Analytic energy derivatives in manybody methods. I.

First derivatives
E. A. Salter, Gary W. Trucks, and Rodney J. Bartlett

Citation: The Journal of Chemical Physics 90, 1752 (1989); doi: 10.1063/1.456069
View online: http://dx.doi.org/10.1063/1.456069
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/90/3?ver=pdfcov
Published by the AIP Publishing

Articles you may be interested in

Manybody methods for excited state potential energy surfaces. II. Analytic second derivatives for excited state
energies in the equationofmotion coupled cluster method
J. Chem. Phys. 103, 8931 (1995); 10.1063/1.470083

Implementation of analytical first derivatives for evaluation of the manybody nonadiabatic wave function with
explicitly correlated Gaussian functions
J. Chem. Phys. 96, 9013 (1992); 10.1063/1.462259

Derivation of manybody potential among charged particles in the Smatrix method

J. Math. Phys. 33, 2303 (1992); 10.1063/1.529602

Analytic energy derivatives in manybody methods. II. Second derivatives

J. Chem. Phys. 90, 1767 (1989); 10.1063/1.456070

Analytic Function Method for the ManyBody Problem

J. Math. Phys. 11, 2463 (1970); 10.1063/1.1665411

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Mon, 24 Nov 2014 10:46:46
 Analytic energy derivatives in many-body methods. I. First derivatives
E. A. Salter, Gary W. Trucks, and Rodney J. Bartlett
Quantum Theory Project, University ofFlorida, Gainesville, Florida 3261 1
(Received 13 June 1988; accepted 19 October 1988)
The theory of analytic energy derivatives is developed for the coupled cluster (CC) model
using diagrammatic techniques. Explicit expressions for the derivative energy and response
density for thefull coupled-cluster singles, doubles and triples (CCSDT) model are presented.
Analytic derivatives for the finite-order MBPT models through MBPT ( 4) and the recently
proposed "quadratic" CI models are derived as special cases of the theory. First derivatives of
the energy correspond to first-order response properties and molecular gradients; the analytic
expressions for the derivative energy are given in terms of the response (or "relaxed") density
for efficient evaluation. The theory of analytic second derivatives of the CC/MBPT energy is
presented in part II.

J. INTRODUCTION to be simple dot products between the response or "relaxed"

The analytical evaluation of energy derivatives is one of MBPT/CC density matrix and the one-electron perturba-
the foremost achievements of modern ab initio quantum tion matrix of interest.
chemistry.1-5 It is a necessary element in the prediction of Molecular gradients are treated as special cases of a re-
molecular structure and properties. Derivatives of the ener- sponse property, wherein the basis functions are dependent
gy with respect to nuclear displacement parameters, called upon the nuclear displacement perturbation. The basis func-
"molecular gradients," represent the forces acting on the tions, usually Gaussian, are fixed to the nuclei, and when the
nuclei and are necessary input for search algorithms which nuclei move, the basis functions move as well. Molecular
locate critical points on a molecule's potential energy sur- gradient expressions have therefore, in addition to a re-
face. Once a critical point has been found, the second deriva- sponse-property term, terms involving the derivative of the
tives of the energy, also evaluated analytically, or perhaps by atomic orbital basis functions in two-electron integrals and
finite-difference of analytic molecular gradients, character- in overlap integrals.
ize the point as a minimum (a stable molecular structure) or The evaluation of energy derivatives of variational
a saddle point (a transition state) or neither. Second deriva- methods can be simplified in ways, which, in general, the
tives of the energy make possible the prediction of harmonic evaluation of energy derivatives of many-body methods can-
vibrational frequencies and intensities, as well. not. The energy is generally a function of the molecular orbi-
Without analytical derivative methods, a search of the tal coefficients c, the coefficients in the determinantal expan-
potential energy surface for polyatomic molecules with more sion of the wave function C, and the one- and two-electron
than four atoms (and no symmetry) would be effectively integrals (l) involving the atomic orbital basis set functions.
impossible. Derivatives of the energy with respect to exter- The derivative of the energy is, in general,
nal field perturbation parameters are directly associated aE [c(X),c(X),l(X)] = aE ~ + aE ac + aE ~.
with the one-electron properties of the molecule such as mo-
aX ac ax ac ax aI ax
ments, field gradients, spin densities, etc. The prediction of
molecular structure and properties using highly correlated In the case of the SCF model, the wave function is a single
many-body methods [coupled-cluster theory (CC) and determinant and the energy is optimal with respect to
many-body perturbation theory (MBPT)] have become changes in the molecular orbital coefficients; that is,
standard in quantum chemical studies.6-17 In this work, the aE I Jc = O. Hence, it is not necessary to evaluate aclaX, the
general diagrammatic theory and procedures for the effi- derivative of the molecular orbital coefficients with respect
cient evaluation of analytic first derivatives of the to the perturbation parameter. If X represents a nuclear dis-
CC/MBPT energy through the full coupled-cluster singles, placement parameter, the third term indicates that it is nec-
doubles, and triples model (CCSDT) are presented. In addi- essary to evaluate aE / aI, that is, how the SCF energy varies
tion, the analytic derivative of the proposed "quadratic" CI relative to the one- and two-electron integrals, and in turn to
method is extracted as a special case of the CC derivative evaluate aI I aX, or how the integrals vary relative to the nu-
theory. clear displacement parameter. These two such contributions
One-electron response properties are often computed in distinguish the generalized Hellman-Feynman theorem
the framework of approximate correlated models as deriva- from the ordinary Hellman-Feynman theorem. Similarly, in
tives rather than as expectation values. For many approxi- the case of the MCSCF model, we have aE/aC=O and
mate models of correlation, however, the expectation value JE I Jc = 0, and again only the third term remains. In CI
and derivative are not the same. In fact, the derivative is the methods, the energy is optimal with respect to the determin-
expectation value plus additional terms which usually bring antal coefficients only; that is, we have aE lac = 0 and
the derivative closer to the full CI property value. First-or- aEIJc#O.
der response properties in the MBPTICC model are shown In the standard coupled-cluster method6-14 and its fi-

1752 J. Chem. Phys. 90 (3), 1 February 1989 0021-9606/89/031752-15$02.10 © 1989 American Institute of Physics

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Mon, 24 Nov 2014 10:46:46
 Salter, Trucks, and Bartlett: Energy derivatives in many-body methods. I 1753

nite-order MBPT approximations,8,9,15-17 the energy is not well. 29 The most recent gradient work 21 ,27 has incorporated
optimal with respect to the molecular orbital coefficients or the procedure of Handy et al. 30 for the elimination of singu-
the determinantal coefficients. Thus, it is necessary to take larities which arise when there are degenerate or near-degen-
into account the derivative of the CC energy with respect to erate orbital eigenvalues or has eliminated the singularities
the SCF coefficients and with respect to the determinantal by factorization. 28
coefficients, and in turn, the derivative of the SCF coeffi- In this discussion of the theory of analytic derivatives in
cients and the determinantal coefficients with respect to the many-body methods, the full CCSDT derivative energy ex-
perturbation parameter. It might appear that, for the evalua- pressions are derived using diagrammatic rules developed
tion of several properties and/or the 3N molecular gradients for second-quantized operators. The expressions apply to all
(for a system of N nuclei), it is necessary to evaluate ac / aX CCSDT-n 13 approximations to the CCSDT model, to
and ac/aX for each atomic displacement or each external MBPT models through MBPT( 4) and to the recently pro-
field perturbation of interest. However, the equation govern- posed "quadratic" CI methods 31 (QCI) since the latter are
ing the response of the SCF coefficients is a linear equation, simply restricted CC methods and are consequently special
as is the equation governing the response of the CC deter- cases of the general theory. The finite-order MBPT deriva-
minantal coefficients. Consequently, we were able to tive models through full fourth order are obtained by iter-
show l8 ,19 that the first and second terms of the CC/MBPT ation of the CC derivative equations. The theory of first de-
energy derivative expression can be transformed into a com- rivatives will prepare for the development of the theory of
putationally efficient form by using the Z-vector method, analytic second derivatives presented in the following paper.
previously used only to avoid solving CPHF equations for
several perturbations. 20 The determinantal coefficient de- II. THE COUPLED-CLUSTER MODEL
rivatives for each perturbation need not be determined; only Consider the Schrodinger equation corresponding to a
a single, perturbation-independent linear equation must be molecular system, where the electronic Hamiltonian H(X),
solved. This has made it possible to efficiently compute its eigenfunction 'I'(x) , and its eigenvalue E(X) are depen-
CCSD and CCSDT-1 gradients. 21 The full CCSDT gradient dent upon some parameter X:
equations presented here are far more complex, and we have
(1)
been required to develop more powerful derivational proce-
dures to obtain them. To facilitate the derivation, we intro-
duce a second-quantized operator A; the A-amplitude equa-
H(X) = - 112I Vi - I
i ~a
Za r;;; 1+ IxQ;
i
+I
i»
rij I.
tion becomes the desired perturbation-independent linear The parameter may control the displacement of an atomic
equation. nucleus in the molecular system, or it may be the strength of
Analogously, the evaluation of the SCF coefficient de- an external one-electron potential energy perturbation QjJ
rivatives for each perturbation can be avoided 20; once again, such as a uniform electric field applied in the z-direction: xQ;
it is necessary to solve only a single, perturbation-indepen- = xz;. In correlated methods, we seek the solution to Eq.
dent linear equation in order to evaluate the energy deriva- ( 1) in the space of the N-electron determinants (antisymme-
tive. A special property, the response or "relaxed" density, trized products) formed from a basis of orthonormal one-
emerges naturally from the transformation to perturbation- electron spin-orbitals {cp(X)}. The solution is built upon an
independent expressions. approximate single-determinant wave function, called the
Analytic gradients for the SCF, I MCSCF,2 CI,3 and reference function and denoted by 10). The total energy of
MBPT(2)4 models have been available for some time. The the molecular system, within the Born-Oppenheimer or
analytic evaluation of CC/MBPT gradients is partly built "fixed nuclei" approximation, is obtained by adding the nu-
upon the CC response theory22; molecular orbital and atom- clear repulsion energy to the electronic energy E(X)'
ic basis relaxation, ignored in that approach, were subse- The reference energy is defined as the expectation value
quently incorporated. 18,19,24 The first reports of the theory of of the Hamiltonian over the reference wave function
MBPT(3), D-MBPT(4), and CCD gradients 25,26 were not (OIH(X) 10). Subtraction of the reference energy from each
suitable for routine applications since they did not take ad- side of Eq. (1) yields
vantage of the linear nature of the CC derivative equations.
[H(X) - (OIH(X) 10) ]'I'(X)
The general theory of CC derivatives was later presented in
an efficient form, 18,19 and, building upon this theory, analyt- = [E(X) - (OIH(X) 10) ]'I'(X)
ic CCSD and CCSDT-1 gradients were subsequently imple- or
mented. 21 Analytic SDQ-MBPT ( 4) gradients have been re- HN(X)'I'(X) = aE(X)'I'(X)' (2)
cently reported,27 but the implementation did not take
H N (X) is the "normal product" form of the Hamiltonian
advantage of the linear nature of the SCF derivative equa-
and is represented in terms of the second-quantized opera-
tions. Also recently, we reported the theory and application
tors by
of a computationally efficient approach to MBPT (3) gradi-
ents. 28 In the report, we presented the detailed theory of
MBPT (2) and MBPT ( 3) response densities for the first
HN(X) = ~(hpq + +
(Pillqi»){p+q}
time and included values ofMBPT (2) and MBPT (3) dipole
moments as examples of analytic property evaluation. We (3)
have studied properties at the SDQ-MBPT( 4) level as r,s

