Gauge-origin independent magneto-optical activity within coupled cluster response theory
Sonia Coriani,a)Christof Ha¨ttig,b)and Poul Jørgensen
Department of Chemistry, Aarhus University, Langelandsgade 140, DK-8000 A˚ rhus C, Denmark Trygve Helgaker
Department of Chemistry, University of Oslo, P. O. B. 1033, N-0315 Oslo, Norway 共Received 2 May 2000; accepted 9 June 2000兲
A gauge-origin invariant formulation of the frequency-dependent Verdet constant V() of magneto-optical rotation and of the Faraday B term of magnetic circular dichroism for coupled-cluster wave functions is derived within the framework of variational response theory.
Working expressions suitable for implementation in ab initio program packages are presented.
These expressions have a structure similar to that of the expressions for the first hyperpolarizability and the two-photon transition moment, respectively, for the Verdet constant and theB term. The approach is general and can easily be extended to other similar frequency-dependent properties as well as to other wavefunction models. Pilot results at the CCSD level are presented for V() of HF and H2. © 2000 American Institute of Physics.关S0021-9606共00兲30233-1兴
I. INTRODUCTION
The discovery of magneto-optical activity共MOA兲dates back to 1845, when Faraday1,2observed that the application of a magnetic field in the direction of propagation of a plane polarized light beam passing through a piece of heavy glass affected the behavior of the beam. The effect consisted of a rotation共兲of the plane of polarization of the ray induced by the longitudinal magnetic field; it was later found to occur in any sample and shown to be proportional to the magnetic field strength. The proportionality constant—which is fre- quency and temperature dependent—is known as the Verdet constant V().3When the frequency approaches the absorp- tion region of the sample, the light is attenuated and the rotation is accompanied by the development of an ellipticity 共兲. Both aspects, rotation and ellipticity, are related to a differential behavior induced by the magnetic field of the right 共R兲and left共L兲circularly polarized components of the plane polarized light beam. Traditionally, the rotation is as- sociated with the anisotropy of the refractive indices nL⫺nR and the ellipticity with the anisotropy of the absorp- tion coefficients nL⬘⫺nR⬘. The latter effect is also called mag- netic circular dichroism 共MCD兲. Magneto-optical rotation 共MOR兲dispersion and MCD are also referred to as the ‘‘Far- aday effect.’’
The fundamental equations relating the observed rotation and/or ellipticity to the microscopic properties of the sample have been derived in several equivalent ways. Buckingham and Stephens4 used a conventional refractive-index ap- proach, relating the complex rotation to the anisotropy of the complex refractive indices for the right and left circularly polarized components. The latter are expressed through the
Maxwell equations in terms of averaged complex induced oscillating molecular moments, which are calculated quantum-mechanically. Focusing on the MCD, Stephens5–7 rederived the equations for the ellipticity by means of semi- classical radiation absorption theory. Equivalent equations for the rotation and ellipticity have also been obtained by the refringent scattering approach, as shown for example by Barron.8 Both the conventional refractive-index approach and the refringent-scattering approach have been success- fully applied to several optical effects such as natural optical activity, linear birefringence, and Raman optical activity.
As for other optical properties, electron correlation is expected to play an important role in the ab initio calculation of the molecular properties characterizing the Faraday effect—that is, the Verdet constants and the Faraday A,B, andC terms. In addition, the problem of the unphysical de- pendence on the origin of the vector potential共gauge-origin兲 for magnetic properties calculated with conventional finite basis sets needs to be addressed.
In the past, the calculation of molecular Verdet constants has appeared at various levels of ab initio theory 共see, for example, Refs. 9–13兲. The Verdet constant has been calcu- lated either from a spectral-representation expression,13or as combination of quadratic dipole–dipole–magnetic dipole re- sponse functions.9–12 Using the Becquerel approximation,14 which is exact only in the atomic case, propagators or linear response function expressions have also been used.9,15These calculations have mostly been limited to systems that are gauge-origin independent by symmetry; no general solution has so far been presented for a gauge-origin invariant calcu- lation of Verdet constants. Indeed, in a recent paper, Peder- sen et al. question the possibility of obtaining gauge-origin independent Verdet constants using a conventional coupled cluster ansatz for the wave function.16
Theoretical calculations of the Faraday terms of the MCD—in particular the B term—have mainly been per-
a兲Permanent address: Dipartimento di Scienze Chimiche, Universita` degli Studi di Trieste, Via Giorgieri 1, I-34127 Trieste, Italy.
b兲Present address: Forschungszentrum Karlsruhe, Institute of Nanotechnol- ogy, P.O. Box 3640, D-76021 Karlsruhe, Germany.
