0=Xi;ajaii(ia)1haij+XI

52 ].Inadditiontoitspointwisecharacterrepresentedbythedeltafunction,itdiersfromtheaboveleast-squaresformbythepresenceofthedenominator.Theusualwaytowritetheaboveequationistointroducethenon-interactingdensity-densityresponsefunction, Thisistheusualoptimized-eectivepotentialprocedureofTalmanandShadwick[ 15 22 ],gearedtowardrepresentingtheVXoperatoronanumericalgrid.Theweightfactor,wia(1);isalsointroduced.Notethematrixelementscanbewrittenincongurationspaceas(VX)ai=hajbVXjii 85


5{18 )thenbecomes, (5{24) Usingweightedleastsquares 5{16 ).Asiscustomaryinpractice,thecomputationalformfortheOEPequationcanbeobtainedbytwiceapplyingtheone-electronprojector, 53 ]oftheresponsefunctionisthen 5{18 )thatresultsintheYcolumnmatrix. 86


52 ],unliketheinnitebasisanalogueofX(1,2),whichhastobesingular[ 52 54 ].Thelatterfollowsbecausetheinnitebasiscannotaccountforthearbitraryconstant.However,inanitebasis,Xisill-conditionedbecausethenull-spaceoffunctionsorthog-onaltoVXcanalwaysbeaddedtoVX,butwithzeroweight,andthisrequiresthattheOEPequationsproduceacomputed,numericalzero.Thismakestheequationsdiculttoinvert,recommendingasingular-valuedecomposition(SVD)procedure.ThisSVDshouldnotbeconfusedwiththeneedtoexcludelineardependencyinthecomputationalbasis,whichcanalwaysbeachievedindependently;butthat,too,wouldbeaccommodatedinpracticebytheSVDprocedure.TheroleofXistoimposethepoint-wiseclosenessof^Kand^VX,and,assuch,itiscriticaltothedeterminationofOEPpotentials.IntheabsenceofX,asattemptedbyStaroverov,etal.[ 45 ]unphysicalresultscanbeobtained.Accordingtotheirprocedure[ 45 ],togetHFenergiesanddensitiesfromtheOEPmethod(i.e.tondthetrivialsolutionofEqn( 5{26 )),thenumberofauxiliaryfunctionsusedmustbegreaterorequaltosomenumber,Nmin,whichisthenumberofnon-zeroKiamatrixelements.Inthiscaseequation(8)ofref[ 45 ]willhaveoneormoresolutions.Toillustrate,theuseoftheresponsematrixapproach,wereportinFigure 5-1 theresultsfortheNeatomexchangepotentialsusingthethreebasissetsusedbyStaroverov,etal,plusafourthuncontractedRoosbasisthatweprefer.TheauxiliarybasesarethesameasthemolecularonesandinallcasesthenumberofauxiliaryfunctionsislargerthanthecorrespondingNmin,necessaryforthesolutionofequation(8)ofref[ 45 ].ThetotalenergyvaluesfromTable 5-1 andcorrespondingpotentialsfromFig 5-1 arecompletelyoppositetotheresultsofref[ 45 ]becauseofthebrokenpointwiseclosenessandtheabsenceoftheself-consistentsolution.NotethatoncethepointwiseclosenessisimposedviaX,andaself-consistentsolutionobtained,thereisnoproblemfortheproperSVD-basedOEPprocedure. 87


ExchangepotentialsofNeatom,obtainedindierendbasissets Table5-1. Total(ina.u.)andorbital(ine.v.)energiesofNeatom HOMOBasisNauxNminE(OEP)E(ref[ 45 ])OEPHF AUG-CC-PVTZ179-128.533191-128.533273-33.58-23.16AUG-CC-PV5Z2515-128.545887-128.546786-14.01-23.14AUG-CC-PV6Z2518-128.546287-128.547062-17.80-23.14ROOS-ADZP6139-128.545016-128.546596-23.74-23.14 InspiredbyanobservationofJiang[ 55 ]ofHOMOandLUMOdegeneracyarisinginOEP,Rohretal[ 49 ]oeredaspeciccombinationofmolecularandauxiliarybasis,thatshowedtheireectforHe,thenusingthedirectoptimizationalgorithmproducesdegenerateHOMOandLUMOorbitalenergies.WeperformedOEPcalculationfortheHeatomusingtheoptimizedauxiliarybasisfromTable1ofref[ 49 ],whichcausedtheirdegeneracy,butinsteadofusingthedirectminimizationalgorithmofBroyder-Fletcher-Goldfarb-Shanno(BFGS)thesolutionoftheOEPequationissolvedbytheinversionofXwiththeSVDprocedure.TheseresultsarepresentedinTable 5-2 ,demonstratingthat Table5-2. TotalandorbitalenergiesofHeatom OEPHFOEP2-KSOEP2-sc Totalenergy,a.u.-2.86115365-2.86115365-2.90524738-2.89488588HOMOenergy,e.v.-24.970-24.970-24.529-24.787LUMOenergy,e.v.3.59115.5613.6713.618 88


5-2 ,eventhenotoriouslyunstablesecond-orderOEPprocedure,basedonHKS0hasnoparticulardiculties,either.Sincethedirectminimizationproce-dureofYangandWu[ 46 ]isbasedonthesameequationastheregularOEPmethod,alloftheproblems,describedinref[ 49 ],arecausedbyusingtheBFGSalgorithminsteadofsolvingtheOEPFredholmequationwiththeSVDprocedurefortheX.Thatmightappeartobesurprisingsinceh=l,wouldmeanthattheresolventoperatorismanifestlysingular,clearlynotthatappropriatetoanon-degenerateperturbationtheoryfunctional,whichhasalwaysbeentheintentofsuchorbitaldependentexpressions.However,itisalsoapparentfromthestandard,chain-ruledierentiationusedbyRohretal[ 49 ],namely 44 ]whomadesuchanapproximationintheGreen'sfunctionwherethedenominatorconsistsofpqbeforesimplifyingthefromtotheoccupied-virtualseparationinvokedhere[ 16 54 57 ].)SeeGritsenko,etal[ 56 ].Suchaconstantenergydenominatorisarelativelypainlessapproximationhere,sinceavgappearsonbothsidesofeqn.(18),makingitdisappearfromtheequationforVX:Oncethatisdone,aresolutionoftheidentitycanbeinvokedtoeliminatethevirtualorbitalstogive, 89


