Hartree–Fock and Kohn–Sham atomic-orbital based time-dependent response theory
Helena Larsen,a)Poul Jo”rgensen, and Jeppe Olsen
Department of Chemistry, University of Aarhus, Langelandsgade 140, 8000 A˚ rhus C, Denmark Trygve Helgakerb)
Department of Chemistry, University of Cambridge, Lensfield Road, Cambridge, CB2 1EW, United Kingdom
共Received 31 July 2000; accepted 24 August 2000兲
A reformulation of general time-dependent Hartree–Fock and Kohn–Sham response theories that refers strictly to the atomic-orbital basis is presented. It is based on a recently proposed exponential parametrization of the one-electron atomic-orbital density matrix. In the presented formulation, only matrix multiplications and additions of sparse matrices are needed to compute the response functions and linear scaling with system size may, therefore, be obtained. Thus, this formalism is well suited to the computation of dynamic and static properties for large molecules at the Hartree–
Fock and Kohn–Sham density-functional levels of theory. © 2000 American Institute of Physics.
关S0021-9606共00兲30643-2兴
I. INTRODUCTION
Currently, there is an intense and multidisciplinary effort aimed at developing photonic materials and technologies.
Whereas electronic techniques are based on the transport of electrons, photonic techniques use photons for the transport and storage of information. The photonic techniques exploit nonlinear interactions between molecules and electromag- netic fields and computational chemistry may, therefore, con- tribute to the development of photonic materials by supply- ing accurate nonlinear susceptibilities. As the photonic materials typically are large organic molecules or polymers, it is important to develop computational methods that are able to compute nonlinear properties for such molecules. For reviews of nonlinear properties of large molecules, we refer to Ref. 1.
In response function theory we determine the time- development of an observable when the molecular system is subjected to, for example, an external electric or magnetic field. This field may oscillate with a given frequency that causes the wave function and the observed properties to be- come frequency dependent. It provides an efficient method for the calculation of response properties like nonlinear sus- ceptibilities for small and medium sized molecules and much effort has, therefore, been devoted to the development of this technique. In the present paper we present a novel method for the computation of response functions within the self- consistent field 共SCF兲 theories Hartree–Fock and Kohn–
Sham density-functional theory that may scale linearly with the size of the system and thus is suitable for the computa- tion of nonlinear properties of large molecules.
Much work has been done to develop density-functional and ab initio methods that are able to handle very large mo-
lecular systems. Ideally, these methods should scale linearly with the number of basis functions, N. Especially, the SCF theories have been considered, since these methods provide a good compromise between relatively low computational cost and reasonable accuracy—not only for the energy but also for molecular properties. The computational cost of the two- electron integrals is effectively reduced to scale linearly with the system size using the fast multipole method共FMM兲and integral prescreening techniques.2–7 The N3 scaling of the diagonalization of the density matrix can be reduced using either localized orbitals or by optimizing the density matrix directly without constructing orbitals.8–14In the methods de- veloped so far, the main concern has been how to optimize the total energy, although a few methods concerning re- sponse properties have already been suggested.15
The response functions are usually obtained in the molecular-orbital 共MO兲 basis.16,17 However, since it is not the wave function correction but rather the expectation val- ues that represent the time development of the observables of interest, response functions may be determined directly in the atomic-orbital共AO兲basis. We describe how this may be done expressing response functions in terms of the Fock, overlap and density matrices in the AO basis. Matrix multi- plications and additions involving these sparse matrices al- low linear scaling to be achieved.
The density-based time-dependent response functions are derived exploiting the general exponential parametriza- tion of the AO density matrix introduced by Helgaker et al.
in Refs. 18 and 19. After a brief introduction to time- dependent response theory, we derive a density-based formu- lation of the time-dependent variation principle in Sec. III.
Next, in Sec. IV, we consider the exponential parametriza- tion of the density matrix in the time-dependent case, and in Secs. V–VIII we derive the response functions. Although the derivation allows the construction of response functions to arbitrary order, we restrict our discussion to the linear and
a兲Electronic mail: [email protected]
b兲Permanent address: Department of Chemistry, University of Oslo, P.O.B.