3561
0021-9606/2000/113(9)/3561/12/$17.00 © 2000 American Institute of Physics
formed by means of sum-over-states procedures. Recently, we presented an investigation of the MCD of two naturally gauge-invariant systems, reformulating theBterm as a single residue of the dipole–dipole–magnetic dipole quadratic re- sponse function in a SCF and MCSCF parameterization.17A gauge-invariant method for calculating the Bterm was pre- sented in the early 1970s by Seaman and Linderberg.18Self- consistent molecular orbitals at nonzero magnetic field were determined in the Pariser–Parr–Pople model, starting with an atomic orbital basis of London gauge including orbitals 共LAOs or GIAOs兲,19–21 and then employed to calculate transition moments according to the single-excited configuration-interaction and the random-phase approxima- tions. TheBterm was then obtained by extrapolating to zero magnetic field the ratio between the field-dependent transi- tion strength and the magnetic field strength—that is, essen- tially by numerical differentiation.
Inspired by Seaman and Linderberg’s approach, we present here a gauge-origin invariant formulation of magneto-optical activity within the framework of coupled cluster共CC兲response theory.22–26Our approach relies on the use of LAOs in connection with a reformulation of both the Verdet constant and the FaradayB term as total derivatives with respect to the strength of an external static magnetic field of, respectively, the dipole polarizability and dipole transition strength. For the latter properties, we employ the appropriate CC response expressions defined with respect to a SCF wave function optimized within the magnetic field.
For the Verdet constant, this represents the first attempt to extend the London-orbital approach to the calculation of a frequency-dependent quadratic response property. To illus- trate the validity of the proposed approach pilot results for the Verdet constant of HF at the CCSD level are finally presented. Also, the effect of using LAOs when calculating Verdet constants which are gauge-origin independent by symmetry is tested through preliminary calculations on H2. II. THE BASIC EQUATIONS
A. Conventional formulation
To treat MOR and MCD in a unified way, one may in analogy with Ref. 4 define the complex optical rotation˜ per unit of path length
˜⫽⫹i⫽
2c⌬˜ ,n 共1兲
where ⌬˜n⫽(nL⫺nR)⫹i(nL⬘⫺nR⬘) is the anisotropy of the complex refractive index. As shown for example in Refs. 4 and 8, the complex optical rotation for plane-polarized light traveling in the z direction of a space-fixed frame is related to the complex polarizability ␣˜␣ 共frequency argument im- plied兲,
˜⫽140cN关i具␣˜xy⫺␣˜y x典兴, 共2兲 where具¯典indicates an appropriate statistical average. In the nonabsorptive region, the complex polarizability ␣˜␣ is de- fined as
␣
˜␣⫽2
ប
再
j兺
⫽n jn2⫺jn2R共具n兩d␣兩j典具j兩d兩n典兲⫹i
兺
j⫽n jn2⫺2I共具n兩d␣兩j典具j兩d兩n典兲冎
⫽␣␣⫺i␣␣⬘ ⫽␣˜␣* , 共3兲 where d indicates the electric dipole moment operator and Eq. 共2兲reduces to
⫽⫺120cNI具␣˜xy典⫽120cN具␣xy⬘ 典. 共4兲 These definitions are readily generalized to the absorptive regions by introducing the line-shape functions f and g共see, for example, Ref. 8兲. For simplicity, we here restrict our- selves to the nonabsorptive region.