J. Chem. Phys., Vol. 90, No.3, 1 February 1989

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Mon, 24 Nov 2014 10:46:46
 1754 Salter, Trucks, and Bartlett: Energy derivatives in many-body methods. I

By convention, the indices iJ,k,I, ... represent the spin-orbi- uct operators over the reference (or Fermi vacuum), only
tals which are occupied in the reference function 10). The the fully contracted products are nonzero.
indices a,b,c,d, ... correspond to the virtual (or unoccupied) The representation of the Fock operator in the orbital
spin-orbitals. The indices p,q,r,s, ... are unrestricted. basis is given by the matrix elements
The symbols hpq and {pqllrs) in Eq. (3) represent inte-
grals wherein the one- and two-electron operators of the
Hamiltonian act upon the orbital basis functions:
So we may represent H N (X) as
hpq = I ¢:( 1)( - 1I2vi - ~Zaral\ + XQI)
HN(X) = IJpq {p+q} +~I {pqllrs){p+q+sr} (4)
P.q 4 p.q.
X ¢q (1 )aT\aU1 r.s

{pqllrs) = II ¢:(1)¢:(2)[ri;I(1-PI2)]
or
HN(X) =iN(X) + WN(X)·
X¢r (1 )¢s (2)aT 1aU1aT2aUZ' If any SCF wave function (canonical or not) is chosen
The variables U \ and T I are the coordinates of spin and space as the reference function, (OIH(X) 10) is the SCF energy,
of electron number one, respectively. The operator P 12 is the aE(X) is the correlation energy, and the Hamiltonian sim-
permutation operator; thus the integral {pqllrs) is antisym- plifies to
metric with respect to the exchange of p with q or r with s.
The integral hpq is explicitly dependent upon X if a one-elec- HN(X) = I/;j{i+j} + Ilab{a+b}
;j a,b
tron potential energy perturbation Q\ is present. Of course,
if the perturbation is a nuclear displacement, the hpq inte-
grals depend upon X through the operator Za / ral as well. In
+ -1 I {pqllrs){p+q+sr},
4 P.q.
general, both hpq and {pqllrs) are implicitly dependent upon r.s
X through the orbital basis functions, since the basis func- since the matrix elements lai and /;a are zero for SCF orbi-
tions vary with molecular geometry or external field tals. For the particular choice of canonical SCF orbitals, we
strength. have, by definition,iij = l>ijEi andlab = l>abEa' where Ep de-
Excited determinants are represented by a string of cre- notes the molecular orbital eigenvalue of orbital p.
ation and annihilation operators acting on the reference The CC method6-l4 proceeds by choosing an exponen-
function, whereby an electron in an occupied orbital is "an- tialform of '11 (X). That is,
nihilated" and an electron is subsequently "created" in a
virtual orbital: 'I1(X) = eT(XlIO), (5)
where reX) is a sum of excitation operators
{a + i} 10) = IS)
singly excited determinant,
{a+ib +j}IO) = ID) doubly excited determinant,
T(X) = T1(X) + T 2(X) + T3 (X) + ....
{a+ib +jc+k}IO) = IT) triply excited determinant, The excitation operators are given by

T1(X) = I t~(x){a+i},
Brackets enclosing a second-quantized operator string i,a

{ABeD .. .} denote a normal product of operators. The nor-

mal product is defined by moving all a,b, ... and i+ ,j+ , ...
operators to the right of the operator string by use of the a.b
anticommutation relations:
Tn (X) = - -12 "ab"'(){
L t ij'" X a +'b+'
I ]" •• } , •.• ,
[a+ ,b) + = I>ab and [i+ J] + = I>ij' (n!) iJ•...
a.b....
The normal-ordered form of the Hamiltonian 32 in Eq. (3) is
where the t amplitUdes are to be determined. Substitution of
obtained by application of Wick's theorem to the ordinary
Eq. (5) into Eq. (2) yields
second-quantized Hamiltonian. A contraction between op-
erators A and B defined as HN (x)eT(Xl 10) = aE(x)eT(XlIO)
AB=AB- {AB} or
is only nonvanishing when (6)
f+}=l>ij and ';;b+ =I>ab' or
The generalized Wick's theorem 32 states that a product of (7)
operator strings in which some operator strings are already where c indicates the limitation to connected diagrams. 7

in normal product form is given as: the normal product of By projecting Eq. (7) on the left by (01 , we obtain the
the overall string plus the sum of all possible normal prod- CC energy expression:
ucts with contractions, excluding contractions between pairs
aE(X) = (01 [HN(x)eT(Xl]cIO). (8)
of operators within an original normal product. A corollary
is that for the expectation value of a product of normal prod- The t amplitUdes are determined by solving the set of cou-

J. Chern. Phys., Vol. 90, No.3, 1 February 1989

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Mon, 24 Nov 2014 10:46:46
 Salter, Trucks, and Bartlett: Energy derivatives in many-body methods. I 1755

V_Ox + to_ox + }---x + A--x

FIG. 2. The diagrammatic repre-
sentation of coupled-cluster exci-
<pqllrs> {ptqtsr} = tation operators T. (X), T2 (X),
and T,(X)'
V_V + f-~ + }--i + f- A + 44

+ to_V + }--1 + }--{ + 1--4

FIG. 1. The diagrammatic representation of HN(X).
FIG. 3. The diagrammatic repre-
6E sentation of coupled-cluster en-
pled equations obtained by projection of the singly, doubly, ergy.
triply, ... excited determinants (S,D,T, ... ):
(SI[HN(x)eT<X)]cIO) =0, (9)
(D I [HN(x)eT<X)]c 10) = 0, V__ " \Lo--" \1..-0 v..o V--·
2a 2b 2c 3a
v-"3b V
3c
(TI[HN(x)eT<X)]cIO) =0,

Various CC models result from choices of truncation of

o-oJ) W-- Vn
4 Sa 5b 5c 6a 6b
T(X)' For example, the CCD model is defined by the choice
T(X) = T2 (X), the CCSD model is defined by T(X)
= T, (X) + T2 (X), and the CCSDT model is defined by Vn \NJ
T(X) = T, (X) + T2 (X) + T3 (X). In the limit of including 6c 7
all excitation operators in T(X), the full CI wave function
and energy are recovered. FIG. 4. The diagrammatic representation of the T. (X) amplitUde equation
Diagrammatic rules 32 are used to identify all of the fully in the CCSDT model.
contracted products in the CC amplitude equations and en-
ergy expression; the diagrams are then interpreted algebrai-
cally. TheoperatorsIN(x) and WN(X),whichcomposethe
normal product Hamiltonian H N (X), are represented dia-
grammatically in Fig. 1. The CC excitation operators 20 2b 2e 2d 2e 30 3b
T, (X), T2 (X), and T3 (X) are represented by the diagrams
in Fig. 2. Please note that although there are lines in the
"middle" of the T3 diagram as opposed to the lines at the
"edges," we mean for all virtual (or occupied) lines to be 40 4b 4e 50 5b
equivalent to one another. The diagrams representing each
of the fully contracted products in the CC energy expression
are shown in Fig. 3. The diagrammatic equations for the
~-oY ~ xVJjj xWJ) 'U-V
5e 5d 60 6b 6e
T, (X), T2 (X), and T3 (X) amplitudes are shown in Figs. 4,
5, and 6, respectively. The explicit algebraic expressions for
the CCSDT model are presented elsewhere. '3
For the choice of canonical SCF orbitals as the orbital
VJJV
6d
V:3J
6e
VJN
6f
VJln 6g
basis, the terms involving the IN (X) operator simplify be-
causelpq is zero except along the diagonal where the orbital
eigenvalues are. The low-order iterative solution to the CC V-V
7b
VV
7e
rroJlj)
80
\tiW
8b
V£yV
8e
equations lead to the MBPT (2), MBPT (3), and MBPT ( 4 )
models. The same procedure will be employed in this work
to define the MBPT analytical derivative equations. Similar-
ly, all restricted CC methods like CCSD-l, CCSDT-l, etc. 90 9b ge 9d
are defined by retaining a subset of the full CCSDT terms,
usually by imposing a criterion requiring the method to be at
least complete through some order in the correlation pertur- ~-'Wjj \,N-V W-V 'yJ-W
bation for classes of Tn operators. ge 100 lOb II
The "quadratic" C1 3 ' (QCI) is simply a restricted CC
method; it is not CI since it is not variational nor an eigenval- FIG. 5. The diagrammatic representation of the T2 (X) amplitude equation
ue equation, and it is an approximate cluster expansion that in the CCSDT model.