1033 Blindern, N-0315 Oslo, Norway.
8908
0021-9606/2000/113(20)/8908/10/$17.00 © 2000 American Institute of Physics
quadratic response functions. In Sec. IX, we discuss how to solve sets of linear equations needed to obtain the response functions. After a brief discussion of Kohn–Sham theory in Sec. X, we give some concluding remarks in Sec. XI.
II. INTRODUCTION TO TIME-DEPENDENT HARTREE–FOCK THEORY
Consider a molecular system described by a time- independent Hamiltonian, Hˆ
0. When a general field W(t) is applied to the system, the system interacts with the field. The interaction operator may be denoted Vˆt. We assume that W(t) vanishes at t⫽⫺⬁. The interaction operator then also vanishes at t⫽⫺⬁ and can be expressed as
Vˆt⫽
冕
⫺⬁⬁
Vˆexp关共⫺i⫹兲t兴d, 共1兲 where is a positive infinitesimal that ensures V⫺⬁ is zero.
From the Hermiticity of Vt, it follows that:
共V兲†⫽V⫺. 共2兲
The Hamiltonian, Hˆ , for the total system becomes
Hˆ⫽Hˆ0⫹Vˆt. 共3兲
In Hartree–Fock theory, ⌿ is represented by a single- determinant wave function
⌿共x1,x2,...,xM,t兲⫽共M !兲⫺1/2det兩1共x1,t兲2共x2,t兲¯
⫻I共xI,t兲¯M共xM,t兲兩, 共4兲 where I is the Ith occupied molecular spin orbital and where xI denotes the space and spin coordinates of the Ith electron. For ease of notation, we suppress the explicit coor- dinate and time dependence ofand⌿. We assume that the molecular spin orbitals are orthonormal
具I兩J典⫽␦IJ. 共5兲
The zero-order wave function is obtained at t⫽⫺⬁ where no perturbation is applied
⌿0⫽共M !兲⫺1/2det兩1 02
0¯M
0兩 共6兲
and is assumed to be optimized. This is ensured when the occupied molecular spin orbitals are eigenfunctions of the zero-order Fock operator20
Fˆ
0I 0⫽II
0. 共7兲
HereI
0 is a time-independent molecular spin orbital andI
its energy. Fˆ0 is given by Fˆ
0⫽hˆ0⫹
兺
I 共JˆI0⫺KˆI0兲, 共8兲where hˆ0 is the one-electron operator hˆ0⫽⫺1
2ⵜ2⫺
兺
A 兩RAZ⫺Ar兩, 共9兲 and JˆI0 and KˆI
0 are the Coulomb and exchange operators, respectively,
JˆI0J
0共x1兲⫽
冕
共I0兲*共x2兲r112I0共x2兲dx2J
0共x1兲, 共10兲
Kˆ
I 0J
0共x1兲⫽
冕
共I0兲*共x2兲r112J0共x2兲dx2I
0共x1兲. 共11兲 The integrations are over the space- and spin coordinates of electron 2.
The time development of the Slater determinant, Eq.共4兲, may be determined by one of the time-dependent variation principles as discussed in the next section. From the time development of the wave function, we may determine the time development of the expectation value of a given opera- tor, Aˆ . This time development is conveniently expressed in terms of response functions
具Aˆ典⫽具⌿0兩Aˆ兩⌿0典⫹
冕
⫺⬁⬁
exp共⫺i1t兲具具Aˆ ;Vˆ1典典1d1
⫹1 2
冕
⫺⬁⬁
冕
⫺⬁⬁
exp关⫺i共1⫹2兲t兴
⫻具具Aˆ ;Vˆ1,Vˆ2典典1,2d2d1⫹¯, 共12兲 where 具具Aˆ ;Vˆ1典典1 and 具具Aˆ ;Vˆ1,Vˆ2典典1,2 are the linear and quadratic response functions, respectively. The response functions are defined to be symmetric with respect to inter- change of the frequencies—for example,
具具Aˆ ;Vˆ1,Vˆ2典典1,2⫽具具Aˆ ;Vˆ2,Vˆ1典典2,1. 共13兲 In the following, we discuss how the response functions may be determined using an exponential parametrization of the density matrix in the AO basis.