59 ]Ofcourse,usingtheusualnitebasissetcomputationaltoolsofquantumchemistry,thedistinctionislessimportant.Thisformulahasrecentlybeenapplied[ 58 ].BecausetheOEPequationisapoint-wiseidentityhavingtobezeroformallyforallxinanitebasissetitcorrespondstoamany-to-fewmappinginthegeneralcase.ThedependenceuponthedeltafunctionmakestheOEPproceduresensitivetotheauxiliarybasissetusedinthecalculation.However,theSVDprocedurehandlesthenullspacefunctionsplusanypotentiallineardependencyinthebasissetinafairlyautomaticway[ 60 ].IftheXmatrixisremoved,thisisnotthecase,andsomealgorithmscanresultinunphysicalortrivialsolutionsundercertaincombinationsofauxiliaryandmolecularbases.Choosingtoinvoketheaverageenergydenominatorandtheresolutionoftheidentitybeforetheleast-squaresmminimization,willnotgiveEqn(24).However,onceweeliminatethedeltafunctionthatmakestheapproximationpoint-wise,wereducetheproblemtotheminimizationofthevariance minVar=minXihij(bK+bVX)2jiiXi;jhijbK+bVXjjihjjbK+bVXjii(5{34)thatcanbeusedtodeneaVX,butonethatisnotpoint-wise,and,consequently,doesnotsatisfyalltheconditionsabove.ThisfurthersimplicationoftheweightedleastsquaresapproachinEqn( 5{26 )hasalsobeenconsideredrecently[ 58 ].Ultimately,oneprimarycriteriaforthebestexchange-onlypotentialshouldbethesatisfactionoftheJanaktheorem.AsshowninTable 5-1 ,andpointedoutpreviously[ 16 48 ]thisisdiculttoachieveinanynormalbasisset. 90


5.2.1TheoryandImplementationWithintheadiabaticapproximationtheTDOEPequationshavethefollowingmatrixform( 5{35 ) 5{36 ) where with (g)=2Xi;a;b(Kba+VOEPba) ("i"a)("i"b)+2Xi;j;a(Kia+VOEPia)(+) ("i"a)("j"a)2Xi;j;a;b(+) (5{39) Oncetheeigenvalueproblem( 5{35 )issolved,wehavethesetofexcitationenergies,!. 91


5{40 )shouldbesolved 5{40 )shouldbesolvedfortheimaginaryfrequenciesi!fromwhichthecoecientscanbycalculatedby 3(xx(i!)+yy(i!)+zz(i!)).Integrationofequation( 5{43 )wascarriedoutbyGauss-Chebyshevquadrature.Toensurethecorrect1 30 ]wasused.Asaseedpotential,weusedtheoneproposedbySlater[ 38 ] 36 ].Forallcalculationstheuncontracted 92


27 ]hasbeenused.ResultsforexcitationenergiesarepresentedinTable 5-3 Table5-3. Excitationenergies(V-valencestate,R-Rydberg) TDHF"OEPi"OEPaTDOEPSVWNCCSDExp CN2(V) AsfollowsfromthersttwocolumnsofTable 5-3 ,TDHFandTDOEPxproduceresultsofapproximatelythesameaccuracy,aswouldbeexpectedbythefactthatthelocalexchangeinTDOEPismeanttobeakindofleast-squaresttotheHFnon-localexchangepotential.Thiscertainlymakestheoccupiedorbitalsquitesimilar,butthespectrumofunoccupiedorbitaleigenvaluesisverydierent,asshowninTable 5-4 ,sinceTDOEPxwillgenerateaRydbergtypeseriesinsteadofanythingliketheHFvirtuals.Asiswellknown,thelatteraredeterminedinapotentialofn-electrons,makingthemappropriateforelectronattachedstates,whiletheoccupiedonesfeelapotentialofn-1electrons.Tothecontrary,theorbitalsobtainedinOEPxhavethesamepotentialforanelectronintheoccupiedandunoccupiedorbitals,whichiswhythelattermorenearlysimulateRydbergstates,assomeoftheunoccupiedorbitalswillhavenegativeorbitalenergies.Ofcourse,inanitebasisaslongasthespaceseparatelyspannedbytheoccupiedandtheunoccupiedorbitalsisthesame,therewouldbenodierenceintheresults,thoughthediagonalvalues(butnotthetrace)wouldchange.Thesignicant 93


Orbitalenergies(ine.v.)ofNeatom OEPxHFEOM-CCSD 1s-839.67-891.78-871.112s-47.36-52.52-48.112p-23.74-23.14-21.283s(3p)-5.213.663.703p(3s)-3.113.943.624p5.6616.0815.66 discrepancieswiththeexperimentaldatacanbeexplainedbythefactthatelectroncorrelationeectsarenotintroducedbyeitherTDHForTDOEP.Togetsomemeasureofthecorrelationeect,coupledcluster(CC)resultsfromEOM-CCSDarereported[ 61 ],andtheSlater-Vosko-Wilk-Nusiar(SWVN)functionalisalsoused.Thelatterproducesgoodresultsforvalencestates,buttheenergiesoftheRydbergstatesareunderestimated.ThisunderestimationisaconsequenceoftheincorrectasymptoticbehavioroftheSWVNpotential,forwhichJanak'stheoremisnotsatised(seeTable 5-5 )andisawell-knowncharacteristicthatstandardexchange-correlationpotentialsdonotoeraKoopmans-typeapproximationfortheorbitalenergies.HoweverforOEPexchange-onlyandwithcorrelation,ithasbeenproventhattheorbitalenergiesoerameaningfulKoopmansapproximationtoprincipalIP'sforthevalenceandmid-valencestates[ 16 ],soRydbergstatesshouldpotentiallybebetterdescribed.Ofcourse,denitivecomparisonwouldrequireOEPwithcorrelationaspresentedelsewhere[ 62 ],andthisisseenintheOEP2-scorbitalenergies[ 16 48 ],butafullTDOEPtreatmentofexcitedstates[ 62 ]requiresthecorrelationkernel.Thisdoesnotappeartobeaviablerouteyet. Table5-5. Ionizationenergies(ine.v.) HOMOEHFOEPSVWNHFOEPSVWNExp CN14.1614.619.7916.2216.3214.8013.60CH310.4611.195.388.988.9910.099.84N15.5316.148.4113.8913.8915.0014.53 ResultsforthepolarizabilitiesandC6coecientsarepresentedinTables 5-6 and 5-7 .Inallcases,thestaticpolarizabilitiesobtainedwithTDOEPareveryclosetothe 94


63 ]andDreuwetal[ 51 ].Accordingtothelatteranalysis[ 51 ]eveniftheexactexchange-correlationfunctionalwereknown,itwouldstillbeimpossibletogetaproperdescriptionofcharge-transferexcitationswithinTDDFT.ConsidertheexampleofHe...Be.Thecharge-transfer1statecalculatedwithTDHFandTDOEP,takenwiththeTamm-Danco(mono-excitedCI)approximationforsimplicity.ForbothatomstheuncontractedROOS-ADZPbasissetwasused.Potentialcurvesofthe1charge-transferexcitedstatearepresentedonFig 5-2 .Thepotentialcurve,calculatedwiththecongurationinteractionsinglesmethodexhibitsthecorrect1 5{36 )areequaltozeroandtheexcitationfrequencyisequaltothedierenceoftheLUMOorbitalenergyoftheBeatomandtheHOMOenergyoftheHeatom.WhatwasnotaddressednumericallyinDreuwetal[ 51 ]wastheeectofhavingaproperself-interactionandasymptoticbehaviorinDFT,whichcanonlybeachieved 95


A1charge-transferedexcitedstateofHe...Be withOEPmethods.WereportthoseinFig 5-3 .Obviously,theorbitalenergydierence Figure5-3. LUMO-HOMOorbitalenergydierence behavesverydierentlyforOEPandHFsinceinbothcasestheHOMOorbitalenergycorrespondstoanionizationpotentialapproximation,buttheLUMOenergycorrespondstoaRydbergstateapproximationfortheOEPmethodandtoanapproximateelectron 96