J. Chem. Phys., Vol. 90, No.3, 1 February 1989

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Mon, 24 Nov 2014 10:46:46
 1756 Salter, Trucks, and Bartlett: Energy derivatives in many-body methods. I

\[j-VjJ ll.E = (0IHNT2 10),

2b (S 1 [HN(T\ + T2 + T\T2) ]cIO) = 0,
(D 1[HN(l + TI + T2 + T~/2) ]cIO) = O.
\LJ-~
2c
\L5J.lY
2d
\f5Njj
2.
Similarly, QCISD(T) is a restricted CCSDT model, very
close to the CCSD + T (CCSD) model discussed else-
where. 14
Vjjjj--x \jj£jj--x V!siY
30 3b 3e The choice of a particular CC or MBPT model defines
an approximate solution to Eq. (2). The approximate total
energy of the molecular system is obtained by adding the
\fjjjJ
3d
\8iY 'j---yy
3. 40
nuclear repulsion energy and reference energy to the ap-
proximate ll.E x. In the absence of analytic methods, the de-
rivative of the total energy at X = 0 is computed by finite
V--Vjj V--VJj V--Vjj difference of the total energy evaluated at a small positive
4b 4e 4d
and a small negative value of X. In the next section, we will
obtain analytical expressions for the evaluation of
\Lj--0LY
50
\;Wj[y
5b
~
5e
(all.E / aX) 1x ~ 0' so that together with analytic derivatives of
the reference energy and the nuclear repulsion energy, we
will have the analytic derivative ofthe total energy.

5d 5. 5f III. FIRST DERIVATIVES OF THE COUPLED-CLUSTER

ENERGY
A. Derivation of general expressions
~
5g
V--Ujy
60
V--L:lY 6b Let us return to Eq. (6), where the Hamiltonian is in
normal-ordered form and the wave function is the exponen-
1----\jjjjj fi\[jjjj 'NJ£Y tial of the coupled-cluster model:
6e 6d 68
e- T<xlHN(x)eT<XlIO) = ll.E(X) 10). (10)

~ 'y.fjjjj \tttYY Let us define the following:

6h
6f 6g
T= T(X) Ix~o, TX = aT(X) I '
w-w 70
W--Vjj VWY
7b 7e
HN =fN +
aX x~o
WN = [fN(X) + WN(x)] Ix~o,

H>Jv =f>Jv + W>Jv = (afN(X) + aWN(X»)1 '

ax ax x~o
7d 80 8b
ll.E=ll.E(X)lx~o, ll.EX= all.E(x)
ax x~o
I .
'vtiADJ8e
~
8d
~
8.
Let the symbollP ) represent the space of orthonormal deter-
minants in which the amplitudes of T(X) and ll.E(X) are
determined. Let the symbol 1<1» represent the excited deter-
Q-WJjjj ff-WJjJj ~ minants only:
90 9b ge
i = IP)(PI,
(11 )
~~
9d 9.
W-\:CyJ
100
IP) = 10) + 1<1»·
Projection on the left by (P 1 and evaluation at X = 0 yields
the unperturbed coupled-cluster expressions:
lOb (01 (HNeT)c 10) = ll.E (12)
and
FIG. 6. The diagrammatic representation of the T, (X) amplitude equation
in the CCSDT model. (13)
Of course, the unperturbed CC model (X = 0) is a special
case of the perturbation-dependent CC model discussed in
does not include all quadratic cluster terms. Hence QCISD the previous section. Equations ( 12) and (13) are Eqs. (8)
is simply an approximation to CCSD, the same as the and (9) evaluated at X = 0, respectively.
CCSD-l method of long ago, \0 except for the T\ T2 term in The derivative ofEq. (10) with respect to X, evaluated
the T\ equation and the exclusion of the T i term in the at X = 0, is given by the following, where we have used the
energy expression: fact that excitation operators commute with one another:

J. Chern. Phys., Vol. 90, No.3, 1 February 1989

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Mon, 24 Nov 2014 10:46:46
 Salter, Trucks, and Bartlett: Energy derivatives in many-body methods. I 1757

{[ (HNeT)C,TX] + (HX,eT)JIO) = aEXIO). (14) CCSDT model we have A = Al + A2 + A 3. We require

that the A amplitudes satisfy the condition
Or more simply,
(OIAI<I» = - (01 CHNeT)c 1<1»(<I>I CHNeT)c - aEI<I»-I.
{[ (HNeT)cTX]c + CHX,eT)JIO) = aEXIO).
By defining A in this manner, we obtain an expression
Projection on the left by (P Iyields an expression for aEx and
for aEx that is not dependent upon T x. We have
a set of linear equations which determine the derivative t-
amplitudes: (01 (HX,eT)cI O) + (OIAI<I»(<I>I (HX,eT)c 10) = aEx,
which can be evaluated if we first solve

and (OIAI<I» (<1>1 (HNeT)c - aE 1<1» + (01 (HNeT)c 1<1» = O.

Analogous steps have been often used in recent derivative
literature to reduce the number of linear equations to be
For the CCSDT model, Eq. (16) represents three coupled, solved from 3N to just one. The method is sometimes re-
linear equations for the T f, T J, and T J amplitudes. If the ferred to as the Z-vector method. 2o,24
unperturbed CCSDT amplitudes are available, the deriva- We now wish to find equivalent forms ofthe above ex-
tive amplitude equations can be constructed and solved. The pressions which can be readily subjected to the usual
derivative energy aE x can then be evaluated by Eq. (15). diagrammatic rules. Recalling that i = IP) (P I = 10) (01
However, it is not necessary to solve Eq. (16) for the + I<1» (<I> I, the derivative energy can be written as
derivative t amplitudes in order to evaluate aE x. Returning aEx= (01 (HX,eT)cI O) + (OIA(HX,eT)cIO), (17)
to Eq. (14), we insert i = IP) (P I into the terms involving
and the A equations can be written as
T x and left project by (P I to obtain
(01- TXIP)(PI(HNeT)cI O) + (01 (HX,eT)cI O) (OIA(HNeT)C 1<1» -!l.E (OIAI<I» + (01 (HNeT)c 1<1» = O.
If we consider
+ (OI(HNeT)cIP)(PITXIO) = aEx
(01 (HNeT)cAI<I»
and
= (01 (HNeT)cIP)(PIAI<I»
(<1>1- TXIP)(PI(HNeT)cI O) + (<I>I (HX,eT)c 10)
= (01 (HNeT)cIO)(OIAI<I»
+ (<1>1 CHNeT)c IP) (P ITXIO) = o.
But T x is an excitation operator, hence (01 T XIP) = O.
+ (01 (HNeT)C 1<1» (<I> IA I<I> )
Furthermore, T is the solution to the' unperturbed = aE (OIAI<I» + (01 (HNeT)c 1<1» (<I>IAI<I»,
equation (<1>1 (HNeT)c 10) = O. Also, recall that then it is clear that we may write the A equations in terms of
(01 (HNeT)c 10) = aE. Now we have the commutator [A,(HNeT)C]' Of the commutator, of
course, only the connected product is nonzero. Finally, the
(01 (HX,eT)c 10) + (01 (HNeT)c 1<1»(<I>I TXIO) = aEx A equations become
and
(<1>1- TXIO)aE+ (<I>I (HX,eT)cI O)
(01 [ A(HNeT)c ] c 1<1» + (01 (HNeT) c1<1» (<I>IAI<I»

+ (<1>1 (HNeT)C 1<1» (<1>1 TXIO) = O. + (01 (HNeT)C 1<1» = O. (18)