III. DENSITY-BASED TIME-DEPENDENT HARTREE–FOCK THEORY
In this section, we derive a general equation for the time development of the density matrix. We begin by deriving the variation principle in the MO basis. Next, we consider a density-based formulation that refers strictly to the AO basis.
A. Time-dependent equations in the MO basis
The solutions to the time-dependent Schro¨dinger equa- tion depend upon an overall phase factor, which is redundant in the limit where Vˆt becomes static共time-independent兲. To obtain an approximate formulation that comprises both the static limit and the dynamic case, it is pertinent to eliminate the overall phase factor. This may be accomplished by using the Langhoff–Epstein–Karplus time-dependent variation principle in phase-isolated form21
Re具␦⌿兩
冉
Hˆ⫺it冊
兩⌿典⫽0, 共14兲where ␦⌿ is an allowed variation in ⌿. Furthermore, the time-dependent wave function should fulfill the orthonormal- ization constraint
具␦⌿兩⌿典⫹具⌿兩␦⌿典⫽0. 共15兲
The phase of⌿ is not defined by Eq. 共15兲. However, since we express our results in terms of response functions 共or equivalently densities兲, which do not depend on the phase factor, it is not necessary to determine the phase explicitly.
Let us consider the terms in Eq.共14兲that contain Hˆ 具␦⌿兩Hˆ兩⌿典⫹具⌿兩Hˆ兩␦⌿典⫽␦具⌿兩Hˆ兩⌿典. 共16兲 Expanding in terms of orbitals, we obtain
␦具⌿兩Hˆ兩⌿典⫽
兺
I 兵具␦I兩Fˆ⫹Vˆt兩I典⫹具I兩Fˆ⫹Vˆt兩␦I典其,共17兲 where Fˆ is the time-dependent Fock operator. It is obtained as a straightforward generalization of Fˆ0 where the orbitals depend explicitly on time.
Next, we consider the terms in Eq.共14兲that contain the partial derivatives with respect to time. Since /t behaves like a one-electron operator, we obtain
Re i具␦⌿兩⌿˙典⫽Re i
冉 兺
I 具␦I兩˙I典⫹
兺
I⫽J具␦I兩I典具J兩˙J典冊
. 共18兲This expression may be simplified using the orthonormaliza- tion constraints on the MOs, Eq. 共5兲
具␦I兩I典⫹具I兩␦I典⫽2 Re具␦I兩I典⫽0, 共19兲 具˙J兩J典⫹具J兩˙J典⫽2 Re具˙J兩J典⫽0. 共20兲 The terms 具␦I兩I典 and 具˙J兩J典 are thus pure imaginary and the last sum in Eq. 共18兲therefore vanishes. Combining Eqs. 共17兲and共18兲, we obtain the time-dependent equations for the MOs
Re
兺
I 具␦I兩Fˆ⫹Vˆt⫺i t兩I典⫽0. 共21兲 B. The time-dependent MO equations in matrix
representation
The molecular spin orbitals, I, can be expanded in AOs,
I⫽
兺
CI. 共22兲Substitution in the first term of Eq.共21兲gives
Re
兺
I 具␦I兩Fˆ兩I典⫽Re兺
I ␦CI†fCI, 共23兲where CI is a column vector containing the elements CI from Eq. 共22兲, and f is the Fock matrix in the AO basis
f⫽h⫹
兺
D共g⫺g兲, 共24兲where hand g are given by
h⫽
冕
*共x兲冉
⫺12ⵜ2⫺兺
A 兩RAZ⫺Ar兩冊
共x兲dx, 共25兲g⫽
冕 冕
*共x1兲*共x2r兲12共x1兲共x2兲dx1dx2. 共26兲 The second term in Eq.共21兲is treated in the same way; for the last term, we obtain
Re i
兺
I 具␦I兩t兩I典⫽Re i兺
I ␦CI†SC˙I, 共27兲where
S⫽具兩典. 共28兲
Thus, Eq.共21兲may be written as Re
兺
I ␦CI†
冉
f⫹Vt⫺iSt冊
CI⫽0. 共29兲The variations ␦CI are not independent as the MOs must satisfy the orthonormality condition
C†SC⫽1, 共30兲
which leads to the constraints
␦CI
†SCJ⫹CI†S␦CJ⫽0 共31兲
for the first-order variations. Introducing the Hermitian La- grange multipliers,IJ, Eqs.共29兲and共31兲may be combined to give the unconstrained Lagrangian equations
Re
兺
I ␦CI†再 冉
f⫹Vt⫺iSt冊
CI⫺兺
J SCJJI冎
⫽0. 共32兲As the variations␦CIand␦CI
†may be considered as linearly independent, Eq. 共32兲is satisfied when
冉
f⫹Vt⫺iSt冊
CI⫺兺
J SCJJI⫽0. 共33兲A similar equation is obtained for the complex conjugate.