5{36 )ascorrections,itisapparentthatTDDFThasaqualitativelywrongbehaviorineachorderwhendealingwithcharge-transferexcitations.IfonearguesthattheoccupiedorbitalsintheHFandOEPcalculationsarenearlythesame,andtheyspantheoccupiedspace,thentheunoccupiedspacewouldbethesame,too.However,evenifthespaceswerethesame,thedierencebetweenthenon-local^Kandlocal^Vxissucienttokeepthelatterlocalexchangefromeverbeingabletodescribeanelectronattachedstate.Thus,itisnotpossibletoovercomethelocaloperator'sfailureeveninthefullspaceforthecalculations. 16 52 ],anddoesrequiretheuseofageneralized(SVD)inverseortheequivalent.ItwasalsoshownthatundertherightconditionstheexactHFresultcanbeobtainedbyanOEP-likeprocedureasatrivialsolutionofaweightedleastsquaresprocedure.Anypoint-wiseconditionhastobeamanytofewmappinginanitebasisset.TheeliminationofdenominatorsfromworkingexpressionsforOEPx[ 20 44 56 ]canbedoneasshown,toobtainthecomputationaladvantageofusingonlytheoccupiedor-bitalsintheOEPprocedure.Thedirectoptimizationalgorithmasimplemented,however,hasledtowork[ 45 49 ]thathassuggestedincorrectconclusions.TheTDOEPmethodwasimplementedforspin-polarizedopen-shellsystems.Resultsforseveralopen-shellsystemsshowverysimilartime-dependentbehaviorforthetime-dependentresponseoftheOEPandtheHFpotentials.Thecorrectlong-range 97


64 { 66 ]Theproblemofcharge-transferexcitedstatesinTDDFTrelatestotheinabilityofthezeroth-orderorbitalenergyapproximations,astheHFvirtualorbitalsandTDOEPexcitedorbitalsshowqualitativelydierentbehavior.IftheHFandOEPoperatorswerethesame,apropertreatmentusingthewholespaceshouldbeabletoovercomethislimitation,eveniftheusualzeroth-orderapproximationoeredbyTDDFTwouldbeapoorone.However,thisdierencearisesfromtheformoftheoperator,localversusnon-local,and,assuch,remainsafundamentalproblemfortheDFTmethoditself,ashasbeenobserved[ 51 ]Inconclusion,forallthepropertiesconsideredinthepaper,exceptforcharge-transferexcitedstates,thereseemstobenosignicantdierencebetweentheresultsofTDHFandexchange-onlyTDOEP,asonemightexpect.Hence,thefocusshouldremainonthecorrelationpotentialasitisinabinitiodft[ 16 17 48 62 ]. 98


Staticpolarizabilities(ina.u.) TDHFTDOEPSVWNCCSD CNxx11.9811.9115.6312.50yy11.9811.9115.6312.50zz18.5518.4125.0023.4214.1714.0818.7516.166.576.509.3810.94CO+xx6.666.669.389.38yy6.666.669.389.38zz12.2212.1514.0612.508.528.4910.9410.425.545.504.693.13CH3xx15.0615.0018.7515.63yy14.0413.9818.7515.63zz15.0615.0018.7515.6314.7214.6618.7515.631. Table5-7. IsotropicC6coecients(ina.u.) TDHFTDOEP CN...CN93.9792.06CH3...CH394.2793.40N...N19.2819.21 99

PAGE 100

20 ][ 21 50 ].NumericalresultsinthelatterpapersshowareasonabledescriptionofbothvalenceandRydbergexcitedstates.Reasonablestaticanddynamicpolarizabilitiesarealsoobtained.GoodresultsofTDOEPxforexcitedstatesandpropertiesareobtainedpartlyduetothecorrectasymptoticbehavioroftheexchangepotential,butalsoduetotheeliminationoftheself-interactionerror.Inparticular,OEP-TDDFTxtendstobesuperiortostandardapproacheslikeTDDFTbaseduponthelocaldensityapproximation(LDA)orBecke-Lee-Yang-Parr(BLYP)functionals.Similarly,exchange-onlyOEPwithexactlocalexchange(EXX)[ 22 23 ]hasbeenshowntogreatlyimproveband-gapsinpolymers[ 23 ]AnotheradvantageofOEPbasedmethodsisthatsincevirtualorbitalsintheexchange-onlyDFTaswellastheoccupiedorbitalsaregeneratedbyalocalpotential,whichcorrespondstotheN-1particlesystemduetothesatisfactionoftheself-interactioncancellation,thedierencesbetweenorbitalenergiesofvirtualandoccupiedorbitalsoeragoodzeroth-orderapproximationtotheexcitationenergies[ 16 17 ].ThisisnotpossibleinthecaseofHartree-Focktheory,withoutaddingaVN1potential[ 67 68 ],sinceoccupiedorbitalsaregeneratedbyanN-1particlepotentialandtheenergiesofunoccupiedorbitalscomefromN-electronpotential,andthereby,approximateelectronanities.InKohn-ShamDFT

PAGE 101

^Vxc(r)=Exc 21 50 ].However,onceourobjectiveisthird-ordermolecularproperty,theessentialnewelementinTDDFTisthesecondexchange-correlationkernel,whichinanadiabaticapproximationisdenedasthesecondfunctionalderivativeoftheexchange-correlationpotentialwithrespecttothedensity 37 62 ].InthispaperweusethatformalismandapplyittotheexchangesecondkernelforOEPxbasedtime-dependentDFT.Afterdiscussingsomepropertiesofthesecondkernel,wereportnumericalresultstoobtainhyperpolarizabilities,comparedtothosefromHartree-Fockandcoupledclustersinglesanddoubles(CCSD).Thereisalonghistoryofthetreatmentofhyperpolarizabilitiesandassociatednon-linearoptical(NLO)properties.See[ 69 ]forareview.Inparticular,issuesoftheincorrect 101

PAGE 102

64 65 ]pointstothenecessityofanimprovedtheory.Arigorous(abinitio)DFTanaloguestartswithexchange-only,TDOEPx. 6.1.1Time-DependentDensityFunctionalTheoryResponsePropertiesWhenanexternaltime-dependentelectriceldisapplied,theperturbationcanbewrittenintheform( 6{2 ) (A+B)U=h(6{3)ThematricesA,Bhavethestructure,givenbyEqn( 5{36 )andhis 70 ] 102

PAGE 103

6{6 ) Fpq=hpq+Xa;i(Apqai+Bpqai)Uai ij=Fij 6{8 ) (6{9) Takingintoaccountthefactthat (6{11) where 103

PAGE 104

6{11 )canbewrittenas (6{13) andP1(r2;r6;r7)canbepresentedasasumofdiagrams(F1)-(F6)of( 6{14 ) (6{14)Inthesamewaythethirdtermofeq(16)canberewrittenas (6{15) whereP2(r1;r6;r7)givesthesumofdiagrams(F7)-(F12)of( 6{16 ) (6{16)ThediagrammaticexpressionforQ(r6;r7)containsterms(V1)-(V6)and(FX1)-(FX8)of( 6{17 )and( 6{18 )[ 62 ] (6{17) 104

