This page intentionally left blank

**MANY-BODY METHODS IN C H E M I S T RY A N D P H Y S I C S
**

Written by two leading experts in the ﬁeld, this book explores the manybody methods that have become the dominant approach in determining molecular structure, properties, and interactions. With a tight focus on the highly popular many-body perturbation theory (MBPT) and coupledcluster (CC) methods, the authors present a simple, clear, uniﬁed approach to describe the mathematical tools and diagrammatic techniques employed. Using this book the reader will be able to understand, derive, and conﬁdently implement the relevant algebraic equations for current and even new CC methods. Hundreds of diagrams throughout the book enhance reader understanding through visualization of computational procedures, and the extensive referencing and detailed index allow further exploration of this evolving area. This book provides a comprehensive treatment of the subject for graduates and researchers within quantum chemistry, chemical physics and nuclear, atomic, molecular, and solid-state physics. I s a i a h S h a v i t t, Emeritus Professor of Ohio State University and Adjunct Professor of Chemistry at the University of Illinois at UrbanaChampaign, developed eﬃcient methods for multireference conﬁgurationinteraction calculations, including the graphical unitary-group approach (GUGA), and perturbation-theory extensions of such treatments. He is a member of the International Academy of Quantum Molecular Science and a Fellow of the American Physical Society, and was awarded the Morley Medal of the American Chemical Society (2000). R o d n e y J . B a r t l e t t, Graduate Research Professor at the Quantum Theory Project, University of Florida, pioneered the development of CC theory in quantum chemistry to oﬀer highly accurate solutions of the Schr¨ odinger equation for molecular structure and spectra. He is a member of the International Academy of Quantum Molecular Sciences and a Fellow of the American Physical Society (1986). He was awarded a Guggenheim Fellowship (1986), the ACS Award in Theoretical Chemistry (2007) and the Schr¨ odinger Medal of WATOC (2008).

Cambridge Molecular Science As we enter the twenty-ﬁrst century, chemistry has positioned itself as the central science. Its subject matter, atoms and the bonds between them, is now central to so many of the life sciences on the one hand, as biological chemistry brings the subject to the atomic level, and to condensed matter and molecular physics on the other. Developments in quantum chemistry and in statistical mechanics have also created a fruitful overlap with mathematics and theoretical physics. Consequently, boundaries between chemistry and other traditional sciences are fading and the term molecular science now describes this vibrant area of research. Molecular science has made giant strides in recent years. Bolstered both by instrumental and theoretical developments, it covers the temporal scale down to femtoseconds, a time scale suﬃcient to deﬁne atomic dynamics with precision, and the spatial scale down to a small fraction of an ˚ Angstrom. This has led to a very sophisticated level of understanding of the properties of small-molecule systems, but there has also been a remarkable series of developments in more complex systems. These include: protein engineering; surfaces and interfaces; polymers colloids; and biophysical chemistry. This series provides a vehicle for the publication of advanced textbooks and monographs introducing and reviewing these exciting developments. Series editors Professor Richard Saykally University of California, Berkeley Professor Ahmed Zewail California Institute of Technology Professor David King University of Cambridge

**MANY-BODY METHODS I N C H E M I S T RY AND PHYSICS
**

MBPT and Coupled-Cluster Theory

I S A I A H S H AV I T T

Professor Emeritus The Ohio State University and Adjunct Professor of Chemistry University of Illinois at Urbana-Champaign

and R O D N E Y J . B A RT L E T T

Graduate Research Professor of Chemistry and Physics University of Florida

CAMBRIDGE UNIVERSITY PRESS

Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo, Delhi, Dubai, Tokyo Cambridge University Press The Edinburgh Building, Cambridge CB2 8RU, UK Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521818322 © I. Shavitt and R. J. Bartlett 2009 This publication is in copyright. Subject to statutory exception and to the provision of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published in print format 2009

ISBN-13 ISBN-13

978-0-511-59643-8 978-0-521-81832-2

eBook (NetLibrary) Hardback

Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

To Vera and Beverly

Contents

Preface 1 Introduction 1.1 Scope 1.2 Conventions and notation 1.3 The independent-particle approximation 1.4 Electron correlation 1.5 Conﬁguration interaction 1.6 Motivation 1.7 Extensivity 1.8 Disconnected clusters and extensivity 2 Formal perturbation theory 2.1 Background 2.2 Classical derivation of Rayleigh–Schr¨ odinger perturbation theory 2.3 Projection operators 2.4 General derivation of formal time-independent perturbation theories 2.5 Similarity transformation derivation of the formal perturbation equations and quasidegenerate PT 2.6 Other approaches 3 Second quantization 3.1 Background 3.2 Creation and annihilation operators 3.3 Normal products and Wick’s theorem 3.4 Particle–hole formulation 3.5 Partitioning of the Hamiltonian vii

page xi 1 1 2 3 7 9 10 11 15 18 18 18 27 29 46 53 54 54 55 67 71 75

viii

Contents

4

5

6

7

8

9

Normal-product form of the quantum-mechanical operators 3.7 Generalized time-independent Wick’s theorem 3.8 Evaluation of matrix elements Diagrammatic notation 4.1 Time ordering 4.2 Slater determinants 4.3 One-particle operators 4.4 Two-particle operators Diagrammatic expansions for perturbation theory 5.1 Resolvent operator and denominators 5.2 First-order energy 5.3 Second-order energy 5.4 Third-order energy 5.5 Conjugate diagrams 5.6 Wave-function diagrams 5.7 Fourth-order energy 5.8 Linked-diagram theorem 5.9 Numerical example 5.10 Unlinked diagrams and extensivity Proof of the linked-diagram theorem 6.1 The factorization theorem 6.2 The linked-diagram theorem Computational aspects of MBPT 7.1 Techniques of diagram summation 7.2 Factorization of fourth-order quadruple-excitation diagrams 7.3 Spin summations Open-shell and quasidegenerate perturbation theory 8.1 Formal quasidegenerate perturbation theory (QDPT) 8.2 The Fermi vacuum and the model states 8.3 Normal-product form of the generalized Bloch equations 8.4 Diagrammatic notation for QDPT 8.5 Schematic representation of the generalized Bloch equation 8.6 Level-shift and wave-operator diagrams 8.7 Incomplete model space Foundations of coupled-cluster theory 9.1 Coupled-cluster theory for noninteracting He atoms 9.2 The coupled-cluster wave function

3.6

80 85 86 90 90 91 92 111 130 130 131 131 132 134 135 138 152 153 156 165 165 172 177 177 180 182 185 185 192 194 195 198 203 227 251 251 254

1 Expectation value for a CC wave function 11.4 258 272 279 292 292 295 297 299 308 321 328 340 347 347 352 361 366 367 376 381 385 396 406 406 411 415 418 422 425 427 429 431 431 .9 The CC correlation-energy derivative Additional aspects of coupled-cluster theory 12.5 Monitoring multiplicities in open-shell coupled-cluster calculations 12. doubles and triples (CCSDT) equations 10.8 The nature of the solutions of CC equations The equation-of-motion coupled-cluster method for excited. ionized and electron-attached states 13.1 Introduction 9. triples and quadruples (CCSDTQ) equations 10.7 Coupled-cluster eﬀective-Hamiltonian diagrams 10.4 The CC energy functional 11.7 Response treatment of the density matrices 11.5 Diagrammatic derivation of the CCD equations Systematic derivation of the coupled-cluster equations 10.1 The connected form of the CC equations 10.6 Coupled-cluster singles.1 Spin summations and computational considerations 12.4 Brueckner orbitals and alternative treatments of T 12.7 Noniterative approximations based on the CC energy functional 12.2 Coupled-cluster theory with an arbitrary singledeterminant reference function 12.6 Eﬀective-Hamiltonian form of the Λ equations 11.8 The perturbed reference function 11.3 The response treatment of properties 11.8 Results of various CC methods compared with full CI Calculation of properties in coupled-cluster theory 11.3 Systematic generation of CC diagrams 10.6 The A and B response matrices from the viewpoint of CCS 12.2 The general form of CC diagrams 10.3 Generalized many-body perturbation theory ˆ1 12. doubles.2 Reduced density matrices 11.5 Coupled-cluster singles.3 9.4 The coupled-cluster singles and doubles (CCSD) equations 10.5 The Λ equations 11.Contents ix 10 11 12 13 The coupled-cluster doubles (CCD) equations Exponential Ansatz and the linked-diagram theorem of MBPT 9.

6 EOM-CC treatment of frequency-dependent properties 454 14 Multireference coupled-cluster methods 462 14.2 Hilbert-space state-universal MRCC 465 14.5 Intermediate-Hamiltonian Fock-space MRCC 490 References 496 Author index 521 Subject index 524 .3 Hilbert-space state-speciﬁc MRCC 471 14.x Contents 13.4 Fock-space valence-universal MRCC 475 14.5 EOM-CC treatment of higher-order properties 449 13.2 The EOM-CC Ansatz 432 13.1 Introduction 462 14.4 EOM-CC treatment of ionization and electron attachment 445 13.3 Diagrammatic treatment of the EE-EOM-CC equations 437 13.

A. Hence a correct description of electron correlation has been the focal point of atomic. maintaining extensivity – a critical rationale for all such methods – requires many-body methods that employ quite diﬀerent mathematical tools for their development than those that have been customary in quantum chemistry. diagrammatic xi . This property distinguishes many-body methods from the conﬁguration-interaction (CI) tools that have commonly been used for many years. to know all there is to know about the properties and interactions of molecules. These approaches have become the methods of choice in quantum chemistry. For many readers. primarily the Schr¨ odinger equation. This situation has now changed. in principle. molecular and solid state theory for over 50 years. oﬀering a uniﬁed. In the last two decades the most prominent methods for providing accurate quantum chemical wave functions and using them to describe molecular structure and spectra are many-body perturbation theory (MBPT) and its coupled-cluster (CC) generalizations. transparent and precise approach to the derivation and implementation of the relevant algebraic equations. this is manifested as a correlation of their motions. diagrammatic techniques are found to be extremely powerful. Achieving this goal. a task that was not possible for most of the past half-century. a property known as extensivity (or size-extensivity ). Accurate quantitative answers to this question would allow us. Today it is well known that the many-body instantaneous interactions of the electrons in molecules tend to keep electrons apart. however. however. However. Mulliken over a half-century ago.Preface “What are the electrons really doing in molecules?” This famous question was posed by R. requires a very accurate solution of the quantum-mechanical equations. primarily due to the development of numerically accurate many-body methods and the emergence of powerful supercomputers. owing to their accuracy and their correct scaling with the number of electrons. In particular.

the monograph by Lindgren and Morrison (1986). making it diﬃcult to understand with conﬁdence the detailed one-to-one correspondence between the diagrams and the various terms of the operable algebraic equations. solid-state and nuclear physics. . In order to address this situation. This book is directed at graduate students in quantum chemistry. quasidegenerate perturbation theory (QDPT) and multireference CC (MRCC). completely unambiguous procedures to derive all the relevant algebraic equations diagrammatically. this book presents a uniﬁed. Monkhorst and Freeman (1992). The ambiguity associated with some diagrammatic approaches is completely eliminated. With a modest investment of time and eﬀort.xii Preface methods have seemed to be used arbitrarily. among others. physical chemistry and atomic. an extensive bibliography and a detailed index are included. It introduces direct. molecular. An interesting historical account of the development of coupled-cluster theory was provided by Paldus (2005). ionized and electron attached states as well as the analytical gradient theory for determining structure. easily applied and uniﬁed approach. excited states and properties derived from expectation values and from response methods. detailed account of the highly popular MBPT and CC quantum mechanical methods. For this purpose. and the book focusing on diagrammatic many-body methods by Harris. including detailed treatments of self-consistent ﬁeld theory and conﬁguration interaction. the recent developments in multireference approaches. can be found in parts of the book by Szabo and Ostlund (1982). Additional useful sources include. in one simple. Selected numerical illustrations are also presented to assess the performance of the various approximations to MBPT and CC. Furthermore. this book will teach anyone to understand and conﬁdently derive the relevant algebraic equations for current CC methods and even the new CC methods that are being introduced regularly. It can serve as a textbook for a two-semester course on many-body methods for electronic structure and as a useful resource for university faculty and professional scientists. are also presented. including the equation-of-motion CC (EOM-CC) method for excited. Useful introductory material for the book. which emphasizes atomic structure and includes the treatment of angular momentum and spin coupling. chemical physics. whose unpublished (but widely distributed) Nijmegen lectures introduced many researchers to this methodology. Finally. All these equations are readily developed from the same simple diagrammatic arguments used throughout the book. new theory has had to be developed. This book also addresses the theory for each of these topics. vibrational spectra and density matrices. in order for a quantumchemical approach to be able to describe molecular structure.

Preface xiii Many colleagues. Josef Paldus. Ajith Perera. In particular we would like to thank the following: Erik Deumens. Steven Parker. students and others have helped us in various ways during the writing of this book. Tom Hughes. Timothy van Zandt. Monika Musial. Andrew Taube and John Watts. Denis Girou and Sebastian Rahtz. using the package pstricks 97. and editors. Joshua McClellan. We are grateful to its author. John Stanton. for providing such a versatile graphics facility. J¨ urgen Gauss. . Marcel Nooijen. The graphics in this book were generated in LaTeX. Peter Szalay.

.

b. many-electron) methods for electron correlation. Applications to the electronic structure of atoms and molecules began with the work of Kelly (1963. indeﬁnite or variable numbers of particles.b) and applied to many-body systems by Hugenholtz (1957) and by Goldstone (1957). 1964a. Other important contributions to the methodology. in scattering problems and in the study of inﬁnite systems because it easily handles problems involving inﬁnite.1 Introduction 1. 1968). and coupled-cluster (CC) theory. and were formalized by Goldstone (1957).b). are due to Dyson (1949a. ﬁrst in ﬁeld theory and then in the theory of nuclear structure. These include Rayleigh–Schr¨ odinger perturbation theory (RSPT). 1958a. The diagrammatic approach was introduced into ﬁeld theory by Feynman (1949a. Further extensions address properties other than the energy. Many-body perturbation theory and its linked-diagram formalism were ﬁrst introduced by Brueckner and Levinson (1955) and by Brueckner (1955). and also excited states and multireference CC and MBPT methods. This formalism is particularly useful in ﬁeld theory. 1 . or MBPT ).b).1 Scope The book focuses primarily on many-body (or better. particularly in its diagrammatic representation (referred to as many-body perturbation theory. The many-body algebraic and diagrammatic methods used in electronic structure theory have their origin in quantum ﬁeld theory and in the study of nuclear matter and nuclear structure. and molecular applications using ﬁnite analytical basis sets appeared in the work of Bartlett and Silver (1974a.b) and Frantz and Mills (1960). their relationship to conﬁguration interaction (CI) is included. The second-quantization formalism was ﬁrst introduced in a treatment of quantized ﬁelds by Dirac (1927) and was extended to fermion systems by Jordan and Klein (1927) and by Jordan and Wigner (1928). Wick (1950). Hubbard (1957.

and the equation-ofmotion (EOM) coupled-cluster method for excited-state calculations is described in Chapter 13. 1969) and C´ ıˇ zek and Paldus (1971). Coupled-cluster theory is discussed in Chapters 9 and 10. The application of the many-body and diagrammatic techniques to perturbation theory is described in Chapter 5. We discuss the limitations of these methods and the need for the perturbation-theoretical and many-body methods that form the subject of the rest of this book. A historical account of its origins and development was given by Paldus (2005). Finally. Additional references to the development and extensions of the many-body methods are given in the relevant chapters. The rest of this chapter provides some background material. The many-body technique of second quantization is introduced in Chapter 3. We also provide a preliminary introduction to the cluster ideas that form the basis of coupled-cluster theory. The coupled-cluster method also has origins in nuclear structure theory. with the seminal papers of Coester (1958) and Coester and K¨ ummel (1960). The calculation of properties in the coupledcluster method is described in Chapter 11. and this is followed by proof of the crucial linked-diagram theorem in Chapter 6 and a discussion of some practical aspects of many-body perturbation-theory calculations in Chapter 7. A detailed exposition of formal perturbation theory is given in Chapter 2. including a brief discussion of the independent-particle model and the conﬁgurationinteraction method. again including several forms of the derivations in order to provide better understanding. ˇ ıˇ It was introduced to electronic structure theory and formalized by C´ zek ˇ (1966. Readers in need of a more extensive introduction are referred to the excellent book by Szabo and Ostlund (1982). Open-shell and quasidegenerate perturbation theory is presented in Chapter 8.2 Introduction More complete accounts of the history of these methods have been given by Lindgren and Morrison (1986) and by Lindgren (1998). multireference coupled-cluster methods are presented in Chapter 14.2 Conventions and notation Throughout this book we use atomic units. setting m = e = = 1 where m and −e are the mass and charge of the electron and = h/2π is Planck’s . and the diagrammatic representation is described in Chapter 4. Several additional aspects of coupled-cluster theory are discussed in Chapter 12. A number of diﬀerent derivations and approaches are included in this exposition in order to provide a broad foundation for the terminology and techniques employed in this ﬁeld. 1.

and upper-case letters (A. etc. . . B.) are used for many-particle entities. using a mixture of English. Greek and Latin roots. φ. Operators are designated by a caret ˆ. . a CI calculation that includes all single. as listed in Table 1. P . For more complete discussions and detailed derivations the reader is referred to other sources. these constants are omitted from the expressions in this book but their implied presence is needed for proper dimensionality. lower-case letters (a. . .1 lists some alternative excitation-level names that have been used. . With a few exceptions. . respectively.and upper-case letters. . etc. .1. Ψ.). . The wave function . ψ. such as Szabo and Ostlund (1982) or standard textbooks. . ˆ i. F etc. Vectors and matrices are represented by boldface lower.) or by an Greek upper-case letter (Λ. Ω.). b. H ˆ .and two-particle entities. etc.3 The independent-particle approximation In this section we brieﬂy summarize several aspects of the procedures used to obtain starting approximations for correlated molecular electronic structure calculations.1. . For example. . in view of the need to provide a unique initial letter for each level. over a roman letter (ˆ a. by a script upper-case letter (H. Terminology for excitation levels Level 1 2 3 4 5 6 Symbol S D T Q P H Name singles doubles triples quadruples pentuples hextuples Alternative mono-excited bi-excited tri-excited tetra-excited penta-excited hexa-excited constant. Most electronic structure calculations begin with a relatively simple approximation based on the independent-particle model. As is customary in quantum chemistry.) are used for one. The fourth column in Table 1. double and triple excitations is described as CISDT.3 The independent-particle approximation 3 Table 1. . . ﬁrst in conﬁguration interaction (CI) and then in coupled-cluster (CC) theory. 1. Φ. etc. The acronyms used to specify excitation-level combinations included in the diﬀerent computational models have evolved. .1.

. Therefore. i. . the SD wave function (1. . ψN (N ) (1. ψ1 (1) ψ2 (1) · · · ψ1 (2) ψ2 (2) · · · 1 Φ= √ .4 Introduction for such a model is a single Slater determinant (SD). is called ˆ f in terms of these spinorbitals is diagonal. (1. The HF solution is thus the limiting result (the HF limit ) of the self-consistent ﬁeld procedure as the basis set approaches completeness.e. . Iterative procedures are reˆ and quired to obtain consistency between the spinorbitals used to deﬁne f the spinorbitals obtained as its eigenfunctions.. a function of the space and spin coordinates of the µth electron (typically a product of a spatial orbital and a spin function). 2. and the terms Hartree–Fock and self-consistent ﬁeld are used interchangeably. . The particular representation of the wave function in which the spinorbitals are solutions ˆ (so that the matrix representation of of (1. ψ1 (N ) ψ2 (N ) · · · = A ψ 1 ψ 2 . and the self-consistent ﬁeld form (also known as Hartree–Fock–Roothaan or matrix Hartree–Fock ). The most commonly used independentparticle model is the Hartree–Fock (HF) or self-consistent ﬁeld (SCF) wave function. N } among themselves. We shall follow this practice in this book. which achieves the absolute minimum of the energy of an SD wave function (1.. .2). The minimization can be achieved by solving a set of coupled one-electron eigenvalue equations for the spinorbitals. the distinction has unfortunately been lost. In current usage. however.1) with respect to any variation of the spinorbitals (subject only to appropriate restrictions in the restricted HF case) and usually require numerical (ﬁnite diﬀerence) methods of solution as employed by Hartree and others for atomic wave functions. i = 1. ψN (1) ψN (2) . the corresponding spinorbitals (including † It was common to distinguish between the original type of Hartree–Fock solution.2) f ˆ depends on all the spinorbitals (this depenin which the Fock operator f dence is given explicitly later in this section). . .2) to a matrix eigenvalue equation for the eigenvectors of expansion coeﬃcients. . ψN .† in which the spinorbitals are varied to minimize the energy expectation value of the single-determinant wave function. ψi |f the canonical HF wave function . . This second approach converts the operator eigenvalue equation (1. . both commonly referring to the basis-set expansion approach. . in which the spinorbitals are expanded in a basis set and the lowest energy solution within the space generated by that basis set is sought.1) is invariant under unitary transformations of the occupied spinorbitals {ψi . . . are eigenfunctions of f ˆ|ψj = εi δij ). .1) where ψi (µ) is a spinorbital. ˆψi = εi ψi . any unitary transformation of the occupied spinorbitals provides an alternative representation of the same SD wave function. and A is the antisymmetrizer. Because a determinant is invariant to unitary transformations of its columns or rows. N! .

in which the individual spinorbitals are localized to the regions of individual atoms or bonds. In the usual form of the unrestricted Hartree–Fock (UHF) model there are no restrictions other than that each spinorbital is a product of a spatial orbital and a spin-up (α) or spin-down (β ) spin function. restrictions can be placed on the symmetry properties of the spatial orbitals requiring them to belong to irreducible representations (irreps ) of the point group of the molecule. The RHF model is most commonly used for closed-shell molecules near their equilibrium geometry. in which one or more spinorbitals are unpaired. which are the eigenfunctions of ˆ. When applied to open-shell cases. Such localized forms of the solution often oﬀer advantages of simpler interpretation and provide a basis for more compact descriptions of correlated wave functions. The orbital energies εi . leading to the restricted Hartree–Fock (RHF) model in which each spatial orbital can accommodate at most two electrons. All other representations of the HF solution are called noncanonical ˆ with two interand produce a block-diagonal matrix representation of f nally non-diagonal blocks representing the occupied and virtual spinorbital spaces. respectively.3 The independent-particle approximation 5 the unoccupied or virtual spinorbitals obtained as additional eigenfunctions ˆ that are not used in the SD wave function) are the canonical spinorof f bitals .† Independently of whether spin restrictions are used. if any. The degree of freedom provided by the invariance of the HF wave function under unitary transformations of the occupied spinorbitals (and. Such restrictions can often cause diﬃculties in the descriptions of potential-energy surfaces. when symmetry-restricted solutions at highsymmetry points may be higher in energy than symmetry-unrestricted solutions at the same points (a phenomenon referred to as symmetry breaking ) † Except where stated explicitly otherwise. these are deﬁned by the restrictions. of the unoccupied spinorbitals) is sometimes used to transform the spinorbitals to a localized form. that are placed on the spinorbitals {ψi }. The most common restriction constrains pairs of spinorbitals to share the same spatial orbital. . are invariant under unitary transformations but are associated one-to-one f with the canonical spinorbitals only. For closed-shell molecules near their equilibrium geometry the UHF and RHF solutions are generally equivalent and produce the same set of doubly occupied spatial orbitals. the treatment in this book is in terms of unrestricted spinorbitals. Several variants of the Hartree–Fock approach are in common use. separately.1. it is often referred to as restricted open-shell Hartree–Fock (ROHF). and the terms orbitals and Hartree–Fock (or SCF) generally refer to spinorbitals and UHF. This form is often used for open-shell states and sometimes for the description of bond dissociation.

There are however cases in which HF does not provide an adequate zero-order function. because the energy diﬀerences of interest in chemical and spectroscopic processes are a fraction of one percent of the total electronic energy and because the accuracy of the HF model tends to vary considerably between diﬀerent structures and diﬀerent electronic states. It is instructive to consider an alternative derivation of the Hartree–Fock model that provides physical insight into the reasons for its success and introduces some important concepts. The averaged interaction energy of an electron in spinorbital ψi . In most cases.2) can be obtained by a physically motivated argument. this model does not usually provide adequately accurate solutions by itself.1). In most cases. Nevertheless. when the total electron distribution is described by the single-SD wave function (1. the same Fock equation (1. also referred to as multiconﬁgurational SCF (MCSCF). describing the motion of each electron in the time-averaged ﬁeld of the other electrons. but the use of such multiconﬁgurational zero-order functions requires multireference methods in the post-HF stage and introduces additional complications for the many-body methods that are the principal topic of this book. The multiconﬁgurational Hartree–Fock (MCHF) model. The diﬃculty in the solution of the electronic Schr¨ odinger equation is principally due to the interelectron repulsion terms 1/rµν in the Hamiltonian. . though. often accounting for more than 99% of the total electronic energy and 95% of the wave function. Those terms couple the motions of the diﬀerent electrons and prevent separation of the equation into individual one-electron equations. We shall do this in terms of the unrestricted model. because of its generality and its relatively simple notation. Instead of invoking the variational principle and minimizing the energy of a single-determinant wave function with respect to the spinorbitals. these are due primarily to near degeneracies between several Slater-determinantal contributions to the wave function. In such cases a multideterminantal (“multiconﬁgurational”) function can provide a better zero-order wave function.6 Introduction and thus do not connect continuously with the solutions for distorted (lowersymmetry) structures. can be particularly eﬀective in providing good zero-order solutions in such cases. it provides a satisfactory zero-order solution that can then be used as the starting point for the post-Hartree–Fock methods discussed in this book. a Hartree–Fock solution provides an excellent initial approximation for the electronic wave function and energy of a molecular system. It is therefore natural to seek an approximate solution in which the instantaneous interelectron interaction terms are replaced by averaged interactions.

It also provides the basis for deﬁning the concept of electron correlation.2) and the Fock operator is given by N ˆ= h ˆ+ f i=1 ˆi − K ˆ i) . as discussed in the next section. and thus the removal of the restriction on the summation.3) when i = j . The Coulomb and exchange operators are deﬁned by ˆi (1)φ(1) = ψi (2) 1 ψi (2) φ(1) . 1. (J (1.4) the integration being over the coordinates of electron 2 only. the equation for each electron takes the form (1. K r12 2 (1. Because of the canwhere h cellation of the Coulomb and exchange terms in (1. J r12 2 1 ˆ i (1)φ(1) = ψi (2) φ(2) ψi (1) . the Fock operators for all the spinorbitals are equal and we have only one eigenvalue equation for the spinorbitals.4 Electron correlation The purpose of all many-body methods is to describe electron correlation. The iterative and coupled nature of the equations is due to the dependence of the Coulomb and exchange operators on all the occupied orbitals. ψi (1)|J where the restriction on the summation in the ﬁrst line can be ignored because of the cancellation between the ﬁrst (Coulomb or direct ) term and the second (exchange ) term when i = j .3) = j =1 ˆj − K ˆ j |ψi (1) . The average-interaction approach leads to exactly the same equations as the energy-minimization approach and serves to provide a physical rationale for the Hartree–Fock model. the Schr¨ odinger equation becomes separable.5) ˆ is the one-electron operator in the Hamiltonian. with the diﬀerent spinorbitals obtained as diﬀerent eigenfunctions of that operator. deﬁned as representing the diﬀerence between the Hartree–Fock description .4 Electron correlation 7 is obtained as N ˆ|ψi = ψi |u j (j =i) N ψi (1)ψj (2) 1 ψi (1)ψj (2) − ψj (1)ψi (2) r12 (1.1. With the replacement of the instantaneous electron repulsion by the average form.

this dynamic eﬀect is not the only type of error in the Hartree–Fock model. i. Kutzelnigg 2003). basis set).) We are interested in the correlation correction to the energy.7) which determines the electron density and all other properties of molecules. Unfortunately. complete. the understanding provided by these concepts is important in designing methods for obtaining satisfactory solutions of the Schr¨ odinger equation. near degeneracies between single-conﬁguration descriptions cause the single-conﬁguration Hartree–Fock model to be deﬁcient even as a zero-order approximation for the wave function and energy. reﬂecting the instantaneous correlation in electron motions due to their mutual repulsion. the multireference extension of the perturbation and coupled- .e. ∆Ecorr = Eexact − EHF (1. especially for excited states and other open-shell states and even for ground-state closed-shell molecules when bonds are stretched to near breaking (if the RHF model is used as the reference). reﬂecting the eﬀect of near degeneracies and other substantial inadequacies of the single-conﬁguration model. In many cases. called the correlation energy . The naming of the correlation correction reﬂects the fact that the Hartree– Fock model describes the motion of the electrons in the average ﬁeld of the other electrons. in an inﬁnite. and also in the correction to the wave function..8 Introduction of the electronic wave function and the exact solution of the nonrelativistic time-independent Schr¨ odinger equation (L¨ owdin 1959. and nondynamic correlation . Nondynamic correlation is handled most eﬃciently by using a multiconﬁgurational zero-order description such as multiconﬁgurational SCF (MCSCF). Ψexact = ΦHF + χcorr . While it is diﬃcult to provide a quantitative separation between these two components of the correlation error. However. (1. The two components of the correlation error respond best to two diﬀerent types of treatment. Dynamical correlation is handled eﬃciently by the perturbation methods and coupled cluster approaches described in the rest of this book. neglecting the instantaneous correlation in the motions of the electrons due to their mutual repulsion.6) (where EHF implies the exact solution of the Hartree–Fock problem. (Note that this deﬁnition is not necessarily unique. Therefore we distinguish between the two components of the correlation eﬀect: dynamic correlation. because of the diﬀerent types of Hartree–Fock model that can be used as the reference point. unlike the situation in the conﬁguration interaction model.

CI with single (S) and double (D) excitations. Nevertheless. which states that the Hamiltonian matrix element between the SCF function and a single excitation vanishes.5 Conﬁguration interaction 9 cluster models is quite diﬃcult. This theorem holds for both canonical and noncanonical forms of the SCF function and is a consequence of the block-diagonal nature of the Fock operˆ (Section 1.. single-reference coupledcluster theory. f ij conﬁguration etc. As a ator f i . and general methods for this type of approach are still not widely available. Φab is a double-excitation eigenfunction of the Fock operator. This number k -fold excited conﬁgurations is O(mk nk ) (as long as k grows very rapidly with k . = 0 for higher than double excitations). this represents a reasonable approximation in most cases. so that a complete solution is impractical and the CI expansion needs to be truncated.. If we start with the self-consistent ﬁeld (SCF) wave function and orbitals (the Hartree–Fock solution limited to the space spanned by a given basis set). producing CISD.5 Conﬁguration interaction The simplest approach to treating electron correlation is by the conﬁgurationinteraction (CI) method.a Cia Φa i + i<j. (1.1. Usually the truncation is made after the double-excitation level. 1.8) based on an SCF function (unrestricted or closed-shell restricted) as the initial term is quite small. a<b ab ab Cij Φij + · · · (up to N excitations).9) ΦSCF |H i = 0.8) where Φa i is a singly excited conﬁguration in which an occupied orbital φi of the SCF wave function has been replaced by a virtual orbital φa (an orbital not occupied in the SCF function.. when carried to a high enough level. This is due to the Brillouin theorem .. The contribution of single excitations to a CI expansion (1. has proved to be capable of overcoming the nondynamic correlation problem to a considerable extent.e. which can also be chosen to be an ˆφa = εa φa ). Because ΦSCF is (usually) a reasonable starting ˆ has no more than approximation to Ψ and because the Hamiltonian H ab. ˆ two-electron operators (so that ΦSCF |H |Φij. i. If m is the number of SCF occupied orbitals and n is the number of virtual orbitals (m + n equals the number of basis functions) then the number of m). ˆ |Φa (1.3) given by fia = ΦSCF |H ˆ |Φa = 0 (see Chapter 3). we can write the CI expansion of an N -electron wave function in the (unnormalized) form Ψ = ΦSCF + i.

to a much smaller extent. 1. in fact much smaller than even the small contributions in the SCF case. they are obtained by transforming the orbitals so as to diagonalize the one-particle density matrix of the correlated wave function. The contribution of single excitations to the CI expansion (as well as to perturbation expansions and coupled-cluster wave functions) can be eliminated completely by a suitable transformation of the orbitals. The exact solution.6 Motivation The motivation for studying perturbation theory (PT) and other many-body (MB) techniques comes from the following principal sources. because it is the interaction of the single excitations with the doubles that is responsible for almost all of the contribution of the single excitations to the expansion. for ˆ matrix is not block diagonal.10 Introduction consequence of the Brillouin theorem. most PT formulations provide properly extensive descriptions. Unlike truncated CI. These arguments do not apply to the restricted open-shell SCF (ROHF) case. For two-electron systems. the Brillouin theorem does not which the f hold. Such a transformation involves some mixing of occupied and virtual orbitals and therefore the initial term in the transformed expansion is no longer exactly equivalent to the SCF function. within the given basis set. This is called full CI and is invariant under any linear transformation of the orbitals. Natural orbitals (NOs) are very similar to Brueckner orbitals. pairs of occupied and virtual orbitals are mixed to eliminate the corresponding single-excitation contributions in ﬁrst order. The resulting orbitals are called Brueckner orbitals and can be obtained by an iterative procedure in which. In natural-orbital-based expansions for systems of more than two electrons the contribution of single excitations tends to be very small. making it relatively easy to eliminate single excitations from correlated two-electron wave functions. a CI expansion with single excitations only (CIS) based on an unrestricted or closed-shell restricted SCF function provides no improvement over SCF. triples) that the singles acquire nonzero coeﬃcients Cia in the expansion.7. brieﬂy. 1. It is only through the interaction of the single excitations with the doubles (and. the natural orbitals are the same as the Brueckner orbitals. up to N excitations. This process must be applied to an expansion that includes at least double excitations. . and a CIS wave function can provide a useful improvement over SCF. in each iteration. is obtained by using all terms of the CI expansion. These concepts are discussed in Section 1.

This aspect is closely related to point 2 above and allows. This property is most easily understood for a system such as a uniform electron gas or a system of noninteracting unis.g. as we shall see.7 Extensivity 11 a method is extensive if the energy of a system computed by this method scales correctly with system size. although redundantly. improved results.2. The PT and related MB techniques are generally nonvariational. these convergence problems can be circumvented by going to a multireference (MR) version such as QDPT or MRCC but. this is done. they can combine the most important quadruple-excitation contributions with the double excitations). Perturbation theory formulations can incorporate the most important higher-excitation eﬀects at a relatively low order (e. For arbitrary systems a precise deﬁnition is more diﬃcult (Nooijen. 1. in coupled-cluster (CC) methods. this introduces additional complications. Systematic diagrammatic techniques allow high-eﬃciency organization of the calculations. for example. but by leaving out the relatively unimportant parts of these higher excitations they avoid the labor that would be required if we tried to include these higher-excitation eﬀects in full. and in many other cases convergence can be slow. so that we have no bounds on the energy. . compared with CI. size-extensive ) if the energy of a system computed with this model scales correctly with the size of the system (Bartlett and Purvis 1978. including the summation of some types of term to inﬁnite order. at comparable levels of eﬀort. 1. Bartlett 1981). These problems can be circumvented in some cases by reorganization of the series.. such as N He atoms. 2. 3. which cannot separate these diﬀerent components of the higher excitations. Shamasundar and Mukherjee 2005) and will be dealt with more fully in Sections 5. The PT series does not always converge and thus cannot be applied in certain situations. A quantum-mechanical model is said to be extensive (commonly.10 and 9. 2. as we would have to do in CI. in principle.7.1.1. An example illustrating the lack of extensivity of truncated CI is presented in Section 1.7 Extensivity In thermodynamics. There also are disadvantages of such methods. a system property is called extensive if its value for a uniformly distributed system is proportional to the size of the system. for which E (N He) = N E (He). Alternatively.

φ0 |h ˆ |χ = β .11) Since this is a two-electron system. ˆ .13) (1. The elements of the corresponding Hamiltonian matrix will be written in the form ˆ |φ0 = ε0 . without single-excitation terms (see Section 1. φ0 |χ = 0 (intermediate normalization). computed in equivalent ways. but now we use a simple example to demonstrate the lack of extensivity of truncated CI. φ0 |h ˆ |χ = α . We shall discuss this question in more detail later.10) These two properties of the computed energy are not quite the same.12 Introduction A closely related concept is that of size-consistency . hψ (1.5). As deﬁned by Pople.14) . Davidson and Silver 1977). N H2 molecules) as examples of electron-pair bonds and examine the behavior of the energy as a function of N (Sasaki 1977.12) (1. a model is size-consistent if the energies of two systems A and B and of the combined system AB with A and B very far apart. in which case χ would be the only double-excitation conﬁguration. but in many cases they go hand-in-hand. (1. we could consider a calculation within the space spanned by just two basis functions. χ|h (1. equivalently. We shall consider a model system of N noninteracting He atoms (or. its exact wave function can be written in terms of natural orbitals as a linear combination of a reference function φ0 and all double excitations relative to it. wave First we shall examine the case of one He atom with Hamiltonian h function ψ and energy ε. Binkley and Seeger (1976). Let us express the wave function in the unnormalized form ψ = φ0 + cχ . alternatively. χ|χ = 1 . for which ˆ = εψ . satisfy E (AB) = E (A) + E (B) . and χ is the appropriate normalized aggregate of all double excitations relative to φ0 . where φ0 |φ0 = 1 .

because of the lack of interaction among the atoms and between φ0 and any single excitation from it. with Hamiltonian N ˆ = H i=1 ˆ (i) . c (1. ∆ε ∆ε2 − (α − ε0 )∆ε − |β |2 = 0 .20) where the arguments 1. (1. The reference function can where h be written Φ0 = Aφ0 (1)φ0 (2) · · · φ0 (N ) . .1. The double-excitation functions are of the form Φi = Aφ0 (1)φ0 (2) · · · φ0 (i − 1)χ(i)φ0 (i + 1) · · · φ0 (N ) . (1.18) which is a quadratic equation for the correlation energy ∆ε. β∗ = ε − α = ε 0 + ∆ε − α .19) ˆ (i) is the Hamiltonian for the ith atom. Now consider a system of N noninteracting He atoms. The second equation is β ∗ + αc = εc . (1. it can actually be ignored. h (1. (1. N identify the atoms. . or single excitations. .) The CI doubles (CID) wave function is then N ΨCID = Φ0 + i=1 Ci Φi (1. 2.15) which can be solved for the exact energy ε and the coeﬃcient c. (1.7 Extensivity 13 and the CI eigenvalue equation is ε0 β β∗ α 1 c =ε 1 c . because of the lack of interaction.16) Substituting c = ∆ε/β from the ﬁrst equation gives |β |2 = ε0 + ∆ε − α .21) (We do not need to consider mixed double excitations.17) ∆ε ≡ ε − ε0 = βc . . The antisymmetrizer A exchanges electrons between diﬀerent two-electron factors φ0 . The ﬁrst equation in the 2 × 2 system is ε0 + βc = ε .22) .

24) (1.. . . . (1.28) . . .27) is quadratic in the correlation energy ∆E . . with solution ∆E = α − ε0 ± 2 α − ε0 2 2 + N |β |2 .. and all the Ci = C are equal by symmetry).. Φ0 |H ˆ |χ = β . . We obtain the following matrix elements: Φ0 |Φ0 = 1 . Φ0 |Φi = 0 . ∆E ∆E 2 − (α − ε0 )∆E − N |β |2 = 0 . . E0 + α − ε0 (1. ˆ |Φ 0 = N ε 0 = E 0 . (1. . . ..25) . . . . C C (1.. .. . . . C Substituting C = ∆E/N β from (1. ˆ |Φi = φ0 |h Φ0 |H ˆ |Φj = δij [(N − 1)ε0 + α] Φi |H = δij (E0 + α − ε0 ) . From the other (equivalent) equations we get β ∗ + (E0 + α − ε0 )C = EC or β∗ = ε0 + ∆E − α . 1 1 C C C = E C . .26) Equation (1. . .14 Introduction (it is unnormalized. Φi |Φj = δij . The CI equations are given by β β E0 β ∗ E 0 + α − ε0 0 β∗ 0 E0 + α − ε0 .. β 0 0 .27) (1. . ∗ 0 0 β From the ﬁrst equation we get E0 + N βC = E or ∆E ≡ E − E0 = N βC .25) gives N |β |2 = ε0 + ∆E − α .23) .

31) = Φ0 + i=1 cΦi + i<j c Φij + · · · + c Φ123. 4 1 1 2 4 2 N (N − 1)c ∼ 2 N c .8 Disconnected clusters and extensivity 15 The behavior of ∆E as N → ∞ is then found to be √ ∆ε √ N ∆E −→ −|β | N = N →∞ |c| (1.16)).1. Of course.. The relative contributions to |Ψ|2 of diﬀerent excitation levels are: zero excitations. 2 N which can also be written as Ψ = Φ0 + i cΦi + 1 2 i.30) etc.32) Thus the coeﬃcients of the higher than double excitations are not independent parameters but are closely related to the coeﬃcients of the doubles. Obviously. (1.. doubles.33) etc. sextuples.j. quadruples. N c2 . instead of the correct answer ∆E = N ∆ε. No matter how small c2 is. the larger the system becomes the poorer will be the truncated CI expansion ΨCID .k c3 Φijk + · · · (1.j c2 Φij + 1 3! i.8 Disconnected clusters and extensivity The N -He-atoms example provides a simple explanation of the lack of extensivity of CID in terms of separated pairs (disconnected clusters). the coeﬃcients of the higher-excitation conﬁgurations . 6 1 1 3 6 3! N (N − 1)(N − 2)c ∼ 6 N c . 1.N . 1. then Ψ = Aψ (1)ψ (2) · · · ψ (N ) = A[φ0 (1) + cχ(1)][φ0 (2) + cχ(2)] · · · [φ0 (N ) + cχ(N )] N (1.29) (using ∆ε = βc from (1. in this case we know the correct answer. We shall return to this example in the discussions of other computational models. Therefore this truncated CI expansion is not extensive. If we deﬁne Φij = Aφ0 (1) · · · φ0 (i − 1)χ(i)φ0 (i + 1) · · · φ0 (j − 1)χ(j )φ0 (j + 1) · · · φ0 (N ) (1. As previously stated. if N is large enough then the higherexcitation contributions eventually become dominant.

. T ˆ4 . etc. a quadruple excitation represents the correlation of four electrons at a time.32). T ˆ3 .36) . obtained from the expansion of ˆ ˆ may be limited to T ˆ2 or may contain an exponential operator eT (where T ˆ3 etc.. 1 T ˆ3 1ˆ2ˆ ˆ1 T ˆ2 T contributions 2 T1 . as well as additional disconnected cluster ter contributions T 1 ˆ2 ˆ ˆ ˆ3 . T additional connected-cluster operators T If we bring the N He atoms closer and let them interact. ˆ1 . . (1.). it will have additional connected clusˆ1 . The CI quadruple excitation expansion of eT (where T contribution is represented by the operator ˆ4 + 1 T ˆ2 ˆ4 = T C 2 2 + 1 ˆ4 4! T1 ˆ1 T ˆ3 + 1 T ˆ2ˆ +T 2 1 T2 . . (1. In fact. not from CID. T1 T2 . sextuples.. so that if we knew the correct double-excitation coeﬃcients c. While a double-excitation conﬁguration represents the correlation of a pair of electrons (a two-electron cluster ). T ˆ3 . being made up of two simultaneous but independent twoelectron clusters. is generally ˆ2 . etc.35) ˆ ˆ2 In the noninteracting case 1 2 T2 is the only contribution to C4 .34) equivalent to the expansion (1. T 3! 1 . T . . the two-electron ˆ2 corresponding to a linear comclusters are represented by an operator T bination of double-excitation operators and the higher-order disconnected clusters are represented by products of connected-cluster operators. i. ˆ4 . But the four-electron correlations in this example really represent two separate two-electron correlations occurring simultaneously. based upon the order of PT in which the contribution ﬁrst appears (for an SCF reference state). the order of importance. T 1 ˆ2 2 T2 .16 Introduction (quadruples. T ˆ3 . ˆ1 . but even in the interacting case it will still be by far the most important. .) in this example are not independent parameters. the full CI wave function will no longer be represented exactly as a product of one-atom wave functions ˆ2 + (1 + T 1 ˆ2 2! T2 + 1 ˆ3 3! T2 + · · · )Φ0 = eT2 Φ0 ˆ (1. such as 1 ˆ2 2 T2 for disconnected quadruple excitations. 2 T1 T2 . T 1 ˆ3 3! T2 . obtained from the ˆ ˆ =T ˆ1 + T ˆ2 + · · · ). In the language of coupled-cluster theory. .e. we would know the complete wave function and would be able to obtain the exact energy. those obtained from a full CI calculation. . This type of four-electron cluster is called a disconnected cluster.

8 Disconnected clusters and extensivity 17 In CI we cannot compute the individual components of (1.35) separately ˆ4 . Furthermore.1. . because in these methods it is not combined with the much ˆ4 . it is the disconnected clusters that are the more diﬃcult T key for extensivity (as is clear from the N -He-atoms example). Thus ˆ 2 is much easier to compute than T but in PT and CC we can. and T 2 1 ˆ2 we can obtain the major eﬀect of the quadruple excitations 2 T2 more easily in PT and CC.

2. 2. Each diﬀerent approach has contributed to the lore and the language of many-body theory. We can quantities (components of H make a considerable amount of progress in this way before considering the ˆ .1 Background There are two stages in the study of perturbation theory and related techniques (although they are mixed intimately in most derivations in the literature). If we want to be able to read the literature in this ﬁeld. We separate the Hamiltonian into a zero-order 18 . integrals over orbitals etc. notations and derivations that have been used. carried out in terms of the total Hamiltonian and total wave function (and total zero-order wave function). in particular. and each tends to illuminate some aspects better than the other approaches. orbitals. Another aspect of the study of many-body techniques is the large variety of approaches. where all detailed form of H expressions are obtained in terms of orbitals (one-electron states) and oneand two-electron integrals.2 Classical derivation of Rayleigh–Schr¨ odinger perturbation theory 2. The ﬁrst is the formal development.1 The perturbation Ansatz We begin with a classical textbook derivation of formal Rayleigh–Schr¨ odinger perturbation theory (RSPT). we shall occasionally derive some results in more than one way and. Therefore. without attempt to express anything in terms of one. we shall derive the basic perturbation-theory equations and their many-body representations in several complementary ways.and two-body ˆ .). We shall try to keep these separate for a while and begin with a consideration of formal perturbation theory. we should be familiar with several alternative formulations. The second is the many-body development.2 Formal perturbation theory 2.

H while for the zero-order problem (assumed soluble) we write (0) ˆ 0 Φn = En H Φn . The exact (unknown) solutions will be written ˆ Ψn = En Ψn .8) . we shall write this in the form ˆ =H ˆ 0 + λV ˆ.2. (2. (0) lim En = En . ˆ )Φn . ˆ (H − En )χn = (∆En − V (2.2 Classical derivation of RSPT 19 part and a perturbation. (2.7) and then rewrite the Schr¨ odinger equation in the forms ˆ (Φn + χn ) = En (Φn + χn ) .6) λ→0 (If there are degeneracies. ˆ =H ˆ0 + V ˆ.5) If Φn is nondegenerate.1) where λ is an “order parameter” that is used to classify the various contributions by their order and set to λ = 1 at the end. (2. (0) ∆En = En − En . it is possible to number the solutions in such a way that λ→0 lim Ψn = Φn .2) (2.4) with Φm |Φn = δmn .3) (2. (2. H (0) ˆ 0 Φn + V ˆ Φn + Hχ ˆ n = En H Φn + ∆En Φn + En χn .6) is still satisﬁed. H (2. H When convenient. it is possible to choose the zero-order solutions so that (2.) Let us deﬁne the diﬀerences χn = Ψn − Φn .

we partition Ψn into two parts. For example.20 Formal perturbation theory 2. . proportional) to Φn and the other orthogonal to it. If we have a particular solution for such an equation. (2. Thus χn is not fully determined. Af ˆ is a linear operator.10) This choice means that we extract only that part of the correction χn to Φn that is essentially diﬀerent from Φn .2 Indeterminacy of the solution If we assume for the moment that ∆En (and thus En ) is known then (2. (2. This degree of freedom can be used to force χn to be orthogonal to Φn . since ˆ ˆ = Af ˆ 0 + cAh Af = g+0 = g.8) is an inhomogeneous linear diﬀerential equation for χn . ˆ − En )f = 0 . satisfying ˆ 0=g Af while h is a solution of the homogeneous equation ˆ =0 Ah then f = f0 + ch (where c is an arbitrary constant) is also a solution of the original equation. it is convenient to use intermediate normalization : Φn |Φn = 1 .9) (H and still have a solution. As the choice of normalization is arbitrary. we can add to it any multiple of the solution of the corresponding homogeneous equation. if the inhomogeneous equation is ˆ = g. since the right-hand side is known. χn |Φn = 0 .9) that is consistent with En .e. since we can add to it an arbitrary multiple of Ψn = Φn + χn . In other words.11) Ψn |Φn = Φn + χn |Φn = 1 + 0 = 1 . g is a known function and f is the unknown where A function and if f0 is a particular solution. (2. one parallel (i.2. χn |Φn = 0 . the only solution of (2. Ψn |Ψn = 1 + χn |χn .

14) En = Ψn |Ψn (which has an error of order 2 for a wave function error of order ). we use the order parameter λ and expand: (1) 2 (2) Ψn = Φn + χn = Ψ(0) n + λΨn + λ Ψn + · · · (0) (0) (1) (2) En = E n + ∆En = En + λEn + λ2 En + ··· (Ψ(0) n ≡ Φn ) . (2.2. ∆En = Φn |V ˆ |Ψ n . then the error in En is also of order and may be greater than the error in the variational estimate ˆ |Ψn Ψn |H (2.17) (2.) 2.2 Classical derivation of RSPT 21 2. = Φn |V (2. Subtracting the second equation from the ﬁrst. we get ˆ =H ˆ 0 + λV with H (0) (1) (2) (1) 2 (2) ˆ − En ˆ 0 + λV − λEn − λ 2 En − · · · )(Ψ(0) (H n + λΨn + λ Ψn + · · · ) = 0 .11) we get ˆ |Ψn = En Φn |Ψn = En . (2. En = Φn |H (2.2.16) . we have (0) ˆ −H ˆ 0 |Ψn ∆En = En − En = Φn |H ˆ |Ψ n .13).4 Order-by-order expansion To proceed further. with an error of order . When Ψn is exact this En is also exact but if Ψn is approximate. Φn |H (0) (0) ˆ 0 |Ψn = H ˆ 0 Φn |Ψn = En Φn |H Φn |Ψn = En .13) The last equation is sometimes referred to as the transition matrix element form (or projected form ) for the energy.12) Thus ˆ |Ψn .3 Energy expressions From the intermediate-normalization assumption (2.15) Substituting into the Schr¨ odinger equation ˆ − En )Ψn = 0 (H ˆ . (Note that if Ψn is a full CI expansion based on Φn and its excitations then only the coeﬃcients of the single and double excitations are needed to determine the exact energy from (2.2.

20) which becomes (0) m) m−1) ˆ 0 )Ψ( ˆ Ψ( −H =V − (En n n (m) m−1 l=0 (m−l) (l) En Ψn .22 Formal perturbation theory Equating coeﬃcients of powers of λ gives for λ0 . the mth-order equation (0) m) (1) m−1) ˆ )Ψ( ˆ 0 − En )Ψ( = (En −V + (H n n m−2 l=0 (m−l) (l) En Ψn . We are free to require (1) Φn |Ψn = 0. λ1 and λ2 . (second order) (2. respectively: ˆ 0 − E (0) )Ψ(0) = 0 (H n n (0) (1) ˆ ˆ (0) (H0 − En )Ψ(1) n = (En − V )Ψn (0) (1) (2) (0) ˆ 0 − En ˆ (1) (H )Ψ(2) n = (En − V )Ψn + En Ψn and in general.25) . For λ1 we get we apply Φn | to each equation and (2. the intermediate-normalization condition. This is equivalent to adding an arbitrary multiple of Φn (the solution of the corresponding (0) (1) ˆ 0 − En )f = 0) to Ψn . (ﬁrst order). Φn |H n n n ˆ 0 we have By the Hermitian property of H (0) (1) ˆ 0 − En ˆ |Φn (H )Φn |Ψ(1) = En − Φn |V n =0 ≡ Vnn and so (1) ˆ |Φn = Vnn . (zero order). = δl0 giving (m) m−1) ˆ |Ψ ( = Φn |V .18) (2. for λm . The same can be done homogeneous equation (H for each order m: (0) m) m−1) ˆ 0 |Ψ( ˆ |Ψ ( −H = Φn |V − Φn |En n n =0 m−1 l=0 (m−l) l) En Φn |Ψ( n .22) ˆ 0 − E (0) |Ψ(1) = Φn |E (1) − V ˆ |Φn . Vij = Φi |V (1) (1) (2. En n (2. (2. = Φn |V En (2.23) where we set ˆ |Φ j .21) In order to get expressions for En integrate.24) Thus we have obtained En without knowledge of Ψn and can now solve (1) the inhomogeneous diﬀerential equation for Ψn .19) (2.

then solve for Ψn etc. while always maintaining Φn |Ψn 2. in principle.27) (0) In this equation the l = 0 contributions are to be interpreted as akn = Φk |Φn = δkn . we can obtain each En from the previous Ψn and (m) (m) = 0 (m > 0). though they are easier than eigenvalue equations.. (m) This result provides a system of equations for the akn coeﬃcients.26) = (to be determined).2. The following choice of intermediate normalization can thus be made for each order: m) =0 Φn |Ψ( n m) a( nn = 0 (m > 0) . corresponding to the arbitrariness of adding any multiple of the zero-order solution Φn (of the zero-order homogeneous equation for the same n). This arbitrariness appears for each (m) order Ψn separately.2. This exploits the fact that the set of eigenfunctions of any semibounded Hermitian operator form a complete set: m) = Ψ( n k (m) akn (m) (m) akn Φk = k m) Φk |Ψ( n (m) m) |Φk Φk |Ψ( . To obtain akn we multiply the mth-order equation by Φk | and integrate: (0) m) ˆ 0 | Ψ( −H = Φk |En n =(En −Ek ) Φk | (0) (0) j m−1) ˆ |Ψ( Φk |V n ˆ |Φj Φj |Ψn Φk |V (m−1) m−1 − l=0 (m−l) l) En Φk |Ψ( .5 Expansion in zero-order functions The inhomogeneous diﬀerential equations for Ψn are not easy to solve. (2. n =akn (l) Thus (0) (En − (0) (m) Ek )akn = j (m−1) Vkj ajn m−1 − l=0 (m−l) En akn .2 Classical derivation of RSPT (m) (m−1) 23 Thus. to be solved order by order. One way to proceed is (m) to expand the unknown Ψn in terms of the known zero-order solutions Φk . (l) (2.28) . but the ﬁrst thing to notice is that we have no equa(m) tion for ann . this coeﬃcient is arbitrary. (m > 0) . n (2. Consequently m) a( nn = δm0 .

to raise the order of the “bra” part while lowering the order of the “ket” part.6 Wigner’s 2n + 1 rule Wigner’s rule (Wigner 1935) says that knowledge of Ψn for l = 1. so that akn = (1) Vkn (0) En − Ek (0) (n = k ) . (2. m (l) allows us to determine directly the En for l = 1. akn Vnk k (1) (2) ˆ |Ψ(1) En = Φn |V = n = k Vnk Vkn E n − Ek (0) (0) = k |Vkn |2 E n − Ek (0) (0) (2. From this we get the second-order energy.. (0) − Ek akn = En (2) (0) −1 j (1) (2) ajn Vkj − En akn − En akn (1) (1) (0) = j Vkj Vjn (0) (En − (0) (0) Ek )(En − (0) Ej ) − Vkn Vnn (0) (0) (En − Ek )2 (k = n).24 (0) Formal perturbation theory Since akn = δkn . which is also well known. (l) (l) .31) . . so we will look at more systematic techniques.30) where k ≡ k (k=n) .29) Thus we have the well-known result Ψ(1) n = k Vkn (0) En − Ek (0) Φk (2. 2. . it is somewhat cumbersome and not very convenient for the manybody techniques that we shall want to use. However. 2.2. e. repeatedly until the two orders are equal (if m is odd) or diﬀer by unity (if m is even). . the Wigner 2n + 1 rule and the Hylleraas variational principle for the ﬁrst-order wave function. This result can be obtained by a series of transformations of the mth-order energy (m) m−1) ˆ |Ψ( . the ﬁrst-order equation becomes: (0) (1) − Ek )akn = Vkn − En akn = Vkn (En (0) (1) (0) (n = k ). before we do that. . 2m + 1. . (2. using the diﬀerential equations contribution (2. En = Φn |V n for Ψn .g. we will consider two more aspects of this analysis. 2. This process can be continued in the same manner to higher orders.32) However. . . .25).

using the ﬁrst-order equation (2. (1) .21): (m) m−1) ˆ |Ψ( En = Φn |V n ˆ Φn |Ψ(m−1) = V = = = = n (0) (m−1) (1) ˆ 0 )Ψ(1) (En − H + En n |Ψn (0) (m−1) ˆ Ψ(1) +0 n |En − H0 |Ψn m−2 (m−2) (m−l−1) ˆ | V | Ψ − En Ψ(1) n n l=0 m−2 (m−l−1) ˆ (m−2) − En Ψ(1) n |V |Ψn l=1 m−1) Φn |Ψ( n (l) Ψ(1) n |Ψn (l) Ψ(1) n |Ψn (2. e.2 Classical derivation of RSPT 25 We shall work out one step of this repetitive sequence explicitly.7 The Hylleraas variation principle for the ﬁrst-order wave function Instead of expanding Ψn in the zero-order functions Φk .34) (since the l = 0 term vanishes). (2.g.2.. respectively. we get Wigner’s results: (2m) m−1) ˆ m) = Ψ( |V |Ψ( − En n n m−1 m j =1 l=1 m (2m+1) En m (2m−j −l) j) (l) En Ψ( n |Ψn .2. (2. the ﬁrst-order wave function and the second-order energy can also be obtained by a variational method due to Hylleraas (1930) (see.36) = m) ˆ (m) Ψ( n |V |Ψn − j =1 l=1 (2m−j −l+1) j) (l) En Ψ( n |Ψn . If this is repeated k times we get (m) k) ˆ (m−k−1) En = Ψ( − n |V |Ψn k m−k−1 j =1 l=1 (m−j −l) j) (l) En Ψ( n |Ψn .33) and then the (m − 1)th-order equation in its usual form (2. (2. pp.19) in the form (0) ˆ 0 )Ψ(1) + E (1) Φn ˆ Φn = (En −H V n n (2. Bethe and Salpeter (1957). 122–3).37) 2.35) Taking the cases in which m is replaced by 2m or 2m + 1 and k by m − 1 or m.

26 Formal perturbation theory (1) We begin by multiplying the ﬁrst-order equation (2. (2. (2.42) . Thus Ψn and En can be obtained by making J2 [Ψ] stationary. J2 [Ψ] = 2 Re Ψ|V n n then obviously (2) J2 [Ψ(1) n ] = En . (2) ˆ − E (1) |Ψ(1) En = Φn |V n n (note that we have not invoked Φn |Ψn (1) = 0). (2.38) n |V − En |Φn + Ψn |H0 − En |Ψn If we deﬁne a functional ˆ − E (1) |Φn + Ψ|H ˆ 0 − E (0) |Ψ . n |V − En |Φn + Ψn |H0 − En |Ψn To this equation we add the equation for the second-order energy.40) The variation of this functional is ˆ − E (1) |Φn + Φn |V ˆ − E (1) |δ Ψ δJ2 [Ψ] = δ Ψ|V n n (0) (0) ˆ ˆ + δ Ψ|H0 − En |Ψ + Ψ|H0 − En |δ Ψ . (2. Requiring that δJ2 [Ψ] = 0 for any δ Ψ (including δ Ψ∗ ) results in (1) (0) ˆ − En ˆ 0 − En (V )Φn + (H )Ψ = 0 .41) for which Ψ = Ψn is a solution (since the above relation is equivalent to (1) (2) the ﬁrst-order equation).19) by Ψn | and integrating: (0) (1) (1) ˆ ˆ Ψ(1) = Ψ(1) n |H0 − En |Ψn n |En − V |Φn or (1) (1) ˆ (0) (1) ˆ 0 = Ψ(1) .39) (2. The result is (2) (1) (1) (1) (0) (1) ˆ ˆ ˆ En = Ψ(1) + Ψ(1) n |V − En |Φn + Φn |V − En |Ψn n |H0 − En |Ψn (1) (1) ˆ (0) (1) ˆ = 2 Re Ψ(1) . It is easily veriﬁed that J2 [Ψ + cΦn ] = J2 [Ψ] for any constant c. and this degree of freedom can be used to enforce Ψ|Φn = 0. If the function Ψ is constrained at the outset to be orthogonal to Φn then J2 [Ψ] can be written in the form (0) ˆ |Φn + Ψ|H ˆ 0 − En J2 [Ψ] = 2 Re Ψ|V |Ψ (1) ( Ψ|Φn = 0) .

J2 [Ψ] = En (2. We deﬁne the projection operators in terms of the orthonormal zero-order functions: ˆ = |Φ0 Φ0 | .47) . instead of n.3 Projection operators Before we proceed to alternative derivations of the perturbation theory equations. the exact energy E0 ≡ E and their order-by-order contributions Ψ(m) and E (m) . we shall use the index 0. P ˆ=ˆ ˆ= Q 1−P i=0 |Φi Φi | = i |Φi Φi | (2. Therefore (2) J2 [Ψ] ≥ En (0) (En ≤ Ek (0) for all k = n) . and only if. energy (provided that Φ0 is an exact eigenfunction of H 2.46) Thus J2 [Ψ]. we ﬁnd that the third term can be written as the negative of the second term. χ is the corresponding eigenfunction.44) Using the ﬁrst-order equation. n (2. canceling it. can be used in a variational approach for ﬁnding approximations to the ﬁrst-order wave function and second-order energy. we shall introduce two projection operators. with an arbitrary trial function Ψ containing adjustable parameters.45) (0) (0) ˆ 0 then the integral χ|H ˆ 0 − En |χ is If En is the lowest eigenvalue of H nonnegative and is zero if. and this provides (2) an upper bound to En in the case of a state having the lowest zero-order ˆ 0 ). To simplify the notation.3 Projection operators 27 (0) ˆ 0 then J2 [Ψ] is an Next we show that if En is the lowest eigenvalue of H (2) upper bound for En .43) (2. We shall omit this index entirely when there is no danger of confusion. We write Ψ in the form Ψ = Ψ(1) n + χ. Then (1) ˆ J2 [Ψ] = J2 [Ψ(1) n ] + 2 Re χ|V − En |Φn (0) (0) ˆ 0 − En ˆ 0 − En + 2 Re χ|H |Ψ(1) + χ|H |χ . so that we have (2) (0) ˆ 0 − En + χ|H |χ . (2. and particularly when we refer to the exact wave function Ψ0 ≡ Ψ.2. to indicate the state in which we are interested (regardless of whether it is the ground state or some excited state).

Q (2. (2. Hermitian and idempo- ˆ 2 = |Φ0 Φ0 |Φ0 Φ0 | = |Φ0 Φ0 | = P ˆ. (A more general expression for these operators. Note that h|h −1 = S−1 is the inverse of the overlap matrix S = h|h .52) ˆ projects out the orthogonal complement to Φ0 . When P i.51) ˆ annihilates the ˆ extracts the Φ0 component from Ψ.28 Formal perturbation theory i (where tent : ≡ i (i=0) ).50) then ˆΨ = P i ai |Φ0 Φ0 |Φi = i ai |Φ0 δ0i = a0 Φ0 . Ψ=ˆ 1Ψ = (P (2. (2. is given by the form ˆ = |Φ0 Φ0 |Φ0 −1 Φ0 | . Any function The operator Q can then be written as a sum of the two projections: ˆ+Q ˆ )Ψ = P ˆΨ + Q ˆΨ .e. ˆ 2 = (ˆ 1−P 1−P 1−P Q ˆQ ˆ=Q ˆP ˆ=ˆ P 0. ˆ Ψ = (ˆ ˆ )Ψ = Ψ − a0 Φ0 = Q 1−P i ai Φi . It is easy to see that this form of the operators satisﬁes the projection operator conditions. any operator can be resolved into four components.) ˆ operates on a general function Ψ expressed in terms of the Φi . Q so that P Φ0 component. (2. P ˆ )2 = ˆ ˆ−P ˆ+P ˆ2 = ˆ ˆ=Q ˆ. Similarly.53) This is often termed the “resolution of the identity” into its components.54) .49) where |h is a row vector consisting of all the functions |Φi (i = 0) and h| is its adjoint column vector. Similarly. A (2. Ψ= i ai Φi . ˆ = (P ˆ+Q ˆ )A ˆ(P ˆ+Q ˆ) A ˆA ˆQ ˆ+Q ˆA ˆP ˆ ˆA ˆP ˆ+Q ˆA ˆQ ˆ +P =P = ˆD A + ˆX . These operators are linear.48) as required for projection operators. P ˆ = |h h|h −1 h| . applicable to nonorthonormal zero-order functions as long as Φ0 is orthogonal to all the other Φi . (2.

[Q.2. The only requirements are that these functions are orthonormal. ˆB ˆ )X = A ˆD B ˆX B (A (2.56) we have ˆH ˆ 0f = P i ˆH ˆ 0 Φi = ci P i (0) ˆ (0) ci E i P Φi = c0 E0 Φ0 . Similarly ˆ 0P ˆH ˆ0 = H Thus P ˆ ˆ ˆ ˆ 1 − P ).57) ˆ (since f is arbitrary). and shall omit this index altogether whenever there is no danger of confusion. so that [P ˆ.or oﬀ-diagonal part ˆX = P or the even part) and A ˆ (also called AO . because in terms of these functions the matrix representation of H diagonal. For a general The projection operators P function f expanded in the zero-order functions Φi .4 General derivation of formal perturbation theories 29 ˆD = P ˆA ˆP ˆ+Q ˆA ˆQ ˆ is the block-diagonal part of A ˆ (also called A ˆE .55) ˆ and Q ˆ commute with H ˆ 0 . The case in which all the Φi are eigenfunctions of H which some simpliﬁcations are possible. Note that ˆD + A ˆX . of H ˆ 0 and that all other Φi span the orthogonal that Φ0 is an eigenfunction of H ˆ 0 .4. H0 ] = 0 (to see this. will be referred to as the diagonal ˆ 0 is case. in terms of which the solutions are to be expanded. . ˆB ˆ )D = A ˆD B ˆX B (A ˆX + A ˆD . f= i ci Φi .4 General derivation of formal time-independent perturbation theories 2. instead of n. or the odd part). H ˆ 0 ] = 0. are eigenfunctions ˆ 0 . (2. ˆ 0P ˆf = H i ˆ 0P ˆ Φi = c0 H ˆ 0 Φ0 = c0 E (0) Φ0 . We shall use λ = 1 and determine the ordering ˆ factors of the contributions to En and Ψn on the basis of the number of V and other considerations. in complement of Φ0 . To be more general. to indicate the state in which we are interested (regardless of whether it is the ground state or some excited state).1 General formalism Now we turn to a more general. where A ˆA ˆQ ˆ+Q ˆA ˆP ˆ is the cross. set Q = ˆ 2. and somewhat more elegant. we shall not always require that all the functions Φi . ci H 0 (2. derivation of the perturbation equations. We shall continue to use the index 0.

−H we introduce an arbitrary parameter ζ by adding ζ Ψ to each side: ˆ 0 )Ψ = (V ˆ − E + ζ )Ψ . Q (2. ζ = E or ζ = E0 ).58) with Φ0 | we obtain ˆ 0 | Ψ + Φ0 |V |Ψ = E Φ0 |Ψ .58) as ˆ 0 Ψ = (V ˆ − E )Ψ .g. as we already know. (ζ − H (2. (H with zero-order solution ˆ 0 Φ0 = E (0) Φ0 .60) Φ0 | ˆ 0 .59) (2. Next. so that ∆E = Φ0 |V |Ψ .61) |Φi Φi |Ψ (2.65) (2. H 0 Operating on (2. Φ0 |H =E0 (0) (2.62) = Φ0 + χ .66) .64) This parameter is a convenience that will allow us to obtain diﬀerent perturbation expansions from the same development by diﬀerent choices of ζ (0) (e. whence ˆΨ . χ=Q (2.58) (2. With intermediate norwhere we have used the Hermitian property of H malization this gives E0 + Φ0 |V |Ψ = E . ˆ to both sides: We next apply Q ˆ (ζ − H ˆ 0 )Ψ = Q ˆ (V ˆ − E + ζ )Ψ . we write Ψ in the form ˆΨ + Q ˆ Ψ = |Φ0 Φ0 |Ψ + Ψ=P i (0) (2.30 Formal perturbation theory We begin the derivation with the Schr¨ odinger equation in the form ˆ0 + V ˆ )Ψ = E Ψ .63) Rearranging the Schr¨ odinger equation (2. Φ0 = P ˆΨ ..

ˆ 0 )−1 . (2.71) ˆ 0 .) ˆ = εφ then f (H ˆ )φ = f (ε)φ. we now have ˆΨ = Q ˆ (V ˆ − E + ζ )Ψ . so that the inverse of Q ˆ (ζ − H ˆ 0 )Q ˆ in any of the eigenvalues of H (0) ˆ 0 then a choice of Q-space exists. |Φi Φi |ζ − H (2. Q (2.) This inverse is written in the form ˆ 0 (ζ ) = R ˆ Q ˆ0 ζ −H ≡ i j ˆ 0 )−1 |Φj Φj | |Φi Φi (ζ − H (2.4 General derivation of formal perturbation theories 31 ˆ and its commutation with H ˆ 0 . (If E0 is a well-separated eigenvalue of H (0) ζ in the neighborhood of E0 would be satisfactory.68) which simpliﬁes in the diagonal case to ˆ= ˆ (ζ − H ˆ 0 )Q Q i |Φi (ζ − Ei ) Φi | .66).72) which rationalizes the notation ˆ Q ˆ0 ζ −H .2. ˆ (ζ − H ˆ 0 )Q Q (2.67) ˆ operates entirely in Q-space and in terms of the ˆ (ζ − H ˆ 0 )Q The operator Q zero-order functions has the expansion ˆ= ˆ (ζ − H ˆ 0 )Q Q i j ˆ 0 |Φj Φj | . (0) (2. In the diagonal case it can be expressed and is called the resolvent of H as ˆ 0 (ζ ) = R i j |Φi Φi |(ζ − Ej )−1 |Φj Φj | = (0) i |Φi Φi | (ζ − Ei ) (0) .69) Going back to the rewritten Schr¨ odinger equation (2. here f (H ˆ 0 ) = (ζ − H (If Hφ .70) We restrict the possible choices of ζ to those that do not coincide with ˆ 0 in Q-space. the operator on Using the idempotency of Q the left-hand side can be rewritten as ˆ (ζ − H ˆ 0 )Ψ = Q ˆ 2 (ζ − H ˆ 0 )Ψ = Q ˆ (ζ − H ˆ 0 )Q ˆΨ .

substituting in Ψ = Φ0 + Q ˆ 0 (ζ )(V ˆ − E + ζ )Ψ .) that space they are their own inverses. we get Applying R ˆ ζ −H0 ˆΨ = R ˆ 0 (ζ )(V ˆ − E + ζ )Ψ . in the complete Hilbert space.l |Φi |Φi i. Projection operators have no inverses outside the space onto which they project. This would be a formal solution to the problem except that the unknown E appears on the right-hand side.74) (2.l ˆ 0 )−1 Φi (ζ − H j ˆ 0 ) Φl |Φj Φj | (ζ − H = = i ˆ 0 ) Φl ˆ 0 )−1 (1 − |Φ0 Φ0 |)(ζ − H Φi (ζ − H ˆ.75) This can be seen as an iterative equation for Ψ.32 Formal perturbation theory ˆ 0 (ζ ) as given in the previous equation is the inverse It is easy to see that R ˆ in Q-space. the mth term in the sum does not simply correspond to the . It is not A necessarily a general inverse. i. ∞ (2. substituting the entire right-hand side of (2.77) (provided that the series converges). if we continue in the same manner.70). we have and thus.76) Ψ= m=0 ˆ 0 (ζ )(V ˆ − E + ζ) R m Φ0 (2. For the same reason. Q ˆ ˆ 0 (ζ ) = Q to both sides of (2.l ˆ 0 )|Φl Φl | |Φk Φk |(ζ − H Φl | Φl | (2.j ˆ 0 )−1 |Φj Φj | |Φi Φi |(ζ − H k.73) = i. within ˆQ ˆ=Q ˆ . |Φi Φi | = Q because the sums exclude i = 0. l = 0. since. since ˆ (ζ − H ˆ 0 )Q of Q ˆ Q ˆ0 ζ −H ˆ (ζ − H ˆ 0 )Q ˆ Q = i.e. (Note that the inverse of an operator ˆ in Q-space is any operator B ˆ that satisﬁes A ˆB ˆ = B ˆA ˆ = Q ˆ .. e.g. Q ˆ Ψ. For example.75) into the Ψ on the right-hand side of the same equation. we have ˆ 0 (ζ )(V ˆ − E + ζ )Φ0 Ψ = Φ0 + R ˆ 0 (ζ )(V ˆ − E + ζ )R ˆ 0 (ζ )(V ˆ − E + ζ )Ψ +R and. Ψ = Φ0 + R (2.