The second equation is solved for (<I>ITXIO): Equations ( 17) and (18) summarize the CC energy de-
rivative. In Eq. (17), !l.EX is expressed in terms of unper-
(<I>I TXIO) = - (<I>ICHNeT)c -aEI<I»-1
turbed t amplitudes, the derivative integrals of H N' and the A
X (<1>1 CHX,eT)c 10 ), amplitudes. Equation ( 18) is the coupled, linear set of equa-
tions which determine the A amplitudes. The equations are
and we then substitute for (<1>1 TXIO) in the aE x expression:
general in that they apply to CC models of any choice of
(OI(HX,eT)cIO) - (OICHNeT>cI<I» truncation of T (and the corresponding choice of A and pro-
X (<1>1 (HNeT)C - aE 1<1» -1(<1>1 (H X,eT)c 10) = !l.E x. jection space IP) ). Also, the equations apply to CC models
where non-Hartree-Fock orbitals are chosen as the spin-
Now we define an antisymmetrized, perturbation-indepen-
orbital basis, as well as when SCF orbitals are chosen. The
dent, deexcitation operator A. Let A = Al + A2 + A3 + ... ,
equations apply generally to CC models for open-shell sys-
where
tems, where unrestricted Hartree-Fock (UHF) orbitals are
~ /l,1 ai {.+}
AI = ~ I a,
~ /l,Hiab {.+
A2 = -I ~ I a]
'+b} , ... used, or other different orbitals for different spin cases.
i,a 4 iJ The diagrammatic representation of the m operator
a,b and the A operators are given in Figs. 7 and 8, respectively.
An = --12 ~ ~ 1 ij'"
/l, ab .. · {.+
I a]'+b • .. } , ....
Please note that as with the representation of the T3 opera-
(n!) iJ.... tor, we mean for the virtual (or occupied) lines of the A3
a.b, ... operator to be equivalent to one another. The fully contract-
For every Tn in T we will have a corresponding An in A. ed products which contribute to the CCSDT derivative ener-
That is, for the CCD model we have the truncation A = A2 . gy expression are represented diagrammatically in Fig. 9.
For the CCSD model we have A = Al + A 2, and for the The number of derivative energy diagrams require us to

J. Chem. Phys., Vol. 90, No.3, 1 February 1989

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Mon, 24 Nov 2014 10:46:46
 1758 Salter, Trucks, and Bartlett: Energy derivatives in many-body methods. I

fpcf { ptq} = V~~X + t===x + f==x + A==x

<pqllrs>x {ptqtsr} =

'<toY + r=='Y + }:={ + }=A + A7~

+ t··Y + } •• -! + f--{ + }--A
1-8 1-9 1-10 I-II 1-12 1-13

FIG. 7. The diagrammatic representation of HNY.

show a truncated set here, but the full set of diagrams, along 1-14 1-15
with their algebraic interpretations are provided in the
PAPS supplement to this paper.44
In the CCSDT model, Eq. (18) represents three cou- 2-2 2-3 2-4 2-5 2-6 2-7 2-8
2-1
pled, linear equations which determine the amplitudes of the
AI' A2, and A3 operators:
(01 [A(HNeT)c]cI S ) + (OI(HNeT)cIS) =0,
(01 [A(HNeT)c]cI D ) + (OI(HNeT)cIS)(SIAID) 2-9 2-10 2-11 2-12 2-13

+ (01 (HNeT)C ID) = 0,

(01 [A(HNeT)c]cIT) + (OI(HNeT)cIS)(SIAIT)
2-14 2-15 2-16 2-17 2-18

+ (OI(HNeT)cID)(DIAIT) =0.
The above equations are represented diagrammatically in
Figs. 10-12. Again, the number of diagrams require us to
present a truncated set of diagrams here, but all diagrams, 2-19 2-20 2-21 2·22 2-23 2-24

along with their algebraic interpretations are provided in the

PAPS supplement. 44 Note that although there are discon-
nected parts of the A2 and A3 amplitude equations, there are
no closed and disconnected parts; the energy derivative is 2-25 2-26 2-27 2-28 2-29

necessarily linked. For the CCSD model and restricted Har-

tree-Fock (RHF) orbitals, the Al and A2 equations are of
course equivalent to that obtained by Scheiner et al. 27
Comparison of the A-amplitude equations with the un- 2-30 2-31 2-32 2-33

perturbed t-amplitude equations (i.e., Figs. 4-6, evaluated at

X = 0) reveals that the A-amplitude equations possess sever-
al diagrams which are "upside-down" versions of diagrams
2-34 2-35 2-36 2-37
in the t-amplitude equations. The algebraic interpretations
of such diagrams are related by complex conjugation; if the
spin-orbital basis functions are real, the expressions are
3-1 3-2 3-3 3-4
equivalent. For the linearized CC models, there is a one-to-
one correspondence between the diagrams in the A- and t-
amplitude equations; hence, the A amplitudes are equivalent [O==[j)
to the unperturbed t amplitudes. Consider the LCCD model, 3-5 3-6 3-7

where T = T2 and the quadratic T2 contributions to T2 are

not permitted. Specifically, the T2 equation to be solved con-
3-8 3-9 3-10

ffi a--~ ~ 0 ~n··U

3-11 3-12 3-13

\ ij {'t 'tb}--
/\ob I OJ
FIG. 8. The diagrammatic repre-
sentation of the deexcitation op-
erators AI , A" and A 3 • m 3-14
R]
3-15
rn 3-16

FIG. 9. The diagrammatic representation of coupled-cluster energy deriva-

tive (truncated).

J. Chern. Phys., Vol. 90, No.3, 1 February 1989

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Mon, 24 Nov 2014 10:46:46
 Salter, Trucks, and Bartlett: Energy derivatives in many-body methods. I 1759

I\---X
/\----7\ 1\ x I\---x l\ x f--A
2 3

3 6
"A--~
4
h--A 1v\-A fv\7\ f\7\--x f\7\--x
5 6 7 8 9

~-7\ '1\.--] ~--1\

10
A--n L-A " 12
h7\ 10
frl\ AO-7\ 12
~-~ ~-~
13 14

~ ~ (A-7\
13 14 15
V--o Von 16 17
~ 15
n--" Al\--o n--"
16 17 18
~ 19

AI}--x Af}--x
18 19
f\1J !\fJ 9 9
20 21 22 23
nCn !\7fJ ~ ~--IJ
20 21 22 23

Al- n
24
/r--x R R
25 26 27
~--X ~--X
28 29
AfL---A ~ ~~
~ H n ~

~--X
30
A"W--x
31
[I}--1\
32
W\ 33
N/t AW--l\ ~-D Ri7 ~--o
28 29 30 31 32

Q 34
Alb 35
[j-A
36
Q G7\
37 38
33 34 35 36 37

FIG. 11. The diagrammatic representation of the A, equation in the

39 40 41 42 43
CCSDT derivative model (truncated).

~ Af}-]
44 45 linear equations; thus, for the CCD model, the A2 equation
must differ from the T2 equation in that the quadratic dia-
al.a7\46
~ 47 48 49
grams (5a)-(5d) in Fig. 5 are absent. Instead, we find the
similar, related diagrams (21 )-(27) in Fig. 11. Ifwe denote
the quadratic contributions to T2 by the function
A--W] ~Q Q ':/ (l,t X t), the related contributions to A2 are given by
50 51 52 53 Q':/(t,I XA) + Q't/(t,A XI).
For models including T I , we have previously identified
~ 1JI\--J) 1JI\--J) ~n] the linear contributions of TI to T 2 , represented by diagrams
54 55 56 57
(3a) and (3b) in Fig. 5; analogously there are linear contri-

%gA1b~
butions of AI to A2 -diagrams (4) and (5) in Fig. 11. For
models including T 3 , we have previously identified the linear
58 59 60 61
contributions of T3 to T2 , represented by diagrams (4a)-

~ 62
a--~ 63
W' 64 65
( 4c) in Fig. 5; analogously there are linear contributions of
A3 to A2 -diagrams (33 )-( 35). Consider the constant part

mmm 66 67 68 69
1\
~ (LJ A m x
I
/\--1\

!\A-,\ f\7\--" ~- -1\ ~-7\

70 71 72 73

WTA qp qjA []A 4 5 6

!\ !\ h--x !\ !\ A--x AnA A A--A A

74 75 76 77 78

FIG. 10. The diagrammatic representation of the A, equation in the 10

CCSDT derivative model (truncated). " 9

sistsonlyofdiagrams (1) and (2a)-(2e) in Fig. 5. The cor-

f\/\ !\ !\ ~--x f\ /\ ~nx
12 13 14

responding A2 equation will be restricted to diagrams (1)

and (8)-(12) in Fig. 11; each diagram matches a diagram in FIG. 12. The diagrammatic representation of the A, equation in the
the T2 equation. The A equations are, as already mentioned, CCSDT derivative model (truncated).