PAGE 105

Functionalderivativesfromtheset(V1)-(V6)canbepresentedbytwotypesofdiagrams.Thersttypecontainstheexchangekernel,diagrams(F13)-(F18)of( 6{19 ) (6{19)Thesecondtypecontainstheexchangepotential,givingdiagrams(VX1)-(VX8)of( 6{20 ) (6{20) Thenotationr4nr5nr6meansthatforeachparticulardiagramallsixpermutationsofr4,r5andr6shouldbetaken.Thefunctionalderivativesfromdiagrams(FX1)-(FX8)leadtotheset(SX1)-(SX14)of( 6{21 )and( 6{22 ). (6{21) 105

PAGE 106

InterpretationofalldiagramswhichcontributesintothesecondkernelanddetailsoftheimplementationaregiveninAppendix B 38 ] 6{20 )willcanceldiagrams(SX7),(SX8)and(SX11)-(SX14)of( 6{21 )and( 6{22 )becauseoftheHOMOconditionandthebehavioroftheHartree-Fockexchangeforthecaseoftwoelectrons.Diagrams(VX6)and(VX7)willcanceldiagrams(SX1)and(SX2)becauseofEqn( 6{23 ).Diagrams(SX5),(SX6),(SX9)and(SX10)willcanceldiagrams(F1)-(F4),(F7)-(F10)and(F13)-(F16)becauseofEqn( 6{24 ).Finallythesumofthediagrams(SX3)and(SX4)cancelthesumof(F5),(F6),(F11),(F12),(F17)and(F18).Thusforthespecialcaseoftwoelectrons 106

PAGE 107

6{25 )mightbeusedasahintfordevelopingnewdensityfunctionals,asnoneoftheexistingfunctionalsexhibitsuchbehavior. 27 ]areusedandequlibriumgeometriesaretakenfromref[ 36 ].TheresultscalculatedwithdierentmethodsarepresentedinTable 6-1 .Intherstfourcolumnsarestatichyperpolarizabilitiesobtained Table6-1. Hyperpolarizabilitiesofseveralmolecules(ina.u.) HFOEP-HFOEPxLDACCSD LiHzzz312.130314.910312.126621.093691.406xxz201.150207.868201.641404.297204.171COzzz-31.016-24.864-29.455-66.40627.343xxz-3.073-4.660-3.118-13.2035.859CNzzz-1.2990.211-3.906-1.953164.02xxz-17.651-15.646-18.121-62.50068.359H2Ozzz-6.713-6.239-7.812-13.672-5.859xxz-0.497-0.604-1.215-6.138-3.870yyz-10.866-10.32-11.847-23.437-7.645 withHartre-Focknon-localexchangeandwiththetwolocalexchangeoperators:OEP-HF,whichmeansequation( 6{5 )withoutthesecondkernelterm,andOEPx,whichmeansalltermsinequation( 6{5 ).ThehyperpolarizabilityvaluescalculatedwiththeOEPxmethodareingoodagreementwiththecorrespondingHartree-Fockvalues,andgenerallynobetter.ThisgoodagreementcanbeexplainedbythefactthatOEPxmethodisfreefromtheself-interactionerror,hasthecorrectlong-rangeasymptoticbehavior,andthecorrectHOMOvalues.AshyperpolarizabilitiesareresponsepropertiesofthirdordertheircalculationwithintheDFTmethodrequiresverypreciseexchange-correlationpotentials.OnlypotentialswhichsimultaneouslysatisfyalltheoremsandconditionspertainingtotheDFTexchangepotentialscanbeexpectedtoreproduceeventheHartree-Fock 107

PAGE 108

66 ]itwaspointedoutthatnoneoftheexistinghybridandasymptoticallycorrectedfunctionalsarecapableofreproducingHartree-Fockhyperpolarizabilities.Thesemi-empiricalmethod,developedthere[ 66 ]istheonlyknownwaytogetreasonablehyperpolarizabilitieswithconventionalfunctionals.However,theOEPxpotentialsandkernelsaretheonlyrigorous,purelyabinitiowaytoobtainhigh-orderproperties,withinasolelyDFTframeworkoflocalpotentialsandkernels.ForthecalculationofhyperpolarizabilitieswiththeHartree-Fockmethod[ 71 ]thelasttermineq( 6{5 )shouldbedropped.ItsimportancefortheDFTschemecanbeestimatedfromthesecondcolumnofTable 6-1 ,whereOEPhyperpolarizabilitieswerecalculatedwithoutthelasttermofequation( 6{5 ).FortheexampleoftheLiHmoleculetherelativecontributionofthelasttermissmall.However,goingfromclosedshelltoanopen-shell,likefortheCNmolecule,therelativecontributionofthethirdkernelbecomesmuchgreaterand,sometimes,canevenchangethesignofhyperpolarizabilities.ThedierencesbetweentheOEPxandthecoupledclustersinglesanddoubles(CCSD)methodarebecauseintheexchange-onlyOEPmethodelectroncorrelationhasnotbeentakenintoaccount.Forthecalculationofhigh-orderproperties,whichcanbecomparedwithexperimentaldata,accountingfortheelectroncorrelationiscritical[ 69 72 73 ],andgenerallyCCSDisfairlygood,thoughaddingtriplesoersomeimprovement.Nevertheless,thepresentresultsshowwhatkindofpropertiesexchangepotentialsandkernelsmusthavetobeabletodescribehigher-orderpropertieswithintheframeworkofDFT,andasmentionedpreviously,noneofthecurrentsetoffunctionalsforstandardDFTshowsthisbehavior. 108

PAGE 109

16 ]IthasbeenshownthattherigorousexchangetreatmentinTDOEPandTDHFprovidehyperpolarizabilitiesthatcloselyagreewitheachother,asonewouldexpect,althoughthiswillnothappenforcharge-transferexcitedstates[ 51 74 ].ThisclearagreementwithTDHFisincontrasttoconventionalexchangefunctionalsthatseverelyoverestimatehyperpolarizabilities.OEPpotentialsandkernelsarenotonlycapableofdescribingexcitationenergies,butalsohigher-orderproperties.SincenoneofthestandardDFTfunctionalsshowsthecorrectanalyticpropertiesofthesecondkernel,theOEPformulasandresultscanbeusedforthetestingandcalibrationofnewdensityfunctionals. 109

PAGE 110

(A{1) where "i"a)=Zdr'p(r)'q(r)(r) (A{2) (VC11)+(VC22)=2Xa;b;c;i;j;k(+) ("i+"j"a"b)("i"c)("k"b) (A{3) (VC12)=2Xa;b;c;d;i;j (A{4) (VC13)+(VC31)=2Xa;b;c;i;j;k(+) ("i+"j"a"b)("i+"k"a"b)("i"c) (A{5) 110

PAGE 111

("i+"j"a"b)("i+"j"a"d)("i"c) (A{6) (VC15)=2Xa;b;c;d;i;j (A{7) (VC16)+(VC25)=2Xa;b;c;i;j;k(+) ("i+"j"a"b)("i"c)("k"b) (A{8) (VC17)=4Xa;b;c;d;i;j (A{9) (VC18)+(VC28)=4Xa;b;c;i;j;k(+) ("i+"j"a"b)("i"c)("k"a) (A{10) (VC19)+(VC110)=4Xa;b;c;i;j;k(+) ("i+"j"a"b)("i"c)("k"c) (A{11) (VC111)+(VC37)=4Xa;b;c;i;j;k(+) ("i+"j"a"b)("k+"j"a"b)("k"c) (A{12) (VC112)+(VC48)=4Xa;b;c;d;i;j(+) ("i+"j"a"b)("i+"j"d"b)("i"c) (A{13) 111