It is instructive to see how the BWPT equations would be evaluated (ignoring the problem with E ). which produces Brillouin–Wigner PT (BWPT) (Brillouin 1932. using ∆E = Φ0 |V ∞ ∆E = m=0 ˆ R ˆ 0 (ζ )(V ˆ − E + ζ) Φ0 |V (0) m |Φ0 .81) . Wigner 1935).(and/or third-) order correction. Use the corrected E to compute a better second.80) but the equations are still implicit. since the unknown E appears on the r. (2. Put in an estimate for E .79) m ∆E = m=0 ˆ R ˆ 0 (E )V ˆ Φ0 |V |Φ0 More explicitly. In this case ˆ 0 (E ) = R ˆ Q ˆ0 E−H = i |Φi 1 E − Ei (0) Φi | . they can be solved iteratively. In principle. (2. + Φ0 |V (2.2. the energy expansion becomes ˆ |Φ0 + Φ0 |V ˆR ˆ 0 (E )V ˆ |Φ0 ∆E = Φ0 |V ˆR ˆ 0 (E )V ˆR ˆ 0 (E )V ˆ |Φ0 + · · · . However. assuming the diagonal case for simplicity.4 General derivation of formal perturbation theories 33 mth-order correction to the wave function. until self-consistency is obtained. The ﬁrst is the choice ζ = E . (0) ˆ |Φ0 and then compute the secondsuch as the ﬁrst-order result E0 + Φ0 |V order correction.s. etc. of diﬀerent orders. In any case.4. this procedure is generally too cumbersome to be very practical.78) which is an implicit equation for E = E0 + ∆E . ˆ − E + ζ )m and leads to ˆ 0 (ζ )(V This choice cancels −E + ζ in R ∞ Ψ= m=0 ∞ ˆ 0 (E )V ˆ R m Φ0 (BWPT). (2. 2. since E contains contributions ˆ |Ψ we have.h. formally.2 Brillouin–Wigner perturbation theory Diﬀerent choices of ζ can now be made in order to obtain diﬀerent types of perturbation theory (PT) expansions..

because of the absence of the so-called renormalization terms – the addi−1 tional sums m l=0 involving the lower-order energies in RSPT. it does not provide a true order-by-order expansion for the energy (because of the presence of the inﬁnite-order E ).4. In this basis the complete Hamiltonian for one atom is ˆ = (P ˆ (P ˆP ˆQ ˆP ˆQ ˆ+Q ˆ )h ˆ+Q ˆ) = P ˆh ˆ+P ˆh ˆ+Q ˆh ˆ+Q ˆh ˆ h (2.j V0i Vij Vj 0 (E − Ei )(E − Ej ) (0) (0) (BWPT) (2.34 Formal perturbation theory and the second-order energy becomes ˆR ˆ 0 (E )V ˆ |Φ 0 E (2) = Φ0 |V = i ˆ |Φi (E − E (0) )−1 Φi |V ˆ |Φ0 Φ0 |V i V0i Vi0 i (0) = E − Ei . BWPT resembles CI in many respects. . are the zeroorder energies associated with the zero-order functions Φ0 . and take the second-order energy as an example: E (2) = i |Vi0 |2 E − Ei (0) .j ˆ |Φi (E − E )−1 Φi |V ˆ |Φj (E − E )−1 Φj |V ˆ |Φ0 Φ0 |V i j (0) (0) = i. 2.83) etc. φ0 (i) and χ(i) for the ith atom. . We shall use a basis of two conﬁgurations per atom. In fact. Ei . and related to this is a very fundamental problem with BWPT.85) .84) where E is the exact ground-state energy and E0 .3 Demonstration of non-extensivity of ﬁnite-order BWPT We shall consider again the model problem of N noninteracting He atoms.7. (2. as in Section 1.82) which is similar to the RSPT result except for the presence of E rather than (0) E0 in the denominator. (0) (0) (2. its lack of extensivity when stopped at any ﬁnite order. However. for third order: E (3) = i. The form of these equations is actually simpler than that of the RSPT equations (ignoring the problem of the unknown E in the denominators). . Similarly. Φi . and successive orders of the BWPT calculation are closely related to the successive iterations used to solve the matrix eigenvalue problem in some CI procedures. . . .

88) (2.7) Ei (0) = E0 + α − ε0 . The zero-order Hamiltonian is H ˆ 0 = |Φ0 Φ0 |H ˆ |Φ0 Φ0 | + H i i h0 (i). ˆ0 = etc. . (2. (2. . etc. Φij = Aφ0 (1) · · · χ(i) · · · χ(j ) · · · φ0 (N ) .90) ˆ also given as ˆ |Φi Φi | |Φi Φi |H + i<j ˆ |Φij Φij | + . using the notation of Section 1. h and the perturbation v ˆ is v ˆ = |φ0 β χ| + |χ β ∗ φ0 | .4 General derivation of formal perturbation theories 35 ˆ = |φ0 φ0 |. |Φij Φij |H (0) = |Φ0 E0 Φ0 | + i |Φi Ei (0) Φi | + i<j |Φij Eij Φij | + · · · (0) (2. j = 0. in which (see Section 1. ˆ 0 = |φ0 ε0 φ0 | + |χ α χ| . h =1 =0 (2. H 0 0 ˆ 0 |Φi = H j (0) |Φj Ej (2.87) ˆ 0 with eigenvalues ε0 and α.86) The zero-order Hamiltonian for the atom can be taken as the diagonal part.93) Φj |Φi + 0 = δij (0) (E0 + α − ε0 )|Φi .89) ˆ 0 |χ = |φ0 ε0 φ0 |χ + |χ α χ|χ = α|χ . (0) Eij = E0 + 2α − 2ε0 . . Q ˆ = |χ χ|) and. (here P ˆ = |φ0 ε0 φ0 | + |φ0 β χ| + |χ β ∗ φ0 | + |χ α χ| . h The basis for N atoms is the set of functions Φ0 = Aφ0 (1)φ0 (2) · · · φ0 (N ) .2. . h (2. Φi = Aφ0 (1) · · · χ(i) · · · φ0 (N ) . (0) (0) (2. .).92) This zero-order Hamiltonian has the above basis functions as eigenfunctions: ˆ 0 |Φ0 = |Φ0 E (0) Φ0 |Φ0 + 0 = E (0) |Φ0 .7.91) (note that the sums exclude i = 0. Note that φ0 and χ are eigenfunctions of h respectively: ˆ 0 |φ0 = |φ0 ε0 φ0 |φ0 +|χ α χ|φ0 = ε0 |φ0 . .

96) We can proceed in one of two ways. . as follows. i<j + i |Φi β Φ0 | + i<j ∗ |Φij β Φi | + (2.97) 2. N ∆ε − α + ε0 N ∆ε − α + ε0 thus N →∞ lim E (2) = |β |2 = constant . In neither case have we obtained the extensive result. .g.. .36 Formal perturbation theory etc. E (2) → −|β | N . It is the fact that we have E rather than E (0) in the denominator that is responsible for the lack of extensivity.. We can use E (2) instead of E − E0 = ∆E (note that E (1) = V00 = 0 in this example): E (2) = thus E (2) = 1 2 (α − ε0 ) ± 1 2 N |β |2 . Φi |H l Φi |h(l)|Φjk = 0 (i = j. with the same limiting behavior as N → ∞. giving the second-order energy E (2) (0) = i |V0i |2 = N ∆ε − α + ε0 N i=1 |β |2 N |β |2 = .. (0) (2. . 2 (α − ε0 )2 + 4N |β |2 . or ∆E = E − E0 = N ε − N ε0 = N ∆ε. E (2) − α + ε0 (E (2) ) + (ε0 − α)E (2) − N |β |2 = 0 . (2. The denominators are E − Ei (0) (2. and all other indices are identical in the two Φ.. The only matrix elements we need for the second-order energy are ˆ |Φ i = β V0i = Φ0 |V (which are independent of N ).. | in V ket and bra diﬀers by exactly unity. · · · Φ.98) just √ as in CID.. e.95) = E − (E0 + α − ε0 ) = ∆E − α + ε0 . We can use our knowledge of the exact E = N ε.. functions. ∆ε (0) (2. E (2) ∝ N . 1. In all other cases the corresponding matrix elements ˆ ˆ |Φjk = vanish. i = k ).94) ˆ the number of excited atoms in the In all nonzero terms |Φ. The perturbation consists of the cross terms: ˆ = V i |Φ0 β Φi | + i<j |Φi β Φij | + i<j ∗ |Φj β Φij | |Φij β ∗ Φj | + .

because ˆ 0 ∆E |Φ = ∆E R ˆ 0 |Φ = 0 R (2. simplifying the notation further by (0) ˆ 0 (E (0) ) ≡ R ˆ 0 and omitting omitting the argument E0 from the resolvent R 0 the subscript 0 from Φ0 ..h.101 for each ∆E on the r. Schr¨ odinger 1926) by setting ζ = E0 so that ζ − E = −∆E .s.100) Ψ= m=0 ∞ ˆ 0 (E (0) )(V ˆ − ∆E ) R 0 m Φ0 (RSPT). (2. (2. This presence can have a true expansion in orders of λ (or orders of V be removed by substituting the formula in 2.78).h.4 General derivation of formal perturbation theories 37 2. Expanding the ﬁrst few orders for ∆E . since ∆E appears on the r.99) (2.4 Formal Rayleigh–Schr¨ odinger perturbation theory Returning to the general results (2.104) . We then have ∞ Ψ= m=0 ∞ ˆ 0 (V ˆ − ∆E ) R m Φ (RSPT). We then have ∞ (0) (RSPT). iteratively. we get ˆ − ∆E )|Φ ˆ |Φ + Φ|V ˆR ˆ 0 (V ∆E = Φ|V ˆR ˆ 0 (V ˆ − ∆E )R ˆ 0 (V ˆ − ∆E )|Φ + Φ|V ˆR ˆ 0 (V ˆ − ∆E )R ˆ 0 (V ˆ − ∆E )R ˆ 0 (V ˆ − ∆E )|Φ + · · · + Φ|V (2.101) m ∆E = m=0 ˆ R ˆ 0 (E (0) )(V ˆ − ∆E ) Φ0 |V 0 |Φ0 ˆ 0 . but we Thus we have removed the unknown E in the denominators in R ˆ still have ∆E in V − ∆E .103) First we note that the ﬁnal ∆E in each term can be ignored. (2.4. Because of the presence of ∆E we do not yet ˆ ).2. and collecting terms of each order of λ by counting the number ˆ factors in each term.77). in both equations.s. we will now derive the Rayleigh– Schr¨ odinger perturbation theory (RSPT) (Rayleigh 1894. of V We shall now proceed to do that.102) m ∆E = m=0 ˆ R ˆ 0 (V ˆ − ∆E ) Φ|V |Φ We will examine the energy expression ﬁrst. (2.

E (2) = Φ|V ˆ |Φ = Φ|V ˆR ˆ 0 (V ˆ |Φ .. and we have ˆR ˆ0V ˆ |Φ ˆR ˆ0W E (3) = Φ|V 2ˆ ˆR ˆ0W ˆR ˆ0V ˆ |Φ − E (2) Φ|V ˆR ˆ0W ˆR ˆ0 E (4) = Φ|V V |Φ .106) .38 Formal perturbation theory ˆ 0 contains Q ˆ ). (2.110) (2.107) ˆ − E (1) appears very frequently. in orders Arranging this result according to the number of V of λ) we have ˆ |Φ ˆ |Φ + Φ|V ˆR ˆ0V ∆E = Φ|V ˆR ˆ0V ˆR ˆ0V ˆ |Φ − Φ|V ˆ |Φ Φ|V ˆR ˆ 2V ˆ + Φ|V 0 |Φ 2ˆ ˆR ˆ0V ˆR ˆ0V ˆR ˆ0V ˆ |Φ − Φ|V ˆR ˆ0V ˆ |Φ Φ|V ˆR ˆ0 + Φ|V V |Φ ˆ |Φ Φ|V ˆR ˆ 0 (V ˆ )R ˆ0V ˆ |Φ ˆR ˆ0 + R ˆ0V − Φ|V 3ˆ ˆ |Φ 2 Φ|V ˆR ˆ0 + Φ|V V |Φ + · · · Thus ˆ |Φ = V00 . (2.e. Thus (since R ˆ |Φ + Φ|V ˆR ˆ 0 (V ˆ |Φ ˆ |Φ + Φ|V ˆR ˆ0V ˆ − ∆E )R ˆ0V ∆E = Φ|V ˆR ˆ 0 (V ˆ − ∆E )R ˆ 0 (V ˆ − ∆E )R ˆ0V ˆ |Φ + · · · + Φ|V ˆ |Φ + Φ|V ˆR ˆ0V ˆR ˆ0V ˆ |Φ ˆ |Φ + Φ|V ˆR ˆ0V = Φ|V ˆR ˆ 0 Φ|V ˆ |Φ R ˆ0V ˆR ˆ0V ˆ |Φ − Φ|V ˆR ˆ 0 Φ|V ˆ |Φ R ˆ0V ˆ |Φ − · · · − Φ|V ˆR ˆ0V ˆR ˆ0V ˆ |Φ ˆR ˆ0V + Φ|V ˆR ˆ0V ˆ |Φ R ˆ0V ˆ |Φ R ˆ0V ˆR ˆ 0 Φ|V ˆ |Φ − Φ|V ˆR ˆ 0 Φ|V ˆR ˆ0V ˆ |Φ − Φ|V ˆ |Φ + · · · ˆ |Φ R ˆ 0 Φ|V ˆ |Φ R ˆ0V ˆR ˆ 0 Φ|V (2.105) + Φ|V ˆ factors (i. and it is convenient to The combination V use a special symbol for it.109) (2. ˆ =V ˆ − E (1) .108) (2. W so that Wij = Vij − δij E (1) . ˆR ˆ 0 (V ˆ − Φ|V ˆ |Φ )R ˆ0V ˆ − E (1) )R ˆ0V E (3) = Φ|V ˆR ˆ 0 (V ˆ − E (1) )R ˆ 0 (V ˆ − E (1) )R ˆ0V ˆR ˆ 2V ˆ |Φ − E (2) Φ|V ˆ E (4) = Φ|V 0 |Φ . E (1) = Φ|V ˆR ˆ0V ˆ |Φ .

as shown at the end of this subsection. ˆR ˆ0R ˆ0V E (2) Φ|V (2. E (2) = D0i i E (3) = ij V0i Wij Vj 0 . for the wave function we obtain (the general form is followed by the diagonal-case formula in each equation): ˆ |Φ = ˆ0V Ψ(1) = R i |Φi Vi0 . If we now look back at E (4) we see that the renormalization term there can be written as ˆ0V ˆ |Φ = E (2) R ˆ Φ|R ˆ0V ˆ Φ = E (2) Ψ(1) |Ψ(1) . V0i Vi0 . since they are simply accommodated by a shift of ˆ . D0i D0j V0i Wij Wjk Vk0 − E (2) D0i D0j D0k V0i Vi0 2 . D0i |Φi Wij Vj 0 .2.112) E (4) = ijk i We see that each E (m) contains a principal term.111) for the denominators in the resolvent. D0i D0j D0k . . In the diagonal case we can make these equations more explicit.) origin for V Similarly. m − 1 factors R ˆ 0 and some renormalization terms . D0y D0z (2.113) ijk···yz ˆ. (It is convenient not to consider the E (1) terms in W powers R 0 as renormalization terms.4 General derivation of formal perturbation theories 39 etc. the various orders of the energy can be written in the diagonal case in the form: E (1) = V00 .115) .114) 2ˆ ˆR ˆ0W ˆR ˆ0V ˆ |Φ − Φ|V ˆR ˆ0V ˆ |Φ R ˆ0 ˆ0W V |Φ Ψ(3) = R Wij Wjk Vk0 Vi0 = |Φi − E (2) |Φi . Using the abbreviation D0i = E0 − Ei (0) (0) (2. D0 i (2. ˆR ˆ0W ˆR ˆ0W ˆ ···W ˆR ˆ0V ˆ |Φ = Φ|V V0i Wij Wjk · · · Wyz Vz 0 . with m factors V ( l ) these contain lower-order energies E (2 ≤ l ≤ m − 2) and factors involving ˆ ˆ k (k > 1). replacing V ˆ by W ˆ . . D0i D0j ˆR ˆ0V ˆ |Φ = ˆ0W Ψ(2) = R ij (2. D0i D0j D0k D0i 2 ijk i etc. W ˆ .

˜ =H 0 (2. E 0 i. as follows: ˆ =H ˆ0 + V ˆ H ˆ 0 + V00 ) + (V ˆ − V00 ) = (H ˆ +W ˆ .118) V00 = Φ0 |V 0 ˆ ˜ 0 is the new zero-order Hamiltonian and W ˆ is the perturIn this form H bation.5 The general (non-diagonal) case ˆ 0 is non-diagonal.40 Formal perturbation theory so that ˆR ˆ0W ˆR ˆ0V ˆ |Φ − E (2) Ψ(1) |Ψ(1) ˆR ˆ0W E (4) = Φ|V V0i Wij Wjk Vk0 − E (2) Ψ(1) |Ψ(1) (RSPT) .116) This is a well-known form for E (4) . with eigenvalues V00 + E (0) . (2. but it can be obtained by inversion of the matrix representation of ˆ (E (0) − H ˆ 0 )Q ˆ . the In the general case. Higher orders remain unchanged. W ˆ |Φ0 = E (1) . However. Note that E (1) = V00 could be eliminated completely from the analysis if ˆ H0 shifted by V00 . which are diﬀerences of eigenvalues. explicit inversion of that matrix is not actually Q 0 ˆ . W00 = 0 . If we ˆ0W required since the quantities needed are matrix elements of R write ˆ ˆ =R ˆ0W (2.119) . H ˆ =V ˆ − V00 .117) where ˆ ˜0 = H ˆ 0 + V00 . With this new splitting of the Hamiltonian we have ˆ 0 + V00 |Φ = E (0) + V00 ˜ (0) = Φ|H E 0 0 (which equals ESCF if Φ is the SCF wave function) and ˜ (1) = Φ|V ˆ − V00 |Φ = V00 − V00 = 0 ..e.121) (2.4. are unchanged. 2.122) X (2.120) (2. The factors D0i eigenfunctions of the new H i in the denominators. In the diagonal case the original zeroth-order functions Φi are still ˆ ˜ 0 . the ﬁrst-order energy vanishes. in which the representation of H ˆ resolvent R0 is not given explicitly in terms of the zeroth-order energies (0) Ei . = D0i D0j D0k ijk (2.

e. Paldus and C´ zek 1975).125) ˆ 0 is sparse (as often happens in actual If the matrix representation of H applications).123) ˆ can be obtained by the solution of and the matrix representation X of X the system of linear equations (E0 1 − H0 )X = W or. Note that only the k = 0 column of X is required for R case we have Wi0 = Vi0 (i = 0). Once X has been obtained. E (2) = Φ|V ˆX ˆ 2 |Φ . E (4) = Φ|V with similar equations for the wave function.128) (i.6 Bracketing procedure for RSPT If the process of deriving the terms in the series (2. the energy expressions can be determined from ˆX ˆ |Φ . Huby 1961. we ﬁrst simplify the notation further. Obviously.2. in which ˆ0V easy.4. explicitly.127) 2. writing ˆR ˆ0 · · · R ˆ ≡ Φ|V ˆR ˆ0V ˆR ˆ0 · · · R ˆ |Φ ˆ0V ˆ0V ˆR ˆ0V V (2.124) (2. Then the principal term for E (m) is ˆR ˆ0V ˆR ˆ0V ˆ ···R ˆ0V ˆ V . D0i (2. To show this result.106) for ∆E is continued. omitting the Φ| from the bra and |Φ from the ket). Xij = Wij (0) E0 (2. ˆ0 Q Q 0 (2. the iterative solution of the linear equation system is relatively ˆ |Φ . E0 δij − (H j (0) (0) (2. ˇ ıˇ a general result emerges (Brueckner 1955. in the diagonal case.126) − (0) Ei = Wij .4 General derivation of formal perturbation theories 41 then ˆ E (0) − H ˆX ˆ =Q ˆW ˆ . E (3) = Φ|V ˆX ˆ 3 |Φ − E (2) Ψ(1) |Ψ(1) . ˆ 0 )ij Xjk = Wik .

at fourth order we obtain ˆR ˆ0V ˆR ˆ0V ˆ ˆR ˆ0V E (4) = V ˆR ˆ0 V ˆR ˆ0V ˆ − V ˆR ˆ0V ˆR ˆ0 V ˆ ˆ R ˆ0V ˆ R ˆ0V − V ˆR ˆ0 V ˆ − V ˆR ˆ0 V ˆ R ˆ0V ˆ . always excluding the V ends but including multiple brackets and brackets within brackets (properly nested). n!(n − 1)! ˆ opWe can reduce the proliferation of terms substantially by using the W ˆ ˆ erator for all interior positions and omitting all single-V or single-W brackˆ that appears next to a bra or etings.42 Formal perturbation theory ˆ . W ˆ = ···V ˆ . ˆ R ˆ0 V ˆ R ˆ0V ˆR ˆ0V + V (2. Each term is given a sign factor (−1)l .131) For example. Thus. for the ﬁfth order contribution to the energy we get ˆR ˆ0W ˆR ˆ0W ˆR ˆ0V ˆ − V ˆR ˆ0 V ˆ R ˆ0W ˆR ˆ0V ˆ ˆR ˆ0W ˆR ˆ0V E (5) = V ˆR ˆ0W ˆR ˆ0V ˆR ˆ0W ˆR ˆ0 V ˆ R ˆ0V ˆ − V ˆR ˆ0 V ˆR ˆ0V ˆ R ˆ0V ˆ − V ˆ0W ˆ0V ˆR ˆ0W ˆR ˆ0W ˆR ˆ0W ˆR ˆ0V ˆ − V ˆR ˆ0V ˆ V ˆR ˆ 0 (R ˆ +W ˆR ˆ 0 )R ˆ = V 2ˆ ˆR ˆ0V ˆ V ˆR ˆ0 ˆR ˆ0W V − V ˆR ˆ0W ˆR ˆ0W ˆR ˆ0W ˆR ˆ0V ˆ − E (2) { Ψ(1) |Ψ(2) + Ψ(2) |Ψ(1) } = V − E (3) Ψ(1) |Ψ(1) . All the other terms are obtained by inserting and contains m operators V ˆ operators and around bracket pairs in all possible valid ways around V ˆ ˆ s at the two ˆ ˆ V R0 · · · V products in the principal term. note that we have here just four terms. (2.132) . where l is the number of inserted bracket pairs. We should also remember that any W ˆ: ket (on its inside) may then be replaced by V ˆ ··· = V ˆ ··· . (2.130) It can be shown that the total number of terms at the nth order is (2n − 2)! . ···W (2.129) At ﬁfth order there are already 14 terms (a principal term plus 13 bracketings) that include nested bracketings such as 2 ˆ ˆ ˆ2 ˆ ˆ R ˆ0V ˆ = V ˆ V ˆR ˆ0 ˆR ˆ0 V ˆR ˆ0 V ˆ R ˆ0V V V R0 V V = E (1) Ψ(1) |Ψ(1) 2 .

4 General derivation of formal perturbation theories 43 2.140) − E (3) i − E (2) ij V0i Wij Vj 0 D0i D0j 1 1 + D0i D0j V0i Vi0 . |Φi − E (2) D0i D0j D0k D0i 2 i j.7 Summary of formal RSPT results We shall now collect the results for several orders of RSPT. The general result (in terms of R ﬁrst three orders of the wave-function contributions are: ˆ |Φ = ˆ0V Ψ(1) = R i |Φi Vi0 .4. D0i V0i Wij Vj 0 .2.134) 2ˆ ˆR ˆ0W ˆR ˆ0V ˆ |Φ − Φ|V ˆR ˆ0V ˆ |Φ R ˆ0 ˆ0W V |Φ Ψ(3) = R W W V V ij jk k0 i0 = . E (1) = Φ|V ˆ |Φ = ˆR ˆ0V E (2) = Φ|V i (2.135) ˆ |Ψ(m−1) .133) ˆ0W ˆR ˆ0V ˆ |Φ = Ψ(2) = R i. D0i 2 .138) ˆR ˆ0W ˆR ˆ0W ˆR ˆ0V ˆ |Φ − E (2) Ψ(1) |Ψ(1) E (4) = Φ|V V0i Wij Wjk Vk0 V0i Vi0 − E (2) = D0i D0j D0k D0i 2 ijk i (2. D0i D0j (2.j |Φi (2.139) ˆR ˆ0W ˆR ˆ0W ˆR ˆ0V ˆ |Φ ˆR ˆ0W E (5) = Φ|V − E (2) Ψ(1) |Ψ(2) + Ψ(2) |Ψ(1) V0i Wij Wjk Wkl Vl0 = D0i D0j D0k D0l ijkl − E (3) Ψ(1) |Ψ(1) (2.k (2. and the ﬁrst six The energy formulas satisfy E (m) = Φ|V orders are: ˆ |Φ = V00 . D0i D0j ˆR ˆ0W ˆR ˆ0V ˆ |Φ = E (3) = Φ|V i. D0i Wij Vj 0 .136) (2.j (2.137) V0i Vi0 . giving both the ˆ 0 ) and the diagonal-case explicit formulas.

44 Formal perturbation theory E (6) ˆR ˆ0W ˆR ˆ0W ˆR ˆ0W ˆR ˆ0V ˆ |Φ ˆR ˆ0W = Φ|V − E (2) − E (3) Ψ(1) |Ψ(3) + Ψ(2) |Ψ(2) + Ψ(3) |Ψ(1) Ψ(1) |Ψ(2) + Ψ(2) |Ψ(1) − E (4) Ψ(1) |Ψ(1) ˆ 0 |Ψ(1) + (E (2) )2 Ψ(1) |R = ijklm V0i Wij Wjk Wkl Wlm Vm0 D0i D0j D0k D0l D0m V0i Wij Wjk Vk0 D0i D0j D0k V0i Wij Vj 0 D 0i D 0j 1 1 1 + + D0i D0j D0k (2. ˆ |Ψ(3) − E (2) E (6) = Ψ(2) |W − E (3) (2. D 0i 3 − E (4) i V0i Vi0 (2) 2 2 + (E ) D 0i i The steps in the derivation of the formula for E (6) are shown in Fig. In the case of the energies the most convenient forms are based on the Wigner formulas (2. = Ψ(1) |W Ψ(1) |Ψ(2) + Ψ(2) |Ψ(1) ˆ |Ψ(2) − E (2) E (5) = Ψ(2) |W − E (3) Ψ(1) |Ψ(1) . 2.144) (2.141) − E (2) ijk − E (3) ij 1 1 + D 0i D 0j V0i Vi0 .146) (2. (2.145) (2.36). (2. Ψ(2) = R ˆ |Ψ(2) − E (2) R ˆ 0 |Ψ(1) . =R ˆ |Ψ(1) . E (3) = Ψ(1) |W ˆ |Ψ(2) − E (2) Ψ(1) |Ψ(1) . The wave functions and energies at each order can also be expressed in terms of lower-order quantities by substitution of the formulas for the lowerorder wave functions and energies. E (2) = Φ|V E (4) (2.147) Ψ (4) ˆ |Ψ(1) . The corresponding results take the forms: ˆ0W ˆ |Ψ(1) .149) Ψ(1) |Ψ(2) + Ψ(2) |Ψ(1) . ˆ0W Ψ(3) = R ˆ0W ˆ |Ψ(3) − E (2) R ˆ 0 |Ψ(2) − E (3) R ˆ 0 |Ψ(1) .37).1.143) (2.142) (2.148) Ψ(1) |Ψ(3) + Ψ(2) |Ψ(2) − E (4) Ψ(1) |Ψ(1) .

unlike BWPT. Steps in the derivation of the formula for E (6) . (2.2. 2.8 Extensivity of Rayleigh–Schr¨ odinger perturbation theory Rayleigh–Schr¨ odinger perturbation theory (RSPT) is extensive order by order.3 for BWPT.151) . E0 − Ei (0) (0) = ε0 − α .4.150) by the corresponding RSPT denominator. but replace the BWPT denominator (2.1. 2. Fig. which has the inﬁnite-order energy in all denominators.4 General derivation of formal perturbation theories ˆR ˆ0W ˆR ˆ0V ˆR ˆ0W ˆR ˆ0W ˆR ˆ0W ˆR ˆ0V ˆ − V ˆR ˆ0 V ˆ R ˆ0W ˆR ˆ0W ˆR ˆ0V ˆ E (6) = V ˆR ˆ0 V ˆ R ˆ0W ˆR ˆ0V ˆ − V ˆR ˆ0W ˆR ˆ0W ˆR ˆ0 V ˆ R ˆ0V ˆ ˆR ˆ0V ˆR ˆ0V ˆR ˆ0W V ˆR ˆ0W ˆR ˆ0W ˆR ˆ0V ˆ R ˆ0W ˆR ˆ0V ˆ − V ˆR ˆ0W ˆR ˆ0 V ˆR ˆ0V ˆ R ˆ0V ˆ ˆR ˆ0 V V ˆR ˆ0W ˆR ˆ0V ˆR ˆ0V ˆR ˆ0W ˆR ˆ0V ˆ R ˆ0V ˆ + V ˆR ˆ0 V ˆ R ˆ0 V ˆ R ˆ0V ˆ ˆR ˆ0 V − V ˆ R ˆ0V ˆ R ˆ0V ˆ ˆR ˆ0 V ˆR ˆ0V ˆR ˆ0 V + V ˆR ˆ0W ˆR ˆ0W ˆR ˆ0W ˆR ˆ0V ˆ ˆR ˆ0W = V ˆ ˆR ˆ0V − V ˆ0W ˆ0W ˆR ˆ 0 |R ˆR ˆ0W ˆR ˆ0V ˆ + V ˆR ˆ0W ˆR ˆ 0 |R ˆR ˆ0V ˆ V ˆR ˆ0W ˆR ˆ0W ˆR ˆ 0 |R ˆ ˆ0V + V ˆ0W ˆ0V ˆR ˆ 0 |R ˆR ˆ0V ˆ + V ˆR ˆ0W ˆR ˆ 0 |R ˆ V 2 3ˆ ˆR ˆ0 V V 45 ˆR ˆ0V ˆ ˆR ˆ0W − V ˆ0V ˆR ˆ0W ˆR ˆ0V ˆ V ˆR ˆ 0 |R ˆ + V ˆR ˆ0V ˆ ˆR ˆ0W − V ˆ0V ˆ V ˆR ˆ 0 |R ˆ 2 ˆR ˆ0V + V ˆR ˆ0W ˆR ˆ0W ˆR ˆ0W ˆR ˆ0W ˆR ˆ0V ˆ = V − E (2) − E (3) ˆ 0 Ψ(1) Ψ(1) |Ψ(3) + E (2) R ˆ 0 Ψ(1) |Ψ(1) + Ψ(2) |Ψ(2) + Ψ(3) + E (2) R Ψ(1) |Ψ(2) + Ψ(2) |Ψ(1) − E (4) + E (2) Ψ(1) |Ψ(1) 2 Ψ(1) |Ψ(1) 2 ˆ 0 |Ψ(1) + E (2) Ψ(1) |Ψ(1) + (E (2) ) Ψ(1) |R . This is related to the fact that RSPT is a true order-by-order expansion.96) E − Ei (0) = ∆E − α + ε0 (2.4. If we consider again the example of N noninteracting He atoms and take the second-order energy expression obtained in subsection 2.

Hij = Φi |H (2. (2. we have ˆU ˆ |Φj = Φi |U ˆ |Ψj ˆ −1 H ˆ −1 H Φi |U ˆ −1 |Ψj Ej = Φi |U = Φi |Φj Ej = δij Ej .153) for the energy of a system made up of noninteracting subsystems A and B must hold in each order.152) which is proportional to N . this extensivity is due to the fact that in a true order-by-order expansion the exact relationship E (A + B) = E (A) + E (B) (2. ˆ is equivalent Finding the eigenvalues and eigenfunctions of an operator H ˆ to ﬁnding a basis {Ψi } in which H is diagonal: ˆ |Ψj = Ei δij . ε0 − α (2.156) The inverse transformation is ˆ −1 |Ψi .5 Similarity transformation derivation of the formal perturbation equations and quasidegenerate PT The following alternative derivation of the RSPT equations (Van Vleck 1929.155) (2.158) (2. 2.157) . A more complete demonstration of the extensivity of RSPT will await development of its many-body form but. Ψi |H ˆ in some other basis. |Φi = U so that. ˆ |Φi = |Ψ i = U j ˆ |Φ i = |Φj Φj |U j |Φj Uji .46 Formal perturbation theory the second-order energy becomes E (2) = N |β | 2 . If we have a representation of H ˆ |Φ j . if the {Φi } basis is orthonormal. basically. (2.154) ˆ that transforms the basis {Φi } then we need the transformation operator U into the basis {Ψi }. Shavitt and Redmon 1980) is of interest because it is most easily generalized to quasidegenerate PT (QDPT) and thus to a form of multireference PT (MRPT).

163) ˆ in the form and write U ˆ =P ˆU ˆP ˆ+P ˆU ˆQ ˆ+Q ˆU ˆP ˆ+Q ˆU ˆQ ˆ U ˆU ˆQ ˆ+Q ˆU ˆP ˆ+Q ˆU ˆQ ˆ.159) ˆ . If both which is a similarity transformation of the original Hamiltonian H the {Φi } and {Ψi } bases are orthonormal then the transformation connecting ˆ −1 . P ˆ=ˆ ˆ= Q 1−P i |Φi Φi | . i. where U00 = Φ0 |U ˆU ˆQ ˆ parts of U ˆ are arbitrary. ˆ −1 H Φ0 |U (i = 0) .e.3. so that enough to ﬁnd a transformation U ˆU ˆ |Φi = Φi |U ˆU ˆ |Φ0 = 0 ˆ −1 H ˆ −1 H Φ0 |U ˆU ˆ |Φ0 = E0 . (2. a known basis. ˆ |Ψj = U ˆ Φi |H ˆ |U ˆ Φj = Φi |U ˆ †H Ei δij = Ψi |H so that the eﬀective Hamiltonian can be expressed as ˆU ˆ.5 Similarity transformation derivation of PT and QDPT 47 ˆ as the elements of the (diagonal) Thus we can obtain the eigenvalues of H representation in the known orthonormal basis {Φi } of the eﬀective Hamiltonian ˆ eﬀ = U ˆU ˆ. Thus the problem of ﬁnding the eigenfunctions and eigenvalues ˆ is equivalent to that of ﬁnding an operator U ˆ such that the repreof H − 1 ˆ ˆ ˆ sentation of U H U in some given basis {Φi } is diagonal. To selves. = |Φ0 U00 Φ0 | + P (2.162) ˆ transforms the other Φi among themwithout being concerned about how U ˆ . ˆ = |Φ0 Φ0 | . If we are only interested in one particular eigensolution E0 . U be written in the form ˆU ˆ |Φj . ˆ eﬀ = U ˆ †H H (2.2. and the diagonalized representation of H ˆ can ˆ† = U them is unitary. (2. it is ˆ that block-diagonalizes H ˆ eﬀ .160) Note that the original basis is any basis we choose. ˆ and is while the transformed basis is the desired set of eigenfunctions of H unknown. ˆ −1 H H (2. since they do not aﬀect the The U00 and Q ˆU ˆQ ˆ block diagonalization (U00 changes only the normalization of Ψ0 .161) (2. If we succeed ˆU ˆ |Φi are the eigenvalues and U ˆ −1 H ˆ |Φi are the corresponding then Φi |U eigenfunctions.. |Ψ0 . while Q . In this case there is considerable arbitrariness in the choice of U be more speciﬁc we shall use the projection operators of Section 2.164) ˆ |Φ0 .

Then ˆ =ˆ ˆU ˆQ ˆ+Q ˆU ˆP ˆ U 1+P and ˆ Φ0 = U ˆP ˆ Φ0 = Φ0 + Q ˆU ˆP ˆ Φ0 . To proceed further. to achieve block diagonalization. If we wish U the minimum change (in the basis or in the eﬀective Hamiltonian) essential ˆU ˆP ˆ = P ˆ (i. Ψ0 = U (2.173) . U ˆ =U ˆD + U ˆX .172) H ˆ 0 is diagonal ( Φi |H ˆ 0 |Φj = E (0) δij ).165) (2. ˆ 0 = (H H ˆD = H ˆ0 + V ˆD . H ˆ 0 )X = 0 . ˆ. The operator that. (A more general deﬁnition relevant to quasiˆ part degenerate PT will be given at the end of this section. such that H the Hamiltonian as ˆ. ˆ =H ˆ0 + V (2.168) ˆ operates on any function Φ that is not orthogonal to Φ0 we obtain When ΩP a function proportional to Ψ0 .167) ˆU ˆP ˆ Φ0 is the perturbative correction to Φ0 (using intermediate so that Q normalization). eﬀ ˆX H =ˆ 0 (2.e. ˆ |Φ = U ˆP ˆ |Φ = U ˆ |Φ0 Φ0 |Φ = |Ψ0 Φ0 |Φ . U ˆD = 1 ˆ =ˆ ˆX . H (2. U U 1+U ˆX such that and so we are looking for U eﬀ ˆU ˆ =H ˆD ˆ eﬀ = U ˆ −1 H H .48 Formal perturbation theory ˆ to make merely mixes the other Ψi (i = 0) among themselves). In the persent case we have ˆ=U ˆP ˆ=P ˆ+Q ˆU ˆP ˆ. the ΩQ part being left undeﬁned.e. Q ˆU ˆP ˆ+Q ˆU ˆQ ˆ=ˆ P 1. (H ˆX = V ˆX .) Only the ΩP ˆ of Ω is of interest.171) ˆ eﬀ is block diagonal).166) (2.. when operating on a zero-order function Φ0 . and then set where H i ˆ 0 )D . ΩP ˆ can be written in the form Using the notation of (2. produces the corresponding perturbed function Ψ0 is called the wave operator and is commonly denoted by Ω. ΩP (2. so that U00 = 1).169) (2. it is natural to choose P ˆU ˆQ ˆ=Q ˆ .54).170) (2. we ﬁrst partition (i.

180) (2. we have ˆ 0. U X X ˆ =W ˆ (1) + W ˆ (2) + · · · W (2.181) (2. provides the desired energy corand P rection. U ˆ (1) = −V ˆX .) We have (This notation is unrelated to W ˆ =W ˆD .182) . (2. ˆ 0U ˆX H ˆ −V ˆU ˆX + W ˆ +U ˆX W H The diagonal part of this equation is ˆD − V ˆX + W ˆX U ˆ ˆ 0 = −V or ˆX U ˆX . U ˆ (m−1) + ˆ (m) = −V H X X m−1 l=1 ˆ (l) W ˆ (m−l) U X (m ≥ 2) .177) (2. W ˆ = ∆E0 P ˆ−P ˆH ˆ 0P ˆ = (E0 − E (0) )P ˆ. ˆ 0.174) ˆ =V ˆ − E (1) in previous sections. ˆ and W ˆ we rewrite Eq. (2. called the level-shift operator. H (2. U ˆX ] = −V ˆX − V ˆD U ˆX W [H ˆ.179) (2. H X ˆD U ˆ 0.159) for H ˆ eﬀ in the form In order to solve for U ˆH ˆ eﬀ = H ˆU ˆ.175) ˆW ˆP ˆ . If we expand in orders of V ˆX = U ˆ (1) + U ˆ (2) + · · · . we get Using U 1+U ˆ0 + W ˆ =H ˆ0 + H ˆX + V ˆ0 + U ˆX H ˆ +U ˆX W ˆ 0U ˆ +V ˆU ˆX . W ˆX = 0. H that is. U or ˆ +V ˆU ˆ. since they have been left out of U 1 eﬀ ˆ ˆ ˆ and W = H − H0 ).2.176) ˆX = U ˆ −ˆ (the zero-order parts are zero. ˆH ˆ0 + U ˆW ˆ =H ˆ 0U U ˆ =ˆ ˆX . ˆX − U ˆ 0 = −V ˆ.178) (2.5 Similarity transformation derivation of PT and QDPT 49 ˆ as the correction to H ˆ 0 needed It is also convenient to deﬁne an operator W ˆ eﬀ : to obtain H ˆ eﬀ = H ˆ0 + W ˆ . ˆW ˆP ˆ=P ˆH ˆ eﬀ P P 0 (2. ˆ =V ˆD + V W while the oﬀ-diagonal part is ˆX + U ˆ .

using W Equation (2. =R (2. U ˆ (m) = − V ˆ (m−1) + H X X m−2 l=1 ˆ (l) W ˆ (m−l) U X (m ≥ 2) (2. we proceed as follows: where A ˆA ˆP ˆ=Q ˆ [H ˆ 0. in the form ˆ (m) can be rewritten. U ˆ 0.188) Equating equal orders of the perturbation on both sides.186) ˆ 0 is the zero-order resolvent operator R ˆ 0 (E (0) ) for state 0. we can now write (2.184) (where the sum is empty for m = 2). = (Q 0 so that ˆU ˆP ˆ = −Q ˆ E (0) − Q ˆ ˆH ˆ 0Q Q 0 ˆ0A ˆP ˆ. i. = −R −1 (0) ˆA ˆP ˆ Q (2.183) in the form ˆU ˆ (1) P ˆ=R ˆ0V ˆP ˆ. U [H (2. Q ˆU ˆP ˆ−Q ˆU ˆP ˆH ˆ 0P ˆ. ˆ 0. U ˆX ]P ˆ. (2.183) ˆ (1) = V ˆD .182) and (2. where R 0 ˆ E (0) − Q ˆH ˆ 0Q ˆ0 = Q ˆ R 0 −1 ˆ= Q ˆ Q (0) E0 ˆ0 −H .189) .50 Formal perturbation theory and ˆD . ˆ (1) = V W ˆ (m−1) ˆ (m) = V ˆX U W X (m ≥ 2) .187) ˆX + U ˆ .180). (2. putting A have ˆU ˆP ˆ=R ˆ 0 (V ˆX − U ˆ )P ˆ ˆX + V ˆD U ˆX W Q ˆ0V ˆP ˆ+R ˆ0V ˆQ ˆU ˆP ˆ−R ˆ0U ˆP ˆW ˆP ˆ. ˆH ˆ 0Q =Q ˆH ˆ 0Q ˆ )(Q ˆU ˆP ˆ ) − (Q ˆU ˆP ˆ )E .185) ˆ is any operator expression satisfying A ˆ=A ˆX .e. In order to solve an equation of the form ˆX ] = A ˆ. Q ˆ=R ˆ0V Q ˆ− ˆU ˆ (m−1) P ˆU ˆ (m) P Q m−1 l=1 ˆ0U ˆ (l) P ˆW ˆ (m−l) P ˆ R (m ≥ 2) (2.182) for U X ˆD . we ˆ = −V ˆX − V ˆD U ˆX W Applying this result to (2.

in degenerate or quasidegenerate perturbation as part of the P theory (QDPT) . The PT equations then have the same form as before (but with the modiﬁed ˆ. ˆU ˆP ˆ and P ˆW ˆP ˆ determine the perturbed wave function These equations for Q through ˆU ˆP ˆ Φ0 Ψ0 = Φ0 + Q (2. ReP ˆU ˆP ˆ equations into the higher-order peated substitution of the lower-order Q ˆ and W ˆ equations gives the usual perturbation theory expressions. the wave operator is determined from (cf. ˆ=ˆ ˆ= Q 1−P i |Φi Φi | . (2. ˆ=P ˆ+R ˆ 0 (V ΩP (2. Using (2.e. ˆR ˆ0V E0 = Φ0 |V (2) (2. for U example.188). the transformation matrix Q . (2.194) The principal merit of this derivation is that it can easily be generalized to a case in which more than one zero-order states are to be handled together ˆ block. ˆH ˆ eﬀ P P (2.195) where the Greek subscripts sum over the model states (the quasidegenerate set of zero-order states) and the Roman subscripts over all others. ˆW ˆ (1) P P ˆ=P ˆV ˆQ ˆU ˆ (m−1) P ˆ ˆW ˆ (m) P P (2.191) and the perturbed energy through (2.175). ˆ=P ˆV ˆQ ˆU ˆ (1) P ˆ=P ˆV ˆR ˆ0V ˆP ˆ.196) ˆU ˆP ˆ and the wave operator ΩP ˆ=P ˆ+Q ˆU ˆP ˆ. In that case ˆ= P α |Φα Φα | .190) (m ≥ 2) . i.168)) ˆ +V ˆQ ˆU ˆ −U ˆP ˆW ˆ )P ˆ. from which deﬁnition of P we may obtain the (non-Hermitian) eﬀective Hamiltonian matrix ˆ=P ˆH ˆ 0P ˆ+P ˆW ˆP ˆ.192) ˆ part (corresponding to a column) of U ˆ (m) and the ˆU ˆ (m) P Only the Q X ˆW ˆ (m) P ˆ part (corresponding to an element) of W ˆ (m) are ever needed.2.5 Similarity transformation derivation of PT and QDPT 51 and ˆ=P ˆV ˆP ˆ. ˆW ˆ (2) P P which corresponds to ˆ |Φ 0 .193) (2. Q ˆ ) and they deﬁne a model-space matrix P ˆW ˆP ˆ .

200) we have i ˆ α |Φi . The QDPT approach and the generalized Bloch equations are discussed in detail in Chapter 8.201) In (2. (2.179). Lindgren 1978. The solutions are given by (compare Eqs. H ˆ 0 ] = (V ˆX + V ˆD U ˆX (V ˆD + V ˆX U [U ˆ=Q ˆ (V ˆ Ω)P ˆ−Q ˆ ΩP ˆ (V ˆ Ω)P ˆ. since P ˆ ΩP ˆ =P ˆ and P ˆ [Ω. H ˆ 0 ]P Q (2. in which the model states are exactly degenerate in zero order.202) .199) for all the model states. known as the generalized Bloch equation (Lindgren 1974.180): ˆX ) − U ˆX ) . H ˆ 0 ]P using ΩP we may add ˆ=P ˆV ˆ ΩP ˆ−P ˆ ΩP ˆV ˆ ΩP ˆ ˆ [Ω. While the PT equations (2. (2. (2.197) ˆ = 0. This equation is a generalization of the original Bloch equation (Bloch 1958.200) and Wαβ = Vαβ . Also. = Φi |R Rα (2. Kvasniˇ cka 1974.190)) i Uiα = Rα Viα . (2. 1977b. ˆ [Ω. Lindgren and Morrison 1986). H (2. is easily recovered from (2. Bloch and Horowitz 1958) of degenerate perturbation theory. Wαβ = i (m) (1) (m−1) Vαi Uiβ (m ≥ 2). their solution in the quasidegenerate case is complicated by the fact that instead ˆ 0 we now have a set of resolvents of a single resolvent R (0) ˆα = ˆ 0 (Eα )≡R R ˆ Q ˆ0 Eα − H (0) (2.52 Formal perturbation theory A more familiar form of the QDPT equations. H ˆ 0 ]P P to (2.197) to get the generalized Bloch equation in the form ˆ=V ˆ ΩP ˆ − ΩP ˆV ˆ ΩP ˆ ˆ 0 ]P [Ω. ˆ =U ˆP ˆ =P ˆ+Q ˆU ˆP ˆ .182). ˆX .189. 2. (m) Uiα (1) m−1 = j (m−1) i Rα Vij Ujα − l=1 β i Rα Uiβ Wβα (l) (m−l) (m ≥ 2) .198) (see Lindgren and Morrison 1986).183) have the same general form in the quasidegenerate and degenerate cases as in the nondegenerate case. (2. which is often the case for calculations on open-shell states.

. (2. This gives the ﬁnal energies Eγ and the transformation matrix C. Young and Sampanthar (1967)). when it operates on the manifold of model states. This variant was in fact used in the original derivation of the many-body form of perturbation theory (Brueckner 1955. Goldstone 1957). ˆ eﬀ Cβγ = Cαγ Eγ . Actually. where some important simpliﬁcations and cancellations will appear through the use of diagrammatic techniques.203) to complete the diagonalization of the P -space. generates the manifold of perturbed states.2. H αβ β (2. the ﬁnal solutions of the quasidegenerate (or degenerate) problem are determined by solving the non-Hermitian eigenvalue problem. but this derivation cannot be presented without the introduction of several additional concepts that complicate the presentation considerably and so it will not be discussed further in this book. Raimes (1972) or March.g.6 Other approaches At this point it is appropriate to begin consideration of the many-body form of RSPT. there is another important variant of the derivation of the formal equations of PT: this uses time-dependent PT (see e. converts the states generated by ΩΦα into the exact eigenstates of H 2.6 Other approaches 53 ˆU ˆP ˆ and P ˆW ˆP ˆ matrices have been determined to satisfactory When the Q accuracy. a diagonalization of the eﬀecˆ is then needed to obtain the transformation that ˆH ˆ eﬀ P tive Hamiltonian P ˆ.204) It is therefore clear that the wave operator can be more generally deﬁned as the operator that. The wave functions are then |Ψγ = β ˆ |Φβ Cβγ U |Φβ + β i = |Φi Uiβ Cβγ .

2. Kutzelnigg and Koch 1983). The totality of the antisymmetric Hilbert spaces for N = 1. Slater determinants) and operators in a compact convenient notation and also provide an eﬃcient way of manipulating such functions and operators. φi |φj = δij . . for which the basis functions are products of N one-particle basis functions. the ordinary form 54 . . To avoid complications.2) The number of functions in the basis need not appear explicitly in the formalism (but of course will aﬀect any computation). . say {φi } = {φ1 .3 Second quantization 3. (3. without explicitly specifying N .1 Background Second-quantization techniques evolved primarily for problems in which the number of particles is not ﬁxed or known and in the context of independentparticle models (such as the Hartree–Fock model). very often the Hartree– Fock spinorbitals. (3.1) In general. and we are looking for a representation of functions and operators in this Fock space. . since we are dealing with fermions. . the functions {φi } are spinorbitals. we assume that the basis is orthonormal . 1984. Obviously. . we will restrict the many-particle functions to be antisymmetric and thus assume a many-particle basis constructed of Slater determinants made up of the one-particle basis functions. Such techniques provide a method of representing independent-particle-model wave functions (i. particles. φ2 . 2. The second-quantization notation assumes the existence of an unspeciﬁed number of functions in a given ﬁxed one-particle basis . is often referred to as a Fock space (Kutzelnigg 1982. . The given one-particle basis generates Hilbert spaces for N = 1.} . .e. However. .

c i.4) are equivalent notations for the same function. (3.2 Creation and annihilation operators 55 of the Hamiltonian. of the basis spinorbitals φ1 . .z ≡ Aφi φj φk · · · φz ≡ |φi φj φk · · · φz ≡ | ijk · · · z .2. The Slater determinant (SD) Φ is represented in second-quantized form by specifying the occupancies (or occupation numbers ) n1 . X i ˆi .1 Deﬁnitions We begin by considering the representation of a normalized Slater determinant Φ = Φijk. does not satisfy this requirement because N appears in the summation limits. 2. . .2 Creation and annihilation operators 3. . . c ˆi . Here we shall use a ˆ† ˆi and.3. . φ2 . n2 . X ˆi . ˆ i† . Obviously. a They are deﬁned in terms of their action on SDs: a ˆ† i |jk · · · z = | ijk · · · z .3) ˆ µ and v ˆ and µ. N N ˆ = H µ=1 ˆµ + h µ<ν v ˆµν (3. . a ˆ† ˆ† creation operator for spinorbital φi . in the determinant. ν ˆµν are the one-particle and two-particle terms in H where h are particle labels. ˆ i.5) The determinant itself (and various operators on it) are represented in terms of a set of creation and annihilation operators. a annihilation operator for spinorbital φi . ni (Φ) = 0 if φi is empty (not present) in the SD Φ 1 if φi is occupied (present) in the SD Φ (i = 1. 3. (3. .. and we shall normally use one of the last two forms. The various forms of (3.) . (3. The notation for these varies: † ˆ †.4) where A is the antisymmetrizer and each φ is a spinorbital in our one-particle basis. . ˆ i. i.6) It is convenient to arrange the spinorbitals in an SD in lexical order as | ijk · · · z . a ˆi | ijk · · · z = |jk · · · z . when there is no possibility of confusion. ˆi. where i < j < k < ··· < z.. . i .

56

Second quantization

and therefore it is necessary to determine the eﬀects of the creation and annihilation operators when the aﬀected orbital is not (or is not being placed) at the beginning of the SD. Then, because

ˆ) ˆ | ijk · · · z = (−1)σ(P | ijk · · · z , P

(3.7)

**ˆ permutes the i, j, k, . . . , z and σ (P ˆ ) is the parity of the permutation where P ˆ , we have P
**

ηp a ˆ† p | ijk · · · z = (−1) | ijk · · · p · · · z ,

a ˆp | ijk · · · p · · · z = (−1)ηp | ijk · · · z ,

(3.8)

where ηp is the number of orbitals preceding φp in the SD (after the operaˆp ). We also have tion, for a ˆ† p , and before the operation, for a a ˆ† i |Φ = 0 a ˆi |Φ = 0 if ni (Φ) = 1 , if ni (Φ) = 0 . (3.9)

We can now describe an SD as the result of the successive operation of several a ˆ† p on the vacuum SD, | , which contains no spinorbitals (N = 0): ˆ† ˆ† ˆ† a ˆ† z | = | ijk · · · z . ia ja k ···a (3.10) ˆp between two SDs Φ and Next, consider the matrix elements of a ˆ† p and a Φ . It is clear (owing to the orthonormality of the spinorbitals) that Φ |a ˆp |Φ = ±1 if np (Φ) = 1, np (Φ ) = 0, ni (Φ) = ni (Φ ) (i = p) , 0 otherwise .

(3.11) That is, the matrix element is nonzero only if Φ and Φ have the same orbital occupancies except that φp has an electron in Φ but not in Φ (thus Φ has one fewer electron than Φ). The sign depends on the relative order of the spinorbitals in the two SDs. Let us order the orbitals the same way in Φ and Φ , except that in Φ we put φp at the beginning, Φ = |pij · · · z , Φ = | ij · · · z (3.12)

(at most, this introduces a factor of −1 in the matrix element). Then the matrix element is ij · · · z |a ˆp |pij · · · z = ij · · · z | ij · · · z = 1 . We get the same answer by operating with a ˆ† p on the bra: a ˆ† p ij · · · z |pij · · · z = pij · · · z |pij · · · z = 1 . (3.14) (3.13)

3.2 Creation and annihilation operators

57

If Φ and Φ do not satisfy the above occupancy conditions then both matrix elements will vanish. Therefore, in all cases, ˆp |Φ = a ˆ† Φ |a p Φ |Φ and similarly ˆ† ˆ p Φ |Φ . Φ |a p |Φ = a (3.16) This shows that a ˆ† ˆp and so justiﬁes the notation. p is the operator adjoint to a In general, we can write ˆ† ij · · · z |a ˆp = a p ij · · · z | = pij · · · z | , ˆp pij · · · z | = ij · · · z | , pij · · · z |a ˆ† p = a (3.17) (3.15)

so that a ˆ† ˆp act as annihilation and creation operators, respectively, p and a when acting to the left. ˆp is particularly interesting: The operator product a ˆ† pa a ˆ† ˆp |Φ = np (Φ)|Φ = pa |Φp 0 if φp is in Φ , otherwise . (3.18)

This combination is called the number operator for orbital (or spinorbital) φp , is written n ˆ p , so that ˆ† ˆp , n ˆp = a pa (3.19)

and has all the SDs as eigenfunctions, with eigenvalues equal to the corresponding occupation numbers (0 or 1). 3.2.2 Anticommutation relations Now we shall derive anticommutation relations for the creation and annihilation operators. Consider the following: ˆ† a ˆ† pa q |ijk · · · = |pqijk · · · ˆ† a ˆ† qa p |ijk · · · = |qpijk · · · = −|pqijk · · · . (3.20)

This is true for any SD |ijk · · · ; if φp or φq already exists in |ijk · · · , both products will give zero. Therefore, in general, a ˆ† ˆ† ˆ† ˆ† pa q = −a qa p, ˆ ˆ† ˆ† ˆ† ˆ† ˆ† [ˆ a† p, a q ]+ ≡ a pa q +a qa p = 0, (3.21)

ˆ B ˆ ]+ ≡ A ˆB ˆ +B ˆA ˆ is the anticommutator of A ˆ and B ˆ and is somewhere [A, ˆ ˆ times written as {A, B }. Note that if p = q the relation still holds, since

58

Second quantization

ˆ B ˆ ]+ = [B, ˆ A ˆ]+ , then a ˆ† ˆ† 0 and thus [ˆ a† ˆ† 0. Note also that [A, pa p = ˆ p, a p ]+ = ˆ ˆ B ˆ ] = −[B, ˆ A ˆ]. unlike the ordinary commutator, for which [A, Next consider the following: ˆq |qpij · · · = a ˆp |pij · · · = |ij · · · , a ˆp a ˆp |qpij · · · = −a ˆq a ˆp |pqij · · · = −a ˆq |qij · · · = −|ij · · · . a ˆq a (3.22)

If φp and/or φq are in the interior of the SD then they can always be brought forward to the beginning with, at most, a change of sign, the same change in both cases. Furthermore, if φp and/or φq are not present in the SD then both lines give zero. Thus, in general, a ˆp a ˆq = −a ˆq a ˆp or [ˆ ap , a ˆ q ]+ = ˆ 0. ˆq : Now consider a ˆ† p and a a ˆ† ˆq |qij · · · = a ˆ† pa p |ij · · · = |pij · · · . (3.25) (3.24) (3.23)

This operator product replaces φq by φp , even if φp was in the interior of the SD, since a ˆ† ˆq |ij · · · q · · · = (±1)2 |ij · · · p · · · = |ij · · · p · · · . pa However, (p = q ) (3.27) (this combination also replaces φq by φp , but with a change of sign), and thus ˆ ˆ q ]+ = 0 (p = q ). (3.28) [ˆ a† p, a If p = q we have a ˆ† ˆp |pij · · · = |pij · · · , pa ˆ† a ˆp a p |pij · · · = 0 , and if φp does not appear in the SD, a ˆ† ˆp |ij · · · = 0 , pa ˆ† ˆp |pij · · · = |ij · · · . a ˆp a p |ij · · · = a Thus in all cases ˆp + a ˆp a ˆ† a ˆ† pa p |··· = |··· , (3.31) (3.30) (3.29) a ˆq a ˆ† ˆq |pqij · · · = −a ˆq |qpij · · · = −|pij · · · p |qij · · · = a (3.26)

3.2 Creation and annihilation operators

59

and therefore ˆp a ˆ† p, a

+

= a ˆp , a ˆ† p

+

=ˆ 1.

(3.32)

To summarize, the anticommutation relations for the creation and annihilation operators are: [ˆ ap , a ˆ q ]+ = ˆ 0, ˆ ˆ† [ˆ a† p, a q ]+ = 0 , [ˆ a† ˆ q ]+ p, a = [ˆ ap , a ˆ† q ]+ ˆpq , =δ

(3.33)

ˆpq is the Kronecker-delta operator. (More generally, if the spinorwhere δ ˆq ]+ = p|q .) bitals are not orthonormal then one obtains [ˆ a† p, a

3.2.3 Representation of operators Next we want to consider the representation of general operators and matrix elements. As previously noted, any SD can be represented as the result of the operation of a string of creation operators on the vacuum state, ˆ† |ij · · · = a ˆ† ia j ···| , we also have (using the adjointness of a ˆ† ˆi ), i and a

† ij · · · | = a† ˆj a ˆi . i aj · · · | = | · · · a

(3.34)

(3.35)

Any annihilation operator operating on the vacuum state gives zero, a ˆi | = 0 , and similarly ˆi | = 0 . |a ˆ† i = a For consistency, the vacuum state is taken to be normalized, | = 1, as seen, for example, from ˆ† 1 = i|i = |a ˆi a i| = = = | [ˆ ai , a ˆ† ˆ† ˆi | i ]+ − a ia |ˆ 1| − |a ˆ† a ˆi |

i

(3.36)

(3.37)

(3.38)

| −0= | .

(3.39)

60

Second quantization

**In general, if we have two SDs ˆ† ˆ† i† |I = |i1 i2 · · · iN = ˆ 1 i2 · · · iN | ,
**

† ˆ† † j2 · · · ˆ |J = |j1 j2 · · · jN = ˆ j1 jN |

(3.40)

**i† for a ˆ† then, using the more compact notation ˆ i for a ˆi and ˆ i,
**

† ˆ† † i2ˆ jN | . I |J = |ˆ iN · · · ˆ i1 ˆ j1 j2 · · · ˆ

(3.41)

If we now move ˆ i1 all the way to the right (N transpositions), there are two possible outcomes: 1. No jp is the same as i1 , and therefore

† ˆ† †ˆ j1 j2 · · · ˆ i1 | (−1)N = 0 , i2 ˆ jN I |J = |ˆ iN · · · ˆ

(3.42)

**since ˆ i1 | = 0. 2. One of the j ’s, say jp , is the same as i1 , so that
**

† † †ˆ †ˆ †ˆ ˆi j − ˆ ˆ = [ˆ i1 , ˆ jp ]+ − ˆ jp jp 1−ˆ jp i1 ˆ jp i1 = δ i1 = ˆ i1 1 p

(i1 = jp ) (3.43)

and

† ˆ† † ˆ† p−1 ˆ† j1 j2 · · · ˆ i2ˆ jp −0 I |J = |ˆ iN · · · ˆ −1 jp+1 · · · jN | (−1)

(3.44)

(the last term is zero because we have an ˆ i1 left over that goes all the way to the right). i3 etc., we get either zero, if there are any If we next do this for ˆ i2 and then ˆ iq without a matching jp , or (−1)τ , if the two sets of indices i1 , i2 , . . . , iN and j1 , j2 , . . . , jN are identical except possibly for their order, in which case τ is the parity of the line-up permutation, which permutes, say, the j ’s to match the i’s. The foregoing provides a demonstration of how matrix elements involving creation and annihilation operators can be manipulated. An important element of such manipulations is commuting enough operators to bring all annihilation operators to the right of all creation operators, since ˆ† ˆp a ˆq | = 0 . |a ˆ† ia j ···a Now consider a symmetric one-electron operator

N

(3.45)

ˆ= F

µ=1

ˆ f µ,

(3.46)

ˆ ˆ where µ identiﬁes the electron on which f µ operates, and the fµ ’s are idenˆ tical except for the identity of the electron on which each operates (thus F

3.2 Creation and annihilation operators

61

operates in the same way on all electrons, which is why it is called a symˆ between metric one-electron operator). Now consider a matrix element of F two SDs: ˆ |j1 j2 · · · jN ˆ |J = i1 i2 · · · iN |F I |F =

µ

**ˆ i1 i2 · · · iN |f µ |j1 j2 · · · jN ˆ φi1 (1)φi2 (2) · · · φiN (N )|f µ
**

µ ˆ P

=

ˆ φj (1)φj (2) · · · φj (N ) , (−1)σ(P ) |P 1 2 N (3.47)

ˆ

where the bra and ket in the last line each comprise a speciﬁc spinorbital ˆ permutes the j ’s and σ (P ˆ ) is the product, not a determinant, and where P ˆ . The well-known result is that parity of P

N

ˆ |J = I |F

k=1

ˆ) ˆ|ik (−1)σ(P ik |f

ˆ if there exists a permutation P ˆ such that P j1 j2 · · · jN = i1 i2 · · · iN , ˆ such that if there exists a permutation P ˆ P j1 j2 · · · jN = i1 i2 · · · ik · · · iN (ik = ik ) (one noncoincidence) , otherwise (two or more noncoincidences) . (3.48)

ˆ) ˆ|i (−1)σ(P ˆ |J = ik |f I |F k

ˆ |J = 0 I |F

**It is easy to verify that the same result will be obtained for the operator ˆ|l a k |f ˆ† ˆl = ka
**

k,l k,l

fkl a ˆ† ˆl , ka

(3.49)

where the sums are over all the spinorbitals in the basis. For simplicity, let us assume that the two sets of indices i1 , i2 , . . . , iN and j1 , j2 , . . . , jN have been “lined-up” in such a way that ip = jp for as many p = 1, 2, . . . , N as possible. Then, if there are zero noncoincidences, I

k,l

fkl a ˆ† ˆl I ka

=

k,l

**fkl I |a ˆ† ˆl |I ka fkl δkl nl (I )
**

k,l N

=

=

k∈I

fkk =

p=1

ˆ|ip , ip |f

(3.50)

ˆl |I . For one since k = l creates a noncoincidence between I | and a ˆ† ka

62

Second quantization

**noncoincidence ip = jp , the only term in the k, l sum that will make I | ˆl |J will be l = jp , k = ip , so that coincide with a ˆ† ka I
**

k,l

fkl a ˆ† ˆl J ka

=

k,l

fkl I |a ˆ† ˆl |J ka

ˆ i† = fip jp I |ˆ p jp |J ˆ|jp . = fip jp = ip |f

(3.51)

ˆl that will For two or more noncoincidences there is no single pair a ˆ† ka † render I |a ˆk a ˆl |J = 0, so we get zero for the matrix element. Thus, all ˆ ˆ = N f matrix elements of F µ=1 µ within the Fock space generated by the one-electron basis are the same as those of we can write

N

ˆ† ˆl k,l fkl a ka

and within that space

ˆ= F

µ=1

ˆ f µ =

k,l

ˆ|l a k |f ˆ† ˆl . ka

(3.52)

**Similarly, for a symmetric two-electron operator
**

N N

ˆ= G

µ<ν

g ˆµν =

1 2 µ=ν

g ˆµν

we obtain ˆ= G

1 2 i,j,k,l

i(1)j (2)|g ˆ12 |k (1)l(2) a ˆ† ˆ† ˆl a ˆk . ia ja

(3.53)

Noting the order of the arguments (1, 2) in the matrix element and the order of the annihilation operators a ˆl a ˆk , we have ˆ† ˆl a ˆk = (ˆ a† a† ˆl )ˆ ak ) = (ˆ a† ˆk )(ˆ a† ˆl ) − δjk a ˆ† ˆl , a ˆ† ia ja i (ˆ ja ia ja ia (3.54)

using the anticommutation relations (3.33) with two transpositions. This form of the operators is independent of N . To show that this form for the two-body operator is correct, consider ﬁrst the zero-noncoincidence case (perfectly lined up), ˆ |I = I |G

1 2 ijkl

ij |g ˆ|kl I |a ˆ† ˆ† ˆl a ˆk |I . ia ja

(3.55)

3.2 Creation and annihilation operators

63

**The only nonzero contributions will be for l and k values that appear in |I , say k = ip , l = iq (p < q ): ij |g ˆ|ip iq I |a ˆ† ˆ† ˆip a ˆiq |i1 i2 · · · ip · · · iq · · · ia ja (3.56)
**

(p−1)+(q −2) . = ij |g ˆ|ip iq I |a ˆ† ˆ† ia j |i1 i2 · · · (−1)

ˆ† ˆ† ˆ† To get a nonzero result, a ˆ† ia ja ia j must restore ip and iq to their proper places; this can be done in two ways: 1. set i = ip , j = iq , resulting in ip iq |g ˆ|ip iq I |i1 i2 · · · ip · · · iq · · · (−1)(p−1)+(q−2) (−1)(p−1)+(q−2) = ip iq |g ˆ|ip iq ; 2. set i = iq , j = ip , resulting in iq ip |g ˆ|ip iq I |i1 i2 · · · ip · · · iq · · · (−1)(p−1)+(q−2) (−1)(p−1)+(q−1) = − iq ip |g ˆ|ip iq = − ip iq |g ˆ|iq ip . (3.58) (3.57)

This same pair of contributions will also be obtained for k = iq , l = ip . The total matrix element can be written in either of two forms ˆ |I = I |G =

i<j (i∈I,j ∈I ) 1 2 i∈I j ∈I

( ij |g ˆ|ij − ij |g ˆ|ji ) ( ij |g ˆ|ij − ij |g ˆ|ji ) . (3.59)

The case i = j does not contribute because of the cancellation of the direct and exchange terms in this case. It is convenient to deﬁne the antisymmetric integral ij |g ˆ|kl so that ˆ |I = I |G

1 2 i∈I j ∈I A

≡ ij |g ˆ|kl − ij |g ˆ|lk , ij |g ˆ|ij

(3.60)

A.

(3.61)

This is, of course, the same as the ordinary result (the Slater–Condon rules) obtained by ﬁrst-quantized techniques. Similarly, for a single noncoincidence ip = ip , i.e. for |I = |i1 i2 · · · ip · · · , |I = |i1 i2 · · · ip · · · , (3.62)

64

Second quantization

ˆ |I from a we get contributions to I |G ˆ† ˆ† ˆl a ˆk as follows: ia ja for for for for i = ip , k = ip , j = l = iq (q = p), i = ip , l = ip , j = k = iq (q = p), j = ip , l = ip , i = k = iq (q = p), j = ip , k = ip , i = l = iq (q = p), ip iq |ip iq ; − ip iq |iq ip ; iq ip |iq ip ; − iq ip |ip iq . (3.63)

Here the ﬁrst two terms are equal to the last two terms, respectively. So the total result is ˆ |I = 2 × 1 I |G ip j |g ˆ|ip j − ip j |g ˆ|jip 2

j ∈I

=

j ∈I

ip j |g ˆ|ip j

A

(single noncoincidence).

(3.64)

**Again, the terms in which j = ip or j = ip will cancel, since ˆ|ip ip ip ip |g ˆ|ip ip ip ip |g
**

A A

= ip ip |g ˆ|ip ip − ip ip |g ˆ|ip ip = 0 , = ip ip |g ˆ|ip ip − ip ip |g ˆ|ip ip = 0 ,

(3.65)

and the result agrees with the Slater–Condon rules for a single noncoincidence. For two noncoincidences jp = ip , jq = iq we get ˆ |I = i i |g I |G p q ˆ|ip iq

A

(two noncoincidences) ,

(3.66)

and for more than two noncoincidences we get zero, also in agreement with the Slater–Condon rules. The electronic Hamiltonian consists of one- and two-electron parts, ˆ =H ˆ1 + H ˆ2 , H with ˆ1 = H

µ

(3.67)

ˆµ , h

ˆ2 = H

µ<ν

v ˆµν ,

(3.68)

**where (using atomic units) ˆ µ = − 1 ∇2 − h 2 µ
**

A

ZA , rµA

v ˆµν =

1 . rµν

(3.69)

**The second-quantized form of the Hamiltonian is then ˆ =H ˆ1 + H ˆ2 = H
**

i,j

ˆ |j a i|h ˆ† ˆj + ia

1 2 ijkl

ij |v ˆ|kl a ˆ† ˆ† ˆl a ˆk , ia ja

(3.70)

3.2 Creation and annihilation operators

65

where ij |v ˆ|kl ≡ i(1)j (2)|v ˆ12 |k (1)l(2) . (3.71)

**The antisymmetric two-electron integral for v ˆ is often abbreviated as follows: ij |v ˆ|kl − ij |v ˆ|lk = ij |v ˆ|kl Since ij kl = − ij lk , ˆk = −a ˆk a ˆl , a ˆl a (3.73)
**

A

≡ ij kl .

(3.72)

**we can also write the second-quantized Hamiltonian in the form ˆ = H
**

ij

hij a ˆ† ˆj + ia

1 4 ijkl

ij kl a ˆ† ˆ† ˆl a ˆk . ia ja

(3.74)

This form is the one we shall ﬁnd most convenient to use in the subsequent development. The second-quantized forms of the operators are the second-quantization analogs of the forms ˆ= A

I,J

ˆ|J J | = |I I |A

I,J

|I AIJ J | ,

(3.75)

except that in the ﬁrst-quantized form the number of electrons appears through the number of spin orbitals in each of |I and |J , while the secondquantized form is independent of N .

3.2.4 Invariance under unitary transformations Next we wish to show the invariance of the second-quantized form of an operator under a unitary transformation of the basis. Let the transformation ˆ , so that be eﬀected by a unitary operator U ˆ |φi = |φi = U

j

ˆ |φi |φj φj |U (3.76)

=

j

|φj Uji ,

where ˆ |φi , Uji = φj |U ˆ −1 = U ˆ† , U ˆ −1 U

ij ∗ = Uji .

(3.77)

. Then the transformation can be written in the form |φ = |φ U . j ˆi a (3.80) (3. (3. with the transformation φj | = φ| = U φ | . |φ = |φ1 |φ2 · · · (3. . i Uji (3. φ | = U† φ| . the inverse transformation being |φ = |φ U−1 = |φ U† or |φj = i † |φi Uij = i ∗ |φi Uji . .79) (3. .82) φi | . (3.81) The corresponding bras are organized as column vectors φ1 | φ| = φ2 | = |φ † .78) and a matrix U with elements Uij .66 Second quantization Deﬁne the row vectors |φ = |φ1 |φ2 · · · .83) Now consider the one-electron operator ˆ = F kl ˆ|φl a φk |f ˆ† ˆl ka † ˆ|φ U † a Uki φi |f ˆl j jl ˆk a kl ij = = ij ˆ|φ φi |f j k a ˆ† k Uki l † Ujl a ˆl = ij † ˆ|φ a φi |f ˆj .84) . . φ1 | φ | = φ2 | = |φ † . .

ˆ C. Then the normal product (or normal-ordered product) of such operators. (note the consistency of the adjoint notation). .90) ˆ (3. is written n[A rearranged product of operators such that all creation operators are to the left of all annihilation operators with a phase factor corresponding to the parity of the permutation producing the rearrangement: ˆB ˆC ˆ · · · ] = (−1)σ(P ) a ˆ †ˆ ˆv ˆ.3 Normal products and Wick’s theorem 3. b† · · · u n[A where ˆ (A ˆB ˆC ˆ ···). a ˆ †ˆ b† · · · u ˆv ˆ=P (3. Then a† U ˆ a†=ˆ ˆ a = U† ˆ a or or ˆ a† = ˆ a † U† ˆ a = Uˆ a (as for the kets) . (3.3 Normal products and Wick’s theorem 67 where a ˆi † = k a ˆ† k Uki .86) These deﬁnitions of the transformed creation and annihilation operators are consistent with their usual deﬁnitions. (3. . ˆB ˆC ˆ · · · ] and is deﬁned as the relative to the physical vacuum. For example. ˆ .88) 3. respectively: a† ˆ† ˆ a† = (ˆ 1 a 2 ···). Let A. a ˆj = l † Ujl a ˆl . a ˆ1 a ˆ a = ˆ2 . . we can organize the creation and annihilation operators as row and column vectors.87) (as for the bras) . a ˆj † | · · · = |j · · · = i |i · · · Uij = i a ˆ† i | · · · Uij .85) Again. (3. . in particular of vacuum expectation values |A of products of creation and annihilation operators (see Harris. Monkhorst and Freeman 1992).89) .3. stand for various creation and annihilation operators.1 Normal products of operators The following deﬁnitions and analyses are designed to help in the evaluation ˆB ˆ ···| of matrix elements.3. ˆ B. (3.

(3. ˆ we deﬁne operators. n[ˆ aˆ b† ] = −ˆ b† a ˆ. thus all forms of a normal product are equivalent. ˆ .92) (3. ˆ B. . A (3. In order to be able to formulate this we need to deﬁne the contraction (or pairing) of ˆ B.95) .68 Second quantization ˆ being a permutation acting on the operators A. so that we have ˆ = H ij hij n[ˆ a† ˆj ] + ia 1 4 ijkl ij kl n[ˆ a† ˆ† ˆl a ˆk ] .93) The usefulness of the normal-product form is that its vacuum expectation value is zero: ˆB ˆ · · · ]| = 0 |n[A ˆB ˆ · · · ] is not empty .3. we have actually put them into normal-product form. since any rearrangement of the creation operators among themselves and/or the annihilation operators among themselves is permissible but would always be accompanied by an appropriate change in the phase factor. and σ (P ˆ ) its parP ity.91) and the δ -term is not present in the deﬁnition of the normal product. their contraction as ˆB ˆ≡A ˆB ˆ − n [A ˆB ˆ] . ˆab − ˆ a ˆˆ b† = [ˆ a. n[ˆ aˆ b† ] = −ˆ Note that in writing the second-quantized forms of quantum mechanical operators (including the Hamiltonian) in the previous section. . n[ˆ a† ˆ n[ˆ aˆ b] = a ˆˆ b = −ˆ ba ˆ. ia ja (3.2 Contractions In order to be able to compute expectation values of general operator strings. ˆ] = −a ˆ= c ˆ= a ˆˆ ˆˆ ˆ†ˆ ˆ †ˆ ˆ† d ˆ† d bc ˆ† d ˆ† c bd ˆ† a bd ˆ† c b = −c ˆ† a b. Examples are as follows: b] = a ˆ†ˆ b.94) 3. For a pair of creation or annihilation operators A. n[A ˆab in addition to the change tor a ˆ with a creation operator ˆ b† adds a term δ in sign. since permuting an annihilation operaIn general. ˆ b † ]+ − ˆ b† a ˆ=δ b† a ˆ. if [A (3. This deﬁnition is not unique. we will take advantage of Wick’s theorem (Wick 1950). b† a ˆ. n[ˆ a† ˆ ˆB ˆC ˆ ···] = A ˆB ˆC ˆ · · · .

the contraction of two operators is equal to a number (zero or one).g.3 Time-independent Wick’s theorem The time-independent form of Wick’s theorem states: A product of a string of creation and annihilation operators is equal to their normal product plus the sum of all possible normal products with contractions. Young and Sampanthar 1967). The result is obviously either zero or plus or minus the normal product with all the contracted pairs omitted. An outline of the proof is given in the next subsection.3. . ˆ b† ]+ = δab .97) (3. two pairs etc. e. a ˆ †ˆ ˆ †ˆ ˆ †ˆ a ˆˆ b=a ˆˆ b−a ˆˆ b = 0. March. all possible contractions of one pair. This expression is also written symbolically in the form ˆB ˆC ˆD ˆ · · · = n [A ˆB ˆC ˆD ˆ ···] + A ˆB ˆC ˆD ˆ ···]. Thus. it is essentially straightforward (see.99) the sum being over all possible contractions.3.3 Normal products and Wick’s theorem 69 Speciﬁcally. n[ A (3. A normal product with contractions is deﬁned as follows: ˆB ˆC ˆ ···R ˆ ···S ˆ···T ˆ···U ˆ · · · ] = (−1)σ R ˆT ˆS ˆU ˆ · · · n[A ˆB ˆC ˆ ···] n[A (3. a ˆˆ b† = a ˆˆ b† − (−ˆ b† a ˆ) = [ˆ a.98) Thus. 3. a ˆ †ˆ b=a ˆ †ˆ b−a ˆ †ˆ b = 0. ˆB ˆC ˆD ˆ ··· A ˆB ˆC ˆD ˆ · · · ] + n[A ˆB ˆC ˆD ˆ · · · ] + n[A ˆB ˆC ˆD ˆ ···] ˆB ˆC ˆD ˆ · · · ] + n [A = n[A ˆC ˆD ˆ · · · ] + n[A ˆB ˆC ˆD ˆ · · · ] + · · · + n[A ˆB ˆC ˆD ˆ ···] + ··· ˆB + · · · + n[A ˆB ˆC ˆD ˆ · · · ] + n [A ˆB ˆC ˆD ˆ · · · ] + n[ A ˆB ˆC ˆD ˆ ···] + ··· + n [A (3.96) where all the contracted pairs have been put in front of the normal product and σ is the parity of the permutation indicated by ˆB ˆC ˆ ···R ˆ ···S ˆ···T ˆ···U ˆ ··· A ˆT ˆS ˆU ˆ ··· A ˆB ˆC ˆ ··· R . While the complete proof is somewhat tedious. Symbolically. the four possibilities are: b† = a b† − a b† = 0 . are included.