A linear dependence on the magnetic field is brought explicitly into Eq. 共4兲through the linear dependence of the antisymmetric polarizability ␣xy⬘ on the magnetic field and the orientational effect of the magnetic field on the perma- nent molecular magnetic moments. The polarizability␣␣⬘ is expanded in the magnetic field according to
␣␣⬘ 共B兲⫽␣␣⬘ 共0兲⫹␣␣⬘共m,␥兲B␥⫹O共B2兲, 共5兲 where B represents the magnetic field strength. Implicit sum- mation over repeated indices is assumed here and through- out. The sum-over-states expression for the higher-order po- larizability ␣␣⬘(m),␥ is obtained from the definition of ␣␣⬘ by substituting the magnetic field-dependent eigenstates j ,n and their frequency separationsjn⫽j⫺n by the correspond- ing perturbational expansions in terms of zero-field eigen- states and frequency separations,8,27
␣␣,␥⬘共m兲⫽⫺ 2
ប2j
兺
⫽n再
共22jn⫺jn2兲2共m␥j⫺m␥n兲I共具n兩d␣兩j典具j兩d兩n典兲⫹共2jn⫺2兲I冋
k兺
⫽n 具k兩mkn␥兩j典共
具n兩d␣兩j典具j兩d兩k典⫺具n兩d兩j典具j兩d␣兩k典
兲
⫹k兺
⫽j 具j兩mk j␥兩k典共
具n兩d␣兩j典具k兩d兩n典⫺具n兩d兩j典具k兩d␣兩n典兲 册冎
⫽⫺2
បj
兺
⫽n再
ប共22jn⫺jn2兲2A␣␥⬘共jn兲⫹共jn2⫺2兲B␣␥⬘共jn兲冎
. 共6兲Here m␥ is a Cartesian component of the magnetic dipole operator and m␥k⫽具k兩m␥兩k典. The states n, j,k and the frequency separations kn, k j, jn refer now to the unpertubed system (B⫽0). The orientational effect on the magnetic moment is
accounted for either by means of classical weighted Boltz- mann average with the potential energy U
⫽⫺m␥nB␥or by means of an unweighted quantum statistical average.8 For a fluid sample, the resulting rotation with re- spect to the molecular frame becomes
⫽ 1
120cNBz⑀␣␥
冉
␣␣,␥⬘共m兲⫹kT1 m␥n␣␣⬘冊
⫽V共兲Bz,共7兲 where we have introduced the Verdet constant V()
⫽/Bz. The symbol ⑀␣␥ was used for the Levi–Civita tensor. Note that, if n is the ground state, the second term in parentheses is zero for closed-shell systems as the magnetic dipole is quenched. An analogous expansion for ␣˜␣ fol- lowed by orientational averaging leads to a similar equation for the generalized complex rotation Eq.共2兲.
The 共averaged兲Faraday A,B, and Cterms, used to ra- tionalize the MCD, are easily identified from the tensors so far introduced by splitting the total rotation 共or more rigor- ously the complex rotation兲 into contributions from a par- ticular ‘‘transition’’ n→j —that is, ⫽兺j⫽n(n→j ). Ac- cording to the current convention,8,28
共n→j兲⫽⫺0cNBz
3ប
再
ប共22jn2⫺jn2兲2A共n→j兲⫹ 2
jn
2 ⫺2
冋
B共n→j兲⫹kT1 C共n→j兲册 冎
, 共8兲A共n→j兲⫽16⑀␣␥共3A␣␥⬘共jn兲兲
⫽12⑀␣␥共m␥j⫺m␥n兲I共具n兩d␣兩j典具j兩d兩n典兲, 共9兲 B共n→j兲⫽16⑀␣␥共3B␣␥⬘共jn兲兲
⫽⑀␣␥I
冋
k兺
⫽n 具kប兩m␥kn兩n典具n兩d␣兩j典具j兩d兩k典⫹k
兺
⫽j具j兩m␥兩k典
បk j 具n兩d␣兩j典具k兩d兩n典
册
, 共10兲C共n→j兲⫽12⑀␣␥m␥nI共具n兩d␣兩j典具j兩d兩n典兲. 共11兲 For molecules with a nondegenerate ground state n and a nondegenerate state j only theBterm is nonvanishing. In the following, we are concerned with the calculation of B(n→j ) and of the Verdet constant V().