J. Chern. Phys., Vol. 90, No.3, 1 February 1989

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Mon, 24 Nov 2014 10:46:46
 1760 Salter, Trucks, and Bartlett: Energy derivatives in many-body methods. I

of T) and linear contributions of T) , T 2 , and T3 to T) , repre- involving the diagonal elementsfr = E7 andf~a = E~-the
sented by diagrams (1), (3a)-(3c), (2a)-(2c) and (4) in derivative SCF eigenvalues. The results of CPHF theory),34
Fig. 4. Analogous diagrams are present in the AI equation are now summarized for canonical SCF orbitals {c,6(X)} ex-
where AI' A z , and A3 take the roles of T), T 2 , and T 3, panded in a primitive AO basis set {f/i(X)}.
respectively. The analogous diagrams are diagrams (1), In this section, all orbitals are assumed to be unper-
(3)-(5), (25)-(27), and (78) in Fig. 10. Finally, consider turbed; i.e., evaluated at X = 0, unless otherwise shown as a
the linear contributions of T2 and T3 to T 3, represented by function of X. Derivative orbitals, evaluated at X = 0, are
diagrams (1a), (1b), and (3a)-(3e) in Fig. 6. (Note: there indicated by the superscript X:
are no linear contributions of TI to T 3 .) Analogous dia-
grams are present in the A3 equation: diagrams (4), (5), and Ip(x» =c,6p(X), Ip) = Ip(x»lx=o,
(8 )-(11) in Fig. 12.
The equivalence of A and Tin the linearized CC models IpX> = a Ip(x» I '
can be used to advantage in the evaluation of the finite-order
ax x=o
MBPT derivatives, since the low-order iterates of A and T l,u(x» = f/ill (X), l,u) = l,u(x» Ix= 0'
are linear and therefore equivalent. Only the fourth-order
contributions from quadruples Q( 4) will require special I,uX) = a l,u(x» I .
consideration. It is apparent that once the equations for AI ax x=o
and A z for the CCSD model have been programmed, the AI
As usual, the labels a,b, ... and iJ, ... are used to represent the
and A2 equations for the CCSDT-l model follows readily.
SCF virtual orbitals and occupied orbitals, respectively,
The same routines developed for inclusion oflinear triples in
whilep,q,r, ... are unrestricted. The labels,u,v,A.,O', ... are used
the CCSDT-1 t-amplitude equations can be employed for the
to represent the primitive AO basis functions. The derivative
A equations. The CCSDT-1 triples routines naturally take
of an SCF orbital, evaluated at X = 0, is given by a linear
advantage of the fact that since T3 amplitudes are linearly
combination of unperturbed SCF orbitals and a term arising
dependent upon the T2 amplitudes, it is not necessary to
from the derivative of the primitive AO basis functions:
store the T3 amplitudes at any time during the CC iterations.
Rather, the backcontributions of T3 to T2 and to TI are
computed as needed. laX) = L Ufo If) + L UIa Ik) + L Cila I,uX) , (19)
For the QCISD model (assuming SCF orbitals for con- I#a k II

venience), the energy derivative is given by the following

diagrams in Fig. 9: (1); (1-3) through (1-7); (1-9) through liX) =L U;;lf) + L UIil k ) + L Cllil,uX). (20)
(1-11); (2-1) through (2-8); and (2-12) through (2-15). I k#i II

See Table Al for the algebraic expressions. 44 The AI equa-

The SCF coefficients for the unperturbed molecular orbital p
tioninFig.1Oistruncatedtodiagrams (3)-(5), (12)-(14),
(26) and (27). The A2 equation in Fig. 11 reduces to dia- are given by CliP' The coefficients q are called the CPHF U:
coefficients. The virtual-occupied block of the CPHF coeffi-
grams (1), (3 )-(12), and (21)-(27). See Tables A2 and A3
cients are determined by solving the linear equation
for the algebraic representation of the diagrams; the terms
are in the same order as the diagrams and correspond one-to- ~(A -I) ai.bj
UXai -- £..i --- ,
Qt (21 )
one. bj Ej - Eb
In order to evaluate the CCSDT derivative energy
expression given in Table AI, the remaining perturbation- where
dependent parts of aP-the contributions arising from the A . = 1+ ...:.<a_b-lI=IY.:....)....:..+....:..<:....::aJ~·III_·b...:....)
ai bj
derivative integrals in thef~ and W~ operators-must be Ea - Ei
made explicit. Recall that the integrals involve the one-elec- and Ep is the unperturbed orbital eigenvalue of orbital p.
tron spin-orbital basis functions {c,6(X)}' Since the orbital Once the virtual-occupied block is determined, the occu-
basis functions are usually linear combinations of some pied-occupied and virtual-virtual blocks are then deter-
primitive set of atomic orbital (AO) basis functions {f/i(X)}, mined as well:
the orbital basis functions are in general dependent upon X in
two ways: the primitive AO functions may be dependent UIi = - 1 [QIi +L Uim «kmllig) + <kgllim) )],
upon X; the coefficients may be dependent upon X. Deriva- (E k - Ei ) g,m

tive integral programs provide the derivatives of one- and (22)

two-electron integrals over the primitive basis functions.
Coupled-perturbed Hartree-Fock (CPHF) theoryl,34 pro- UIa = - 1 [QIa +L U:m«(mllag) + «gllam) )].
(EI-E a ) g,m
vides the derivatives of the SCF orbital coefficients. In the (23)
following sections, we exclusively consider the choice ofSCF
The occupied-virtual CPHF coefficients are determined by
orbitals as the spin-orbita1 basis.
the derivative of the SCF orthonormality condition:
B. Expressions for the choice of SCF reference (24)
For canonical SCF orbitals, all terms in Table Al involv- The elements of the general one-electron perturbation ma-
ing the derivative integrals of f~ vanish, except for those trix Q X are given by

J. Chern. Phys., Vol. 90, No.3, 1 February 1989

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Mon, 24 Nov 2014 10:46:46
 Salter, Trucks, and Bartlett: Energy derivatives in many-body methods. I 1761

TABLE I. The occupied-occupied and virtual-virtual blocks of the

CCSDT density matrix.'

+ "(C
£". p.p
)*C P
vq Au
a(p)"a IIva) I ' (25) Dij + 0~bt~~ - nA.~Zctit:'
= - A.~tf
p.,v X X=O
A,u Dab = A~tt - yt gct~/ + nAg~ct~kb

where
• Summation over all but target labels is implied.
~(€p + €q)
{
when p and q are both
~pq = virtual or both occupied.
€q otherwise
ality. Rather than introduce singularies which must be later
The symbol PAu represents an element of the SCF density eliminated, we follow the development of Handy et al. for
matrix in the AO basis. The matrix h~q consists of the deriva- MBPT(2) gradients 30 by choosing noncanonical SCF orbi-
tive of the matrix elements of the one-electron part of the tals such that
Hamiltonian in the primitive, perturbation-dependent basis
{"'(X)}, but transformed into the unperturbed orbital basis
Ufa = - ~fa If), (28)
{tfo}: UIi = -~SI;lk), (29)

hJ,q = I(Cp.p)*CVq a (p(x)lh(x)lv(x» I . 36

as found in early studies by Moccia and Pulay. I The prob-
lematic denominators of the occupied-occupied and vir-
p.,v aX x=o
tual-virtual CPHF coefficients for canonical orbitals are not
Similarly, the matrix ~ is the derivative of the overlap ma-
present; the virtual-occupied CPHF coefficients are deter-
trix of the perturbation-dependent primitive basis functions,
mined as in Eq. (21). For the choice of noncanonical orbi-
transformed into the unperturbed orbital basis:
tals, however, the off-diagonal elementsf}b andf~ are not
= "'(C )*Cvq a(p(x)lv(x»
SXpq£".p.p aX I . (26) zero. These matrix elements are given by
~v x=o
For a uniform external field perturbation in the z-direc- i}b = Q}b + I U~m «amllbg) + (agllbm», (30)
g,m
tion, the elements of the perturbation matrix Q are simply
Q~q = h ~q = (Plzlq). g,m
For the delta-function perturbation35 located at the point r, Equations (28 )-( 31) and (24) are to be applied to the
we have CCSDT energy derivative expression given in Table AI.
First, gathering together the common derivative two-elec-
tron integrals and the nonzero derivative Fock matrix ele-
The derivative of the SCF orbital eigenvalues, the diag- ments, we define the intermediates rand D for a more com-
onal elements of the Fock matrix, are given by pact expression:

€} = Q}a +I U~m «amllag) + (agllam», AEX = D;j'~ + DaJ}b

g,m
+ r(ij,ab) (ijllab)X + r(ab,cd) (ab Iled)X
g,m + r(kl,ij) (klljij)X + rUa,bi) (jallbi)X
Now we could use the above identities to develop the + r(ai,be) (aillbe)X + rUk,ia) (jk lIia)X,
expression for AEX, except that the occupied-occupied and
virtual-virtual CPHF coefficients [Eqs. (22) and (23)] where summation over all labels is implied. The D and r
possess denominators involving the difference of unper- intermediates for the CCSDT model are defined in Tables I
turbed occupied or virtual eigenvalues. Such denominators and II, respectively. For the CCSDT-1 model, the contribu-
must be eliminated by subsequent factorization of the deriv- tions to the r's which arise from terms [2-27], [2-28], and
ative energy expression; otherwise the expression will not be [2-29] are excluded. In addition, the contributions from
applicable to molecular systems with degeneracies or near terms involving A3 are excluded, with the exception of terms
degeneracies. Since CC theory and its finite-order MBPT [3-1], [3-2], [3-8], and [3-9]. We will show that the D
approximations are invariant under unitary transformations intermediates so defined are the occupied-occupied and vir-
so long as occupied and virtual orbitals are not mixed, we tual-virtual blocks of the density matrix.
may make a particular choice of noncanonical SCF orbitals Now, using the identities from CPHF theory, we ex-
to simplify the derivative expressions without loss of gener- pand the derivative integrals, and obtain the following:
I

AEX = Dij [Q~ + U;m «iellim) + (imllie»] + Dab [Q}b + U;m «aellbm) + (amllbe) >] + [r(ij,ab) - rUi,ab)]
X (U;;(ejllab) - 1/2S~;(mjllab» + [r(ij,ab) - r(ij,ba)]( [ - U}m - S}m] (ijllmb) - 1/2S;a(ijlleb»
+ ... (several terms similar to the last two)
+ AO derivative part. (32)

J. Chem. Phys., Vol. 90, No.3, 1 February 1989

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Mon, 24 Nov 2014 10:46:46
 1762 Salter, Trucks, and Bartlett: Energy derivatives in many-body methods.