Using standard matrix notation, Eq.共33兲becomes
冉
f⫹Vt⫺iSt冊
C⫽SC, 共34兲which is the form given in Ref. 20. Note carefully that C is a rectangular matrix containing in column I the expansion coefficients CI of the Ith occupied orbital. In the time- independent limit, Eq.共34兲reduces to the well-known equa- tion fC⫽SC⑀, withIJ⫽⑀IJ. Eq.共34兲determines the time- development of the transformation matrix C. Next, we will rewrite this equation to obtain an equation for the time- development of the density matrix.
C. Time-dependent equations for the AO density matrix
Multiplying Eq.共34兲with C†S from the right, we obtain
再
共f⫹Vt兲C⫺iStC冎
C†S⫽SCC†S. 共35兲Likewise, the complex conjugate equations can be rewritten to obtain the conjugate transpose of Eq.共35兲. Subtracting the two sets of equations, we obtain
共f⫹Vt兲CC†S⫺SCC†共f⫹Vt兲⫺i共SC˙ C†S⫹SCC˙†S兲⫽0.
共36兲 Introducing the one-electron matrix in the AO basis
D⫽CC†, 共37兲
these equations can be written as
共f⫹Vt兲DS⫺SD共f⫹Vt兲⫽iSD˙ S. 共38兲 Note that the time-dependent variation principle in Eq. 共38兲 refers exclusively to the AO basis and that, in the time- independent limit, Eq.共38兲reduces to the standard SCF con- ditions
f0DS⫺SDf0⫽0, 共39兲 where f0 is the time-independent Fock matrix in the AO basis. Furthermore, as Eq.共38兲defines the time development of the SCF density matrix, it constitutes the SCF Liouville equation in an orthonormal basis.
For future convenience, it is advantageous to rewrite Eq.
共38兲slightly. First note that, for all matrices M
TrEM⫽M, 共40兲
where E is a unit matrix with elements given by
关E兴⫽␦␦. 共41兲
The ()th element of Eq.共38兲can, therefore, be written as TrE兵共f⫹Vt兲DS⫺SD共f⫹Vt兲其⫽iTrE共SD˙ S兲. 共42兲 Introducing the S commutator
关A,B兴S⫽ASB⫺BSA, 共43兲
we may write Eq. 共42兲in a short-hand notation as
Tr共f⫹Vt兲关D,E兴S⫽iTrE共SD˙ S兲. 共44兲 This is the fundamental equation that we shall use to derive the response functions.