B. Derivative formulation
The Verdet constant and the FaradayBterm may alter- natively be expressed as derivatives of property expressions with respect to the magnetic induction. Expanding the rota- tion in the field according to29
具典B⫽具典⫹Bz
再 冓 冉dBdz冊
0冔
⫹kT1 共具mzn典⫹具典具mzn典兲冎
⫹O共B2兲
⫽Bz
再 冓 冉dBdz冊
0冔
⫹kT1 具mzn典冎
⫹O共B2兲, 共12兲
where the averages on the r.h.s. are taken at zero magnetic field, we can immediately relate the Verdet constant for a closed-shell system to the derivative of the frequency- dependent antisymmetric polarizability 共i.e., the imaginary part of the complex polarizability兲,
V共兲⫽
冓 冉dBdz冊
0冔
⫽ 1
120cN⑀␣␥
冉
d␣␣⬘ dB共⫺␥;兲冊
0⫽⫺ 1
120cN⑀␣␥I
冉
d␣˜␣dB共⫺␥;兲冊
0共13兲 and identify the FaradayB(n→j) term of a closed-shell sys- tem without degenerate states from the derivative of a one- photon transition strength
B共n→j兲⫽1
2⑀␣␥
冉
dI共具n兩d␣dB兩j典具␥j兩d兩n典兲冊
0⫽1
2⑀␣␥I
冉
dSdBn j␣␥冊
0. 共14兲
The derivative formulation is advantageous for several reasons. First, contrary to the conventional sum-over-states representation, it does not require that all excited states are explicitly calculated. Rather, one can simply identify the property expressions to be differentiated in terms of linear response functions and corresponding residues, and then dif- ferentiate these response function expressions for a given wave function approximation to obtain the Verdet constant and the Faraday B term. Second, we also demonstrate that, with London atomic orbitals and using response function ex- pressions where the orbitals are allowed to relax in the mag- netic field, V() andBmay be obtained gauge-origin inde- pendent without consideration whether the Ehrenfest theorem is satisfied for one-electron operators in the chosen wavefunction approximation. Difficulties in ensuring gauge- origin independence are encountered if V() andBare cal- culated in terms of quadratic response functions and residues of quadratic response functions.10
III. GAUGE-ORIGIN INVARIANT MOA WITHIN CC RESPONSE THEORY
As physical observables, the Verdet constant V() and the FaradayBterm must be independent of the choice of the gauge origin. In approximate calculations, however, gauge- origin independence is not necessarily satisfied. In the next subsections, we describe in detail how gauge-origin indepen- dent expressions for V() and Bcan be obtained from the previous derivative formulation of the response properties and the use of London orbitals. We restrict ourselves to the CC model but emphasize that the same strategy may be ap- plied for other approximate wave function models.
First we summarize how the polarizability and oscillator strength can be calculated in CC response theory. We then discuss the use of LAOs to parametrize the magnetic field dependence in the polarizability and oscillator strength. In
particular, we demonstrate that the magnetic-field orbital- relaxed property expressions to be differentiated are indepen- dent of the chosen connection scheme and identical to that obtained using the symmetric connection. As the molecular orbital integrals in the symmetric connection30 are indepen- dent of the gauge origin, we obtain a gauge-invariant Verdet constant and Bterm.
A. The ansatz
In CC response theory, the frequency-dependent dipole polarizability and the dipole oscillator strengths can be for- mulated as derivatives with respect to the 共frequency- dependent兲 electric-field strengths of the real part of the time-averaged quasienergy Lagrangian26,31 and the time- averaged transition moment Lagrangians,32both of which are generalizations of the energy Lagrangian introduced for the treatment of time-independent properties.33 To obtain these derivatives, a time-dependent CC state is used in which the orbitals are not allowed to relax with respect to the electric perturbation.26,31,32 The time-dependent Hamiltonian em- ployed can be written as
H共t,E兲⫽H⫹V共t,E兲
⫽H⫺d␥关E␥共␥兲exp共⫺i␥t兲
⫹E␥共⫺␥兲exp共i␥t兲兴 共␥⫽x,y ,z兲, 共15兲 where H is the Hamiltonian for the unperturbed system, E() the electric field strength, and d the static electric di- pole moment operator.
The resulting expressions for the electric-field-unrelaxed frequency dependent polarizability26,31 and for the one- photon transition strength26,32are