PAGE 112

("i+"j"a"b)("i"d)("i"c) (A{14) (VC115)+(VC215)=4Xa;b;c;i;j;k(+) ("i+"j"a"b)("i"c)("k"a) (A{15) (VC116)=4Xa;b;c;i;j;k (A{16) (VC21)=2Xa;b;i;j;k;l( (A{17) (VC23)+(VC32)=2Xa;b;i;j;k;l(+) ("i+"j"a"b)("i+"l"a"b)("k"a) (A{18) (VC24)+(VC42)=2Xa;b;c;i;j;k(+) ("i+"j"a"b)("i+"j"a"c)("k"a) (A{19) (VC26)=2Xa;b;i;j;k;l( (A{20) (VC27)=4Xa;b;i;j;k;l( (A{21) 112

PAGE 113

("i+"j"a"b)("k"c)("k"a) (A{22) (VC211)+(VC47)=4Xa;b;c;i;j;k(+) ("i+"j"a"b)("i+"j"b"c)("k"c) (A{23) (VC212)+(VC38)=4Xa;b;i;j;k;l(+) ("i+"j"a"b)("l+"j"a"b)("k"a) (A{24) (VC213)+(VC214)=4Xa;b;i;j;k;l(+) ("i+"j"a"b)("k"a)("l"a) (A{25) (VC216)=4Xa;b;c;i;j;k (A{26) (VC33)+(VC35)=Xa;b;i;j;k;l(+) ("i+"j"a"b)("i+"k"a"b)("l+"j"a"b) (A{27) (VC34)+(VC45)=Xa;b;c;i;j;k(+) ("i+"j"a"b)("i+"k"a"b)("i+"j"b"c) (A{28) (VC36)+(VC43)=Xa;b;c;i;j;k(+) ("i+"k"a"b)("i+"k"c"b)("i+"j"c"b) (A{29) 113

PAGE 114

("i+"k"a"b)("i+"l"a"b)("i+"j"a"b) (A{30) (VC311)+(VC412)=2Xa;b;c;i;j;k(+) ("i+"k"a"b)("i+"j"a"c)("i+"j"a"b) (A{31) (VC312)+(VC411)=2Xa;b;c;i;j;k(+) ("i+"k"a"b)("i+"k"a"c)("i+"j"a"c) (A{32) (VC313)+(VC314)=4Xa;b;c;i;j;k (A{33) (VC44)+(VC46)=Xa;b;c;d;i;j(+) ("i+"j"a"c)("i+"j"a"b)("i+"j"d"b) (A{34) (VC49)+(VC410)=2Xa;b;c;d;i;j(+) ("i+"j"a"c)("i+"j"a"b)("i+"j"a"d) (A{35) (VC413)+(VC414)=4Xa;b;c;i;j;k (A{36) (VC51)+(VC61)=2Xa;b;c;i;j;k(+) ("i+"j"a"b)("i"c)("k"a) (A{37) 114

PAGE 115

(A{38) (VC53)+(VC81)=2Xa;b;c;i;j;k(+) ("i+"j"a"b)("i+"k"a"b)("i"c) (A{39) (VC54)+(VC712)=2Xa;b;c;d;i;j(+) ("i+"j"d"b)("i+"j"a"b)("i"c) (A{40) (VC55)+(VC78)=2Xa;b;c;d;i;j(+) ("i+"j"a"b)("i+"j"a"d)("i"c) (A{41) (VC56)+(VC68)=2Xa;b;c;i;j;k(+) ("i+"j"a"b)("i"c)("k"b) (A{42) (VC57)=2Xa;b;c;d;i;j (A{43) (VC58)+(VC66)=2Xa;b;c;i;j;k(+) ("i+"j"a"b)("i"c)("k"a) (A{44) (VC59)=2Xa;b;c;d;i;j (A{45) 115

PAGE 116

("k+"j"a"b)("i+"j"a"b)("i"c) (A{46) (VC511)+(VC512)=2Xa;b;c;i;j;k(+) ("k+"j"a"b)("i"c)("k"c) (A{47) (VC513)+(VC514)=2Xa;b;c;d;i;j(+) ("i+"j"a"b)("i"c)("i"d) (A{48) (VC515)+(VC615)=2Xa;b;c;i;j;k(+) ("i+"j"a"b)("i"c)("k"b) (A{49) (VC516)=2Xa;b;c;i;j;k (A{50) (VC62)=2Xa;b;i;j;k;l (A{51) (VC63)+(VC71)=2Xa;b;c;i;j;k(+) ("i+"j"a"b)("i+"j"a"c)("k"a) (A{52) (VC64)+(VC88)=2Xa;b;i;j;k;l(+) ("i+"j"a"b)("l+"j"a"b)("k"a) (A{53) 116

PAGE 117

("i+"j"a"b)("i+"l"a"b)("k"a) (A{54) (VC67)=2Xa;b;i;j;k;l (A{55) (VC69)=2Xa;b;i;j;k;l (A{56) (VC610)+(VC79)=2Xa;b;c;i;j;k(+) ("i+"j"a"b)("i+"j"c"b)("k"c) (A{57) (VC611)+(VC612)=2Xa;b;c;i;j;k(+) ("i+"j"c"b)("k"c)("k"a) (A{58) (VC613)+(VC614)=2Xa;b;i;j;k;l(+) ("i+"j"a"b)("l"a)("k"a) (A{59) (VC616)=2Xa;b;c;i;j;k (A{60) (VC72)+(VC73)=Xa;b;c;d;i;j(+) ("i+"j"a"c)("i+"j"a"b)("i+"j"d"b) (A{61) 117

PAGE 118

("i+"j"a"c)("i+"j"a"b)("i+"k"a"b) (A{62) (VC75)+(VC86)=Xa;b;c;i;j;k(+) ("i+"j"a"c)("i+"k"a"c)("i+"k"a"b) (A{63) (VC76)+(VC87)=Xa;b;c;i;j;k(+) ("i+"j"a"c)("j+"k"a"c)("j+"k"a"b) (A{64) (VC77)+(VC84)=Xa;b;c;i;j;k(+) ("i+"j"a"c)("i+"j"a"b)("j+"k"a"b) (A{65) (VC710)+(VC711)=Xa;b;c;d;i;j(+) ("i+"j"a"c)("i+"j"a"b)("i+"j"a"d) (A{66) (VC713)+(VC714)=2Xa;b;c;i;j;k (A{67) (VC82)+(VC83)=Xa;b;i;j;k;l(+) ("i+"k"a"b)("l+"j"a"b)("i+"j"a"b) (A{68) (VC811)+(VC812)=Xa;b;i;j;k;l(+) ("i+"j"a"b)("i+"l"a"b)("i+"k"a"b) (A{69) 118