4 Outline of proof of Wick’s theorem In a normal-ordered product p ˆ† q ˆ† · · · u ˆv ˆ all contractions vanish since in such ˆ† · · · contractions. if a string of ˆ· · · t a product there can be no · · · s . bc ˆ† d ˆ† f a†ˆ bc ˆ† d ˆ† f +n[ˆ a†ˆ (3. as is obvious from ˆe ˆ| = 0 . aˆ b† c ˆd ˆf aˆ b† c ˆd ˆf aˆ b† c ˆd ˆf + n[ˆ (3.102) A more complex example is given by: ˆ† e ˆ† =n[ˆ ˆ† e ˆ† ] + n[ˆ ˆ† e ˆ† ] + n[ˆ ˆ† e ˆ† ] + n[ˆ ˆ† e ˆ† ] ˆd ˆf aˆ b† c ˆd ˆf aˆ b† c ˆd ˆf aˆ b† c ˆd ˆf aˆ b† c ˆd ˆf a ˆˆ b† c ˆ† e ˆ† ] + n[ˆ ˆ† e ˆ† ] + n[ˆ ˆ† e ˆ† ] + n[ˆ ˆ† e ˆ† ] ˆd ˆf aˆ b† c ˆd ˆf aˆ b† c ˆd ˆf aˆ b† c ˆd ˆf + n[ˆ aˆ b† c ˆ† e ˆ† ] + n[ˆ ˆ† e ˆ† ] + n[ˆ ˆ† e ˆ† ] + n[ˆ ˆ† e ˆ† ] aˆ b† c ˆd ˆf aˆ b† c ˆd ˆf aˆ b† c ˆd ˆf aˆ b† c ˆd ˆf + n[ˆ ˆ† e ˆ† ] + n[ˆ ˆ† e ˆ† ] + n[ˆ ˆ† e ˆ† ] . ˆ···B ˆ ···C ˆ ···D ˆ ···| = |A ˆ···B ˆ ···C ˆ ···D ˆ · · · ]| .3. bc ˆ† d ˆ† f |a ˆ† ˆ =0 =0 (3. since they vanish. |n[A (3. its vacuum expectation value is zero.70 Second quantization The importance of the above result is that the vacuum expectation value of any normal product with contractions is zero unless all operators are contracted.101) where we have omitted all contractions except those of the form u ˆv ˆ† . The reason is that each contraction contributes a factor of zero or ±1 and. Since no fully contracted term survives. ˆd ˆf aˆ b† c ˆd ˆf |a ˆˆ b† c These examples demonstrate the power of Wick’s theorem. if an uncontracted normal product remains. (3.104) 3.100) where the sum is over all possible fully contracted normal products. For example. Thus. Thus. ˆe ˆ = n[ˆ ˆe ˆ] + n[ˆ ˆe ˆ] + n[ˆ ˆe ˆ] bc ˆ† d ˆ† f a†ˆ bc ˆ† d ˆ† f a†ˆ bc ˆ† d ˆ† f a†ˆ bc ˆ† d ˆ† f a ˆ †ˆ ˆe ˆ] + n[ˆ ˆe ˆ] . the vacuum expectation value of this operator product is zero.103) for which the vacuum expectation value is ˆ† e ˆ† | = |n[ˆ ˆ† e ˆ† ]| = δab δcd δef .

109) (3.4 Particle–hole formulation 3.107) again satisfying Wick’s theorem. (3.4 Particle–hole formulation 71 operators is already in normal product form we have p ˆ† q ˆ† · · · u ˆv ˆ = n[ˆ p† q ˆ† · · · u ˆv ˆ] = n[ˆ p† q ˆ† · · · u ˆv ˆ] + all contractions n[ˆ p† q ˆ† · · · u ˆv ˆ] . it is more convenient to begin with a ﬁxed reference state |0 ≡ |Φ0 = |ijk · · · n . (3.106) 3. s ˆ† ]+ − s ˆ† r ˆ ···u ˆv ˆ p ˆ† q =p ˆ† q ˆ† · · · δrs · · · u ˆv ˆ−p ˆ† q ˆ† · · · s ˆ† r ˆ· · · u ˆv ˆ ˆ† · · · r ˆs ˆ† · · · u ˆv ˆ] + n[ˆ p† q ˆ† · · · r ˆs ˆ† · · · u ˆv ˆ] .105) since all terms in the sum vanish. (3. = n[ˆ p† q (3. Now consider the case where we have two annihilation operators to the left of one of the creation operators: ˆ† · · · u ˆ† · · · u ˆ† s p ˆ† q ˆ† · · · r ˆs ˆt ˆv ˆ=p ˆ† q ˆ† · · · r ˆs ˆt ˆv ˆ−p ˆ† q ˆ† · · · r ˆt ˆ· · · u ˆv ˆ ˆ† · · · u ˆ† s ˆ† r ˆ† · · · r ˆs ˆt ˆv ˆ−p ˆ† q ˆ† · · · r ˆt ˆ· · · u ˆv ˆ+p ˆ† q ˆ† · · · t ˆs ˆ· · · u ˆv ˆ =p ˆ† q ˆ† · · · u ˆ† · · · u ˆ† · · · u ˆ† · · · r ˆs ˆt ˆv ˆ] + n[ˆ p† q ˆ† · · · r ˆs ˆt ˆv ˆ] + n[ˆ p† q ˆ† · · · r ˆs ˆt ˆv ˆ] . Thus Wick’s theorem holds in this case.4. = n[ˆ p† q All other contractions vanish. since all other contractions vanish. This procedure can be continued for all pairs of operators out of normal order.108) . Consider next the case where one pair of operators is out of normal order: ˆ† · · · r ˆs ˆ† · · · u ˆv ˆ=p ˆ† q ˆ† · · · [ˆ r. so Wick’s theorem still holds.3.1 The reference state Instead of referring all SDs and their matrix elements back to the vacuum state ˆ† ˆ† i† |I = |i1 i2 · · · iN = ˆ 1 i2 · · · iN | .

. while the other spinorbitals a. . λ. c. . γ. a ˆ |0 = 0 . b. (3. r. . . and other variations thereof. q. k. without restriction). are called particle states . the letters a. ˆ†ˆ |Φa i|Φ0 = |ajk · · · n i ≡a |Φab jˆ i|Φ0 = |abk · · · n ˆ†ˆ b† ˆ ij ≡ a i|Φ0 = |jk · · · n |Φi ≡ ˆ ˆ† |Φ0 = |aijk · · · n |Φa ≡ a etc. . for either. e. We shall use the letters i. . . . for either. (double excitation) . ba = a ˆ b† ˆi . the level of the highest occupied orbital in the reference function). p. for holes. . m. ν. µ.110) (electron removal) . q. j. in contrast with the physical vacuum | . to indicate indices restricted to hole states.g. . . . c.g. . . . n that are occupied in |0 are called hole states (they appear explicitly only when an electron is excited out of them by.113) ˆ Thus ˆ b† i creates a vacancy in state i while bi eliminates such a vacancy. b.72 Second quantization and deﬁne other SD’s relative to it. ˆ ˆa . n. for holes. (single excitation) .) The spinorbitals i. . r. .112) It is convenient to deﬁne a new set of operators. e. k. . respectively. . (electron attachment) (3. . Note also that ba ba ab |Φab ij = |Φji = −|Φij = −|Φji . to indicate indices restricted to particle states and the letters p. (Some sources use |0 for the physical vacuum and |Φ or |Φ0 for the Fermi vacuum. to indicate any state (either hole or particle. b. .111) The reference state |0 is also called the Fermi vacuum (corresponding to the Fermi level. The motivation for this notation is that all pseudo-annihilation operators . . 0|a ˆ† = 0 . . . (3. (3. β. . for particles. p. . . . r. q. 0|ˆ i = 0.) Using this notation we ﬁnd that ˆ i† |0 = 0 . creating a hole in the reference state). . (Other common notations are: α. via ˆ ˆ† bi = a i . . . for particles. while the hole pseudo-creation and pseudo-annihilation operators are equivalent to the ordinary hole annihilation and creation operators. ˆ i. i =a ˆ b† ˆ† a =a a. j. . The particle pseudo-operators are identical to the ordinary particle operators. a. . sometimes called pseudocreation and pseudo-annihilation operators (or quasi-operators).

in the last two rows (since lm nn = ll mn = 0). ˆ bp |0 = 0 . . (3.4 Particle–hole formulation 73 operating to the right on the Fermi vacuum state give zero and all pseudocreation operators operating to the left on the Fermi vacuum state also give zero. We may also write Eref in the more familiar form Eref = l hll + l<m lm lm . that the summations are over lm ˆ = −ˆ l† m Note that we have used ˆ l† m the hole states only and that the sum over l and m includes both l > m and l < m. The energy of the reference state can be computed as follows: ˆ |ijk · · · n Eref = ijk · · · n|H = pq ijk · · · n|p ˆ† q ˆ|ijk · · · n hpq +1 2 pqrs ijk · · · n|p ˆ† q ˆ† s ˆr ˆ|ijk · · · n pq |v ˆ|rs ijk · · · n|ˆ l† ˆ l|ijk · · · n hll = l +1 2 l =m ijk · · · n|ˆ l† m ˆ †m ˆˆ l|ijk · · · n lm|v ˆ|lm ijk · · · n|ˆ l† m ˆ †ˆ lm ˆ |ijk · · · n lm|v ˆ|ml l =m +1 2 = l hll + hll + l 1 2 lm 1 2 lm lm|v ˆ|lm − lm lm . 0|ˆ b† p = 0.115) = ˆ †ˆ ˆ †m ˆˆ l. as well as l = m. (3. (3.116) where the restriction l < m cancels the factor 1 2.114) Here we will continue to use the ordinary creation and annihilation operator notation relative to the physical vacuum but our particle and hole states will be deﬁned relative to the Fermi vacuum. 1 2 lm lm|v ˆ|ml .3.

b† = a a a (3. . A We then ﬁnd that the only nonzero contractions are ˆ i†ˆ j = δij . contractions and Wick’s theorem so that they relate to a reference state (the Fermi vacuum) instead of the physical vacuum. Contractions relative to the Fermi vacuum will be denoted by brackets above the operators instead of below. . . ba = a the product is in normal order if all the ˆ b† p operators are to the left of all the ˆ bp operators. Here we shall use the alternative notation {ABC · · · } for the normal product relative to the Fermi vacuum. it is often written as ˆ † ˆ† bu ˆ bp bq · · · ˆ bv . . . To distinguish the new type of normal product from the previous type. Since ˆ b p |0 = 0 . are to the left of all pseudo-annihilation operators operators a ˆ† .121) (3. . . N [ABC · · · ] = (−1)σ(P )ˆ (3. . and ˆ i† .119) instead of n[ABC · · · ] when the ordering is relative to the physical vacˆ ) of the phase factor is the parity of the permutation uum.. b† i =a † ˆ bi = a ˆi = ˆ i† .2 Normal products and Wick’s theorem relative to the Fermi vacuum Now we are in a position to modify the concepts of normal products.117) ˆ ˆa = a ˆ. and we have ˆB ˆ=A ˆB ˆ − {A ˆB ˆ} . A product of creation and/or annihilation operators is said to be in normal order relative to the Fermi vacuum |0 = |ijk · · · n if all pseudo-creation i. The power σ (P † ˆ ˆ b† from ABC · · · to ˆ bp ˆ q · · · bu bv .74 Second quantization 3.120) . . (3. .118) the Fermi-vacuum expectation value of a normal-ordered product of such operators vanishes. and ˆ a ˆ. (3. Using the notation ˆ ˆi = ˆ i. . ˆ ˆ† = a ˆ† . . 0|ˆ b† p = 0. a ˆˆ b† = δab .4.

. which we shall call orbital energies . ˆB ˆC ˆ ···R {A Here σ is the parity of the permutation. (3.5 Partitioning of the Hamiltonian The most convenient form for the zero-order part of the Hamiltonian is that of a diagonal one-electron operator. ˆB ˆC ˆD ˆ · · · = {A ˆB ˆC ˆD ˆ ···} + A all contractions ˆB ˆC ˆD ˆ ···}. . as given below: ˆB ˆC ˆ ···R ˆ ···S ˆ···T ˆ···U ˆ ··· A ˆT ˆS ˆU ˆ ··· A ˆB ˆC ˆ ··· R . but both choices strongly aﬀect the convergence of the perturbation series.}. ˆ0 = H p εp p ˆ† p ˆ. (3. p = 1. whenever we talk of the vacuum we will be referring to the Fermi vacuum and whenever we talk of normal products or contractions. unless explicitly stated otherwise.122) Wick’s theorem has exactly the same form in this system as in the physical vacuum system.125) In principle. 3.3.123) as indicated. {A (3. the sum is over all possible contractions of one pair. 2. p = 1. two pairs etc. 2. the choice of the energy parameters {εp . we are referring to these concepts relative to the Fermi vacuum. . . 0|A (3.124) where the sum is over all fully contracted normal products. is arbitrary and can be made independently of the choice of the set of orthonormal spinorbitals {φp . Obviously. so that ˆ···B ˆ ···C ˆ ···D ˆ · · · |0 = 0|A ˆ···B ˆ ···C ˆ ···D ˆ · · · }|0 . With the total Hamiltonian split . . . the usefulness of this theorem is at least partly due to the fact that the Fermi vacuum expectation value of a normal product vanishes unless it is fully contracted. From here on.} used in the construction of the Slater determinants.5 Partitioning of the Hamiltonian 75 A normal product with contractions is also deﬁned in the same way as in the case where it is relative to the physical vacuum: ˆ···S ˆ···T ˆ···U ˆ · · · } = (−1)σ R ˆT ˆS ˆU ˆ · · · {A ˆB ˆC ˆ ···}.

(3. u ˆ= i ˆi − K ˆi .126) Most commonly the orbital energies are chosen as the diagonal elements εp = fpp (3. the perturbation is given by ˆ 0) + H ˆ2 ˆ = (H ˆ1 − H V = pq (hpq − εp δpq )ˆ p† q ˆ+ 1 4 pqrs pq rs p ˆ† q ˆ† s ˆr ˆ.74). simplify the perturbation operator by canceling cerˆ 2 .131) ˆ could. Using the Coulomb and exchange operators deﬁned tain contributions to H in (1. in principle. (3. (3. be chosen arbitrarily. explicitly. fpq = p|f (3. ˆ |q = upq = p|u i pi qi .128) ˆ is deﬁned in terms of an auxiliary one-electron operator The Fock operator F N ˆ= U µ=1 u ˆµ = pq upq p ˆ† q ˆ (3.130) ˆ|q = hpq + upq .and second-quantized forms) N ˆ= F µ=1 ˆ f µ = pq fpq p ˆ† q ˆ.4).133) are over the hole-state spinorbitals that deﬁne the reference state |0 used as the zero-order function for the perturbation treatment.127) of a Fock operator (shown here in both ﬁrst. (3. J (3. we deﬁne u ˆ as in (1. at the same time.132) or.3). (3.76 Second quantization into a one-electron and a two-electron part according to (3. .129) by ˆ=H ˆ1 + U ˆ= F pq (hpq + upq )ˆ p† q ˆ= pq fpq p ˆ† q ˆ. it is generally chosen in While U a particular way designed to generate an appropriate zero-order Hamiltonian and.132).133) The sums in (3.67) and (3.

(3.138) ˆ is block diagonal in terms of the occupied In the noncanonical HF case. (3.137) pi pi . However.139) When the orbital energies are given by (3. this choice makes the F operator equal to the Fock operator of Hartree–Fock theory and makes the reference energy Eref .140) (3.141) (3. the operator F o ˆ ˆ F of F : ˆ=F ˆd + F ˆo .134) ˆ |q = fpq − εp δpq .127).127) holds. equal to the HF energy EHF . fpq = p|f the perturbation (3.142) ˆd fpp p ˆp ˆ= fpq p ˆq ˆ= p=q † ˆo = F . and if (3.5 Partitioning of the Hamiltonian 77 Deﬁning the operator ˆ = F pq fpq p ˆ† q ˆ. pq (3. (3. (3. Furthermore.135) (fpq − upq )ˆ p† q ˆ+ 1 2 pqrs pq |v ˆ|rs p ˆ† q ˆ† s ˆr ˆ. neither have we assumed that the orbital energies are equal to the diagonal elements of the Fock operator. (3.3. when a Hartree–Fock function is ˆ used as the reference function. in the canonical Hartree–Fock case this Fock operator is diagonal and is given completely in terms of the orbital energies fpq = εp δpq and εp = hpp + i (canonical HF) (3. (3.134).115). F and virtual spinorbital blocks: fia = 0 (noncanonical HF) . F F = p ˆ= f ˆd + f ˆo . f † d † fpq p ˆq ˆ. becomes equal to the oﬀ-diagonal part practice). (3.136) In these deﬁnitions we have not assumed that the spinorbitals are Hartree– Fock spinorbitals for the state being studied or for any other state.127) (this is the most common ˆ .126) becomes ˆ =F ˆ −U ˆ +H ˆ2 V = pq (3. pq o † fpq p ˆq ˆ.

. and the eigenvalues are sums of {i. . and the Hartree–Fock scheme is usually a good way of achieving this goal. fpq = p|f (3.144) pq |v ˆ|rs p ˆ† q ˆ† s ˆr ˆ.148) . .147) (for any number of electrons). ˆ1 = F ˆo − U ˆ= V pq o (fpq − upq )ˆ p† q ˆ. more generally. F the perturbation. the choices ˆ o = 0. If the reference state is |Φ0 = |jk · · · n (3. . ˆ2 = ˆ2 = H V 1 2 pqrs pq |v ˆ|rs p ˆ† q ˆ† s ˆr ˆ= 1 4 pqrs pq rs p ˆ† q ˆ† s ˆr ˆ. = p|f fpq o ˆo |q = (1 − δpq )fpq .} = {p} = ˆ 0 . it will generally be assumed that the orbital energies ˆ is determined by are given by (3. In the canonical Hartree–Fock representation.133). ˆ with certain contributions to the two-electron to the cancellation of −U o ˆ ˆ ) as the one-electron part of ˆ part V2 . . leaving just F (or. . leading venience of having a known complete set of eigenfunctions of H to the diagonal case of PT. Ultimately. all the one-electron part V F ˆ and F ˆ of the one-particle basis. as we shall see later.125) we ﬁnd that all possible Slater determinants made up of the spinorbitals {φ1 . .146) Unless stated otherwise. .145) The perturbation thus consists of one-electron and two-electron parts. a. (3. (3.133) will lead.} are eigenfunctions of H the corresponding orbital energies. The deﬁnition of u ˆ according to (3.127) and that the auxiliary operator U (3. Indeed. b. The use of a diagonal zero-order Hamiltonian is motivated by the conˆ 0 . the orbital energies and the operators U ˆ are motivated by the desire to make the perturbation V small. .143) In this case ˆd ˆ0 = F H and ˆ +H ˆ2 ˆ =F ˆo − U V = pq o (fpq − upq )ˆ p† q ˆ+ 1 2 pqrs (3. . in which ˆ1 is canceled. using (3. j.78 Second quantization where d ˆd |q = fpp δpq . ˆ 0 |rst · · · = H p εp p ˆ† p ˆ|rst · · · = (εr + εs + εt + · · · )|rst · · · (3. φ2 .

we have ˆ |Φ0 = Φ0 Φ0 |V pq o (fpq − upq )ˆ p† q ˆ+ 1 2 pqrs 1 2 ij pq |v ˆ|rs p ˆ† q ˆ† s ˆr ˆ Φ0 ij ij ij =− i uii + ij ij = − 1 2 (3. H 0 E0 = i (0) εi (3.152) o = 0).3. (3. (3.150).) For any other SD |Φab ij ··· we have ··· (0) ··· ˆ 0 |Φab + εa + εb + · · · − εi − εj − · · · |Φab H ij ··· = E ij ··· . (3. from (3. is often referred to .151) are the traces of matrix representations over the occupied spinorbital space and thus are invariant under unitary transformations of the occupied spinorbitals. In fact. we can also write the reference energy in the form Eref = 1 2 i (hii + εi ) .152) ··· have the additional contribution i fii .151) The various sums in (3. (3.154) The partitioning of the Hamiltonian based on an HF reference function and Fock operator. Eref = Φ0 |H (3. using (3.153) (In the more general case.150).150) = = i uii + 1 2 ij εi − ij ij .127) is not assumed.5 Partitioning of the Hamiltonian 79 then ˆ 0 |Φ0 = E (0) |Φ0 = (εj + εk + · · · + εn )|Φ0 .115).133). Eref = i hii + 1 2 ij ij ij 1 2 ij 1 2 ij = i (fii − uii ) + εi − i i ij ij ij ij (3. in which (3.115).150) is consistent with (since fii ˆ |Φ0 = Φ0 |H ˆ0 + V ˆ |Φ0 = E (0) + E (1) . This zero-order energy is not the same as Eref . Following the same analysis as in (3. and so we ﬁnd that (3.150)–(3.149) (the sum is over the hole states). particularly in the canonical HF case. Combining the ﬁrst and last line of (3.

1 One-electron operators Let us consider a one-electron operator ˆ= F pq ˆ|q p p|f ˆ† q ˆ. Binkley and Seeger 1976). Pople. 3. with eigenvalues Hii : H ˆ 0 |Φi = H j |Φj Hjj Φj |Φi = |Φi Hii . p ˆ† q (3. the Hilbert-space basis functions |Φi are eigenfunctions of this ˆ 0 . ˆ = {p ˆ† q ˆ} + p ˆ† q ˆ. Other partitionings of the Hamiltonian can also be used as a basis for an RSPT expansion. and Hamiltonian H00 − Hii . the diagonal part of the Hamiltonian in any convenient Hilbert-space representation can be used as the zero-order ˆ 0: Hamiltonian H ˆ0 = H i ˆ |Φi Φi | = |Φi Φi |H i |Φi Hii Φi | . In particular. (3. The Epstein–Nesbet perturbation expansion can also be obtained as a result of inﬁnite-order summations of certain classes of terms in the Møller–Plesset series though unlike the MP choice. (3. The eﬀect of these two types of partitioning on PT convergence was examined by Bartlett and Shavitt (1977b). In this case H the Hii are not expressible as sums of orbital energies. The RSPT expansion based on MP partitioning and Hartree–Fock orbitals has been referred to as Møller–Plesset perturbation theory (Binkley and Pople 1975. the EN expansion is not invariant under rotations among the occupied or the unoccupied orbitals.6 Normal-product form of the quantum-mechanical operators 3.155) ˆ =H ˆ −H ˆ0 = V i=j ˆ |Φj Φj | . (3.158) .80 Second quantization as Møller–Plesset (MP) partitioning (Møller and Plesset l934). Nesbet 1955) and leads to a perturbation expansion in which the denominators contain diﬀerences of diagonal matrix elements of the full ˆ 0 is not a one-electron operator.156) This type of partitioning is called Epstein–Nesbet (EN) partitioning (Epstein 1926.6. |Φi Φi |H Obviously.157) Using Wick’s theorem.

ˆN = F ˆ − 0|F ˆ |0 . fpq {p ˆ† q ˆ} = F (3. Note that F ˆN = F ij j} + fij {ˆ i† ˆ ab fab {a ˆ †ˆ b} + ia fia {ˆ i† a ˆ} + ia i} fai {a ˆ†ˆ (3. i|f where FN is the normal-product form of the operator (3. we have Since 0|F ˆ |0 = 0|F i ˆ|i . particle–particle and hole–particle terms. when it is equal to 1.164) ˆ o |0 = 0.159) ˆN + =F i ˆ|i . since 0|F 3.163) =− ij fij j i + ab ˆˆ† fab a ˆ b+ ia †ˆ i† a fiaˆ ˆ + ia i. ˆ= G 1 2 pqrs pq |g ˆ|rs p ˆ† q ˆ† s ˆr ˆ.161) ˆN represents the diﬀerence between F ˆ and its Fermi-vacuum exso that F pectation value.157). (3. F (3.2 Two-electron operators Next consider a two-electron operator. d ˆN = F p o ˆN = F p=q ˆ d − 0|F ˆ d |0 . i|f ˆ=F ˆN + 0|F ˆ |0 . fpp {p ˆ† p ˆ} = F ˆ o − 0|F ˆ o |0 = F ˆo. F (3.160) ˆN |0 = 0.162) ˆN contains hole–hole. ˆN = F pq ˆ|q {p p|f ˆ† q ˆ} . fai a ˆ†ˆ and can be separated into diagonal and oﬀ-diagonal parts: d o ˆN = F ˆN ˆN F +F . and thus ˆ = F pq ˆ|q p|f ˆ {p ˆ† q ˆ} + i ˆ|i i|f (3.165) .3.6. (3.6 Normal-product form of the quantum-mechanical operators 81 The contracted term vanishes unless p and q are the same hole state (call it i).

so ˆ = G 1 2 pqrs pq |g ˆ|rs {p ˆ† q ˆ† s ˆr ˆ} + ipq pi|g ˆ|qi {p ˆ† q ˆ} ij |g ˆ|ij − 1 2 ij 1 2 ij − ipq pi|g ˆ|iq {p ˆ† q ˆ} + pi|g ˆ|qi pq i 1 2 ij A ij |g ˆ|ji ij |g ˆ|ij A. (3. respectively.168) The second and fourth terms are equal to the third and ﬁfth. (3.82 Second quantization Noting that p ˆ† q ˆ† = p ˆq ˆ= 0. (3.167) pq |g ˆ|rs {p ˆ† q ˆ† s ˆr ˆ} + pi|g ˆ|qi {p ˆq ˆ} − ipq † 1 2 1 2 ipq ip|g ˆ|iq {p ˆ† q ˆ} ip|g ˆ|qi {p ˆ† q ˆ} +1 2 −1 2 ipq ipq pi|g ˆ|iq {p ˆq ˆ} + † 1 2 ij ij |g ˆ|ij − 1 2 ij ij |g ˆ|ji . and therefore ˆ = G 1 2 pqrs ˆ i†ˆ j = δij .171) ˆ is Since the Fermi-vacuum expectation value of G ˆ |0 = 0|G 1 2 ij ij |g ˆ|ij A. (3. a ˆ †ˆ b = 0.169) pq |g ˆ|rs {p ˆ† q ˆ† s ˆr ˆ} = 1 4 pqrs pq |g ˆ|rs ˆ† q ˆ† s ˆr ˆ} A {p (3.172) . thus removing the factor 1 2 .166) (3. ˆN + =G where ˆN = G and pq |g ˆ|rs A 1 2 pqrs {p ˆ† q ˆ} + (3.170) = pq |g ˆ|rs − pq |g ˆ|sr . we obtain ˆ† s ˆr ˆ = {p ˆ† q ˆ† s ˆr ˆ} + {p ˆ† q ˆ† s ˆr ˆ} + {p ˆ† q ˆ† s ˆr ˆ} p ˆ† q +{p ˆ† q ˆ† s ˆr ˆ} + {p ˆ† q ˆ† s ˆr ˆ} + {p ˆ† q ˆ† s ˆr ˆ} + {p ˆ† q ˆ† s ˆr ˆ} = {p ˆ† q ˆ† s ˆr ˆ} + p ˆ† r ˆ{q ˆ† s ˆ} + q ˆ† s ˆ{p ˆ† r ˆ} −p ˆ† s ˆ{q ˆ† r ˆ} − q ˆ† r ˆ{p ˆ† s ˆ} + p ˆ† r ˆq ˆ† s ˆ− p ˆ† s ˆq ˆ† r ˆ.

is ˆ2 = (V ˆ2 )N + V ˆN + 0|V ˆ2 |0 .173) ˆN + G ˆ + 0|G ˆ |0 . ij ij = − 0|U ij (3. (3.173). (3. For the zero-order part.128)) ˆ1 = (V ˆ1 )N + 0|V ˆ1 |0 . A ˆ} = {p ˆ† q pq gpq {p ˆ† q ˆ} .146).174).178) pi qi .175) The one-electron part of the perturbation is given by (see (3. V where ˆ2 )N = (V ˆN = V pq 1 4 pqrs (3.177) pq rs {p ˆ† q ˆ† s ˆr ˆ} . using (3. V o ˆ1 )N = F ˆN ˆN = (V −U pq ˆo − upq ){p (f ˆ† q ˆ} . (3. vpq {p ˆ† q ˆ} .174) gpq = 3. using (3. .125) and Wick’s theorem we have ˆ 0 − E (0) = ˆ 0 )N = H (H p εp p ˆ† p ˆ− i εi = p εp {p ˆ† p ˆ} . i vpq = ˆ2 |0 = 0|V 1 2 ij ij ij . pq ˆ |0 . as partitioned in Section 3. =G N where GN is a normal-product one-particle operator: ˆ = G N pq i pi|g ˆ|qi pi|g ˆ|qi i A.176) ˆ1 |0 = − 0|V i uii = − The two-electron part.3.3 The normal-product Hamiltonian Next we look at the Hamiltonian operator. (3.5. (3.6.6 Normal-product form of the quantum-mechanical operators 83 we have ˆ =G ˆN + G pq i pi|g ˆ|qi A ˆ |0 {p ˆ† q ˆ} + 0|G (3.

(3. the matrix elements of the shifted perturbation operator (3. is then found to be a (this is the operator denoted by W normal-product operator (relative to the Fermi vacuum): o ˆN = F ˆN ˆN V +W = p=q fpq {p ˆ† q ˆ} + 1 4 pqrs pq rs {p ˆ† q ˆ† s ˆr ˆ} . The shifted perturbation operator. ˆ N or simply W will be denoted by W ˆ =W ˆN = W 1 4 pqrs pq rs {p ˆ† q ˆ† s ˆr ˆ} .127) is not assumed. In the more general case.181) ˆ and −U ˆN . the operator W operator of Chapter 2. in Section 3. and ˆN will greatly facilitate the computation the normal-product property of V of its matrix elements.182) are of central importance in the perturbation expansion. ˆN +W =F (3.5). V made up of the two-particle normal-product operator only: ˆ = ˆN = W V 1 4 pqrs pq rs {p ˆ† q ˆ† s ˆr ˆ} (canonical HF).179) (this is the cancellation using the cancellation between V N alluded to before (3. in which the Fock operator is diagonal.180) is indeed the same as the W ˆ In this case.183) As seen in Chapter 2.185) . ˆN is In the canonical HF case.179) ˆ2 )N will play a very important role in the formalism.184) ˆ deﬁned in (3.84 Second quantization so that vpq = upq . (3. (3. V (3. we have (using fpq = fpq − εp δpq ) ˆN = F ˆN + W ˆN V = pq fpq {p ˆ† q ˆ} + 1 4 pqrs pq rs {p ˆ† q ˆ† s ˆr ˆ} (3.182) V ˆ in Chapter 2). (3.147). and The operator (V ˆ.180) The total perturbation is then given by o ˆN + 0|V ˆ1 |0 + W ˆN + V ˆ + 0|V ˆ2 |0 ˆ =F ˆN −U V N o ˆ N + 0|V ˆ |0 . ˆN = U ˆN . which we deﬁne as ˆ − 0|V ˆ |0 ˆN = V (3. in which (3.

Finally.190) (3. This is needed since we shall have to evaluate matrix elements of the . obtaining ˆ 0 |0 + V ˆ − 0|V ˆ |0 ˆ − 0|H ˆ |0 = H ˆ 0 − 0|H H ˆ 0 − E (0) + V ˆN . The Schr¨ odinger equation for this operator is ˆ N Ψ = ∆E Ψ . H and the computed energy is ∆E = E − Eref (3. leaving V ˆ. we can write the total normal-product Hamiltonian as ˆN + W ˆ =F ˆd + F ˆo + W ˆ =F ˆd + V ˆN .189) (the correlation energy in the Hartree–Fock case).4.191) Most of the discussion in this book is based on the normal-product Schr¨ odinger equation (3. the ﬁrst nonvanishing contribution is the second-order energy. ˆ =H ˆ0 + V gously to the last paragraph of subsection 2. V ˆN is the ˆ 0 − E (0) = (H In this scheme H ˆ ˆ perturbation and H − Eref = HN is the full Hamiltonian and is a normalproduct operator. Starting with H we subtract ˆ |0 = 0|H ˆ 0 |0 + 0|V ˆ |0 0|H from both sides. (3. (3.3. ˆ − Eref = H H . analoﬁrst-order energy correction. ˆN = F H N N (3.7 Generalized time-independent Wick’s theorem To complete this phase of the analysis we need one more theorem.188) (3.4. We also ﬁnd that the zeroand ﬁrst-order energies vanish in this form since they are vacuum expectation values of normal-product operators. the generalized Wick’s theorem dealing with products of normal products of operators.187) ˆ 0 )N is the zero-order Hamiltonian.191) of the normalproduct Hamiltonian. 3.7 Generalized time-independent Wick’s theorem 85 and ˆ |0 = 0|V i fii − 1 2 ij ij ij .186) It is also possible to rewrite the Hamiltonian in a form that eliminates the ˆN as the total perturbation.189) and the decompositions (3.

etc. {ˆ i† ˆ ˆ† e ˆ† e ˆ† · · · m ˆˆ l} = d ˆ† · · · m ˆˆ l.. since j† · · · ˆ j† · · · ˆ ba ˆ} = ˆ i†ˆ ba ˆ. B ˆi B ˆj . {d (3.195) . since contractions of pairs of operators that are already in normal order vanish (as seen from the deﬁnition (3. no contractions between pairs of operators within the same original normal product need be included: ˆ2 · · · }{B ˆ2 · · · }{C ˆ2 · · · } · · · ˆ1 B ˆ1 C ˆ1 A {A ˆ2 · · · B ˆ2 · · · C ˆ2 · · · } + ˆ1 A ˆ1 B ˆ1 C = {A ˆ2 · · · B ˆ1 B ˆ2 · · · C ˆ2 · · · } .194) where the sum is over contractions of one pair at a time. such as A Note that the case in which the original product contains some individual creation or annihilation operators not within any normal product is also included in the scope of the generalized Wick’s theorem. (3. b ˆ ˆNˆ ˆ† ˆF b† ˆ j |0 Φa i |FN |Φj = 0|i a = pq ˆ|q 0|{ˆ p|f i† a ˆ}{p ˆ† q ˆ}{ˆ b† ˆ j }|0 . and the prime on the summation sign indicates that no “internal” contractions.192) Here we have a vacuum expectation value of a product of three operator strings.8 Evaluation of matrix elements As a ﬁrst example we shall use the generalized Wick’s theorem to evaluate a matrix element of a normal-product one-electron operator between two singly excited determinants. as for example in ··· ˆ de··· ˆˆW ˆ† e ˆd ˆ†ˆ† ˆ† · · · m ˆˆ l|0 .193) The generalized Wick’s theorem states that a general product of creation and annihilation operators in which some operator strings are already in normal-product form is given as the overall normal product of all the creation and annihilation operators plus the sum of all overall normal products with contractions except that. each of which separately is in normal-product form. ˆi A ˆj . are to be included. erators A 3. Φab ij ··· |W |Φlm··· = 0|i j · · · ba (3. since for such opˆ = {A ˆ}. etc.120) of a contraction). two pairs. ˆ1 A ˆ1 C {A (3.86 Second quantization ˆ between various Slater determinants (not just normal-product operator W the reference SD)..

197) Speciﬁcally. a ˆˆ b† = δab . ˆ b† ˆ contractions are of the type ˆ i†ˆ j = δij . (3.200) .198) The last case leads to the obvious result a ˆ a ˆ a ˆ Φa i |F |Φi = 0|F |0 + Φi |FN |Φi = j ˆ|j + a|f ˆ|a − i|f ˆ|i = j |f j =i ˆ|j + a|f ˆ|a . a ˆ ˆ Φa i |FN |Φj = − j |f |i b ˆ ˆ Φa i |FN |Φi = a|f |b a ˆ ˆ ˆ Φa i |FN |Φi = a|f |a − i|f |i . In this case the only non-zero where we have used ˆ i† a ˆ = {ˆ i† a ˆ} . and ˆ i† a j . ˆ b ˆˆ ˆ† ˆW b† ˆ Φa j |0 i |W |Φj = 0|i a = 1 2 pqrs j }|0 . one of ˆ i† and a 2. the only ways in which such terms are obtained through contractions between diﬀerent normal products are as follows: ˆ with one of q ˆ.196) q ˆˆ b† = δqb . respectively. Noting that only fully contracted terms survive in the (Fermi) vacuum expectation value. giving b ˆ ˆ ˆ Φa i |FN |Φj = − j |f |i δab + a|f |b δij . 1. p ˆ† ˆ (3. ˆp ˆ† q ˆˆ b† ˆ ˆp ˆ† q ˆˆ b† ˆ The only two possibilities then are ˆ i† a j. p ˆ† . (a = b) .199) j |f ˆ between As another example we shall evaluate the matrix element of W the same two conﬁgurations. (i = j ) . ˆ i† q ˆ = δiq . and † 3. j = δpj . the remaining a ˆ or ˆ i† with ˆ b† or ˆ j . pq |v ˆ|rs 0|{ˆ i† a ˆ}{p ˆ† q ˆ† s ˆr ˆ}{ˆ b† ˆ (3. (3. the remaining q ˆ or p ˆ with the remaining ˆ b† or ˆ j. (3. a ˆp ˆ† = δap .8 Evaluation of matrix elements 87 j = {ˆ b† ˆ j }.3.

202) The only way in which we can fully contract the operators in the vacuum matrix element without internal contractions inside each of the three normal products is to contract the two operators in the ﬁrst normal product with one operator each in the second and third products respectively. but for any particular contracted product the number of intersections is either always even. Obviously x is not unique in general. p ˆ† and q ˆ† with k 1. r ˆ or s ˆ with c ˆ† or ˆ b† . we must contract two of the p ˆ† q each of the pairs to the left and right. The possibilities are: ˆ ˆ or r ˆ. or always odd. ˆ or ˆ ˆ with q j. since the contraction lines can be drawn with diﬀerent intersection patterns. . (3. and introduce three contractions between the latter two products. giving a sign factor of +1.201) = aj |v ˆ|ib − aj |v ˆ|bi = aj ib . ˆ i† with k † and c ˆ. ˆ j or k ˆ† or p ˆ† . giving a factor of −1. It can be veriﬁed that the sign factor for a fully contracted product of creation and annihilation operators is (−1)x . The options for non-zero contractions then are: i† a j |0 + 0|ˆ i† a j |0 0|ˆ ˆp ˆ† q ˆ† s ˆr ˆˆ b† ˆ ˆp ˆ† q ˆ† s ˆr ˆˆ b† ˆ + 0|ˆ ˆp ˆ† q ˆ† s ˆr ˆˆ b† ˆ ˆp ˆ† q ˆ† s ˆr ˆˆ b† ˆ i† a j |0 + 0|ˆ i† a j |0 = −1 ˆ|bi + 2 aj |v 1 2 Noting again that the only non-zero contractions are of the types ˆ i† ˆ j = ˆˆ b† = δab and that only fully contracted terms survive in the (Fermi) δij and a ˆ b Φa i |W |Φj = 1 2 pqrs pq |v ˆ|rs aj |v ˆ|ib + 1 2 ja|v ˆ|bi − 1 2 ja|v ˆ|ib (3.88 Second quantization ˆ† s ˆr ˆ operators with vacuum expectation value. a ˆ with ˆ b† or c ˆ† . A more diﬃcult example is the matrix element between a single excitation and a double excitation: ˆ bc Φa i |W |Φjk = 1 2 pqrs ˆˆ pq |v ˆ|rs 0|{ˆ i† a ˆ}{p ˆ† q ˆ† s ˆr ˆ}{ˆ b† c ˆ† k j }|0 . s ˆ and r ˆ with ˆ b† 2. where x is the number of intersections of the contraction lines. a ˆ with p ˆ† or q ˆ† . ˆ i† with s and ˆ j.

(a = b. we need some help with the detailed evaluation of these formulas.205) Obviously.8 Evaluation of matrix elements 89 Thus we get ˆ bc Φa i |W |Φjk = 1 2 pqrs pq |v ˆ|rs ˆˆ ˆˆ i† a i† a 0|ˆ ˆp ˆ† q ˆ† s ˆr ˆˆ b† c ˆ† k ˆp ˆ† q ˆ† s ˆr ˆˆ b† c ˆ† k j |0 + 0|ˆ j |0 ˆˆ ˆˆ + 0|ˆ ˆp ˆ† q ˆ† s ˆr ˆˆ b† c ˆ† k ˆp ˆ† q ˆ† s ˆr ˆˆ b† c ˆ† k j |0 + 0|ˆ j |0 i† a i† a ˆˆ ˆˆ ˆp ˆ† q ˆ† s ˆr ˆˆ b† c ˆ† k ˆp ˆ† q ˆ† s ˆr ˆˆ b† c ˆ† k j |0 + 0|ˆ j |0 + 0|ˆ i† a i† a ˆˆ ˆˆ i† a i† a ˆp ˆ† q ˆ† s ˆr ˆˆ b† c ˆ† k ˆp ˆ† q ˆ† s ˆr ˆˆ b† c ˆ† k j |0 + 0|ˆ j |0 + 0|ˆ + eight more terms from possibility 2 . for example ˆ bc Φa i |W |Φik = ak bc ˆ ac Φa i |W |Φjk = − jk ic ˆ ac Φa i |W |Φik = ak ac − ik ic .204) Thus we can get a non-zero result only if at least one hole or particle in Φbc jk matches the hole or particle in Φa i . (i = j. and this help is provided by the diagrammatic representation described in the following chapter. (3. (3. (3.203) from which we obtain 1 ˆ bc ˆ|ci δab + jk |v Φa ˆ|ci δab + kj |v ˆ|ic δab − jk |v ˆ|ic δab i |W |Φjk = 2 − kj |v + kj |v ˆ|bi δac − jk |v ˆ|bi δac − kj |v ˆ|ib δac + jk |v ˆ|ib δac + aj |v ˆ|cb δik − aj |v ˆ|bc δik − ja|v ˆ|cb δik + ja|v ˆ|bc δik − ak |v ˆ|cb δij + ak|v ˆ|bc δij + ka|v ˆ|cb δij − ka|v ˆ|bc δij = − jk ic δab − jk bi δac + aj cb δik + ak bc δij . k ) . . c) .3.

4 Diagrammatic notation 4. and this is indicated in the diagrams by means of a time axis t 90 . Initial applications to RSPT were also in time-dependent form. to provide certain systematics for the discussion and manipulation of the various surviving terms. in the form of Feynman diagrams or graphs. The purposes of the diagrammatic notation are thus: 1. It leads to a time sequence in the application of various operators. the second-quantization treatment can be cumbersome and error-prone. 3.8. to elucidate certain cancellations in these sums.1 Time ordering As we saw in Section 3. perform some partial summations etc. The generation and manipulation of the various expressions in this treatment are made much easier by the introduction of a systematic diagrammatic representation. Diagrammatic notation originated in quantum ﬁeld theory. but in them the time dependence was introduced artiﬁcially by a gradual switching-on of the perturbation from t = −∞ to t = 0 using a switching-on function eαt (with ˆ and varies from ˆ =H ˆ 0 + λV α > 0). in an explicit time-dependent format. 2. This function takes the place of λ in H 0 as t → −∞ to 1 as t → 0. to make it easy to list all non vanishing distinct terms in the perturbation sums. including classiﬁcations that allow us to group together various types of contribution.

with labels identifying the spinorbitals: Φa i = i a Φab ij = i a j b (the horizontal position of the lines has no signiﬁcance). from right to left. The only diﬀerence is a 90◦ rotation of the diagrams. say Φa i . for ˇ ıˇ example. the following (non-standard) notation may be used: ˆ† |0 = |Φa = a a |Φi = ˆ i|0 = i . but the sequence in which operators act (i. by Paldus and C´ zek (1975). t corresponding to the way in which we normally write a sequence of operators acting on a function.4.e. from right to left) still furnishes us with a time axis.e. by a position on the time axis at which there are no lines or other symbols. only the sequence is signiﬁcant. Thus.2 Slater determinants We begin with the representation of a Slater determinant (SD). pointing upward for particles and downward for holes. If we want to indicate speciﬁcally the ket and bra forms. The reference state (the Fermi vacuum) is represented by nothing. The actual time at which each event occurs (i. This was used. but we shall use the more common vertical arrangement. is represented by vertical or diagonal (see the next section) directed lines. if we want to represent the result of the operation of an ˆ on a function |Φab . we begin with a representation of Φab at the operator U ij ij ˆ bottom. We shall continue to use the time-independent formulation. i. an operator acts) is irrelevant. Any other SD.2 Slater determinants 91 for the sequence of events. 4. leading to ˆ |Φab at the top. as mentioned above. a representation of the result U ij Another common arrangement is to place the time axis horizontally. followed by a representation of the operator U above it.e.

(4. The horizontal double line represents the point of operation of the normal-product operator. However. |Φa ˆ†ˆ i = {a (4.3. and below or above it we have the Fermi vacuum. q in (4. (4. b|u ˆ|c {ˆ b† c ˆ} . We begin with a particle–particle (pp) term. Note that there is a phase ambiguity in the last example.1 Representation of one-particle operators and contractions Next we consider the representation of operators.1) acting on a singly excited Slater determinant i}|0 .92 Diagrammatic notation i}|0 = |Φa ˆ†ˆ i = {a i a ˆ† ˆ} = Φa i | = 0|{i a i a |Φab ˆ †ˆ b† ˆ jˆ i}|0 = {(ˆ a†ˆ i)(ˆ b† ˆ j )}|0 = ij = {a i a j b etc. the way in which we shall use the diagrams will be independent of this choice of phase.3 One-particle operators 4.2) The action and representation of the individual terms in the sum over p.1) will depend on whether p and q are particle or hole indices.3) . since this diagram ba could equally represent |Φab ij or |Φij . We begin with a oneelectron operator in normal product form. say ˆN = U pq p|u ˆ |q { p ˆ† q ˆ} . we can indicate which particle index appears above which hole index: i a j b 4. provided reasonable consistency is maintained. If we want to be more speciﬁc.

. Represented at the bottom is |Φa i . e.g.4. the horizontal arrangement has no signiﬁcance and we could equally well have used b × a i or. b a × i × as a multiplicative We associate the integral b|u ˆ|a with the vertex factor. can be represented by using diﬀerent markers. while the ket corresponds to the entering line. the × being operator is marked by the interaction line (or vertex ) a marker for the particular operator u ˆ.g. Note that the bra spinorbital in the integral corresponds to the line leaving the vertex. which is represented by the diagram (4. .5) b i a × Note that the case b = a is included.4) Thus we get a non-zero contribution only from particle–particle terms of the type ˆ}|Φa ˆ|a |Φb b|u ˆ|a {ˆ b† a i = b|u i . Diﬀerent one-electron operators #. Again.3 One-particle operators 93 for which we obtain (using the generalized Wick’s theorem) i}|0 = b|u ˆ|c {ˆ b† c i}|0 + b|u ˆ|c {ˆ b† c i}|0 ˆ}{a ˆ†ˆ ˆa ˆ†ˆ ˆa ˆ†ˆ b|u ˆ|c {ˆ b† c ic ˆ|0 + b|u ˆ|c δac {ˆ i}|0 = b|u ˆ|c ˆ b† a ˆ†ˆ b†ˆ = 0 + b|u ˆ|c δac |Φb i . etc. (4. the resulting determinant. . e. The point of action of the and at the top |Φb i × .

we obtain: ˆ}{a ˆa ˆa i}|0 = j |u ˆ|k {ˆ j†k i}|0 + j |u ˆ|k {ˆ j†k i}|0 ˆ†ˆ ˆ†ˆ ˆ†ˆ j |u ˆ|k {ˆ j†k ˆa ˆa iˆ j † |0 + δij j |u = − j |u ˆ|k k ˆ†ˆ ˆ|k {k ˆ† }|0 ˆ}|0 ˆ|k {a ˆ† k = 0 − δij j |u ˆ|k |Φa = −δij j |u k . ˆ}. j |u ˆ|k {ˆ j†k |Φa i . in (4. Using the generalized Wick’s theorem. This representation of contractions tells us automatically which pairs of lines may be contracted (ˆ ˆˆ b† ).1). The only nonvanishing term of this type is ˆ}|Φa = − i|u ˆ|k |Φa i|u ˆ|k {ˆ i† k i k . and where a creation operator is represented by an × or double line ) arrow leaving its point of action (a vertex and an annihilation operator is represented by an arrow entering its point of action. where the left–right order of the operators corresponds to the top–bottom order of the directed lines. (4.7) (4. Any line that is not contracted i† ˆ j or a or terminated by a vertex (or by the double line) must proceed unchanged in the appropriate direction throughout the diagram.6) . a ˆˆ b† : a b δab .94 Diagrammatic notation We can view this diagram as a graphical representation of the contraction process: ˆ} : × b|u ˆ|c {ˆ b† c b c → × b c c ˆa ˆ† = δac a i → × b i a |Φa i : a i The only non-zero contractions are then represented as follows: i ˆ i†ˆ j : j δij . acting on Next we consider a hole–hole (hh) term.

2 Rules of interpretation We should note the following rules for the interpretation of the one-particle vertices.3. 1. entering the vertex) is above the j line (the creation operator. in the representation of the operator. For both the pp and hh cases. leaving the vertex) and a minus sign appears. {ˆ j†k j † . 4.4. the bra in the integral associated with the vertex corresponds to the outgoing line (and the creation operator) while the ket corresponds to the incoming line (and the annihilation operator): k j ˆ} : j |u ˆ|k {ˆ j†k × b|u ˆ|c {ˆ b† c ˆ} : × b c .3 One-particle operators 95 which is represented by k × i This term originates in the contraction a ˆ} : × j |u ˆ|k {ˆ j†k k j → × k j ˆ j †ˆ i = δij i a → × k a i |Φa i : i a ˆ} = −k ˆˆ Note that. hence the k line (the annihilation operator.

Again. Similarly. 3. one is relying on its cancellation with another EPV diagram that results from contractions ˆ (which are excluded if within the creation and annihilation operators of U ˆ we use UN ). (4. represented by a diagram with a “bubble ”. So in this case we have the additional hh terms ˆa i}|0 = δjk j |u ˆ|k |Φa ˆ†ˆ j |u ˆ|k {ˆ j†k i .9) namely. since this gives i × i a i|u ˆ|i {ˆ i†ˆ ˆ |i ˆ iˆ i† |Φa ˆ|i |Φa i}|Φa i = − i|u i = − i|u i : In formalisms that do not take advantage of normal ordering and Wick’s ˆ rather than U ˆN . the contheorem and that therefore compute terms for U tribution of the i|u ˆ|i integral is i|u ˆ|i ˆ i†ˆ i|Φa i = 0. the creation operator associated with U outgoing line (a line leaving the vertex) and the annihilation operator corresponds to an incoming line (a line terminating at the vertex). × k i a (k = i). terms of the type + k |u ˆ|k |Φa i (k = i). The case k = i is included in the hh diagram. this is true for both the pp and hh cases. We have to associate a phase factor −1 with the case in which one hole is replaced by another (the hh case). The contribution − i|u ˆ|i |Φa i .8) and the corresponding diagram is called an exclusion-principle-violating (EPV) diagram. Note This diagram represents the diﬀerence between U i i ia ˆ†ˆ i|0 = 0 that the case k = i should be excluded. according to the ordinary interpretation of the diagram. since it corresponds to ˆ i†ˆ . 4. is included nevertheless. (4.96 Diagrammatic notation ˆN corresponds to an 2. ˆ |Φa and U ˆN |Φa .

on ··· the same path. since it originates in {ˆ i†ˆ i}|Φa i = −ii |Φi = a −|Φi . beginning with ph: j }{a ˆ†ˆ i}|0 |Φa ja ˆ†ˆ i}|0 ˆ|j {ˆ b† ˆ b|u ˆ|j {ˆ b† ˆ i = b|u †ˆ† ˆˆ = b|u ˆ|j a ˆ b j i|0 = b|u ˆ|j |Φab ij (no contraction can contribute in this case). and so we do not need to include bubble diagrams. In principle we should have omitted the cases j = i and/or b = a. × showing that the resulting determinant is |Φab ij (for the phase factor. (4. are in the same vertical position in |Φab ij ··· ).3. . but if we leave it in. but it is not necessary to do this explicitly since EPV diagrams of this type.3 The complete one-particle operator ˆN |Φa . Next we will look at the We have looked at the pp and hh terms of U i ph and hp terms. they would just provide the added term k |u ˆ|k k a a ˆ ˆ ˆ |Φa i = 0|U |0 |Φi = (U − UN )|Φi . represented by b j i a . the ﬁrst EPV diagram (on the left in the above ˆˆ† a ﬁgure) is all right as it stands.4.e.3 One-particle operators 97 and is an EPV term (containing two holes i at the same time). if we consistently use the normalproduct form of operators. i. 4. (4. However. it cancels the previously mentioned EPV diagram exactly: i × i − i|u ˆ|i |Φa i + i|u ˆ|i |Φa i a + × i i a = 0. we use the convention that a hole and a particle joined at the same vertex.10) Therefore we eliminate the k = i restriction in both cases and include EPV diagrams in the summation.11) as noted above.

(4.j b × j i a + × i a b|u ˆ|a |Φb i − i|u ˆ|j |Φa j b|u ˆ|j |Φab ij i|u ˆ|a |0 (without restrictions on any of the summation indices). give a vanishing SD anyway. The hp term is i}|0 = j |u ˆ|b {ˆ j †ˆ i}|0 + j |u ˆ|b {ˆ j †ˆ i}|0 b}{a ˆ†ˆ ba ˆ†ˆ ba ˆ†ˆ j |u ˆ|b {ˆ j †ˆ + j |u ˆ|b {ˆ j †ˆ ba ˆ†ˆ ba ˆ†ˆ i}|0 + j |u ˆ|b {ˆ j †ˆ i}|0 = j |u ˆ|b {0 + δij × 0 + δab × 0 + δij δab }|0 ˆ|b |0 = δij δab j |u and is nonzero only for i|u ˆ|a {ˆ i† a ˆ}|Φa ˆ|a |0 . × a j i a . × For both the ph and hp terms. i = i|u which is represented by × i a . .12) showing clearly that the result of the operation involves the vacuum state |0 . the integral associated with follows the same interpretation rules as for the pp and hh cases. note the mnemonics in parentheses: incoming line outgoing line ↔ ↔ annihilation operator creation operator ↔ ↔ ket state bra state (IAK) (OCB) Of the four cases considered. × a i i a . ˆN |Φa is The complete result for U i b i b a × + j × j a i + b.98 Diagrammatic notation × b i i a . only the hole–hole case requires a factor −1.13) (4.

a i a + i. ˆN on a doubly exFor another example.14) Only the ph term contributes here.i × i a = a. (4.j j + i × i.17) . since this is the only term in which we can have the vacuum state below the interaction.16) ˆ there is one more term: For U ˆ =U ˆN + 0|U ˆ |0 = U ˆN + U i × i . (4. (4.a × i a . (4.b × b + a × i.c i a j b k c × + × i a j b + i a j b × + × i b j a + i b j a × . we can represent the operator U ˆN = U a.3 One-particle operators 99 ˆN on the Fermi A very simple example is provided by the operation of U vacuum state: ˆN |0 = U a.15) ˆN as In general.4.i a|u ˆ|i |Φa i . we consider the operation of U cited SD: c i c ˆN |Φab U ij = a × j b + k × k a i j b c + c i a j b × + k i a × k b j + k.

. To avoid having to write summation signs we introduce the convention that any unlabeled line is summed over all possible hole or particle indices. b = δij i a = δij b|u ˆ|a j a i i|u ˆ|j .18) a ˆ Φb j |UN |Φi × + δab × − δab as we already know. respectively.3. (4. the exUsing the markers pectation value can be evaluated from × ˆN |0 = ˆN Z 0|U # = i.20) (The notation will be modiﬁed later to include energy denominators. hole indices for downward arrows and particle indices for upward arrows. each of which is in normal-product form.) No other diagram is possible here.19) 4. × and # for u ˆ and z ˆ. (4.4 Products of one-particle operators Consider the vacuum expectation value of a product of two one-particle operators. ˆN = U pq p|u ˆ |q { p ˆ† q ˆ} and ˆN = Z pq p|z ˆ|q {p ˆ† q ˆ} .16) is to be contracted in all valid ways with whichever state diagram exists below it and any state diagram we desire to be above it. ˆN is written without labels in the form Thus the representation of U ˆN = × U + × + × + × . For example. (4.a i|u ˆ|a a|z ˆ|i . since we have to have the Fermi vacuum at both top and bottom.100 Diagrammatic notation A diagram in (4.

as indicated by the sense of the ˆN |0 = 0 for any normalcircles as they pass through the vertices. Z ˆN .23) .21) Note that bubbles can carry hole labels only. no corresponding diagrams can be constructed).a × + # # − × − # × − # × ˆN . the mixed terms 0|U not be considered (and.3 One-particle operators 101 ˆ. The only diagrams with nonzero vacuum expectation values are (21) and (34). i|u ˆ|i j |z ˆ|j = 0|U i. 4. only the normal product part survives: × ˆ. indeed. diagrams consisting of two or more disconnected parts).e. we would have had the additional If we had used U term × # ˆ |0 0|Z ˆ |0 . For the U are in addition disconnected diagrams in Fig.2 (i. ˆN |0 = 0|U ˆZ ˆ |0 − 0|U ˆN Z 0|U (4. 4. (4.a = 2i Im . As noted previously.j = (4.1. Because of the cancellation of the disconnected terms in such a commutator. which were obtained by contracting the U ˆZ ˆ product there ˆ with the ZN diagrams below in all possible ways. Z ˆN ]|0 0|[U i|u ˆ|a a|z ˆ|i − i|z ˆ|a a|u ˆ|i i|u ˆ|a a|z ˆ|i i. ˆN Z ˆN is represented by the diagrams in The complete operator product U ˆN diagrams at the top Fig. the latter represents ˆ |0 0|Z ˆ |0 . Since 0|U ˆ |0 and 0|U ˆ |0 0|Z ˆN |0 need ˆN |0 0|Z product operator. Z ˆ ]|0 = 0|[U # × = # = = i.4. Z ˆ instead of U ˆN .22) Particularly interesting is the vacuum expectation value of the commutator of two operators.

if all the integrals are real then the vacuum expectation value of the commutator vanishes. . 4. This result is typical of the simpliﬁcations that are possible when working with commutators rather than with individual operator products.1.102 Diagrammatic notation × # 1 × # 6 × # 11 × # 16 × # 21 × # × # × × # 4 × # 8 × # 9 × # 14 × # 19 × # # 24 × × # × # 5 2 × # 7 × # 12 × # 17 × # 22 3 10 × # 15 × # 20 × # 25 # 13 × # 18 23 ˆN Z ˆN . we shall consider several examples of the matrix elements of operator products. First. In preparation for extending the treatment of phase factors. Fig. Obviously. Diagrams representing the operator product U where “Im” indicates the imaginary part.

4. Additional diagrams for U ˆN |0 .24) In this equation we introduce another (nonstandard) notational convention: labels in parentheses identify the corresponding summation indices in the accompanying algebraic expressions. Fig. The above result was obtained by condiagram for ZN (the only one with the vacuum state # below it) with those UN diagrams that can be attached to it without leaving tracting the .4.3 One-particle operators 103 × # 26 × # 31 # × × # × # 29 × # 30 27 28 × # 32 33 × # 34 × # ˆZ ˆ. i|u ˆ|a a|z ˆ|i |0 ia abij a|u ˆ|i b|z ˆ|j |Φab ij (4.2. which can be repreˆN Z let us take the relatively simple example of U sented by ˆN Z ˆN |0 = (i) U (a) (b) × + (a) (i) (j ) × # − aij # a|u ˆ|b b|z ˆ|i |Φa i abd j |u ˆ|i a|z ˆ|j |Φa i + (a) (i) × # + (j ) (i) (b) # (a) ×.

×.104 Diagrammatic notation any lines extending to the bottom. The phase of the resulting SD in the last diagram is obtained from the condition that ··· surviving hole and particle lines at the top are paired vertically in |Φab ij ··· if they are part of the same path.3. At the same time. . but the rules give − i|u ˆ|a b|z ˆ|i (for j = i). diagram 2 contains an EPV term when j = i. use an unrestricted sum for 2 and include 4. in which the intermediate ˆN ) has two holes in state i. We have a total of nine diagrams. of which the last three contribute only in the special case a = b. . Diagram 4 is a true EPV case. 4. The only minus sign is for the combination in the second diagram. which should not contribute either (since it has two holes in i in the intermediate state). but if we apply the diagram evaluation rules we do get a contribution. as noted in subsection 4. Any diagrams that can be brought to congruence . ˆN but before U state (after the operation of Z Clearly this diagram should not contribute. . The surviving with the terms are shown in Fig. .3. In diagrammatic many-body theory we always use option 2. i|u ˆ|a b|z ˆ|i . or to 2. # # # . Other examples of this choice will be encountered later. Thus these two spurious contributions cancel exactly. Among these are some unusual cases. #. The result is the same. restrict the summation in diagram 2 to j = i and leave out diagram 4. We also need to re-emphasize that the horizontal arrangement in a diagram is not signiﬁcant.2. a ˆ ˆ Now consider the example Φb i |UN ZN |Φi . even though the second choice includes the evaluation of two terms that should have been left out. Here we have to connect the fragments i b and i a (note the same hole label i in both fragments) through the operator fragments × × × . × interaction line above the # line. We have the option either to 1.

3. × is equivalent to # # × . Diagrams for the evaluation of the matrix element Φb i UN ZN Φi . by changing the horizontal positions of lines and vertices are not distinct.3 One-particle operators 105 1 i b (c) a × # 2 i # − j b (j ) a j |u ˆ|a b|z ˆ|j i i × 3 i i b a × # b|u ˆ|c c|z ˆ|a c b|u ˆ|i i|z ˆ|a b a 4 # i b i i|u ˆ|a b|z ˆ|i a × 5 × b a # 6 # i i × − i|u ˆ|i b|z ˆ|a 8 × (c) i 9 δab i − b|u ˆ|a i|z ˆ|i × δab (c) # 7 × # i (j ) i j |u ˆ|i i|z ˆ|j δab i (j ) # −δab δab j i|u ˆ|c c|z ˆ|i c δab cj j |u ˆ|c c|z ˆ|j ˆ ˆ a Fig. . For example.4. × # is equivalent to × # . 4.

the sign factor can also be determined from the parity of the number of intersections of the contraction lines. The parity of the unscrambling permutation provides the sign factor.106 Diagrammatic notation 4. diagrams 1.5 Phase factors The example in Fig.3 raises the important question of phase factors.8. even though our current rule does not call for it. Each closed path represents a sequence of creation and annihilation operators connected alternately either by contraction or by belonging to the same original creation–annihilation pair (i.3. corresponding to ˆ j b† ˆ i a ˆ†ˆ ˆ j†a ˆ . As noted in Section 3. 4 and 9 have two closed paths each. taking care that the order of the operators within each contracted pair is not changed. A path is a sequence of connected lines. 4. in diagram 2 the connection pattern is ˆ i†ˆ b ˆ ˆ ˆ b† ˆ b ˆ i† ˆ j† a j a ˆ† ˆ i . 4.e. The need for a minus sign to be associated with diagram 2 can be seen from the following: b}{ˆ j†a bˆ ˆ}{ˆ b† ˆ ˆˆ b† ˆ 0|{ˆ i†ˆ j }{a ˆ†ˆ i}|0 = 0|{ˆ i†ˆ j†a ja ˆ†ˆ i}|0 ˆa ˆ† }|0 = −1 . 2 and 5–8 each consist of a single closed path (or loop).25) Here we have unscrambled the operator product into a sequence of contracted pairs of operators. bˆ b† ˆ = − 0|{ˆ i†ˆ iˆ j †ˆ ja (4. We have put a minus sign in the expressions corresponding to diagrams 2 and 8.. For example. .3. To be able to determine the sign associated with a particular diagram directly from the diagram we need a more general sign rule. including any (external) connection or that indicates particle–hole pairs in the initial or ﬁnal state. while diagrams 3. In developing such a rule we consider the notion of a path in the diagram (see Paldus ˇ ıˇ and C´ zek 1975). a vertex or external connection) in a normal product. In the example of Fig.

say or n ˆ† n ˆ f f . Similarly.3 One-particle operators 107 while in diagram 4 we have ˆ i†ˆ b ˆ ˆ ˆ b† ˆ b ˆ i† ˆ i† a i a ˆ† ˆ i . the rightmost pair is associated with the bottom feature. We shall now show that the sign factor for each loop is (−1)h−1 . The sign factors can thus be computed separately for each loop and then multiplied together. n ˆ† n ˆ f f The general form of the operator sequence is ˆ† n ˆ ˆ p ˆ† q ˆ r ˆ† s ˆ .. f ˆ. where h is the number of hole lines in the loop. which rearranges to decouple the two loops without change of sign. Entangled loops can always be disentangled with an even permutation moving pairs of operators (each corresponding to a vertex or an external connection) past other pairs. The left-most pair of creation–annihilation operators corresponding to any loop must be associated with the topmost feature of the loop. say ˆ ˆ i† a i a or i ˆ ˆ i† a a .4.. corresponding to ˆ i b†ˆ a ˆ†ˆ i ˆ i† a ˆ . since this only requires pairs of operators (vertices) to be moved past other pairs: ˆ b ˆ b† ˆ i† ˆ i ˆ ˆ a ˆ† ˆ i† a i. i† a .

We operators k . represented by the loop c ˆ ˆ j f c Obviously unscrambling the contractions in this product produces a change of sign. . r. but have some pairs of particle operators between them. the interior p ˆ† q † ˆ ˆ ˆ. q. . Thus. Then the diagram must ˆ (providing a factor δin ) and. If the two interior hole operators are adjacent we have the pattern a i ˆ i† a ˆ a ˆ† · · · ˆ f ˆ† ˆ j† ˆ i c ˆ† · · · f j . as in . two pairs of contracted hole operators. c ˆ† .108 Diagrammatic notation where each of p. f This example clearly has a + sign. must have the general form a b ˆ ˆ a ˆ† ˆ ˆ c ˆ† · · · b ˆ b† c i† a ˆ f ˆ† ˆ f i . To allow ˆr ˆ† s ˆ · · · must contain an even number of hole full contraction. represented by the loop i c . for a have a contraction between ˆ i† and n nonzero result.e. for a loop with one hole line (h = 1) we get the phase factor +1 = (−1)h−1 . may refer to either a particle or a hole. Next we consider the case of two hole lines. i. k as well as an even number of particle operators c consider ﬁrst the case of no interior hole operators. as required. . If the two interior hole operators are not adjacent. s. so that we still have a phase factor (for h = 2) equal to (−1)h−1 = −1.

c ˆ ˆ ˆ† ˆ i c d j. h2 . . .3 One-particle operators 109 a ˆ i† a ˆ a ˆ† ··· c ˆ ˆ j† ˆ d ˆ† ˆ ˆ f ˆ† ˆ i c ˆ† · · · f j. (4. we thus get l i=1 (−1)hi −1 = (−1)h−l . For l loops with h1 . The same result is also obtained if more pairs of particle operators appear in between the hole operators.26) where h = l i=1 hi is the total number of hole lines.4. f then we ﬁnd the same sign factor as in the above simpler pattern. giving a phase factor (−1)h−1 for a loop with h hole lines. if more hole lines are introduced then we obtain another factor −1 for each additional hole line.e. respectively. represented by the loop d d j f or i c. alternative external connections) produce equivalent results. Using the phase factor (−1)h−l we can easily see that alternative pairings of open lines (i. . . so that we do not have to indicate the external connections explicitly. a ˆ† . . . represented by the loop ˆ a ˆ† . hl hole lines. . f ˆ c ˆ f ˆ† ˆ ˆ i† a j† d j d i . . Similarly. .

110 Diagrammatic notation For example. 4 and 8: 9 i a a j × # 2 (l = 2) cancels with i × # a a j (l = 1) . which we have used temporarily as a teaching tool. we show how the phase-factor rule helps in demonstrating the cancellation of EPV diagrams.3. (l = 2. Again we use the example of Fig. in general. i j b # a|u ˆ|i b|z ˆ|j a × = i j b # a × i = j b # a × |Φab ij = − a|u ˆ|i b|z ˆ|j |Φab ji . 9 i a a i × # 4 (l = 2) + # i a a × i (l = 2) cancels with . i. h = 2) (l = 1. 4. we evaluate the matrix element Φa i |UN ZN |Φi . 9 i a c i × # 8 (l = 2) cancels with a i i c × # (l = 1) . Finally. in the a ˆ ˆ special case a = b. h = 2) (4. Therefore we can dispense henceforth with specifying external connections in open diagrams. with the unconventional double-bar notation. We can also dispense. We show the cancellation of the EPV components of diagram 9 with the EPV components of diagrams 2. as long as the phase factors are calculated from the formula (−1)h−l for whichever connections are chosen.e.27) Thus the results are independent of how the external connections are made.

4 Two-particle operators 4.e. In each case the integrals involved are the same and the number of hole lines is the same but the number of loops diﬀers by one. The two half-vertices and the interaction line constitute a single vertex. the same point on the time axis). each of which may be a particle line or a hole line. so we get cancellation in each set. with the factor 1 2 .4. The association of line labels with the two-electron integral indices and the creation or annihilation operators follows the same rule as for one-body vertices: bra index ket index ↔ ↔ creation operator annihilation operator ↔ ↔ outgoing line incoming line but with the added feature for the two-body case that electron 1 electron 2 ↔ ↔ left half-vertex right half-vertex Thus for the term pq |rs {p ˆ† q ˆ† s ˆr ˆ} we have the association p ˆ† r ˆ ↔ ↔ left outgoing line .4. for example .4 Two-particle operators 111 2 i a a i × # 8 (l = 1) + a i i a × # (l = 1). and later go over to the . ˆ = W 1 2 pqrs pq |rs {p ˆ† q ˆ† s ˆr ˆ} = 1 4 pqrs pq rs {p ˆ† q ˆ† s ˆr ˆ} . (4.1 Goldstone diagrams for a two-particle operator We now turn to a two-particle operator in normal-product form. Each individual half-vertex will have one incoming and one outgoing line.28) We shall ﬁrst use the pq |rs form. . right incoming line . 4. left incoming line q ˆ† s ˆ ↔ ↔ right outgoing line . This operator will be denoted by antisymmetric form pq rs with factor 1 4 an interaction line connecting two half-vertices at the same level (i.

29) . Note that we have left out diagrams that can be obtained from one of the above diagrams by interchanging the left and right half-vertices. Taking into consideration the diﬀerent possible assignments of particle ˆ operator is represented by and hole lines. as we discuss next. bi|aj {ˆ b†ˆ (4. take the diagram fragment (b) (a) (j ) (i) = 1 2 abij i†ˆ ja ˆ} . e. ˆ So far we have ignored the factor 1 2 in front of the sum for W . b† a (4. = .31) . as well as the collection of equivalent terms.112 Diagrammatic notation The integral indices associated with a two-body vertex are assigned according to the scheme left–out right–out | left–in right–in . the W ˆ = W + + + + + + + + + (4. which corresponds to exchanging the names of electrons 1 and 2. where summations over all appropriate indices are implied by the unlabeled lines. while the corresponding operator product can be described by {(left–out)† (right–out)† (right–in)(left–in)}. (j ) (i) (b) (a) = 1 2 abij ib|ja {ˆ i† ˆ ˆˆ j} .g. As an example.30) There is an equivalent fragment. These equivalences will be taken care of by weight factors. Diagrams employing this representation of the two-body interaction (which is based on non-antisymmetrized integrals) are called Goldstone diagrams.

since ib|ja = bi|aj and ˆ b†ˆ i†ˆ ja ˆ= b† . and so the factor 1 2 must be retained. (4.32) there is no other diagram equivalent to it since its mirror image in a vertical plane bisecting the vertex is congruent with it. ˆ† and u ˆ† . which represent the resolvent operators in the energy expressions. the energy of a many-body system is given by sums of diagrams representing vacuum expectation values. In particular.4. which requires that 1. r. w. However. a rigorous treatment of the weight factor problem cannot deal with diagram fragments. The diagrammatic representation of these denominators will be discussed in Chapter 5. 4. v be hole indices. which requires that 2. we shall consider the vacuum ˆ 2 .4.4 Two-particle operators 113 i† = −ˆ which is exactly equal to it. in either order. but must involve complete diagrams. in order to make sure that we have the correct numerical factors and phases: ˆ 2 |0 = 0|W 1 2 pqrs pq |rs 1 2 tuvw ˆ† u tu|vw 0|{p ˆ† q ˆ† s ˆr ˆ}{t ˆ† w ˆv ˆ}|0 = 1 4 pqrstuvw pq |rs tu|vw contractions ˆ† u 0|{p ˆ† q ˆ† s ˆr ˆt ˆ† w ˆv ˆ}|0 . if we consider the example (b) (a) (d) (c) = 1 2 ˆ† c bd|ac {ˆ b† d ˆa ˆ} . . see (2. q. To take a simple example. contract p ˆ† and q p. contract s ˆ and r ˆ with t s. in either order. (So far we have ignored the question of energy expectation value of W denominators.71). This question will be dealt with further in the next subsection. Thus we agree to use just one of the two equivalent fragments and so remove the factor 1 2 . abcd (4. u be particle indices. The only two possibilities for obtaining nonzero fully contracted terms without using internal contractions among the operators of either of the two original normal prodˆ† u ˆ† s ˆr ˆ} and {t ˆ† w ˆv ˆ} are: ucts {p ˆ† q ˆ† with w ˆ and v ˆ.33) Now we consider how the contractions are to be made. using the generalized Wick’s theorem. t. ˆ ˆ −a ˆj. However.) We shall ﬁrst evaluate this quantity algebraically.2 Vacuum expectation values of products of two-particle operators Most of the diagrams with which we shall deal in many-body perturbation theory involve vacuum expectation values.