TABLE II. The r intermediates. a

r(ij,ab) =!Atb(*) -14A ~~';'t ~1:t ~

+!tijb(*) + to ~~';tktet,!
+!t ftJ - ~A ~~';t k1:n t ~/
+!A ~t~~/ + 1).. '::!t;:t ,,1:t C;::j
+ A ~t~~tJ(*) - AA. :~';t itat i!
+ !A~tij·tZ(*) + to- ~~';t k~! t ~e
+!A ~tk~tj(*) - AA. ~~';t k~~t!J
+A~tktftj - M ~~;t ,,1:1:n t j
+ !A~t1[:rJ + 1.tA ~:;;ft "tel ~ t t
+lA~t~rtZ - -f2,t ~~';t k1:n t ~t J
+ !A ~t 1:;bt j - !"t :~';t ,,1:I;t ~
+!A ~~t~~ttb(*) +:itA ~;dtk~!t;t1
-!A ~~tij·ttr(*) + 1).. ~":ckt ~et:k t j
+ n,A ~~tkrtijd(.) - ¥l ~~';t ~et kjt ~
- !A ~~t k~t{f(.) +!A ~';:t~etC:::ktj
-F~~t~1t'ttJ + i\A de~lt ~et:k t t
+AA~~tijdt,d -!A ~7:t1!tk~t;
-!A ~~t~it1tJ + aA ~ke';t~etjt~t~
-A.~~tk~t1tr + 1,1. ~";:t1!t~t;tZ
+ AA ~; t %rt 1t;
+!A ~;t1tjtktr
nab,cd) = AA tbtijd(*) rUa,bi) =A~tJ(.)
+ !A~btJt1 + A ~:t~J(.)
+ itA. ~~tijkd -A~~tjt:

+ !A.~k;et;kt1 + aA. ~~c t 1{/

r(kl,ij) = AA tb t kr( *) - ¥l~'1cttjtr
+!A Y.,tZt, -~A~~t~tt1
+ 1...4 ~;;t~,!kl
+ aA. ~t:kti
r(ai,bc) = 0~tJ7(.)
-A~tJtk +A~tM
+ 1.1 ~btlj'k - !A%ttJ7
+F;;"t~tk -W:d t 1M
+!A;;"tJkti + A~~t;bt~
-A;;"tljt~ - !Al,t'd t Jk t 1
+~~bt;t%ti -!A %t~t1t'j
-n...t~dtt.~t: + M~kJet;,bt~
- M~dttkdt~ - !A~kJet1:ttl

+ !A;;"dt~tt + MJtJet~'it7
-AA~dbt;tt:1 + All !~dt k1t t
+!2 ~,%tljat~k - !A !Het tft jr
-!A~dbtljdtZt~ + !A !~dt k1t JI ~
- aA. ~'jt:lt Jlt + ~!~dtk~t1tj
-!A ¥Xt)·(.)
+ ~A ~dttj

a Summation over all but the target labels is implied. The symbol (*) indicates the QCISD terms.

J. Chem. Phys., Vol. 90, No.3, 1 February 1989

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Mon, 24 Nov 2014 10:46:46
 Salter, Trucks, and Bartlett: Energy derivatives in many-body methods. I 1763

Contributions arising from the derivatives of the primitive TABLE IV. The I and X intermediates used in the evaluation of CCSDT
and MBPT energy derivatives."
AO basis functions when Eqs. (19) and (20) are applied to
the derivative two-electron integrals are called the "AO de- Iij = - l/2[rUk,ab) - r(kj,ab) j(ik lIab)
rivative part." The AO derivative part is not explicitly
- l/2[rUI,mk) - r(lj,mk) + r(mkJ/) - r(mk,lj») (illlmk)
shown here because the following discussion is concerned -l/2[rUa,bk) +r(kb,aj)](iallbk)
with manipulations of the other terms. Rather, the AO de- - 1/2r(aj,bc) (aillbc)
rivative part is shown in final, computational form in Table - l/2[rUk,la) - r(kj,la») (ik lila)
III. In Table III, we define a perturbation-independent - 1/2r(lkJa) (Ik Ilia)
quantity r(llvpu) by the MO-to-AO transformation to the
r intermediates, as in previous work. 19 The transformation lab = - l/2[r(ij,bc) - r(ij,cb») (ijllac)
of 3N sets of derivative integrals to the MO basis is thereby - l/2[r(be,cd) - r(eb,cd) + r(cd,be) - r(cd,eb) ](aellcd)
avoided. Instead, only the one transformation in the con- - l/2[rUb,ci) + r(ic,bj») (jallci)
struction of the perturbation-independent r(llvpu) is nec- - l/2r (bi,cd) (aillcd)
- 1/2[r(di,bc) - r(di,cb)](dillac)
essary. The AO derivative part of each of the 3N gradients is
- l/2rUk,ib) (jkllia)
computed by the dot product of r(llvpu) with the AO de-
rivative integrals as they are generated for each of the 3N Iai = - l/2[r(kj,ab) - r(kj,ba») (kjllib)
degrees of freedom. -l/2[r(ab,cd) - r(ba,cd) + r(cd,ab) - r(cd,ba»)(ibllcd)
We gather together the common terms of Eq. (32) by -l/2[rUa,bk) + r(kb,aj») (jillbk)
defining computational intermediates I and X such that - 1/2r(aj,bc) (ijllbc)
'-l/2[r(dj,ac) - r(dj,ca)](djllic)
tl.EX= IDijQ~ + IDabQ~b + 2IXaiU~i - l/2rUk,/a) (jk llli)
iJ a,b a,;

+I IijS~ +I IabS~b + 2I IaiS~i + l/2Dkj «kallii) + (killia»

ij a,b a,;
+ l/2Dcb «callbi) + (cillba»
+ AO derivative part. + l/2[r(ij,cb) - rUi,cb») (ajllcb)
The intermediates X and I are summarized in Table IV. Now + l/2[r(i/,kj) - r(Ii,kj) + r(kj,il) - r(kj,li)](alllkj)
recall that the CPHF coefficients are the solutions to a linear + l/2[r(ic,bj) + rUb,ci») (acllbj)
+ 1/2r(di,bc) (dallbc)
equation. Hence, we use the Z-vector method 2o,24 to elimi-
+ l/2[r(ikJb) - r(kiJb») (ak llib)
nate the CPHF coefficients from the expression
+ l/2r Uk,ib) (jk lIab )
2IXaiU~i = 2IXai I(A -I)ai,bj Qt = 2IDbj Qt, 'Summation over all but the target labels is implied. The occupied-occu-
a,i ai bj €j - €b bj pied and virtual-virtual blocks of the CCSDT density matrix D are given
in Table I. The r intermediates are defined in Table II.

TABLE III. The AO derivative part ofthecoupled-c1uster and MBPT ener-

gy derivatives.'

where the summation over the labels a and ihas been carried
AO derivative part = I ,(,u(x)v(x) Ip(x)u(x) ) "r(Il Vpu) ,
out in the term at the far right. The virtual-occupied block of
p,V
p,o
the matrix Dbj is the solution to the perturbation-indepen-
where dent linear equation:
r(Il Vpu) = Ir(ij,ab)(CpaCOb - CpbCoa)C!iC~
iJ (33)
a,b

+ Ir(ab,cd)(CpcC"" - CpdC",,)C::"C~b Finally we have the computational form of the energy deriv-
a.b ative:
c.d

+ rX(ij,kl)(CpkCu/ - Cp/Cuk)C!iC~
'J
tl.EX = IiJ DijQ~ + I a,b
DabQ~b + 2I DaiQ~i
a.;
1<./

+ IrUa,bi)(CpbCOi - CpiCub)C!jC!, + IIijS~ + IIabS~b +2IIaiS~i

i,a ij a,b a,i
j,b
+ AO derivative part. (34)
+ Ir(ai,bc)(CpbCoc - CpcCob)C::"C~i
i,a In the special case of the applied external field, Eq. (34)
b.c reduces to
+ IrUk,ia)(CpiCoa - CpaCOi)C!jC~k
i,a
j,k

, (Ilvlpo}l' denotes the derivative of an ordinary AO two-electron integral, Thus, for the delta-function perturbation [see Eq. (27)] we
evaluated at X = o. have