PAGE 119

(A{70) (VC91)+(VC118)=2Xi;j;k;afaifkj(+) ("i"a)("j"a)("k"a) (A{71) (VC92)+(VC103)=Xi;j;a;bfaifbj(+) ("i"a)("j"a)("j"b) (A{72) (VC93)+(VC102)=Xi;j;a;bfaifbj(+) ("i"a)("i"b)("j"b) (A{73) (VC94)+(VC125)=2Xi;j;a;bfaifba(+) ("i"a)("j"a)("j"b) (A{74) (VC95)+(VC96)=Xi;j;k;afaifka(+) ("i"a)("j"a)("k"a) (A{75) (VC97)=2Xi;j;a;bfaifaj (A{76) (VC98)+(VC153)=4Xi;j;k;a;bfai(+) ("i"a)("j"a)("k"b) (A{77) 119

PAGE 120

("i"a)("j"a)("k"b) (A{78) (VC910)+(VC1310)=2Xi;j;k;a;bfai(+) ("i"a)("j"a)("k"b) (A{79) (VC101)+(VC127)=2Xi;a;b;cfbifac(+) ("i"a)("i"b)("i"c) (A{80) (VC104)+(VC115)=2Xi;j;a;bfibfij(+) ("j"a)("i"a)("i"b) (A{81) (VC105)+(VC106)=Xi;a;b;cfbific(+) ("i"a)("i"b)("i"c) (A{82) (VC107)=2Xi;j;a;bfiafib(+) ("i"a)("i"b)("j"a) (A{83) (VC108)+(VC154)=4Xi;j;a;b;cfib(+) ("i"a)("i"b)("j"c) (A{84) (VC109)+(VC142)=2Xi;j;a;b;cfib(+) ("i"a)("i"b)("j"c) (A{85) 120

PAGE 121

("i"a)("i"b)("j"c) (A{86) (VC111)+(VC112)=2Xi;j;k;afijfaj(+) ("i"a)("j"a)("k"a) (A{87) (VC113)+(VC114)=2Xi;j;a;bfijfaj(+) ("j"a)("i"a)("i"b) (A{88) (VC116)=2Xi;j;k;afjkfki(+) ("i"a)("k"a)("j"a) (A{89) (VC117)+(VC128)=2Xi;j;a;bfijfab(+) ("i"a)("i"b)("j"b) (A{90) (VC119)+(VC129)=2Xi;j;a;bfaifjb(+) ("i"a)("j"a)("i"b) (A{91) (VC1110)=2Xi;j;a;bfaifib (A{92) (VC1111)+(VC155)=4Xi;j;k;a;bfai(+) ("i"a)("j"a)("k"b) (A{93) 121

PAGE 122

("i"a)("j"a)("k"b) (A{94) (VC1113)+(VC1311)=2Xi;j;k;a;bfij(+) ("i"a)("j"a)("k"b) (A{95) (VC1114)+(VC143)=2Xi;j;k;a;bfij(+) ("i"a)("j"a)("k"b) (A{96) (VC1115)+(VC145)=2Xi;j;k;a;bfai(+) ("i"a)("j"a)("k"b) (A{97) (VC1116)+(VC137)=2Xi;j;k;a;bfai(+) ("i"a)("j"a)("k"b) (A{98) (VC121)+(VC122)=2Xa;b;c;ifabfib(+) ("i"a)("i"b)("i"c) (A{99) (VC123)+(VC124)=2Xi;j;a;bfibfab(+) ("i"a)("i"b)("j"a) (A{100) (VC126)=2Xa;b;c;ifabfac (A{101) 122

PAGE 123

(A{102) (VC1211)+(VC156)=4Xa;b;c;i;jfib(+) ("i"b)("i"a)("j"c) (A{103) (VC1212)+(VC151)=4Xa;b;c;i;jfab(+) ("i"b)("i"a)("j"c) (A{104) (VC1213)+(VC1312)=2Xa;b;c;i;jfab(+) ("i"b)("i"a)("j"c) (A{105) (VC1214)+(VC141)=2Xa;b;c;i;jfab(+) ("i"b)("i"a)("j"c) (A{106) (VC1215)+(VC1410)=2Xa;b;c;i;jfia(+) ("i"b)("i"a)("j"c) (A{107) (VC1216)+(VC138)=2Xa;b;c;i;jfia(+) ("i"b)("i"a)("j"c) (A{108) (VC131)+(VC132)=2Xi;j;k;a;bfai(+) ("i"a)("j"b)("k"b) (A{109) 123

PAGE 124

("i"a)("j"b)("j"c) (A{110) (VC135)+(VC149)=2Xi;j;k;a;bfai(+) ("i"a)("j"b)("k"b) (A{111) (VC136)+(VC146)=2Xa;b;c;i;jfai(+) ("i"a)("j"b)("j"c) (A{112) (VC1313)+(VC1514)=4Xa;b;c;i;j;k(+) ("i"a)("j"b)("k"c) (A{113) (VC1314)=2Xa;b;c;i;j;k (A{114) (VC1315)+(VC1414)=2Xa;b;c;i;j;k(+ (A{115) (VC147)+(VC148)=2Xi;j;k;a;bfaj(+) ("j"a)("i"b)("k"b) (A{116) (VC1411)+(VC1412)=2Xa;b;c;i;jfaj(+) ("j"a)("i"b)("i"c) (A{117) 124

PAGE 125

("i"b)("j"a)("k"c) (A{118) (VC1415)=2Xa;b;c;i;j;k (A{119) (VC157)(VC158)=8Xi;j;k;a;bfai(+) ("i"a)("j"b)("k"b) (A{120) (VC159)+(VC1510)=8Xa;b;c;i;jfai(+) ("i"a)("j"c)("j"b) (A{121) (VC1511)=8Xi;j;k;a;bfai (A{122) (VC1512)=8Xa;b;c;i;jfai (A{123) (VC1513)=16Xa;b;c;i;j (A{124) (V1)+(V2)=2Xi;a;b(+) ("i"a)("i"b) (A{125) 125

PAGE 126

("i"a)("j"a) (A{126) (VC5)=2Xi;j;a (A{127) (VC6)=2Xi;a;b (A{128) 126

PAGE 127

6{5 )anauxiliarybasissetdecompositionisused B{2 ) where "i"a(B{3)Twoseparategshouldbeconstructedforeachspincomponent.Expressionsfor,andcanbewrittenintermsofthediagrams(F1)-(F18). =2Xi;j;afia(<'a'j><'i'j>+<'a'j><'i'j>) ("i"a)("j"a)2Xi;a;bfia(<'i'b><'a'b>+<'i'b><'a'b>) ("i"a)("i"b)+2Xi;j;afij<'a'i><'a'j> (B{4) wheretheintegralsinvolvingtheexchangekernelcanbecalculatedusingtheformula 127