35) ˆ verThe diagrammatic description of this matrix element involves two W tices. with the vacuum at top and bottom. and so are the second and third. corresponding to the four contraction schemes in (4.34): . (4. The two vertices must therefore have the forms for the top vertex and for the bottom vertex and can be connected in four diﬀerent ways. one above the other. ˆ 2 |0 = 0|W 1 4 abcdijkl ij |ab cd|kl contractions ˆ† ˆ ˆ}|0 i†ˆ j †ˆ 0|{ˆ ˆc ˆ† d ba lk = 1 4 abcdijkl ˆ† ˆ ˆ|0 i† ˆ j †ˆ ij |ab cd|kl 0|ˆ ˆc ˆ† d ba lk + abcdijkl ˆ† ˆ ˆ|0 ij |ab cd|kl 0|ˆ ˆc ˆ† d ba lk i†ˆ j †ˆ + abcdijkl ˆ† ˆ ˆ|0 ba lk ij |ab cd|kl 0|ˆ ˆc ˆ† d i†ˆ j †ˆ + abcdijkl 1 4 abij ˆ† ˆ ˆ|0 i†ˆ j †ˆ ij |ab cd|kl 0|ˆ ˆc ˆ† d ba lk ij |ab ab|ij − abij = ij |ab ba|ij ij |ab ba|ji abij − abij ij |ab ab|ji + .114 Diagrammatic notation Thus there are four distinct contractions. Thus the result is ˆ 2 |0 = 0|W 1 2 abij ij |ab ab|ij − ab|ji .34) The ﬁrst and fourth terms are equal (by exchange of electrons 1 and 2 in the second integral). (4.

The same situation holds for the diagrams. but we need a factor 1 2 for each diagram or unlinked part of a diagram that can be written in a form symmetric under left–right reﬂection. Note also that the EPV terms. which we derived originally for diagrams with one-body vertices only. will be included separately in the unrestricted sum over i. because of the unrestricted summations. ˆ 2 |0 = 0|W 1 2 abij + ij |ab ab|ij (l = 2. the ﬁrst and fourth diagrams are equivalent (by exchange of the two half-vertices at the top or bottom) and so are the second and third. in which i = j and/or a = b. Each distinct term should be included once but. The minus sign for the exchange diagram is obtained from our phase-factor rule (−1)h−l . a simultaneous exchange p ↔ q. ij |ab ab|ji (l = 1. To provide further examples of diagram construction. applies equally to diagrams containing two-body vertices. More generally. r ↔ s (which corresponds to the exchange of electron labels 1 and 2) does not produce a distinct term. and of rules for diagram interpretation and phase factor and weight factor determination. This rule. j.4 Two-particle operators 115 As in (4. ij |ab ab|ij ji|ba ba|ji which are equal. h = 2) −1 2 abij . It compensates for the fact that terms such as a 1 2 i j b and b 1 2 j i a . b. The restriction to inequivalent diagrams eliminates the double counting for asymmetric diagrams. often referred to as the direct and exchange diagrams. cancel between the direct and exchange terms. some are included more than once and the weight factor is needed to compensate.34).36) in agreement with the algebraic derivation. we note that the original coeﬃcient 1 2 in the sum 1 † † ˆq ˆs ˆr ˆ} was introduced because only distinct terms should pqrs pq |rs {p 2 be summed. respectively. h = 2) (4. The factor 1 2 derives from the fact that each of these diagrams is (or can be redrawn to be) symmetric under reﬂection in a vertical plane through its middle.4. as shown. Thus we are left with two diagrams. we next consider the vacuum expectation value of a product of three two-body . justifying the use of unrestricted summations. a.

we have a direct and an exchange diagram. h = 2. obtained in this case by interchanging the endpoints of the incoming particle lines at the middle vertex (any alternative exchange produces an equivalent diagram). ij |ab ab|dc cd|ij l = 1. the incorporation of the resolvent operators and their concomitant denominators. (i) (j ) (a) (k ) (l) (i) (j ) (a) (k ) (l) (b) = − 1 2 abijkl (b) = 1 2 abijkl ij |ab kl|ij ab|kl l = 2. symmetric .116 Diagrammatic notation ˆ 3 |0 . we obtain the hole ladder diagrams. This aspect will be introduced in Chapter 5. symmetric . Using the second choice for the middle vertex. h = 4. i. referred to as particle ladder diagrams (shown with their algebraic interpretations). To allow full contraction with no the same form as in the 0|W open lines remaining. of which the top and bottom vertices must have ˆ 2 |0 case. The ﬁrst choice produces two diagrams. symmetric ij |ab lk |ij ab|kl l = 1. the middle vertex must have two lines above and two lines below the vertex. . .e. (We should note again that in this chapter we are operators. The second diagram is the exchange version of the ﬁrst. (a) (b) (i) (c) (d) (a) (b) (i) (c) (d) (j ) = − 1 2 abcdij (j ) = 1 2 abcdij ij |ab ab|cd cd|ij l = 2. 0|W leaving out an important diagram interpretation aspect relating to the use of these diagrams in perturbation theory. h = 2.) Each diagram for this matrix element contains three two-body vertices. for which there are the four possibilities . Here also. symmetric . h = 4. . . .

(i) (a) (j ) (k ) (c) (b) = abcijk ji|ab kb|ci ac|jk l = 3. (j ) (i) (a) (k ) (c) (b) =− abcijk ji|ab kb|ci ac|kj l = 2. (a) (i) (b) (j ) (k ) (c) =− abcijk ij |ab bk |ic ac|jk l = 2.4 Two-particle operators 117 The third and fourth choices for the middle vertex. (b) (c) (j ) = abcijk ij |ab bk |ic ac|kj l = 1. asymmetric . h = 3. asymmetric . h = 3. taken together. (b) = abcijk ij |ab kb|ic ac|jk l = 1.4. h = 3. asymmetric . asymmetric . h = 3. referred to as ring diagrams (the name is taken from the shape of the last diagram of this set). asymmetric . (i) (b) (a) (k ) (c) (i) (a) (j ) (k ) (i) (a) (j ) (k ) (i) (a) (k ) (c) (c) (j ) = − abcijk ij |ab kb|ic ac|kj l = 2. (a) (i) (j ) (k ) (b) =− (c) abcijk ji|ab kb|ic ac|jk l = 2. h = 3. h = 3. asymmetric . h = 3. produce eight diagrams. asymmetric . h = 3. asymmetric . (b) = abcijk ji|ab kb|ic ac|kj l = 1. .

including the associated integrals (ignoring weight and phase W factors) is shown in Fig. There is also some diﬃculty in making sure that all those distinct possibilities have been listed exactly once. since it is not always easy to determine whether two diagrams are equivalent. The diagrams using this representaˆ operator are called Hugenholtz diagrams (Hugenholtz 1957). and therefore the phase of the corresponding algebraic interpretation is indeterminate. while the incoming labels appear in the ket part. and there would still be three single. (4. in fact.4. ˆ 2 |0 = 0|W = 1 4 abij ij ab ab ij . The labels on the outgoing lines appear in the bra part of the antisymmetrized integral. The diﬃculties associated with the use of the Goldstone representation can be overcome by basing the analysis on the antisymmetric integrals pq rs (the ﬁnal form in (4. three double and one triple exchange relative to it.3 Hugenholtz diagrams The principal problem with the Goldstone representation is the rapid growth in the number of distinct diagrams as the number of interaction vertices increases. (We shall deal with this phase-factor problem in the next subsection. tion of the W They maintain the usual (Goldstone) form for one-body operators but represent the two-body vertex as a single large dot with two incoming and two outgoing lines (each of which can be a particle or hole line). One could.) The correspondence between the Goldstone and Hugenholtz representations of the various terms in the ˆ operator. an exchange diagram is obtained by exchanging the terminating points of the two incoming lines or the starting points of the two outgoing lines at any two-body vertex. The order of the labels in each part is indeterminate. the advantage of Goldstone diagrams is the straightforward determination of phase factors. reﬂecting the individual listing of each possible exchange. 4.28)). such an approach leads to a much smaller number of distinct diagrams. Since the exchange contribution is incorporated within each antisymmetrized integral.37) .118 Diagrammatic notation If the ﬁrst of these eight diagrams is considered as the direct case then the next three are single exchange. 4.4. the following three double exchange and the last one triple exchange. However. ˆ 2 |0 matrix element has just The Hugenholtz representation of the 0|W one distinct diagram instead of two. In general. take any of these eight as direct.

4.4. The correspondence between the Goldstone and Hugenholtz diagrams for ˆ operator and their associated integrals (the weights and phase factors have the W been left out).4.4 Two-particle operators Goldstone a c i k a b a i b a b i j a b i j a k c i a k c j i b d j l i j Hugenholtz 119 ab|cd a c i k b d j l ab cd or ab dc kl|ij aj |bi aj |ib kl ij or kl ji a b i j aj bi or aj ib c ac|bi a b a ja|ik i j a b c jk |ia i j a a i b j ab|ij i a a i b b j j ab ij or ab ji k jk ia or jk ai i ai bc or ai cb k ja ik or ja ki i ac bi or ac ib ai|bc a i b j ij |ab ij ab or ij ba Fig. .

36). . Instead of the 12 diagrams of the Goldstone representation. reproducing the two-term result obtained with Goldstone diagrams. (b) (a) (c) (j ) (k ) (i) = ± abcijk ij ab kb jc ac ik . respectively: (a) (i) (c) (b) (d) (j ) = ± 1 8 abcdij ij ab ab cd cd ij . which are diagrams without the arrows.120 Diagrammatic notation Expansion of the antisymmetrized integrals in terms of ordinary integrals gives four terms. which are equal in pairs. The power of the Hugenholtz representation is easily seen in the evaluation ˆ 3 |0 . Each pair of equivalent lines contributes a factor 1 2 . corresponding to the particle ladder. it is much easier to list all the distinct diagrams in this scheme. The last diagram replaces all eight Goldstone ring diagrams. The usual approach is ﬁrst to list all Hugenholtz ˆ 3 |0 matrix skeletons . It is obvious that the Hugenholtz representation reduces the number of diagrams considerably. . The weight factor 1 4 is obtained by counting the number of pairs of equivalent lines in the diagram: a pair of lines is equivalent if they connect the same pair of vertices in the same direction. The diaˆ 2 |0 has two such pairs (two equivalent particle lines and two gram for 0|W equivalent hole lines). resulting in a weight factor 1 4. (4. For the 0|W element we have just one skeleton. Furthermore. the hole ladder and the ring diagrams. we of 0|W obtain just three. (i) (a) (k ) (j ) (l) (b) = ± 1 8 abijkl ij ab kl ij ab kl .

The weight factors are those of the corresponding Hugenholtz diagram. The weight factor is also easy to determine.4 Two-particle operators 121 Arrows are then added in all distinct ways to produce all the distinct diagrams. This expansion is not unique.38) . because the four lines connected to a Hugenholtz vertex can be assigned in diﬀerent ways to the two half-vertices of the ASG diagram.4.4. with the same senses of the arrows. It is quite easy to tell whether two diagrams are distinct: any two diagrams are equivalent if they have the same connection pattern. The major disadvantage is that Hugenholtz diagrams do not immediately specify the phase of the corresponding algebraic expression. These antisymmetrized Goldstone (ASG ) diagrams (Brandow 1967) are obtained by expanding each large dot (two-body vertex) of the Hugenholtz representation into a Goldstone-type two-body vertex. two diagrams are equivalent if one can be deformed into the other without changing the vertical sequence of the vertices or the sense of the arrows.4 Antisymmetrized Goldstone diagrams A solution to the phase problem of Hugenholtz diagrams is obtained by converting each distinct Hugenholtz diagram into just one Goldstone-like diagram and interpreting each two-body vertex in this Goldstone diagram in terms of antisymmetrized integrals pq rs instead of ordinary integrals pq |rs . 4. = (a) (i) (j ) (b) or (a) (i) (j ) (b) = = = 1 4 abij 1 4 abij 1 2 abij ij ab ab ij ij |ab − ij |ba ij |ab = −1 4 abij ij ab ab ji ab|ij − ab|ji ab|ij − ab|ji . For example. (4. by counting the number of equivalent line pairs. In other words. but all possible expansions of a given Hugenholtz diagram give the same resulting algebraic expression provided that the signs are interpreted according to the rules for Goldstone diagrams. with a factor 1 2 for each pair of equivalent lines.

. 6. Generate all distinct Hugenholtz skeletons. l = 2. with the usual left–out right–out left–in right–in arrangement. (i) (b) = (a) (k ) (c) (j ) = − abcijk ij ab kb ic ac kj (h = 3. Interpret each one-body vertex in each ASG diagram as in ordinary Goldstone diagrams. n = 3) . l = 2. Assign a phase factor (−1)h−l . where n is the number of equivalent line pairs. 7. Expand each Hugenholtz diagram into an ASG diagram in any of the possible equivalent ways. n = 3) (i) (j ) = (a) (k ) (l) (b) = 1 8 abijkl ij ab kl ij ab kl (h = 4. 2. Interpret each two-body vertex in each ASG diagram in terms of an antisymmetrized integral. n = 0) . 4. 5. Assign a weight factor 1 2 . as for ordinary Goldstone diagrams. two lines are equivalent if they connect the same two vertices in the same direction. ˆ 3 |0 example. l = 2.122 Diagrammatic notation The procedure for generating and interpreting the antisymmetrized Goldstone diagrams can be summarized by the following rules: 1. we obtain the matrix eleApplying these rules to the 0|W ment as a sum of three diagrams: (a) (b) = (i) (c) (d) (j ) = 1 8 abcdij n ij ab ab cd cd ij (h = 3. For each skeleton assign arrows in all distinct ways to generate Hugenholtz diagrams. 3.

For example. l = 3. (4. we obtain the following new ˆ2 )N ): diagrams.29) (which we refer to as (V ˆ2 = (V ˆ2 )N + V + + + + + + + + + . to be summed over hole labels only. is drawn in this shape by convention. except where stated otherwise. the alternative expansion of the third diagram.4. but also involves hole lines only.4 Two-particle operators 123 Any other expansions of the Hugenholtz diagrams would give the same algebraic results. 4. and would . The last ﬁve diagrams (the “wiggles” and the oyster) are the exchange terms for the ﬁrst ﬁve (the bubbles). we must include internal contractions between the four creation or annihilation operators. The products of creation and annihilation operators of the form p ˆ† q last diagram. obviously reduces to an equal algebraic expression. all the Goldstone-type diagrams we use will be ASG diagrams. since only hole operators give nonzero contractions within ˆ† s ˆr ˆ. called an oyster.(4. From now on.28)). (i) (j ) (b) = (k ) (c) abcijk = (a) ji ab bk ic ac jk (h = 3. n = 0) .4. the directions in which the lines leave and enter a vertex indicate that they are hole lines.39) that is not in normal-product form (cf. Therefore the diagrammatic representation of such an operator includes contractions between the four lines attached to the same two-body vertex. to be added to those of (4.40) In all but the last of these diagrams. Dealing ﬁrst with the ordinary Goldstone representation.5 Representation of operators not in normal-product form When we deal with a two-body operator ˆ2 = V 1 2 pqrs pq |v ˆ|rs p ˆ† q ˆ† s ˆr ˆ (4.

The ﬁrst four diagrams in (4. h = 3. The non-normal-product Goldstone diagrams in the representation of ˆ 2 |0 are: 0|V 2 (j ) (a) (i) (k ) (j ) (a) (i) (k ) (j ) (a) (i) (k ) (j ) (a) (k ) (i) (k ) (i) (j ) (k ) (l) = −1 4 ijkl = aijk ij |aj ak |ik l = 3. h = 4.178). h = 3. h = 4. since it is immaterial which of the two hole lines is drawn above or below the vertex line. asymmetric . asymmetric . (i) = aijk ji|aj ak |ki l = 1. asymmetric . symmetric .40) and the corresponding exchange diagrams ˆ = ˆ† q ˆ} operin the second row represent the V pq i ( pi|qi − pi|iq ) {p N ˆ2 |0 = ator of (3. =− aijk ij |aj ak |ki l = 2. The ﬁfth and the remaining diagrams represent 0|V 1 2 ij ( ij |ij − ij |ji ). ij |ji kl|kl l = 3. =− aijk ji|aj ak |ik l = 2. h = 3. symmetric . (j ) = (l) 1 4 ijkl ij |ij kl|kl l = 4.124 Diagrammatic notation be left out in the ASG representation. For the purpose of weight-factor determination the oysters are considered symmetric. h = 3. . asymmetric .

These unlinked diagrams have a vacuum gap. other unlinked diagrams also disappear from these sums because of cancellation with renormalization terms.. symmetric . h = 3. As will be shown in Chapter 5. = 1 4 ijkl ij |ji kl|lk l = 3.4 Two-particle operators 125 (i) (k ) (l) (i) (j ) (k ) (l) (j ) = −1 4 ijkl ij |ij kl|lk l = 3. In the Hugenholtz and ASG representations the above eight Goldstone diagrams are replaced by just two diagrams. h = 4. 1 The weight factor 1 4 for the last four diagrams arises from the factor 2 for each of the two symmetric unlinked parts in each diagram. one for the four linked and one for the four unlinked Goldstone diagrams: (j ) = (a) (i) (k ) = aijk ij aj ak ik l=3 h = 3 . n=0 l=4 h = 4 .e. a Fermi vacuum intermediate state. h = 4. in the 0|V 2 (i) = (k ) (j ) = (l) 1 4 ijkl ij ij kl kl Finally. n = 1) . i. symmetric . n=2 ˆ 3 |0 . Such diagrams are excluded from perturbation theory sums since the primed sums in perturbation theory exclude the reference state. we give the non-normal-product terms for Hugenholtz and ASG forms only: (a) (j ) = (i) (c) (k ) (b) = 1 2 abcijk ij ab ab cj ck ik (l = 3. .4.

h = 4. = (a) (j ) (i) (b) = (k ) (l) abijkl ij ab bk jk al il (l = 4. h = 4. n = 0) . h = 4. h = 3. n = 1) . = (i) (a) (b) (j ) = −1 2 (k ) (l) abijkl ij ab kl jl ab ik (l = 3. n = 1) . h = 3. . (j ) (a) = (i) (b) (k ) (c) = 1 2 abcijk ij aj ak bc bc ik (l = 3. h = 4. n = 1) . n = 0) . (j ) (i) = (a) (k ) (l) (b) = −1 2 abijkl ij aj kl ib ab kl (l = 3. = (a) (i) (j ) 1 2 abcijk ij ab bk ck ac ij (l = 3. n = 1) . h = 4.126 Diagrammatic notation (i) (j ) = (a) (k ) (b) = −1 2 abijkl ij ab kb ij al kl (l) (b) = (c) (k ) (l = 3. (j ) (a) = (k ) (l) (b) (i) = abijkl ij aj ak ib lb lk (l = 4. n = 1) .

n = 3) . h = 4.4 Two-particle operators 127 (j ) (l) = (a) (i) (k ) = (b) abijkl ij aj kl bl ab ik (l = 4. n = 3) . n = 1) . (j ) (a) = (i) (b) (l) (j ) (i) = (a) (l) (m) (a) = (k ) (b) = (k ) (j ) = (a) (k ) (l) (b) 1 8 abijkl 1 8 abijkl (k ) = abijkl ij aj ak bk bl il (l = 4. . n = 0) . (i) (j ) (b) = (l) 1 8 abijkl ij ab ab ij kl kl (l = 4. (k ) =− aijklm ij aj lk ik am lm (l = 4. h = 4. h = 4. (j ) (a) = (l) (i) (k ) (m) = 1 2 aijklm ij aj ak ik lm lm (l = 5. h = 5. n = 0) . h = 5. h = 4. n = 0) . h = 4. = (a) (i) (j ) ij ab kl kl ab ij (l) (l = 4. (i) = ij ij kl ab ab kl (l = 4.4. n = 3) .

(j ) (l) = 1 2 aijklm ij ij kl al am km (l = 5. h = 6. while the two-electron part is represented matically by the ˆo by the standard two-body vertex. The last seven of these diagrams are unlinked. taken toˆ 3 |0 and the normal-product gether. while the ˆ matrix element 0|(V2 )N |0 .6 The RSPT perturbation operator As we saw in subsection 3.128 Diagrammatic notation (j ) = (a) (i) (k ) (m) (i) = (a) (k ) (m) (i) = (k ) (m) (j ) (l) (n) = 1 8 ijklmn = (l) 1 2 aijklm ij aj kl kl am im (l = 5. In the canonical Hartree–Fock case F N ˆN = W ˆ as the sole perturbation. In nonvanishes.4. fai connecting hole states .6. and we are left with W canonical Hartree–Fock the matrix elements fia . n = 3) . represent the diﬀerence between 0|V 2 3 ˆ2 |0 3 . n = 1) . n = 1) . The last diagram represents 0|V ˆ operator of (3. As will be seen in Chapter 5. All 18 diagrams. h = 5.41) ˆN and V ˆ that arises from where we have used the cancellation between U N the choice upq = − i pi qi . all the unlinked diagrams will not contribute to the perturbation treatment and will not need to be considered. The one-electron part is represented diagram× vertex. ij ij kl kl mn mn (l = 6.3. 4 pqrs (4. the normal-product perturbation operator ˆ − 0|V ˆ |0 can be written in the form ˆN = V V o ˆN = F ˆN ˆN V +W = pq o fpq + 1 pq rs {p ˆ† q ˆ† s ˆr ˆ} .178) and/or others represent contributions involving the V N ˆ powers of 0|V2 |0 . h = 5. 4.

135).43).43) ˆ o by Even in the non-HF case it is possible to simplify the expansion of F N performing a semicanonical transformation of the orbitals (Handy. . so that the one-electron perturbation operator has the diagrammatic expansion o ˆN = F × + × (noncanonical HF). (4. ˆ operator of (3. to be discussed in Section 12. which diagonalizes the hole–hole and particle– ˆ separately (leaving the reference function invariant).42) In the general (non-HF) case. Head-Gordon et al. o ˆN F = × + × + × + × (general case). 1989).4 Two-particle operators 129 with particle states also vanish. (4. the full expansion is required. × vertex represents the f More generally. the o ˆ which is equal to f only if εp = fpp holds for all p. Pople.3. particle blocks of F thus eliminating the ﬁrst two diagrams in Eq. (4. This is the form of the partitioned Hamiltonian used in generalized MBPT (Bartlett 1995).4.

ij.....2) ˆ 0 does not change the state on which it operates. (5. (0) (0) (5. except for division Thus R by the energy denominator.. ij... ij.... εab. (5. = E0 − E|Φab.... When R on any state |J other than |0 the result is ˆ 0 |J = R I |I I |J (0) E0 − (0) EI = I |I δIJ (0) E0 − (0) EI = |J 1 (0) E0 − EJ (0) . ˆ0 R |Φab. we need a diagrammatic ˆ 0 . ij. This can be represented point of action of R symbolically as follows: |Φab. (5..... ij... . = εi + εj + · · · − εa − εb − · · · . We note that representation of the resolvent R ˆ0 = R ˆ Q (0) E0 ˆ0 −H = I |ΦI ΦI | E 0 − EI (0) (0) .. a i b j ··· a i b j ··· = |Φab..1 Resolvent operator and denominators In order to represent diagrammatically the various terms in the perturbation expansions for the energy and wave function.1) ˆ 0 operates the sum being over all states |I ≡ |ΦI diﬀerent from |0 .5 Diagrammatic expansions for perturbation theory 5... The only R0 line minus those of the particle lines crossed by the R 130 . ij.4) is equal to the sum of the orbital energies of the hole lines crossed by the ˆ 0 line..3) where the energy denominator εab. and any particle or hole lines present below the ˆ 0 continue unchanged above it.

as in Chapter 2. ˆ0V The equivalence of the two forms in (5.186)) E (1) = i fii − 1 2 ij ij ij = × − (5. rather than the where we have used W ˆ ˆ W = WN of Chapter 4.8) includes a one-electron term. the W ˆ in (5.5. ˆR ˆ0V E (2) = 0|V (5. because R ˆ |0 ˆR ˆ0W The diagrammatic representation of the second-order energy 0|W 2 ˆ is very similar to that of 0|W |0 . an additional resolvent line is implied above the highest vertex.8) follows from R ˆ 0 |0 = 0. In wave function diagrams.3 Second-order energy 131 ˆ 0 |0 = 0.135). it being understood that an energy denominator is to be read oﬀ the diagram between each successive pair of vertices.and ﬁrst-order energies (see Section 3.5) (5.6) ˆ |0 = − 1 E (1) = 0|V 2 ij (using the ASG representation) and. ij ij = − (5. otherwise.3 Second-order energy The correlation energy begins with second order.7) using the ˆ operator of (3. 5. × vertex to represent the f 5. F ˆo . More generally. as in ˆ0 R . except for the addition of the resolvent . which are open at the top. Eref = E (0) + E (1) . so that we have no terms in which the exception to this result is R resolvent line goes through a vacuum gap. if εp = fpp is not assumed.2 First-order energy The zero.137)) ˆ |0 = 0|W ˆR ˆ0W ˆ |0 . W N ˆ |0 = R ˆ0W ˆ |0 . of course.5) are given by E (0) = i εi .8) ˆ = V ˆ − 0|V ˆ |0 . In the canonical Hartree–Fock case the two usages of ˆ are equivalent. The convention is to omit the resolvent lines. we have (see (3. We shall show the resolvent lines explicitly in some cases for pedagogical reasons. in the form (see (2.

5. The absence of single excitations from the second-order energy expression in the HF case is a reﬂection of the Brillouin theorem (Section 1.9) abij (the resolvent line has no eﬀect on the determination of the sign and weight). As for the second-order energy diagrams in the HF case. In the general non-HF case the intermediate states include single excitations |Φa i . In the canonical HF case we get three ASG diagrams. and thus (5. since i|f any HF solution. o ˆ contribution. It is evident that the only allowed intermediate states in this case (those crossed by the resolvent line) are double excitations |Φab ij .132 Diagrammatic expansions for perturbation theory ˆ vertices. . ˆ0 R 1 4 ij ab ab ij εab ij (a) (i) (j ) (b) = (5. (5.12) ˆ 3 |0 . here too all intermediate states are double excitations.11) and there are no direct contributions of single excitations (i. In the non-HF case there are eleven additional ASG diagrams. 5.e.5).138). ˆ |Φa = 0|W ˆ |Φa = 0 0|H i i (for any i. which states that for Hartree–Fock reference functions. none through the second order terms). 6 and 7. ˆR ˆ0W E (3) = 0|W (5. represented by the In the non-HF case we must add the F N diagram × ˆ|a a|f ˆ|i i|f ˆ0 R = . except that we need to include two This result is similar to that for 0|W resolvent lines. for which the Brillouin theorem does not generally apply. a) (5. 5. In the Hartree–Fock case we have just one line between the two W ASG diagram for this order. the third-order energy is given by ˆR ˆ0W ˆ |0 . canonical or otherwise.9) still holds.10) εa (a) (i) i ai × ˆ|a = 0 for This expression vanishes in the noncanonical HF case. Two of these diagrams. shown with their interpretation in Fig.1. This analysis applies to unrestricted HF and to closed-shell restricted HF but not to restricted open-shell HF. shown in Fig.2 (with the resolvent lines omitted).4 Third-order energy According to (2.

cd εab ij εij abcdij ˆ0 R 2 ˆ0 R (a) (i) (j ) (b) = 1 8 ij ab kl ij ab kl . 5. ab εab ij εkl abijkl (k ) (l) ˆ0 R 3 ˆ0 R (a) (i) (b) (j ) = − abcijk ij ab kb ic ac kj . ˆ|i or i|f ˆ|a ) do not involve particle–hole one-electron integrals (such as a|f and contribute also in the noncanonical HF case. Non-HF ASG diagrams for the third-order energy. 5. . ac εab ij εkj (k ) (c) Fig.4 Third-order energy 133 ˆ0 R 1 ˆ0 R (i) (a) (b) (j ) = (c) (d) 1 8 ij ab ab cd cd ij . they involve doubly excited intermediate states only. Canonical HF-case ASG diagrams for the third-order energy.2.5.1. The other diagrams contribute in the non-HF case only and involve some singly excited intermediate states. × × × × × × 9 × × × 14 7 8 × × × 11 × 12 13 × × 4 × × 10 5 × 6 Fig.

5. After these two operations each particle or hole line of the original diagram is again a particle or hole line. it is easy to see that the eﬀect of the transformation on the algebraic interpretation of the diagram is an interchange of the bra and ket indices in each integral with no change in the values of the denominator or the numerical coeﬃcient: × = (b) (c) (b) (c) (i) ˆ0 R (a) × (j ) (5.134 Diagrammatic expansions for perturbation theory The advantages of the Hartree–Fock case in simplifying the formalism seem obvious.5 Conjugate diagrams An interesting diagram symmetry operation consists of turning a diagram upside down (time reversal ) followed by reversal of all arrow directions ˇ ıˇ (particle-hole inversion ) (Paldus and C´ zek 1974). compared with six indices for the three HF-case diagrams. 5. By carrying the original dummy summation label with each line to its new position in the transformed diagram. in its new position in the new diagram. The above non-HF diagrams have between three and ﬁve summation indices each. The diagram resulting from this transformation (in any of its equivalent forms) is said to be conjugate to the original diagram. but in fact have little computational signiﬁcance beyond the proliferation of terms. and thus require signiﬁcantly less computational eﬀort.2. respectively.13) ˆ0 R (j ) = 1 2 ˆ|i ij bc bc aj a|f . the numerical values of the two diagrams are the complex conjugates of each other. bc εa i εij ˆ0 R (i) ˆ0 R (a) abcij abcij Consequently. An example of a pair of conjugate diagrams is given by the non-HF thirdorder energy diagrams 4 and 8 in Fig. . bc a εij εi 1 2 ˆ|a aj bc bc ij i|f .

while the rest are self-conjugate. We shall see many more examples of conjugate diagrams in the fourth-order energy.1.15) has two quasiloops.133). thus forming two quasiloops. in the sense of Chapter 2. two hole lines and two pairs of equivalent lines. The diagram in (5.e. Hermitian ). hence the factor 1 4 (as usual the labels in parentheses are included only as a convenience to identify dummy summation indices in the corresponding algebraic expression. The ﬁrst-order wave function is given by (2. and all lines in this diagram are to be taken as unlabeled). In where. |Φab ij in this case. in which case the sum of the two diagrams is equal to twice the real part of the value of either. we have used W the Hartree–Fock case this is the same as the two-electron normal-product ˆ N and we have just one diagram. In regard to the determination of the weight factor. the pairs 4 and 8. representing double-excitation operator W contributions: (a) (i) (j ) (b) = ˆ0 |Ψ(1) = R 1 4 abij ab ij |Φab ij . The three HF-case diagrams for the third-order energy.6. Fig. 5 and 9 and 10 and 12 are mutually conjugate. taking twice the real part of its value as the total value of the pair. two unlabeled open lines are considered equivalent if they both begin at the same vertex or both end at the same vertex.2.6 Wave-function diagrams 5.14) ˆ = V ˆ − 0|V ˆ |0 . Thus it is enough to evaluate just one of each pair of conjugate diagrams. Among the non-HF third-order diagrams. εab ij (5.1 First-order wave function Before proceeding to higher orders for the energy.6 Wave-function diagrams 135 Each MBPT energy diagram is either self-conjugate (i. we will digress to look at the diagrammatic representation of the wave function. Fig. As stated in Section 5. according to the convention one omits the resolvent line. in which case its value is necessarily real.5. 5. with the understanding that in open diagrams .15) For the purpose of counting loops we consider the open lines to be paired at the top according to the pairing pattern in the corresponding slater determinant. ˆ0W |Ψ(1) = R (5. 5. 5. ˆ |0 .1. are all self-conjugate. or conjugate to another energy diagram of the same order.

which include ﬁve double excitations (9–12 and 19).e. Here we have one quasiloop. ˆ0W |Ψ(2) = R (5. Diagrams 9 and 10 contribute in . has no open lines.17) In the canonical HF case we have eight ASG diagrams. 17. Equation (5. The second-order wave function expression has no renormalization terms. replaces f ai ai 5.2 Second-order wave function The second-order wave function is given by (2. 18) and two triples (15. they are followed by two singles. 16). 5. that case. four singles (13.6. hence the factor +1. 14. such a diagram has two or more disconnected parts. As the name implies.134).16) still holds in the more general case in which f o o ˆ because we have f = f .16) o because a and i cannot be equal) which adds contributions (since fai = fai from singly excited states. This diagram does not contribute in the noncanonical Hartree–Fock case because the fai integrals vanish in ˆ. as in . The quadruple-excitation diagram 8 is an example of a disconnected diagram. two triples and one quadruple. If any part of a disconnected diagram is closed. leading to the additional diagram ˆ0 R (a) (i) = × ai ˆ|i a|f |Φa i a εi (5. i. (3. It will be shown later that unlinked diagrams cancel with the renormalization terms while linked disconnected diagrams remain in the wave function expansion. one hole line and no equivalent lines.135).136 Diagrammatic expansions for perturbation theory there is one such line above the highest vertex as well as between each pair of successive vertices. The non-HF contributions to |Ψ(2) are represented by the eleven diagrams in Fig.3. The ﬁrst three diagrams represent doubly excited ﬁnal states. 5. then the diagram is called unlinked. In the non-HF case we need to include a one-electron vertex.4. ˆR ˆ0W ˆ |0 . shown with their algebraic interpretation in Fig. and no unlinked diagrams appear.

ab εabc ε ijk il cd kl ab ij |Φabcd ijkl . the noncanonical HF case but diagrams 11–19 do not because they involve ˆ|i or i|f ˆ|a . bc εab ij εik aj cb cb ij |Φa i .5. cb εa ε i ij kj ib ab kj |Φa i . ad εabc ijk εij lc jk ab il |Φabc ijk . Canonical HF-case ASG diagrams for the second-order wave function. which vanish in this case. all disconnected wave-function diagrams can be re-expressed as products of connected wave-function diagrams each consisting of one of their disconnected parts. 5. This factorization is at the .3. ab εab ij εkl ak cj cb ik |Φab ij . Diagrams integrals of the form a|f 15.6 Wave-function diagrams 137 1 (i) (a) (b) (c) (d) (j ) = 1 8 abcdij ab cd cd ij |Φab ij . 16 and 19 are disconnected but not unlinked and are legitimate parts of the wave-function expansion. As will be shown in Section 9. cd εab ε ij ij kl ij ab kl |Φab ij . ab εa i εkj bc dk ad ij |Φabc ijk . ab εabcd ijkl εij 2 (a) (i) (j ) (k ) (l) (b) = 1 8 abijkl 3 (i) (a) (j ) (c) (a) (c) (b) (i) (k ) (j ) (k ) (b) = − abcijk 4 (i) (j ) = 1 2 abcij 5 (a) (b) =− 1 2 abijk 6 (a) (i) (j ) (b) (c) (k ) (d) (j ) (k ) (c) (l) (k ) (l) (d) = 1 4 abcdijk 7 (i) (a) (b) =− 1 4 abcijkl 8 (a) (i) (j ) (c) (b) = 1 16 abcdijkl Fig.4.

5. These diagrams have equal numerators but. where new features will appear for the ﬁrst time. non-HF ASG diagrams for the second-order wave function.139). of (5. As an example we consider diagrams 15 and 16 of Fig. (2.138 Diagrammatic expansions for perturbation theory × 9 10 × × 11 12 × 13 × × 14 × 15 × 16 17 × × 18 × × × 19 × Fig. heart of the exponential Ansatz of coupled-cluster theory (Chapter 9).h. This is the ﬁrst case in which we have . This is an example of a type of factorization that is very common in MBPT.4. Combining their denominator factors we get 1 1 1 + abc c = abc abc ab εijk εij εijk εk εijk 1 1 + c ab ε εij k ab 1 εc 1 k + εij = abc ab c = ab c εijk εij εk εij εk ab abc because εc k + εij = εijk .1 Energy formula Now we return to the energy expansion and look at E (4) .s. 5.4.7. their denominators are diﬀerent. because of the diﬀerent time sequences of the vertices.7 Fourth-order energy 5. 5. Thus the sum of these two second-order diagrams can be factored into a product of two independent sums corresponding to their two ﬁrst-order disconnected parts: × + × = × × (5. and other examples will be seen in the discussion below of the fourth-order energy and in Chapters 6 and 7.18) (note that the upper cross on the r.18) is a multiplication sign).

Three of these skeletons (4-0-0.7.5.3. 0-0-4) are unlinked. In the last line of the ˆ0W ˆR ˆ0W using |Ψ(1) = R (2) ˆ 0 operators equation the E expression has been inserted between the two R 2 (1) (1) ˆ in R0 (which derived from Ψ |Ψ ).4. so k + l + m = 4. and the additional diagrams for other cases are derived in subsection 5. The ﬁrst unlinked skeleton has . The resulting 15 skeletons are shown in Fig.4. and in order to avoid missing some terms or including redundant equivalent terms.5. Because of the greater complexity when dealing with four vertices. where it will be shown that it cancels with the unlinked diagrams of the principal term. The canonical HF case is discussed in the present subsection. we classify them according to their connection patterns.7. generalizes this result to all orders. To facilitate systematic generation of the Hugenholtz skeletons for the principal term of E (4) . l lines to the third vertex and m lines to the bottom vertex. The bracket insertion form of the renormalization terms will help in demonstrating this cancellation. we shall follow the procedure outlined in subsection 4. 5. each vertex must have four connecting lines. 0-4-0. showing the bracket insertion form of the renormalization term. to be proved in Chapter 6. so that the ﬁnal expressions are sums of linked diagrams only. 5. = 0|W (5.19) ˆ |0 and E (2) = 0|W ˆ |0 . using the designation k -l-m to describe a skeleton (or diagram) in which the top vertex is connected by k lines to the second vertex (counting from the top). We have ˆR ˆ0W ˆR ˆ0W ˆ |0 − E (2) Ψ(1) |Ψ(1) ˆR ˆ0W E (4) = 0|W 2 ˆ ˆR ˆ0W ˆR ˆ0W ˆ |0 − 0|W ˆR ˆ0W ˆ |0 0|W ˆR ˆ0 ˆR ˆ0W W |0 = 0|W ˆR ˆ0W ˆR ˆ0W ˆR ˆ0W ˆR ˆ0W ˆ |0 − 0|W ˆR ˆ 0 0|W ˆ |0 R ˆ0W ˆ |0 . Since we are dealing with the canonical HF case.7 Fourth-order energy 139 a renormalization term.7. The linked-diagram theorem.2 Diagrams for E (4) in the canonical HF case We shall now derive the diagrams that describe the principal term of the fourth-order energy. The renormalization term will be discussed in subsection 5.4 and begin by listing all Hugenholtz skeletons. In the derivation of the diagrammatic expression for the principal term of E (4) we shall ﬁnd some unlinked diagrams (diagrams that have disconnected closed parts) and will show that these exactly cancel the renormalization term. for both the energy and the wave function. The same requirement of four lines at each vertex determines the remaining connections that do not involve the top vertex.

The classiﬁcation based on the intermediate state has important practical ramiﬁcations.6. The other two unlinked skeletons generate diagrams that cancel with the renormalization term. The corresponding antisymmetrized Goldstone diagrams. because the diagrams for the diﬀerent classes allow diﬀerent factorization patterns for the corresponding sums. 5. 5. Hugenholtz skeletons for the principal term of E (4) for canonical HF. Only the middle intermediate state. These factorizations and other computational considerations are discussed in Chapter 7. leading to substantial diﬀerences in the computational eﬀort required for their evaluation. a vacuum gap and thus cannot contribute.10 classiﬁed by the type of intermediate states involved.140 4-0-0 Diagrammatic expansions for perturbation theory 3-1-0 3-0-1 2-2-0 2-1-1 2-0-2 1-3-0 1-2-1 1-1-2 1-0-3 0-4-0 0-3-1 0-2-2 0-1-3 0-0-4 Fig.6.7–5. we obtain the 41 Hugenholtz diagrams in Fig. 5. with two incoming and two outgoing lines at each vertex.4. using the same numbering as in Fig. 5. . diﬀers in its excitation level between the diﬀerent classes of HF-case diagrams.7. while the other (linked) skeletons generate from two to six distinct diagrams each. are shown in Figs.5. and are represented by the denominators in the corresponding algebraic expressions. between the second and third vertices. Ignoring the vacuum-gap skeleton and inserting arrows in the other skeletons in all distinct ways. as will be shown in subsection 5. The two unlinked skeletons (33 and 41) generate just one diagram each. These intermediate states are deﬁned by the (omitted) resolvent lines that cross the diagrams between the vertices.

. 5. Hugenholtz diagrams for the principal term of E (4) for canonical HF.7 Fourth-order energy 3-1-0 3-0-1 2-2-0 2-1-1 141 1 2 3 4 5 6 2-0-2 7 8 9 1-3-0 10 11 12 1-2-1 13 14 15 16 17 18 1-1-2 19 20 21 22 23 24 1-0-3 25 26 0-4-0 27 0-3-1 28 29 30 0-2-2 31 32 33 0-1-3 34 35 0-0-4 36 37 38 39 40 41 Fig.6.5.

5.8). 29 and 30. The double-excitation diagrams 14 and 15 are particle and hole ladders. include a singly excited intermediate state (between the second and third vertices). Double-excitation ASG diagrams for the principal term of E (4) .142 Diagrammatic expansions for perturbation theory 1 2 3 4 Fig. 5. respectively. shown in Fig. Single-excitation ASG diagrams for the principal term of E (4) .10). 16 that include a triply excited intermediate state (Fig. which will be shown to cancel with the renormalization term. 8 and 9. 10 and 11. The ﬁrst four diagrams. 5 and 6. The latter include the two unlinked diagrams. 33 and 41. 5 6 7 8 9 10 11 12 13 14 15 16 Fig.7. . All other diagrams are self-conjugate. 23 and 24.9) and nine that include a quadruply excited intermediate state (Fig.8. 21 and 22. 17 and 18.7. 5. 5. 5. 5. Diagrams 7 and 16 are usually referred to as ring diagrams. There are 12 diagrams with doubly excited intermediate states only (Fig. There are eight conjugate pairs: diagrams 1 and 2.

5. Triple-excitation ASG diagrams for the principal term of E (4) . a designation that can be understood by considering their equivalent forms.5. . and .7 Fourth-order energy 143 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Fig.9.

since a Hugenholtz skeleton may produce more than one distinct ASG skeleton. Similarly. respectively.11. 5. in order. to the upside-down images of those with a single one-electron vertex at the bottom (the ﬁrst six in the third row). Only six of these skeletons. generating 22 diagrams (those that do not involve a particle– ˆ|a or a|f ˆ|i ). The other double-excitation diagrams involve more complicated connection patterns. Each of these skeletons generates between one and four distinct diagrams.11 are equivalent to the upside-down image of other skeletons and these generate sets of pairs of conjugate diagrams. 5. It is left as an exercise for the reader to generate the non-HF diagrams from the skeletons. since each Hugenholtz diagram produces a single distinct ASG diagram.144 Diagrammatic expansions for perturbation theory 33 34 35 36 37 38 39 40 41 Fig. (To generate ASG diagrams from Hugenholtz skeletons. Thus the skeletons with a single one-electron vertex at the top (the ﬁrst six in the ﬁrst row) are equivalent. contribute in the hole one-electron integral of the form i|f noncanonical HF case. The alternative sequence through ASG skeletons is less straightforward.) Some skeletons in Fig. 5. the six skeletons with a single one-electron vertex at the second level are symmetry related to the . and generate 18 pairs of conjugate diagrams. resulting in a total of 162 linked diagrams.7. the preferred sequence is to generate Hugenholtz diagrams ﬁrst and then convert these to ASG diagrams.10.3 Non-HF diagrams for E (4) For the non-HF case we obtain the 69 additional linked Hugenholtz skeletons shown in Fig. Quadruple-excitation ASG diagrams for the principal term of E (4) . 5.

5. while the skeletons . Among the skeletons with two one-electron vertices there are two sets of ﬁve skeletons that are symmetry related to two other sets. six diagrams with that vertex at the third level. Non-HF linked Hugenholtz skeletons for the principal term of E (4) . generating another 18 pairs of conjugate diagrams.11. 5. generating 10 pairs of conjugate diagrams each.7 Fourth-order energy × × × × × × × × 145 × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × × Fig.

The large number of non-HF diagrams re-emphasizes the advantages of the HF case in simplifying the formalism but. each generating a single diagram. We begin with the canonical HF case. Non-HF unlinked Hugenholtz diagrams for the principal term of E (4) .20) abcd ijkl . In all. The two unlinked diagrams. diﬀer only in time ordering. There also are three pairs of conjugate diagrams generated from skeletons that are equivalent to their own upside-down image (two pairs with two one-electron vertices and one pair with four). 5. There also are six unlinked non-HF skeletons (not counting three with a vacuum gap). abcd cd εab ij εijkl εkl (5. 5. 33 and 41 (Fig. These diagrams are shown in Fig. A formal proof of the cancellation of all unlinked diagrams in all MBPT energy and wave function expressions with the corresponding renormalization terms will be given in Chapter 6.4 Cancellation of the unlinked diagrams in E (4) We now proceed to show that the unlinked diagrams in the fourth-order energy cancel with the renormalization term. the computational eﬀort involved in evaluating all the non-HF terms is small compared with the eﬀort required for the HF terms. As will be shown later. we obtain 71 pairs of conjugate diagrams and 20 self-conjugate diagrams. generating six pairs of conjugate diagrams each.12. again. with three one-electron vertices consist of two sets of three skeletons that are symmetry related to two other sets. they cancel with the non-HF part of the renormalization term.146 Diagrammatic expansions for perturbation theory × × × × × × × × × × × × × × × × Fig.12.10). 5. 5. Their algebraic expressions have the same numerators and diﬀer only in one factor in the denominator: ˆ0 R (a) ˆ0 R (i)(j ) (b) (c) ˆ0 R (k ) (l) (d) = 1 16 ij ab kl cd ab ij cd kl .7.

cd εkl εij εij εkl εij εkl the sum becomes 1 16 ij ab ab ij kl cd cd kl abcd ijkl εab ij εcd kl 2 ij ab ab ij 2 εab ij 1 4 kl cd cd kl εcd kl . . abcd ijkl Noting that εabcd 1 1 1 ijkl ab cd + kl = ab cd (εij + εkl ) = ab cd . εcd kl (5. abcd ab εab ij εijkl εij (5.7 Fourth-order energy 147 ˆ0 R (a) ˆ0 R ˆ0 R (i) (j ) (b) (c) (k ) (l) (d) = 1 16 abcd ijkl ij ab kl cd cd kl ab ij .5.21) Adding them together term by term we obtain 1 16 ij ab ab ij kl cd cd kl abcd εab ij εijkl 1 1 + ab cd εkl εij .22) The ﬁrst factor can be described diagrammatically in the form ˆ0 R (a) ˆ0 R (i) (j ) (b) . 1 = 4 abij cdkl The second factor on the right-hand-side of this expression is just the secondorder energy in the HF case. ˆ 0(c) E (2) = R (d) (k ) (l) = 1 4 cdkl kl cd cd kl .

In the non-HF case there are six additional unlinked diagrams. More explicitly. This result can be represented diagrammatically in the form ˆ0 R + = ˆ0 R (5.148 Diagrammatic expansions for perturbation theory which is equal to Ψ(1) |Ψ(1) in the HF case (the two resolvent lines represent the squared denominator). Note that no denominator line spans both parts of this diagram. At the same time.12.28) × . exactly canceling the renormalization term. 5.24) ij | kl ab cd 4 4 ε ε ij kl abij cdkl The sum of the two unlinked diagrams has thus been shown to be equal to E (2) Ψ(1) |Ψ(1) . noting that cd Φab ij |Φkl cdkl cd kl ij ab ba ij ab ji ba ji = − − + cd ab ba ab εkl εij εij εji εba ji = ij ab ij ab ij ab ij ab + + + ab ab ab εij εij εij εab ij ab ij εab ij (5. we can rewrite the ﬁrst factor in the form ij ab cd kl 1 1 = Ψ(1) |Ψ(1) . . shown in their Hugenholtz form in Fig. (5. (5. The complete diagrammatic description of E (2) is × + E (2) = .23) =4 cd (because Φab ij |Φkl = δik δjl δac δbd − δik δjl δad δbc − δil δjk δac δbd + δil δjk δad δbc ). Φab |Φcd (5.25) where “insertion” of the E (2) diagram simply means multiplication by E (2) .26) × while the ﬁrst-order wave function is |Ψ(1) = and its norm can be described by Ψ(1) |Ψ(1) ˆ0 R = ˆ R0 ˆ0 R + ˆ R0 + × × (5. there are additional diagrams in the expressions for E (2) and Ψ(1) |Ψ(1) in the renormalization term.27) ˆ0 R .

The complete renormalization term can therefore be described by E (2) Ψ(1) |Ψ(1) = × + × + × + × × × × × . × (5. The remaining three terms are canceled by the six non-HF unlinked diagrams of Fig.30) . one due to Ψ(1)| and the other to |Ψ(1) . as follows: (a) (i) (j ) (b) (c) × + (a) (k ) × = 1 4 1 4 (i) (j ) (b) (c) × (k ) × 1 1 + ab c εk εij abc ijk ˆ|c ab ij c|f ˆ|k ij ab k |f abc εab ij εijk ˆ|c c|f ˆ|k ij ab ab ij k |f 2 c (εab ij ) εk = abc ijk = (a) (i) (j ) (b) (c) × (k ) . as shown earlier. 5.29) it being understood that there is a resolvent line both above and below each insertion.29) is the HF part.12. must be shown explicitly in this equation to indicate the squared denominator). which is cancelled by the two HF-case unlinked diagrams.5.7 Fourth-order energy 149 (the repeated resolvent lines. The ﬁrst term in (5. (5.

of Fig. there are four equivalent ways to assign the pair of equal particle labels in the latter diagram. focusing on the role of exclusion principle violating (EPV) terms. .33) a i j b b k l d =+ ij ab kl bd bd kl ab ij .34) We have not included the weight factors. 5. abbd ab εab ij εijkl εij (5. respectively.150 Diagrammatic expansions for perturbation theory and. (5. 5.7. contributing to diagrams 39 and 41.5 Role of the EPV terms We shall now examine the cancellation of unlinked diagrams in more detail. abbd ab εab ij εijkl εij (5. one linked and one unlinked. × × + × × × × × + × × × × = × × = × × × × . since both are EPV terms. however. In unrestricted summations there would be a coeﬃcient 1 4 for the linked term 1 and a coeﬃcient 16 for the unlinked term.31) Thus we have obtained full cancellation of the unlinked diagrams with the renormalization term.10: b a i j b k l d =− ij ab kl bd bd kl ab ij . so that the weight factors balance at the end.32) × . (5. since all lines are labeled. Consider two EPV quadruple-excitation terms. These two terms obviously cancel and this is as it should be. similarly.

b a i i c k l d =− ii ab kl cd bd kl ac ii = 0. (5. . Thus the unlinked EPV terms that are canceled by the renormalization term are not legitimate contributions in the ﬁrst place and have only been included because of the presence of cancelling linked EPV terms. abcd ac εab ii εiikl εii a i i b c k l d =+ ii ab kl cd cd kl ab ii = 0. by the renormalization term.36) abcd ab εab ij εijkl εij Note that having equal labels on equivalent lines produces vanishing contributions. abcd ab εab ii εiikl εii and thus such diagrams can be ignored in considering the cancellations.35) abcd ac εab ij εijkl εij a i j b c k l d =+ ij ab kl cd cd kl ab ij .37) where S11 = Ψ(1) |Ψ(1) and the subscript DJ (disjoint) indicates the sum of those contributions in which the diagrams for E (2) and S11 have no labels in common. (5. (5. After cancellation of all the unlinked contributions.7 Fourth-order energy 151 This same cancellation does not occur for the non-EPV terms (b = c): b a i j c k l d =− ij ab kl cd bd kl ac ij . We can divide up the renormalization term into two parts: −E (2) S11 = (−E (2) S11 )CJ + (−E (2) S11 )DJ .5. including the EPV terms. while CJ (conjoint) indicates the sum of the contributions in which the two diagrams have one or more labels in common. the “illegitimate” linked EPV terms in the principal term remain.

b (−E (2) (−E (2) S11 )DJ = − − j + + j j b b S11 )CJ = b + + ··· j + b b + j j (5.8 Linked-diagram theorem The cancellation of all the unlinked diagrams against the renormalization terms in Rayleigh–Schr¨ odinger perturbation theory is formalized by the linked-diagram theorem (Goldstone 1957). which states that the energy and . (5. leaving behind the linked EPV terms. (5.38) while the conjoint part cancels the EPV unlinked-diagram contributions. they must be equal to each other. (non-EPV terms only). so that these linked EPV terms can be regarded as the remnant of the renormalization term that was not canceled by “legitimate” (non-EPV) unlinked terms.40) (4) 5. The total quadruple-excitation contribution can therefore be expressed in either of two ways: EQ = (non-EPV linked quadruple-excitation diagrams) + (EPV linked quadruple-excitation diagrams) = (non-EPV linked quadruple-excitation diagrams) + (−E2 S11 )CJ .152 Diagrammatic expansions for perturbation theory The disjoint part of the renormalization term cancels the non-EPV unlinked-diagram contributions.39) (all the linked EPV terms). Since either (−E (2) S11 )CJ or the sum of the linked EPV terms cancels the sum of the unlinked EPV contributions.

as a sum of linked diagrams only.9 Numerical example In order to get a feel for the relative magnitudes of the various contributions to the energy.5.910 mEh . we shall consider the third-order wave function. (5. 5. we shall look at an example from a set of calculations for the H2 O molecule at its equilibrium geometry in a (5s4p2d/3s1p) Slater-type basis (39-STO) (Bartlett. Again. as will be discussed in Section 5.10. Wilson 1979. Bartlett.41) (5. whose sum can be expressed as an insertion. Shavitt and Purvis 1979. This cancellation is even more striking if we look just at the quadruple-excitation contribution. in the case of the energy. (5. in each order. through fourth order. is equivalent to a limitation to connected diagrams). L L. We see in the table that the fourth-order renormalization term cancels 61% of the principal term in this example. As an example of the cancellation of unlinked diagrams in the wave function expansion. are given in Table 5. The cancellation of the unlinked diagrams is closely related to the extensivity of the Rayleigh– Schr¨ odinger perturbation theory. E (n) = |Ψ (n) = ˆ (R ˆ0W ˆ )n−1 |0 0|W ˆ0W ˆ )n |0 (R . Purvis et al. |Ψ(3) = R (5. The sum of all quadruple-excitation diagrams in the principal term is −13.42) where the subscript L indicates the limitation to linked diagrams (which.9 Numerical example 153 the wave function can be expressed. Explicitly. . Proof of this theorem is deferred to Chapter 6.43) In the canonical HF case we ﬁnd two unlinked diagrams. The non-HF case is left as an exercise for the reader.135)): ˆ0W ˆR ˆ0W ˆR ˆ0W ˆR ˆ0W ˆ |0 − R ˆ 0 0|W ˆ R ˆ0W ˆ |0 . Cole. it is understood that a resolvent line occurs both above and below the insertion.1. the lowest wave-function order in which unlinked diagrams and a renormalization term occur (see (2. The correlation energy contributions for this example. 1987). + = .44) which is equal to the renormalization term. Bartlett 1981.

014 −296.780 −3. The results in Table 5. Shavitt and Purvis 1979. While the computational eﬀort for the fourth-order quadruples is approximately proportional to the sixth power of the number of orbitals.116 mEh .130 +17. Purvis et al.1.g. Bartlett. Shavitt. the eﬀort for the triples is approximately proportional to the seventh power.206 −11. We can use the H2 O 39-STO example to obtain a quantitative estimate of the role of the EPV terms in the cancellation of the unlinked diagrams.324 −7.5. Bartlett.116 −11. early fourth-order MBPT calculations (see e. leaves a linkeddiagram contribution of +3. 1979) left out the triples. Wilson 1979. using a 39-STO calculation (Bartlett.154 Diagrammatic expansions for perturbation theory Table 5. In fact. Purvis et al.206 mEh . producing a result sometimes referred to as MBPT(4)SDQ or SDQ-MBPT(4). Bartlett 1981.7.035 E (2) +E (4) Another breakdown of E (4) : principal term renormalization term E (4) −28. including the quadruples without the triples may actually produce less accurate results than leaving both out.1 show not only that the triples’ contribution may be greater in magnitude than that of the linked quadruples but also that the two contributions may be of opposite sign. .014 deducting the value of the unlinked diagrams. Thus removal of the unlinked quadruple-excitation diagrams in E (4) leads to a change in sign of the quadruple-excitation contribution.033 −4. Cole. Using an approximate evaluation of the conjoint part of the renormalization term based on a CEPA-3 estimate (Bartlett. Contributions to the correlation energy in the water molecule in millihartrees (mEh ). −17. It will be shown in Chapter 7 that the determination of the quadrupleexcitation contribution to the fourth-order energy requires substantially less computational eﬀort than the determination of the triple-excitation contribution. 1987) E (2) E (3) E (4) −281. as discussed in subsection 5. In fact.863 +3.241 singles doubles triples linked quadruples +E (3) −2.

015 (−E (2) S11 )DJ +10.5.015 Cancellation EPV +7.015 Fig. Role of EPV terms in the cancellation of unlinked quadruple-excitation diagrams in E (4) for the 39-STO H2 O calculation (energies in millihartrees).116 (b) Linked non-EPV quadruple-excitation terms −3.895 (−E (2) S11 )CJ +7.895 Final result: (4) EQ = 3.101 Unlinked quadrupleexcitation diagrams non-EPV −10.206 0 Unlinked non-EPV quadrupleexcitation terms Cancellation −10.101 Final result: (4) EQ = 3.101 non-EPV −3. 5. .13.9 Numerical example (a) Linked quadrupleexcitation diagrams 0 EPV −7.206 155 −E (2) S11 +17.

the total energy is. If the Rayleigh–Schr¨ odinger perturbation theory expanˆ ˆ ˆ sion. proportional to N . (4) (linked non-EPV quadruple excitations) (linked EPV quadruples = (−E (2) S11 )CJ ) 5. so that the linked quadruple-excitation contribution can be divided up as follows: EQ = −3.10 Unlinked diagrams and extensivity 5. leaving the linked diagrams (both EPV and non-EPV terms) as the ﬁnal quadruple-excitation energy. (This does not hold for BWPT.101 mEh .206 mEh . justifying the interpretation of the linked EPV terms as the uncanceled remnant of the renormalization term. In part (b) we leave out the EPV terms. Thus. with size (extent) proportional to some count parameter N (the number of identical subunits).4. as discussed in subsection 5. The two interpretations of the role of the EPV terms in the cancellation of the renormalization term and unlinked quadruple-excitation diagrams in fourth-order MBPT.13.10. we ﬁnd (−E (2) S11 )CJ = 7. leaving the linked non-EPV terms plus the conjoint part of the renormalization term as the ﬁnal quadruple-excitation energy. This conjoint part can be re-expressed in the form of linked EPV terms.156 Diagrammatic expansions for perturbation theory Shavitt and Purvis 1979). here the unlinked non-EPV terms are canceled by the disjoint part of the renormalization term. 5.015 mEh .1 Extensivity implications For an extensive system. in which E is not expanded in powers of λ. converges for a range of values of λ then it must also provide a total converged energy that is proportional to N for each value of λ in that range. (5.7.895 mEh + 7. order by order. E = E (0) + λE (1) + λ2 E (2) + · · · . (−E (2) S11 )DJ = 10. In part (a) of the ﬁgure all the unlinked quadruple-excitation diagrams (the EPV and non-EPV terms) are canceled by the complete renormalization term. if the energy is expanded in powers of λ. with H = H0 + λV . are given for the 39-STO H2 O calculation in Fig. of course.3.101 mEh = 3.5.45) then each order E (n) must be proportional to N so that the RSPT energy is extensive. as demonstrated in subsection 2.) .

In fact. Thus nonzero contributions to the sum represented by any diagram are obtained only from terms in which the labels of the four or two lines connected to each vertex all belong to the same atom. q ) all belong pq rs and p|f to the same atom.4. s (or p. so nonzero contributions are obtained only from terms in which all labels in this connected piece belong to the same atom. ε0 − α (5. The key point is that. 5.2 The N-dependence of diagrams We can use the example of N noninteracting He atoms to examine the dependence of all MBPT diagrams on the size of the system. To ﬁnd the behavior of S11 we go back to the example of N noninteracting He atoms considered in subsections 2. As a result of this analysis we ﬁnd that −E (2) S11 ∝ N 2 . The factor E (2) is obviously proportional to N .4. with which the renormalization term cancels.10.48) which implies that the unlinked diagrams in E (4) . i.46) and thus S11 = Ψ(1) |Ψ(1) = i. the integrals ˆ|q for this example vanish unless p.3 and 2.5. on the number of atoms N . (5.10 Unlinked diagrams and extensivity 157 Now consider the renormalization term in E (4) . r. i.e.j β |β |2 β∗ Φi |Φj =N ε0 − α ε0 − α (ε0 − α)2 (5. For this example the RSPT ﬁrst-order wave function is given by |Ψ(1) = i |Φi ˆ |Φ0 Φi |V E 0 − Ei (0) (0) = i |Φi β∗ . are also proportional to N 2 . as discussed below in subsection 5. −E (2) S11 . .10. It is only by this cancellation that extensivity can be restored.e.49) (5.2.8. q. because the atoms do not interact. and so any model for the energy that consists of a sum of linked diagrams is properly extensive. it can be shown more generally that linked energy diagrams are always extensive (provided that the complete diagram is evaluated).47) or S11 ∝ N . In any connected piece of a diagram all vertices form a connected network.

.e. where k is the number of closed connected pieces in the diagram. The linked-diagram theorem states that each order of the energy is given by a sum of closed linked (and therefore connected) diagrams only. a nonzero contribution requires all ﬁxed and internal labels of each connected piece to belong to the same atom. to products of the individual connected pieces (with an extra denominator for each insertion line). The factorization decouples the connected pieces of the disconnected diagrams by eliminating denominators involving more than one connected piece. so the sum over all internal labels is independent of N .8). As examples we show two seventh-order unlinked diagrams. the sums over all disconnected diagrams can be rewritten in terms of sums of products of connected diagrams and insertions.7. diagrams each consisting of one closed part. to be stated in detail and proved in Chapter 6.4) and in the third-order wave function (Section 5. We have seen examples of this factorization in the demonstration of the cancellation of unlinked diagrams in the fourth-order energy (subsection 5. For a closed linked diagram (or insertion) the sum over the labels of all atoms is thus proportional to N .e. one open and one closed: . For an open linked diagram. for both the wave function and the energy. shows that diﬀerent time-orderings of an unlinked and/or disconnected diagram can be collected into sets (deﬁned by the relative positions of the top vertices of their separate connected pieces) whose sums are equal to connected diagrams with insertions. The factorization theorem therefore leads to the conclusion that the value of any MBPT diagram. i. while each order of the wave . i. The ﬁrst (open) diagram has two closed parts and its value (for the coefﬁcient of any speciﬁc term in the wave function expansion) is proportional to N 2 while the second (closed) diagram has three closed parts and a value proportional to N 3 . with ﬁxed labels on the open lines (representing the coeﬃcient of a given Slater determinant). The sum of the terms of a connected diagram or insertion over the labels belonging to any He atom in our example is independent of the presence of the other atoms and therefore of N . As a result.158 Diagrammatic expansions for perturbation theory The factorization theorem. is proportional to N k .

) In a certain sense.. 5. in each order. ij. mEh .. in each order.10 Unlinked diagrams and extensivity 159 function is given by a sum of open diagrams with no closed parts. etc. the full-CI energy and the inﬁnite-order converged PT energy n E (n) (if the PT series converges) are identical and provide the “exact” solution for the Schr¨ odinger equation within the space generated by the given basis set. mEh .576 mEh . as a sum of linked diagrams only. In fact. is independent of N (though the number of such terms is strongly dependent on N ). ECISDT (triples added). in the expansion of the wave function. This behavior is not atypical for a nonvariational method like PT. We see that both E2 and E3 are lower than the CI energies. Truncated PT sums. While the discussion of the N -dependence of diagrams has been presented here in terms of the example of N noninteracting He atoms. or CISD in the non-HF case. It therefore implies that each order of the energy is proportional to N .. At the same time. The full-CI energy can also be approached by a sequence of CI calculations with increasing excitation-level conﬁgurations: ESCF . again conﬁrming the extensivity of RSPT. the coeﬃcient of each term |Φab. ECID (with all double-excitation conﬁgurations). E2 and E3 can be seen as approximations to CID.780 E3 = E (2) + E (3) =−285.021 ECISD =−275. (Note that if only singles are used then ECIS = ESCF for SCF orbitals because of the Brillouin theorem. Thus the full-CI result provides a benchmark against which other methods using the same basis set can be measured. the result is much more general: the order-by-order extensivity property of Rayleigh– Schr¨ odinger perturbation theory requires that the energy and wave function be expressed.5..3 Relationship to conﬁguration interaction (CI) Within any given basis set. ECISDTQ (quadruples added). (5. ECISD (singles and doubles).10.021 ECID =−274. n En = m=2 E (m) .50) present successive approximations to the full-CI correlation energy Efull-CI . mEh . for the H2 O 39-STO calculation we ﬁnd (Bartlett and Shavitt 1977b). . E2 = E (2) =−281. since they involve only a sum over doubly excited conﬁgurations (singly excited conﬁgurations must be added for the non-HF case).

576 mEh given above.324 mEh 17.033 mEh −274. so that they only span doubly excited states |ΦI . this result can be worked out as follows (Bartlett and Shavitt 1977b. from fourth-order MBPT.116 mEh 12. These terms are canceled in full CI but. Note that the renormalization term is made up entirely of double-excitation contributions (in E (2) and in S11 . In fact.021 mEh . while CI is inherently an inﬁnite-order method. This analysis is not meant to imply that CID or CISD is a more desirable approximation than any particular order of RSPT. truncated CI at any level of truncation short of full CI necessarily retains non-extensive contributions. which are eventually canceled by higher-excitation non-EPV unlinked-diagram contributions (the EPV terms and the conjoint parts of the renormalization terms are absent .792 mEh −285. ECISD = −275.229 mEh −2.262 mEh This result compares well with the values ECID = −274. then we should get an approximation to CID only. triple and quadruple excitations as well. because the cancellation involves contributions from diﬀerent excitation levels. Shavitt and Purvis 1979): double excitation diagrams in E (4) + renormalization term (−E (2) S11 ) total double-excitation part of E (4) E (2) + E (3) fourth-order RSPT approximation to CID single excitation diagrams in E (4) fourth-order RSPT approximation to CISD −4. These non-extensive contributions reﬂect the implicit presence in the CI energy of the disjoint parts of renormalization terms. but the latter contains contributions from single. Bartlett.160 Diagrammatic expansions for perturbation theory A better approximation for ECID (or ECISD ) can be obtained. separately) and thus constitutes part of the RSPT approximation to CID. If we take the fourth-order RSPT formula ˆR ˆ0W ˆR ˆ0W ˆR ˆ0W ˆ |0 − E (2) S11 E (4) = 0|W ˆ 0. In fact. The loss of extensivity is due to the presence of non-extensive terms at each truncation level of the CI energy. and limit the implied sums over intermediate states in R ˆ0 = R I |ΦI ΦI | E 0 − EI (0) (0) .021 mEh −272. in principle. truncation at any excitation level sacriﬁces the important property of extensivity.

and possibly single. the non-extensive contributions in the CISD energy reﬂect the implicit presence of the disjoint part of the fourth-order renormalization term (−E (2) S11 )DJ . excitations. the non-extensivity is commonly ascribed to uncanceled unlinked-diagram contributions retained in truncated CI. These missing unlinked diagrams in the CI energy are a manifestation of the missing disconnected clusters in the CI wave function. For example. This approach is. 5.116 mEh −292.10 Unlinked diagrams and extensivity 161 in CI calculations). ∆EQ ≈ (1 − C0 (5.576 mEh −17.and quadruple-excitation contributions. the basis for the Davidson correction (Davidson 1974. the unlinked quadrupleexcitation diagrams (see Fig. In the above example this approach would give ECISD E (2) S11 estimate of ECISDQ −275. This contribution would be canceled in higher-level CI by non-EPV unlinked quadruple-excitation contributions. but it is retained in CISD and is ˆ2 primarily a manifestation of the lack of the disconnected 1 2 T2 clusters from the CISD wave function (see Section 1.692 mEh This is a very reasonable value. in fact. The non-extensivity eﬀects in truncated CI increase rapidly with the size of the system. though it is likely to overestimate the quadruple-excitation contribution somewhat (the true CISDQ correlation energy is probably about −289 mEh ). this arises from double. This correction is based on the assumption . respectively.8).51) where ∆ED and ∆EQ are the double. A simple strategy to improve the CID (or CISD) energy is to add an estimate of the quadruple-excitation contribution. Thus adding +E (2) S11 is equivalent to adding the unlinked diagrams + .5.13). because of the higher N -dependence of the non-extensive terms. Langhoﬀ and Davidson 1974) 2 )∆ED . As a ﬁrst try we can simply remove the −E (2) S11 renormalization-term contribution since this cancels the largest quadruples contribution in E (4) . in CI and C0 is the coeﬃcient of the SCF conﬁguration in the normalized CID expansion. Since the renormalization terms are expressible in terms of unlinked diagrams.

576) = −14. Buenker.52) (5.9724222 )(−275. since only (−E (2) S11 )DJ is truly canceled by the non-EPV unlinked quadruple-excitation diagrams. Applying the Davidson-correction formula to the H2 O 39-STO CISD calculation gives 2 )∆ED = (1 − 0. ∆ESD and C0 from CISD are generally used instead of the CID values. (5. Pople. 1983. The various CEPA (coupled electron pairs approximation) modˇ ıˇ els (CEPA-n. Seeger and Krishnan 1977. . Meyer 1974.852 mEh . Burton. or an estimate for it. (1 − C0 (5. Furthermore. n = 0–5) (Kelly 1964a. Actually. providing an approximation for CISDQ. Since the principal extensivity error in CISD is due to the implicit presence of the disjoint part of the fourth-order renormalization term and since those contributions can only be canceled by unlinked quadruple-excitation terms.56) but we cannot easily obtain this value.990 mEh . giving the last form in the equation. these corrections go a long way toward restoring extensivity to CISD and are therefore also known as extensivity corrections. in which the coeﬃcient of ΦSCF is equal to 1. Levy and Berthier 1976. Martin. this denominator is usually ignored. The Davidson correction is one of a class of so-called quadruples corrections (see also Davidson and Silver 1977. 2 C0 (5. (5.53) The denominator in the next-to-last form of (5.162 Diagrammatic expansions for perturbation theory (see Meunier.015 mEh .116 mEh . Fran¸ cois and Gijbels 1990) that constituted an attempt to account for the quadrupleexcitation CI contributions on the basis of CISD data. E (2) S11 overestimates the correction. C´ zek 1966. Ahlrichs. Meissner 1988. Bruna et al. ΨCID = ΦSCF + χD . A better correction would be (E (2) S11 )DJ = −10.54) which may be compared with E (2) S11 = −17. Bartlett and Shavitt 1977a) that ∆ED ≈ E (2) and S11 = Ψ(1) |Ψ(1) ≈ χD |χD = 2 1 − C0 2 ≈ 1 − C0 .55) 2 )/C 2 instead of 1 − C 2 then we would have obtained a If we had used (1 − C0 0 0 correction equal to −15. however. from the CISD calculation.53) (Siegbahn 1978) signals the conversion from the full normalization of the CI wave function to the intermediate normalization appropriate for perturbation theory.

015 mEh −285.5 mEh compared with −288.10 Unlinked diagrams and extensivity 163 Table 5.5. Rosenberg. since the remaining (uncancelled) true quadruples terms have been ignored.001555 0. Shavitt and Purvis 1979.290566 −0. Purvis et al.287858 −0. Kutzelnigg 1977. using E (2) S11 instead of (E (2) S11 )DJ compensates to some extent for this omission. Correlation energy of H2 O in the 39-STO basis (energies are in millihartrees) (Rosenberg and Shavitt 1975.973176 −0. Ermler and Shavitt 1976. . Staemmler et al.014504 −0. Bartlett. giving about −289. Including these terms would have added about −3.288525 −0. 1987) From Da calculation CI correlation energy Contribution of single excitations C0 (coeﬃcient of ΦSCF in normalized CI) Davidson correction CI correlation energy + correction MBPT(2) MBPT(3) MBPT double excitations to inﬁnite order MBPT(4)SDQ MBPT(4) CCD CCSD CCSDT-1 a b From SDb calculation −0.296035 −0. Bartlett. 1975.014990 −0.9 mEh .285021 −0.972422 −0. Cole. Lischka.28826 −0.281780 −0.295408 −0.286210 Doubles Singles and doubles.591 mEh This analysis does not include the full quadruple-excitations contribution.275576 −0.576 mEh (E (2) S11 )DJ = −10.288172 −0. Bartlett 1981.2. Thus.172 mEh for the fourth-order MBPT SDQ energy. if we could have used (E (2) S11 )DJ we would have obtained ECISD = −275.274021 0. Koch and Kutzelnigg 1981) are various ways to estimate (E (2) S11 )DJ on the basis of pair energies (we shall not describe these methods here).

Bartlett. MBPT and coupled-cluster (CC) results are included. 1987) are summarized in Table 5. The coupled-cluster methods will be discussed in Chapters 9 and 10. . Purvis et al. Bartlett. the CI. Rosenberg. Shavitt and Purvis 1979. Ermler and Shavitt 1976.2.164 Diagrammatic expansions for perturbation theory Various correlation-energy results for the H2 O 39-STO CISD calculation (Rosenberg and Shavitt 1975. Cole. Bartlett 1981.

at least one of which is closed.25). embodied in (5. with a modiﬁcation in the principal part due 165 . using skeletons rather than full diagrams to reduce clutter since the validity of the factorization is independent of the arrow directions: × × + × = × + × + + × × + × . it can be applied to a diagram with an insertion. We ﬁrst give an example. we will derive a generalization of the factorization described in subsection 5. Frantz and Mills 1960. because the total number of denominator factors would not be the same in the product and in the unlinked diagrams resulting from the expansion. 1977). However. states that a product of two connected open diagrams is equal to a sum of disconnected diagrams.4.1 The factorization theorem As a ﬁrst step in the formal proof of the linked-diagram theorem. It cannot be applied to a simple product of an open and a closed diagram or of two closed diagrams. each made up of the two diagrams of the original product with all possible time orderings of the vertices of one of the disconnected parts relative to those of the other. Brandow 1967. This generalization. referred to as the factorization theorem (Hugenholtz 1957.6 Proof of the linked-diagram theorem 6. The theorem can be extended straightforwardly to products of more than two open diagrams.7. which represents a product of two diagrams.