IV. PARAMETRIZATION OF THE DENSITY MATRIX In Refs. 18 and 19 we presented a general exponential parametrization of the density matrix in the AO basis. In this section, we briefly review this parametrization and consider how the time development may be introduced. The density matrices in the AO basis fulfill the symmetry, trace, and idempotency conditions, which in the spin–orbital basis are given by
D†⫽D, 共45兲
TrDS⫽Ne, 共46兲
DSD⫽D. 共47兲
The density matrices in the AO basis are related by the transformation18,19
D共X兲⫽exp共⫺XS兲D exp共SX兲, 共48兲 where X is an anti-Hermitian matrix. Note that D(X) fulfills Eqs.共45兲–共47兲. Finally, X should comply with the projection relation
X⫽P共X兲⫽PXQ†⫹QXP†, 共49兲 where P⫽DS and Q⫽1⫺DS, in order to eliminate redun- dancies, which are nonvanishing choices of X for which D(X)⫽D(0)⫽D. Note that the projectors are constructed from the matrix D(0). The above parametrization of the AO density matrix is equivalent to the one in the MO basis where similar symmetry, trace and idempotency conditions are ful- filled. Furthermore, Eq.共48兲reduces to a standard orthogonal transformation in the MO basis when S⫽1.18
The transformed density matrix, D(X), may be evalu- ated using the asymmetric Baker–Campbell–Hausdorff 共BCH兲expansion according to Ref. 18
D共X兲⫽D⫺关X,D兴S⫹12关X,关X,D兴S兴S⫺¯. 共50兲 Introducing the operator Xˆ by
Xˆ M⫽关X,M兴S, 共51兲
D(X) can be written in the more compact form as
D共X兲⫽n
兺
⫽⬁0 共⫺n!1兲nXˆnD. 共52兲 The time development of the AO density matrix is intro- duced by X(t) which may be written asX共t兲⫽⬎
兺
共X共t兲E⫺X*共t兲E兲⫽兺
mXm共t兲Om, 共53兲 where the diagonal elements, X, vanish since we have used a phase-isolated variation principle. In Eq. 共53兲, we have introduced the short-hand notation
Xm共t兲⫽
再
⫺XX*共t共兲t兲 mm⬎⬍00 共54兲and similarly
Om⫽
再
QQm†m⫽⫽EE mm⬍⬎0.0 共55兲The time evolution of the density matrix is determined by Eq. 共44兲. Using Eqs.共53兲and共55兲, this equation may be expressed as
Tr共f共X兲⫹Vt兲关D共X兲,Om†兴S⫽iTrOm†SD˙共X兲S, 共56兲 which for t⫽⫺⬁ 共and X⫽0) becomes the standard SCF equation, Eq.共39兲. Note that f(X) depends on X through the density matrix
f共X兲⫽h⫹G共D共X兲兲, 共57兲 where, for convenience, we have introduced
G共D共X兲兲⫽
兺
D共X兲共g⫺g兲. 共58兲V. GENERAL EQUATIONS FOR THE TIME DEVELOPMENT
The time development of the AO density matrix can be determined from Eq. 共56兲. Inserting the expression for the Fock matrix Eq. 共57兲into Eq.共56兲, we obtain
Trh关D共X兲,Om†兴S⫹TrG共D共X兲兲关D共X兲,Om†兴S
⫹TrVt关D共X兲,Om†兴S⫽i TrOm†SD˙共X兲S. 共59兲 Let us first consider the evaluation of the left-hand side of Eq.共59兲. Inserting the expressions for D(X) and X from Eqs.
共52兲and共53兲, the first term can be written as
Trh关D共X兲,Om†兴S⫽n
兺
⫽⬁0 共⫺n!1兲nTrh⫻
冋冉
⫽兿
n1 Oˆl冊
D,Om†册
S ⫽兿
n1 Xl共t兲, 共60兲 where summation over the repeated index l is understood 共the Einstein summation convention兲. Likewise, the second term becomesTrG共D共X兲兲关D共X兲,Om†兴S
⫽n
兺
⫽⬁0 k兺
⫽n0 共n共⫺⫺k1兲兲!k!n TrG冉冉
⫽n兿
⫺1k Oˆl冊
D冊
⫻
冋冉
⫽兿
k1 Oˆj冊
D,Om†册
S冉
⫽n兿
⫺k1 Xl共t兲冊冉
⫽兿
k1 Xj共t兲冊
,共61兲 where terms of identical order in X have been collected.