PAGE 128

("i"a)("i"b)("i"c)F1=(<'i'c><'b'c><'a'b>+<'i'c><'b'c><'a'b>+<'i'c><'b'c><'a'b>+<'i'c><'b'c><'a'b>+<'i'c><'b'c><'a'b>+<'i'c><'b'c><'a'b>) (B{6) ("i"a)("j"a)("k"a)F2=(<'a'k><'k'j><'i'j>+<'a'k><'k'j><'i'j>+<'a'k><'k'j><'i'j>+<'a'k><'k'j><'i'j>+<'a'k><'k'j><'i'j>+<'a'k><'k'j><'i'j>) (B{7) ("i"a)("i"b)("j"b)F3=(<'i'j><'j'b><'a'b>+<'i'j><'j'b><'a'b>+<'i'j><'j'b><'a'b>+<'i'j><'j'b><'a'b>+<'i'j><'j'b><'a'b>+<'i'j><'j'b><'a'b>) (B{8) ("i"a)("j"a)("j"b) (B{9) ("i"a)("i"b)("j"a)F4=(<'i'b><'a'b><'j'a>+<'i'b><'a'b><'j'a>+<'i'b><'a'b><'j'a>+<'i'b><'a'b><'j'a>+<'i'b><'a'b><'j'a>+<'i'b><'a'b><'j'a>) (B{10) 128

PAGE 129

("i"a)("i"b)("j"a)F5=(<'a'j><'i'j><'i'b>+<'a'j><'i'j><'i'b>+<'a'j><'i'j><'i'b>+<'a'j><'i'j><'i'b>+<'a'j><'i'j><'i'b>+<'a'j><'i'j><'i'b>) (B{11) ("i"a)("i"b)("i"c)F6=(<'i'c><'b'c><'i'a>+<'i'c><'b'c><'i'a>+<'i'c><'b'c><'i'a>+<'i'c><'b'c><'i'a>+<'i'c><'b'c><'i'a>+<'i'c><'b'c><'i'a>) (B{12) ("i"a)("j"a)("k"a) (B{13) Thesecondpartisrepresentedbythesumofdiagrams(SX1)-(SX6) ("k"b)("j"b)("i"a)F7=(<'b'k><'k'j><'i'a>+<'b'k><'k'j><'i'a>+<'b'k><'k'j><'i'a>+<'b'k><'k'j><'i'a>+<'b'k><'k'j><'i'a>+<'b'k><'k'j><'i'a>) (B{14) ("j"c)("j"b)("i"a)F8=(<'j'c><'c'b><'i'a>+<'j'c><'c'b><'i'a>+<'j'c><'c'b><'i'a>+<'j'c><'c'b><'i'a>+<'j'c><'c'b><'i'a>+<'j'c><'c'b><'i'a>) (B{15) 129

PAGE 130

("k"a)("j"b)("i"a)F9=(<'a'k><'b'j><'i'a>+<'a'k><'b'j><'i'a>+<'a'k><'b'j><'i'a>+<'a'k><'b'j><'i'a>+<'a'k><'b'j><'i'a>+<'a'k><'b'j><'i'a>) (B{16) ("i"c)("j"b)("i"a)F10=(<'b'j><'i'c><'i'a>+<'b'j><'i'c><'i'a>+<'b'j><'i'c><'i'a>+<'b'j><'i'c><'i'a>+<'b'j><'i'c><'i'a>+<'b'j><'i'c><'i'a>) (B{17) ("i"c)("j"b)("i"a)F11=(<'i'c><'a'c><'b'j>+<'i'c><'a'c><'b'j>+<'i'c><'a'c><'b'j>+<'i'c><'a'c><'b'j>+<'i'c><'a'c><'b'j>+<'i'c><'a'c><'b'j>) (B{18) ("k"a)("j"b)("i"a)F12=(<'i'k><'a'k><'b'j>+<'i'k><'a'k><'b'j>+<'i'k><'a'k><'b'j>+<'i'k><'a'k><'b'j>+<'i'k><'a'k><'b'j>+<'i'k><'a'k><'b'j>) (B{19) TogetnalexpressionforalltermsD1-D14shouldbesummedup. 130

PAGE 131

[1] A.SzaboandN.S.Ostlund,ModernQuantumChemistry,IntroductiontoAdvancedElectronicStructureTheory(McGrow-Hill,Inc.,NewYork,1990). [2] R.J.Bartlett,inModernElectronicStructureTheory,PartI,editedbyD.R.Yarkony(WorldScienticPublishingCo.,Singapore,1995). [3] F.E.Harris,H.J.Monkhorst,andD.L.Freeman,AlgebraicandDiagrammaticMethodsinMany-FermionTheory(OxfordUniversityPress,NewYork,1992). [4] P.-O.Lowdin,J.Chem.Phys.19,1936(1951). [5] P.HohenbergandW.Kohn,Phys.Rev.136B864(1964). [6] W.KohnandL.J.Sham,Phys.Rev.A140A1133(1965). [7] E.RungeandE.K.U.Gross,Phys.Rev.Lett,52,997(1984). [8] E.K.U.GrossandW.Kohn,Adv.Quantum.Chem.21,255(1990). [9] M.E.Casida,inRecentAdvancesinDensityFunctionalMethods,Part1,editedbyD.P.Chong(WorldScientic,Singapore,1995). [10] M.Petersilka,U.J.Gossmann,andE.K.U.Gross,Phys.Rev.Lett76,1212(1996). [11] J.Lindenberg,Y.Ohrn,PropagatorsinQuantumChemistry(JohnWiley&Sons,Inc,NewJersey,2004). [12] S.IvanovandR.J.Bartlett,J.Chem.Phys.114,1952(2001). [13] I.Grabowski,S.Hirata,S.IvanovandR.J.Bartlett,J.Chem.Phys.116,4415(2002). [14] A.GorlingandM.Levy,Phys.Rev.A50,196(1994). [15] J.D.TalmanandW.F.Shadwick,Phys.Rev.A1436(1976). [16] R.J.Bartlett,V.LotrichandI.Schweigert,J.Chem.Phys123,062205(2005). [17] R.J.Bartlett,I.Grabowski,S.HirataandS.Ivanov,J.Chem.Phys.122,034104(2005). [18] L.M.FrantzandR.L.Mills,Nucl.Phys.15,16(1960). [19] Q.Zhao,R.C.MorrisonandR.G.Parr,Phys.Rev.A50,2138(1994). [20] A.Gorling,Phys.Rev.A573433(1998). [21] S.Hirata,S.Ivanov,I.GrabowskiandR.J.Bartlett,J.Chem.Phys.116,6468(2002). 131