Fig. Obviously. any complex structure may be added above the insertion and we would still obtain a sum of the same number of unlinked diagrams.1. Therefore.1. This form of the theorem is used to prove that the renormalization terms of RSPT cancel against the unlinked diagrams. restoring these parts does not aﬀect the validity of the expansion. applying the original factorization theorem and then restoring the deleted parts in the product and in all the unlinked diagrams in the expansion. In this case the sum over all time orderings is restricted to those orderings in which the top vertex of the inserted diagram remains at the level of the insertion. beginning with the innermost . An example is shown in Fig. A factorization-theorem example of the expansion of a diagram with an insertion as a sum of unlinked diagrams (resolvent lines are shown in the product form to show that a resolvent remains on each side of the original insertion). In the case of nested insertions the factorization theorem can be applied successively to the individual insertions. One obtains the modiﬁed theorem directly from the original theorem by cutting oﬀ the top of the principal part just above the insertion (including the extra resolvent) and the top vertex of the insertion before the expansion of the product. 6. Any structure that exists above the insertion remains unchanged in all diagrams in the sum. to the presence of an extra resolvent line just above the insertion.166 Proof of the linked-diagram theorem ˆ0 R ˆ0 R ˆ0 = R ˆ0 R ˆ0 R ˆ0 R × ˆ0 R = + + + + + . 6.

For example. D2 . following Frantz and Mills (1960) (see also Manne (1977) and Harris. . This is described . then diﬀer only in their denominators. . respectively. Dn and D1 . beginning with the lowest insertion. . owing to the sign change associated with each insertion. When deriving the linked-diagram theorem we do not need to continue the expansion to the outer insertion.. Thus we have to show that the sum of the appropriate denominator factors (1/εab. ) for the disconnected diagrams is equal to the product of the denominator factors of the product. an extension of the expansion to multiple insertions in the same principal part can also be done sequentially. To prove the factorization theorem we consider the product of two diagrams. . For example. . = + . both open at the top. and let the corresponding denominators be written (from the bottom up) as D1 .1 The factorization theorem 167 insertion and proceeding outwards. who deal specifically with the case of a diagram with an insertion) and note that the algebraic expressions for the diagram product and for all the diﬀerent time orderings in the sum of the comparably labeled disconnected diagrams have the same numerator. Similarly. produces a zero factor. ij... . we have = + . respectively.6. and thus the number of denominator factors. Again. . because the cancellation of the inner insertion with the corresponding unlinked terms within the outer insertion (shown on the right-hand side). it is unnecessary to proceed further because of the zero factor resulting from the ﬁrst expansion. in the two diagrams of the product be n and m. . Monkhorst and Freeman (1992). Let the number of resolvent lines. D2 .. Dm .

we shall consider the case with n = 3. i.1) As an example.168 Proof of the linked-diagram theorem schematically by Dn Dn−1 Dn−2 . . . summed over all possible relative orderings of their vertices without change of ordering within each disconnected part. .e. m = 2. The sum S32 consists of 10 terms. We denote the sum of the denominator factors of the corresponding unlinked diagrams. involving all possible time orderings of the two sets of vertices relative to each other: ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R → → → → → → → → → → D2 D1 D3 D2 D1 1 D1 D2 D3 (D3 + D1 )(D3 + D2 ) D3 D2 D1 D2 D1 1 D1 D2 (D2 + D1 )(D3 + D1 )(D3 + D2 ) . D3 D2 D1 vertex m vertex m − 1 vertex 3 vertex 2 vertex 1 vertex 3 vertex 2 vertex 1 The same scheme also applies to the case of a diagram with an insertion if we ignore all vertices and resolvents in the principal part above the insertion and add a ﬁxed vertex above Dm in the other part. by Snm . We need to show that this sum is equal to the product of the denominator factors of the original diagram product. Snm = 1 . . D3 D2 D1 vertex n vertex n − 1 Dm Dm−1 Dm−2 . D1 D2 · · · D n D 1 D 2 · · · D m (6.

1 The factorization theorem 169 ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R → → → → → → → → → → D3 D2 D1 D2 D1 1 D1 (D1 + D1 )(D2 + D1 )(D3 + D1 )(D3 + D2 ) D3 D2 D1 D2 1 D1 (D1 + D1 )(D2 + D1 )(D3 + D1 )(D3 + D2 ) D1 ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R → → → → → D3 D2 D1 D2 D1 1 D1 D2 (D2 + D1 )(D2 + D2 )(D3 + D2 ) ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R → → → → → D3 D2 D2 D1 1 D1 (D1 + D1 )(D2 + D1 )(D2 + D2 )(D3 + D2 ) D1 ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R → → → → → D3 D2 D1 D2 1 D1 (D1 + D1 )(D2 + D1 )(D2 + D2 )(D3 + D2 ) D1 .6.

in which there is no ﬁrst factor. D1 D2 D3 D 1 D 2 It can be veriﬁed laboriously that they actually do so by bringing all 10 terms to a common denominator. in each term of the sum Snm the contribution to the denominator from the topmost level of the diagram will be the factor Dn + Dm .1)) 1 . and succeeding factors will have the form Dk + Dl . Returning to the general case. and n = 0. where we have factors such as Dk (if the ﬁrst part extends below the bottom of the second part) or Dl (if the second part extends below the bottom of the ﬁrst part).2) . Both these cases satisfy the theorem represented by (6.170 Proof of the linked-diagram theorem ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R ˆ0 R → → → → → → → → → → → → → → → D3 D2 D1 D2 D1 1 D1 (D1 + D1 )(D1 + D2 )(D2 + D2 )(D3 + D2 ) D3 D2 D1 D2 1 D1 (D1 + D1 )(D1 + D2 )(D2 + D2 )(D3 + D2 ) D1 D3 D2 D1 D2 1 D1 D2 (D1 + D2 )(D2 + D2 )(D3 + D2 ) D1 These 10 terms should add up to (see (6.1) since Sn0 = 1 D1 D2 · · · D n (6. until we reach the region in which the two parts do not overlap. in which there is no second factor. We extend the deﬁnition of Snm to the cases m = 0.

m−1 .m + Sn. The contribution from the topmost level in both classes is the factor 1 .m + Sn.1 The factorization theorem 171 and S0m = 1 . Dn + D m D1 D2 · · · D n D 1 D 2 · · · D m (6. . . The two classes can be described schematically by ˆ0 → R ˆ0 → R Dn Dn−1 . we can substitute the corresponding products and obtain Sn−1. Dm . D1 D 1 (6.m−1 ) . . those in which the top vertex occurs in the ﬁrst part and those in which it occurs in the second part.m−1 = 1 1 1 + Dm Dn D1 · · · Dn−1 D1 · · · Dm−1 Dn + Dm = . . Therefore Snm = 1 (Sn−1. Equations (6. Now consider the case n ≥ 1. m ≥ 1. (6. D1 D2 · · · D n D 1 D 2 · · · D m (6. .5) If the theorem holds for Sn−1. Dn + D m In the diagram class on the left this factor is multiplied by Sn−1. while in the class on the right it is multiplied by Sn.6) so that Snm = 1 1 (Sn−1.2).m−1 ) = .1) holds for all n.m and Sn. It is also very easy to verify the S11 case: S11 = 1 1 1 + = D1 (D1 + D1 ) D1 (D1 + D1 ) D1 + D 1 1 1 + D1 D 1 = 1 . as well as for S20 and S02 .m (when we allow all possible relative orderings of the remaining vertices). .m−1 .m + Sn.4) so that the theorem holds for all n + m ≤ 2. . Dn + D m (6. showing that if (6. D1 D2 · · · D m (6.3) show that the equation holds for n + m = 1. . and ˆ0 → R ˆ0 → R Dn .6. and divide the corresponding set of unlinked diagrams into two classes.3) Our proof will be inductive. . Dm Dm−1 .7) . m such that n + m = k − 1 then this equation holds also for n + m = k .

7. who showed that all diagrams that contain unlinked parts or insertions. to any depth. The direct diagrammatic expansion of RSPT using the bracketing procedure described in subsection 2. it must hold for all n. cancel mutually leaving an expansion in terms of linked diagrams only. the diagrams in such a . 2. without change of ordering within each part (such as the set of unlinked diagrams in the expansion in Fig. Let us group together all those of a given order which have the same open part and the same closed part and for which the top vertex of the closed part is at the same level relative to the open part.1) includes diagrams that contain insertions.e.4. who proved it explicitly for speciﬁc low orders of the perturbation expansion and also asserted that it must be true at all orders. Each unlinked diagram may itself contain insertions and each insertion may contain unlinked diagrams and further insertions. diagrams that contain separate closed parts. Since the theorem holds for n + m ≤ 2. It was ﬁrst stated by Brueckner (1955). 6. or any combinations of these.6 (and see Fig. m. in terms of both time-dependent and time-independent diagrammatic techniques. and because of the change of sign associated in the perturbation expansion with each insertion.2 The linked-diagram theorem The linked-diagram theorem states that the wave function and the correlation energy are each given by a sum of linked diagrams only. i. as discussed in subsection 5. It was proved formally for all orders by Goldstone (1957). provided that all EPV terms are included in the summations for each diagram. for the factorization theorem to hold. as well as unlinked diagrams. The EPV terms in the disconnected or unlinked diagrams are counterbalanced in the MBPT expansions by the inclusion of such terms in the summations for the linked diagrams as well. all summations over internal lines must be unrestricted.172 Proof of the linked-diagram theorem satisfying the theorem.1). including any EPV terms. one separate closed part and no insertions. We will consider ﬁrst all unlinked wave function diagrams that contain a linked (but possibly disconnected) open part. This derivation is closely tied to the factorization theorem in the form applicable to diagrams with insertions. because of the incorrect size dependence of the unlinked diagrams and the renormalization terms. with all possible relative orderings of the other vertices of the closed part relative to those of the open part.5. 6. According to the factorization theorem. An elegant time-independent derivation was presented by Manne (1977). It should be emphasized that.

including cases with insertions that contain unlinked diagrams or other insertions.6. If the lowest item is an insertion that contains no further insertions or unlinked diagrams then. it cancels against all the corresponding unlinked diagrams with a lowest unlinked part having a top vertex at the same level. linked) closed part.2 The linked-diagram theorem 173 group cancel with the diagram in which the separate closed part is placed as an insertion into the open part at the original level of the top vertex of the closed part. This remainder diagram may be linked or unlinked. consider the following twelfth-order wave-function diagram containing an unlinked principal part and two insertions. may have any number of separate closed parts and may itself contain insertions. following Manne we classify all the diagrams of a given order according to the nature of the lowest-level separate item they contain. this lowest-item insertion contains other insertions or unlinked diagrams then we consider the remainder diagram as a ﬁxed multiplicative factor and apply the same procedure to the diagram within the insertion. to any depth. For the purpose of this extension. If the lowest item is a separate closed part.e. one of which . here a separate item refers to an insertion at that level or to a separate closed part with top vertex at that level. leaving an expansion in terms of linked diagrams only. Proceeding in this manner until we reach the innermost insertion we ﬁnd that all unlinked diagrams and all diagrams that contain insertions cancel. we collect together all diagrams of a given order that have that same lowest closed part with top vertex at the same level relative to the remainder diagram. The same procedure can be applied to groups of diagrams with the top vertex of the closed part at any other level relative to the open part. If. The same procedure can be applied to unlinked energy diagrams consisting of two closed parts where the part with the higher top vertex plays the same role as the open part in the previous paragraph. In this case all these diagrams cancel with a corresponding diagram with an insertion at the speciﬁed level. however. and the same remainder diagram. We will refer to the original diagram minus the lowest separate item (but retaining the extra resolvent) as the remainder diagram. thus proving the cancellation of all two-part unlinked closed diagrams with the corresponding insertions. as well as to unlinked diagrams that contain insertions. by the argument of the previous paragraph. As an example. consequently all unlinked wave-function diagrams with one open part cancel with all linked wave-function diagrams with one insertion containing a single (i. This approach can then be extended to diagrams with any combination of separate closed parts and insertions.

comes in with a sign change). In this example there is only one other diagram that can be grouped with it. provided that all summations in all diagrams are unrestricted. because it has the lowest top vertex within the insertion. these and all other diagrams that contain unlinked parts or insertions can be left out of the expansion.174 Proof of the linked-diagram theorem contains an unlinked diagram: The lowest separate item in this diagram is the lower insertion. the sum of these two diagrams cancels against the corresponding diagram with an inner insertion (which. . like all insertions. Since this insertion contains an unlinked diagram. The lowest separate item within this insertion is the separate closed part at its right. we consider the rest of the diagram (which is itself unlinked and contains an insertion) as a multiplicative factor and focus our attention on the contents of the lower insertion. diﬀering from it only in the relative levels of the lower vertices within the insertion: According to the factorization theorem. regardless of the value of the original remainder diagram: All three of these diagrams would appear in the straightforward RSPT expansion of the twelfth-order wave function following the bracketing procedure and.

12) It is easy to verify that all the closed diagrams that can be formed by adding a new top vertex to the upwards-open linked n-vertex diagrams are linked (because all disconnected parts of the open diagram must be closed by the single added vertex) and constitute the complete set of all closed linked (n + 1)-vertex diagrams. (6.9). (6.12) is consistent with (6.9) for |Ψ and ∆E are entirely equivalent to the Schr¨ odinger equation.10) in (6.75) in a form appropriate for RSPT.8) ˆ0 ≡ R ˆ 0 (E0 ).8). In this derivation the linked-diagram expansions for the wave function and energy are substituted into the recursive form (2. where R The implicit equations (6.8).2 The linked-diagram theorem 175 We shall consider brieﬂy an alternative derivation of the linked-diagram theorem along the lines of Goldstone’s time-independent derivation (Harris.6.9) (6.11) ∆E = n=1 ˆ (R ˆ0W ˆ )n 0 0W L where the subscript L indicates that the summations are limited to linked diagrams only (note that the n = 0 term is missing in the summation for ˆ |0 = 0). ˆ [(R ˆ0W 0|W (6. obtaining ∞ ∆E = n=1 ˆ )n |0 ]L . (6. To prove this assertion we ﬁrst rewrite (2. .10).9). . We need to prove that these equations are satisﬁed by the linked-diagram expansions ∞ |Ψ = n=0 ∞ ˆ0W ˆ )n |0 (R L . (6. We are going to prove this assertion ∆E in (6. Therefore (6. ∆E = 0|W (6.11) because 0|W by substituting (6.11) into the recursive equations (6.9) and showing that the latter are then satisﬁed. W ˆ =V ˆ − E (1) and ∆E = E − Eref = E − E0 − E (1) .75) of the Schr¨ odinger equation. and the factorization theorem is used to show that this expansion satisﬁes the equation. We ﬁrst substitute (6. Monkhorst and Freeman 1992).10) (6. ˆ − ∆E )|Ψ ˆ 0 (W |Ψ = |0 + R with ˆ |Ψ .

proving that this expansion satisﬁes (6.8) and the Schr¨ odinger equation. We may therefore rewrite (6. (6. In fact.14) where the subscript U indicates restriction to unlinked terms. . The linked-diagram theorem accounts for the transition from RSPT to MBPT.13) consists of all the upwards-open (n + 1)-vertex diagrams that can be formed by adding one vertex (and the corresponding resolvent) to all upwards-open linked nvertex diagrams. The factorization theorem can then be used to show the cancellation of the last two sums in this equation. providing both an eﬃcient computational tool and a framework for the easy introduction of inﬁnite-order summations of diagrams via the coupled-cluster methods described in Chapters 9 and 10. because each term in the third sum can be described by an open diagram with an insertion above its top vertex.8). maintaining extensivity.10). this diagram cancels the contributions to the second sum from the sum of corresponding unlinked two-part open diagrams in which the top vertex of the closed part is the top vertex of the entire diagram. The remaining terms of the right-hand side are equivalent to the linked-diagram expansion (6. In this form. As a result of the linked-diagram requirement and the inclusion of EPV terms.176 Proof of the linked-diagram theorem Next we substitute (6.10) in (6. MBPT assumes the simple. there are no renormalization terms and no restricted summations in the MBPT expansions.13) in the form ∞ ∞ |Ψ = |0 + n=0 ˆ (R ˆ0W ˆ )n |0 ˆ0W R L + n=0 ∞ ˆ [(R ˆ0W ˆ )n |0 ]L ˆ0W R ˆ0W ˆ 0 (R ˆ )n |0 ∆E R n=0 U − L . and is a major simpliﬁcation of ordinary RSPT. recursive form of BWPT while. RSPT has been reduced to its essentials. Each resulting diagram either is linked or is unlinked with a single separate closed part (if the added vertex closed a disconnected part of the n-vertex open diagram) and has the top vertex of the closed part as the top vertex of the entire diagram. unlike BWPT.13) Each term of the ﬁrst sum over n in the second line of (6. (6. resulting in ∞ |Ψ = |0 + n=0 ∞ ˆ 0 (W ˆ )n |0 ˆ − ∆E ) (R ˆ0W R ∞ L = |0 + n=0 ˆ0W ˆ (R ˆ0W ˆ )n |0 R − L n=0 ˆ0W ˆ 0 (R ˆ )n |0 ∆E0 R L .

see Fig. First we examine nh . which may often be too large overall number of terms is still of order nh p for practical computation.1 Techniques of diagram summation The straightforward summation of individual MBPT diagrams is usually an ineﬃcient process.5) can also serve to reduce the number of diagrams that need to be evaluated. Using symmetries with respect to permutation of the particle lines is nh p indices can reduce the number of distinct terms by some factor but the h n p .8. 5. The identiﬁcation of conjugate diagrams (Section 5. The principal approach involves the evaluation and reuse of various intermediate quantities (partial sums). Another technique is the factorization of sums of certain classes of diagrams. diagram 14: (a) (c) (e) (i) (j ) (b) (d) = (f ) 1 16 ij ab ab cd cd ef ef ij abcdef ij cd ef εab ij εij εij . We will focus on the canonical HF case. (7. We will consider some examples from the fourth-order energy in discussing eﬃcient summation techniques. using a procedure analogous to the use of the factorization theorem to factor sums of unlinked diagrams. the the double-excitation diagrams. Since typically np largest number of terms is represented by the particle ladder. the number of terms in the sum represented by a diagram with h internal hole lines and p internal h n p . If the one-electron basis contains nh hole states (orbitals occupied in Φ0 ) and np particle states (virtual orbitals). A variety of techniques are used to reduce the computational eﬀort.7 Computational aspects of MBPT 7. since diagrams containing one-particle vertices have fewer lines than the corresponding diagrams with only two-particle vertices.1) 177 .

cd = −S dc = −S cd = The last form of this expression uses the relationships Sij ij ji dc and S cc = S cd = 0. the (1) . consider diagram 7 of Fig. j own εcd ij lines and the resolvent line are actually common to the two parts. loosely referred to as n6 . d. because the c.6. in either ket (bottom) or bra (top) form.2). j after insertion of the denominator corresponding to the middle resolvent line produces the complete sum for the diagram. The evaluation of Sij h p 2 2 2 contributions). As a wave-function diagram. Note also that each term in the partial sums contains a ﬁrst-order wave-function coeﬃcient such as ef ij /εef ij . f are dummy summation indices). d. e. 1 16 cd |2 |Sij cdij εcd ij = 1 4 cd |2 |Sij c>d i>j εcd ij .8: (a) (i) (c) (k ) (j ) (l) (b) = (d) abcdijkl ij ab al cj kb id cd kl . (there are nh np sums to be computed. coeﬃcient of |Φef ij in |Ψ For another example.178 Computational aspects of MBPT 2 n 6 terms (loosely The straightforward summation of (7.2) is a second-order wave-function diagram. The complete The ﬁnal sum is much faster.3) . Summation of the product of these two partial sums over c. However. i. Note that each half-diagram in (7. we have to remove a factor 1 4 (for two common pairs of equivalent lines) and one denominator. we can split this diagram into two parts. However. bc cd εab ij εil εkl (7. 5. each half would include a factor 1 8 (for three pairs of equivalent lines) and its denominator (see subsection 5. being proportional to nh p 2 4 procedure thus amounts to an nh np process.2) cd ef ef ij εef ij cd = Sij ef (noting that a.1) would involve nh p referred to as an n8 process). each requiring np 2 n 2 . leaving the above result. b. i. which arise from the use of antisymmetric twoSji ij ii cd is proportional in eﬀort to n 2 n 4 electron integrals. separated by the middle resolvent line: (a) c ˆ0 R c (e) i i j j (b) ∼ d d ∼ (f ) ab ij ab cd ∗ ab cd = (Sij ) εab ij (7.

4) of the previous two examples. These examples demonstrate how the same partial sum can be used in the evaluation of more than one diagram.9. εab ij (7. and take diagram 25 as an example: (a) (b) (i)(k ) (c) (e) (d) (j ) = 1 2 abcdeijk ij ab ak cd db ke ce ij . consider diagram 8 of Fig. 5.7.6) Because of the triple-excitation factor in the denominator.8: (a) (c) (i) (j ) (k ) (b) (d) = − (e) 1 4 abcdeijk ij ab ab cd kd ie ce kj . Fig. order involves at most an nh p Next we consider fourth-order triple-excitation diagrams.4) 3 n 3 process.5) Using the partial sums (7. we are forced to use a six-index partial sum. cd ce εab ij εij εjk (7. εcd kl (7.7) 3 n 4 (∼ n7 ) computational process. 5. εbc il bcil For one more example of a double-excitation diagram.1 Techniques of diagram summation 179 In this case the partial sum will be bc = Til dk kb id cd kl . The complete sum for requiring an nh p . (7. Using the partial summation technique. the evaluation of all the double-excitation diagrams in fourth 2 n 4 (∼ n6 ) computational process. bcd = Uijk e db ke ce ij . the complete sum for this diagram is given by − 1 4 cd )∗ T dc (Sij ij cdij εcd ij . and the complete sum is given by an nh p cb )∗ T bc (Tli il .2). bcd ce εab ij εijk εij (7.

9). Applying the same approach to the fourth-order quadruple-excitation diagrams would have resulted in an n8 process because of the quadrupleexcitation factor in the denominator. the quadruple-excitation diagrams can actually be evaluated in an at most n6 process. a factorization similar to that derived in the factorization theorem eliminates the quadruple-excitation denominator factor and reduces the computational eﬀort for these diagrams to an n6 process. 5.180 Computational aspects of MBPT this diagram is then 1 2 cbd )∗ U bcd (Ujik ijk bcdijk εbcd ijk . This is accomplished by a technique similar to that used in the proof of the factorization theorem.9.2 Factorization of fourth-order quadruple-excitation diagrams As stated in Section 5. evaluation of the quadruple-excitation diagrams in the fourth-order energy (Fig. Adding diagrams 34 and 40 in Fig. Both types of diagram involve summations over eight indices and.8) ij ab kl cd ab ik cd jl abcdijkl 1 cd ab εab ij εjl εik As a result of this factorization the quadruple-excitation denominator is . 5. leaving the triple-excitation contribution as the most computationally demanding part of a fourth-order energy calculation.10) requires less computational eﬀort than needed for the triple-excitation contributions (Fig. 7. adding together diagrams that diﬀer only in the time ordering of their lowest two vertices.2.10 we get (j ) (i) (a)(b) (j ) (k ) =− =− 1 4 1 4 (c)(d) (l) 1 abcd εab ij εijkl + (i) (a)(b) (k ) (c)(d) (l) ij ab kl cd ab ik cd jl abcdijkl 1 1 + ab cd εjl εik . (7. 5. as is shown below in Section 7. However. while partial summations reduce the computational eﬀort for the triple-excitation diagrams to order n7 .

9) 3 n 2 (∼ n5 ) process for each partial sum. The computation of the partial sums in this case is an nh p 2 process. ijkl The above factorizations take care of six of the seven HF-case linked . respectively.11) Tijkl = cd The ﬁnal sum is an n4 process. εcd εcd ij kl (7. process and the ﬁnal sum is an np Next we consider the sum of diagrams 36 and 37: (a) (i) (c) 1 16 1 16 (k ) (l) (d) (b) (j ) + (a) (k ) (i) (c) (d) (l) 1 abcd εab ij εijkl (j ) (b) = = ij ab kl cd ab kl cd ij abcdijkl 1 1 + ab cd εij εkl . (7.7. by reversing the directions of 2n 3 all the arrows. εab ij kl cd cd ij . which can be obtained from 34 and 40. 1 16 Tijkl Sijkl . The This procedure is just an nh p 2 ﬁnal sum is an nh process. εcd jl Tjk = abi ij ab ab ik . Tjk Sjk .2 Factorization of fourth-order quadruple-excitation diagrams 181 eliminated. jk The same procedure can be used for the sum of diagrams 35 and 39.10) ij ab ab kl kl cd cd ij abcdijkl 1 cd ab εab ij εij εkl In this case the partial sums involve n6 processes: Sijkl = ab ij ab ab kl . allowing the use of partial summation in the forms Sjk = cdl kl cd cd jl . εab εab ij ik (7.

Bartlett and Shavitt 1977b). bcjk The summation methods described here are not the only feasible processes.g.10). 5.13) which are n6 processes. εac εij ik ab (7.3 Spin summations So far the formalism has been speciﬁed in terms of spinorbitals. (7. and the choice of procedure depends also on the maximum utilization of sums computed in lower orders (see e. can be handled by writing it as a sum of two equivalent diagrams and factoring the sum as before: 1 (a) 2 1 2 (j ) (d) (l) (c) (b) 1(a) + 2 (i) (k ) (c) (d) 1 abcd εab ε ij ijkl 1 ac bd εab ij εik εjl = (i) (k ) (j ) (l) (b) = ij ab kl cd db lj ac ik abcdijkl 1 1 + bd εac ε ik jl . integration over the spin variables is easily carried out and results in signiﬁcant economies in the calculations. εbd jl = ai ij ab ac ik . The remaining diagram. and no bar over . and no attempt has been made to consider the eﬀects of spin. The simplest way in which spin aﬀects the PT summations is that some integrals vanish because of spin orthogonality.182 Computational aspects of MBPT quadruple-excitation fourth-order diagrams (Fig. since the nonrelativistic Hamiltonian does not contain spin coordinates. and the ﬁnal sum is an n4 process: 1 2 bc bc Tjk Sjk .12) 1 = 2 ij ab kl cd bd jl ac ik abcdijkl The partial sums can be chosen as bc = Sjk dl bc Tjk kl cd bd jl . Thus if we indicate the spin factor of a spinorbital by putting a bar over β spinorbitals. number 38. 7. However.

s.3 Spin summations 183 α’s. pq rs = − pq |v ˆ|sr . ˆ|rs . pq rs = pq |v ˆ|rs − pq |v ˆ|sr . (7. such as one consisting of UHF spinorbitals. however. pq rs = pq rs = 0.15) (7. results in some reduction in computational eﬀort. Spin integration in the case of a general spinorbital basis. much more signiﬁcant economies result if we have a spinadapted basis.14) abij abij abij . once with spin α (say p) and once with spin β (p). pq rs = − pq |v pq rs = pq |v ˆ|rs . Thus. Taking the second-order energy in the canonical RHF case as an example.14). where the integrals on the r. and using (7. in which each spatial orbital appears twice.7. we have pq rs = pq |v ˆ|rs − pq |v ˆ|sr . pq rs = pq rs = pq rs = pq rs = 0 . we have: (a) = (i) (j ) 1 4 (b) 1 ij ab ab ij + ij ab ab ij + ij ab ij ij εab ij + ij ab ab ij + ij ab ab ij + ij ab ab ij = = 1 4 1 2 1 2 ij ab ab ij + 2 ij |v ˆ|ab ab|v ˆ|ij + 2 ij |v ˆ|ba ba|v ˆ|ij εab ij 1 ij |v ˆ|ab ab|v ˆ|ij − ij |v ˆ|ab ba|v ˆ|ij − ij |v ˆ|ba ab|v ˆ|ij εab ij + ij |v ˆ|ba ba|v ˆ|ij + ij |v ˆ|ab ab|v ˆ|ij + ij |v ˆ|ba ba|v ˆ|ij (7. and pq rs = pq rs = pq rs = pq rs = 0. are over the spatial factors only. pq rs = pq |v ˆ|sr . 10 of the resulting integrals vanish completely and four are reduced to a single spatial integral. out of the 16 possible combinations of spin assignments to the four orbitals in an antisymmetric two-electron integral.h.15). as in RHF.

but even when summing over spinorbitals in a UHF-type treatment the number of nonzero contributions is reduced very substantially by spin orthogonality and by the requirement of spin continuity along each loop. More general diagram interpretation rules can be derived to obtain the spin-independent form directly from the diagram. This factor arises because in each closed loop of a Goldstone diagram all spin factors must be equal. for which we simply get an extra 2l factor for the summation over spatial orbitals. where the summations are over the distinct spatial orbitals only and zero terms have been left out. Summing over spatial orbitals reduces the range of each summation by a factor of two. Thus (a) (i) (j ) (b) = abij 2 ij |v ˆ|ab ab|v ˆ|ij − ij |v ˆ|ab ba|v ˆ|ij ab εij ij |v ˆ|ab [2 ab|v ˆ|ij − ba|v ˆ|ij ] . εab ij (7. at least for the Goldstone diagram case (see e. Paldus and ˇ ıˇ C´ zek 1975). we ﬁnd that the ﬁrst two terms in the brackets are equal (after summation). Since a. Thus a viable procedure for spin-adapted bases is to expand each ASG diagram into ordinary Goldstone diagrams and to sum them over orbital labels only. and they may be either all α or all β .184 Computational aspects of MBPT 1 = 2 abij 1 2 ij |v ˆ|ab ab|v ˆ|ij + 2 ij |v ˆ|ba ba|v ˆ|ij εab ij − ij |v ˆ|ab ba|v ˆ|ij − ij |v ˆ|ba ab|v ˆ|ij .16) = abij Similar treatments hold for other terms. and so are the third and fourth. . with a factor of 2l for each diagram. b are dummy summation indices and can be interchanged.g.

8 Open-shell and quasidegenerate perturbation theory 8. These problems commonly arise in the case of open-shell states because diﬀerent distributions of the open-shell electrons among the open-shell orbitals. as is the case for many excited states or in situations involving bond breaking. Many open-shell high-spin states can be treated eﬀectively with singlereference-determinant methods using either unrestricted or restricted openshell HF reference determinants because the spin restrictions exclude alternative assignment of the electrons to the open-shell orbitals. Several common series-extrapolation techniques can be used to speed up the convergence of a perturbation expansion or to obtain an approximate limit of a divergent series. Even when exact zero-order degeneracies are not present but two or more closelying zero-order states contribute strongly to the wave function. low-spin states. 185 . Bartlett and Br¨ andas 1972. however. Bartlett and Shavitt 1977b. are possible. Br¨ andas and Goscinski 1970. require alternative approaches.1 Formal quasidegenerate perturbation theory (QDPT) As is well known. unless spin or symmetry restrictions eliminate all but one of the degenerate determinants from the expansion. ordinary Rayleigh–Schr¨ odinger perturbation theory breaks down when applied to a state that is degenerate in zero order. all with the same or very similar total zero-order energies. The results of such an extrapolation usually improve as more of the early terms of the series become available. the RSPT expansions tend either to diverge or to converge very slowly.g. Reid 1967. such as open-shell singlets. Approaches based on Pad´ e approximants (closely related to continued fractions) have been applied in some studies (e. Goscinski 1967. The breakdown is due to singularities arising from the vanishing of denominators involving diﬀerences in energy between the reference determinant and determinants that are degenerate with it.

The equation governing degenerate perturbation theory is the Bloch equation (Bloch 1958. Ω=P and it follows that ˆΩ = P ˆ. A more general solution to these problems is found in a multireference version of perturbation theory. in which no single E0 is identiﬁed. which.3) and Ω is the wave operator. Ψα = ΩΦα = ΩP (8.8) ˆP ˆ .198): ˆ=V ˆ ΩP ˆ − ΩP ˆV ˆ ΩP ˆ. is ˆ=ˆ ˆ= Q 1−P I |ΦI ΦI | = I ˆI . called open-shell or degenerate perturbation theory in the case of exactly degenerate zero-order states and quasidegenerate perturbation theory (QDPT) in the more general case. Q ˆ I = |ΦI ΦI | . the space spanned by all the functions ΦI that are not in the model space. ˆ 0 )ΩP ˆ=V ˆ ΩP ˆ − ΩP ˆV ˆ ΩP ˆ.7) (8. so that Since the ΩQ ˆ.6) . Ω = U ˆ block and a Q ˆ block. produces the space spanned by the perturbed wave functions ˆ Φα .5. it is to be expected that each eigenfuncˆ will contain sizable contributions from more tion of the full Hamiltonian H (8.5) ˆ part of Ω is irrelevant it is convenient to set it to zero.186 Open-shell and quasidegenerate perturbation theory Swain 1977). Bloch and Horowitz 1958).4) The projector onto the orthogonal or complementary space. when operating on the model space. ˆ 0 ]P [Ω. Q (8. Kvasniˇ cka 1974). Ω = ΩP With intermediate normalization Ω is split into two components. P ˆα = |Φα Φα | . where U ˆ is the transformation operator As noted in Section 2. that block-diagonalizes the Hilbert space into a P ˆ The model functions Φα are eigenfunctions of H0 but. the space spanned In these equations P by the reference functions (model functions ) Φα . H (8. ˆ= P α |Φα Φα | = α ˆα . (E0 − H (8.1) It is easily generalized to the quasidegenerate case (Lindgren 1974. P (8. P (8.2) ˆ is the projector onto the model space. ˆ+Q ˆΩ . in the form seen in (2. because of their degeneracy or quasidegeneracy.

11) (8. This equation was derived in Section 2. The transformation matrix C is determined. tonian H The relationships between the diﬀerent functions and the wave operator Ω can be summarized in the following set of equations: ˆ ΩΦα .9) where Φα = β Φβ Cβα (8. as seen from (8. we obtain ˆ 0 and P tion of H ˆ 0 Φα = ΩV ˆ ΩΦα . ˆ ΩΦα = P ˆ Ψα = Φα . ˆ .8) and the commutaˆ . by a ﬁnal diagonalization of the eﬀective Hamilˆ eﬀ . together with the energies. will now be given. a ﬁnal transformation among the perturbed functions Ψα is ˆ: required to produce the eigenfunctions Ψα of H Ψα = β Ψβ Cβα = β ΩΦβ Cβα = ΩΦα .17) .8. The intermediately normalized wave operator cannot directly produce this mixing since.14) Since the generalized Bloch equation contains the exactly degenerate case as a special case. as discussed further below.9) for the exact eigenfunctions Ψα of H can be written in the form ˆ ΩΦα = Eα ΩΦα . H (8. Eα ΩΦα − ΩH (8.5 using a similarity transformation approach. we can rewrite (8. As discussed in contributions from the Q Section 2.10) are the properly mixed zero-order functions for the problem. Ψα = Φα + Q ˆ Ψα = Φα .12) (8.16) Operating on (8. also called bonnes fonctions (Bloch 1958). the Schr¨ odinger equation Using (8. with H ˆ 0 Φα = Eα Partitioning the Hamiltonian as H Φα . P ΩΨα = ΩΦα = Ψα . P ˆ ΩΦα . following the procedure of Lindgren and Morrison (1986) with some modiﬁcations. it merely adds ˆ -space to each model function.13) (8.16) from the left with Ω and using (8. we may conﬁne our attention to the generalized equation.7). (Eα − H (8.15) (0) ˆ =H ˆ0 + V ˆ . Ψα = Φα + Q ˆ Ψα = Φα . A more conventional derivation.15) in the form ˆ 0 )ΩΦα = V ˆ ΩΦα .1 Formal quasidegenerate perturbation theory (QDPT) 187 than one model function.5. ˆ ΩΦ α = P ˆ Ψα = Φα . P ΩΨα = ΩΦα = Ψα . (8. P (8.

which is the generalized Bloch equation (8.2) with Ω = ΩP ˆ Operating on (8. can be of (8. furthermore.19) from the left with Q ˆ [Ω. the P ˆ . we obtain ˆH ˆ ΩΦα = Eα Φα . As in Section 2.159) but these deﬁnitions are equivalent in P be shown. H ˆ 0 ]P ˆα = Q ˆV ˆ ΩP ˆα − Q ˆ ΩV ˆ ΩP ˆα .7). ˆ 0P = H where ˆ =P ˆW ˆP ˆ=P ˆV ˆΩ W (8. as will now deﬁnition (2. obtaining ˆ =H ˆ 0 +V order part and a correction by substituting H ˆ eﬀ = P ˆH ˆ 0Ω + P ˆV ˆΩ H ˆ+W ˆ .24) (8. since the Eα no longer appear.25) . H (8.10).18) This equation must be true for all Φα . obtaining replace Φα with P ˆ Ω − ΩV ˆΩ.22) ˆ -space and shows that the correctly This equation operates entirely in P mixed zero-order functions Φα and thus the transformation coeﬃcients Cβα ˆ . also called the reaction operator. (8.23) (8. P Deﬁning an eﬀective Hamiltonian ˆ eﬀ ≡ P ˆH ˆ ΩP ˆ=P ˆH ˆΩ. In order to obtain the Q ˆ ΩP ˆ part we operate on the generalized Bloch P ˆ and from the right with P ˆα : equation (8. we eliminate the term containing Eα and obtain an implicit equation for Ω.188 Open-shell and quasidegenerate perturbation theory Subtracting (8. ˆ 0 )Φα = (V ˆΩ ˆ − ΩV ˆ Ω)Φα .21) of H ˆ -space.21) (8. H0 ] = V ˆ.16). ˆ eﬀ has an appearance that is diﬀerent from the The deﬁnition (8. for W ˆ ΩP ˆ part of the wave operator is given simply by As seen in (8. Q (8. it must be true for any linear combination of the Φα .19) is the level-shift operator.5. H we ﬁnd that ˆ eﬀ Φα = Eα Φα . This equation ˆ corresponds to (2. ˆ 0 Ω + ΩH (−H (8.20) (8. as well as the corresponding eigenenergies Eα of H ˆ -space eigenvalue problem for the non-Hermitian obtained by solving the P eﬀ ˆ eﬀective Hamiltonian H . [Ω.17) from (8.9).15) from the left with P . so we may ˆ .179). we split the eﬀective Hamiltonian into a zeroˆ into (8.21).

186).33) etc. Q ˆαV ˆ Ω(n) P ˆα = R ˆQ ˆ Ω(n−1) P ˆα − Q and ˆ (1) = P ˆV ˆP ˆ. ˆ+Q ˆ Ω(1) + Q Ω=P ˆ =W ˆ (1) + W ˆ (2) + · · · W ˆ (0) = 0). the zero-order resolvent for the model function Φα . Φα = Φ0 .26) order by order. where the lower-case Greek indices run over model states and the upperˆ -space) states.27) ˆ in orders of V ˆ. is given by where R (0) ˆ )− 1 Q ˆ≡ ˆα = Q ˆ (Eα ˆH ˆ 0Q R −Q ˆ Q (0) ˆ0 Eα − H .28) (8.26) ˆ α .31) n−1 m=1 (8.31) can be obtained by deﬁning the matrix elements ˆ |Φ α .35) .34) where |χα = I |I ΩIα (8.8. (8.1 Formal quasidegenerate perturbation theory (QDPT) 189 Applying the procedure of (2.30) These equations are equivalent to (2.30).. Expanding Ω and W ˆ Ω(2) + · · · . we can solve (8. (8. and are valid for both QDPT and nondegenerate RSPT. The perturbed case Latin indices run over the orthogonal (Q wave functions are then given by |Ψα = Ω|Φα = |Φα + I |I ΩIα = |Φα + |χα .189). (8. ˆ ΩP ˆα − R =R (8. (8. W ˆV ˆQ ˆ Ω(n−1) ˆ (n) = P W (n ≥ 2) . (2. the commutator equation is replaced by the following equation involving a resolvent: ˆ ΩP ˆα − R ˆ ΩP ˆα = R ˆαV ˆ α ΩV ˆ ΩP ˆα Q ˆαV ˆ α ΩW ˆP ˆα . Wβα = Φβ |W ΩIα = ΦI |Ω|Φα (8. A more explicit form of (8. for which there is only one model state. (2. obtaining (since W ˆαV ˆ Ω(1) P ˆα = R ˆP ˆα .32) (8.190).29) ˆ α Ω(m) W ˆ (n−m) P ˆα R (n ≥ 2) (8.185).

DαI (8. this expression simpliﬁes to I = Rα 1 (0) Eα − (0) EI = 1 .39) corresponds to the renormalization term in nondegenerate RSPT but. Successively replacing high-order terms on the right-hand sides of (8.38) The diagonal case is assumed in the diagrammatic formulation that follows. Wβα = I (n) (1) VβI ΩIα (n−1) (n ≥ 2) . (8. We also deﬁne I (0) ˆ )− 1 Q ˆ |ΦI ≡ ΦI ˆ α |ΦI = ΦI |Q ˆ (Eα ˆH ˆ 0Q = ΦI |R −Q Rα ˆ Q (0) Eα − H0 ΦI . (8. ΩIα = Rα (1) ΩIα = J (n) I Rα VIJ ΩJα (n−1) n−1 − m=1 β I Rα ΩIβ Wβα (m) (n−m) (n ≥ 2) (8. in which this term cancels unlinked-diagram contributions to the principal term in the manybody form of PT.39).36) where the last form is short-hand notation for the more complete expression. these equations can be expressed fully in terms of resolvent operators and matrix elements of the . unlike the nondegenerate case. in which H Φα states.37) where (0) DαI = Eα − EI . as will be shown explicitly in the diagrammatic formulation that follows.190 Open-shell and quasidegenerate perturbation theory is the perturbation correction to |Φα . (8.40) The last term in (8. In either the diagonal or the non-diagonal case we then have I VIα .40) by expressions in terms of lower-order terms. certain parts of the renormalization term survive in the form of “folded diagrams” in the degenerate and quasidegenerate cases. ˆ 0 is diagonal in the space of all the ΦI and In the diagonal case.39) and Wβα = Vβα . (0) (8.

43) = Wβα = ΩIα = − + (3) − (8. I I Rα Rβ VIβ Vβα .41) (8. βγ I I VβI Rα Rγ VIγ Vγα .44) result from substitution (3) of (8.44) The ﬁrst two terms on the right-hand side of (8.1 Formal quasidegenerate perturbation theory (QDPT) 191 ˆ . Deﬁning the cumulative nth-order eﬀective Hamiltonian by n ˆ0 + ˆ eﬀ(n) ≡ H H m=1 ˆ (m) W (n ≥ 1) . which is found from (8. The results for the second and third orders are perturbation operator V Wβα = I (2) ΩIα (3) (2) I VβI Rα VIα .8.42) into the principal term of ΩIα . (8. H γβ βα β (8.42) (8. The third term corresponds to the m = 1 contribution to the renormalization term. I J Rα VIJ Rα VJα − J β I J VβI Rα VIJ Rα VJα − IJ Iγ I J K Rα VIJ Rα VJK Rα VKα − JK Jβ I I J Rα Rβ VIβ VβJ Rα VJα Jβ I J I Rα VIJ Rα Rβ VJβ Vβα I I J Rα Rβ VIJ Rβ VJβ Vβα Jβ I I I Rα Rβ Rγ VIγ Vγβ Vβα .46) the coeﬃcients Cβα that deﬁne the transformation of the nth order perˆ and turbed functions to the proper eigenfunctions of the full Hamiltonian H also the corresponding nth order eigenvalues Eα the non-Hermitian matrix eigenvalue problem (n) (n) are obtained by solving n (n) ˆ eﬀ(n) C (n) = Cγα Eα . which are based on expressions in terms of resolvents and perturbation-operator components only.39). (8. These explicit forms are easily connected to the diagrammatic representations. while the last two terms correspond to the m = 2 contribution (since (2) ΩIβ generates two terms). (8.47) .45) so that n eﬀ(n) Hβα (0) ˆ eﬀ(n) |Φα = δβα Eα = Φβ |H + m=1 Wβα (m) (n ≥ 1) .

3). and obviously. 8. j. with the second-quantization notation for operators. v. as before. as in nondegenerate PT. In QDPT there are multiple reference determinants.132) and (3. refer to the hole and particle states. the partitioning of the Hamiltonian and the normal-ordered form of its components are the same in QDPT as in nondegenerate PT (Section 3. are used for the valence states while the labels i. model states and the virtual states are unoccupied in all the model states. Nevertheless. which determines the Fermi level separating the one-electron states into particle and hole states. for example. In nondegenerate many-body PT the Fermi vacuum state. . respectively.5 and subsection 3. Usually this is a closed-shell state. The core states are those spinorbitals that are occupied in all the reference determinants (model states). . the summations . in which the number of electrons does not appear explicitly. The model states are then described by the action of valence-state creation operators on the Fermi vacuum. .1. ˆ† v ˆ† w ˆ † |0 . b. is then placed between the core and valence states. (3. . . it contains fewer electrons than the target states of the calculation. in which the boundary between the hole and particle states is put either within or at the top of the valence space.192 Open-shell and quasidegenerate perturbation theory 8.6. .129). as shown schematically in the level diagram in Fig. In the present treatment we shall limit ourselves to the ﬁrstmentioned choice for the Fermi vacuum. In these cases the model states are described by the action of valence-state annihilation (and possibly creation) operators on the Fermi vacuum. c. We ˆ.1. Commonly (Sandars 1969). the same operator deﬁnitions can still be used for the Fermi vacuum state and for the states of interest in the calculation. so the assignment of the Fermi vacuum is not immediately obvious. The Fermi level.2 The Fermi vacuum and the model states The one-electron states (spinorbitals) in QDPT are divided into three classes. With the Fermi level placed at the bottom of the valence space. k. . consisting of all the one-electron states shared by all the model states.48) Other choices for the Fermi vacuum are possible. and a. w. With hole and particle states deﬁned as in Fig. the valence states are occupied in some. the Fermi vacuum state contains fewer electrons than the model states and so cannot be one of them. but not all.133). the boundary between the particle and hole states. |Φuvw = u (8. is the reference determinant. . Here the labels u. . the Fermi vacuum state in QDPT is chosen as a determinant made up of all the core states. shall also continue to use the same deﬁnition of the auxiliary operator U see (3. 8.

but they are often used because of the resulting simpliﬁcations. .e. .) . even if the size of the model space is not excessive.130). .6. adversely impacting the feasibility of the calculation. j. If we want this operator to be diagonal (as in canonical HF) or block diagonal (as in noncanonical HF). all possible assignments of the valence electrons in the valence states are included in the model states. .133) being over the hole (core) states. is the Fock operator for the perturbation operator. The operator F Fermi vacuum state.3 in the normal-ordered form of the ˆ .132). such a model space is often inordinately large. in order to take advantage of the resulting elimination or reduction of diagrams involving one-electron operators. . . Second. (3.8. . . (3. Classiﬁcation of one-electron states in QDPT. . . Valence (u. First. a complete model space often includes many high-energy determinants that are not expected to contribute signiﬁcantly to the wave function.) . v. this choice retains the simpliﬁcations seen in subsection 3. Such high-energy determinants result when multiple valence electrons are placed in the highest valence orbitals. . While the choice of a complete model space simpliﬁes the formalism.2 The Fermi vacuum and the model states 193 . 8. which are often . i. Core Holes (i. . we need to use the HF wave function and one-electron states corresponding to the Fermi vacuum state.1. in (3. In the most common form of diagrammatic QDPT the model space is chosen to be complete. . it has two important disadvantages.) Fig. Virtual Particles (a. . b. . It should be realized that the HF spinorbitals of the Fermi vacuum state and their orbital energies may be less than optimal for rapid convergence of the perturbation expansion for the quasidegenerate states (an alternative choice may be the use of MCSCF spinorbitals of the model space).

only the ﬁrst-order part is modiﬁed. Intruder states are Q with zero-order energies lying within the range of model-state zero-order energies. =W (8.6. i. Nevertheless the relative formal simplicity of this approach is instructive. will automatically be in normal order. between the model states and all the orthogonalspace states. i.6. while H The other operators obtained by contractions based on the generalized Wick’s theorem. in the zero-order energies.7. ˆ 0 + 0|V ergy in the nondegenerate theory. and few applications of such methods to electronic structure problems have been made. ˆ ≡W W (8. Such states usually generate small denominators in the perturbation expansions.29) for W ˆ W (n) ˆ (n) − δn1 0|V ˆ |0 .3). It is also convenient to deﬁne a modiﬁed level-shift operator ˆ − 0|V ˆ |0 .49) where |0 represents the Fermi vacuum. leading to divergence or extremely slow convergence.50) is the analog of the correlation enˆ |0 is the analog of Eref .51) and the eﬀective Hamiltonian can be expanded as ˆ eﬀ = H ˆ 0 + 0|V ˆ |0 + W ˆ H (1) ˆ (2) + W ˆ (3) + · · · .194 Open-shell and quasidegenerate perturbation theory ˆ -space) determinants. and so the resulting diagrammatic methods will be described in detail in Sections 8. .e. ˆ − 0|V ˆ |0 .3 Normal-product form of the generalized Bloch equations In order to use in degenerate MBPT the generalized Wick’s theorem and the diagrammatic techniques employed so eﬀectively in the nondegenerate case. The convergence of a PT expansion is generally enhanced by having a wide gap.e. the various components of the wave operator and the level-shift operator. the core state.52) The modiﬁed level-shift operator (8. +W (8. leading higher in energy than some of the external (Q ˆ -space states to the phenomenon of intruder states.4–8.50) ˆ . 8. ˆN = V V (8. As a result of these diﬃculties. Formalisms for incomplete model spaces in QDPT are discussed in Section 8. Such a gap is usually not achievable with complete model spaces. even when they do not lead to intruder states. it is most convenient to use the normal-product form of the perturbation operator (see subsection 3. In the series expansion (8. complete-model-space methods rarely provide practical solutions for problems involving quasidegenerate states.

= Vβα − δβα 0|V ˆN V Iα = VIα .3 and to apply the contractions based on the generalized Wick’s theorem in the diagrammatic development of QDPT.4 Diagrammatic notation for QDPT The diagrammatic notation for hole and particle states is the same in QDPT as in nondegenerate PT. (8. we shall omit the N subscript and the prime on the perturbation and level-shift operators and their matrix elements.30). = VIJ − δIJ 0|V ˆN V βα ˆ |0 .31) remain valid if V relevant elements of these operators are given by ˆN V IJ ˆ |0 . Note in particular that the equation in (8. n ≥ 2. For example. The ˆ and W ˆ (1) are replaced by V (8. the order-by-order generalized Bloch equations ˆN and W ˆ (1) .6.4 Diagrammatic notation for QDPT 195 With these deﬁnitions. In the rest of this chapter we shall assume that the various operators and their matrix elements have been replaced by their modiﬁed versions but.53) (8.8. (8. 8. However.39) for ΩIα the modiﬁcations in the principal and renormalization terms cancel.1). Wβα = Wβα − δβα 0|V The detailed forms of the generalized Bloch equations (8. 8. . These are conventionally denoted by upward-pointing double arrows (Sandars 1969). since summations over particle states include the valence as well as the virtual states (using the arrangement in Fig.44) also remain valid if the matrix elements in them are replaced by these modiﬁed (n) versions. In the equation in (8.54) (1) (1) ˆ |0 . We shall therefore be able to use the normal-ordered perturbation operator in the form given in subsection 3.40) for Wβα . in a case with three valence electrons the model state Φuvw can be represented diagrammatically by u v w . we also need a notation for spinorbitals that are restricted to valence states.39)–(8. is not aﬀected by the modiﬁcations because the oﬀ-diagonal elements VαI (n) are not modiﬁed. Additional notation is required if the Fermi vacuum is chosen to include all or some of the valence orbitals. in order not to encumber the notation.

any labels on incoming valence lines may be equal to labels on the outgoing lines.196 Open-shell and quasidegenerate perturbation theory ˆ (the QDPT analog of the enComponents of the level-shift operator W ergy) may be represented schematically by the diagram β ˆ |α α| = |β β |W ˆ W α where the heavy lines represent many-electron states (in this case. Some valence lines in a diagram may pass through it without interaction. the corresponding matrix element Wβα is recovQ ered by placing this operator between the respective ket and bra states. However. To avoid confusion between the generic Φα notation and the speciﬁc Φuvw notation.57) y w. ˆ† v ˆ† w ˆ † |0 and |β = |Φxyz = If the model states are |α = |Φuvw = u † † † ˆz ˆ |0 then the operator can be represented schematically by the diagram |x ˆy x ˆ |Φuvw w ˆ |Φuvw {x ˆ† z ˆ† Φxyz |W ˆv ˆu ˆ = Φxyz |W ˆ† y ˆ† z ˆ† w ˆv ˆu ˆ} = x ˆ† y u y ˆ W v w (8. the model states Φα and Φβ ) rather than one-electron states. Again. Clearly. and the same must apply to the outgoing valence lines.55) . z . otherwise the corresponding determinants would describe null states. for ˆ -space states. as in x ˆ |Φuv w ˆ |Φuv {x ˆ† w ˆ † Φxy |W ˆv ˆu ˆ = Φxy |W ˆ† y ˆ† w ˆ†w ˆv ˆu ˆ} = x ˆ† y u ˆ W v (8. It will sometimes be convenient to identify this diagram just with the matrix element Wβα instead of the full operator. for model states and |I etc. it will sometimes be convenient to identify this diagram just with the ˆ |Φuvw .56) Note that the creation and annihilation operator product is in normal order. we designate the many-electron basis states by |α etc. all the labels on the incoming matrix element Φxyz |W valence lines should be diﬀerent from each other. (8. Obviously.

A speciﬁc matrix element ΩJα is obtained by placing the sum (8.61) w u v . Wave-operator diagrams are the QDPT analogs of wave-function diagrams in nondegenerate PT.60) where a label in parentheses (or no label) indicates a dummy (summation) index. participate in determining the identity of the initial and ﬁnal states involved but do not aﬀect the matrix-element value. (I ) |χα α| = I |I ΩIα α| = Ω α = Ω α . with the number of outgoing lines equal to the number of incoming lines.4 Diagrammatic notation for QDPT 197 Such noninteracting valence lines. respectively. hole and valence lines. at the top then preservation of the number of electrons requires that nv + n p − n h = n v . (8.8. so that the diagram in (8. They can be represented schematically in the form I |I I |Ω|α α| = |I ΩIα α| = Ω α ˆ -space state and may represent a mix of where the top line designates a Q particle. (8. To take a speciﬁc example. nh are the number of open valence. Speciﬁcally. called passive lines.59) .58) represents the perturbation correction |χα to the model state |α . They are usually omitted from the diagram.58) ˆ -space Usually wave-operator diagrams are summed over the outgoing Q states |I . with the understanding that the diagram represents all allowed assignments of ˆ diagrams all open lines valence labels to the omitted passive lines. =a ˆ†ˆ ˆ† x ˆ† |0 then if |α = |Φuvw and |I = |Φabcx iˆ b† c i (a) (i) iˆ b† c Φabcx |Ω|Φuvw {a ˆ†ˆ ˆ† x ˆ† w ˆv ˆu ˆ} = i aibcx (b) (c) (x) Ω . (8. np .60) between the bra state J | and the ket state |α . particle and hole lines. (8. For W must be valence lines. if nv is the number of entering valence lines at the bottom of the diagram and nv .

In particular. all of which are included among the QDPT diagrams. not the value.39) can be represented schematically by the two diagram equations I Ω(1) α = I ˆα R I ˆ V α (8. the number of incoming valence lines is limited to the range nv = 0.62) . . This linked-diagram theorem for QDPT was ﬁrst proved by Brandow (1966. Nv . the operator product is in normal order. . Using this broadened deﬁnition. 1. nv = 0 represents the diagrams of nondegenerate PT. It is understood that the balance of Nv − nv valence lines in each diagram is made up of implied passive lines. All QDPT diagrams can be classiﬁed according to the number of incoming valence lines nv . where Nv is the number of valence electrons. of the matrix element and are usually omitted from the diagrams.5 Schematic representation of the generalized Bloch equation The explicit form of the generalized Bloch equation (8. assuming a complete model space. The number of omitted passive lines is Nv − nv .2). . . Outgoing open lines (other than passive lines) in wave-operator diagrams may include valence lines in addition to general particle lines. just as in nondegenerate PT. open valence lines are disregarded in determining the linked or unlinked status of a diagram. For a problem involving Nv valence electrons. 8. A disconnected part of a diagram is considered unlinked if it has no open lines other than valence lines. As in the case of the W diagrams. 1967). The deﬁnition of linked and unlinked diagrams needs to be broadened in QDPT. such lines have their origin in the secondary term of the generalized Bloch equation (8.198 Open-shell and quasidegenerate perturbation theory ˆ Here. In other words. see also Sandars (1969) and Lindgren (1974). too. the cancellation of unlinked diagrams holds in QDPT. As will be shown later. in the case of the level-shift diagrams the nondegenerate PT (nv = 0) energy diagrams contribute equally to all diagonal elements of the eﬀective Hamiltonian (and do not contribute to oﬀ-diagonal elements). passive valence lines aﬀect only the identity.

we can see a problem in using the rule ˆ I .e. DαI (8.62) for Ω(1) ﬁrst.63) Ω(n−1) α ˆ (n−m) W α Again. This value can be obtained from the above rule if we fold the valence lines for state α upwards and treat them as hole lines. indicate that these are dummy ˆ equations (8. In the current case we need (0) (0) the denominator DαI = Eα − EI . The W ically in the form β β ˆ (1) = W α β ˆ .40) can be represented schemat(summation) indices.65) The circle around the double arrow is intended to indicate that the folded valence lines represent particle states rather than hole states. as in the following schematic diagram: I ˆI R α ˆ V α = ˆ |α α | |I I |V (0) Eα (n ≥ 2) . the negative of the zero-order energy of state I relative to the Fermi vacuum. Similarly. around J and β . V α β ˆ (n) W α = ˆ V (I ) Ω(n−1) α Considering (8.5 Schematic representation of the generalized Bloch equation 199 and I ˆ V (J ) n−1 I Ω(n) α = I ˆα R ˆI R α − m=1 I Ω(m) (β ) (n ≥ 2) .e. In nonfor the determination of energy denominators in the resolvent R α degenerate PT the denominator was determined as the sum of the orbital energies of the hole lines crossed by the resolvent line minus the sum of the orbital energies of the particle lines crossed by it. i. (8. (8.8. This rule resulted in a (0) (0) denominator value D0I = E0 − EI . . the negative of the zero-order energy of state I relative to the model state α. the parentheses. i.64) |I ΩIα α| = (1) − (0) EI = |I VIα α| .

Looking at the schematic representation of the secondary term in (8. In nondegenerate PT the W operator just equals the number E (n−m) . as will be shown next. In the QDPT case there will ˆ (n−m) operator is disconnected and therestill be diagrams in which the W fore unlinked. not for the purpose of contracting successive operators in accordance with Wick’s theorem. For the purpose of the latter contractions the valence lines continue to act as particle lines rather than hole lines.200 Open-shell and quasidegenerate perturbation theory for the principal term (Ω(n) )1 in the Ω(n) equation. and its diagrammatic representation has no open lines.66) DαI Jα Obviously. called a folded diagram.63). see Section 8. On the other hand. But the terms in which the W is connected to the Ω(m) operator survive and give rise to a new type of diagram speciﬁc to QDPT. However. the ﬁrst term on the right-hand side of (8. The secondary term (Ω(n) )2 in the wave-operator equation (8. it is understood that these lines are to be folded only for the purpose of determining the appropriate denominators. the folded valence lines of model state α will be crossed by any J) ˆα resolvent line within the Ω(n−1) block (including its own top resolvent R and will contribute to the corresponding denominators.4).63). and these contributions will still cancel the unlinked diagrams in the principal term (using the broadened deﬁnition of unlinked diagrams ˆ (n−m) operator in QDPT. The term “folded diagrams” is applied instead to diagrams representing the secondary term in the Bloch equation for the wave operator. we have I ˆ V (J ) α = J I ˆα R |I (n) ΩIα 1 α| = |I Ω(n−1) VIJ (n−1) Ω α| . Therefore this energy factor is disconnected from the wave-operator factor and can be represented as an insertion that cancels the unlinked diagrams in the principal term. the model state α that enters the W ˆI ) any resolvent lines within the Ω(m) block (including its top resolvent R β must refer to the β state that enters that block. in the conventional drawing of QDPT diagrams the incoming valence lines are left unfolded. (8. The correct denominators .63) repreˆ (n−m) sents new challenges speciﬁc to QDPT. on the one hand we see that the resolvent that appears above the Ω(m) block in that diagram refers not to the model states β that enter that block but to ˆ (n−m) block below it.

5 Schematic representation of the generalized Bloch equation 201 can be recovered. In general. in the following way: ˆI R α |I (n) ΩIα 2 n−1 α| = m=1 I (β ) Ω(m) n−1 (n−m) (m) Wβα ΩIβ DαI α ˆ (n−m) W = −|I m=1 β α|. the resolvent lines within within the Ω(m) and W the Ω(m) block are to be interpreted relative to the model state β entering it. employing the usual rule. are to be in (8. as well as the valence lines entering the W block. ignoring the descending part of the .67) However. The minus sign in the interpretation of the diagram originates from the minus sign in front of the secondary term in the generalized Bloch ˆ (n−m) blocks may contain folds equation. we have a −1 factor from each internal fold in a diagram. since the Ω(m) and W of their own (originating from lower-order factors inside them). Further points need to be made concerning the scope of the resolvent lines ˆ (n−m) blocks. ˆI R α I (β ) Ω(m) α ˆ (n−m) W it being understood that the denominators are read from the fully folded diagram (8. Obviously.67).8. ignoring the α state. conventionally this schematic diagram is drawn without folding the incoming α lines. by folding the valence lines ˆ (n−m) connecting the two blocks. however.67). Resolvent lines within the W interpreted relative to model state α. as shown schematically by the thin horizontal lines ˆ (n−m) block. (8.

It can replace the folding of the incoming valence lines in all diagrams as well as the folding of the intermediate valence lines in the folded diagrams that represent the I is renormalization terms. With this notation the schematic form (8. to be described in Section 8. (8.63) of the generalized Bloch equation can be rewritten as I Ω(1) α = I ˆ V α (8.and P ˆ -space states. Thus the lower part of the resolvent Q line crosses valence lines only and. while the state crossed by the upper part may include valence lines. it consists of the folding of the resolvent lines instead of the state lines. If any folded line is crossed twice by a resolvent line (once as the folded line descends and again as it ascends). the summation over particle labels in it has to be restricted in many diagrams to prevent it from becoming a model state.62). An alternative to the folding of the state lines in the QDPT diagrams was introduced by Kucharski and Bartlett (1988) in the context of the Hose– Kaldor approach for incomplete model spaces.202 Open-shell and quasidegenerate perturbation theory β folded line emerging from it. The denominator DαI is obtained from the orbital energies of the state lines crossed by the resolvent line according to → → ← DαI = i εi − a εa − u εu . as also shown in (8.67). (8. respectively. as represented schematically by I α .68) where the arrows on the summation signs limit the summations to the state lines crossed by the folded resolvent line in the indicated direction. its eﬀects cancel in the corresponding denominator. The states crossed by the upper and lower parts of the resolvent line must be ˆ . In this procedure a resolvent term such as Rα represented by a directed line that crosses the state lines for state I from left to right and then curves down to cross the state lines for state α from right to left.69) .7. This approach can be used just as well in the context of the diagrammatic representation discussed here for complete model spaces.

ij. β |H α ˆ 0 |I = δJI E (0) . No resolvents are involved in these diagrams. 8.1 Zero and ﬁrst order ˆ eﬀ is just the zero-order The zero-order part of the eﬀective Hamiltonian H ˆ 0 . This notation is particularly eﬀective in describing the renormalization terms without resort to folded state lines.72) εi (8. which is diagonal in the model functions {α} and the Hamiltonian H orthogonal-space functions {I }. ) = E0 + εa + εb + · · · − εi − εj − · · · . The secsince the only open lines allowed in W ond of these diagrams is the non-HF contribution.. which vanishes if canonical HF spinorbitals of the Fermi-vacuum state are used. I |H (8.6.6 Level-shift and wave-operator diagrams 8....8..71) The zero-order energies of these functions are equal to the sum of the orbital energies of the one-particle states they contain: E (0) (Φuv.73) is the zero-order energy of the Fermi-vacuum state. where E0 = i (8. ) = E0 + εu + εv + · · · .52) is given by the ﬁrstThe 0|V order energy diagram (5. . The modiﬁed ﬁrstˆ (1) is given by the ASG diagrams order level-shift operator W w u x + v u w ×. J |H I ˆ 0 |α = 0 .74) ˆ diagrams are valence lines.. (8.6 Level-shift and wave-operator diagrams 203 and I I I Ω(n) α = ˆ V (J ) Ω(n−1) α n−1 Ω(m) (β ) (n ≥ 2) . ˆ 0 |α = δβα E (0) .6) for the Fermi-vacuum state.70) − m=1 ˆ (n−m) W α Note that the leading resolvents within the Ω(n−1) and Ω(m) blocks in (8. respectively. (8. ˆ |0 term in the eﬀective Hamiltonian (8.70) are represented by lines that fold down to cross the model states α and β . E (0) (Φab.

where x. generating a ﬁrst-order correction for the |Φxy. Φuvxy. it is important to restrict the summations over the outgoing particle labels in the ﬁrst and third of these diagrams to prevent the upper state from becoming a model state.16). Φxy.. | εab uv ab ab ... The secˆo |u {w ˆ†u ˆ}. the last of these is the non-HF contribution.. while the summation over a in the third diagram must be over virtual labels only. diagrams in which the number of incoming valence lines exceeds the number of valence electrons are to be ignored.16) provides a contribution ˆ|i a|f i} = {a ˆ†ˆ εa i ˆ|i axy. plus the ASG diagrams (a) u (b) + v u (b) (a) (i) + u (a) ×. provides a contribution ond diagram in (8. . a|f |Φi Φxy.74)...15) provides a contribution 1 4 ab ij 1 iˆ b† ˆ j} = {a ˆ†ˆ ab 4 εij ab ij |Φabxy.. provides a contribution wx uv to the matrix elements wx uv {w ˆ†x wxyz. .. including those cases in which one or both of the Φ incoming line labels u..75) provides a contribution 1 2 ab uv 1 {a ˆ †ˆ ˆu ˆ} = b† v ab εuv 2 ab uv |Φabxy... Speciﬁcally. the ﬁrst diagram in (8.. (1) uxy. ˆ ˆ |W |Φ . | ij ab εij abij abij to Ω(1) .4. model states for all x = y = · · · . The ﬁrst diagram in (8. whose value is w|f o wxy. (5.15). whose value is ˆ† v ˆu ˆ}.204 Open-shell and quasidegenerate perturbation theory In all QDPT diagrams it is to be understood that additional passive valence lines are implied that make the total number of incoming valence lines equal to the number of valence electrons. | εa i ai ai to the same ﬁrst-order correction... . Similarly....74). y. .. subsection 4. (1) ˆ |W |Φuvyz.75) Again. x. Thus labels a and b in the ﬁrst diagram cannot be allowed to take on valence values at the same time... label an appropriate number of passive lines.6).. However. not including the case w|f |u to the matrix elements Φ u = w (because of the oﬀ-diagonal nature of the one-electron perturbation. Because summations over particle labels include valence label values. The ﬁrst-order wave operator is given by the corresponding nondegenerate PT diagrams (5. v are equal to the outgoing line labels w. (8. the diagram (5.. The nondegenerate PT diagram (5.

8.6 Level-shift and wave-operator diagrams

205

to Ω(1) (terms in which both a and b take on valence values are excluded), generating a ﬁrst-order correction for the model state |Φuvxy... . The second diagram in (8.75) provides a contribution 1 2 ba ui 1 {a ˆ†ˆ ˆ} = iˆ b† u ab 2 εiu ba ui abxy... |Φi Φuxy... | εab iu

abi

abi

to Ω(1) , generating a ﬁrst-order correction for the |Φuxy... model state, and the third diagram provides a contribution ˆo |u a|f {a ˆ† u ˆ} = εa u ˆo |u a|f |Φaxy... Φuxy... | εa u

a

a

(in which the summation over a is restricted to virtual label values) to the same ﬁrst-order correction.

8.6.2 Second-order level-shift operator The matrix element of the second-order level-shift operator is given by (8.41). It can be expressed schematically in the folded-resolvent-line notation: β ˆ (2) W α = β ˆ V (I ) ˆ V α As in nondegenerate PT, a convenient way to generate higher-order QDPT ASG diagrams is to begin with Hugenholtz diagrams. This approach makes it easy to avoid redundant equivalent diagrams. The QDPT Hugenholtz diagrams for the second-order level-shift operator are given in Fig. 8.2. The ﬁrst and second rows in this ﬁgure represent the canonical HF and non-HF diagrams, respectively. The relevant ASG diagrams are shown in Fig. 8.3. As in other cases, the corresponding nondegenerate PT diagrams (5.9), (5.10), which are not shown here, are also to be included. Together, all these diagrams represent (8.41). In the diagrams in Fig. 8.3 without intermediate hole lines (diagrams 3, 6, 10, 11, 14), it is necessary to restrict the summations over the intermediate-particle labels so that not all these labels take on valence label ˆ projector in the resolvent). values at the same time (because of the Q

.

(8.76)

206

Open-shell and quasidegenerate perturbation theory

1

2 × ×

3

4 ×

5 × ×

6 × × × 13 14

7

× 9 10 11 12

× 15

8

Fig. 8.2. Quasidegenerate PT Hugenholtz diagrams for the second-order level-shift operator.

1

2

3

4 × ×

5

× 9 × × × × 15 10

6 ×

7

8 × ×

11

12

13

14

Fig. 8.3. Quasidegenerate PT ASG diagrams for the second-order level-shift operator.

8.6 Level-shift and wave-operator diagrams

207

For examples of the folded-resolvent-line notation for these diagrams we consider diagrams 1 and 2 of Fig. 8.3: 1 v (a) (i) u (b) = 1 2 vi ab ab ui {v ˆ† u ˆ} , εab ui

iab

2 (i) u

v (j ) (a) = u =− 1 2 (i)

v (j ) (a)

ija

ij ua va ij {v ˆ† u ˆ} . εav ij

The shortened resolvent line in the second form of diagram 2 reﬂects the fact that lines crossed twice by the folded resolvent line make no net contribution to the corresponding denominator. Obviously the same denominators are obtained with conventional resolvents and folded incoming valence lines. The weight factor is due to the presence of one pair of equivalent internal lines in each diagram. 8.6.3 Second-order wave operator The matrix elements of the second-order wave operator are given by (8.42) and are represented schematically in the folded-resolvent-line notation by I I Ω(2) α = ˆ V (J ) ˆ V α − I ˆ V (β ) ˆ V α . (8.77)

The corresponding ASG diagrams consist of the nondegenerate PT diagrams in Figs. 5.3 and 5.4 plus the diagrams in Figs. 8.4 and 8.5. This

208

Open-shell and quasidegenerate perturbation theory

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

Fig. 8.4. Quasidegenerate PT ASG HF diagrams for the second-order wave operator.

set of diagrams is the ﬁrst that contains folded diagrams (diagrams 28–30 in Fig. 8.4 and 58–61 in Fig. 8.5), representing the secondary contribution in (8.63). The diagrams in each of these ﬁgures are arranged in order of increasing number of incoming valence lines (separately for principal-term and folded diagrams). Diagrams 11, 12, 20–22, 25–27, 39-42, 45, 46, 51–57 are disconnected but not unlinked. The folded diagrams 28, 30, 60 are the

8.6 Level-shift and wave-operator diagrams

209

× × 31 32 × × 38 ×× × 45 46 × 47 × × × 52 53 × × 58 × 39 40 × 33

×

×

× ×

34

35 × × 41 × 42

36 × × × 43 × ×

37

× 44 ×

× 48 49

50 × ×

51

× 54 × × 59 60 61 55 56 × 57

×

Fig. 8.5. Quasidegenerate PT ASG non-HF diagrams for the second-order wave operator.

ﬁrst examples of wave-operator diagrams that contain outgoing valence lines ˆ part of the secondary term). The summations over (originating in the W particle labels in many of these diagrams must be restricted to comply with ˆ projectors in the resolvents. the Q No disconnected folded diagrams have been included in Figs. 8.4 and 8.5, because all such diagrams for the second-order wave operator are unlinked

210

Open-shell and quasidegenerate perturbation theory

and cancel against certain contributions to disconnected principal-term diagrams. In fact, folding any of the disconnected principal-term diagrams for the second-order wave operator produces “folded” diagrams in which there ˆ part in the second term of the second are no folded lines that connect the W equation in (8.39) with the rest of the diagram. The result is best described in the folded-resolvent-line notation. An example of the folding process for these disconnected diagrams and of the relevant cancellations is provided by the folding of diagram 25 of Fig. 8.4: (i) − − → ˆ (1) W =− 1 2 ab iu xy vw

iabxy ab εabxy iuvw εiu fold

(a)

(b) (x)

(y )

= u v w

{a ˆ†ˆ ˆ† y ˆ† w ˆv ˆu ˆ} . iˆ b† x

Comparing this diagram with the disconnected principal-term diagram 26 of Fig. 8.4, (i) (a) (b) (c) (d) = u v w 1 2 cd vw ab iu ˆ† w {a ˆ†ˆ ˆ† d ˆv ˆu ˆ} , iˆ b† c ab εabcd ε iuvw iu

iabcd

it is seen that the whole of the unlinked folded diagram cancels with those terms of the disconnected principal-term diagram in which c and d take on valence-label values (i.e. c = x, d = y ). The canceled terms in the latter diagram represent unlinked contributions to that diagram. Such cancellations occur between all unlinked folded diagrams and all unlinked contributions to the disconnected principal-term diagrams 21, 26, 27, 40, 46, 52, 54–57. The other disconnected diagrams have no unlinked contributions, either because both their disconnected parts contain hole lines (diagrams 11, 12, 22, 41, 42) or because the restriction that the intermediate state must not be a model state eliminates potential unlinked contributions to them (diagrams 20, 25, 39, 45, 51, 53). The result is that no unlinked diagrams or unlinked terms in disconnected diagrams survive. Thus the

8.6 Level-shift and wave-operator diagrams

211

summations over ﬁnal states in the disconnected diagrams must be restricted to reﬂect that result. As an example of the folded-resolvent-line notation for linked folded diagrams we consider diagram 28 of Fig. 8.4: (w) = (x) u v (i) (a) (b) (x) u =− 1 2 v (w)

(i)

(a)

(b)

iabwx

ab ix xw uv {a ˆ†ˆ ˆ†v ˆu ˆ} . iˆ b† w ab εabw ε iuv ix

No restriction over the a and b summations is needed here because of the presence of the hole line i. The algebraic interpretation of QDPT diagrams, including weight factors, phase factors and more complex types of folded diagram, are discussed in detail in the next two subsections.