␣␣共␣,兲⫽⫺12Cˆ⫾兵d␣td共兲⫹Ftd␣共␣兲td共兲
⫹dtd␣共␣兲其 共16兲 with␣⫽⫺⫽, and
Sn j␣⫽12兵Mn←j d␣
Mdj←n⫹共Mnd← jMjd←␣n兲*其, 共17兲 where the left and right one-photon dipole transition mo- ments Mnd←jand Mdj←n are given by26,32
Mnd←j⫽¯d共⫺j兲Ej共j兲⫽dEj共j兲⫹M¯ j共j兲d, 共18兲 Mdj←n⫽E¯j共⫺j兲d. 共19兲 The operator Cˆ⫾ enforces time-reversal symmetry on the CC polarizability by symmetrizing with respect to simulta- neous complex conjugation and inversion of the sign of the frequency. The vector-matrix notation of Refs. 26 and 32 was used. The various matrices and vectors are defined as26,31,32
d⫽具¯兩d兩CC典, 共20兲
d⫽具⌳兩关d,兴兩CC典, 共21兲 F⫽具⌳兩关关H,兴,兴兩CC典, 共22兲 and
¯d共d兲⫽d⫹Ftd共d兲, 共23兲
where
兩CC典⫽exp共T兲兩HF典, 共24兲
具¯兩⫽具HF兩† exp共⫺T兲⫽具兩exp共⫺T兲, 共25兲 具⌳兩⫽具HF兩
冉
1⫹兺
t¯†冊
exp共⫺T兲. 共26兲The cluster operator is given by T⫽兺t, with an excitation operator and t the associated amplitude共for each excitation level 兲. To zero order with respect to the time- dependent electric-field perturbation, the amplitudes t are ob- tained from the CC equations,
具兩exp共⫺T兲H exp共T兲兩HF典⫽0. 共27兲 The zero-order Lagrange multipliers t¯ are the solutions of the set of linear equations
⫹t¯A⫽0, 共28兲
where
⫽具HF兩关H,兴兩CC典, 共29兲 and the CC Jacobian A is given by
A⫽具¯兩关H,兴兩CC典. 共30兲 The left and right excitation vectors E¯j(⫺j) and Ej(j) are the solutions of the linear equations
E¯jA⫽jE¯j, AEj⫽jEj 共31兲 under the biorthonormality condition E¯jEf⫽␦j f. The fre- quency argument in E¯jand Ejis conventional.32,34The first- order responses td(d) of the amplitudes to the electric field are obtained from the equations
d⫹关A⫺d1兴td共d兲⫽0. 共32兲 The intermediate vector M¯ j(j) is found by solving32
M¯ j共j兲共A⫹j1兲⫽⫺FEj共j兲. 共33兲 In second quantization, the dipole and Hamiltonian operators are expressed as
d⫽
兺
pq dpqEpq, 共34兲H⫽
兺
pq hpqEpq⫹12pqrs兺
gpqrsepqrs, 共35兲where Epq⫽兺⫽␣,ap†aq, epqrs⫽EpqErs⫺␦qrEps, and dpq, hpq and gpqrs are the one- and two-electron integrals over the orthonormal basis of molecular orbitals.
B. The introduction of the magnetic field
In the presence of an external static magnetic field, the polarizability and the oscillator strength may be obtained in a two-step approach, replacing the unperturbed Hamiltonian in Eq.共15兲by the Hamiltonian for the system in presence of the static magnetic field
H共B兲⫽H⫺mO,␥B␥⫺ 12O,␥␦B␥B␦ 共␥,␦⫽x,y ,z兲 共36兲
and treating all states as magnetic-field dependent. The time- dependent Schro¨dinger equation for the interaction between the polarized light and the system is thus solved for a static Hamiltonian that contains the interaction with the magnetic field.
In the Hamiltonian, both the magnetic dipole moment operator mO and the diamagnetic magnetizability operator
O refer to an arbitrarily chosen gauge origin O. Alterna- tively, the Hamiltonian H(B) can be written in terms of the vector potentials of the electrons with respect to the gauge origin Ai⬅A(riO)⫽12B⫻(ri⫺O) as
H共A兲⫽1
2
兺
i 共⫺iⵜi⫹Ai兲2⫺兺
i,K rZiKK⫹12
兺
i⫽j r1i j. 共37兲 The dependence of the Hamiltonian H(B) on an arbitrarily chosen gauge origin is a manifestation of the gauge-origin problem, which hampers the calculation of magnetic molecu- lar properties within approximate methods.Thus, in the presence of a magnetic field, the polarizabil- ity and the oscillator strength are still defined according to Eqs. 共16兲–共33兲, where all contributions become magnetic- field dependent, for example, the Jacobian matrix becomes
A共B兲⫽具¯共B兲兩关H共B兲,共B兲兴兩CC共B兲典, 共38兲 where 兩CC(B)典⫽exp T(B)兩HF(B)典 and analogously for 具¯ (B)兩.
In terms of an orthonormal set of field-dependent orbit- als, the second-quantization form of the Hamiltonian H(B) becomes30,35
H共A兲⫽
兺
pq hpq共A兲Epq共A兲⫹12pqrs兺
gpqrs共A兲epqrs共A兲,共39兲 where the field dependence in the one-electron part arises from the implicit field dependence in the orbitals and the creation/annihilation operators as well as from the explicit dependence in the one-electron operator itself. Note that, in the second quantization formalism, also the dipole moment operator becomes magnetic field dependent, through the im- plicit dependence in the magnetic field of the orbitals and of the creation/annihilation operators.