J. Chern. Phys., Vol. 90, No.3, 1 February 1989

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Mon, 24 Nov 2014 10:46:46
 1764 Salter, Trucks, and Bartlett: Energy derivatives in many-body methods. I

dE x = 2: Dpq¢;;(r)¢;q (r), E. Summary of computational strategy

P.q
The overall strategy for computing MBPT/CC gradi-
and we identify the matrix D as the matrix representation of
ents is:
the correlation correction to the electron density. When add-
( 1) Obtain SX and QX for all 3N degrees of free-
ed to the SCF density matrix, the total response density ma-
dom. [See Eqs. (25) and (26).]
trix is obtained. Since the molecular orbitals are allowed to
(2) (a) If an nth-order MBPT gradient is desired,
"relax" in the presence of the perturbation, the response
solve the unperturbed MBPT problem by low-
density necessarily contains all orbital relaxation effects, as
order iteration of the CC t-amplitude equa-
opposed to a density constructed by taking the expectation
tions. Save the t-amplitude iterates through
value over a given unperturbed wave function. We have ap-
(n - 1)th order in the correlation perturba-
plied such an idea in MBPT(3)28 and MBPT(4)29 applica-
tion.
tions. Such a density matrix for MBPT models has also been
(b) Or if a CC gradient is desired, solve the unper-
suggested by Amos.37 It is well known that iteration of the
turbed CC problem by converging the CC t-
CPHF equation [Eq. (21)] and subsequent substit4tion
amplitude equations. Save the converged t
into the SCF second derivative energy expression yields cer-
amplitUdes.
tain classes of double-perturbation diagrams to all orders. 34 (3) (a) If an MBPT gradient is desired, construct the
Similarly the first derivative of the MBPTICC energy with r intermediates and the occupied-occupied
respect to a one-electron perturbation parameter is actually and virtual-virtual blocks of the density ma-
an infinite sum of diagrams in double perturbation theory. 38 trix D from the low-order iterates of the CC t-
amplitude equations. (See Tables V-VIII.)
(b) Or if a CC gradient is desired, it is necessary to
C. QCISD energy derivatives solve the linear equation for the A-amplitudes
The equations that determine the A I and A2 amplitudes using the converged t amplitudes. Then con-
for the QCISD model were identified in the previous section. struct the r intermediates as given in Table II
The occupied-occupied and virtual-virtual blocks of the and construct the occupied-occupied and vir-
density matrix given in Table I apply to the QCISD model as tual-virtual blocks of the density matrix D as
well, except that the terms involving triples amplitudes are given in Table I.
to be ignored. The appropriate r intermediates for the ( 4) Transform the r intermediate to the AO basis;
QCISD model are given by only those terms noted in Table compute the AO derivative contributions for
II. With r intermediates so constructed, the rest of the theo- each of the 3N degrees offreedom. (See Table
ry applies as already given. That is: Table III applies to the III. )
AO derivative part of the gradient; Table IV applies to the X (5) Using the r intermediates and the occupied-
and I intermediates; Eq. (33) determines the virtual-occu- occupied and virtual-virtual blocks of the re-
pied block of the density matrix; and finally, the derivative laxed density matrix D, construct the interme-
energy is as given in Eq. (34). diates I and X as given in Table IV.
(6 ) Solve Eq. (33) to obtain the virtual-occupied
block of the density matrix.
D. MBPT energy derivatives (7) Evaluate the simple dot products in Eq. (34)
Now we extract the finite-order MBPT derivative from for each of the 3N degrees of freedom; add to
the CCSDT derivative energy expression. Inspection of Eq. the corresponding AO derivative contribu-
(34) reveals that for the derivative of the nth-order correla- tions to obtain the total derivative of the corre-
tion energy, the nth-order MBPT density matrix D and the lation energy for each degree of freedom.
nth-order I intermediates are required. We specify the con- {Note: In Eq. (34) the density D, the pertur-
struction of the occupied-occupied and virtual-virtual bation matrix Q x, and overlap derivative ma-
blocks of the density matrix D and the r intermediates ap- trix S x are in terms of the MO basis. The 3N
propriate for various MBPT models in Tables V through sets of S x and Q x matrices are initially pre-
VIII. The X and I intermediates are to be subsequently con- pared in the AO basis before an MO transfor-
structed as given in Table IV, using the r's and the two mation. [See Eqs. (25) and (26).] It is more
blocks of D so defined. The appropriate nth-order virtual- efficient to avoid 3N sets of transformations by
occupied block of D is obtained by solving Eq. (33) using the
resultant X intermediate. Finally, the appropriate nth-order
AO derivative part is obtained by using the same set of r
intermediates. The MBPT(2), MBPT(3), and MBPT(4) TABLE V. Summary for the MBPT (2) mode1."
density and gradient summaries are, of course, equivalent to
that of our earlier reports.28.29.39.42 Dij = !t ;;(1) t :~(1)
The expressions in the MBPT tables (Tables V-VIII) Dab = - !tijC(l)tijb(l)
are in terms of the low-order iterates of T and the function r(ij,ab) = !tijb(l); all other r's = 0
Q't/ (A, B X C) . We summarize the low-order iterates of A
and T and define related quantities in Table IX. • Summation over all but target labels is implied.

J. Chem. Phys., Vol. 90, No.3, 1 February 1989

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Mon, 24 Nov 2014 10:46:46
 Salter, Trucks, and Bartlett: Energy derivatives in many-body methods. I 1765

TABLE VI. Summary for the MBPT( 3) mode!. TABLE VIII. Summary for the LCCD [or D-MBPT ( 00 ) ] mode!."

Dij = !t;:(l)tJ:~(2) + !~tjt(2)tk~(l) Dij = !tj:( 00 )tiS( (0)

Dab = - !tijc(l)tijb(2) - !~tijC(2)tijb(l) Dab = - !tijc(oo)tijb(oo)
r(ij,ab) = !t;/(2) r(ij,ab) = !tijb( (0)
r(ab,cd) = ~tijb(l)tijd(l) r(ab,cd) = Atijb( 00 )tij'( (0)
r(kl,ij) = 'tijb(l )t kt(1) r(kl, ij) = Atijb( 00 )tJ:t( (0)

rUa,bi) = t~f(l )tfS( I); r(ai,bc) = rUk,ia) = 0 rUa,bi) = t~k( 00 )t~J( (0); r(ai,bc) = rUk,ia) = 0

"Summation over all but target labels is implied. "Summation over all but target labels is implied. The symbol t ijb( 00 ) repre-
sents the converged LCCD amplitudes.

instead transforming the perturbation-inde- in the alternate expectation value coupled-cluster40 (XCC)
pendent matrices D and I to the AO basis. The and unitary41 (UCC) methods, recently proposed. Further-
contributions ofEq. (34) can then be comput- more, since the so-called "quadratic" CI method also corre-
ed equivalently as dot products of matrices in sponds to a restricted CC theory, the gradient equations for
the AO basis.} it are presented as another special case.
(8) Option: Since the MBPTICC relaxed density All CC/MBPT gradient equations can be implemented
is available, one may readily compute one- quite efficiently in what we have called the "response den-
electron response properties or plot the elec-
tronic density of the molecular system.
TABLE IX. The low-order iterates of A and T."

First order:
IV. CONCLUSION
tijb(1) = A Zb (1) = (ijllab )IDijb
A diagrammatic derivation of the detailed spin-orbital
expressions for the full CCSDT analytical first derivatives
has been developed, with all the explicit algebraic expres- Second order:
sions presented. All approximate CC or MBPT methods are = A Zb (2) = t ijb(l) + ~t ijb(2)
t ijb(2)
included as special cases. Equivalent equations have been
~tijb(2) = Utij'(I)(abllcd) + !tJ:t(1) (ijllkl)
implemented for MBPT(2),4 CCSD,21 MBPT(3),27,28
MBPT( 4 ),39 and CCSDT-l.21 The equations are important + P(ab W)t~~( I) (;cllbk) ]IDijb

Third order:
tijb(3) = t:Jb(1) + ~tijb(2) + ~tijb(3;SDn + ~tijb(3;Q)
TABLE VII. Summary for the MBPT ( 4) mode!' a A tb (3) = t ijb( I) + ~t ijb(2) + ~t ijb( 3;SDn + ~ Zb (3;Q)
~tijb(3;Q) = Qijb[I,t(1)Xt(1)]lDijb
Dij = - tj(2)tf(2)
~Zb(3;Q) = {Qijb[t(l),!Xt(l)] + Qijb[t(l),t(I)Xlj}IDijb
+ Yj:(l)tk~(3) + !~tj:(2)tk~(2) + !~tj:(3)tk~(l)
= ~t ijb(3;Q) + Q ijb[ t(1 ),t(1) X t(l) ]
+ !Qjt[t( I ),t( I) Xt( I) ]tk~( I)
t(1)=tijb(l) 1= (ijllab)
Qa,/(A,B XC) =!A ~[Bij'CJ:t - 2(BijcC'f:{ + B7J'CiJ)
Dab = t~(2)t~(2)
- 2(B~:Cl + B'ftbqt) + 4(Bk~C~ + B'f:{CljC)]
- !tijc(1)tijb(3) - !~tijC(2)tijb(2) - !~tijC(3)t:t(1) ~tijb(3;SDn = [!~tijd(2)(abllcd) + !~tkt(2)(ijllkl)
- !QijC[t(1 ),t(1) X t(1)] t :t(1) + P(ab lij)~t ~~ (2) (;cllbk)

+ P(ij)t ~(2) (cjllab ) - P(ab)t J: (2) (ijllkb )

r(ij,ab) = ~tijb(3) + !Qijb[t(1),t(1)Xt(l)] + !P(ab)t~'t'(2) (cdllbk) - !P(ab)t.~~(2) (;cllkl) ]IDijb

t~(2) =A~(2) = [-!tJ;<(1)(b~laj) +!tJk(1)(ibllik)]ID~
r(ab,cd) = Atijb(1)tijd(2) + A~tijb(2)tijd(l)
r(kl,ij) = At:/(1)t kt(2) + AM ijb(2)t %t(1)
tij",:(2) =AZ!c(2) = [-P(albclklij)tta(1)(dkllbc)
rUa,bi) = t ~f(1)t ~J(2) + ~t ~f (2)t ~J( I) + P(clab liljk)tlib( I) (;k Illc) ]IDij",:
r(ai,bc) = tj(2)tJ;«1) + !t:k'f(2)ttj(1)
rUk,ia) = - t~(2)tJk(1) - !tjft(2)t{,b(1) " Summation over all but target labels is implied. P( *) refers to various op-
erators which generate terms with permuted labels. See the notes in Tables
a Summation over all but target labels is implied. A3 and A4 or Ref. 13 for the definitions of such operators.