8.6.4 Weight factors and phases ˆ or Ω, the free lines have When computing a speciﬁc matrix element of W ﬁxed labels and so may be ignored for the purpose of counting equivalent lines. Thus, in these cases the weight factor is 1 2 for each pair of equivalent internal lines. However, when summing over the orthogonal-space states in wave-operator diagrams to obtain the perturbative corrections to the model states, as in

n) = |χ( α I

|I ΩIα ,

(n)

(8.78)

open lines at the top of the Ω diagrams are included in the calculation of the weight factors. Each pair of open lines of the same type (hole, particle or valence) connected to the same vertex contributes a factor 1 2 . All open ˆ lines in W diagrams and at the bottom of Ω diagrams are always valence lines that specify a model state and are never summed over; therefore they do not contribute to the weight factors. The phase factors in QDPT diagrams depend on the order of the creation and annihilation operators in the operator product associated with the diˆ diagrams is relatively simple, since all operators agrams. The case of W

212

Open-shell and quasidegenerate perturbation theory

are valence-state operators. Each pair of open-line labels on the same continuous path in the diagram is represented by a creation operator and an annihilation operator in corresponding positions (counting inwards from the two ends) in the operator product. With this ordering of the operators, the phase factor is given by the usual PT rule as (−1)h−l , ignoring the open ˆ matrix element is determined lines. The identity of the corresponding W from the operator product based on the constructions ˆ† v ˆ† w ˆ † · · · |0 , |Φuvw... = u Φxyz... | = 0| · · · z ˆy ˆx ˆ. (8.79)

Indices representing the appropriate number of passive lines are added in corresponding positions. The same diagram value is assigned to matrix elements with all allowed values of the passive-line indices, i.e. all values which cause no duplication of indices in either the bra or ket states. As examples, we consider diagrams 4 and 5 in Fig. 8.3, shown ﬁrst in the folded-resolvent-line notation: w (i) u v w (i) u x (j ) v = 1 2

ij

(a)

x =

ia

wi ua ax iv {w ˆ†x ˆ† v ˆu ˆ} , εax iv

(8.80)

ij uv wx ij {w ˆ†x ˆ† v ˆu ˆ} . εwx ij

(8.81)

ˆ (2) |Φuvyz... , Both diagrams contribute to the matrix element Φwxyz... |W where y, z, . . . label passive lines. In the ﬁrst diagram we have one hole line and one loop, giving a phase factor +1, and no equivalent internal lines. In the second we have two hole lines and no loops, again giving a phase factor +1, and one pair of equivalent internal lines, giving a weight 1 2 . Alternatively, the denominators can be derived by folding the incoming valence lines: u w (i) x (a) u w (i) x (j ) v

v

Alternative assignments of operator sequences and phase factors can be based on the folded form of these diagrams. Taking the ﬁrst of the above

8.6 Level-shift and wave-operator diagrams

213

diagrams as an example, we can connect pairs of incoming and outgoing valence lines whose labels occupy corresponding positions in the initial and ﬁnal model states with dotted lines, forming quasiloops, and pair the corresponding creation and annihilation operators:

u

w (i)

x (a) v

=

ia

wi ua ax iv {(w ˆ † (ˆ x† v ˆ)ˆ u)} . εax iv

In this case we have included these quasiloops in the loop count and have counted the folded lines as holes (just as for the denominator determination), so this diagram now has three loops and three hole lines, and a phase factor +1, just as before. The alternative quasiloop connection,

u

w (i)

x (a) v

=−

ia

wi ua ax iv {(w ˆ † (ˆ x† u ˆ)ˆ v )}, εax iv

has two loops and a phase factor −1 and is equal in value to the previous form because of the odd permutation of the operators. This form of the diagram ˆ |Φvuyz... , which is consistent can be considered as contributing to Φwxyz... |W with the previous assignment because of the odd permutation of the indices in the ket side of the matrix element. As in the case of nondegenerate PT diagrams, this analysis shows that the choice of quasiloop connections does not aﬀect diagram values and justiﬁes the omission of these connections from the diagrams. The folding of the incoming valence lines could also have been done from the bottom of the diagram, as in w u (i) (a)

x v .

214

Open-shell and quasidegenerate perturbation theory

Since the resolvent line crosses the folded u line twice, the denominator is unaﬀected. To specify the phase of an orthogonal-space state we deﬁne it by

g... ˆ† g = (ˆ a†ˆ ˆ )f ˆ† · · · |0 , |Φab...ef i)(ˆ b† ˆ j ) · · · (ˆ e† m ij...m

(8.82)

where the superscript indices may represent either particle or valence labels. The labels in an orthogonal-space state fall into two classes: there are zero or more pairs, each pair consisting of a particle (or valence) label vertically above a hole label in the state speciﬁcation, such as (a, i), (b, j ), (e, m) in (8.82), and there are unpaired (excess) particle (or valence) labels, such as f , g . In a wave-operator matrix element ΩIα each unpaired particle or valence label in the ﬁnal state I is considered as paired with a valence label in the initial (model) state α if the two are in corresponding positions in the unpaired ﬁnal-state labels and the initial-state labels (ignoring all the g... |Ω|Φuv... , f is paired paired labels). Thus, in the matrix element Φab...ef ij...m with u and g is paired with v . Note that the operator product in (8.82) is in normal order and that creation and annihilation operator pairs such as (ˆ a†ˆ i) may be moved together freely across any other operators in the sequence without aﬀecting the resulting function or the normal order. In determining the phase factor for a wave-operator diagram, the simplest approach is to place paired labels on the same continuous path in the diagram. Each open hole line is then connected by a dotted line with its paired open particle line, creating a quasiloop, and the usual (−1)h−l rule for the phase factor is applied. As examples we consider diagrams 8 and 22 of Fig. 8.4, (i) (j ) (a) (b) (k ) (c) =− 1 4 kc ij ab uk ˆ†ˆ † ˆ † c j )(ˆ au {(b i)(ˆ ˆ)} , ab εabc uij εuk

abcijk

u (a) u v which contribute to Φbcax... |Ω|Φux... and Φbdacx... |Ω|Φuvx... , respectively, ij ij where x, . . . label passive lines. Each diagram has two pairs of equivalent (b) (i) (d) (c) (j ) = 1 4 ab ui cd vj ˆ†ˆ ˆ† ˆ † † a (ˆ {(b i)(d j )(ˆ cv ˆ)ˆ u)} , cd εabcd uvij εvj

abcdij

. u (a) (b) (i) (j ) (k ) (c) =− 1 4 abcijk kc ij ab uk {(ˆ c†ˆ ˆ)}..... ij ij These two forms now have three and one loops. which represent the secondary term in the generalized ˆ diagrams. counting the folded incoming valence lines as hole lines for the purpose of phase-factor determination as we have previously done for level-shift diagrams. corresponding to even permutations of the creation and annihilation operators and of the superscript indices in the bra state between the two forms.6 Level-shift and wave-operator diagrams 215 open lines. arise when the ΩIβ factor and/or the Wβα contributing to the secondary term in (8. and both have four hole lines. using a diﬀerent pairing scheme. 8.. Thus.. Repeated secondary terms. i)(ˆ a† ˆ j )(ˆ b† u ab εabc ε uij uk which contribute to Φbcax. hence the phase factor −1. c j )(ˆ au ab εabc uij εuk or.67) by its own secondary term produces the doubly folded diagram shown here .24) or (8.5 Folded diagrams Folded diagrams. and a phase factor of +1. In the ﬁrst diagram we have two loops and three hole lines. Normal order of the operator product is maintained as long as the valence annihilation operators are placed at the end of the operator string. occur for both Ω and W their folds from their Ω component. we can fold the incoming valence lines and connect all paired indices with dotted lines. |Ω|Φux.6.39) contain lower-order secondaryterm contributions of their own. respectively.8. respectively. An additional phase factor −1 is required for each fold.40). Alternatively. and Φcabx.. thus (i) (j ) (k ) (c) u (a) (b) =− 1 4 abcijk kc ij ab uk ˆ†ˆ † ˆ † {(b i)(ˆ ˆ)}. The number of folds is not determined by the number of internal folded lines but by the number of nested secondary terms represented in the diagram. The second diagram has two loops and two hole lines. replacing the Ω(m) factor in (8. contributing a weight factor 1 4 . and (m) (n−m) factor thus multiple folds. (8. . the W ˆ diagrams inherit Bloch equation. |Ω|Φux. giving a phase factor −1 in both cases.

(8.83) This type of doubly folded diagram ﬁrst occurs in the third-order wave operator and the fourth-order level-shift operator diagrams. (8. ˆV ˆQ ˆ Ω(n−m−1) factor in (8. replacing the W its own secondary term produces the doubly folded diagram shown here schematically in its fully folded form: ˆI R α ˆ V I (β ) ˆJ R α (J ) (γ ) Ω(n−m−l−1) Ω(m) (Ω(n−m−1) )2 (n−m−l−1) (l) α ˆ (l) W = ml Jβγ |I (m) ΩIβ VβJ ΩJγ Wγα DαI DαJ α| .84) This type of doubly folded diagram ﬁrst occurs in the fourth-order wave operator and the ﬁfth-order level-shift operator diagrams. .216 Open-shell and quasidegenerate perturbation theory schematically in its fully folded form: I ˆα R I ˆI R β (γ ) Ω(l) (Ω(m) )2 = ml βγ ˆ (n−m) W α ˆ (m−l) W (β ) |I ΩIγ (l) Wγβ (m−l) Wβα (n−m) DαI DβI α| .67) by ˆ (n−m) = P Similarly.

in which the identity of the intermediate states at the various level is more apparent: (a) I ˆ I : Rα ˆI R β (b) (j ) × ΩIβ (2) J: R ˆJ β β: K: α: u ˆK R α v (x) (i) (y ) (c) Wβα w (2) The initial. Note that the y line is passive .83). The doubly folded schematic diagrams of (8.3.6 Level-shift and wave-operator diagrams 217 The folded-resolvent-line notation is particularly eﬀective in representing all kinds of folded diagrams. We shall ﬁrst display it in the folded-resolventline notation.84) take the following forms in this notation: ˆI R α I ˆI R β Ω(l) (γ ) ˆ (m−l) W (β ) ˆ (n−m) W α Ω(m) 2 ˆI R α I Ω(m) (β ) ˆ (J ) V and ˆJ R α Ω(n−m−l−1) (γ ) ˆ (l) W α Ω(n−m−1) 2 An example of a singly folded diagram that illustrates some details of the interpretation of such diagrams is provided by the following fourthˆ (2) components in the order wave-operator diagram containing Ω(2) and W (2) (2) IΩ W ˆα expression −R Iβ βα . We have seen its application to an Ω(2) singly folded diagram in subsection 8.6. (8. and the resolvents required are also indicated. ﬁnal and intermediate states are identiﬁed on the left-hand side.8.

DIβ = Eβ − EI (0) (0) DJβ = Eβ − EJ = εab uj . It is easily seen that these denominators are correctly recovered from the fully folded form of the diagram: ˆI R α v (y ) ˆK R α (x) ˆI R β ˆJ R × (j ) ΩIβ (2) w u (a) (b) (i) (c) Wβα (2) β ˆ (2) section rather than conThe x line must come out of the top of the W necting its two end vertices with a straight line and the internal resolvent K within the W ˆ (2) section crosses the ascending part of the x line but not Rα the descending part nor any other line on the left-hand side of the diagram. The value of this diagram is given by − 1 2 ˆ|x iy vc xc iw ab uj j |f ijxabcy ab ab xc εaby uvw εux εuj εiw {a ˆ †ˆ ˆ† w ˆv ˆu ˆ} . b† y The factor 1 2 is due to the pair of equivalent lines a. (0) − EK = εxc DKα = Eα iw . . = εab ux . b. The negative sign is due to the single internal fold.218 Open-shell and quasidegenerate perturbation theory ˆ (2) diagram. There are two hole lines and no loops. The for the Ω(2) diagram while the u line is passive for the W denominators generated by the various resolvents are: (0) − EI DIα = Eα (0) (0) (0) (0) = εaby uvw . Openline labels on the same continuous path are associated with creation and annihilation operators in corresponding positions in the operator sequence (counting inwards from the two ends).

The upper parts of resolvent lines may not cross model states.84). |γ = Speciﬁcally.6 Level-shift and wave-operator diagrams 219 As an example of a doubly folded diagram we consider a fourth-order wave-operator diagram of the type described schematically in (8. |β = |Φzw . The fully folded form is I ˆα R (w) J ˆα R u v (i) (a) (b) (z ) (c) J ˆγ R (d) (x) (y ) ΩJγ (1) Wγα (1) (ΩJα )2 (2) ˆI R β ΩIβ (1) . so the labels c and d cannot both take valence index values at the same time.84): ˆI I : Rα β: (i) ˆI R β (z ) (c) J: ˆJ R α ˆJ R γ γ: (x) (y ) Wγα α: (u) (v ) (1) (a) (b) ΩIβ (1) (w) (d) ΩJγ (1) (ΩJα )2 (2) The many-electron states involved. |α = |Φuv . are identiﬁed on the left-hand side of the diagram. these states are |I = |Φabw i |Φxy . corresponding to the designations in the schematic diagram (8. . |J = |Φcd .8. Each intermediate model state (β and γ in this diagram) corresponds to a fold and is crossed only by the lower parts of the folded resolvent lines.

second and third vertices.. (x. We shall restrict ourselves here to discussing the generation of third-order level-shift-operator diagrams. y ). It provides a contribution to the matrix elements Φtw. The notation (ijk/lmn) will be used to describe a class of diagrams with i. |W 8. justifying the designation of the open line emerging from ˆ vertex as a valence line (w). ab cd xy εabw iuv εiz εuv εcd ˆ (5) |Φuv. The whole right-hand side of this diagram. we consider a ﬁfth-order level-shift-operator diagram obtained by adding an interaction vertex on top of the previous wave-operator diagram: t I ˆα R w u v Wγα (1) 2 (1) (i) (a) (b) J ˆα R (c) (d) (x) (z ) R J ˆγ ˆI R β 1 8 ΩIβ = (1) (y ) ΩJγ (Ω(2) )Jα ixyzabcd it ab ab iz zw cd cd xy xy uv ˆ† † {t w ˆ v ˆu ˆ} . including the top V (4) represents Wβα . j and k open valence lines entering the ﬁrst. Its value is given by 1 8 ab iz zw cd cd xy xy uv {a ˆ†ˆ ˆ†v ˆu ˆ} . one hole line. We shall consider principal-term diagrams (those without internal folds) and folded diagrams separately.. We shall classify the diagrams according to the number of open valence lines that enter and leave each of the three vertices. respectively . (c. This diagram has three equivalent pairs that V (a. one quasiloop involving the lines labeled a. making high-order calculations diﬃcult.6 Third-order level-shift operator The number of distinct diagrams in QDPT grows very rapidly with the order of the treatment.. b).220 Open-shell and quasidegenerate perturbation theory ˆ vertex. i and two internal folds.6. iˆ b† w ab εcd εxy εabw ε iuv iz uv cd iwxyzabcd Finally. . d). For the principal-term diagrams it is easiest to follow a similar procedure to that used for the second-order level-shift diagrams and begin with the Hugenholtz diagrams..

1. m and n open valence lines leaving these vertices. (200/110). (020/110). We shall list the linked diagrams for just one representative . (011/002). Obviously i + j + k = l + m + n. The various classes of diagrams can be organized in sets.1 and 5. For example. (101/002).2 correspond to the class (000/000) and are included in the expansions for the diagonal elements of the level-shift operator in the open-shell case. (011/020). (002/011).8. Representative classes.6 Level-shift and wave-operator diagrams 221 Table 8. numbers of distinct linked ASG diagrams in each class and set and total number of principal-term diagrams for the third-order level-shift operator Valence electrons 0 1 2 Representative classes (000/000) (001/100) (001/001) (002/200) (002/110) (002/101) (011/110) (011/011) (012/210) (012/201) (012/111) (012/021) (111/111) (022/211) (112/211) Classes in set 1 6 3 6 6 12 6 3 6 6 12 6 1 6 6 Diagrams in class HF 3 2 2 1 1 1 2 1 1 1 1 0 2 1 1 non-HF 11 7 4 1 3 2 3 2 0 1 1 1 0 0 0 Diagrams in set HF 3 12 6 6 6 12 12 3 6 6 12 0 2 6 6 10 108 non-HF 11 42 12 6 18 24 18 6 0 6 12 6 0 0 0 20 181 3 4 Folded Total (counting downwards from the top vertex) and l. (110/020). 5. (110/200) and (101/200) form one such set. (200/101). the 12 classes (002/101). The diﬀerent classes in a set all have the same number of diagrams and can be obtained from each other by permutation of the three vertices and/or by reversing all arrows (which includes converting all entering open valence lines to leaving open valence lines and vice versa). The closed-shell third-order energy diagrams of Figs. (020/011). respectively.

of third-order diagrams (including 30 folded diagrams. demonstrates the diﬃculty of carrying out higher-order QDPT calculations.. | ab εa ε v ij ijab ijab ˆ (3) |Φuv. The generation of the rest of the normal third-order diagrams and their interpretation are left as an exercise for the reader. Clearly. The phase factor is due to the presence of two hole lines and one loop... .. As an example we show the generation of a Hugenholtz diagram of the class (110/200) from the corresponding diagram in the representative class (002/101) by permutation of vertices (1 → 2 → 3 → 1) and reversal of arrows. The denominator reﬂects the folding of the incoming valence lines u and v . No disconnected diagrams are included since such diagrams would be unlinked. Obtaining all other Hugenholtz diagrams from the representative classes.222 Open-shell and quasidegenerate perturbation theory class from each set.... as shown later). 289. the number of distinct diagrams in each class and set and the total number of diagrams of HF and non-HF type. contributing to the matrix elements Φwx. including the interconversion of entering and leaving open valence lines. followed by its conversion to an ASG diagram: (002/101) (200/110) (110/200) w → → → u x (a) (i) (j ) (b) v This ASG diagram represents the contribution − 1 2 wx ua ij vb ab ij {w ˆ†x ˆ† v ˆu ˆ} ab a εv εij =− 1 2 wx ua ij vb ab ij |Φwx. The principal-term Hugenholtz diagrams for the representative classes are shown in Fig. . 8. the total number. |W to W over the summation sign indicates that the sum over a exclude valence-index values in order to avoid intermediate model states. The weight factor reﬂects the equivalence of the two hole lines i and j .1 lists the representative classes. Table 8. converting them to ASG diagrams and giving them an algebraic interpretation are straightforward.6 (apart from the closed-shell class (000/000)). in comparison with just 14 diagrams for the closed-shell case. The prime ˆ (3) . Φuv..

6.8. . Quasidegenerate PT Hugenholtz diagrams for representative classes of the principal term of the third-order level-shift operator.6 Level-shift and wave-operator diagrams (001/100) × × × × (002/200) × × × × × × × × × × × (001/001) × × × 223 (002/110) × (002/101) × × × × (011/110) × × × (011/011) × × (012/111) × × (012/021) (111/111) × × (012/210) (012/201) × × (022/211) (112/211) Fig. 8.

Quasidegenerate PT ASG folded diagrams for the third-order level-shift operator. 8.4 and 8.224 Open-shell and quasidegenerate perturbation theory × 28a 28b 28c 28d 28e × × × 28f 28g 29a 29b 29c 30a × × × 30b 30c 30d 30e 30f 58a × × 58b × 58c × × 58d × × 59a × × × 59b × 60a × 60b × × 60c × 60d × × × 60e 61a 61b × × × 61c Fig. The resulting set of folded ASG diagrams is . The folded third-order level-shift diagrams are best obtained by adding an interaction vertex to the top of the folded second-order wave-operator diagrams (diagrams 28–30 and 58–61 of Figs.7. 8. the Hugenholtz version should be examined in each case. To avoid generating equivalent diagrams.5) in all allowed ways.

. There are no equivalent internal lines.7.. wx εwxz ε iuv iy ˆ (3) |Φuvt. where This diagram contributes to the matrix elements Φxwt.6 Level-shift and wave-operator diagrams 225 displayed in Fig. Their schematic form in the folded-resolvent-line notation is I I Ω(3) α ˆ V ˆ = V ˆ V α (J ) (K ) ˆ V ˆ − V ˆ V α I (J ) (β ) ˆ V (β ) ˆ − V ˆ V α (J ) ˆ − V ˆ V (β ) α I ˆ V I (J ) ˆ V ˆ + V ˆ V (γ ) (β ) α (8.6. |W the indices beginning with t represent passive lines. three singly folded terms and one doubly folded term.7 Third-order wave operator Because of the proliferation of the number of diagrams with increasing order.44). shown here in both conventional and folded-resolvent-line form: (x) = (y ) u v (y ) u = iyz w x (i) (z ) × (w) (i) (z ) × v ˆ|z xw yi yz uv i|f {x ˆ† w ˆ†v ˆu ˆ} . . Even the number of skeletons is too large to present conveniently.8. The numbering of these diagrams identiﬁes the folded second-order wave-operator diagrams from which they have been generated. 8. butions of Q I .. 8. As an example of how to interpret the folded diagrams we shall consider diagram 28f.. This expression contains one principal term. The minus sign due to the fold is canceled by the minus sign due to the presence of one hole line (i) and no loops. Instead. we shall just give a schematic description of the ﬁve diﬀerent terms appearing in the expression for this operator in (8.85) It can be seen that the expression for Ω(3) contains all possible distriˆ -space and model-space states for the two internal state lines. we shall not present actual diagrams for the third-order wave operator.

An outer resolvent line that crosses the ﬁnal and initial states is always present.226 Open-shell and quasidegenerate perturbation theory Furthermore. in which we have folded the incoming model-state line labeled α in each term in order to be able to show the scope of all resolvent lines: I I Ω(3) α = (J ) ˆ V (K ) ˆ V α − I ˆ V (J ) ˆ V ˆ V + (β ) ˆ V ˆ V (γ ) (8. The opening and closing parentheses of each pair represent the upper and lower parts of the folded resolvent line. interspersed with state labels. paired according to the ordinary algebraic convention. Summations over intermediate states are implied. Note that the second and fourth terms in (8. without intervening parentheses. A compact notation for these schematic diagrams can be obtained using sets of matched parentheses. all possible arrangements of the three folded resolvent lines ˆ -space state and the lower are included in which the upper part crosses a Q part crosses a model state.87) I ˆ V α ˆ V (β ) . the resolvent lines do not cross each other.86) Perturbation operators are implied between the state labels in such a way ˆ -space-state label is immediately followed by a V ˆ operator and that each Q each model-state label is immediately preceded by such an operator. Using the conventional notation the same expression takes the following form.85) contain the same sequence of intermediate states but diﬀer in the states that are crossed by the middle resolvent line. (8. respectively: ΩIα = (I (J (Kα))) − (I ((Jβ )α)) − ((Iβ )(Jα)) − ((I (Jβ ))α) + (((Iγ )β )α) . no two resolvent lines cross the same pair of states and each state is crossed by at least one resolvent line. ˆ V α − (J ) ˆ V (β ) ˆ V I ˆ V α I − ˆ V (β ) ˆ V α (J ) (3) .

However.7 Incomplete model space 8. For each such Fermi vacuum. in which the folded-resolvent-line notation was ﬁrst introduced. for each diﬀerent Fermi vacuum used. while the fourth term repreˆα (ΩJα )2 . although. Kucharski and Bartlett 1988). as will be shown in the following analysis. the hole states consist of the core states plus the valence hole states of that vacuum.8.7 Incomplete model space 227 Note the scope of the middle resolvent line in the third term (compare (8. in which (ΩJα )2 = β R Jβ βα (2) (1) I ˆ R (Ω )1 W sents (the subscripts 1 and 2 refer to the principal and β α Iβ βα secondary terms. 1980. Thus all elements in the same column of the matrices for the wave and level-shift operators use the same Fermi vacuum. valence and virtual states. like most QDPT applications. The diﬀerence in the operator sequences in the second and fourth IV ˆα terms is due to the fact that the second term represents the sum J R IJ (2) (2) J Ω(1) W (1) . in (8. Hose and Kaldor 1984. The one-electron states in the HK formalism are classiﬁed into core. respectively. However.39)).g. the valence orbitals are classiﬁed as valence holes (the valence states occupied in that particular Fermi vacuum) and valence particles. A similar approach to multireference coupledcluster theory was formulated by Jeziorski and Monkhorst (1981). In such cases it results in incomplete cancellation of unlinked diagrams. 8. not beyond third order. The incoming model state is used as the Fermi vacuum for all matrix elements in which that state is the initial (ket) state. but the procedure is nevertheless quite practical and has been used successfully for various applications (e. Similarly.7. it can partly overcome the problems of intruder states and of model spaces that are too large for practical applications. When the Hose–Kaldor (HK) approach is applied to a complete model space it generates the same results. order by order (but not diagram by diagram).67)). the . Kaldor 1984. In presenting this formalism we shall use the diagrammatic approach of Kucharski and Bartlett (1988).1 The Hose–Kaldor approach A formalism for the use of incomplete model spaces in QDPT was introduced by Hose and Kaldor (1979. 1982). as in the previously described approach. but diﬀerent columns use diﬀerent Fermi vacuums. as the approach presented in the previous sections because it solves the same generalized Bloch equation. Their approach uses diﬀerent Fermi vacuum states for the calculation of the diﬀerent matrix elements of the wave and level-shift operators. because it can be applied easily to incomplete model spaces.

The selection of the model states is a major factor in determining the convergence behavior of the perturbation series. In skeleton representations a bar across the state line is used to indicate lines restricted to valence states. .2 The one-electron interaction ˆ operators that simpliﬁes the exˆN and V The cancellation between the U N pression for the perturbation in nondegenerate MBPT. . particle states consist of the virtual states plus the valence particle states. . Virtual Particles (a. because each denominator factor (Q in the various terms contributing to the level-shift and wave-operator matrix elements represents an energy diﬀerence between a model state and an external state. j.183). 8. .) Valence holes (u. . . In the diagrammatic notation. y.228 Open-shell and quasidegenerate perturbation theory . but it should be remembered that the sequence of the valence states in this ﬁgure does not reﬂect their energy order and that the classiﬁcation of the valence states depends on the model state used as the Fermi vacuum.7. v. . .2. can .) Core Fig. . Classiﬁcation of one-electron states in the Hose–Kaldor approach. This classiﬁcation is shown schematically in Fig. . .8. . a large energy gap between the model states on the one hand and all external ˆ -space) states on the other is desirable. 8. . . As noted in Section 8.176)–(3. . . valence states are distinguished by double arrows. (3. .8. . Valence particles (x.) Holes (i. . . 8. b.) . . Small denominator factors increase the magnitude of high-order contributions and retard (or even prevent) convergence. .

e. in the deﬁnition vpq = i pi qi .8. The one-electron component of the normal-product perturbation operator in the HK formalism is therefore given by the operator o ˆN ˆN − U ˆN = ˆN = F +V G pq gpq {p ˆ† q ˆ} .90) The last form in (8. is part of the deﬁnition of the Fock operator (3. (8. given in (3. is the most general. the summation over i is over the spinorbitals constituting the Fermi vacuum used for the deﬁnition of normal products for the current matrix element.7 Incomplete model space 229 no longer be used in the same way in the HK formalism of quasidegenerate PT. .130) and therefore the summation over i runs over the occupied spinorbitals in the determinant used as the reference state for that Fock operator. we use fpq = fpq − εp δpq = hpq + upq − εp δpq and deﬁne gpq = fpq + vpq − upq = hpq − εp δpq + i pi qi . However.89) More generally. being independent of the deﬁnition of the Fock operator and its reˆ are lationship to the orbital energies.88) where o ˆ|q = fpq + vpq − upq . owing to the varying deﬁnitions of the Fermi vacuum for the diﬀerent columns of the level-shift and wave-operator matrices and to the diﬀerence between these Fermi-vacuum deﬁnitions and the corresponding deﬁnition for the Fock operator. The deﬁnition upq = i pi qi .90). The reason for this lies in the diﬀerent summation ranges in the deﬁnitions of upq and vpq in the latter case. if we do not require εp = fpp .178). (3. A typical choice for the reference determinant is just the core state. (8. i. which include the current valence hole states. regardless of whether this determinant is an HF wave function for any particular state. in which the summation is over the hole states of the current Fermi vacuum (core plus valence holes). gpq = p|g (8. the determinant consisting of all the spinorbitals occupied in all the model states. Obviously the operators v ˆ and g diﬀerent for the diﬀerent columns of the level-shift and wave-operator matrices.133).

usually diﬀer between model states. depends upon the model state. However. which enter the diagonal elements of the ˆ eﬀ .230 Open-shell and quasidegenerate perturbation theory If the core determinant is used for the deﬁnition of the Fock operator. the initial state in the generalized Bloch equations (8.44) is the current vacuum state.152)). which beneﬁted from the cancellation of half the sum − i uii with 1 1 ˆ is not equal ij ij ij = 2 i vii (see (3.91) the sum over u running over the valence hole states for the current Fermi vacuum.7). reﬂecting the diﬀerences in the corresponding sets of hole states. it being understood that the deﬁnition of the corresponding interaction varies for the diﬀerent columns of the level-shift and wave-operator matrices.3 First-order diagrams The zero-order energies for the various model states are given by the usual formula. The diagonal elements of the level-shift operator are vacuum expectation values and are represented by closed diagrams that are mostly similar to those of nondegenerate MBPT.92) ˆ operator by the vertex In the diagrams we shall designate the G ⊗ . Summations over the ﬁnal states in wave-operator diagrams must exclude model states. there are open valence lines at the top of all diagrams for oﬀ-diagonal elements of the level-shift operator. (8. however. (8. as in the nondegenerate theory. and thus the cancellation is incomplete. In nondegenerate MBPT the ﬁrst-order energy was given by (5.6) or (5. In the present case v 2 to u ˆ (and is diﬀerent for diﬀerent model states).7. Instead we ﬁnd that the diagonal elements of the ﬁrst-order . All wave-operator diagrams are open at the top. the scope of the summation. In this case o + gpq = fpq u pu qu . reﬂecting the diﬀerences between the ﬁnal and initial states. i εi .39)–(8. |α = |0 . Thus the zero-order energy values. the diﬀerence between vpq and upq is given by vpq − upq = u pu qu . 8. and there are no lines below the lowest vertex in the diagrams of this formalism. eﬀective Hamiltonian matrix H Since the initial (ket) state in each matrix element serves as its own Fermi vacuum in the HK formalism.

the label values for the open lines are restricted to exclude states that are not part of the model space. (8. = x|g ˆ|u {x ˆ† u y = xy uv {x ˆ† y ˆ† v ˆu ˆ}. with matrix element ˆ (1) |0 = x|g ˆ|u .93) = −1 2 ij ij ij + i using (8. The remaining part of the ˆ (1) of Section 8.94) ˆ |0 of the unmodiﬁed perand represents the vacuum expectation value 0|V turbation (which is diﬀerent for each model state). The ﬁrst-order wave operator is represented by the same diagrams as the ﬁrst-order wave function in the nondegenerate case. The ﬁrst diagram represents a single replacement for the ﬁnal (bra) state relative to the initial (ket) state. When an incomplete model space is used. ﬁrst-order level-shift operator is the modiﬁed operator W (1) ˆ |0 = 0.15). but with . Diagrammatically this result is represented by (compare (5. Φxy uv |W (8. (8.7 Incomplete model space 231 level shift operator are given by fii − uii + 1 2 vii i = i − vii + 1 2 vii + i (fii + vii − uii ) gii .7)) − + ⊗ = −1 2 ij ij ij + i i|g ˆ|i (8.8.96) with the creation or annihilation operators acting on the respective initial state.90). and |Φx u and |Φuv are deﬁned relative to this particular ket state. (5.97) The second diagram represents a double replacement. Φx u |W (8.95) (8.16). it has no diagonal contribution here because 0|W The oﬀ-diagonal part of the ﬁrst-order level-shift operator is represented by the two diagrams x x u ⊗ u v ˆ}. (5.3. There are no ﬁrst-order contributions for more than two replacements.98) In both these equations |0 represents the ket state for the respective maxy trix element. with matrix element ˆ (1) |0 = xy uv .

99) (a) (b) = 1 4 (8. giving a total of 21 diagrams. εa i ab ij {a ˆ †ˆ b† ˆ jˆ i}.7.9.) Skeletons 1.100) abij These diagrams are similar to those for the oﬀ-diagonal elements of the ﬁrstorder level-shift operator. (8.e. for an incomplete model space it is necessary to retain those contributions to the corresponding unlinked . diagrams are considered unlinked if they have at least one disconnected part in which all open lines are valence lines. it should be remembered that the particle or hole classiﬁcation and the modiﬁed one-electron interaction are speciﬁc to each column of the wave operator matrix. 6. being determined by the corresponding ket state as the Fermi vacuum. i. 12–14 generate one diagram each. Note that no resolvent line is present above the top vertex for any level-shift operator diagram.96).2–8. Skeletons 1. In the case of a complete model space this last restriction means that at least one open-line label in each term in these sums must not take on a valence state value. 3–6 represent single replacements. The deﬁnition of linked and unlinked diagrams is the same in the HK formalism as in the QDPT formalism for the complete model spaces described in Sections 8. 5. 8. the number of replacements cannot exceed the number of valence hole states or the number of valence particle states. (8. All the disconnected skeletons in Fig. The summations over intermediate states must exclude model states. 9. εab ij (8. and they are excluded when a complete model space is used because they involve intermediate model states. 2. 8. 11 generate two diagrams each and skeleton 7 generates three diagrams. (However. However. 8.95).232 Open-shell and quasidegenerate perturbation theory a modiﬁed one-electron interaction: (a) (i) ⊗ (i) (j ) = ai a|g ˆ|i {a ˆ†ˆ i}. 11–13 represent triple replacements and 14 represents quadruple replacements.4 Second-order diagrams Skeletons for the second-order level-shift operator are shown in Fig. 8.9 (skeletons 10 and 12–14) are unlinked. 7–10 represent double replacements. but with single arrows instead of double arrows and with an implied resolvent line above the vertex.6. skeletons 3. 10. 2 are for diagonal elements. Again. 4. and that model states must be excluded from the sum over ﬁnal states in wave-operator diagrams.

In a complete model space this exclusion implies that a is restricted to virtual states.9. 8. we shall ﬁnd that ⊗ ⊗ 1 2 ⊗ ⊗ 9 ⊗ 10 11 3 ⊗ ⊗ 4 5 ⊗ 12 6 ⊗ ⊗ 7 ⊗ 8 ⊗ 13 14 Fig.9: y x u v (a) ⊗= a y |g ˆ|a xa uv {x ˆ† u ˆy ˆ† v ˆ} .) . εxa uv (8. in which V ˆN of (8. The disconnected diagrams are not unlinked and are included even in the complete model space case. Skeletons for the second-order level-shift operator in the Hose–Kaldor formalism. Fig. 8. ˆ represents The second-order wave operator is given by (8.7 Incomplete model space 233 diagrams in which the intermediate state is not in the model space but the ﬁnal state is in the model space.101) The ﬁnal state |Φxy ˆ† u ˆy ˆ† v ˆ} on the initial uv . 5. In this case the ﬁnal states.42). They are seen to be similar to the skeletons for the oﬀ-diagonal part of the second-order level-shift operator.4. but in an incomplete model space a may take on a valence index value provided that the resultant intermediate state |Φxa uv is not part of the model space. Figs.9. while the summation over a is restricted to exclude intermediate model states. must exclude model states. (A short bar across a line indicates limitation to active orbitals. represented by the action of {x state. must be a model state.10. 8.3 and 5.49). and are shown in skeleton form in Fig. 8. except that no bars are present on any of the lines and a resolvent line is implied above the top vertex. we take one of the diagrams generated from skeleton 8 of Fig. The diagrams reprethe modiﬁed (normal-ordered) operator V senting the principal term of this operator are the same in the HK formalism as in nondegenerate MBPT. However.8. as well as all intermediate states. As an example for the interpretation of the second-order level-shift diagrams.

⊗ 1r ⊗ ⊗ 8r 9r 2r ⊗ 3r ⊗ ⊗ 4r ⊗ 10r ⊗ 11r 5r ⊗ ⊗ 6r 7r 12r Fig. The other contributions remain and are described by folded diagrams.234 Open-shell and quasidegenerate perturbation theory ⊗ ⊗ 1 ⊗ ⊗ 8 9 2 3 4 ⊗ ⊗ 5 ⊗ 10 ⊗ 11 ⊗ ⊗ 6 7 12 Fig. 8. with the lower branch below the lowest vertex. . 8. The skeletons for these folded diagrams are shown in the folded-resolvent-line notation in Fig. Skeletons for the renormalization term of the second-order wave operator in the Hose–Kaldor formalism. Skeletons for the principal term of the second-order wave operator in the Hose–Kaldor formalism. 8.10. certain contributions to the disconnected diagrams are canceled by unlinked renormalization diagrams. An additional unfolded resolvent line at the top of each diagram is implied.42)) vanish because for the normal-ordered perturbation operator we have V00 = 0.11. this line can also be drawn in folded form. An additional (unfolded) resolvent line at the top of each of these diagrams is implied but not shown. In the renormalization term those contributions in which the intermediate state is the same as the initial state (|β = |α = |0 in (8.11.

8. 8.) It is seen that the whole of diagram 8r cancels with those terms of diagram 8 in which b and j take on valence values (i. the cancellation includes only those terms of diagram 8 for which Φb j is in the model space. Unlike the disconnected diagrams among the principal term skeletons seen in Fig. Second. but with an incomplete model space those terms of this diagram in which the intermediate state x |Φxy uv is in the model space. 8. ab a εij εi (8. those terms of diagram 8r for which Φa i is in the model space. In the case of an incomplete model space there are two complications.10.10 and 8r of Fig.11 are unlinked and cancel with certain contributions to the disconnected principal-term diagrams. 8. b = y and j = u).103) a|g ˆ|i y |g ˆ|u {a ˆ†ˆ ˆ} iy ˆ† u a εay ε iu i (8. are not canceled because they do not have corresponding contributions in diagram 8. but Φay iu is not.e. (i) =− ayiu (j ) = abij b|g ˆ|j a|g ˆ|i {a ˆ†ˆ iˆ b† ˆ j} . contribute to the second-order wave operator. Therefore the diagram represented by skeleton 2r.7 Incomplete model space 235 The intermediate state in these renormalization diagrams must be a model state. We can illustrate this cancellation by considering diagrams 8 of Fig. 8.11: 8 (a) (b) (i) ⊗ ⊗ 8r (y ) (a) (u) ⊗ ⊗ (the minus sign for diagram 8r is due to the fold. the disconnected folded diagrams in Fig. leaving behind terms for which j and b take on valence label values but Φb j is not in the model space. v |g ˆ|y xy uv {x ˆ† u ˆ} .104) . The canceled terms are just those that cause diagram 8 to become unlinked. while the ﬁnal state |Φu is not. xy εx ε uv u (x) (u) ⊗ (v ) = (y ) xyuv (8. First. because Φy u in diagram 8r must be in the model space.102) cannot contribute when the model space is complete. while the ﬁnal state cannot be a model state.

17. There are no unlinked diagonal-element diagrams but.12 and 8. no such unlinked diagrams remain for a complete model space but for an incomplete model space the only contributions of unlinked folded diagrams that remain are those for which both disconnected parts. organized by . Figs. 8. unlike nondegenerate MBPT. For the unlinked skeletons. 5.6). classes (200) and (400) are omitted because they involve vacuum gaps and classes (020) and (040) are used instead as representative classes. taken separately. The linked and unlinked Hugenholtz skeletons for the representative classes are shown in Figs.236 Open-shell and quasidegenerate perturbation theory In a similar way. The notation (klm) will be used to describe a class of skeletons with k. while the entire ﬁnal state is not in the model space.14.15–8. They are repeated in skeleton form in Fig. second and third vertex. Again. there are diagonal renormalization contributions (as there are in the complete-model-space formalism described in Section 8.6. respectively. For the folded diagrams. The diﬀerent classes in a set all have the same number of skeletons and diagrams and can be obtained from each other by permutation of the three vertices. Since the lower disconnected part of the disconnected principal-term diagrams must not be a model state (because it determines the intermediate state). taken separately.1 and 5. in which the upper disconnected part. represent model states. The numbers of ASG skeletons and diagrams in one representative class in each set and the total number of skeletons and diagrams are listed in Table 8.7. respectively (counting downwards from the top vertex). represents a model state. in order to facilitate later discussion of the renormalization terms that can be derived from them.13.2. The linked principal-term ASG skeletons for the oﬀ-diagonal elements of the third-order level-shift operator are shown in Figs. we use a scheme similar to that used in subsection 8.2.5 Third-order level-shift diagrams In order to identify all the distinct skeletons for the principal term of the third-order level-shift operator. The various classes of linked skeletons are organized in sets. l and m open lines connected to the ﬁrst. 11r and 12r cancel those terms of diagrams 11. 8. a characteristic of the QDPT case is that intermediate model states are excluded from the summations for these diagrams. the only surviving terms are those in which neither disconnected part of a disconnected principal-term diagram. taken separately. diagrams 10r. respectively. 8. 10 and 12. 8. separately for linked and unlinked skeletons. The ASG diagrams for the principal term of the diagonal elements belong to class (000) and are the same as those for nondegenerate MBPT. represents a ﬁnal model state.6 and divide the skeletons into classes and sets.

numbers of distinct ASG skeletons and diagrams in each class and set and total number of principal-term skeletons and diagrams for the third-order level-shift operator (vacuum-gap skeletons are omitted) Number of replacements Representative class No. Linked 0 1 2 (000) (200) (110) (310) (220) (211) (330) (321) (222) (332) 1 3 3 6 3 3 3 6 1 3 8 4 7 3 2 4 1 2 1 1 34 Unlinked 1 2 (020) (040) (220) (211) (420) (411) (321) (222) (422) (431) (332) (442) (433) (444) 2 2 3 3 6 3 6 1 3 6 3 3 3 1 2 2 2 2 1 2 1 4 2 1 1 1 1 1 23 2 2 2 4 1 4 2 10 4 2 2 1 2 1 39 4 4 6 6 6 6 6 4 6 6 3 3 3 1 64 4 4 6 12 6 12 12 10 12 12 6 3 6 1 106 14 6 18 6 7 18 2 8 8 4 91 8 12 21 18 6 12 3 12 1 3 96 14 18 54 36 21 54 6 48 8 12 271 Diagrams Number in set Skelets.8. Representative classes.7 Incomplete model space 237 Table 8. Diagrams 3 4 Total linked 3 4 5 6 Total unlinked .2. of classes in set Number in class Skelets.

the principal-term ASG skeletons are shown for all classes. similarly organized. and for convenience in our later consideration of renormalization skeletons.12. in Figs.12 as well as for the corresponding skeletons of the other two classes in that set. The unlinked principal-term skeletons are shown. (121) and (112). not just the representative classes.238 Open-shell and quasidegenerate perturbation theory (200) (000) ⊗ ⊗ ⊗ (110) ⊗ ⊗ ⊗ (310) ⊗ ⊗ ⊗ ⊗ (330) ⊗ ⊗ (321) ⊗ (222) (332) ⊗ ⊗ ⊗ (220) ⊗ ⊗ (211) ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ Fig. 8. 8. 8. This (211) Hugenholtz . Linked Hugenholtz skeletons for representative classes of the principal term of the third-order level-shift operator in the HK formalism.20. In order to be able to identify all non-contributing diagrams. In the conversion of Hugenholtz skeletons to ASG skeletons it is important to realize that some Hugenholtz skeletons may require more than one ASG skeleton to describe the full set of diagrams that can be generated from those skeletons. In the present case this situation occurs for the ﬁrst skeleton of the (211) class in Fig. the number of replacements for the ﬁnal state relative to the initial (vacuum) state.18–8.

skeleton generates six diagrams: . Unlinked Hugenholtz skeletons for representative classes of the principal term of the third-order level-shift operator in the HK formalism. 8. 8.8.7 Incomplete model space (020) ⊗ ⊗ ⊗ ⊗ (420) ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ (422) ⊗ ⊗ ⊗ (332) (442) ⊗ ⊗ (433) ⊗ ⊗ (444) ⊗ ⊗ ⊗ (431) (411) ⊗ ⊗ (321) (040) ⊗ (220) ⊗ ⊗ ⊗ (222) ⊗ (211) ⊗ 239 ⊗ ⊗ Fig.14.13. ⊗ ⊗ ⊗ 1 2 3 4 5 6 ⊗ ⊗ 7 ⊗ ⊗ ⊗ ⊗ 8 ⊗ ⊗ ⊗ Fig. Diagonal linked ASG skeletons for the principal term of the third-order level-shift operator in the Hose–Kaldor formalism.

Two inequivalent ASG skeletons are needed to represent the same six diagrams: .240 (200) Open-shell and quasidegenerate perturbation theory (020) ⊗ ⊗ 9 10 ⊗ ⊗ 15 (110) ⊗ ⊗ 21 (101) ⊗ ⊗ 28 (011) ⊗ ⊗ ⊗ 35 36 37 38 39 ⊗ ⊗ ⊗ ⊗ 40 ⊗ ⊗ ⊗ 41 29 30 ⊗ 31 ⊗ 32 ⊗ ⊗ ⊗ 33 ⊗ ⊗ ⊗ 34 22 23 ⊗ 24 ⊗ 25 ⊗ ⊗ ⊗ 26 ⊗ ⊗ ⊗ 27 ⊗ 16 ⊗ 11 (002) ⊗ ⊗ 17 18 19 ⊗ ⊗ 20 ⊗ ⊗ 12 13 14 Fig. 8.15. Single-replacement linked ASG skeletons for the principal term of the third-order level-shift operator in the Hose–Kaldor formalism.

8.7 Incomplete model space (310) ⊗ ⊗ 44 (301) ⊗ 45 46 ⊗ ⊗ 53 ⊗ ⊗ 59 (211) ⊗ ⊗ 66 ⊗ ⊗ 71 72 ⊗ 73 74 ⊗ ⊗ 75 67 68 ⊗ ⊗ 60 54 (220) ⊗ 55 ⊗ ⊗ 47 (130) 241 42 ⊗ ⊗ 50 ⊗ ⊗ 56 (202) ⊗ 43 (103) 48 (031) ⊗ 49 ⊗ ⊗ 51 (013) 52 57 (022) ⊗ 58 ⊗ 61 62 63 64 (121) 65 ⊗ 69 ⊗ 70 (112) ⊗ 79 ⊗ 76 77 78 80 Fig. 8. Double-replacement linked ASG skeletons for the principal term of the third-order level-shift operator in the Hose–Kaldor formalism. .16.

8. 62. 8. even when the model space is incomplete. 102. while ﬁve ASG skeletons are shown for each class in this set in Fig. Most skeletons in Figs. 103. so that none of the unlinked principal-term skeletons 100–107 remains. 104. two or four diagrams each. which generates eight diagrams. which generate three diagrams each. The exceptions are skeletons 1.14–8. Triple. 146. 144. 60. 8. 106 and 107 of Fig. 142.2. see Table 8.and quadruple-replacement linked ASG skeletons for the principal term of the third-order level-shift operator in the Hose–Kaldor formalism.20 generate one. These skeletons are included in the ﬁgure in order to facilitate the consideration of renormalization skeletons that can be derived from them. and skeleton 96. 8. 105 always cancel fully with renormalization skeletons.17.18 can never contribute to the principal term of the third-order levelshift operator. Thus there are only four Hugenholtz (211) skeletons in Fig. The unlinked diagrams represented by skeletons 102. .12. 138–140. It will be shown later that skeletons 100. because their lower intermediate state is the same as their ﬁnal state and is thus a model state. 8.242 (330) Open-shell and quasidegenerate perturbation theory (303) ⊗ 82 83 (231) ⊗ 86 87 (132) ⊗ 91 (222) 92 (332) ⊗ 88 (033) ⊗ ⊗ 85 (213) (321) ⊗ 81 (312) 84 ⊗ 89 (123) 90 ⊗ 93 (323) 94 (233) 95 96 97 98 99 Fig.16.

7 Incomplete model space (020) ⊗ ⊗ 100 ⊗ ⊗ 101 (004) ⊗ ⊗ 105 (202) ⊗ ⊗ 110 ⊗ ⊗ 111 (121) ⊗ ⊗ ⊗ 115 ⊗ 116 ⊗ ⊗ ⊗ 117 106 (022) ⊗ ⊗ 112 ⊗ ⊗ 113 (112) ⊗ ⊗ ⊗ 118 ⊗ 119 107 108 ⊗ ⊗ ⊗ 102 (002) ⊗ ⊗ ⊗ 103 (220) ⊗ ⊗ ⊗ ⊗ 109 (211) ⊗ (040) 243 104 114 Fig.3. When the model space is complete. any unlinked third-order skeleton that has no line directly connecting the top and bottom vertices to each other necessarily involves a model state either as the upper or lower intermediate state and cannot contribute to the principal term.) As examples we consider the diagrams derived from skeletons 108 and 111: ⊗ x u y (i) v = ai x|g ˆ|u iy av a|g ˆ|i {x ˆ† y ˆ† v ˆu ˆ} y a εv εi (8.8.18. 8.and double-replacement unlinked ASG skeletons for the principal term of the third-order level-shift operator in the Hose–Kaldor formalism.105) (a) ⊗ . Single. (A classiﬁcation of skeletons according to their contribution status is given in Table 8.

106) . Triple-replacement unlinked ASG skeletons for the principal term of the third-order level-shift operator in the Hose–Kaldor formalism. 8. ay y εiv εv (8. and x (a) (i) ⊗ ⊗ u v y = ai ix au a|g ˆ|i y |g ˆ|v {x ˆ† y ˆ† v ˆu ˆ} .19.244 (420) Open-shell and quasidegenerate perturbation theory (402) ⊗ ⊗ 120 (024) ⊗ ⊗ ⊗ 127 121 (411) ⊗ 122 123 (141) ⊗ ⊗ 129 (312) ⊗ ⊗ 133 (123) ⊗ ⊗ 136 ⊗ ⊗ 137 124 (240) (204) ⊗ (042) ⊗ 125 (114) 126 128 (321) ⊗ ⊗ 130 (231) ⊗ ⊗ ⊗ 134 (222) ⊗ ⊗ 138 139 135 (213) ⊗ 131 (132) ⊗ ⊗ 132 ⊗ ⊗ ⊗ 140 ⊗ 141 Fig.

and sextuple-replacement unlinked ASG skeletons for the principal term of the third-order level-shift operator in the Hose– Kaldor formalism. 8. quintuple. Quadruple-.7 Incomplete model space (422) ⊗ 142 (224) ⊗ ⊗ 146 (341) (314) ⊗ ⊗ 150 (332) 151 (323) ⊗ ⊗ 154 (424) ⊗ 158 (343) 159 (334) 160 (444) 155 (244) ⊗ 156 ⊗ 157 (433) 152 (233) ⊗ 153 (442) 147 (143) ⊗ ⊗ 148 149 (134) ⊗ ⊗ 143 144 (431) ⊗ 145 (413) (242) 245 ⊗ ⊗ 161 162 163 Fig. .20.8.

(161. 155. These two diagrams can contribute only when the model space is incomplete and then only if the intermediate state Φy v is not in the model space. 131r). 103r). In both xy cases the ﬁnal state Φuv must be in the model space. 155r–161r. 128. 162. 118r). (112. 133r). 152. 133–136. 87r. 44r. 81r. All the other unlinked skeletons cancel with unlinked renormalization skeletons. 10r. 140–143. (155. 23r. 46r. (152. 93r. 140r). 112r. 108r–110r. 134r.3). 55r. 14r. 107r). (128. 135r. 137. . (139. Skeletons pairs that cancel with each other but may have residual contributions when the model space is incomplete (asterisks indicate skeletons that have no residual contributions even in an incomplete model space): (100*. 106r). 163r. 156–160. Contribution status of third-order level-shift skeletons Unlinked principal-term skeletons that never contribute because their lower intermediate state is always a model state: 102. 138. 32r. 100r. (129. Renormalization skeletons (both linked and unlinked) that do not contribute when the model space is complete but may have residual contributions when the model space is incomplete (skeletons that have no line connecting the top two vertices to each other): 5r. 125–127. 123. 39r. 147r–150r. 122. 50r. 42r. 162r). 153r). 105r. and terms in which ya the intermediate state Φa i (in the ﬁrst diagram) or Φvi (in the second) is a model state are excluded from the sum. 154. Unlinked principal-term skeletons that may have residual contributions when the model space is incomplete (skeletons that have no line connecting the top and bottom vertices to each other): 108. 27r. 114r–117r. 132. With an incomplete model space the cancellations are incomplete. 144. 104r. 73r. (137. (109. 111r). (122. 154r). 68r. 146r). 163. 119r). (101*. as will be shown in the following discussion of the renormalization terms. 88r.246 Open-shell and quasidegenerate perturbation theory Table 8. 102r). 22r. 132r. 111. 152r. 151r). 150. 123r). 145–149. 116. 112. 117.3. 136r). 151. (132. 101r. (105*. 141r–145r. 130r). 130. (117. 48r. (104*. 161. 97r. 113–115. 124. 131. and no unlinked contributions of any kind remain when the model space is complete (see Table 8. 103. 107. 84r. 113r). 80r. 153. 118–121. 125r). (150. 106. 126r–129r. 60r. 139. (144. (116. 120r–122r. 129. 124r. (124. 110. 137r–139r. Unlinked principal-term skeletons that cancel with renormalization skeletons but may have residual contributions when the model space is incomplete: 109.

and then only from those terms in the sum for yz which Φyz vw is in the model space while Φuw is not. . including the diagonal skeleton 5r (see Fig.7 Incomplete model space 247 Each principal-term third-order skeleton. three hole lines and two loops (one of them a quasiloop). Again. and then only for those terms in the sum for which Φxz uw is in the model space.107) yz v 2 yzvw εuw εu ⊗ x u ⊗ v y = − y |g ˆ|v zw w|g ˆ|z xz uw {x ˆ† u ˆy ˆ† v ˆ} . (In all cases. the ﬁnal state Φxy uv must be in the model space. Those renormalization skeletons. has a corresponding renormalization skeleton. whether linked or unlinked. some because they connect to the bottom vertex and the others because they are part of the ﬁnal-state designation.) The second diagram has three hole lines. (8.8. The resulting negative sign is canceled by the negative sign due to the fold.108) (z ) (w) The ﬁrst diagram has one pair of equivalent lines. 8. three loops and no equivalent lines. which is easily obtained from it by restricting the lower intermediate state to be a model state and replacing the lower resolvent line by a folded resolvent line surrounding the middle vertex. w εx u εz (8. both linked and unlinked. these will be referred to using the numerical labels of the corresponding principalterm skeletons with an added ‘r’.14). It is therefore unnecessary to show all the renormalization skeletons. that have no line directly connecting the top two vertices to each other cannot contribute when the model space is complete (Table 8. In these cases all lines are valence lines. There are 24 linked and 44 unlinked renormalization skeletons of this type. This diagram can only contribute in an incomplete model space.3). and the diagram can only contribute when Φx u is not in the model space. Similar considerations apply to the other skeletons of this type. As examples we shall consider one of the two diagrams generated by skeleton 23r and the diagram generated by skeleton 110r: x u ⊗ (v ) (y ) = (z ) (w) xw yz v |g ˆ|u yz vw 1 {x ˆ† u ˆ} . making the upper intermediate state a model state. the ﬁnal state Φx u must be in the model space.

133r): 109 v x u y ⊗ (i) ⊗ = (a) ai yi va x|g ˆ|u a|g ˆ|i {x ˆ† y ˆ† v ˆu ˆ} . because such terms are excluded from diagram 100. Φab ij cannot be in the model space when Φaby iju is not. As examples we shall consider diagrams derived from the skeleton pairs (109. This cancellation will be illustrated by skeletons 100 and 102r: 100 y (a) (i) (j ) (b) u ⊗ = 1 4 ij ab y |g ˆ|u ab ij abij ab εaby iju εij {y ˆ† u ˆ} . (8.248 Open-shell and quasidegenerate perturbation theory As noted earlier. but may have residual terms surviving the cancellation when the model space is incomplete. 103.109) 102r (a) (i) (j ) (b) y u ⊗ =− 1 4 ij ab ab ij y |g ˆ|u abij ab εaby iju εij {y ˆ† u ˆ} . 106 and 107 can never contribute.110) In both diagrams. but any terms in diagram 102r for which Φab ij is in the model space (while Φaby iju is not) remain. 111r) and (132.111) xa a εui εi . and terms in which Φaby iju is in the model space are excluded. in this case the cancellation of the two diagrams is complete. 104 and 106r and 105 and 107r. All remaining unlinked principal-term and renormalization skeletons cancel in pairs. skeletons 100. Combined with the fact that principal-term skeletons 102. Similar behavior is obtained for the unlinked skeleton pairs 101 and 103r. Φy u must be in the model space (since it is the ﬁnal state). (8. 101. (8. 104. as listed in Table 8. 105 cancel completely with renormalization skeletons. Thus the whole of diagram 100 is always canceled by corresponding terms in diagram 102r. we ﬁnd that none of the unlinked principal-term skeletons 100–107 survive for any model space. When the model space is complete.3.

each factor represents a zero-order energy difference between a model state and an external state. and the sums are restricted to terms in which Φxz iw is not in the model space. (8. 132 x (i) u v ⊗ ⊗ u v x (i) ⊗ ⊗ Φxyz uvw z w = y 1 2 iy uv x|g ˆ|i z |g ˆ|w {x ˆ† y ˆ† z ˆ† w ˆv ˆu ˆ} . terms in the sum in which xz Φx i is a model function (but Φiw is not) are excluded from diagram 132 and thus remain uncanceled in diagram 133r. Similarly. However. x εxz ε iw i (8. as in (8.108).112) In both diagrams must be in the model space. involve diﬀerences in valence-orbital energies only and are thus likely to be relatively small.7 Incomplete model space 249 111r x u v ⊗ ⊗ y (i) (a) = − ai yi va a|g ˆ|i x|g ˆ|u {x ˆ† y ˆ† v ˆu ˆ} . in those terms that contribute to the sum. x εxz ε iw i (8. terms in the sum in which Φa i is is not) are excluded from diagram 109 and thus a model function (but Φxa ui remain uncanceled in diagram 111r. xa a εui εi (8. No uncanceled terms of these types remain when the model space is complete. Nevertheless.114) y z w =− 1 2 iy uv z |g ˆ|w x|g ˆ|i {x ˆ† y ˆ† z ˆ† w ˆv ˆu ˆ} . However. cases in x which Φu is not a model state are excluded from diagram 111r and thus remain uncanceled in diagram 109. As long as the choice . and the sums are restricted to terms in which Φxa ui is not in the model space. The denominator factors in some residual terms for incomplete model spaces.113) Φxy uv i 133r i In both these diagrams must be in the model space. Similarly.107). No uncanceled terms of these types remain when the model space is complete.8. cases z in which Φw is not a model state are excluded from diagram 133r and thus remain uncanceled in diagram 132.

As in the case of the complete-model-space approach described in Sections 8.6. and they are rarely practicable beyond the third order of the energy. . Coupled with the summation restrictions. convergence of the perturbation series need not be seriously impaired. the number of diagrams generated in the HK approach is also quite large and increases very rapidly with the order of the calculation.250 Open-shell and quasidegenerate perturbation theory of model space maintains signiﬁcant zero-order energy diﬀerences between all model states on the one hand and all external states on the other. this situation makes QDPT calculations diﬃcult.2–8.

For this purpose. t (9. N }. In this basis we have two orthonormal two-electron functions for each atom. {φ0 (i). (9. We shall use the notation and basis set of Section 1. electrons at a time). The eﬀect of T 251 . The t the basis functions for the other atoms.4 for BWPT and RSPT. the functions on one atom are orthogonal to those on any other atom. . Since the atoms are noninteracting. t ˆi χ(i) = 0 .3) ˆ2 indicates that it is a sum of operators that excite two (the subscript 2 on T ˆ2 on the N -atom functions Φ0 . 2. χ(i). i = 1.7 in the context of CI and in Section 2. ˆi φ0 (j ) = φ0 (j ) . we look again at the model problem of N noninteracting He atoms.1 Coupled-cluster theory for noninteracting He atoms Before beginning the formal development of coupled-cluster (CC) theory and its diagrammatic tools. χ(i)} by ˆi φ0 (i) = τ χ(i) .2) For the N -atom wave function we deﬁne the operator ˆ2 = T i ˆi t (9.1) ˆi operator has no eﬀect on where τ is a constant to be determined.9 Foundations of coupled-cluster theory 9. which we examined in Section 1. we will introduce some essential elements of coupled-cluster theory in a conceptual form. . ˆi for the ith atom We deﬁne a single-atom two-electron excitation operator t in the space spanned by {φ0 (i). Φi . t ˆi χ(j ) = χ(j ) t (i = j ) .7. . where φ0 (i) is the zero-order function for atom i and χ(i) represents a two-electron excitation from φ0 (i). . Φij .

7) is easily found: ˆ2 Φ0 = T i ˆi Φ0 = τ t i Φi . . automatically accounts for the correct relationship between the coeﬃcients τ. Φij (since Φii = 0).34)).5) (compare (1.252 Foundations of coupled-cluster theory etc. . Φij . (Section 1. which is a characteristic of the coupledcluster approach. ˆ2 in this example is the coeﬃcient τ . in the form ˆ − E )eT2 Φ0 = 0 . . We know that the exact solution for N noninteracting He atoms can be written in the form Ψ = Φ0 + i τ Φi + 1 2! ij τ 2 Φij + 1 3! ijk τ 3 Φijk + · · · (9. In the CC approach we write the wave function in this exponential form and derive ˆ2 in this case) equations for the determination of the cluster operators (T and the corresponding energy. Φijk .7) onto Φ0 | and Φi | (the diﬀerent values of i all lead to the same ﬁnal equation): ˆ2 ˆ − E )eT |Φ0 = 0 .8) . Therefore (compare (1. Φi |(H (9.6) The exponential Ansatz Ψ = eT2 Φ0 .4) etc. Φ0 |(H ˆ2 ˆ − E )eT |Φ0 = 0 . τ 2 . ij ˆ2 Φi = T j ˆj Φi = τ t j (j =i) 2 ˆ2 T Φ0 = τ i 3 ˆ2 T Φ0 = τ 2 ij Φij = τ j ˆ2 Φi = τ 2 T ˆ2 Φij = τ 3 T ijk (9. and to obtain The only unknown in T equations for τ and the corresponding energy E we project the Schr¨ odinger equation. of the various excitation levels. ˆ (9. (H ˆ (9.32)). ˆ2 Φ0 + Ψ = Φ0 + T 1 ˆ2 2! T2 Φ0 ˆ + 1 ˆ3 3! T2 Φ0 + · · · = eT2 Φ0 .

13) (9. 2 ∗ (9.15) Since τ is independent of N . (9. in the form eT2 Φ0 . ε = ε0 + ∆ε .10) ˆ |Φ0 Φ0 |He ˆ2 T = ˆ |Φ0 + Φ0 |H ˆT ˆ2 |Φ0 + Φ0 |H i ˆT ˆ22 |Φ0 + · · · Φ0 |H = E0 + τ ˆ |Φ0 Φi |He ˆ2 T ˆ |Φi + 0 + · · · = N ε0 + N τ β . we obtain a pair of simultaneous equations for τ and E : N (ε0 + βτ ) = E . (9. Substituting in (9.16). (9. (1.23) we obtain the following matrix elements: Φ0 |EeT2 |Φ0 Φi |EeT2 |Φ0 ˆ ˆ ˆ2 |Φ0 + · · · = E .9) 2 1 ˆ2 |Φ0 + E ˆ2 |Φjk + · · · = E Φi |Φ0 + E Φi |T Φi |T 2 jk = 0 + Eτ j Φi |Φj + 0 + · · · = Eτ . We can write E = Nε . as we see from the ﬁrst equation. coupled-cluster theory (limited in this case to clusters of double ˆ excitations) gives the exact answer.s.11) Φ0 |H 1 2 = ˆ |Φ0 + Φi |H ˆT ˆ2 |Φ0 + Φi |H ˆ |Φj + 1 τ 2 Φi |H 2 j ˆT ˆ22 |Φ0 + · · · Φi |H ˆ |Φjk + 0 + · · · Φi |H = β∗ + τ ∗ ∗ jk 1 2 2τ j ˆ |Φi + = β + τ Φi |H ˆ |(Φij + Φji ) Φi |H (9. Thus.17) for c and ∆ in the full-CI solution for a single He atom. leaving a quadratic equation for τ that is independent of N : β ∗ + (α − ε0 )τ − βτ 2 = 0 . it is easy to see that the equations obtained here for τ and ∆ are equivalent to the equations (1. thus τ = c and the CC solution is indeed the exact solution.32). for the N -atom problem. containing N as a factor but. = E Φ0 |Φ0 + E Φ0 |T (9.8). 1 2 (9.16) In fact.12) = β + τ [(N − 1)ε0 + α] + τ 2 (N − 1)β.9. for the noninteracting .h.14) The second equation includes two terms.. as required for extensivity. their sum cancels against τ E on the r. β + τ [(N − 1)ε0 + α] + τ (N − 1)β = τ E .1 Coupled-cluster theory for noninteracting He atoms 253 Using (9. (1. ∆ε = βτ .4) and (1. we see from the ﬁrst equation that E is proportional to N . τ N ε0 and τ 2 N β .

5 we introduced the wave operator Ω. the correct wave function will have additional (but relatively smaller) terms. • The amplitude equations are nonlinear. is arguably the most convenient and powerful form of Ω. • Instead of the eigenvalue equation of CI. containing powers and products of the various amplitudes.254 Foundations of coupled-cluster theory He atoms. however. we obtain a set of simultaneous algebraic equations by projecting the Schr¨ odinger equation onto the zeroorder function and onto the functions obtained by the operation of the cluster operators on the zero-order function.19) ˆ is an excitation operator. • The ﬁrst equation (called the energy equation ) provides an expression for the energy in terms of the coeﬃcients or amplitudes appearing in the cluster operators. allowing the amplitudes to be determined by the solution of these remaining simultaneous equations. ˆ Ψ = eT Φ0 . If we bring the atoms closer and allow them to interact. ˆ (9. which converts a zeroorder wave function Φ0 into the exact wave function Ψ: ΩΦ0 = Ψ . One form for Ω is given by a linked-diagram expansion. After the amplitudes have been obtained.2.18) We will see shortly that Ω may also be written quite generally as Ω = eT . (9. The energy cancels in the other equations (the amplitude equations ). Iterative methods are generally required for their solution. Before a formal derivation. in addition to the exponential Ansatz for the wave function and the resulting extensivity. This form. . the energy can be calculated from the energy equation.2 The coupled-cluster wave function 9. which is the coupled-cluster where T choice. 9.17) Ω|Φ0 = |Φ0 + k=1 ˆ )k |Φ0 ˆ0W (R L . ∞ (9. The following characteristic features of the CC approach. let us consider some of the consequences of this choice. should be noted.1 The exponential Ansatz and extensivity In Section 2.

If the zero-order wave function for the system is separable.21) (9. Equation (9. where Φ0 = |0 is the reference function and ˆ=T ˆ1 + T ˆ2 + T ˆ3 + · · · . Ψ(A · · · B ) = eT (A···B ) Φ0 (A · · · B ) = eT (A)+T (B ) Φ0 (A)Φ0 (B ) = eT (A) Φ0 (A)eT (B ) Φ0 (B ) = Ψ(A)Ψ(B ) ˆ ˆ ˆ ˆ ˆ (9. T (9.24) . and if T ˆ(A · · · B ) = T ˆ(A) + T ˆ(B ) . RSPT) does not satisfy the wave-function multiplicative separability condition (9. T then the total wave function is multiplicatively separable.22).22) (operators on the A and B subsystems commute with each other).23) as required for extensivity.e. This separability of the wave function ensures additivity of the energy.25) ˆ (9. but it always gives extensive results for the energy.2 The coupled-cluster wave function 255 The exponential Ansatz for the wave operator immediately manifests the extensivity property.2 The cluster operators The coupled-cluster wave function is written in the form Ψ = eT Φ0 .2.22) and is the preferred deﬁnition of extensivity. systems are noninteracting).20) (overall antisymmetrization of the wave function is unnecessary since the ˆ for the system is additive. Consider a system A · · · B composed of two noninteracting subsystems A and B and expressed in terms of orbitals localized on the two subsystems. ˆ (A · · · B )Ψ(A · · · B ) = [H ˆ (A) + H ˆ (B )]Ψ(A)Ψ(B ) H ˆ (A)Ψ(A)]Ψ(B ) + Ψ(A)[H ˆ (B )Ψ(B )] = [H = [E (A) + E (B )]Ψ(A · · · B ) . (9.23) is a weaker condition than (9.9. Unlike CC. 9. Φ0 (A · · · B ) = Φ0 (A)Φ0 (B ) (9. ﬁnite-order MBPT (i.

. If the exponential wave operator is expanded in a Taylor series.28) ˆ2 = T ˆ3 = T 1 (2!)2 1 (3!)2 tab ˆ †ˆ b† ˆ jˆ i= ij a ijab ˆ= tabc ˆ†ˆ iˆ b† ˆ jc ˆ† k ijk a ijkabc 1 36 ˆ} tabc ˆ†ˆ iˆ b† ˆ jc† k ijk {a ijkabc etc. {a ij..31) so that the (2!)2 = 4 contributions from permutations of the i. are coeﬃcients to be determined.) = (−1)σ(P )+σ(P ) tab.) P P (ij.. (9. ˆ†ˆ iˆ b† ˆ j ···} ij. a ˆ†ˆ we also require that ab ba ba tab ij = −tji = −tij = tji . the number of electrons. ˆN but in actual applications T fold cluster operator T ˆ ˆ usually at the T2 or T3 level. ij... are deﬁned to be antisymmetric in the hole indices and (separately) in the particle indices. all tij. (9..... usually referred to as the amplitudes for the corresponding operators. j and a. this repetition being oﬀset by the ab. and the strings of creation and annihilation operators are automatically in normal-ordered form (as before. The 1/(m!)2 factor accounts for the redundancy created by the unrestricted summations.256 Foundations of coupled-cluster theory ˆ1 ) two-body (T ˆ2 ) etc.. Furthermore. tˆ ˆ (ab..32) ˆ ).30) (9. ˆ operator for an N -electron system. (9. terminates with the N The T ˆ has to be truncated. ij... where tab. σ (P ˆ ) are their pariˆ and P ˆ are permutation operators and σ (P where P ties. .. (9.. as discussed further below.33) . ˆm = T 1 (m!)2 tab. ˆ ˆ2 + ˆ + 1T eT = 1 + T 2 1 ˆ3 3! T + ··· ..26) (9.. since for example iˆ b† ˆ j = −a ˆ† ˆ jˆ b†ˆ i = −ˆ b†ˆ ia ˆ† ˆ j =ˆ b† ˆ ja ˆ†ˆ i. cluster operators are given by The one-body (T ˆ1 = T ia i= ta ˆ†ˆ ia ia i} ta ˆ†ˆ i {a 1 4 tab ˆ†ˆ iˆ b† ˆ j} ij {a ijab (9. producing m-fold excitations. b indices will produce four equal terms. where m ≤ N ..29) comprises m pairs of creation and annihilation operators.. ab... factor 1 4 .. More generally. the braces { } indicate normal ordering). In general. ˆ ˆ (9. since any permutation of the m hole indices or of the m particle indices does not produce a distinct contribution..27) (9..25)..

(9.4). which are a direct consequence of the exponential form of the wave operator. including ROHF.25). This theorem. This property of eT1 is known as the Thouless theorem. states that matrix elements of the Hamiltonian between the Hartree–Fock function and singly excited determinants vanish: ˆ Φa i |H |Φ0 = 0 (HF case).) are of much less importance in general. Contributions to the wave (note that the diﬀerent T ˆ function of the form Tm Φ0 are called connected-cluster contributions while ˆ2 ˆ ˆ those involving products of cluster operators. such as 1 2 T2 Φ0 or T1 T2 Φ0 . CC treatments that include ˆ1 and its products are relatively insensitive to the choices of Φ0 and the T orbitals (within a given basis set). and it is thus equivalent to a transformation ˆ of the orbital basis (see subsection 3.5). These disconnected-cluster terms. ˆ4 Φ0 . The connected T tribution is important particularly in systems with high electron densities. If Φ0 tant connected-cluster contribution to the wave function is T is a Hartree–Fock wave function (canonical or otherwise) for the state of ˆ1 Φ0 is quite small. the most imporˆ2 Φ0 . including molecules containing multiple bonds. This insensitivity is due to the fact ˆ that the eﬀect of the operator eT1 is to transform Φ0 to another Slater determinant (Thouless. is 1 2 T2 Φ0 (though higher powers of T2 can become more ˆ3 Φ0 conimportant as the size of the system increases). T ˆ6 Φ0 etc. T Compared with other correlation methods. as a consequence of the interest.2 The coupled-cluster wave function 257 ˆ is expanded as in (9.9. In other cases. Because of the two-electron nature of the Hamiltonian. at least in the ˆ2 ˆ Hartree–Fock case. 1960).35) ˆ1 Φ0 may be important. T tributions (T though they can be important in special situations.34) + ··· ˆm operators commute).2. Higher-order connected conˆ5 Φ0 . The most important disconnected-cluster contribution. are disconnected-cluster contributions. are responsible for the extensivity of the CC wave function. which is valid for UHF and closed-shell RHF wave functions. . the contribution of T Brillouin theorem (Section 1. we obtain and T ˆ1 Φ0 + T ˆ2 Φ0 + · · · Ψ = Φ0 + T 1 ˆ2 ˆ1 T ˆ2 Φ0 + 1 T ˆ22 Φ0 + · · · + T1 Φ0 + T + 2 1 ˆ3 3! T1 Φ0 2 + 1 ˆ2ˆ 2 T1 T2 Φ0 ˆ ˆ2 +1 2 T1 T2 Φ0 + 1 ˆ3 3! T2 Φ0 + ··· (9.

in ˆ is truncated to which T ˆCCD = T ˆ2 .38) An excellent approximation to the exact wave function is usually provided by the CCSDT model.37) The most common extension of this model is coupled-cluster singles and doubles (CCSD). employing second quantization and the generalized Wick’s theorem.37) provides the simplest coupled-cluster approximation. T (9.3. involving T ˆ ˆ2 Φ0 + 1 T ˆ2 ΨCCD = eT2 Φ0 = Φ0 + T 2 2 Φ0 + 1 ˆ3 3! T2 Φ0 + ··· . Following a section that introduces the diagrammatic representation of the cluster operators and demonstrates the connection between the exponential Ansatz of coupled-cluster theory and the linked-diagram theorem of MBPT. First. we use an algebraic derivation.5. 9.39) ˆ3 . we use conﬁguration-space techniques. We can derive the CCD equations for the T . ˆCCSDT = T T (9. including the Slater–Condon rules for matrix elements of the Hamiltonian between Slater determinants.1.3.3.2.3 The coupled-cluster doubles (CCD) equations We shall derive the equations for the coupled-cluster doubles model in several ways. Next. this model Owing to the high computational cost of including T is often approximated in a number of ways.1 Coupled-cluster doubles equations: conﬁguration-space derivation Equation (9. which adds the triple-excitation clusters: ˆ1 + T ˆ2 + T ˆ3 . (9. in subsection 9. in subsection 9. as commonly employed in the conﬁguration-interaction formalism. 9. the ˆ2 coupled-cluster doubles model. ˆCCSD = T T (9. however.258 Foundations of coupled-cluster theory The simplest CC approach is that of coupled-cluster doubles (CCD).36) The CCD wave function includes all connected and disconnected clusters ˆ2 only. we derive the CCD equations diagrammatically in Section 9. in order to clarify the correspondence between the various approaches and to demonstrate the advantages of the diagrammatic procedures. deﬁned by ˆ1 + T ˆ2 .

r µν µ>ν εi . odinger equation. f µ ˆd |q = εp δpq . = H where ˆ0 = F ˆd = H µ (9.43) ˆ − Eref . ˆ o = 0 and F Note that.3.6. ˆd = F ˆ. p|f p|u ˆ|q = i ˆ= U pi qi .27) and (9. we write out the full Hamiltonian in ﬁrst-quantization form.5 and subsection 3. E0 = 1 .41) ECCD = Φ0 |H since Φ0 |ΨCCD = 1 by the choice of intermediate normalization. in the canonical HF case.3 The coupled-cluster doubles (CCD) equations 259 amplitudes and energy by applying the Slater–Condon rules in a conﬁguration-space formalism.40) Now projecting from the left with the reference function Φ0 | we obtain the energy. p|f F = µ ˆo ˆo . using the analysis in Section 3. ˆ2 = H Eref = E0 + E (1) . (9.9. the energy expression becomes ˆ (1 + T ˆ2 )|Φ0 = Eref + ECCD = Φ0 |H i>j a>b ab ˆ |Φab Φ0 |H ij tij = Eref + i>j a>b ij ab tab ij (9. ˆ |ΨCCD .42) ˆ2 can contribute. F . to give ˆ N ΨCCD = ∆ECCD ΨCCD . we obtain Inserting ΨCCD into the Schr¨ ˆ ΨCCD = ECCD ΨCCD . At this point it is convenient to subtract Eref from both sides of the Schr¨ odinger equation. H (9. µ ˆo |q = (1 − δpq ) p|f ˆ|q . To bring out the correspondence with perturbation ˆN = H where H theory clearly. Using the expansions in (9.37). i E (1) = − 1 2 ij ij ij . We have ˆN = F ˆ−U ˆ +H ˆ 2 − Eref H ˆ0 + F ˆo − U ˆ +H ˆ 2 − Eref . H (9. as is easily seen from the Slater– (no other powers of T Condon rules). f µ u ˆµ .44) ˆd .