C. The London orbitals and the relationship between different connections
Suppose that we have performed a Hartree–Fock calcu- lation at zero magnetic field and obtained the wave function 兩HF典and a set of optimized orbitals
p共0兲⫽
兺
C共0兲p, 共40兲where is a conventional Gaussian orbital centered on nucleus M at position RM. For any given magnetic induction B, a basis of atomic field-dependent orbitals, the London orbitals,19兵其 can be defined according to
共B兲⫽exp共⫺iAM Oe •r兲, 共41兲
where AeM O is the vector potential at the position RM O⫽RM⫺O of the nucleus M with respect to the chosen gauge origin,
AMOe ⫽12B⫻RM O, 共42兲 and r is the electron coordinate. Note that in absence of the perturbation (0)⫽.
The LAOs can be employed to define an orthonormal- ized basis of molecular field-dependent orbitals, from which an Hamiltonian operator defined at all values of the field can be constructed. This may be accomplished in two steps.
First, from the LAOs and the MO coefficients Cn(0) of the optimized orbitals at zero magnetic field, the unmodified mo- lecular orbitals共UMOs兲are constructed as30
q共B兲⫽
兺
共B兲C共0兲q. 共43兲Second, assuming that their overlap matrix S is nonsingular, the UMOs are further transformed into a set of orthonormal- ized molecular orbitals 共OMOs兲,30
p共B兲⫽
兺
q q共B兲Rq p共B兲. 共44兲The transformation matrix R⫽S⫺1/2U, where U is unitary defines the different connection schemes36,37which establish a one-to-one correspondence between orbitals at different values of the field. If U⫽1, the OMO basis in Eq. 共44兲cor- responds to a symmetric orthogonalization of the UMO basis—the symmetric connection.30For any general connec- tion scheme,
p
g⫽
兺
r rsUr pg ⫽兺
r rs关e⫺xg兴r p, 共45兲where the unitary matrix Ug was expressed in terms of an anti-Hermitian matrix xg. The indices ‘‘g’’ and ‘‘s’’ are used to distinguish between a general connection and the symmet- ric connection and explicit reference to the field dependence is suppressed for ease of notation. The orbital indices p, q, r, s are used for unspecified 共occupied or virtual兲 orbitals, i, j, k, l for occupied, and a, b, c, d for vir- tual. Note that the OMOs are not the optimized orbitals in the presence of the external perturbation but that they be- come identical to the optimized orbitals for the unperturbed system in absence of the perturbation.
D. The„effective…relaxed Hamiltonian and dipole operator
We now consider a general connection OMO basis and employ it to expand the Hamiltonian and parametrize the optimized Hartree–Fock state in the presence of the field according to共neglecting the purely nuclear contributions兲
H共B兲⫽
兺
pq hpqg Epqg ⫹12pqrs兺
gpqrsg epqrsg , 共46兲兩HF共B兲典⫽exp共⫺˜g兲兩HFg典, 共47兲
where兩HFg典 is a determinantal wave function built from the creation operators for the general OMO basis, and the opera- tor
˜g⫽
兺
pq nonred.
˜gpqEpqg 共48兲
generates共nonredundant兲orbital rotations among the OMOs.
Note that Epqg ,兩HFg典, and˜pq
g are magnetic field-dependent, but their dependence is suppressed for ease of notation.
To obtain V() andBwe need to take the total deriva- tive of matrix elements such as Eq.共38兲, that is, the deriva- tive of generalized transition expectation values between states defined at the same value of the field. As a conse- quence of Wick’s theorem, we can then neglect the magnetic field 共and gauge兲 dependence of the elementary creation/
annihilation operators corresponding to the OMO basis.30,33 As discussed in Ref. 37, this may be viewed as expanding the elementary operators at a given value of the field in the elementary operators at zero field. The Hamiltonian in Eq.
共46兲and the field-dependent Hartree–Fock state in Eq. 共47兲 can then be written in the ‘‘effective’’ form,
H共B兲⫽Hg⫽
兺
pqhgpqEpq⫹1 2pqrs
兺
ggpqrsepqrs, 共49兲
兩HF共B兲典⫽exp共⫺g兲兩HF典, 共50兲
where Epq, epqrs, and 兩HF典 refer to the optimized basis in absence of the perturbation, and
g⫽nonred.