PAGE 132

S.Ivanov,S.HirataandR.J.Bartlett,Phys.Rev.Lett.83,5455(1999). [23] M.Stadele,J.A.Majewski,P.VoglandA.Goerling,Phys.Rev.Lett.79,2089(1997). [24] A.GorlingandM.Levy,IntJ.QuantumChem.Symp29,93(1995). [25] D.P.Chong,O.V,Gritsenko,E.J.Baerrends,J.Chem.Phys.1161760(2002). [26] R.J.Bartlett,I.Schweigert,V.Lotrich,J.Mol.Sruct.(Theochem)7711-8,(2206). [27] P.O.Widmark,B.J.Persson,andB.Roos,Theor.Chim.Acta79,419(1991). [28] K.Siegbang,ESCAappliedtofreemolecules,p.31,(AmericanElsevierpublishing,NewYork,1969). [29] C.E.Moore,Atomicenergylevels,(U.S.DepartmentofCommerce,Washington,D.C,1949)Vol1. [30] R.Colle,R.K.Nesbet,J.Phys.B342475(2001). [31] S.A.Kucharski,M.Wloch,M.Musial,R.J.Bartlett,J.Chem.Phys,1158263(2001). [32] T.P.Straatsma,E.Apra,T.L.Windus,E.J.Bylaska,W.deJong,S.Hirata,M.Valiev,M.T.Hackler,L.Pollack,R.J.Harrison,M.Dupuis,D.M.A.Smith,J.Nieplocha,V.Tipparaju,M.Krishnan,A.A.Auer,E.Brown,G.Cisneros,G.I.Fann,H.Fruchtl,J.Garza,K.Hirao,R.Kendall,J.A.Nichols,K.Tsemekhman,K.Wolinski,J.Anchell,D.Bernholdt,P.Borowski,T.Clark,D.Clerc,H.Dachsel,M.Deegan,K.Dyall,D.Elwood,E.Glendening,M.Gutowski,A.Hess,J.Jae,B.Johnson,J.Ju,R.Kobayashi,R.Kutteh,Z.Lin,R.Littleeld,X.Long,B.Meng,T.Nakajima,S.Niu,M.Rosing,G.Sandrone,M.Stave,H.Taylor,G.Thomas,J.vanLenthe,A.Wong,andZ.Zhang,"NWChem,AComputationalChemistryPackageforParallelComputers,Version4.7"(2005),PacicNorthwestNationalLaboratory,Richland,Washington99352-0999,USA. [33] H.JiangandE.Engel,J.Chem.Phys.123,224102(2005). [34] J.P.Perdew,K.Burke,M.Ernzerhof,Phys.Rev.Lett.77,3865(1996). [35] K.Raghvachari,G.W.Trucks,J.A.PopleandM.Head-Gordon,Chem.Phys.Lett157,479(1989). [36] D.Bokhan,I.V.Schweigert,R.J.Bartlett,Mol.Phys,103,2299(2005). [38] J.C.Slater,Phys.Rev.81,385(1951). [39] D.J.TozerandN.C.Handy,J.Chem.Phys.108,2545(1998). 132

PAGE 133

J.P.Perdew,R.G.Parr,M.LevyandJ.L.Balduz,Jr.,Phys.Rev.Lett.49,1691(1982). [41] A.Balkova,R.J.Bartlett,J.Chem.Phys.102,7116(1995). [42] Ch.W.Bauschliher,Jr.andP.R.Taylor,J.Chem.Phys.86,2844(1987). [43] J.Schirmer,A.Dreuw,Phys.Rev.A75022513(2007). [44] J.B.Krieger,Y.LiandJ.Iafrate,Phys.Rev.A45101(1992). [45] V.N.Staroverov,G.Scuseria,E.R.Davidson,J.Chem.Phys.124141103(2006). [46] W.YangandQ.Wu,Phys.Rev.Lett,89143002(2002). [47] A.Gorling,Phys.Rev.Lett,83,5459(1999). [48] R.J.Bartlett,I.V.SchweigertandV.F.Lotrich,J.Mol.Struct:Theochem771,1(2006). [49] D.Rohr,O.Gritsenko,E.J.Baerends,Chem.Phys.Lett.432336(2006). [50] S.Hirata,S.Ivanov,R.J.Bartlett,I.Grabowski,Phys.Rev.A71,032507(2005). [51] A.Dreuw,J.L.WeismanandM.Head-Gordon,J.Chem.Phys.119,2943(2003). [52] S.Hirata,S.Ivanov,I.Grabowski,R.J.Bartlett,K.Burke,J.D.Talman,J.Chem.Phys.115,1635(2001). [53] P.O.Lowdin,Phys.Rev.139,A357-A372(1965). [54] S.Ivanov,M.Levy,J.Chem.Phys.119,7087(2003). [55] H.Jiang,E.Engel,J.Chem.Phys.125184108(2006). [56] O.V.Gritsenko,E.J.Baerends,Phys.Rev.A,64042506(2001). [57] S.Ivanov,S.Hirata,I.Grabowskii,R.J.Bartlett,J.Chem.Phys.118461(2003). [58] A.F.Izmaylov,V.N.Staroverov,G.E.Scuseria,E.R.Davidson,G.StoltzandE.Cances,J.Chem.Phys.126,084107(2007). [59] S.KummelandJ.P.Perdew,Phys.Rev.Lett,90043004(2003). [60] I.Schweigert,V.F.LotrichandR.J.Bartlett,J.Chem.Phys,125104108(2006);I.Schweigert,Ph.D.thesis,UniversityofFlorida(2005). [61] R.J.BartlettandM.Musial,Rev.Mod.Phys,79291(2007). [62] D.Bokhan,R.J.Bartlett,Phys.Rev.A,73022502(2006). [63] D.Tozer,J.Chem.Phys.11912697(2003). 133

PAGE 134

S.J.A.vanGisbergen,P.R.T.Schipper,O.V.Gritsenko,E.J.Baerends,B.ChampagneandB.Kirtman,Phys.Rev.Lett.83697(1999). [65] F.A.Bulat,A.Toro-Labbe,B.Champagne,B.Kirtman,W.Yang,J.Chem.Phys.123014319(2005). [66] M.Kamiya,H.Sekino,T.TsunedaandK.Hirao,J.Chem.Phys122234111(2005). [67] R.J.Bartlett,D.M.Silver,Phys.Rev.A101927(1974). [68] H.P.Kelly,Adv.Chem.Phys12314(1963). [69] R.J.Bartlett,H.Sekino,ACSSymposiumSeries628:NonlinearOpticalMaterials23-57(1996). [70] A.M.LeeandS.M.Colwell,J.Chem.Phys.1019704(1994). [71] S.P.KarnaandM.Dupuis,J.Comp.Chem.12487(1991). [72] H.SekinoandR.J.Bartlett,J.Chem.Phys85976(1986). [73] H.SekinoandR.J.Bartlett,J.Chem.Phys983022(1993). [74] D.Bokhan,R.J.Bartlett,J.Chem.Phys.(submitted). 134

PAGE 135

IwasborninAvdeevka,asmalltownlocatedinthesoutheasternpartoftheUkraineintheDonetskregion.IgrewupinthecityofGorlovka,aworking-classminingcommunity.WhenIwasinninthgrade,IacquiredaninterestinchemistryanddecidedthatIwouldbecomeachemist.Duringmylastyearofhighschool,Iwasoneofthewin-nersoftheUkrainianChemistryOlympiad.Asaresult,IwasadmittedtotheprestigiousMoscowStateUniversitywithouttheotherwisenecessaryentranceexaminations.Duringmysophomoreyearofcollege,Idecidedtospecializeinquantumchemistry.AfterIde-fendedmyundergraduatethesisandobtainedmydegreein2002,IbegantoworktowardjoiningDr.RodneyBartlett'sresearchgroupattheUniversityofFlorida.AfterworkingasasystemadministratorforseveralcompaniesinMoscowfor16months,IobtainedthenecessaryfundstopayfortransportationtotheUnitedStatesaswellasforthenecessaryexams,whichincludedtheGraduateRecordExamination(GRE)andtheTestofEnglishasaForeignLanguage(TOEFL).Finally,IjoinedDr.Bartlett'sgroupattheUniversityofFloridaonJanuary21st,2004. 135