From (3. 1 2 T2 . (9.46).50) .43) onto all double excitations. For a given set of indices i.45) or ab 1 ˆ2 ˆ ˆ Φab ij |HN (1 + T2 + 2 T2 )|Φ0 = ∆ECCD tij . in addition to the linear term ˆ2 . The number of resulting equations is exactly the same as the number of amplitudes tab ij to be determined. giving ˆ Φab ij |HN |Φ0 = ab ij .48) (9. Only the two-electron part H the ﬁrst term. which since higher powers of T cannot interact with the double excitations in the bra part of the matrix element. We next apply the Slater–Condon rules to the individual terms in the ˆ N contributes to ˆ 2 of H parentheses in (9. b we have ab T2 ˆ T2 Φab ij |HN e |Φ0 = ∆ECCD Φij |e |Φ0 ˆ ˆ (9.260 Foundations of coupled-cluster theory To calculate the energy we need the amplitudes tab ij .48) we have L0 = −εab ij + 1 2 kl kl kl tab ij .46) ˆ2 This equation contains a quadratic term. (9.49) where εab ij = εi + εj − εa − εb . The second (linear) term is ˆ ˆ Φab ij |HN T2 |Φ0 = k>l c>d cd cd ˆ Φab ij |HN |Φkl tkl (9.47) = ab ab ˆ Φab ij |H0 − Eref |Φij tij + k>l c>d ˆ o ˆ cd cd Φab ij |F − U |Φkl tkl + k>l c>d cd cd ˆ Φab ij |H2 |Φkl tkl = L0 + L1 + L2 . and we can obtain equations for these amplitudes by projecting (9. so that for the ﬁrst term in (9. a. j.154) we get ab ab ab ˆ 0 − E0 )|Φab (H ij = (εa + εb − εi − εj )|Φij = −εij |Φij . (9. but note that there has been no truncation of the exponential operator T ˆ2 produce six-fold and higher excitations.

9. c. in L0 . l. b. a. c>d If we have one coincidence between the hole indices and another between the particle indices. j. of i.51) Next we turn to the two-electron term L2 in (9. except for the additional contributions − kl ab ( jl kl tab ik − il kl tjk ) + cl bc ( bl cl tac ij − al cl tij ) .48). at least two of the k. a. respectively. c. and kl kl tab ij = k>l 1 2 kl kl kl tab ij .48) contributes only if at least three of the indices k.51).50). d are equal to the same number of the indices i. j. b. together with the last term ˆ contributions to L1 in (9. a. we get − kc ac bc bc bk cj tac ik − bk ci tjk − ak cj tik + ak ci tjk (the signs arise from the maximum-coincidence permutations required in the application of the Slater–Condon rules). d indices must equal the same number of i. For cd = ab we get ij kl tab kl . (9. cancel all the u . Careful analysis shows that the above terms also account for most terms arising from the cases in which three or four of the indices k.3 The coupled-cluster doubles (CCD) equations 261 The second (one-electron) term in (9. d are equal to three or four. These additional contributions are expressible in terms of matrix elements of u ˆ and. Application of the Slater–Condon rules to this term gives: L1 = − k ukk tab ij − k o [(fbc c o o ab [(fjk − ujk )tab ik − (fik − uik )tjk ] + o bc − ubc )tac ij − (fac − uac )tij ] . (9. j. arising from the four-coincidences case. arising from the three-coincidences case. l. l. For this term to be nonzero. b. k>l and for kl = ij we get ab cd tcd ij . c.

ˆ2 ) are the All the terms considered so far (i. Nonzero contributions to Q are obtained only when four of the indices k.54). (9.e.h. through terms linear in T same terms that occur in a CID calculation.53) This term introduces quadruple excitations into the wave function but with coeﬃcients that are sums of products of the coeﬃcients of double excitations. and it is obvious that ˆ 2 . Coupled-cluster theory diﬀers from CID primarily because of the quadratic term in (9. d. is equal to ∆ECCD tab ij . There only H are many possibilities for the required four coincidences. l. Q= 1 2 2 Φab ij |HN T2 |Φ0 = 1 2 k>l m>n c>d e>f cdef cd ef ˆ Φab ij |HN |Φklmn tkl tmn .52) vanish. ˆo = 0 and the last four terms in the ﬁrst line of In the canonical HF case f (9.262 Foundations of coupled-cluster theory The total result. e. Consequently.46). (9. j. f in (9.s. n. can contribute. c. for the linear term is ab L = −εab ij tij − k o ab fjk tik + k o ab fik tjk + c o ac fbc tij − c o bc fac tij + k>l ij kl tab kl + c>d ab cd tcd ij (9.52) − kc ac bc bc bk cj tac ik − bk ci tjk − ak cj tik + ak ci tjk . b.42) we see that ∆ECCD = i>j a>b ij ab tab ij (9.55) and. m. cd cd ab ac bd bd ac kl cd (tab ij tkl + tij tkl ) − 2(tij tkl + tij tkl ) (9.54) A very important observation pertains to the right-hand side of the amplitude equation (9. the two-electron part of the Hamiltonian. of (9. since the summation labels are dummy variables. L0 + L1 + L2 = L.46). eliminating the unknown energy from the amplitude . From (9.46). a. this term cancels against the r. the ﬁrst term in Q.53) are equal to i. and it is left as an exercise for the reader to prove that Q= k>l c>d cd cd ab ac bd bd ac − 2(tab ik tjl + tik tjl ) + 4(tik tjl + tik tjl ) .

1977b) and is correct through third order in MBPT. The energy in this approximation corresponds to an inﬁnite sum of double-excitation diagrams (Bartlett and Shavitt. which characterizes the CC methods.5. Adams. Jankowski and Paldus 1979). Because of the absence of the 1 2 T2 term in the CID equations. 9.56) are neˇ ıˇ glected (C´ zek 1966). Rearranging terms and converting to unrestricted summations. As we shall see in the diagrammatic derivation in Section 9. This model has also been called D-MBPT(∞) (Bartlett. CEPA(0) and LCPMET ˇ ıˇ (C´ zek 1966. In this approach.56) Once again we see the emergence of simultaneous algebraic equations (instead of the eigenvalue problem of CI). after cancellation of the unlinked terms the remaining quadratic contributions in (9.3. Since the equations are quadratic they are solved iteratively.9. A CI-like model that corrects for this deﬁciency and restores extensivity is provided by the linearized CCD (LCCD) approach.2 Coupled-cluster doubles equations: algebraic derivation For an algebraic derivation of the coupled-cluster doubles equations we use second quantization and the generalized Wick’s theorem for products of normal-ordered products of operators and write the Hamiltonian in . ∆ECID tab ij is not canceled there and therefore the CI procedure retains contributions arising from unlinked diagrams. the quantity ∆ECCD tab ij corresponds to an unlinked diagram and therefore must cancel ˆ2 in the ﬁnal equations. 1979). (9.3 The coupled-cluster doubles (CCD) equations 263 equations. the CCD amplitude equations can then be written as ab εab ij tij = ab ij + 1 2 cd ab cd tcd ij + tac ik − bk ci o ab fik tjk + k 1 2 kl ij kl tab kl bc − ak cj tbc ik + ak ci tjk o ac fbc tij − o bc fac tij c − kc bk cj tac jk − k o ab fjk tik + c + klcd kl cd 1 cd ab 4 tij tkl ac bd bd ac −1 2 (tij tkl + tij tkl ) ab cd cd ab ac bd bd ac −1 2 (tik tjl + tik tjl ) + (tik tjl + tik tjl ) . Shavitt and Purvis. substituting the amplitudes tab ij obtained in each iteration into the quadratic terms for the next iteration.

61) . Converting to unrestricted summations. is zero. The contribution of the one-electron part of H since there is no way of contracting all the operators in this term without using an internal contraction in one normal product.57) To save work it will be convenient to combine the ﬁrst two terms.59) recalling that the vacuum expectation value of a normal-product operator ˆ N in (9.59) vanishes.40)–(9.3. (9.1.6. so that we have ˆN + W ˆ = ˆN = F H pq fpq {p ˆ† q ˆ} + 1 4 pqrs pq rs {p ˆ† q ˆ† s ˆr ˆ} .60) There are four ways of contracting all the operators between the normal products in the vacuum expectation value in (9. we then have ∆ECCD = 1 16 ijab pqrs pq rs 0|{p ˆ† q ˆ† s ˆr ˆ}{a ˆ †ˆ b† ˆ jˆ i}|0 tab ij .264 Foundations of coupled-cluster theory normal-ordered form according to subsection 3. (9.43).60). equations (9.3: ˆ 0 )N + F ˆo + W ˆ ˆ N = (H H N = p ε p {p ˆ† p ˆ} + p=q fpq {p ˆ† q ˆ} + 1 4 pqrs pq rs {p ˆ† q ˆ† s ˆr ˆ} . as follows: ˆ† q ˆ† s ˆr ˆ}{a ˆ †ˆ ˆ† q ˆ† s ˆr ˆ}{a ˆ †ˆ b† ˆ b† ˆ jˆ i} + {p jˆ i} 0 {p b† ˆ b† ˆ ˆ† q ˆ† s ˆr ˆ}{a ˆ †ˆ ˆ† q ˆ† s ˆr ˆ}{a ˆ †ˆ + {p jˆ i} + {p jˆ i} 0 = δpi δqj δsb δra − δpi δqj δsa δrb − δpj δqi δsb δra + δpj δqi δsa δrb . we get ˆ2 |0 ˆ N (1 + T ˆ2 )|0 = 0|H ˆ NT ∆ECCD = 0|H = i>j a>b 0 pq fpq {p ˆ† q ˆ} + 1 4 pqrs pq rs {p ˆ† q ˆ† s ˆr ˆ} {a ˆ †ˆ b† ˆ jˆ i} 0 tab ij . This results in four equal contributions to the sums. removing one factor 1 4: ∆ECCD = 1 4 ijab ij ab tab ij . (9.58) Proceeding as in subsection 9. using εp = fpp . (9. (9.

The decomposition of L in (9.62) is not the same as in the conﬁgurationspace derivation (9. giving 4 the same result as (9. The one-particle part of L is L1 = 1 4 klcd pq ˆ† ˆ ˆ}|0 tcd .58).47).48) and.48).60) without the 1 ijab sum and without the amplitude factor. The other 12 terms are obtained by interchanging the termination points ˆ† of the contractions connecting to the particle creation operators c ˆ† and d ˆ ˆ and/or to the hole annihilation operators k and l.63) To contract all operators between the diﬀerent normal products in the vacuum expectation value. Next we turn to the amplitude equation (9. we obtain L= k>l c>d cd cd ˆ Φab ij |HN |Φkl tkl = 1 4 klcd ˆ ˆ cd cd Φab ij |FN + W |Φkl tkl (9. In each case we obtain . There are 16 valid ways of accomplishing these contractions.62). The linear term is given by (9.62) = L1 + L2 . we must form three contractions between the ﬁrst and third normal products and one contraction each between the ﬁrst and second and between the second and third. four of which are ˆ† ˆ ˆ} + {ˆ ˆ† ˆ ˆ} ba lk ba lk ˆ}{p ˆ† q ˆ}{c ˆ† d ˆ}{p ˆ† q ˆ}{c ˆ† d i† ˆ j †ˆ i†ˆ j †ˆ 0 {ˆ ˆ† ˆ ˆ} + {ˆ ˆ† ˆ ˆ} 0 i† ˆ j †ˆ i†ˆ j †ˆ + {ˆ ˆ}{p ˆ† q ˆ}{c ˆ† d ˆ}{p ˆ† q ˆ}{c ˆ† d ba lk ba lk = δik δjl δbd δap δcq + δik δjl δac δbp δdq − δik δjq δlp δac δbd − δiq δkp δjl δac δbd . Substituting this result into the sums in (9. in which multiple versions of the same term are generated.46).63) gives fac tcb ij + c d fbd tad ij − l flj tab il − k fki tab kj . not only because of the incorporation of L0 into L1 but also because the cancellations between contributions to diﬀerent components of L in (9. occurs often in the algebraic derivation but is removed automatically in the diagrammatic derivation. j †ˆ fpq 0|{ˆ i†ˆ ˆ}{c ˆ† d ba ˆ}{p ˆ† q lk kl (9. The ﬁrst term is the same as (9.48) are incorporated at the outset by the use of normal-product operators in (9. with the normal-product Hamiltonian given by (9.3 The coupled-cluster doubles (CCD) equations 265 This type of redundancy.9.

64) For the two-particle part of the linear term we have to evaluate L2 = 1 16 pqrs klcd ˆd ˆ† ˆ pq rs 0|{ˆ j †ˆ ˆ}{p ˆ† q ˆ† s ˆr ˆ}{c ˆ† k bˆ i† a l}|0 tcd kl . l. the ﬁrst and third and the second and third normal products.65) To obtain valid contractions in this case we must form two contractions each between the ﬁrst and second. giving rise to equal sums. Combining the equal sums in each case.63). l. pq rs 0|{ˆ i† ˆ ˆ† q ˆ† s ˆr ˆ}{ˆ lk j † }{p kl (9.66) L2b = = 1 8 1 8 pqrs kl ˆ}|0 tab . canceling the factor 1 4 in (9. they can be used to classify the terms into three cases. while the third case can be accomplished in four equivalent ways (depending on which two ket operators ˆ ˆ ˆ are contracted). (b) two pairs of particle-index operators or (c) one pair of each type. pq rs 0|{ˆ ba ˆ}{p ˆ† q ˆ† s ˆr ˆ}{c ˆ† d ij ˆ† ˆ ˆ} 0 tcd . The ﬁrst and second of these cases can be accomplished in two ways ˆ and ˆ j † with ˆ l. d write the three types of terms as L2a = = 1 8 pqrs klcd 1 8 pqrs cd ˆ† k ˆˆ pq rs 0 {ˆ ˆ† s ˆr ˆ}{c ˆ† d ba ˆ}{p ˆ† q l} 0 tcd i†ˆ j †ˆ kl ˆ† }|0 tcd . according to whether we contract (a) two pairs of hole-index operators. L1 = − c ac (fac tbc ij − fbc tij ) − k ab (fjk tab ik − fik tjk ) o ac o bc (fbc tij − fac tij ) + c k o ab o ab (fik tjk − fjk tik ) . b on the other. = (εa + εb − εi − εj )tab ij + (9. (9. or ˆ i† with ˆ l and ˆ j † with (for example.67) . j. The contractions between the ﬁrst and third normal products force identities between two of the ket indices k. c ˆ. Changing the dummy summation indices in some terms and permuting some indices gives the ﬁnal result. d on the one hand and two of the bra indices i. These contractions in turn determine the types of remaining contractions involving the second product. c. contracting ˆ i† with k ˆ k in the ﬁrst term). j †ˆ pq rs 0 {ˆ i†ˆ ˆ}{p ˆ† q ˆ† s ˆr ˆ}{c ˆ† d ba lk kl pqrs klcd (9. a.266 Foundations of coupled-cluster theory contributions that are equal to the above four terms. we can k.

we get four equal terms. where the parity of the number of intersections between the contraction lines determines the sign of each term.69) .68) The signs for all these terms can be determined by the number of interchanges needed to bring the contracted operators out in front of the remaining normal products.9. ab cd . so that L2a = 1 2 cd ab cd tcd ij .3 The coupled-cluster doubles (CCD) equations 267 L2c = 1 4 pqrs klcd ˆ† ˆ ˆ} i†ˆ j †ˆ pq rs 0 {ˆ ˆ}{p ˆ† q ˆ† s ˆr ˆ}{c ˆ† d ba lk ˆ† ˆ ˆ} j †ˆ + {ˆ i† ˆ ˆ}{p ˆ† q ˆ† s ˆr ˆ}{c ˆ† d ba lk ˆ† ˆ ˆ} + {ˆ ˆ† s ˆr ˆ}{c ˆ† d ba ˆ}{p ˆ† q lk i†ˆ j †ˆ ˆ† ˆ ˆ} 0 tcd j †ˆ + {ˆ i† ˆ ˆ† s ˆr ˆ}{c ˆ† d ba ˆ}{p ˆ† q lk kl = 1 4 pqrs kc pq rs ˆ}|0 tac b}{p ˆ† q 0|{ˆ j †ˆ ˆ† s ˆr ˆ}{c ˆ† k ik ˆ}|0 tac − 0|{ˆ i†ˆ ˆ† s ˆr ˆ}{c ˆ† k b}{p ˆ† q jk ˆ}|0 tbc − {ˆ j†a ˆ}{p ˆ† q ˆ† s ˆr ˆ}{c ˆ† k ik ˆ}|0 tbc . Substituting this result into the summations and rearranging the indices in each case to obtain the same order. (9. + 0|{ˆ i† a ˆ}{p ˆ† q ˆ† s ˆr ˆ}{c ˆ† k jk (9. The vacuum expectation value in L2a can be evaluated as ˆ† } + {ˆ ˆ† } 0 {ˆ ˆ}{p ˆ† q ˆ† s ˆr ˆ}{c ˆ† d ˆ}{p ˆ† q ˆ† s ˆr ˆ}{c ˆ† d ba ba ˆ† } + {ˆ ˆ† } 0 + {ˆ ˆ}{p ˆ† q ˆ† s ˆr ˆ}{c ˆ† d ˆ}{p ˆ† q ˆ† s ˆr ˆ}{c ˆ† d ba ba = δbq δap δds δcr − δbq δap δcs δdr − δbp δaq δds δcr + δbp δaq δcs δdr .

(9. the vacuum expectation value for this term becomes ˆ} + {ˆ ˆ} ˆ† s ˆr ˆ}{c ˆ† k ˆ† s ˆr ˆ}{c ˆ† k b}{p ˆ† q b}{p ˆ† q j †ˆ j †ˆ 0 {ˆ ˆ} + {ˆ ˆ} 0 ˆ† q ˆ† s ˆr ˆ}{c ˆ† k ˆ† q ˆ† s ˆr ˆ}{c ˆ† k b}{p b}{p j †ˆ j †ˆ + {ˆ = δbp δcs δjr δkq − δbp δcr δjs δkq − δbq δcs δjr δkp + δbq δcr δjs δkp . in this case bk cj .72) and thus.72) No nonzero contractions are possible between the third and fourth normal products in (9. Taking the ﬁrst term as an example. giving − bk cj tac ik . (9. ˆ N will Turning to the quadratic term.268 Foundations of coupled-cluster theory A similar procedure for the second component of L2 results in L2b = 1 2 kl kl ij tab kl . Applying the same procedure to the other three terms results in L2c = − kc ac bc bc bk cj tac ik − bk ci tjk − ak cj tik + ak ci tjk .52). (9. represent unlinked contributions since the set of contractions involving the ﬁrst normal product is decoupled from the set . to obtain nonzero contributions. we get four equal terms.68) can be fully contracted in four ways. Again substituting this result into the summations and rearranging the indices in each case to obtain the same order. We shall ﬁrst consider the case in which the four operators of the ﬁrst product are contracted with the four operators of the fourth. (9. Thus we need to evaluate Q= 1 8 pqrs k>l m>n c>d e>f ef ˆ† ˆ ˆ}{e ˆ† n pq rs 0|{ˆ i† ˆ ˆ† s ˆr ˆ}{c ˆ† d ˆm ˆ }|0 tcd ba ˆ}{p ˆ† q lk ˆ† f j †ˆ kl tmn . four of the eight operators in the third and fourth normal products have to be contracted with the ﬁrst product. and the similar one in which the four contractions are between the ﬁrst and third normal products. only the two-electron part of H enable the eight creation or annihilation operators of the ket quadrupleexcitation conﬁguration to be fully contracted. and the remaining four with the second product.71) Adding up all the components of L = L1 + L2a + L2b + L2c . This term. each of the four terms in (9. we get the same result as in the conﬁguration-space derivation.70) For the third component of L2 .

n. pq rs 0|{p ˆ† q ˆ† s ˆr ˆ}{c ˆ† d lk kl ij Contracting the remaining operators in the four ways that are possible gives four equal contributions. These two types of contraction produce equal results. and so cancels with the r. and results in 1 2 k>l c>d ab kl cd tcd kl tij .3 The coupled-cluster doubles (CCD) equations 269 involving the second. canceling a factor 1 2 . (9. respectively. The remaining terms in the quadratic contribution fall into four classes. f and the restriction i > j. ˆ i† and ˆ ˆ and ˆ k l.46). The same result is obtained (after renaming the summation indices) for the case in which the four operators of the ﬁrst product are contracted with those in the third product. we see that it is equal to ∆ECCD tab ij . of the amplitude equations. thus canceling the factor 1 2 for the total unlinked contribution. giving Qa = = 1 16 pqrs klcd 1 4 klcd ab kl cd tcd ij tkl .s.h.9. which is later canceled by the four equivalent ways of contracting factor 1 4 the remaining operators. e. Then converting to unrestricted summations adds a . a > b. In class (a) the two hole operators of the ﬁrst product are contracted with j † are contracted with either the third or the fourth product (i. respectively.61). or with m ˆ and n ˆ . Comparing this result with (9. Considering the inequalities in the restricted summations over m. ˆ}{c ˆ† }|0 tcd tab pq rs 0|{p ˆ† q ˆ† s ˆr ˆ}{ˆ lk ˆ† d ij kl (9. while the remaining two operators are contracted with operators in the fourth. Converting to . as we have already seen in the conﬁguration-space derivation. depending on the pattern of contractions of the ﬁrst normal product. respectively. the contractions between the ﬁrst and fourth products can be accomplished in only one way: ef ˆ† ˆ ˆ}{e ˆ† n i†ˆ j †ˆ 0 {ˆ ˆ}{p ˆ† q ˆ† s ˆr ˆ}{c ˆ† d ˆm ˆ } 0 tcd ba lk ˆ† f kl tmn 1 8 pqrs k>l m>n c>d e>f pq rs = 1 8 pqrs k>l c>d ˆ† ˆ ˆ}|0 tcd tab .73) In class (b) one hole and one particle operator of the ﬁrst normal product are contracted with operators in the third product.e. using ordered sums) while the two particle operators are contracted with the fourth or third product.

which introduces an additional factor 16 . for a total of 32 equal terms. we ﬁnd that there are 64 choices for these contractions. or vice versa. There are 16 possibilities in each case: the set of three operators in the ﬁrst product can be contracted with operators in the third or the fourth product. In classes (c) and (d) three operators of the ﬁrst normal product are contracted with operators in the third product and one with an operator in the fourth. In total. d in the second term. ˆ with operators in the fourth giving 16 equal terms. . depending on b ˆˆ b or ˆ j†a ˆˆ b for (c) and ˆ i† ˆ ˆ or ˆ i† ˆ j†a j †ˆ whether the set of three operators is ˆ i† a for (d). necessitating a change of sign to restore them to their original order in the antisymmetric integral.74) klcd where in the ﬁnal expression we have interchanged the summation indices c. In class (c) the set of three operators in the ﬁrst product consists of two particle operators and one hole operator while in class (d) it consists of one particle operator and two hole operators. while the remaining single contraction can be chosen in two ways. and in each case these three contractions can be done in four ways. or vice versa. contracting ˆ i† and a † ˆ ˆ product while j and b are contracted with operators in the third product give 16 more terms equal to the above. The 16 1 obtained by possibilities lead to equivalent results. canceling the factor 16 converting to unrestricted summations.270 Foundations of coupled-cluster theory 1 unrestricted summations. each case can be generated in two distinct ways. the ﬁrst Qc term can be written in the form 1 8 pqrs klcd mnef ef ˆ† ˆ ˆ}{e ˆ† n ba lk ˆ† f pq rs 0|{ˆ ˆ}{p ˆ† q ˆ† s ˆr ˆ}{c ˆ† d ˆm ˆ }|0 tcd i† ˆ j †ˆ kl tmn = −1 8 pqrs klcd ab ˆ† k ˆ}{ˆ pq rs 0|{p ˆ† q ˆ† s ˆr ˆ}{c ˆ† d l}|0 tcd kj tli . klcd (9. Furthermore. Speciﬁcally. there are four ˆ to be contracted with operators in the third product while ways for ˆ i† and a † ˆ ˆ j and b can be contracted with operators in the fourth product in four ways. after renaming the summation indices and performing the remaining contractions we get Qb = = 1 4 pqrs klcd bd bc ad ˆ}{d ˆ† ˆ pq rs 0|{p ˆ† q ˆ† s ˆr ˆ}{c ˆ† k l}|0 (tac ik tjl − tik tjl ) bd bc ad kl cd (tac ik tjl − tik tjl ) = bd bd ac kl cd (tac ik tjl + tik tjl ) . Another b with operators in set of 32 equal terms is obtained by contracting ˆ i† and ˆ † ˆ ˆ are contracted with operators in the fourth the third product while j and a product. As an example.

(9. Similarly. we get the result (9.9. The remaining operators can be contracted in four ways: ˆ† k ˆ}{ˆ ˆ† k ˆ}{ˆ ˆ† q ˆ† s ˆr ˆ}{c ˆ† d ˆ† q ˆ† s ˆr ˆ}{c ˆ† d l } + {p l} 0 {p ˆ† k ˆ}{ˆ ˆ† k ˆ}{ˆ + {p l } + {p l} 0 ˆ† q ˆ† s ˆr ˆ}{c ˆ† d ˆ† q ˆ† s ˆr ˆ}{c ˆ† d = δpl δqk δrd δsc − δpk δql δrd δsc − δpl δqk δrc δsd + δpk δql δrc δsd .1) and algebraic derivation (subsection 9.3 The coupled-cluster doubles (CCD) equations 271 The sign reﬂects the odd number of interchanges needed to move all the contracted operators to the front in pairs (note that the summation index m is changed to l after the contraction). All four terms produce the same value for the sum.75) Similar treatment of case (d) gives Qd = − 1 2 klcd bd bd ac kl cd (tac ij tkl + tij tkl ) .5.4 below we will introduce diagrams for the general cluster operators as part of the formal introduction of the exponential Ansatz and will show their relationship to the diagrams of MBPT.3. Comparisons with the conﬁguration-space derivation (subsection 9. for the second term of Qc we have −1 2 klcd ab kl cd tcd ik tjl and these two terms add up to Qc = − 1 2 klcd cd cd ab kl cd (tab ik tjl + tik tjl ) . In particular.54). and after exchanging and rearranging some of the indices we get −1 2 klcd cd kl cd tab ik tjl . .3. The algebraic derivation just presented is fairly laborious. but it serves to demonstrate many potential simpliﬁcations that are incorporated naturally in the diagrammatic derivation.76) When all the diﬀerent parts of Q are put together. In subsequent developments we will derive all expressions diagrammatically. (9. In Section 9.2) can then be readily made. the CCD equations are rederived diagrammatically in Section 9.

(9. as required.272 Foundations of coupled-cluster theory 9. 18 and 19 above are disconnected but.10).77) ˆN and where V ˆN = F ˆo + W ˆ where the superscripts indicate the order in V N ˆ 0 )N + V ˆN ). 17. Thus r = 1 includes all possible . (9. This relationship demonstrates that the various CC models can be viewed as inﬁnite-order summations of selected classes of MBPT diagrams. We can gather together classes of linked diagrams by the number r of disconnected parts that they contain.4 Exponential Ansatz and the linked-diagram theorem of MBPT In this section we explore the relationship between the CC and MBPT wavefunction expansions and show that the CC form of the wave function can actually be derived from the inﬁnite-order MBPT expansion. is ∞ ΨMBPT = ˆN )n |0 ˆ0V (R n=0 L = Φ0 + Ψ(1) + Ψ(2) + · · · . Only linked diagrams have been included.78) + 2 + + 5 + 9 × 14 × 19 (9. The linked-diagram form of the MBPT wave function. by deﬁnition. ˆ N = (H (noting that H The individual orders of Ψ can be written in terms of the antisymmetrized wave-function diagrams in the form (omitting arrows except where they are needed to avoid ambiguity) Ψ(1) = Ψ(2) = 1 + 6 + 10 + 15 ×+ × 16 ×+ 11 ×+ × × 17 + 7 × + 12 ×+ 18 × + 13 ×+ + × + 3 + 8 ×+ 4 . (6. since they have no closed disconnected parts. they are linked. Diagrams 8.79) × etc.

80) The sum extends over all connected wave function diagrams in all orders. N ˆ= T m=1 ˆm .im {a i1 i2 ..im a1 a2 .82) ˆm |i1 i2 · · · im with antisymmetrized amplitudes a1 a2 · · · am |t a1 a2 .am ta ˆ1†ˆ ˆ2†ˆ ˆm im · · · } . We may further classify the connected diagrams belonging to the operator ˆ by the number m of pairs of external lines they contain..... (9. In particular.. up to N (the T number of electrons).im = †ˆ im · · · ˆ i1 } ˆ1† a ˆ2† · · · a ˆm i2ˆ A {a = (9.... ij {a (9.im a1 a2 .am † 1 a2 ... We represent the Tm operators diagrammatically by ˆ1 = T . ˆ0V (R n=1 (9.81) ˆm is an operator that creates m pairs of hole–particle lines.84) ˆ2 = T . We can represent this class formally by ˆ.4 Exponential Ansatz and MBPT 273 connected wave-function diagrams. i1 a i2 · · · a i1 i2 ...am 1 a2 . T (9.83) ab ba ba ˆ where tab ij = −tji = −tij = tji .am = ta i1 i2 .. ˆ2 = T 1 4 ijab A tab ˆ†ˆ iˆ b† ˆ j} . a connected operator T ∞ ˆ|0 = T ˆN )n |0 C .. (9..9.am −ti2 i1 ..85) .. Here T ˆm = T 1 (m!)2 1 (m!)2 ˆm |i1 i2 · · · im a1 a2 · · · am |t i1 i2 ..im = · · · ..

(Strictly speaking... (9. and . where the subscripts in the last form indicate a restriction to nth-order connected diagrams with m external hole–particle line pairs.im a a ...im = n=1 m ti11i22 .86) ˆ4 = T = .274 Foundations of coupled-cluster theory ˆ3 = T = .am ta i1 i2 .90) . We classify these contributions to each T of the order of perturbation theory in which they occur. (9.88) ˆm diagram represents a sum of perturbation theory connected wave Each T function diagrams.e. A product of these cluster operators is represented simply by putting the ˆm diagrams side by side. (9. ∞ 1 a2 . in terms of the number of interaction lines (vertices) they contain. i.89) C. The order-byˆm operator implies a corresponding expansion for order expansion of the T the amplitudes. The second designation for T emphasizes that all particle–hole pairs are equivalent. T n (9.) Note the use of etc.m ˆN ) |0 ˆ0V = (R . including .. with the understanding that all pairs are equivalent. such as the connected diagrams in (9.. (9. these diagrams represent T ˆ solid horizontal lines to represent the Tm operators.. a fact which is not obvious in the ﬁrst form since the middle pairs appear to be diagrammatically diﬀerent. as in the following example (the corresponding T left–right order is immaterial): ˆ2 T ˆ2 = T ˆ1 = ˆ1 T T . We shall use the ﬁrst designation. instead of the broken ˆ4 ˆ3 and T lines used for interaction vertices. Other variations of these forms have been used.87) ˆm |0 .79) but ˆm in terms extending to all orders.a (n) . ∞ ˆm = T n=1 (n) ˆm T |0 (n) ˆm .78) and (9..

17. including the ﬁrst.1) we can show . 18.95) The ﬁrst class of disconnected diagrams has r = 2 and consists of diagrams with two disconnected parts.94) × ˆ3 |0 = T = × ˆ(2) + T ˆ(3) + · · · )|0 = (T 3 3 + + ··· .4 Exponential Ansatz and MBPT 275 (n) ˆm ˆm diagram with subscript n. (9. × (9. arrows are omitted except as needed to avoid ambiguity): ˆ1 |0 = T = + ˆ = (T 1 + × + (1) ˆ +T 1 (2) ˆ +T 1 (3) + · · · )|0 + × + × + × × × + ··· . Using the factorization theorem (see Section 6. ˆ2 and T ˆ3 in terms of MBPT diagrams are given in ˆ1 . The initial few terms.92) (compare (5.9. as The operator T can be represented by a T in = (1) ˆ(1) .79). 8. Four such diagrams.93) ˆ2 |0 = T = + ˆ(1) + T ˆ(2) + T ˆ(3) + · · · )|0 = (T 2 2 2 + + + + + × + × + · · · . 19.83)). this diagram implies that for T 2 ab ij ab (1) ˆij = t εab ij (9.91) (9.and second-order contributions.15) and (9. are seen in (9. (9. T in the expansion of T the following equations (as before.

+ = = 1 16 1 16 ab ij cd kl ˆd ˆ† ˆ iˆ b† ˆ j }{c ˆ† k {a ˆ†ˆ l} cd εab ε ij kl tij 2 ab (1) cd (1) †ˆˆ† ˆ ˆd ˆ† ˆ ˆ† k tkl {a ˆ ib j }{c l} ˆ(1) = T 2 . ˆ(1) T 2 2 . for m = 2. {a ˆ†ˆ iˆ b† ˆ j }{c ˆ† k cd εabcd ε ijkl kl Bringing these two expressions to a common denominator. r = 2 in ﬁrst order we have = = 1 2 1 2 2 + = 1 2 . and the second can be shown algebraically by inserting summation labels and bringing the resulting expressions to a common denominator. (9. as in the proof of the factorization theorem: (i) (a) (b) (j ) (k ) (c) (d) 1 16 (l) ab ij cd kl ˆd ˆ† ˆ iˆ b† ˆ j }{c ˆ† k {a ˆ†ˆ l} . .97) The ﬁrst equality in (9.96) To illustrate this. T (9. adding them and cd abcd then noting that εab ij + εkl = εijkl gives the desired result. abcd εab ε ij ijkl = (i) (a) (b) (j ) (k ) (c) (d) = (l) 1 16 ab ij cd kl ˆd ˆ† ˆ l} .97) is due to the fact that the left–right order in the diagram is immaterial.276 Foundations of coupled-cluster theory ˆ that the sum of all r = 2 diagrams can be expressed as the square of the T operator: 1 ˆ2 2T = 1 2 m ˆm T 2 = 1 2 m ˆm )2 + (T m>n ˆm T ˆn .

11 = ijklmn abcdef N cd cdef εab ij εkl εklmn + N abef cdef εab ij εijmn εklmn + N abef cdef εef mn εijmn εklmn . hole–hole and particle–hole cases for the second vertex in the left-hand part of the diagram.9.4 Exponential Ansatz and MBPT 277 × × 3 1 2 × × 4 × 5 × × × 6 × 7 × × 8 × 9 × 10 × 13 14 × 11 12 Fig. All third-order disconnected wave-function skeletons are shown in Fig. corresponding to the particle–particle.10. For each arrow orientation the sum of the corresponding diagrams originating in skeletons 9.1. 9. 9.98) . This derivation provides a simple illustration of how the factorization theˆ and of how it is related to the orem is manifested in the amplitudes of T exponential Ansatz of coupled-cluster theory. Third-order disconnected wave function skeletons. 10 and 11 can be written in the form S9. (9. Each of skeletons 1–8 represents a single diagram while skeletons 9–14 represent three diagrams each. with three distinct arrow orientations.1. Next we proceed to consider third-order disconnected contributions.

1. which is a T 2 2 2 ˆ 2 component of 1 2 (T2 ) .11 = = N 1 cd cdef εab ij εkl εklmn + 1 cdef εabef ijmn εklmn 1 εab ij + εef mn N N εab ij 1 cd cdef εab ij εkl εklmn + 1 ef cdef εab ij εmn εklmn = 1 εcdef klmn εcd kl 1 + εef mn = N cd ef εab ij εkl εmn .100) ˆ is a product of creation and annihilation operators. we get S9. 9. which generates a single diagram with just one distinct time order. this diagram can be factored by splitting it into six equal contributions and labeling its lines in six diﬀerent ways: 1 6 ab ij kl cd mn ef ijklmn abcdef abcd abcdef εab ij εijkl εijklmn S1 = + ab ij mn ef kl cd abef abcdef εab ij εijmn εijklmn + + kl cd ab ij mn ef abcd abcdef εcd kl εklij εijklmn + + kl cd mn ef ab ij ef cd abcdef εcd kl εmnkl εijklmn mn ef ab ij kl cd abef abcdef εef mn εijmn εijklmn mn ef kl cd ab ij cdef abcdef εef mn εklmn εijklmn ˆ. Introducing a common denominator. Again. This numerator is diﬀerent for the diﬀerent arrow orientations. Bringing the six terms to a common denominator. R (9.98) but in each case it is the same for the three terms in the sum. the where R numerators are all equal. Using a procedure similar to that for the quadratic term in (9.10.99) Adding up this result for the three arrow orientations gives the product ˆ(2) T ˆ(1) (the T ˆ(1) factor is the same for all three orientations). (9.97). For an r = 3 example we consider skeleton 1 in Fig. this sum can be factored as .278 Foundations of coupled-cluster theory where N is the numerator containing the integrals and operators. (9.

ˆ† k 9. Since the Taylor series expansion of the exponential of the connected part ˆ ˆ2 + 1 T ˆ3 + · · · |0 . represents 3! (T2 ) .102) An important point to note is that while MBPT diagrams carry energy denominators. ˆ (9. This observation leads us to the deduction that the sum of T of all linked wave-function diagrams of all orders containing r disconnected 1 ˆr parts is given by r ! T |0 . All other disconnected terms in ΨMBPT are similarly obtained as products ˆm operators.5 Diagrammatic derivation of the CCD equations The diagrammatic derivation of the coupled-cluster equations makes use of the diagrammatic form of the Hamiltonian (Chapter 4) and of the cluster operators (Section 9. as seen. we can deduce that.101).4). This result can be proved inductively by ﬁrst showing that it is correct for r = 2 and then showing that if it is correct for some value of r it is also correct for r + 1. in CC diagrams at least some denominators are hidden in the amplitudes. of course. ΨMBPT = eT |0 . (9.9. The details are left as an exercise for the reader. for example in (9. in (9. in the inﬁniteˆ + 1T is eT |0 = 1 + T 2 3! order limit.5 Diagrammatic derivation of the CCD equations 279 S1 = 1 6 N εabcdef ijklmn 1 abcd εab ij εijkl + 1 abef εab ij εijmn + 1 abcd εcd kl εklij + = 1 6 N εabcdef ijklmn N εabcdef ijklmn 1 cdef εcd kl εklmn + 1 1 abef εef mn εijmn + 1 cdef εef mn εklmn 1 cd εab ij εkl + ef εab ij εmn + 1 cd εef mn εkl 1 = 6 cd ab εef mn + εkl + εij cd ef εab ij εkl εmn = 1 3! N cd ef εab ij εkl εmn .92). Underlying this development is the generalized Wick’s theorem. which simpliﬁes the evaluation of contractions of .101) 1 ˆ (1) 3 The latter. Thus. the ﬁnal expression corresponds to 1 3! tij ab(1) cd(1) ef (1) †ˆˆ† ˆ ˆd ˆ† ˆ ˆ† n l}{e ˆ† m tkl tmn {a ˆ ib j }{c ˆf ˆ} .

this equation is written simply in the form ˆN = F ×+ (0) ×+ (0) ×+ (+1) ×.104) fab {a ˆ †ˆ b} j} fij {ˆ i†ˆ i} ˆ†ˆ fai {a i† a ˆ} fia {ˆ (the algebraic interpretation is given below each diagram). the excess of particle–hole open-line pairs at the top over the number of such pairs at bottom. Using the implicit summation convention for unlabeled lines.105) The numbers below each diagram denote the excitation level represented by this diagram.280 Foundations of coupled-cluster theory normal-ordered operators. Note that here the vertex × ˆ represents the full FN operator.6) is similarly represented as ˆ = 1 W 4 a abcd b c 1 + d 4 k ijkl l j i + ijab a b j i ˆc ab cd {a ˆ †ˆ b† d ˆ} + 1 2 a abci ˆ} j †ˆ ij kl {ˆ i† ˆ lk b 1 + 2 j ijka i†ˆ jˆ b} ai bj {a ˆ†ˆ + 1 2 a abci i k a c i b i c ic ˆ} ab ci {a ˆ †ˆ b†ˆ ˆˆ ia jk {ˆ i† a ˆ† k j} i† c ˆˆ b} ai bc {a ˆ†ˆ .4. and the normal-product form of the Schr¨ odinger equation. (9. as given in (9.58).103) ˆN = F ˆN + W ˆ. We use the normal-ordered Hamiltonian. where H Following (4. (−1) (9.e. as nature of f in (4. H (9.19).57). the one-electron part of the Hamiltonian is represented diagrammatically as ˆN = F ab a b ×+ ij j i ×+ ai i a ×+ ia i a × (9. ˆ N Ψ = ∆E Ψ . unlike in perturbation theory where it repˆo . resents just the oﬀ-diagonal part F N The two-electron operator (subsection 4. and omitting nonessential arrows. Note that in the HF case the last two sums in (9.104) vanish because of the block-diagonal ˆ. i.16).

this operator takes the form ˆ = W (0) + (+1) + (+2) + (−2) + (+1) . (9.59)) is given by ˆ NT ˆ2 |0 . which . the net excitation level ˆ N and T tion of the H ˆ2 represented by (9. ∆ECCD = 0|H (9. we need to combine ˆ2 operator.110) .107).108) The CCD energy (see (9. ij {a ijab (9.108).e. of the energy diagrams should be zero.107) the excitation level is indicated below each diagram. To complete the representation of the CCD equations.9. Put another way. The only such term is ∆ECCD = . i. + (0) + (−1) + (0) + (−1) (9. Therefore have excitation level −2. (9. we should contract it only with those diagrams from the representation of the Hamiltonian. no open lines should remain after the contracˆ2 operators. the Hamiltonian diagrams with the diagrams that represent the T ˆ2 = T = 1 (2!)2 tab ˆ†ˆ iˆ b† ˆ j} . b. (9.5 Diagrammatic derivation of the CCD equations 281 + 1 2 k ijka i j 1 + a 4 i abij aj b 1 + 4 ijab i aj .109) The diagrammatic representation of this equation must begin and end with the Fermi vacuum state. (9.105). With T which has excitation level +2.106) ˆ} ij ka {ˆ i† ˆ ˆk j†a b† ˆ ab ij {a ˆ †ˆ jˆ i} ba ˆ} ij ab {ˆ i† ˆ j †ˆ In terms of unlabeled diagrams and with the usual rules for the assignment of weight factors.

111) Apart from the absence of a denominator. to get a net +2 excitation level the only H . Now we turn to the CCD amplitude equation (9. giving the same result as (9.108). and the positive sign arises from the factor (−1)h+l .282 Foundations of coupled-cluster theory which is just like the expression for the MBPT(2) energy except that we have a solid line instead of a broken line at the bottom of the diagram. +2 diagram ab ˆ ij |HN |0 = i ab j = ab ij .3. ab ij |HN T2 |0 . right–in. and the tab ij amplitude corresponds to the bottom vertex. The ascending and descending lines are assigned arbitrary particle and hole labels. giving ∆ECCD = 1 4 ijab ij ab tab ij .112) ab using the shorthand notation ab ij | ≡ Φij |. the left-hand side represents a net double excitation. in the order left–out. there is a one-to-one correspondence between the diagrams and the contractions of the second-quantized operators as evaluated using the generalized Wick’s theorem (see subsection 9. thus. left–in. which we rewrite in the form ab ˆ ij |HN (1 ab ˆ2 ˆ2 + 1 T +T 2 2 )|0 = ∆ECCD tij . Other than this redundancy and the corresponding numerical factors. with the addition of the interpretation of the solid line according to (9. 1 i a b j . and the only matching HN term is the . The numerical factors are automatically accounted for in the diagrammatic notation.46). as there are two hole lines and two loops. (9. (9. When in doubt about a diagram or its interpretation. ˆ ˆ The ﬁrst term is simply ab ij |HN |0 . it is recommended that it be derived algebraically as a check.2). The factor 4 is due to the fact that the diagram has two sets of equivalent lines.47). The rules for the interpretation of this closed diagram are eﬀectively the same as in MBPT (but without a denominator). right–out. the T2 operator provides a double excitaˆ N terms that can be tion. The matrix element ij ab corresponds to the top vertex i a b j . and the Hamiltonian diagrams used in the contractions for each term must be chosen to match this requirement. (9. avoiding redundant terms. while the second-quantized algebraic derivation usually retains the redundancy.113) ˆ ˆ ˆ In the linear term. respectively. the usual interpretation rules apply to this diagram. Obviously.

When we include such dummy labels in the diagram to show a correspondence with the summation indices in the algebraic expression. b in these diagrams correspond to the “target” indices in the bra part of the matrix element in the amplitude equation. . There are ﬁve added above the T such terms: i aj L1a + i a b (c) (d) L2a j + a i j (k ) (l) L2b b ×+ i (c) a b L1b b + i a j (c) (k ) L2c b j × (k ) where the labels in parentheses indicate summation indices. as in i a b j or i b a j . otherwise they . This rule requires the inclusion of all distinct permutations of the target indices. The interpretation of these open diagrams introduces a new rule to ensure that all distinct terms are included.5 Diagrammatic derivation of the CCD equations 283 ˆ2 vertex should have excitation level zero. i with a and j with b. so their labels are dummy labels and are usually omitted. These target labels are paired in the diagram in correspondence with the vertical pairing in the bra function ab ij |. .) Internal lines are summed over. as before we place them in parentheses.e. The ﬁxed labels i. a. . i. . . Note that diagram L2c can equally well be drawn as or or their mirror images. creating quasiloops. Two external lines are quasi-equivalent if closing the diagram by a two-particle interaction vertex would make them equivalent. .9. (These two diagrams are equivalent. j. and a. i. j. as will become clear from the rules of interpretation. permutations of hole labels and of particle labels of inequivalent external lines. . The pairing can be indicated by imaginary external connections of the correspondingly labeled lines. b.

117) The factor arises from the fact that the diagram has one pair of equivalent internal lines. i. of the algeˆ operator is symmetric. and noting the two hole lines. ˆ (ab) L1a = P c fbc tac ij = c fbc tac ij − c fac tbc ij .284 Foundations of coupled-cluster theory are inequivalent. Returning to the interpretation of diagram L2a . These factors can also be interpreted as a remnant of ˆm operator. If we had permuted the labels a and b then we would have had one quasiloop. (9. (9. so we get ˆ (ij ) fkj tab = − fkj tab + fki tab .64).114) ˆab permutes the labels a and b.e.115) The overall sign is positive. The remaining part of such factors disappears from the amplitude equations because the restriction to distinct diagrams and the ﬁxed external-line labels in the amplitude equations eliminate redundant terms (except for the case of equivalent internal line pairs).82). This factor is included in the in which P algebraic interpretation of the diagram. along with the diagram as given we have to include one in which a is permuted with b. inserting imaginary external connections between i and a and between j and b.2 (note that the f fki = fik ).116) L1b = −P ik ik jk k k k The L1a and L1b terms together correspond to term L1 . in diagram L1a the indices a and b are inequivalent while i and j are quasi-equivalent. because we have two quasiloops (which are treated as ordinary loops for the purpose of the sign rule) and two hole lines. two lines that connect the same two vertices and go in the same direction. in diagram L1b the indices a and b are quasi-equivalent while i and j are inequivalent. (9. Consequently. we get a positive sign. For example. Similarly. the total contribution is then expressed by means of the operator ˆ (ab) = ˆ ˆab . forming two quasiloops. giving an overall minus sign.3. we have two quasiloops and three hole lines. P 1−P (9. a and b are quasi-equivalent and so are i and j . (9. In the case of diagram L2a . hence no permutation factor is needed and L2a = 1 2 1 2 cd ab cd tcd ij . Forming the two external connections. (9. . used there to the 1/(m!)2 factor in the deﬁnition of the T compensate for the unrestricted summations. braic derivation of subsection 9.

This diagram must be attached in all possible two-particle part W ˆ 2. (9. Since T2 corresponds to quadruple excitations while the target state is a double excitation. (9. easily applied procedure that generates all the combined diagrams and generates each distinct diagram only once. (9. However. Thus we need agrams.69). requiring the permutation factor ˆij . since the particle–hole interaction eliminates both the equivalence of internal lines and the quasiequivalence of external lines. we ˆ N. (9. distinct ways to the representation of T 2 Identifying all the distinct ways in which such connections can be accomplished can be diﬃcult.5 Diagrammatic derivation of the CCD equations 285 resulting in a minus sign. so the result would have been the same and no additional distinct term would have been obtained. The fourth diagram is similar. ab ˆ ˆ 2 ˆ2 Next we consider the quadratic term.118) The ﬁfth diagram is the most complicated. L2b and L2c correspond to the terms (9. we can also use an extension of the Hugenholtz-diagram approach of MBPT to provide an unambiguous. must use a −2 de-excitation-level diagram from the representation of H There is only one such diagram.9. as in the MBPT Hugenholtz diˆ vertex is collapsed into a large dot ( ).3. In this extension each two-particle vertex is collapsed into a medium dot ( ). and a systematic way for dealing with this problem will be presented in Section 10. while each T to combine the diagram elements . (9.70) and (9. but this sign would have been compensated by a change of sign of the antisymmetric integral. so the overall sign is negative and the diagram is interpreted as ˆ (ij |ab) L2c = −P kc ak cj tcb ik =− kc ca cb ca ak cj tcb ik − bk cj tik − ak ci tjk + bk ci tjk .71) of the algebraic derivation. the last diagram in the representation of the ˆ .107).119) ˆ (ab|ij ) = P ˆ (ab)P ˆ (ij ) = (ˆ ˆab )(ˆ ˆij ) = ˆ ˆab − P ˆij + P ˆab P P 1−P 1−P 1−P There are two quasiloops and three hole lines. except that it has four hole lines including the two internal equivalent lines giving L2b = 1 2 kl ij kl tab kl .120) The terms L2a . 1 2 ij |HN T2 |0 .

For example. it is easy to see that Qe = ∆ECCD tab ij and therefore that it cancels with the right-hand side of (9. the detailed expansion of a Hugenholtz diagram is not unique but the result is the same for all valid expansions. This combination can be accomplished in ﬁve distinct ways: for T 2 Qa Qb Qc Qd Qe Expanding these into antisymmetric Goldstone-like diagrams. we get j a a b i a b Qc a b j. In fact. Qe . (9. we could equally well have drawn diagram Qb as a i j b or i a j a j b or i b The ﬁfth quadratic-term diagram. .286 Foundations of coupled-cluster theory ˆ and from W ˆ 2 .121) j Q= i b +i j+ Qa + a i j Qd b+ Qb i Qe As always. As such it should cancel against some other unlinked contribution. is both disconnected and unlinked since it has a closed disconnected part. removing the energy from the amplitude equations.112).

they have the same m value) and are connected to the interaction vertex with the same number (zero or more) of particle lines and the same number of hole lines. so for this case 1 2 P (ij |ab) = P (ij ) = ˆ (ab). Diagram Qb introduces our ﬁnal interpretation rule: an extra factor 1 2 is ˆ required for every pair of equivalent Tm vertices in the diagram. we get a positive sign. as will be seen later in other examples. The cancellation of the factor 1 arising from a pair of equivalent P 2 vertices against a permutation of the labels of two open lines connected to these two vertices (going in the same direction) is quite general.3. The equivalence of vertices is most easily judged in the Hugenholtz-like form of the diagram or in terms of the +/− contraction-pattern notation to be introduced in Section 10. ˆ Qb = 1 2 P (ij |ab) klcd db kl cd tac ik tlj db ˆ kl cd tac ik tlj = P (ab) klcd klcd db kl cd tac ik tlj ˆ (ij ) =P (9. Combined with the four hole lines. four hole lines and two loops). the remainder of these coeﬃcients having been accounted for by the restriction to distinct diagrams (see e. (9.122) (two pairs of equivalent internal lines. Two such vertices are equivalent when they are of the same type (i.123) . resulting in two quasiloops.9. Applying the rules of interpretation to the four uncanceled quadratic-term diagrams.96)).5 Diagrammatic derivation of the CCD equations 287 The sign factor for diagram Qa is obtained by pairing i with a and j with b. we obtain Qa = 1 4 klcd ab kl cd tcd ij tkl (9. j i a b . Diagrams Qc and Qd each have two quasiloops plus one internal loop and four hole lines. resulting in a negative sign. The need for this factor 1 2 in diagram Qb can also be understood from the observation that permuting i with j is equivalent ˆ ˆ in this diagram to permuting a with b.e. The factors 1 2 introduced by this rule can be interpreted as remnants of the inverse factorial coeﬃcients in the power series ˆ expansion of eT .g.

76).124) (one pair of equivalent internal lines.2. in agreement with the results obtained in the algebraic derivation. four hole lines and three loops).288 Foundations of coupled-cluster theory (a pair of equivalent vertices. the label D refers to the double-excitation amplitude equations.126) × D1 D2a D2b × D2c D2d D2e D3a D3b D3c D3d Fig. The diagram numbering in this ﬁgure is part of a systematic scheme that will be followed in subsequent CC equations. (9. ˆ Qc = − 1 2 P (ij ) klcd ab kl cd tcd ki tlj (9. The ﬁnal form of the algebraic amplitude equations for CCD is ˆ (ab) ab ij + P c ˆ fbc tac ij − P (ij ) k 1 2 kl fkj tab ik ˆ (ij |ab) +P kc bd tac ik tjl db kl cd tac lk tij klcd +1 2 cd ab cd kl cd klcd tcd ij + kl ij tab kl kb cj tac ik +1 4 ab tcd ij tkl ˆ (ij ) +P klcd kl cd ˆ −1 2 P (ij ) klcd ab 1 ˆ kl cd tdc ik tlj − 2 P (ab) =0 (for all i > j. (9. (9. . 9. The diagrams for all three terms of the CCD amplitude equation (9.73)– (9. a > b). Antisymmetrized Goldstone diagrams for the CCD amplitude equations.125) (one pair of equivalent internal lines.112) (incorporating cancellation of the disconnected diagram against the righthand side) are collected in Fig. four hole lines and four loops). four hole lines and three loops) and ˆ Qd = − 1 2 P (ab) klcd db kl cd tca kl tij . 9.2.

9. in each CC amplitude equation these denominators would be the same in all terms. but in the meantime we will brieﬂy consider the generation of the fourth-order energy diagrams from the third-order wave-function diagrams in Fig. in Section 10. Using fii = εi etc. and must be included in the interpretation of the amplitude iteration diagrams shown in Fig. (εi + εj − εa − εb )tab ij = ab ij + L(t) + Q(tt) .127). εi + εj − εa − εb εab ij (9.3. as in (9. and the with each of the three diagrams containing second-order T (2) ˆ T2 vertex is expanded in turn in terms of compatible second-order MBPT .127) = ij ab ij ab = . The ﬁrst approximation is to neglect L(t) and Q(tt) and choose tij ab (1) (9.5 Diagrammatic derivation of the CCD equations 289 Note that. ˆ are chosen In terms of perturbation theory. 9. As noted previously. and inserting tij into L(t) provides the double-excitation MBPT(4) energy contributions. Inserttij ing tij ab (1) into Q(tt) generates the quadruple-excitation MBPT(4) energy ab (2) contributions. We therefore set up an iterative scheme for the solution of the equations by separating the diagonal parts out of the second and third terms and moving them to the opposite side of the equation (the remaining oﬀ-diagonal parts vanish in the canonical HF case).3 illustrates this process diagrammatically. However. using the canonical Hartree–Fock case for simplicity.9.and two-electron matrix elements are part of the perturbation.128) below. Figure 9. They resurface indirectly in the iterative form of the equations. When this vertex is contracted used in this case is the −2 vertex ˆ2 vertices. the only Hamiltonian vertex that can be .3. The correspondence of the energy contributions with the MBPT energy terms can be seen (compare Chapter 5). Inserting this approximation into L(t) while neglecting Q(tt) provides ab (2) and the MBPT(3) energy as the next-order contributions. (9. in consistency with MBPT diagrammatic notation. symbolically. we then have.3. We obtain the fourth-order energy diagrams on closing each of the seven ˆ(3) by contracting it with a Hamiltonian vertex diagrams contributing to T 2 above it. the diagonal elements of f ˆ here as part of H0 while all other one. A systematic procedure for generating all the relevant diagrams will be presented later. a resolvent line and its corresponding denominator is implied above the top interaction line of each open diagram. so they may be ignored.128) When inserted into the energy formula this gives the usual MBPT(2) energy.

9.290 Foundations of coupled-cluster theory = (1) ∆E (2) = (1) = (2) ∆E (3) = (2) = (3) + (1) ∆E (4) = (3) = + + + + + + (2) + (2) + (1) + (2) + (1) (1) (1) + (1) = + (1) + (1) + = (1) (1) (1) + + + + + + + + + + + + + Fig. When we contract the ˆ(3) we ˆ(1) )2 diagrams contributing to T Hamiltonian vertex with the four (T 2 2 . 9. we obtain in order the ﬁrst three. the next three and the next six contributions to ∆E (4) shown in Fig. wave-function diagrams. Diagrammatic description of an iteration scheme for the solution of the CCD equations and its relationship to MBPT energies through the fourth-order energy (for the canonical Hartree–Fock case).3. all of which are double-excitation fourth-order energy contributions.3.

(9. Comparing with the MBPT results in Section 5. CCD is correct through third order in the energy and ﬁrst order in the wave function. as discussed in Chapter 7.and third-order MBPT energies contain double-excitation contributions only. (Spin summations.5 Diagrammatic derivation of the CCD equations 291 obtain.7. Taking diagram Qb . We can therefore state that in the Hartree–Fock case. can nh .) Normally np that the most expensive linear term arises from diagram L2a . as an example. which contains four particle labels and two hole labels. . In order to estimate the computational cost of solving the CCD equations. all of which are quadruple-excitation contributions. the procedure corresponds to evaluating ad Sil = kc kl cd tac ik . The computational cost of this term 2 n 4 per iteration. roughly speaking. in order. by considering one amplitude at a time. they can be summed in stages (analogously to the quadruple-excitation contributions to the fourthorder MBPT energy.129) 3 n 3 processes. the same cost as for CISD is therefore proportional to nh p and (without the iteration factor) for the third-order MBPT energy. the next one. Although the quadratic terms have more indices. ˆ (ij ) Qb = P ld ad db Sik tlj . the next two and the last two fourth-order MBPT energy diagrams.e. the next two. we see that all double-excitation and quadruple-excitation ∆E (4) diagrams have been accounted for but that the single-excitation and triple-excitation contributions are missing. We refer to CCD as. an n6 process. (9.123). Thus the quadratic terms do not change the overall i. two nh p dependence of the computational cost on the number of spinorbitals.1). let nh be the number of hole states (occupied spinorbitals) and np be the number of particle states. in which the second.9. so eﬀectively reduce these numbers by up to a factor 2. as shown in Section 7.

3) which has |0 as a right eigenfunction and ∆E as the corresponding eigenvalue. but it should not be confused ˆ eﬀ of the Bloch equations. This operator is commonly referred to as the CC eﬀective Hamiltonian.10 Systematic derivation of the coupled-cluster equations 10. often using the symbol H (“H-bar”). (e−T H ˆ ˆ ˆ ˆ (10. ˆ ˆ ˆ T H = e−T H Ne . as used in QDPT with the eﬀective Hamiltonian H (Chapter 8) and MRCC (Chapter 14). we shall introduce an elegant and more convenient procedure than the heuristic approach that we have used to derive the CCD equations. Similarity transformations play an important role in PT and in CC theories. arrives at the cancellation of the energy in all the amplitude equations very easily at the outset and adds the connectedness condition for the CC amplitude equations. (10. The normal-product form of the Schr¨ odinger equation for a general CC wave function can be written in the form ˆ N − ∆E )eT |0 = 0 . for example. as is seen. While somewhat less transparent than the heuristic approach it is more powerful and compact.1) (10.2) We have thus obtained a non-Hermitian similarity-transformed Hamiltonian.1 The connected form of the CC equations Before deriving additional coupled-cluster equations. such as the equations for the coupled-cluster singles and doubles (CCSD) model. (H We operate on this equation from the left with e−T and obtain ˆ N eT − ∆E )|0 = 0 . in the derivation of quasidegenerate 292 .

T ˆ].5) ˆ N has For reasons that will become clear later (related to the fact that H at most two-particle interactions). each consisting of a product of an even number of creation or annihilation operators (these conditions are satisﬁed ˆ). B ˆ ]. T ] (10. Then the generalized Wick’s theorem. ˆ ˆ B ˆ ˆ + 1B ˆ2 − 1 B ˆ3 + · · · A ˆ2 + ˆ 1+B ˆ + 1B = 1−B e−B Ae 2 3! 2 ˆ) ˆ + (A ˆB ˆ −B ˆA ˆ) + 1 (A ˆB ˆ 2 − 2B ˆA ˆB ˆ +B ˆ 2A =A 2 1 ˆ3 3! B + ··· ˆB ˆ −B ˆ 3A ˆ) + · · · ˆA ˆB ˆ 2 + 3B ˆ 2A + − 3B ˆ + [A. ˆ N and T by the components of H applied to their commutator. taking advantage of the generalized Wick’s theorem and its diagrammatic representation. Since A ˆ and B ˆ each creation or annihilation operators in A .4) (10. T ˆ]. [A. T ].6) ˆB ˆ } represents the sum of all where { } indicates a normal product and {A normal products in which there are one or more contractions between the ˆ and those in B ˆ . we obtain ˆ N + [H ˆ N. Hausdorﬀ 1906). =A 2 3! Applying this expansion to H. An important property of such transformations is that they do not change the eigenvalue spectrum of the operator. this series terminates with the four-fold commutator. T ˆ]. Baker 1905. ˆ B ˆ ].10. The commutator expansion of H can be simpliﬁed. At ˆ and this point we shall consider only one aspect of the simpliﬁcation. For a more explicit form of H we use the Baker–Campbell–Hausdorﬀ expansion (Campbell 1897. gives ˆ B ˆ] = A ˆB ˆ −B ˆA ˆ = {A ˆB ˆ } + {A ˆB ˆ } − {B ˆA ˆ} − {B ˆA ˆ} . + 1 [[[[H 4! 1 ˆ ˆ ˆ ˆ 3! [[[HN . T ˆ]. T ˆ] . B ˆ] + · · · .1 The connected form of the CC equations 293 perturbation theory in Section 2. ˆ B ˆ ] + 1 [ [A. ˆ B ˆ ]. T ˆ] + H=H 2 ˆ N. ˆ B ˆ ] + 1 (A ˆB ˆ −B ˆA ˆ)B ˆ −B ˆ (A ˆB ˆ −B ˆA ˆ) =A 2 ˆB ˆ −B ˆA ˆ)B ˆ −B ˆ (A ˆB ˆ −B ˆA ˆ)]B ˆ + 1 [(A 1 ˆ ˆ3 3! (AB 3! ˆ [(A ˆB ˆ −B ˆA ˆ)B ˆ −B ˆ (A ˆB ˆ −B ˆA ˆ)] + · · · −B ˆ + [A. T ˆ] + 1 [[H ˆ N.5. T ]. B ˆ ] + 1 [ [ [A. (10. Let A ˆ B be two normal-product operators.

ˆ B ˆ ] = {A [A. . Since the only nonzero contractions are of the forms a ˆˆ b† = δab and ˆ i† ˆ j = δij . Furthermore. since a particle creation operator. (10. the expansion (10. . T ˆ]. and their cancellation simpliﬁes the equations considerably. ˆ N.5) are those in which H ˆ N eT H = e−T H ˆ ˆ 1 ˆ ˆˆˆ 3! HN T T T 1 ˆ ˆˆˆˆ 4! HN T T T T ˆ NT ˆ + 1H ˆ ˆˆ ˆN + H =H 2 NT T + ˆ ˆ N eT = H C + . . such as a ˆ† .294 Systematic derivation of the coupled-cluster equations contain an even number of creation or annihilation operators. can only produce a nonzero contraction with a particle annihilation operator a ˆ to its left ˆ and since a hole annihilation operator i can only produce a nonzero contraction with a hole creation operator ˆ i† to its left. . the subscript C also indicates this restriction the following T to connected terms. thus lation operators it can be contracted with at most four T accounting for the termination of the Baker–Campbell–Hausdorﬀ expansion with the four-fold commutator. the only surviving terms in ˆ N is the ﬁrst operator on the left.9) The contraction symbols of the form etc. Tn ] = 0. . ˆm operators (and no nonzero contractions can be obtained between diﬀerent T ˆ ˆ ˆ thus [Tm . . ˆ b† . and ˆm T clusters such as T disconnected terms (disconnected diagrams). indicate a sum over all ˆ N is connected by at least one contraction with each of terms in which H ˆ operators. Since each term in H of the T ˆm operators. . which are present in the CC wave function. .7) The canceled uncontracted terms correspond to disconnected diagrams in the diagrammatic representation of the CC equations.e. . . ˆm cluster operators contain only particle creation operators a ˆ† . ˆ j. we have ˆB ˆ } = {B ˆA ˆ} . the diﬀerent Tm operators commute. (10. T ˆ]. It is important to distinguish between disconnected ˆn Φ0 .8) (10. [[H ˆ the only surviving terms involve contractions between HN and one or more ˆ N contains at most four creation or annihiˆm . which cancel in the expansion ˆ ˆ N eT Φ0 C and in the coupled-cluster equations. in the repeated commutators [. As a result. as might be exˆ]. T pected). {A so that all terms without contraction cancel and ˆB ˆ } − {B ˆA ˆ} . of H = H . . . The T and hole annihilation operators ˆ i. i.

. have ˆm cluster operators. which consists solely of linked (or closed. as well as the associated denominators.2 The general form of CC diagrams ˆ Each term in the perturbation theory equations may contain multiple V factors. ˆ ˆ N eT |0 H C = ∆E |0 . All other product Hamiltonian H Hamiltonian-operator factors. ˆ ˆ N eT |0 0|H C = ∆E . With this derivation the energy has automatically disappeared from the amplitude equations. the terms appearing in coupledvertices cluster equations each contain only one operator derived from the normalˆ N . (10.10) This form of the Schr¨ odinger equation will be used as the basis for the derivation of the various sets of CC equations.. The appropriate general form (2. Q (10. 9. and the corresponding diagrams may contain multiple interaction × and .10. connected) diagrams. respectively (Section 2. In contrast.3. (10. Therefore the diagrams used to represent the various terms in coupled-cluster equations never have more than one interaction vertex each.13) ˆ and Q ˆ are the familiar projection operators onto the reference where P space and its complement.49) of the projection operators applies when nonorthonormal functions Φi are employed. that vertex is the top vertex in the diagram.2) we obtain the connected form of that equation. as can be seen from (10. as can be seen in Fig.11) and (10.11) and when it is projected onto a set of excited functions we get the connected amplitude equation. 10. (10. Furthermore.11). we do not have to work out its cancellation explicitly in each case.12) in the form ˆ HP ˆ = ∆E P ˆ.. . (10.12) ij. been absorbed into the T for example.2 The general form of CC diagrams 295 When the connected-terms expansion of H is substituted into the similaritytransformed form of the Schr¨ odinger equation (10. |HN e |0 C = 0 .3). When it is projected onto the zero-order function we get the energy equation.12).. ˆ ˆ T Φab. It is sometimes convenient to write the CC equations (10. P ˆ HP ˆ = 0.

including a parˆ ity factor (−1)σ(P ) . Because all terms in these equations must be connected. permutations are represented by operators of the form P see (9. 4.1. (The target indices are those that occur in the bra part of the amplitude equation. where h is the number of hole lines and l is the number of loops.. forming quasiloops. (Two internal lines are considered equivalent if they connect the same two vertices. ··· 5. ˆ of labels of inequivalent external 9. .) and particle (a. Associate a factor 1 2 with each pair of equivalent internal lines. Sum over all internal line labels. but not those in . Sum over all distinct permutations P particle lines and of inequivalent external hole lines. For the purpose of counting loops. 10. (10. cancel each factor 1 2 arising from rule 7 above with a permutation of the labels of a pair of external lines connected to the equivalent vertices (going in the same direction).. . . With every T ij. In open diagrams with equivalent vertices. Fig. (9. These sums over ˆ (ij · · · |ab · · · ). With any one-particle interaction vertex fout. there can be no more than four T T diagram in the CC equations (demonstrating again the termination of the commutator expansion of H. . 10.) 8. . 3. examples are provided by the two T and in .e. that would be equivalent if they became internal by the addition of an interaction vertex) are not considered inequivalent for this purpose.296 Systematic derivation of the coupled-cluster equations Interpretation rules for coupled-cluster diagrams 1. Associate with each term a sign −1(h−l) . . paired external lines (lines with labels such as (i. b) that are paired vertically in the bra part of the amplitude equation) are considered connected externally through imaginary extensions. with the four-fold commutator)..) ˆ 7.. . . Lines that are quasi-equivalent (i. Summary of the rules of interpretation for coupled-cluster diagrams. ˆm vertex i a j b associate an amplitude tab.5). (Two ˆ T vertices are considered equivalent if they have the same number of line pairs and are connected in equivalent ways to the interaction verˆ1 vertices in tex.114). With any two-particle interaction vertex associate an antisymmetric two-electron integral left–out right–out left–in right–in . b. Associate a factor 1 2 also with each pair of equivalent Tm vertices. × associate a factor 2. a) or (j.in .) target indices. j. Label external (open) lines with the hole (i. 6. going in the same direction.) Label internal lines with hole and particle indices that are diﬀerent from the target indices.119). and because an interaction vertex has at most four lines and so can connect at most four ˆm vertices in any connected ˆm vertices.

(b) open diagrams. A zero is used to indicate the absence of such lines. for generating distinct antisymmetrized CC diagrams.121). which appear in the amplitude equations (10. the use of Hugenholtz-like diagrams can be very helpful in this process. The Hamiltonian-operator vertices are thus assigned labels as follows: × + × − 0 × × +− ++ −− +− + ++− − +−− 0 + + −− . We now describe another method. fall into two classes: (a) closed diagrams. using simple combinatorial elements. (9. These labels describe the type and number of lines below the interaction vertex and above the cluster-operator vertex (or vertices) of the CC diagrams. This method is based on the assignment of a set of plus and minus labels to each Hamiltonian-operator (interaction) vertex and to each cluster-operator diagram.3 Systematic generation of CC diagrams 297 The coupled-cluster diagrams. like MBPT diagrams. shown here with their interpretations: × i 1 4 ijab a b j i a fia ta i ia 1 2 i a b j ij ab tab ij b ij ab ta i tj ijab These diagrams provide the energy expression at all levels of the coupledcluster method. A plus sign is included for each open particle line and a minus sign for each open hole line.12). As we saw in the construction of the diagrams for the quadratic contribution to the CCD equations. There are only three distinct CC energy diagrams. 10.10. which occur in the energy equation (10.3 Systematic generation of CC diagrams The process of constructing all the distinct diagrams that appear in the CC equations requires careful consideration of all the distinct ways in which Hamiltonian vertices can be added to cluster-operator diagrams. The rules of interpretation for CC diagrams are given in Fig. The diﬀering energy results obtained at diﬀerent CC levels are due to the diﬀering amplitudes obtained for the cluster operators at those levels.1.11). 10.

105). 9.107).3). The interaction vertices must be selected to have the appropriate excitation level. using the + and − labels for guidance.e. ˆN vertex (the third diagram in (9. number of T ˆN vertices (the ﬁrst two diagrams in (9. (9. there are obvious limitations on the ˆm vertices that can be connected by each interaction vertex. diagrams from disconnected clusters of T We will illustrate this process by applying it to the generation of the ˆ(3) equation in Fig.121) (also seen in the T Q (T 2 2 ˆ 2 and the tarSince we are beginning with a quadruple excitation in T 2 ˆ2 . see (9. all the particle or hole lines below it are contracted with (i. Thus the . to produce ﬁnal diagrams of the desired overall excitation level. for each selected interaction vertex. connected to) matching open lines of the CC diagram. ˆ 2 ) diagrams of (9.107)) can connect to (+1) W ˆm vertex and thus cannot be used to generate connected CC only one T ˆm operators. only double-de-excitation (−2) inget diagram is a contribution to T teraction vertices can be used and there is only one of those. Furthermore. in all possible non-redundant ways.107)) can be used to generate the (+2) W connected CC diagrams.105)) nor Note that neither the (+1) F ˆ vertex (the eighth diagram in (9. Then.298 Systematic derivation of the coupled-cluster equations The cluster-operator vertices are given the following labels: +− + + −− + + + − −− + + + + − − −− For repeated cluster operators the labels for the individual vertices are partitioned by vertical lines: +−|++−− ++−−|++−− The key step in each stage of CC diagram construction is the addition ˆm vertices in all possible ways that of an interaction vertex above a set of T generate distinct connected CC diagrams of the required excitation level.105)) or the Thus. (0)-excitation F ˆ vertices (the fourth and ﬁfth diagrams in (9.

121).10. since the two T equivalent to + + | − − and + + −|− is equivalent to −| + +−. i j a b i a b j i a b a j i j b . but it provides the sign determination directly. Furtherˆ2 vertices are equivalent. must be selected for contraction with the top part. The order of the + and − labels in each partition is irrelevant. giving rise to four disˆ2 vertices to the interaction tinct contraction patterns that connect both T vertex: ++|−− + −| + − ++−|− + − −| + . The use of antisymmetrized Goldstone diagrams together with ˆ operators and the above simple combinatoˆ N and T the normal-ordered H rial scheme generates all amplitude diagrams unambiguously and without repetition. together with their + and − labels. ˆ1 is due to the one-electron operator f to T = (1) × .5 based on Hugenholtz-like diagrams. 10. (10. ++−−|++−− Four lines in the bottom part. the only ﬁrst-order wave-function contribution ˆo .4 The coupled-cluster singles and doubles (CCSD) equations As can be seen from (9. the pattern − − | + + is more.14) .4 The coupled-cluster singles and doubles (CCSD) equations 299 diagram parts that need to be connected. The four distinct contraction patterns generate the four connected quadratic diagrams Qa –Qd of (9. This procedure is equivalent to the approach described in Section 9.78). corresponding to two + and two − labels. are + + −− .

including single excitations in truncated CI is not nearly as eﬀective in reducing this sensitivity. including T function greatly reduces the sensitivity to orbital choice in CCSD and higher ˆ approximations.300 Systematic derivation of the coupled-cluster equations The second. (10. Thus we can say that in the HF case third-order energy come from T CCD is correct through ﬁrst order in the wave function and third order in the energy.and contribution to T ˆ2 . f block diagonal and any diagram containing a one-body particle–hole interac× ) vanishes. Therefore CCD is just a manifestation of the block-diagonal nature of f based on HF orbitals will frequently provide quite reasonable approximations. Thus there is no ﬁrst-order wave function tion ( × or ˆ1 . (10. though it includes many higher-order contributions involving disconnected clusters such as disconnected quadruple excitations. which do not satisfy the usual Brillouin theorem either. This property of eT1 is known as the Thouless theorem (Thouless.5. The same is true for restricted open-shell HF (ROHF) reference functions. because ˆ1 . and single excitations can make substantial contributions to the ﬁrst-order wave function and second-order energy. and the only nonzero contributions to the second.16) . ˆ1 and its disconnected cluster products in the wave Furthermore. However. because the eﬀect of eT1 is to transform any single determiˆ nant into another single determinant. Thus the init leaves out most of the disconnected clusters involving T ˆ1 introduces an important ﬂexibility in CCSD and higher CC clusion of T applications. which ˆ. A primary reason for the inclusion of single excitations in many cases ˆ is is that it permits the use of non-HF reference functions. 9. .15) ∆E (2) = ˆ is When a Hartree–Fock reference function (canonical or otherwise) is used. However. Section 1. for which f not block diagonal and the Brillouin theorem does not hold.and third-order contributions to the CC energy (see Fig. 1960). The same conclusion can be reached by use of the Brillouin theorem. In such cases ˆ the matrix elements Φa i |H |Φ0 = fia can be quite large.3) are × + (1) ∆E (3) = (2) + (2) (1) × + (1) (1) . the addition of single-excitation contributions adds 35 diagrams but entails little computational eﬀort and can be very useful in many situations.