兺
pq gpqEpq. 共51兲The elementspq
g of the anti-Hermitian matrixg, which depend implicitly on the field, are nonzero only for the occupied-virtual and virtual-occupied blocks—that is, they are the nonredundant parameters as determined from the共ef- fective兲Hartree–Fock orbital optimization condition, 具HF共B兲兩关Epq共B兲,H共B兲兴兩HF共B兲典⫽0
→具HFg兩exp共g兲关exp共⫺g兲Epqexp共g兲,Hg兴
⫻exp共⫺g兲兩HFg典⫽具HF兩关Epq,Hg兴兩HF典⫽0, 共52兲 where we have introduced the 共effective兲 relaxed Hamil- tonian in the given general-connection OMO basis
Hg⫽exp共g兲Hgexp共⫺g兲. 共53兲 In the following, we shall consider the structure of this op- erator.
Inserting Eq. 共45兲into the definition of the integrals of Hg and introducing the anti-Hermitian operator,
xg⫽
兺
pqxgpqEpq, 共54兲
where xpqg are the elements of the anti-Hermitian matrix xg, the Hamiltonian Hg may be written as
Hg⫽exp共xg兲Hsexp共⫺xg兲, 共55兲 where
Hs⫽
兺
pq hpqs Epq⫹12pqrs兺
gspqrsepqrs 共56兲is the effective Hamiltonian in the symmetric connection.
With this reformulation of Hg, the relaxed Hamiltonian in the given general-connection OMO basis becomes
Hg⫽exp共g兲exp共xg兲Hsexp共⫺xg兲exp共⫺g兲
⫽exp共rd兲exp共s兲Hsexp共⫺s兲exp共⫺rd兲
⬅exp共rd兲Hsexp共⫺rd兲, 共57兲 where
s⫽
兺
pq nonred.pq
s Epq, 共58兲
rd⫽
兺
occi j i jrdEi j⫹兺
abvir abrdEab, 共59兲with the anti-Hermitian matrices sandrdhaving nonzero elements corresponding to the nonredundant and redundant orbital rotations, respectively. The equivalence of the second and third expressions in Eq.共57兲follows since these expres- sions represent two equivalent parameterizations of a general unitary matrix.38
Let us derive a set of equations for the nonredundant parameterss. Inserting Eq. 共57兲in Eq.共52兲, we obtain 具HF兩关Epq,exp共rd兲Hsexp共⫺rd兲兴兩HF典
⫽具HF兩exp共rd兲关exp共⫺rd兲Epqexp共rd兲,Hs兴
⫻exp共⫺rd兲兩HF典⫽0, 共60兲 where pq refers to the nonredundant parameters ( pq⫽ai, ia). Because of the relations
exp共⫺rd兲兩HF典⫽兩HF典, 共61兲
exp共⫺rd兲Epqexp共rd兲⫽nonred.
兺
rs CrsErs 共62兲we find that Eq.共52兲 关or equivalently Eq.共60兲兴is satisfied if 具HF兩关Ers,exp共⫺s兲Hsexp共s兲兴兩HF典⫽0. 共63兲 This corresponds to the effective Hartree–Fock optimization condition in the symmetric connection, from which the non- redundant parameters, and thus the effective relaxed Hs op- erator, are determined.
The redundant parametersrdof Eq.共57兲are not deter- mined by the orbital optimization condition, but are implic- itly given through the choice of g, xg, and s. Since the relaxed Hamiltonian Hgis later used only in contexts where the property considered is invariant to redundant orbital ro- tations, explicit knowledge of these parameters is therefore not required.
Thus, even though the parametersgandsdetermined by Eq.共52兲and Eq.共63兲, respectively, are not the same, the final effective relaxed Hamiltonian in the general connection is for all practical purposes 共i.e., to within a redundant or- bital transformation兲 equivalent to the effective relaxed Hamiltonian in the symmetric connection. This is an impor-
tant result also in the sense that the symmetric connection may not always be the most useful approach from a numeri- cal point of view.39 Similar relations may be derived for other operators. In particular, for the effective relaxed dipole moment operator, we obtain
dg⫽exp共g兲dgexp共⫺g兲
⫽exp共g兲exp共xg兲dsexp共⫺xg兲exp共⫺g兲
⫽exp共rd兲exp共s兲dsexp共⫺s兲exp共⫺rd兲
⬅exp共rd兲dsexp共⫺rd兲, 共64兲 with dg⫽兺pqdpqg Epq and ds⫽兺pqdpqs Epq.