Defining
Eml
1l2¯ln
关n⫹1兴 ⫽共⫺1兲n
n! Trh
冋冉
⫽兿
n1 Oˆl冊
D,Om†册
S⫹k
兺
⫽n0 共n共⫺⫺k1兲兲!k!n TrG冉冉
⫽n兿
⫺k1 Oˆl冊
D冊
⫻
冋冉
⫽兿
n⫺nk⫹1 Oˆl冊
D,Om†册
S, 共62兲
the two first terms of Eq.共59兲can be written as Trh关D共X兲,Om†兴S⫹TrG共D共X兲兲关D共X兲,Om†兴S
⫽n
兺
⫽⬁0 Eml关n⫹1l12¯兴 ln⫽兿
n1 Xl共t兲. 共63兲Writing out the expression for Em关1兴, we obtain
Em关1兴⫽Trf0关D,Om†兴S⫽TrOm†共f0DS⫺SDf0兲. 共64兲 Thus, Em关1兴 is equal to a element of the gradient and vanishes for an optimized state. The elements of the second and third E matrices are given as 共assuming symmetrization of the indices兲
Emn关2兴⫽⫺Trf0关关On,D兴S,Om†兴S
⫺TrG共关On,D兴S兲关D,Om†兴S, 共65兲
Emnk关3兴 ⫽12Trf0关关Ok,关On,D兴S兴S,Om†兴S⫹TrG共关On,D兴S兲
⫻关关Ok,D兴S,Om†兴S⫹12TrG共关Ok,关On,D兴S兴S兲
⫻关D,Om†兴S. 共66兲 Similarly, the last term on the left-hand side of Eq.共59兲may be written as
TrVt关D共X兲,Om†兴S
⫽n
兺
⫽0⬁ 共⫺1兲n
n! TrVt
冋冉
⫽兿
n1 Oˆl冊
D,Om†册
S ⫽兿
n1 Xl共t兲⫽n
兺
⫽⬁0 Vmlt关n1⫹l21¯兴ln⫽兿
n1 Xl共t兲, 共67兲 whereVml
1l2¯ln
t关n⫹1兴 ⫽共⫺1兲n
n! TrVt
冋冉
⫽兿
n1 Oˆl冊
D,Om†册
S. 共68兲
Writing Eq. 共68兲 out for the matrices Vt关1兴 and Vt关2兴, we obtain
Vmt关1兴⫽TrVt关D,Om†兴S⫽TrOm†共VtDS⫺SDVt兲, 共69兲 Vmnt关2兴⫽⫺TrVt关关On,D兴S,Om†兴S. 共70兲 Note that Vmt关1兴 has the same structure as the gradient, Eq.
共64兲.
Finally, let us consider the term on the right-hand side of Eq. 共59兲
iTrOm†SD˙共X兲S⫽iTrOm†S
再
texp共⫺XS兲D exp共SX兲冎
S.共71兲 To simplify this equation, we use the relation
t兵exp共⫺MS兲N exp共SM兲其
⫽n
兺
⫽⬁0 共⫺n!1兲n共MˆnN˙兲⫹n
兺
⫽0⬁ k
兺
⫽0n 共⫺1兲k
k!共n⫺k⫹1兲!兵Mˆk关N,共Mˆn⫺kM˙兲兴S其, 共72兲 derived in the Appendix. Noting that D˙⫽0, Eq.共71兲can be written as
iTrOm†SD˙共X兲S⫽in
兺
⫽⬁0 k兺
⫽n0 k!共n共⫺⫺k1⫹兲k1兲!⫻TrOm†S兵Xˆk关D,共Xˆn⫺kX˙兲兴S其S. 共73兲 Inserting the expression for X from Eq. 共53兲, we obtain
iTrOm†SD˙共X兲S
⫽in
兺
⫽0⬁ k
兺
⫽0n 共⫺1兲k
k!共n⫺k⫹1兲!TrOm†S
再 冉⫽兿
k1 Oˆj冊
⫻
冋
D,冉
n⫽⫺兿
k⫹21 Oˆl冊
Ol1册
S冎
S冉
⫽兿
k1 Xj共t兲冊
⫻
冉
n⫺⫽兿
k⫹21 Xl共t兲冊
X˙l1共t兲⫽in
兺
⫽⬁1 Sml关n⫹1l1兴2¯lnX˙l1共t兲⫽兿
n2 Xl共t兲, 共74兲 whereSml
1l2¯ln
关n⫹1兴 ⫽kn
兺
⫽⫺01 k!共⫺共n⫺1兲kk兲!TrOm†S再 冉⫽n兿
⫺nk⫹1 Oˆl冊
⫻
冋
D,冉
⫽n兿
⫺k2 Oˆl冊
Ol1册
S冎
S. 共75兲The explicit forms for the first two terms in Eq.共75兲become Smn关2兴⫽TrOm†S关D,On兴SS, 共76兲 Smnk关3兴 ⫽⫺TrOm†S关Ok,关D,On兴S兴SS
⫹12TrOm†S关D,关Ok,On兴S兴SS. 共77兲 Thus, Eq.共59兲may be written as
n
兺
⫽0⬁
共Eml
1l2¯ln 关n⫹1兴 ⫹Vml
1l2¯ln
t关n⫹1兴 兲⫽
兿
n1 Xl共t兲⫽in
兺
⫽⬁1 Sml关n⫹1l12¯兴 lnX˙l1共t兲⫽兿