J. Chem. Phys., Vol. 90, No.3, 1 February 1989

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Mon, 24 Nov 2014 10:46:46
 1766 Salter, Trucks, and Bartlett: Energy derivatives in many-body methods. I

sity" approach. 28 ,29 The response density contains all MO 11M. J. Frisch, R. Krishnan, and J. A. Pople, Chern. Phys. Lett. 75, 66
( 1980); R. Krishnan, M. J. Frisch, and J. A. Pople, J. Chern. Phys. 72,
relaxation terms to all orders and provides the many-body 4244 (1980); M. F. Guest and S. Wilson, Chern. Phys. Lett. 73, 607
analog to the density matrix of variational theories. The re- (1980); R. J. Bartlett,H. Sekino,andG. D. Purvis, ibid. 98, 66 (1983); M.
sponse density is efficiently used to evaluate all first-order Urban, J. Noga, and V. Kello, Theor. Chern. Acta 62, 549 (1983).
one-electron properties and to facilitate the evaluation of IRR. J. Bartlett, in Geometrical Derivatives ofEnergy Surfaces and Geometri-
cal Properties, edited by P. Jorgensen and J. Simons (Reidel, Dordrecht,
second-order properties by determining the response density Holland, 1985).
as a function of an external field, for example. 42 In addition, 19G. B. Fitzgerald, R. J. Harrison, and R. J. Bartlett, J. Chern. Phys. 85,
the response density has been used to study correlation ef- 5143 (1986).
fects on bond formation and related qualitative density con- 2°N. C. Handy and H. F. Schaefer III, J. Chern. Phys. 81, 5031 (1981).
21 A. C. Scheiner, G. E. Scuseria, J. E. Rice, T. J. Lee, and H. F. Schaefer III,
siderations. 43 J. Chern. Phys. 87, 5361 (1987); G. E. Scuseria and H. F. Schaefer III,
Chern. Phys. Lett. 146, 23 (1988).
22H. J. Monkhorst, Int. J. Quantum Chern. Syrnp. 11, 421 (1977).
23D. Mukherjee and P. K. Mukherjee, Chern. Phys. 39, 325 (1979).
24L. Adamowicz, W. D. Laidig, and R. J. Bartlett, Int. J. Quantum Chern.
Syrnp. 18, 245 (1984).
2SG. B. Fitzgerald, R. J. Harrison, W. D. Laidig, and R. J. Bartlett, J. Chern.
IJ. Garrett and I. M. Mills,J. Chern. Phys. 49,1719 (1968); P. Pulay, MoL Phys. 82, 4379 (1985).
Phys. 11, 197 (1969); 18, 473 (1970); K. Thompson and P. Swanstrom, 26G. B. Fitzgerald, R. J. Harrison, W. D. Laidig, and R. J. Bartlett, Chern.
ibid. 26, 735 (1973); H. B. Schlegel, S. Wolfe, and F. Bernardi, J. Chern. Phys. Lett. 117, 433 (1985).
Phys. 63, 3632 (1975); 67, 4181 (1977). 21J. Gauss and D. Cremer, Chern. Phys. Lett. 138, 131 (1987)
2A. Kormonicki, K. Ishida, K. Morokurna, R. Ditchfield, and M. Conrad, 2KE. A. Saiter, G. W. Trucks, and R. J. Bartlett, Chern. Phys. Lett. 141,61
Chern. Phys. Lett. 45, 595 (1977); M. Dupuis and H. F. King, Int. J. (1987).
Quantum Chern. 11, 613 ( 1977 ); J. Chern. Phys. 68, 3998 (1978); S. Kato 29G. Trucks, E. A. Salter, C. Sosa, and R. J. Bartlett, Chern. Phys. Lett. 147,
and K. Morokul1la, Chern. Phys. Lett. 65,19 (1979);J. D. Goddard, N. C. 359 (1988).
Handy, and H. F. Schaefer III, J. Chern. Phys. 71, 1259 (1979); M. Page, 30N. C. Handy, R. D. Amos, J. F. Gaw, J. E. Rice, and E. S. Sirnandiras,
P. Saxe, G. F. Adams, and B. H. Lengsfield III, ibid. 81, 434 (1984). Chern. Phys. Lett. 120,151 (1985).
3B. R. Brooks, W. D. Laidig, P. Saxe, J. D. Goddard, Y. Yamaguchi, and 31J. A. Pople, M. Head-Gordon, and K. Raghavachari, J. Chern. Phys. 87,
H. F. Schaefer III, J. Chern. Phys. 77, 383 (1982). 5968 (1987).
4J. A. Pople, R. Krishnan, H. B. Schlegel, and J. S. Binkley, Int. J. Quan- 32J. Paldus and J. Cizek, Adv. Quantum Chern. 9,105 (1975).
tum Chern. Syrnp. 13, 225 (1979). 33S. Kucharski and R. J. Bartlett, Adv. Quantum Chern. 18, 281 (1986).
sp. J0rgensen and J. Simons, J. Chern. Phys. 79,334 (1983). 3~. C. Caves and M. Karplus, J. Chern. Phys. 50, 3649 (1969).
6F. Coester, NucL Phys. 1,421 (1958); F. Coester and H. Kiirnrnel, NucL 35 A. Szabo and N. S. Ostlund, Modern Quantum Chemistry: Introduction to
Phys. 17,477 (1960). Advanced Electronic Structure Theory (Macmillan, New York, 1982), p.
1J. Cizek, J. Chern. Phys. 45, 4256 (1966); Adv. Chern. Phys. 14, 35 139.
(1969); J. Paldus, J. Cizek, and I. Shavitt, Phys. Rev. A 5, 50 (1972); J. 36R. Moccia, Chern. Phys. Lett. 5, 260 (1970).
Paldus, in New Horizons of Quantum Chemistry, edited by P. O. Lowdin 31R. D. Amos, in Ab initio Methods in Quantum Chemistry Part I, edited by
and B. Pullman (Reidel, Dordecht, Holland, 1983), pp. 31-60; J. Cizek, K. P. Lawley (Wiley, Chichester, Great Britain, 1987).
Adv. Chern. Phys. 14, 35 (1969). 3RE. A. Saiter, H. Sekino, and R. J. Bartlett, J. Chern. Phys. 87, 502 (1987).
KR. J. Bartlett and G. D. Purvis, Int. J. Quantum Chern. 14, 561 (1978); R. 39G. Trucks, E. A. Saiter, J. Watts, and R. J. Bartlett, Chern. Phys. Lett. (in
J. Bartlett and G. D. Purvis, Phys. Scr. 21, 255 (1980). press).
9J. A. Pople, R. Krishnan, H. B. Schlegel, and J. S. Binkley, Int. J. Quan- 4°R. J. Bartlett and J. Noga, Chern. Phys. Lett. 150, 29 (1988). This paper
tum Chern. 14, 545 (1978). erroneously states that XCC( 4) and XCC (5) satisfy the generalized Hell-
lOG. D. Purvis and R. J. Bartlett, J. Chern. Phys. 75, 1284 (1981); G. D. rnan-Feynrnan theorem exactly instead of just to fourth and fifth order.
Purvis and R. J. Bartlett, ibid. 76, 1910 (1982). See Ref. 41 for correction.
"R. J. Bartlett, I. Shavitt, andG. D. Purvis, J. Chern. Phys. 71, 281 (1979). 41R. J. Bartlett and J. Noga, Chern. Phys. Lett. (in press).
12R. J. Bartlett, Annu. Rev. Phys. Chern. 32, 359 (1981); R. J. Bartlett, C. 42G. Trucks, E. A. Saiter, J. Noga, and R. J. Bartlett, Chern. Phys. Lett. 150,
E. Dykstra, and J. Paldus, in Advanced Theories and Computational Ap- 37 (1988).
proachesfor the Electronic Structure ofMolecules, edited by C. E. Dykstra 43c. Sosa, G. Trucks, G. D. Purvis III, and R. J. Bartlett, J. MoL Graph. (in
(Reidel, Dordecht, Holland, 1984). press).
13y' S. Lee, S. Kucharski, and R. J. Bartlett, J. Chern. Phys. 81, 5906 «See AlP document no. PAPS JCPSA-90-1752-43 for 43 pages of detailed
(1984), J. Noga and R. J. Bartlett, ibid. 86, 7041 (1987); 89, 3401 (E) coupled-cluster first and second derivative expressions. Order by PAPS
( 1988). number and journal reference from American Institute of Physics, Phys-
14M. Urban, J. Noga, S. Cole, and R. J. Bartlett, J. Chern. Phys. 83, 4041 ics Auxiliary Publication Service, 335 East 45th Street, New York, NY
(1985). 10017. The price is $1.50 for each microfiche (98 pages) or $5.00 for pho-
ISR. J. Bartlett and D. M. Silver, Phys. Rev. A 10,1927 (1974). tocopies of up to 30 pages, and $0.15 for each additional page over 30
16R. J. Bartlett and I. Shavitt, Chern. Phys. Lett. 50, 190 (1977); 59, 157 pages. Airmail additionaL Make checks payable to the American Institute
( 1978). of Physics.

J. Chern. Phys., Vol. 90, No.3, 1 February 1989

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.22.67.107 On: Mon, 24 Nov 2014 10:46:46