E. The effective CC equations and matrix elements In the presence of the magnetic field, the CC wave func- tion is parametrized in the OMO basis as
兩CC共B兲典⫽exp共⫺˜g兲exp共˜T
g兲兩HFg典, 共65兲
where T˜
g⫽兺t˜gg. Analogously, 具¯共B兲兩⫽具HFg兩g†exp共⫺T˜
g兲exp共˜g兲
⫽具g兩exp共⫺T˜
g兲exp共˜g兲, 共66兲
具⌳共B兲兩⫽具HFg兩
冉
1⫹兺
tនgg†冊
兩exp共⫺T˜g兲exp共˜g兲. 共67兲Again, according to Wick’s theorem, we can expand our creation/annihilation OMO basis in the generator basis at the expansion point, only retaining an implicit dependence on the magnetic field in the orbital and wavefunction param- eters,
兩CC共B兲典⫽exp共⫺g兲exp共Tg兲兩HF典, 共68兲
具¯共B兲兩⫽具HF兩† exp共⫺Tg兲exp共g兲
⫽具兩exp共⫺Tg兲exp共g兲, 共69兲 具⌳共B兲兩⫽具HF兩
冉
1⫹兺
t¯g†冊
兩exp共⫺Tg兲exp共g兲, 共70兲where Tg⫽兺tg.
The CC amplitudes are obtained by solving the 共effec- tive兲relaxed CC equations,
具兩exp共⫺Tg兲Hgexp共Tg兲兩HF典⫽0. 共71兲 Introducing the relation between the operators in general and symmetric connection, Eq.共57兲, we get
具兩exp共⫺Tg兲Hgexp共Tg兲兩HF典
⫽具兩exp共⫺Tg兲exp共rd兲Hsexp共⫺rd兲exp共Tg兲兩HF典
⫽0⬅具兩exp共rd兲exp共⫺T⬘兲Hsexp共T⬘兲兩HF典⫽0, 共72兲 and the effective expressions for the vectors and matrix ele- ments 共for the general connection兲 to be differentiated be- come
d⫽具兩exp共⫺Tg兲dgexp共Tg兲兩HF典
⫽具兩exp共⫺Tg兲exp共rd兲dsexp共⫺rd兲exp共Tg兲兩HF典
⫽具兩exp共rd兲exp共⫺T⬘兲dsexp共T⬘兲兩HF典, 共73兲
d⫽具HF兩
冉
1⫹兺
t¯g†冊
exp共⫺Tg兲关dg,兴exp共Tg兲兩HF典⫽具HF兩
冉
1⫹兺
t¯g†冊
exp共⫺Tg兲⫻关exp共rd兲dsexp共⫺rd兲,兴exp共Tg兲兩HF典
⫽具HF兩
冉
1⫹兺
t¯g†冊
exp共rd兲exp共⫺T⬘兲⫻关ds, exp共⫺rd兲exp共rd兲兴exp共T⬘兲兩HF典, 共74兲 A⫽具HF兩† exp共⫺Tg兲关Hg,兴exp共Tg兲兩HF典
⫽具HF兩†exp共rd兲exp共⫺T⬘兲
⫻关Hs, exp共⫺rd兲exp共rd兲兴exp共T⬘兲兩HF典, 共75兲
F⫽具HF兩
冉
1⫹兺
t¯g†冊
exp共⫺Tg兲关关Hg,兴,兴⫻exp共Tg兲兩HF典
⫽具HF兩
冉
1⫹兺
t¯g†冊
exp共rd兲exp共⫺T⬘兲⫻关关Hs,e⫺rderd兴,e⫺rderd兴exp共T⬘兲兩HF典. 共76兲 Since we here consider states that are invariant with respect to rotations among the occupied orbitals and among the vir- tual orbitals, we may instead of the projection manifold 兵具HF兩† exp(rd)其use the manifold兵具HF兩†其. Equation共72兲 and the matrix elements above then reduce to those of the symmetric connection so that the same result is obtained with the general and symmetric connections.
As shown in Ref. 30, the effective Hamiltonian in the symmetric connection Hs is gauge-origin independent.
Gauge independence was the result of the gauge-origin inde- pendence of both the integrals over London orbitals and the connection matrix. It is straightforward to show that, in the symmetric connection, the effective dipole-moment operator is gauge-origin independent as well. As a consequence of the gauge independence of the effective Hamiltonian in the sym- metric connection, the effective s operator is gauge inde- pendent. Hence, Hs and ds are gauge-origin independent, ensuring that the final result is gauge-origin independent.
F. The canonical formulation
We here consider the case where the Hartree–Fock or- bitals are required to be canonical. Thepq
g matrix now con- tains both redundant and nonredundant orbital rotation pa- rameters, which can be determined by solving the effective canonical equations,