n2 Xl共t兲, 共78兲 which is formally equivalent to the time-dependent equations previously derived for SCF and multiconfiguration SCF 共MCSCF兲 wave functions in the MO basis. The order-by- order solution of Eq. 共78兲 is, therefore, similar to the one described in Ref. 16. In the following, we summarize how the response equations and response functions are obtained.VI. FIRST- AND SECOND-ORDER EQUATIONS
The set of parameters X(t) may be expanded in powers of the perturbation
Xn共t兲⫽Xn共1兲共t兲⫹Xn共2兲共t兲⫹¯, 共79兲 where the zero-order coefficient vanishes since the reference state is optimized. The parameters, Xn(i)(t), may be deter- mined by requiring Eq.共78兲to be valid to each order of the perturbation. Let us consider the expressions needed to de- termine Xn(1)(t) and Xn(2)(t). The terms in Eq.共78兲that may contribute to the evaluation of these coefficients are Emn关2兴Xn共t兲⫹Emnk关3兴 Xn共t兲Xk共t兲⫺iSmn关2兴X˙
n共t兲
⫺iSmnk关3兴 X˙n共t兲Xk共t兲⫹Vmt关1兴⫹Vmnt关2兴Xn共t兲⫹¯⫽0. 共80兲
As before summation is carried out over all repeated indices.
Inserting the expansion of Xn(t) from Eq.共79兲and collecting terms of first order, we get
Emn关2兴Xn共1兲共t兲⫺iSmn关2兴X˙
n
共1兲共t兲⫽⫺Vmt关1兴. 共81兲 The second-order equation is given by
Emn关2兴Xn共2兲共t兲⫺iSmn关2兴X˙
n 共2兲共t兲
⫽⫺Emnk关3兴 Xn共1兲共t兲Xk共1兲共t兲⫹iSmnk关3兴 X˙
n
共1兲共t兲Xk共1兲共t兲
⫺Vmnt关2兴Xn共1兲共t兲. 共82兲 To solve the first-order equation, Eq.共81兲, we use the Fourier expansion of Xn(1)(t)
Xn共1兲共t兲⫽
冕
exp共⫺it兲Xn共1兲共兲d. 共83兲Inserting Eqs. 共1兲and共83兲into Eq.共81兲, we obtain
冕
exp共⫺it兲兵共Emn关2兴⫺Smn关2兴兲Xn共1兲共兲⫹Vm关1兴其d⫽0,共84兲 which implies
X共1兲共兲⫽⫺共E关2兴⫺S关2兴兲⫺1V关1兴. 共85兲 Likewise to solve the second-order equation, the Fourier transform
Xn共2兲共t兲⫽
冕 冕
exp关i共1⫹2兲t兴Xn共2兲共1,2兲d1d2共86兲 is introduced. Insertion in Eq.共82兲gives
关Emn关2兴⫺共1⫹2兲Smn关2兴兴Xn共2兲共1,2兲
⫽12共⫺Emnk关3兴 ⫺Emkn关3兴 ⫹1Smnk关3兴
⫹2Smkn关3兴 兲Xn共1兲共1兲Xk共1兲共2兲⫺Vmn1关2兴Xn共1兲共2兲
⫺Vmn2关2兴Xn共1兲共1兲, 共87兲 where Xn(2)(1,2) is defined to be symmetric in 1 and
2.
VII. THE STRUCTURE OF E†2‡ANDS†2‡
Consider Eq.共85兲in more detail. Ordering the matrices Om in Eq. 共53兲 in the order 1,2,...,m,⫺1,⫺2,...,⫺m (m
⬎0) and assuming that integrals and density matrices are real, we can write E关2兴 and S关2兴 in the following forms:
E关2兴⫽
冉
AB BA冊
, 共88兲S关2兴⫽
冉
⫺⌺⌬ ⫺⌬⌺冊
, 共89兲where
Amn⫽⫺Tr兵f0关关QnT,D兴S,Qm兴S⫺G共关QnT,D兴S兲关D,Qm兴S其, 共90兲