C´ ıˇ zek and Jeziorski 1989. which do not vanish in general even when HF orbitals are erators 0|U i used.19) contain disconnected tripleˆ4 . a ˆ 1 ˆ2 1 ˆ3 ˆ ˆ ˆ ˆ i |HN (1 + T2 + T1 + T1 T2 + 2 T1 + 3! T1 )|0 C = 0 . we substitute T and (10. . has to do with the calculation of density matrices and associated properties. ˆ2 into (10. While the value in this equation (because of the two-particle nature of H ˆ2 amplitudes only. ab ˆ ˆ2 + T ˆ2 + 1 T ˆ2 + 1 T ˆ2 T ˆ2 + 1 T ˆ3 + ˆ2 + 1 T ˆ1 + T ˆ1 T |HN (1 + T ij 2 2 2 1 2 1 3! 1 (10.4 The coupled-cluster singles and doubles (CCSD) equations 301 Another important reason for including single excitations. Pople. and should be treated on an equal footing.and double-excitation contributions to such properties enter at the same order of perturbation theory. adversely aﬀect the quality of the results (Watts. Head-Gordon and Pople 1990). the equations take the form ˆ2 + T ˆ1 + 1 T ˆ N (T ˆ2 0|H 2 1 )|0 C = ∆E .10.17) retains this form even in higher CC models.18). Such calculations involve matrix elements of various one-particle opˆ |a . (10. quadruple and higher excitations in higher CC models because these amplitudes are aﬀected by the higher excitations through the coupled amplitude equations.19) The energy equation (10.17) (10. a reason that holds even for unrestricted and closed-shell restricted HF reference functions. is an approximation to CCSD in which all ˆ2 term in (10. in some cases. except the T terms containing T ˇ left out (Paldus. are ˆ1 T ˆ1 in (10. however. 1990.11) ˆ =T ˆ1 + T To derive the CCSD equations. the QCISD model of Pople.17)–(10. it is aﬀected by ˆ1 and T energy depends directly on the T the inclusion of connected triple.18) 1 ˆ4 4! T1 )|0 C = 0.19). However.18) can include up to triple-excitation terms. the omission of these terms has negligible impact on the computational cost of the model and may. The single-excitation equations (10. Simmodel T ilarly. Urban and Bartlett 1995). HeadGordon and Raghavachari (1987).12). the double-excitation equations (10. since clusters of more than two particles cannot contribute to the expectation ˆ N ). ˆ3 or T and quadruple-excitation terms but not T The “quadratic CI” analog of CCSD. Single. in the CCSD ˆ3 = 0 so only disconnected triple-excitation clusters appear. Head-Gordon and Raghavachari 1989. Arranging the resulting terms in approximate order of their importance. Raghavachari.

using the scheme introduced in Section 10. giving the antisymmetric form ∆E = 1 4 ijab a b b a ij ab (tab ij + ti tj − ti tj ) + ia fia ta i . 10. these three diagrams are all the distinct closed diagrams that can appear in any CC energy equation.18) are collected in Fig. The numbering of the diagrams corresponds to the order of the terms in (10. The generation and interpretation of these diagrams is shown in Table 10. The corresponding algebraic energy expression is ∆E = 1 4 ijab ij ab tab ij + ia fia ta i + 1 2 ijab b ij ab ta i tj . The summations are over all the common indices × S1 S2a × S2b S2c × S3a × S3b × S3c S4a S4b S4c S5a S5b S5c S6 ˆ1 Fig. Antisymmetrized Goldstone diagrams representing the CCSD T equations. The diagrams representing the single-excitation CCSD equations (10. 10. (10.18).1).20) As noted in the previous section. (10.1.3 and the rules of interpretation (Fig.2. the ﬁxed labels i and a are understood for all open lines. . (10.22) In the HF case fia = 0 and the last term in (10. 10. the CCSD energy equation can be written × ∆E = + + .302 Systematic derivation of the coupled-cluster equations Diagrammatically.21) The last term can be combined with the ﬁrst after splitting it into two terms and interchanging a and b in one of them.2.22) vanishes.

requires −2 interaction vertex): T + + −− +| + −− −| + +− +−|+− 1 ˆ2 2 T1 S4a S4b S4c −1 2 1 −2 ad kl cd tc i tkl cd kl cd ta k til da kl cd tc k tli (vertices + − | + −.4 The coupled-cluster singles and doubles (CCSD) equations 303 ˆ1 Table 10. requires +1 interaction vertex) 1 (no T 0 0 S1 fai ˆ2 (vertex + + − −. internal hole lines). requires −1 interaction vertex): +|− +| + − −| + − S5a S5b S5c − − a fkc tc i tk d ak cd tc i tk c kl ic ta k tl +− ++− +−− 1 ˆ3 3! T1 (vertices + − | + −| + −. from left to right. requires 0 interaction vertex): T + − +− + − +− S3a S3b S3c − fac tc i fki ta k ak ic tc k ˆ1 T ˆ2 (vertices + − | + + − −. .18)) Interaction Contractions Diagram Interpretationa ˆ vertex. so are b and j etc. subject to the condition that a and i are on the same continuous path.1. open hole lines. within each class of lines (open particle lines. Generation and interpretation of diagrams for CCSD T amplitude equations (see (10.10. the algebraic interpretation of the diagrams assumes that labels are assigned in alphabetical order. internal particle lines. requires −2 interaction vertex): +| + − | − S6 − a d kl cd tc i tk tl + + −− a In this and all subsequent cases. requires −1 interaction vertex): T +− ++− +−− +− ++− +−− S2a S2b S2c 1 2 1 −2 fkc tac ik ak cd tcd ik kl ic tac kl ˆ1 (vertex +−.

19). l. S3b ) and move them a to the opposite side of the equation as (fii − faa )ta i = (εi − εa )ti . c. The two 1 ˆ2 ˆ 1 ˆ3 cubic terms.23) (corresponding to diagrams S3a .25). a). 4! T1 .304 Systematic derivation of the coupled-cluster equations k. The single-excitation CCSD equations are then fai + kc fkc tac ik + fki ta k + kc da tc k tli 1 2 kcd ak cd tcd ik − tc k − 1 2 klcd 1 2 klc kl ic tac kl + c ad tc i tkl fac tc i cd kl cd ta k til − k ak ic − kc kl cd + kcd − 1 2 klcd + klcd kl cd a fkc tc i tk ak cd d tc i tk − klc c kl ic ta k tl − klcd a d kl cd tc i tk tl = 0 (for all i.23) (diagrams S1 .23) As in the CCD case.3. The remaining terms are represented by the diagrams ˆ1 contributions to in Fig. is represented by diagrams D6a –D6c . ε i − εa (10. 1 2 T1 . are represented by diagrams D7a –D7e and 1 ˆ4 D8a . respectively. εi − εa = ia (10. second and eleventh terms of (10. since those were present in CCD. (10. S2a . 10. S5a ) vanish. D4b represent the linear T ˆ2 term. . (10. d that occur in each term.19). we separate the diagonal elements from the ﬁfth and sixth terms in (10. In the canonical HF case the oﬀ-diagonal parts of the ﬁfth and sixth terms also vanish. Diagrams D4a . For the double-excitation part of the CCSD equations. we have already derived the contributions from the ﬁrst three terms of (10. while diagram D9 represents the quartic term. while diagrams D5a –D5h represent the quadratic T ˆ2 other quadratic term. It is then obvious that the ﬁrst approximation to ta i is ti a (1) = fai . in order to set up an iterative solution. 2 T1 T2 and 3! T1 .25) In the Hartree–Fock case the ﬁrst. Anˆ1 T (10. D8b .24) and when this approximation is put into the energy expression it gives the second-order MBPT energy contribution from the single excitations. × × |fia |2 . and so does the second-order energy contribution from the single excitations.

. The generation and interpretation of these diagrams is given in Table 10. There are three such components. × +− ++− +−− from which the eight diagrams D5a –D5h are generated.10. D5d and D7c could also have been drawn in the equivalent forms D5c D5d D7c as can be veriﬁed by examining their + and − contraction patterns (or.3.2. from the corresponding Hugenholtz-like forms). Antisymmetrized Goldstone diagrams representing T ˆ2 equations. ˆ2 contribution. and since we ˆ2 equations. the CCSD T Note that diagrams D5c . ˆ1 T As an illustration of the procedure we shall consider the T The cluster operators in this term provide a +3 excitation.4 The coupled-cluster singles and doubles (CCSD) equations 305 × D4a D4b D5a × D5b D5c D5d D5e D5f D5g D5h D6a D6b D6c D7a D7b D7c D7d D7e D8a D8b D9 ˆ1 contributions to Fig. 10. equivalently. we can only use the −1 need a ﬁnal +2 excitation for the T components of the Hamiltonian.

Generation and interpretation of diagrams for T ˆ to CCSD T2 equations Interaction Contractions Diagram Interpretation ˆ1 (vertex +−. requires −1 interaction vertex): T +− ++− +|− −|+ +| + − −| + + + − |+ +−− −| + − +| − − + − |− 1 ˆ2 T 2 1 D5a D5b D5c D5e D5g D5d D5f D5h ˆ (ij ) −P ˆ (ab) −P ˆ (ij |ab) P ˆ (ab) −1 P 2 ˆ (ab) P ˆ (ij |ab) −P 1 ˆ P (ij ) 2 ab fkc tc i tkj cb fkc ta k tij db ak cd tc i tkj cd kb cd ta k tij db ka cd tc k tij cb kl ic ta k tlj ab kl cj tc i tkl ab kl ci tc k tlj ˆ (ij ) −P (vertices + − | + −. requires +1 interaction vertex): T + − + − D4a D4b ˆ (ij ) P ˆ (ab) −P ab cj tc i kb ij ta k ˆ1 T ˆ2 (vertices + − | + + − −.306 Systematic derivation of the coupled-cluster equations ˆ1 contributions Table 10.2. requires −1 interaction vertex): D8a D8b 1 ˆ P (ij |ab) 2 1 ˆ P (ij |ab) 2 a d ˆ kb cd tc i tk tj = P (ab) a b ˆ kl cj tc i tk tl = P (ij ) a d kb cd tc i tk tj a b kl cj tc i tk tl ++− +−− 1 ˆ4 T 4! 1 +| − |+ +| − |− (vertices + − | + −| + −| + −. requires −2 interaction vertex): +| − |+ D9 1 ˆ P (ij |ab) 4 d a b kl cd tc i tj tk tl = d a b kl cd tc i tj tk tl ++− . requires −2 interaction vertex): D7a D7b D7c D7d D7e 1 ˆ P (ij ) 4 1 ˆ P (ab) 4 ab d kl cd tc i tkl tj = cd b kl cd ta k tij tl = 1 2 1 2 ab d kl cd tc i tkl tj cd b kl cd ta k tij tl + + −− +| + | − − −| − | + + +| − | + − + − | + |− + − | − |+ 1 ˆ3 T 3! 1 ˆ (ij |ab) −P ˆ (ij ) −P ˆ (ab) −P a db kl cd tc i tk tlj d ab kl cd tc k ti tlj a db kl cd tc k tl tij (vertices + − | + −| + −. requires 0 interaction vertex): D6a D6b D6c 1 ˆ P (ij ) 2 1 ˆ P (ab) 2 d ab cd tc i tj = b kl ij ta k tl = d ab cd tc i tj b kl ij ta k tl ++ −− +− 1 ˆ2 ˆ T T 2 1 2 +|+ −|− +|− ˆ (ij |ab) −P a kb cj tc i tk (vertices + − | + −| + + − −.

the P (ij |ab) permutation operator cancels with the factor 1 4 . This cancellation is with the P D8b . The factors 1 4 in diagrams D7a . These two factors cancel with each other (rule 10). since permuting a with b or i with j is equivalent to relabeling the internal lines. The factors 1 2 in diagrams D5e and D5f result from the presence of a pair of equivalent internal lines. resulting in a factor 1 4 and ˆ two pairs of inequivalent external lines. Diagrams D6a . and thus ˆ they acquire factors of 1 2 . it will be instructive to review some of the numerical coeﬃcients and the permutation operators in the table in detail.2 together with those previously ˆ2 amplitude equations for CCSD are obtained for the CCD equations. as shown. Combining all the results in Table 10. c. as shown. respectively. One factor 2 cancels with the permutation operator in each of these diagrams. since the permutation is equivalent to the relabeling of internal lines corresponding to dummy summation indices. D6b each have a pair of equivalent vertices (as shown by the contraction patterns) and a pair of external lines that are inequivalent (because they are connected to diﬀerent vertices).4 The coupled-cluster singles and doubles (CCSD) equations 307 The summations in the interpretation column of Table 10. In diagrams D7c –D7e there are no equivalent lines and no equivalent vertices. Diagram D9 has two pairs of equivalent vertices (two vertices using a + contraction and two using a − contraction).1) and a permutation operator in each case. d present in each term. To clarify the interpretation rules further.10.2 are over the internal labels k. 10. Fig. D8b each have a pair of equivalent vertices. D7b arise because each diagram has one pair of equivalent internal lines and one pair of equivalent vertices (in diaˆ1 vertices while in D7b we use − gram D7a we use + contractions for both T 1 contractions for both). resulting in a factor 1 2 (rule 7. +| − |+ and +| − |−. These factors cancel with part of the P (ij |ab) permutation operator that results from the two pairs of inequivalent external ˆ (ab) in ˆ (ij ) operator in D8a and with P lines. as shown in the table. Again. l. as can be seen from their contraction patterns. the T obtained as ˆ (ij ) DCCD + P c ˆ ab cj tc i − P (ab) k ˆ kb ij ta k − P (ij ) kc db tc i tkj ab fkc tc i tkj ˆ (ab) −P kc cb fkc ta k tij ˆ (ij |ab) +P kcd ak cd ˆ (ij |ab) −P klc cb 1 ˆ kl ic ta k tlj − 2 P (ab) kcd cd kb cd ta k tij . Diagrams D8a .

Thus the cost remains proportional to nh p 6 an n process.126). a > b. doubles and triples (CCSDT) equations The importance of various cluster contributions to the wave function and to the energy is best judged in terms of the order of perturbation theory in which they ﬁrst enter. and (9. The second-order wave function.308 Systematic derivation of the coupled-cluster equations ˆ +1 2 P (ij ) klc ab ˆ kl cj tc i tkl + P (ab) kcd db ka cd tc k tij b kl ij ta k tl kl ˆ (ij ) −P klc kl ci ab tc k tlj + cd d ab cd tc i tj + ˆ (ij |ab) −P kc kb cj kl cd klcd a tc i tk + 1 2 klcd kl cd d ab tc i tj tkl + 1 2 klcd b cd kl cd ta k tl tij ˆ (ij |ab) −P ˆ (ab) −P klcd a db tc i tk tlj ˆ (ij ) −P klcd kl cd d ab tc k ti tlj a db ˆ kl cd tc k tl tij + P (ab) kcd a d kb cd tc i tk tj ˆ (ij ) +P klc kl cj a b tc i tk tl + klcd d a b kl cd tc i tj tk tl =0 for all i > j. Section 10. in which the one-electron ˆ is block diagonal. As can be seen from Eqs. the only ﬁrst-order contributions to the wave function come from singleand double-excitation conﬁgurations. 10. (9. Therefore the lowest-order CC interaction operator f model is CCD. (10. (9. which is included in the .4).93).5 Coupled-cluster singles.94). Furthermore.78).79).79)). or roughly model. As previously discussed (see e. the ﬁrst-order singleexcitation contribution vanishes in the HF case. these models include all the secondand third-order contributions to the energy.g. (9. and the fourth-order energy add contributions from triple and quadruple excitations. followed (particularly in non-HF cases) by CCSD.26) where DCCD stands for all the terms on the left-hand side of the CCD amplitude equations (9. In a CI expansion the most important second-order contribution to the wave function (at least in the HF case) comes from quadruple excitations and consists entirely of the ˆ2 disconnected-cluster 1 2 T2 term (diagram 8 in (9. None of the terms in the CCSD equations raises the dependence of the computational cost on the number of spinorbitals over that of the CCD 2 n 4 per iteration.

79).30) because they Note that T cannot produce connected triple-excitation diagrams with one interaction vertex. see also (9. Therefore the principal contribution to the CC wave function beyond the CCSD model is due to the connected triple-excitation ˆ3 (diagrams 6 and 7 in (9.93)) and the fourth-order energy. The CCSDT the presence of the T amplitude equations are a ˆ i |HN (1 ˆ2 + 1 T ˆ2 ˆ2 + T ˆ1 + T ˆ1 T +T 2 1 + 1 ˆ3 3! T1 ˆ3 )|0 +T C = 0. is the next CCSDT model.30) ˆ1 and its powers have been left out of (10. |ΨCCSDT = exp(T (10.29) abc ˆ 1 ˆ2 ˆ ˆ ˆ ˆ ˆ ˆ ˆ ˆ ijk |HN (T2 + T3 + 2 T2 + T1 T2 + T2 T3 + T1 T3 1 ˆ ˆ2 1 ˆ2 ˆ 1 ˆ3 ˆ ˆ2 ˆ +1 2 T1 T2 + 2 T1 T2 + 2 T1 T3 + 3! T1 T2 )|0 C = 0.10.17).95)). Note also that T that do not contain the one-electron interaction. the disconnected triple-excitation contribution to the second-order wave function (diagrams 18 and 19 in (9. . The CCSDT wave function is ˆ1 + T ˆ2 + T ˆ3 )|0 . which adds the terms containing T level of improvement in CC methodology and is correct through the secondˆ1 contributions order wave function and fourth-order energy.79)) contains a one-electron interaction and vanishes in the HF case.) However. (10.28) ab ˆ ij |HN (1 1 ˆ2 ˆ2 ˆ ˆ2 + 1 T ˆ ˆ +T 2 2 + T1 + T1 T2 + 2 T1 ˆ12 T ˆ2 + 1 T ˆ13 + 1 T ˆ14 + T ˆ3 )|0 ˆ3 + T ˆ1 T + 1T 2 3! 4! C = 0. (10. Thus the ˆ3 to CCSD.e. though ˆ2 amplitudes that appear in the equation will be aﬀected by ˆ1 and T the T ˆ3 operator in the amplitude equations. doubles and triples equations 309 ˆ4 ﬁrst CCD and CCSD models.79). (10. i. see also (9.5 Coupled-cluster singles.27) The energy equation is the same as in the CCSD model. also enter the second-order wave function (diagrams 4 and 5 in (9. This contribution cluster T ˆ2 enters the energy in fourth order. (The connected quadruple-excitation term T enters in the third-order wave function and ﬁfth-order energy. (10. and thus do not vanish even in the HF case. the same order as 1 2 T2 .

represented by the six diagrams in Fig. . 10.32) where SCCSD stands for the entire left-hand side of the CCSD singles equation (10.23). The generation and interpretation of diagrams D10a –D10c .31) The CCSDT singles equation can then be written as 1 4 mnef SCCSD + mn ef taef imn = 0 (for all i. repˆ3 . × D10a D10b D10c D11a D11b D11c ˆ3 contributions Fig. is straightforward.4.310 Systematic derivation of the coupled-cluster equations The single-excitation equation has one new term added to the terms in the corresponding CCSD equation (10.4. (10. The doubles equation has two new terms. 10. giving the algebraic expressions ˆ NT resenting the term H 1 ˆ fme tabe ijm + 2 P (ab) me mef 1 ˆ bm ef taef ijm − 2 P (ij ) mne mn je tabe imn .18). This new term is represented by the diagram = S7 1 4 mnef mn ef taef imn . Antisymmetrized Goldstone diagrams representing the T ˆ to the CCSDT T2 equations. a) . (10.

10. For the algebraic interpretation of these diagrams we introduce an extension of the notation for the permutation operators. (10. .26). Thus we need to include the permutations P ˆ ˆ and Pac . P (10. The triple-excitation CCSDT equation (10. in which all six permutations of i. requiring ˆ (ijk |a/bc). but not Pbc . and putting labels on the external lines. in diagram T4c all three open hole lines are inequivalent.5. we need the operator P ˆ (a/bc|k/ij ) = P ˆ (a/bc)P ˆ (k/ij ). but that b ˆab and c are quasi-equivalent.34) ˆjk . the total permuˆ (k/ij ) = 1 − P ˆik − P Similarly. We therefore deﬁne the permutation operator ˆ (a/bc) = 1 − P ˆab − P ˆac .30) has two linear. In tation operator for the diagram T1a is P contrast. a i j b k c . we see that lines a and b are inequivalent. doubles and triples equations 311 ˆ NT ˆ1 T ˆ3 we have For the diagrams representing H Vertices +−|+++−−− Interaction + + −− Contraction +−|+− −| + + − +| − − + Diagram D11a = f ab mn ef te m tni D11b = ef b ˆ −P (ab) mn ef ta m tinj D11c = af b ˆ (ij ) −P mn ef te i tmnj The CCSDT doubles equation then takes the form DCCSD + m.5 Coupled-cluster singles. j. Taking diagram T1a as an example.10. a > b. which generate the 47 diagrams shown in Fig.33) where DCCSD stands for all the terms on the left-hand side of the CCSD doubles equation (10. four quadratic and three cubic terms and one quartic term.e 1 ˆ fme tabe ijm + 2 P (ab) mef bm ef taef ijm f ab mn ef te m tnij − 1 ˆ 2 P (ij ) mne mn je mn ef tabe imn + mnef ˆ (ab) −P mnef ef b ta m tinj ˆ (ij ) −P mnef af b mn ef te i tmnj =0 for all i > j. k the permutation operator P are included. and so are a and c.

in corresponding order.5. 10. T9d . is shown in Fig. 10. Note that diagrams T7c . m.6. T7d . The summations are over those of the internal indices l.312 Systematic derivation of the coupled-cluster equations × T1a × T3a T3b T3c T1b T2a T2b × T2c T2d T2e T3d T3e T4a T4b T4c T4d T5a T5b T5c × T6a × T6b T5d T5e T5f T5g T6c T6d T6e T6f T6g T6h T7a T7b T7c T7d T8a T8b T8c T8d T8e T9a T9b T9c T9d T9e T10a T10b ˆ3 Fig. 10. T10a . T9c . which gives the complete CCSDT triples equation. d.5. Antisymmetrized Goldstone diagrams representing the CCSDT T equations. e that are present in each term. The algebraic interpretation of the diagrams in Fig.

doubles and triples equations ˆ (k/ij |a/bc) P ˆ (k/ij ) −P ˆ bc dk tad ij − P (i/jk |c/ab) 1 ˆ flk tabc ijl + 2 P (c/ab) 313 fcd tabd ijk ˆ lc jk tab il + P (c/ab) 1 ˆ ab de tdec ijk + 2 P (k/ij ) bc fld tad ij tlk bc lm dj tad il tmk lm ij tabc lmk ˆ (i/jk |a/bc) +P ˆ (i/jk |abc) +P ˆ −1 2 P (i/jk |c/ab) ˆ (i/jk |c/ab) +P ˆ (ijk |a/bc) −P ˆ (i/jk |a/bc) +P ˆ −1 2 P (a/bc) ˆ al id tdbc ljk − P (k/ij |a/bc) ec ˆ lb de tad il tjk − P (ijk |a/bc) de 1 ˆ lc de tab il tjk + 2 P (k/ij |a/bc) ec ˆ ab de td i tjk + P (k/ij |a/bc) bc ˆ al id td j tlk − P (i/jk |abc) ebc 1 ˆ lm de tad il tmjk − 2 P (i/jk ) bc lm dk tad ij tlm bc lm ij ta l tmk dc al id tb l tjk abc lm de tde li tmjk bec lm de tad ij tlmk abc lm de tde ij tlmk dbc fld ta l tijk ebc 1 ˆ lm de tda lm tijk − 2 P (k/ij |a/bc) dec 1 ˆ lm de tab il tjmk + 4 P (k/ij ) ˆ −1 2 P (i/jk |c/ab) ˆ +1 4 P (c/ab) ˆ (i/jk |a/bc) +P ˆ (a/bc) +P ˆ −1 2 P (abc) ˆ (i/jk |abc) −P ˆ (k/ij |a/bc) −P ˆ (i/jk |c/ab) −P ˆ (i/jk |abc) −P ˆ +1 2 P (k/ij |a/bc) ˆ (a/bc) −P ˆ +1 2 P (b/ac) ˆ (k/ij |a/bc) +P =0 dec ˆ lm de tab lm tijk − P (i/jk ) abc ˆ fld td i tljk − P (a/bc) dbc lm id ta l tmjk ebc ˆ al de td i tljk − P (i/jk |a/bc) ebc ˆ la de td l tijk − P (i/jk ) dec 1 ˆ lb de ta l tijk + 2 P (ijk ) abc lm di td l tmjk abc lm dj td i tlmk a bc lm dj td i tl tmk b dc lm id ta l tm tjk e bc lm de tad il tj tmk db ec lm de ta l tim tjk e abc lm de td l ti tmjk a ec ˆ lb de td i tl tjk + P (ijk |a/bc) e bc ˆ al de td i tj tlk + P (i/jk |c/ab) ab ec ˆ lm de td l tim tjk − P (ijk |a/bc) b ec 1 ˆ lm de tad il tm tjk + 2 P (i/jk |c/ab) de bc ˆ lm de ta l tij tmk − P (i/jk ) a ebc 1 ˆ lm de td l tm tijk + 2 P (j/ik ) abc e lm de td i tljm tk a ebc lm de td i tl tmjk a b ec lm de td i tl tm tjk dbe c ˆ lm de ta l tijk tm − P (i/jk |a/bc) a e bc ˆ lm de td i tl tj tmk + P (i/jk |c/ab) (for all i > j > k.5 Coupled-cluster singles.10.6. 10. Fig. The CCSDT T . a > b > c) ˆ3 equations.

4 P (ijk ). Note that each T vertex is ﬁrst order in the non-HF case but second order in the HF case ˆ3 vertices are ﬁrst and ˆ2 and T (canonical or noncanonical). and so the resultand T10b each have a pair of equivalent T 1 ing factors 2 cancel against part of the corresponding permutation operˆ ator. The diagonal part of the one-electron interaction vertex is zero order while the oﬀ-diagonal part is ﬁrst order in general but does not contribute in the canonical HF case and contributes only for particle–particle and hole–hole vertices in the noncanonical HF case. 10. Lowest wave-function orders at which triple-excitation CCSDT diagrams (Fig. diagonal T2a . 1 P ˆ ˆ become −P 2 (j/ik ). T2b . respectively.3.3. respectively. oﬀ-diag. T2c –T2e T3a T3b –T3e T4a –T4d T5a –T5g T6a .5) contribute Diagrams T1a . T10b Non-HF 2 2 3 3 3 3 3 4 4 4 4 4 5 5 Noncanonical HF 2 2 3 3 — 3 4 4 — 5 6 5 7 8 Canonical HF 2 2 — 3 — 3 4 4 — 5 6 5 7 8 ˆ1 vertices. 1 ˆ 1 ˆ 1 ˆ 1 ˆ 1 ˆ 2 P (i/jk |abc). the original factors for these diagrams. T1b T2a . 2 P (ijk |a/bc) and 2 P (i/jk |abc) (in order) 1 ˆ ˆ (k/ij |a/bc).314 Systematic derivation of the coupled-cluster equations Table 10. 4 P (abc). P ˆ (i/jk|c/ab). Speciﬁcally. − 1 2 P (ijk |a/bc). 2 P (b/ac). T6b T6c –T6h T7a –T7d T8a –T8e T9a –T9e T10a . T2b . The lowest energy order to which these diagrams contribute is two higher than . and P The respective lowest wave-function orders at which the triple-excitation ˆ1 CCSDT diagrams contribute are listed in Table 10. while the T second order. P (k/ij |a/bc) ˆ (i/jk|c/ab).

T2b (the only nonzero part of these diagrams in the canonical HF case) and move it to the other side of the equation. T1b and the diagonal parts of T2a . The oﬀ-diagonal parts of diagrams T2a .4) in the CCSDT doubles equations. T2b of the CCSDT triples equations. The CCSDT-1b model adds the remaining terms of the CCSDT doubles equations. T2b . Applying the permutation operators (see the third and fourth terms in Fig. all of which are of second order. The CCSDT-1a model includes the T the CCSDT singles equations. To set up an iterative sequence. depending upon which diagrams involving a T are included in the CCSDT doubles equations (Urban. are included in other cases in order to maintain the invariance of the model to separate unitary transformations of the occupied and virtual orbitals.5 Coupled-cluster singles. Lee.35).4). the solution rethe T quired iteration of all three sets of equations. because ˆ2 amplitudes enter in the right-hand side of (10. 10. to 1985). doubles and triples equations 315 the wave-function order. (10.32).6).10. since at least two interaction vertices are needed to close any of the triple-excitation diagrams. (10. we get a total of six terms: abc abc (εi + εj + εk − εa − εb − εc )tabc ijk = εijk tijk .5) are diagrams T1a . This invariance permits the removal of these oﬀ-diagonal parts by an orbital transformation to semicanonical orbitals. T1b and T2a . Cole et al. which vanish in the canonical HF case. which diagonalizes the hole–hole and particle– .33) but. and diagrams T1a . 10. (10. 10. ˆ3 contribution S7 . Kucharski and Bartlett 1984) and ˆ3 vertex has two variants. The leading terms in the triple-excitation CCSDT equations (Fig. Diagram D10a of the doubles equation vanishes in the HF case. 10. T2b . diagrams D10a –D10c (Fig. The ﬁrst attempt at including triple-excitation eﬀects in the CC method used just the leading (second-order) terms in the CCSDT triples equations: abc ˆ εabc ijk tijk = P (a/bc|k/ij ) d ˆ bc dk tad ij − P (c/ab|i/jk ) l lc jk tab il . Noga. diagrams D11a –D11c (Fig. This model has been called CCSDT-1 (Lee and Bartlett 1984.31). following a similar procedure to that of previous cases we separate the diagonal part of each of diagrams T2a .35) The tabc ijk amplitudes obtained from these equations were then substituted in the CCSDT singles and doubles equations (10.

the full CCSDT involves an nh np step per iteration (diagram T2c . 3 5 Moreover. in turn. in the ROHF case this transformation destroys the equivalence of the spatial parts of the α-spin and β -spin orbitals. However. In all cases the full CCSDT method requires storage of the tabc ijk amplitudes.7). 10. Head-Gordon et al. which. ad ij /εad ij and ab il /εil . besides adding another nh p each iteration dependent on the triples amplitudes of the previous iteration and thus requires the amplitudes to be stored. there is another nh p ˆ3 amplitudes into diagrams D10b diagonal part of T2a . for CCSDT-1b. Noga and Bartlett 1988). The iterative process for the solution of the CCSDT-1 equations in the canonical HF case is shown diagrammatically in Fig. T2b . Initially. Besides the extra power of nh . the CCSDT-1 . While the full CCSDT model has been implemented (Noga and Bartlett 1987. to D11a –D11c ) of the CCSDT doubles equations (see Fig. ˆ3 amplitudes from (10. This diﬃculty can be eliminated by performing a semicanonical transformation. making the computational eﬀort similar to that of a UHFbased calculation by precluding spin summations and essentially doubling the ranges of the various indices. with inclusion of open-shell systems (Watts and Bartlett 1990). the inclusion of the oﬀ-diagonal parts of diagrams 3 n 4 step. the ﬁrst term in Fig. there also is the potential 3 n 3 tabc amplitudes. as already noted. The CCSDT-1 model is invariant under such a transformation provided that the full diagrams T2a . makes the triples amplitudes of T2a .6). the ﬁfth term in Fig. Except be compared with the nh p 3 n 4 step per iteration for the oﬀfor the canonical HF case. T2b are included. 10.6). The insertion of the T 3n 4 and D10c of the CCSDT doubles equations (10.316 Systematic derivation of the coupled-cluster equations ˆ (without eliminating the oﬀ-diagonal blocks) (Handy.7. 3n 4 nh p ad (1) ab (1) In cases other than canonical HF (in this context these cases include restricted open-shell HF).35) in CCSDT-1 includes an The evaluation of the T step per iteration (diagram T1a . Gauss and Bartlett 1993). Fortunately.35) to diagrams S7 of the CCSDT singles equations and D10a – D10c (and. 1989. Watts. which may 2 n 4 process per iteration in CCD or CCSD. Trucks. particle blocks of f Pople. the abc (2) is obtained from the ﬁrst-order approxsecond-order approximation tijk imations tij and til . it is diﬃculty of having to store the ∼nh p ijk possible to avoid this latter diﬃculty in the CCSDT-1 model (and also in the higher-order CCSDT-2 and CCSDT-3 models to be described later) by evaluating the contribution of each tabc ijk amplitude as soon as it is calculated from (10. 10.33) includes yet another nh p step per iteration. are just two-electron integrals ab divided by denominators. 10.

7. 10. The simplest way abc (2) to achieve this reduction would be to use the second-order tijk amplitudes computed from the initial (ﬁrst-order) tij and til amplitudes in all subsequent iterations of the CCSDT singles and doubles equations. this procedure provides just the fourth-order triple-excitation 3 n 4 steps. only one nh p in eﬀect. the contributions to the energy yet it requires two nh p evaluation of triple-excitation contributions to the MBPT(4) energy requires 3 n 4 step.h. However. by “squaring” the tijk abc (2) ad (1) ab (1) quantity in the MBPT(4) procedure (see . 10. doubles and triples equations 317 = + ← + + + + + + + + ← + + + Fig. so a further approximation that eliminates the iterations in the evaluation and substitution of the tabc ijk amplitudes would result in a signiﬁcant reduction in the computational cost. This more economical MBPT evaluation is achieved. of the equations. Typically.126). in the iteration. The horizontal arrows indicate that the value of the expression on the r.s.7. As can be seen from the description of the iterative solution of the CCSDT-1 equations in Fig. One cycle in the CCSDT-1 iteration scheme (for the canonical Hartree– Fock case).s. As noted in the discussion following (9.h.h.s.10. model and other CCSDT models have been developed in order to avoid the very high computational cost of the full CCSDT equations. the number of iterations required to solve the CCSDT-1 equations is between 10 and 20. is used as the next approximation for the l.5 Coupled-cluster singles. a denominator corresponding to a resolvent line above each interaction vertex is to be understood for each diagram on the r.

9). Cole et al. we can employ Instead of using second-order T ˆ2 amplitudes.36) by the ˆ [2] vertex can be expected to provide additional higher-order adjoint of the T 3 corrections and may be justiﬁed in terms of an expectation-value form of the energy formula (Urban. In the same sense we consider converged ˆ1 amplitudes as second order in the HF case but ﬁrst order otherwise. while T ˆ4 amplitudes are second and third order.36) ˆ3 vertex as This formula can be understood by writing the second-order T + (10. It was ﬁrst termed CCSD+T(CCSD). and the eﬀect of diagrams T2a .36) by the improved T 3 energy contributions of diagrams D10b . Noga. tijk = [2] ˆ2 amplitudes as ﬁrst order since their initial appearance counting converged T is in the ﬁrst-order wave function. ˆ3 amplitudes obtained from (10. as can be seen by using the adjoint of (10. (10.35) using converged CCSD T T abc [2] ˆ3 amplitudes t We designate these improved T ijk . Replacing ˆ3 and T T ˆ [2] vertex provides the just the bottom vertex in (10. and is thus only suitable for the canonical HF case. abc(2) .21).1).35) ˆ [2] . 1985).318 Systematic derivation of the coupled-cluster equations Section 7. but is now known frequently as CCSD[T] + . T2b on the T amplitudes. this is illustrated diagrammatically and algebraically by (2) (4) ET = (2) = i>j>k a>b>c tijk abc (2)∗ abc (2) abc tijk εijk . The factor εabc ijk is required to compensate for the fact that each carries one triple-excitation denominator. respectively. 5.37) = (2) and contracting it with its adjoint in all allowed ways.36). Also replacing the top vertex in (10.37) for the top vertex. The resulting expression can simply be added to the CCSD energy to provide an estimate of the CCSDT (or CCSDT-1) energy. ˆ3 amplitudes in (10. giving rise to the 16 MBPT triple-excitation antisymmetrized fourth-order energy diagrams (Fig. This is exactly the contribution that would be obtained by to deﬁne T 3 ˆ2 amplitudes and including the added adding these diagrams to the CCSD T terms in the energy formula (10. This approximation leaves out energy contributions ˆ3 from diagrams S7 and D10a . where the superscript [2] indicates a generalized order. D10c subject to the use of (10.

(10. In the HF case the order of the singles amplitudes becomes [2] instead of [1]. In this case the .41) The generalized orders (in superscript brackets) are for the non-HF case. This procedure includes all possible fourthorder contributions of single and double excitations subject to the simpliﬁed triples expression obtained from (10. An extension of CC4SD[T] incorporating the energy contributions from diagram S7 is known as CCSD(T) or CC4SD(T) (Raghavachari. the resulting energy can be written as ECC4SD(T) = ECCSD + ET = ECCSD + Et + Est + Edt . using a more complete terminology. (10. Trucks. Pople et al.39) Est = = 1 4 ti ia a [1]∗ jkbc [4] bc jk tijk abc [2] = ia ti a [1]∗ j>k b>c bc jk tijk abc [2] . and the Est term is of order (5) while the Edt term vanishes.38) = i>j>k a>b>c tijk abc [2]∗ abc [2] abc tijk εijk .10.40) Edt = = 1 4 tij ijab ab [1]∗ kc [4] × fkc tijk abc [2] = i>j a>b tij ab [1]∗ kc fkc tijk abc [2] . (10. Gauss and Bartlett 1993). Employing an approach similar to that used for the principal triples fourth-order energy contribution above in which lower-order singles and doubles amplitudes are replaced by their converged CCSD values. the brackets in [T] being used to indicate the limitation to the HF case. where Et = = 1 36 tijk ijkabc abc [2]∗ abc [2] abc tijk εijk [4] [4] [4] [4] [4] (10. provided that a semicanonical transformation is ﬁrst carried out to eliminate the oﬀ-diagonal parts of diagrams T2a and T2b (otherwise it would be necessary to iterate the evaluation of the triple-excitation amplitudes). doubles and triples equations 319 or. This notation implies inﬁnite order in S and D and fourth order in T.5 Coupled-cluster singles. 1989) and can be generalized for non-HF cases by adding diagram D10a (Watts.35). CC4SD[T].

as all such ˆ1 and T tributions to the T terms are already included in CCSDT-1b. The CC4SD(T) model is common in quantum chemistry applications. Furthermore. The triples conˆ2 contributions to the T includes all purely T ˆ2 equation do not change. Diagram T2a introduces an nh p step.3). While it is just one of many ﬁfth-order terms that could contribute to the CCSDT energy. see Figs. Thus a sequence of linear approximations can often overestimate the correlation energy. The CCSDT-2 model (Urban. All the approximations considered here for CCSDT are correct through fourth order in MBPT for both HF and non-HF cases. A characteristic of CC theory is that most linear terms. other strategies are recommended. since it requires storage of the triple-excitation amplitudes and since the remaining terms in CCSDT are nonlinear and can be factored. it is typically positive and serves to damp the sometimes excessive negative contribution of ET in diﬃcult cases. while containing many higher-order terms.6. doing so requires the eval3n 5 uation of diagrams T2a –T2e (Fig. 1985.8. Since these diagrams represent the 1 2 T2 term in (10. which implies a restriction to the HF case. they should be the most important nonlinear terms.30). this model ˆ3 equations. Cole et al. A systematic way to proceed would be to include all third-order terms in the triples equations (see Table 10. as can happen for CCSDT-1. adding diagrams T4a –T4d .5 and 10. culminating in the full CCSDT treatment. which led to the above noniterative approximations. like diagrams T1a . since it is inexpensive enough to be usable for a large number of problems while providing results that are usually as accurate as can be obtained by any modern quantum chemistry method. Comparisons of numerical results for the diﬀerent models will be presented later in Section 10. ˆ1 We can take this approach a step further by including all possible T ˆ and T2 contributions to the triples equation. 10. and is the only diﬀerence between CCSD[T] and CCSD(T). The CCSDT-1 model. For higher accuracy we can consider the successive addition of further terms in these equations. However. Noga. tend to produce negative-energy contributions. T1b and T2a –T2e . is usually lower in energy than CCD. includes all second-order terms in the triples equations. of course. That is why LCCD. the linearized version of CCD. 10. In our notation the (T) designation implies suitability to any type of reference function. in fact.5). which is one power of np higher than the costliest terms in CCSDT-1. in contrast with [T]. The computational cost of this approach is. . Noga. while the contributions of the nonlinear terms tend to be positive. almost as high as that of full CCSDT. Consequently. Bartlett and Urban 1987) is an attempt to rectify this problem by adding the quadratic terms T3a –T3e to the triples equations ˆ2 of CCSDT-1.320 Systematic derivation of the coupled-cluster equations Est term is small.

Cole et al. CCSDTQ is correct through sixth order in the energy and third order in the wave function. ˆ1 T +T . Noga. 10. Whereas CCSDT is correct through fourth order in the energy and second order in the wave function. the CCSDT-3 model beneﬁts from the complete inclusion of their contributions by providing additional ﬂexibility. order nh p All the iterative approximations to CCSDT can be characterized by the ˆ approximation to eT implied by each model in each of the singles. doubles. and neither model requires the storage of the T ˆ3 amplitudes. Approximations for eT used in the equations in the CCSDT-n models (Urban. triples and quadruples equations ˆ 321 Table 10. This choice deﬁnes the CCSDT-3 model (Urban.4.28) T ˆ2 . Compared with the CCSDT equations. Since the single excitations are instrumental in ensuring the insensitivity of CC theory to the choice of orbitals. (10.30) T CCSDT-1a eT1 +T2 +T3 ˆ3 eT1 +T2 + T ˆ2 T ˆ ˆ ˆ ˆ ˆ CCSDT-1b eT1 +T2 +T3 eT1 +T2 +T3 ˆ2 T ˆ ˆ ˆ ˆ ˆ ˆ CCSDT-2 eT1 +T2 +T3 eT1 +T2 +T3 eT2 ˆ ˆ ˆ ˆ ˆ ˆ ˆ CCSDT-3 eT1 +T2 +T3 eT1 +T2 +T3 eT1 +T2 ˆ ˆ ˆ ˆ ˆ ˆ ˆ ˆ T7a –T7d and T8a –T8e to those included in CCSDT-2. (10. Noga.29) T ˆ3 . a third-order term conThis model introduces the connected T tributing to the ﬁfth-order energy. 1985). (10. (10. Cole et al. triples and quadruples (CCSDTQ) equations The state-of-the art in CC theory currently includes the CCSDTQ model (Kucharski and Bartlett 1992). 1985. Because of the factorization of the nonlinear terms. ˆ4 cluster. doubles and triples equations.4. ˆ4 appear in the doubles and triples CC amplitude new terms involving T equations.42) (10. the computational cost of both the CCSDT-2 and CCSDT-3 approximations is no worse than 3 n 4 . and a quadruples amplitude equation is added: DCCSDT + TCCSDT + ab ˆ ˆ ij |(HN T4 )C |0 = 0.6 Coupled-cluster singles.43) abc ˆ ˆ ijk |[HN (T4 ˆ4 )]C |0 = 0. Bartlett and Urban 1987).10. as shown in Table 10. Equations ˆ1 .6 Coupled-cluster singles. doubles. Noga.

44) The new term in the doubles equation (10. 10. (10. respectively. respectively. Antisymmetrized Goldstone diagrams representing the T ˆ3 equation to the CCSDTQ T .43) are represented by the diagrams in Fig. we obtain the quadruples equation for the CCSDTQ-1 model (Kucharski and Bartlett 1989.42) is represented by = 1 4 mnef mn ef tabef ijmn . b.9) × T11a T11b T11c T12a T12b T12c ˆ4 contributions Fig.8.9. and e. 1998c) by taking the diagonal parts of diagrams Q2a and Q2b (Fig.322 Systematic derivation of the coupled-cluster equations abcd ˆ 1 ˆ2 1 ˆ2 1 ˆ3 ˆ ˆ ˆ ˆ ˆ ˆ ˆ ˆ ˆ ˆ ijkl |[HN (T3 + T4 + T1 T3 + T1 T4 + 2 T2 + T2 T3 + T2 T4 + 2 T3 ) + 3! T2 1 ˆ2 ˆ 1 ˆ ˆ2 1 ˆ3 ˆ 1 ˆ2 ˆ2 ˆ2 ˆ ˆ ˆ ˆ +1 2 T1 T3 + 2 T1 T4 + 2 T1 T2 + T1 T2 T3 + 3! T1 T3 + 4 T1 T2 ]|0 C = 0. (10. j. the resulting triples equation takes the form TCCSDT + − − − 1 fme tabce ijkm + 2 P (c/ab) cm ef tabef ijkm f abc ef te m tnijk me mef abcf 1 mn kf tijmn + mn 2 P (k/ij ) mnf mnef af bc 1 mn ef te i tmnjk 2 P (i/jk ) mnef ef bc 1 P ( a/bc ) mn ef ta m tinjk 2 mnef =0 for all i > j > k . c and i. 10. Assigning labels in the diagrams from left to right in the order a.45) D12 The new terms in the triples equation (10. k to the open particle and hole lines. 10. The diagrams representing the T Analogously to the derivation of the CCSDT-1 model. n to internal particle and hole lines. a > b > c. (10. f and m.8.46) ˆ4 equation are shown in Fig. 10.

Antisymmetrized Goldstone diagrams representing the CCSDTQ T equation. triples and quadruples equations × Q1a Q3a Q4c Q5a Q6d Q6j Q1b Q3b Q4d Q5b Q6e Q7a Q7f Q8d Q10b Q11c Q12c Q13d Q13i Q15a Q5c Q6f Q7b Q7g Q2a Q3c Q4e × Q6a Q6g Q2b × Q3d Q4f × Q6b Q6h Q7c Q8a Q9b Q10d Q11e Q13a Q13f Q14a Q15c Q6c Q6i Q7d Q8b Q9c Q11a Q12a Q13b Q13g Q4a Q4g × Q2c Q2d × Q4b Q4h Q2e 323 Q7e Q8c Q10a Q11b Q12b Q13c Q13h Q9a Q10c Q11d Q12d Q13e Q13j Q14b Q15b ˆ4 Fig.6 Coupled-cluster singles. 10.9.10. doubles. .

ˆac − P ˆad − P ˆbd − P ˆcd ˆ (a/bc/d) = 1 − P ˆab − P P ˆab P ˆac P ˆbd P ˆcd P ˆab P ˆac P ˆcd + P ˆbd + P ˆad + P ˆad + P ˆad + P ˆad . b. we need to include all distinct assignments of target labels to the lines originally labeled a and d. 10. me ˆ (abc/d|ij/kl) −P + 1 ˆ 2 P (a/bc/d|ij/kl) ˆ +1 2 P (i/jk/l|ab/cd) ˆ (ij/k/l|a/bc/d) −P (10. so that permutations between the labels of these lines should be omitˆ (ab/cd|ijk/l) should be read as ted. which are of third order. Thus. Using rule 10. ˆbd )(1 − P ˆac − P = (1 − P Similarly. d except for any permutations that exchange the labels of equivalent lines (initially b and c). +P In other words. The factors 1 2 in the third and fourth terms on the right-hand side of ˆ2 vertices (10. the “/” notation separates groups of equivalent lines. on the other side. these factors can be canceled with that arising from the permutation of labels a and d.47) In the permutation factors.1). Q1b and Q5a –Q5c .47) (diagrams Q5a and Q5b ) result from the fact that the two T in each of these diagrams are equivalent (interpretation rule 7. The resulting equation is abcd ˆ εabcd ijkl tijkl = P (ab/cd|ijk/l) e cd el tabe ijk md kl tabc ijm m fd bc ef tae ij tkl ef cd mn jk tab im tnl mn bc md el tae ij tmk . the factor P (a/bc/d) implies all permutations of a. resulting in the simpliﬁed form ˆ (ad/bc|ij/kl) P ef fd ˆ bc ef tae ij tkl + P (il/jk |ab/cd) mn cd mn jk tab im tnl . diagrams Q1a . Fig.324 Systematic derivation of the coupled-cluster equations to one side of the equation and retaining only the dominant terms. c. a permutation factor such as P follows: ˆ (ab/cd|ijk/l) = P ˆ (ab/cd)P ˆ (ijk/l) P ˆad − P ˆbc − P ˆbd + P ˆac P ˆil − P ˆjl − P ˆkl ) .

of (10. A higher-level CCSDTQ-2 approximation. Speciﬁcally. the eﬀect here is of fourth order. the to diagram Q1a ). includes the T 2 quadruples equation (Kucharski and Bartlett 1993). for the sum in the ﬁrst quadratic term. as discussed below.47) that the T ˆ2 termine its initial contribution to the energy. diagram Q5a . In contrast.47). as deﬁned by (10.20). However. To deIt is clear from (10. f. i > j > k > l. since one interaction vertex is involved in this diagram. which is correct ˆ 3 terms in the ˆ2 T ˆ3 and T through sixth order in the energy.45). in analogy to CCSDT-1. and. The CCSDTQ-1 model. When substituted in the energy equation (10. we can also introduce a noniterative approximation for the quadruple-excitation contributions to CCSDTQ (Kucharski and Bartlett 1989.6 Coupled-cluster singles.s. we note that it aﬀects the T amplitudes through diagram D12 .h. triples and quadruples equations 325 for these terms. and then evaluate ˆ Q5a = 1 2 P (d/abc|ij/kl) f 4 n 5 procedure.43) (Fig.47). it is more convenient to use the form in (10. the complete CCSDTQ model results in an ∼ nh p 4 n 6 procedure and requires the storage of ∼ n 4 n 4 T ˆ4 amplitudes. doubles. 10. the result is a ﬁfth-order contribution to the energy. resulting in an ∼ nh p quadratic terms in (10. We can obtain the ﬁfth-order energy contribution of abcf f d Sij tkl for all a > b > c > d.47) (diagrams Q5a –Q5c ) can be factored and so do not raise the computational cost. ˆ4 vertex arises in third order.8) result in energy contributions of sixth order for diagrams T11a –T11c and seventh order for diagrams T12a –T12c (eighth order in the HF case) and may thus be left out of CCSDTQ-1. It 4 n 4 amplitudes. we can ﬁrst sum over e.47) plus the inclusion of diagram D12 in the doubles equation. . is correct through ﬁfth order in the energy. As in other cases. In CCSDTQ-1 the to store the T ˆ T4 amplitudes can be calculated on the ﬂy during the processing of the ˆ3 amplitudes need to be stored. Note that the doubles equation.10. abcf ˆ (a/bc) =P Sij e bc ef tae ij for all a > b > c. (10. corresponding 4 n 5 procedure. but the T ˆ T4 contributions to the triples equation (10. each of which involves at requires the evaluation of ∼ nh p most np terms (it is the ﬁrst term on the r. an ∼ nh p h p In non-HF cases we employ semicanonical orbitals to avoid the need to include the oﬀ-diagonal part of diagrams Q2a and Q2b and thus the need ˆ4 amplitudes. to facilitate factorization of the sums. i > j. 1998c). Analogously to the noniterative approximations for the triple-excitation contributions to CCSDT.

∆EQ = f [5] 1 2 ˆ† T ˆ†(1) W ˆ2 ˆ (T ˆ3 + 1 T 0T 2 2 2 2) 0 C . Since this approximation is used as a noniterative addition to a CCSDT calculation. the ultimate rate-determining 3 n 5 process.326 Systematic derivation of the coupled-cluster equations the connected quadruple excitations by substituting diagram D12 . using the factorization techniques we get (5) ˆT ˆ(3) |0 ˆR ˆ0W ∆EQ = 0|W 4 C = (3) = 1 2 1 32 1 32 1 32 + (3) ij ab kl cd kl cd + ij ab εab εcd ij kl 1 1 + cd ab εij εkl ij ab kl cd tabcd ijkl (3) tabcd ijkl = abcd ijkl = abcd ijkl = abcd ijkl ij ab kl cd abcd abcd εijkl tijkl .20). which is an ∼ nh p CC5SDT(Qf ). we can write the result in the compact form ∆EQ = (5) 1 2 ˆ†(1) 0 T 2 2 ˆ2 ˆ (T ˆ3 + 1 T W 2 2) 0 C . The method is termed step is in CCSDT itself. then.49) As in the CCSD(T) case. instead of ∼ nh p 2 4 5 the ∼ nh np process of CCSDTQ-1. into the energy equation (10. indicating that it is ﬁfth order in quadruples and inﬁnite order in SDT. (10. we can assume that closing the diagram with ˆ† vertex instead of its ﬁrst-order approximation would a ﬁnal converged T 2 provide a superior approximation. (10. so it has the same .7.45). It does not change for non-HF cases.50) A very important feature of this approximation is that it requires only an 3 n 4 evaluation for the T ˆ3 part and roughly ∼ n6 for T 2 .47). (10. This is borne out by numerical studies (Kucharski and Bartlett 1998c) and alternative derivations based on the CC energy functional to be discussed in Section 12.48) Substituting for the last two factors from (10. The ﬁnal estimate is the Qf (factorized quadruples) correction. εab εcd ij kl (10.

we can couple (10.2). accepting the ap(10. Attempting to extend the Qf approach from the noniterative CCSDT(Qf ) contribution to the iterative CCSDTQ-1 model.45). which has a less important role in electron corˆ3 would depend on T of T relation than in orbital rotation (see Sections 12. It derives theory. Kutzelnigg 1977. The Qf approximations involve a factorization of a connected-cluster diagram that is reminiscent of the factorization of linked quadruple-excitation diagrams in the fourth-order MBPT energy (Section 7. The type of factorization introduced here is not intrinsic to coupled-cluster ˆ† terms in its deﬁning equations. A lower-level approximation. To reduce the computational effort of the CC5SDT(Qf ) model further.50) to one of the simpler CCSDT-n approximations. However. as was done for ∆EQ in (10. Similar ˆ4 . doubles.2 and 12. labeled CCSDTQf -1 (Kucharski and Bartlett 1998c). At that level we would formally have a CCSDT procedure ˆ6 operators. Arponen and Bishop 1991. incorporates the leading quadruple excitaˆ4 diagrams or tion eﬀect in the amplitude equations without requiring any T ˆ T4 equations. Van Voorhis and Head-Gordon 2000). the ˆ4 contribution to the double-excitation amplitude equation is replaced T 1 ˆ2 ab ˆ † ˆ ˆ by 1 2 ij |T2 [W (T3 + 2 T2 )]|0 C . the ˆ6 connected-cluster contributions to same approach can be applied to the T ˆ the T3 equation. These approximations also indicate the potential for augmentation of the standard CC ˆ4 operator. is obtained by adding 1 ab ˆ † ˆ ˆ 2 4 ij |T2 W T2 |0 C to the CCSD doubles equation. the factorization ˆ3 but. of the T ˆT ˆ4 ]C |0 in diagram proximate nature of this factorization and replacing [W (5) 1 ˆ† ˆ ˆ 1 ˆ2 D12 by { 2 T2 [W (T3 + 2 T2 )]}C |0 . ˆ4 and T with implicit contributions from the connected T The iterative CCSDTQf -1 model and the noniterative CCSDT(Qf ) = CC5SDT(Qf ) and CCSD(TQf ) = CC4SD(TQf ) approximations have been applied to prototypical problems such as the vibrational frequencies of O3 . ˆ2 amplitude equation. CCSDQf .50). 1989. triples and quadruples equations 327 invariance properties as CC theory and the generalized CC4SD(T) of the previous section. Nevertheless. we soon encounter the ˆ4 energy diagram in (10. which does not have T from simpliﬁcations that are inherent in perturbation theory.6). the expectation value CC (XCC) and the extended CC Ans¨ atze (Bartlett.6 Coupled-cluster singles. Kucharski.10.48) canproblem that the factorization of the T not be applied strictly to the corresponding wave-function diagram D12 . unlike T approaches could possibly be applied to T ˆ1 . such approach through alternative methods that involve the T as the unitary CC (UCC). related to Wigner’s 2n + 1 rule (subsection 2. thus approximately incorˆ4 contribution without including any T ˆ3 or T ˆ4 diagrams porating a leading T or equations.3). The resulting model. Noga et al.2.

This expansion separates H into zerobody. The CC eﬀective Hamiltonian can be expanded in terms of contributions with diﬀerent numbers of creation and annihilation operators: 1 ˆ2 ˆ2 ˆ ˆ ˆ1 + T ˆ2 + .52) as follows: i† a ˆ} ≡ χia {ˆ = ×+ . Here the constant term χ0 = ∆E corresponds to the P given diagrammatically in (10.51) can be separated into parts involving diﬀerent allocations of hole and particle indices. For an N -electron system the expansion terminates with the N -body term. 10. one-body. two-body etc. for the one-body term we have χpq {p ˆ† q ˆ} = pq ab χab {a ˆ †ˆ b} + ai i} + χai {a ˆ†ˆ ai χia {ˆ i† a ˆ} + ij j} .51) † † + pqrstu ˆs χpqrstu {p ˆ† q ˆ† r ˆ† u ˆt ˆ} + · · · . These diagrams also provide valuable guidance for tonian H = (H the eﬃcient reuse of intermediates in the calculations.53) i a i a i a . . (10. for which an accuracy of 1 cm−1 relative to full CI was achieved ˆ5 operator had been included after contributions from the fourth-order T (Musial. + 1 T H = HN (1 + T 2 1 + T1 T2 + 2 T2 + · · · ) C = χ0 + pq χpq {p ˆq ˆ} + pqrs † χpqrs {p ˆq ˆs ˆr ˆ} (10. we can diagram the various terms of the one-body contribution (10. Thus.20). . Kucharski and Bartlett 2000).328 Systematic derivation of the coupled-cluster equations (Kucharski and Bartlett 1999) and the harmonic vibrational frequencies of N2 and C2 .52) Using wavy lines to represent the H vertices. χij {ˆ i† ˆ (10. Each term in (10.7 Coupled-cluster eﬀective-Hamiltonian diagrams A major reduction in the number of diagrams for the CC amplitude equations can be achieved by using diagrams representing the CC eﬀective Hamilˆ ˆ N eT )C . ˆ HP ˆ contribution. terms.

the CC singles equation can be written as a i =0 (for all a. An added beneﬁt of the use of these intermediates is the conversion of the nonlinear equations into a pseudolinear equation in which each term contains an integral or intermediate multiplied by a single t amplitude. . This substitution is an example of the use of intermediate quantities in the calculations and provides an eﬃcient iterative strategy for the solution of the CC equations.2 and (10.55) ˆ†ˆ χai {a i} ≡ a i = a i ×+ a a i +a a i ×+ i i a ×+ a i i + i + a + i + i i ×+ a a + a + a i + a i × + a + a i i . (10. is exactly the set that appears in the CC singles equations. b = b b (10.56) The second lines in equations (10.31). (10.57) Thus. The set of diagrams in the diagrammatic expansion of χai .55) utilize the previously determined χia .7 Coupled-cluster eﬀective-Hamiltonian diagrams 329 χab {a ˆ †ˆ b} ≡ a b = a ×+ b a ×+ b a b a b + a + a b ×+ a + a b + a . In fact. replacing the third and ﬁfth diagrams in each case by a single eﬀective-Hamiltonian diagram. i = i i (10. (10.54) j} ≡ χij {ˆ i† ˆ j i j ×+ i j ×+ i j i j i = + j + j i ×+ j + j i + j .10. this matrix element of H represents the entire equation. 10. i).54) and (10. comprising Fig.56).

i†ˆ χij {ˆ j} ≡ j i = j ×+ i j i + j i . eleventh and thirteenth diagrams of (10. tenth.55). the fourth represents the sum of the fourth. which add up to a i .59) If we used both these diagrams in the expansion we would be including twice the tenth and eleventh diagrams.56) and the seventh represents the sum of the ninth and fourteenth diagrams of (10. This representation provides a compact pseudolinear form for the CC singles equation.56) we run into a complication: the third. sixth. To avoid this overcounting we deﬁne an intermediate. ﬁfth. (10. (10.56) as χij {ˆ i† ˆ j} ≡ a i i a j i = j i + j i . tenth.56). sixth. (10.62) the third diagram represents the sum of the third. eleventh and twelfth diagrams add up to i . (10.62) In (10.330 Systematic derivation of the coupled-cluster equations When we try to use intermediates from (10.55) in the expansion (10.63) . eleventh and thirteenth diagrams add up to i a . The twobody matrix elements and intermediates are obtained in a similar manner: ba ˆ} ≡ i†ˆ j †ˆ χijab {ˆ i a b j = i a b j (10. tenth.61) i} ≡ ˆ†ˆ χai {a = a i ×+ a a i + a i + i + i + a + a i + a i .60) Using this intermediate we can rewrite (10.53)–(10. (10.58) a The fourth. ﬁfth and twelfth diagrams of (10. (10.56).

70) i j l cd b + a cd b .64) χaibc {a ˆ†ˆ ˆˆ b} ≡ i† c c i = c i +a b c i .10. (10. (10.67) ˆc ˆ †ˆ b† d ˆ} ≡ χabcd {a b a = d c = a c b a + d c b a + d c d d b + a b.66) k a = k +j a i k a .7 Coupled-cluster eﬀective-Hamiltonian diagrams 331 a i† c χaibc {a ˆ†ˆ ˆˆ b} ≡ b a b c i = a b a b j i j i c i + 1 a 2 b c i .65) j ˆ† a χikja {ˆ i† k ˆˆ j} ≡ i j ˆ† a χikja {ˆ i† k ˆˆ j} ≡ i a c k a = k 1 + j a 2 i k a . (10.69) ˆ} ≡ j †ˆ χijkl {ˆ i† ˆ lk = = k i k i k i l j l k + j i l k + j i j j l + k l + k i j i j l + k l . (10. (10. cd b (10.68) ˆc χabcd {a ˆ †ˆ b† d ˆ} ≡ = = a c a c a c b d b a + d c b + a d d cd b + a b. (10.

73) 1 i 2 a b j j †ˆ χajib {a ˆ† ˆ bˆ i} ≡ i a b j = i a b j +a a b i j + i b + i a b j + i = i a b + 1 i 2 a b j a b . j (10.332 Systematic derivation of the coupled-cluster equations χajib {a ˆ† ˆ bˆ i} ≡ j †ˆ = = i i i a b a b a b j j j + + 1 a 2 1 a 2 i j i j b b + i + i a b a b j j + .71) 1 i 2 a b j j †ˆ ˆ† ˆ bˆ i} ≡ χajib {a = = i i i a b a b a b j j j + + 1 a 2 1 a 2 i j i j b b + + 1 i 2 1 i 2 a b a b j j + .74) j j + i a b j . (10. (10.72) 1 i 4 a b j ˆ† ˆ bˆ i} ≡ χajib {a j †ˆ = = i i i a b a b a b j j j +a +a i j i j b b + + 1 i 2 1 i 2 a b a b j j + . (10.

77) .76) ˆˆ j} ≡ i† a ˆ† k χiajk {ˆ j i k a = j i k a + 1 j k 2 i a.10.75) ic ˆ} ≡ χabci {a ˆ †ˆ b†ˆ = a c a c +a +a b i b i + a c b i + a c ab i b + a c a b c b i i i c i c b c b i + b +× b + c i + c b c ab i + a c b i +a a c + a c i +a a c b i +a i c b i = i + i c a c b + c c b ab i. (10.7 Coupled-cluster eﬀective-Hamiltonian diagrams 333 ic ˆ} ≡ χabci {a ˆ †ˆ b†ˆ a c b i = a c + b i + 1 a 2 c b i + 1 a 2 a c i c i b 1 a b 2 c 1 a 4 b b c i + i+ b + a c b i i a c i = + 1 a 2 c b. i b i+a b +a (10. (10.

10. including the use of permutation .334 Systematic derivation of the coupled-cluster equations j ˆˆ j} ≡ χiajk {ˆ i† a ˆ† k i = j i +j +j +j = j i +j k k a a a i i i k j k i k+ j i j a i k k i j k i a i k i k a + a+ k+ a a+j a+j a+ j k a a+ j a i k a.1.78) k+ j i k a j k + i a i k+j k a i j ˆˆ i† a ˆ† k j} ≡ χiajk {ˆ i = j i +j +j +j = j i k k a a a i a i k i j k i k+× k+ a+j + i i k i a. Fig.79) j a i j k i k k a a + a+ k+ j i a jk jk a + a + j a+j i i k a k a i jk The interpretation rules for these diagrams are the same as those for the ordinary CC diagrams. (10. (10.

(10. and should add up to zero when the CC equations are satisﬁed. For example. for inequivalent open-line labels. 9.10. inˆ (kl) would stead of the intermediate χikja . but labels on lines open at the top are never permuted with labels of lines open at the bottom.2. according operators P to rules 8 and 9. then the permutation operator P have introduced the third diagram of the second line in two equivalent forms. if we had used the full diagrams with equivalent T expression for χikja in the second diagram in the third line of (10.45).80) ij ab This form is untruncated. it condenses to the pseudolinear form ˆ †ˆ b† ˆ jˆ i} ≡ χabij {a = a a i j i j b b ab ab b ij ij b j b j +a i j + i ab a ij + i + i +a j +a j b +a + i b + i +a j +a b ij b . Using the various matrix elements and intermediates. the ﬁnal two-body matrix element χabij expands to a sum of all the CC doubleexcitation diagrams. the permutation operator ˆ (ab) = 1 − P ˆab is applied to the second diagram for χabcd in the second P line of (10. (10.69).4 and (10.70). Figs.56). The intermediates χaibc etc. 10.8 plus one diagram involving T . Obviously. 10. but in the third diagram the same operator cancels with the factor due to the pair of equivalent vertices. which is equal to the sum of all the diagrams for the tripleˆ5 ). can excitation equations (Figs.7 Coupled-cluster eﬀective-Hamiltonian diagrams 335 ˆ (ij · · · |ab · · · ) etc. Like the case of the ﬁnal one-body matrix element χai . For example. and no additional double-excitation diagrams would appear at any level of CC theory. the last three diagrams would not be included in a CCSD calculation and the last diagram would be omitted in CCSDT. The corresponding untruncated three-body matrix element.3. 10. are needed to avoid overcounting and to obtain the correct weights for ˆ vertices.5 and 10.

including the number of eﬀective-Hamiltonian diagrams. jk bc c k jk jk c (10. and the last diagram would be left out in CCSDTQ.336 Systematic derivation of the coupled-cluster equations be expressed in pseudolinear form in terms of the same eﬀective-Hamiltonian matrix elements and intermediates as ˆˆ ˆ †ˆ ˆ† k b† c jˆ i} ≡ χabcijk {a a i j bc k c bc c c c k = i +a +a + i + i + i ab i j i j ab ab ab jk +a k+ i k +a i j ab i j ab ij bc jk bc c k b jk + i +a . Musial.81) The last four diagrams would not be included a CCSDT calculation.82) bc j + a d b ij . To represent the quadruple-excitation equation in a pseudolinear form we deﬁne four three-body intermediates: ˆ} ≡ ˆ †ˆ ˆ†ˆ b† c jˆ id χabcdij {a = a d 1 a 2 d + a b b ij c ij i d c + a d b ij c c . This approach was extended to CCSDTQ by Kucharski and Bartlett ˆ5 ) by Musial. (10. Kucharski and Bartlett (1992) and to CCSDTQP (including T (2002a). Kucharski and Bartlett (2002b) described the general diagrammatic structure of the CC equations and obtained algebraic formulas for the number of diagrams of each type at each level.

Using these intermediates. (10. the untruncated four-body matrix element that deﬁnes the quadruple-excitation equation. which are presented later in this section. 10. is obtained as: ˆ5 or T Fig.10.84) ˆˆ χijaklm {ˆ i† ˆ ˆ† k lm ˆ} ≡ j†a lm j a = k i j lm a. representing all the diagrams in ˆ6 . kl b (10.85) These intermediates are parts of the complete three-body matrix elements of H.7 Coupled-cluster eﬀective-Hamiltonian diagrams 337 ˆˆ χiabjkl {ˆ i† a ˆ †ˆ b† ˆ lk j} ≡ = j k i 1 j k 2 i + j i ab l ab l +j k i a b a i l k ab l+ j l.9 plus diagrams containing T ˆ† ˆ ˆˆ b† c lk jˆ i} ≡ ˆ †ˆ ˆ† d χabcdijkl {a =a + i + i + i + i + i i j ab ab ab ab ab b c jk jk j c jk jk a i j d cd d d l bc k l d l kl + i l +a l + i ab i j ab i j ij ab jk bc jk bc c cd kl cd kl kl d l d d c kl cd c l +a +a d+ i b jk cd l . (10.83) + j i b a k ab ˆˆ χibcajk {ˆ i†ˆ b† ˆ b† k ja ˆ} ≡ jk c i k i = i a a kl b.

(10.88) ˆˆ χibcajk {ˆ i†ˆ ˆ† k b† c ja ˆ} ≡ = i b a j a jk c b a c. The three-body elements are: ˆc ˆ† ˆ b†ˆ χajbcdi {a j †ˆ id ˆ} ≡ a c b j d i = ab i . ab jk cd l l (10.87) j dc ˆ} ≡ k χijakbl {ˆ j†a i† ˆ ˆ† ˆ lˆ bk i a j b l = b j i k l a . because vertex. (10. we also present below the diagrams for the complete three-body and some four-body matrix elements of H.89) ˆ} ≡ a χabcdei {a ie ˆd ˆ †ˆ ˆ†ˆ b† c d = a d e b c e bc i i + a de bc i.86) For later use in Chapters 11 and 13.338 Systematic derivation of the coupled-cluster equations +a + i i j ab bc jk k l cd d + i .90) k ˆˆ lm ˆ} ≡ i†ˆ ˆ† k χijaklm {ˆ j†a i = k i j lm j lm a a+ k i j lm a. Note that there can be no H-matrix elements with four or more lines below the = . except for the single two-body element ˆ T -vertex lines can only be open at the top. (10. (10.91) . i + bc k+ i jk c i a b jk (10.

92) + a bc j + a d b ij j k ˆˆ χiabjkl {ˆ i† a ˆ †ˆ b† ˆ lk j} ≡ i = j k i + i ab l ab l +j k i l + j i a b l jk a i ab k ab l l .93) + j kl b + j i k ab Only three four-body vertices will be needed.7 Coupled-cluster eﬀective-Hamiltonian diagrams 339 ˆ} ≡ a jˆ id χabcdij {a ˆ †ˆ ˆ†ˆ b† c d = a d + b b ij c ij c +a b d j + a d i j c d ai i d bc b ij c c. They are: ˆ †ˆ ˆ} ≡ a χakbcdeij {a ˆ† k ˆ†ˆ b† c jˆ ie ˆd d k b e ij c = k ed ab ij c. (10.94) ˆ} ≡ k b† m i†ˆ ˆ †ˆ ˆˆ lc ˆk χijabkclm {ˆ j†a i l j c ab m = j c i kl ab m.10. (10. (10. (10.95) .

it is still expected to provide realistic indicators of the comparative merits of various approximations. .340 Systematic derivation of the coupled-cluster equations ˆ† ˆ ˆˆ χibcdajkl {ˆ i†ˆ ˆ† d b† c lk ja ˆ} ≡ = i b a j a jk cd l b a l + i + bc kl d+ cd i jk b a cd jk l i a b jk i cd l. the Born–Oppenheimer approximation and relativistic eﬀects. While this type of basis is not suﬃcient for high-accuracy results. Full-CI calculations require all possible excitations among any orbitals that are not frozen. since full CI represents the exact results within the Hilbert space generated by the chosen basis set. only a comparatively few full-CI calculations are available for use as benchmarks and almost none of these use basis sets better than polarized double-zeta (DZP).96) 10. The full-CI energy is a variational upper bound to the exact energy. Thus the full-CI results are solely determined by the basis set. (10. Taking the example of H2 O with the inner-shell electrons on the oxygen atom frozen. a DZP basis contains a minimum of 23 functions for the eight valence electrons. Not only do such comparisons separate correlation-method contributions from basisset eﬀects. is exactly extensive and is invariant under the choice of the molecular orbitals spanning the basis-set space. For N electrons and n orbitals. In fact. The most unambiguous assessment of the quality of the diﬀerent models is provided by comparison with full-CI results using the same basis set.8 Results of various CC methods compared with full CI It is very helpful to have some numerical data on the relative contributions of the various cluster operators and on the eﬀects of the various approximations used in introducing correlation corrections. Consequently. the total number of determinants in a full-CI expansion is given asymptotically by ∼ nN . but they also avoid the questions of experimental uncertainties. corrections for zero-point vibrations. comparisons with full CI are more useful for the assessment of the eﬀectiveness of various correlation methods than comparisons with experimental data.

but not quite negligible since CCSDTQ reduces the error by a factor 30 compared with CISDTQ. We can see the magnitude of this eﬀect by comparing CID with CCD. Furthermore.0Re .5Re and 2. we ﬁnd that the mean absolute error for CCSDT is about 1 mEh . The diﬀerence . MBPT models can be viewed as ﬁnite-order approximations to various levels of CC theory. making the correlation error much greater at stretched geometries than at Re . As pointed out previously. when single excitations are added the improvement in the mean absolute error of CCSD relative to CISD is a factor 3.10. When we add quadruple excitations. is not extensive until the full-CI limit is reached. are not variational. unlike CI. MBPT and CC for DZP (doublezeta + polarization) basis sets at various levels of correlation treatment for the FH and H2 O molecules at their equilibrium bond length Re and at 1. eventually accounting for most of the correlation energy as the size of the system increases. Unlike MBPT. leaving the eight valence electrons to be correlated. since the restricted Hartree–Fock reference function does not separate correctly to neutral atoms. the unlinked diagram error in CISDTQ would be much larger than in the examples presented here. to 0. For much larger molecules (or assemblies of molecules). the CISDTQ model achieves a dramatic reduction in the error compared to CISDT. since it now includes the disconnected quadruple excitation terms that serve to eliminate the most important unlinked term (see subsection 5. the principal diﬀerence between CI methods on the one hand and MBPT and CC methods on the other is that the CI energy retains unlinked diagram contributions and. As conﬁrmed by the presence of some negative-energy errors the MBPT and CC methods.4). both these methods are of inﬁnite order but the mean absolute error of CCD for the examples in Table 10.7.05 mEh . but this deﬁciency is a small price to pay for extensivity. The inner-shell pair of electrons on the heavier atom is frozen in these calculations.8 Results of various CC methods compared with full CI 341 In Table 10.5 results are given for CI. Since the MBPT and CC theories are built upon the linked-diagram theorem. these methods beneﬁt from the elimination of non-extensive terms at the outset.5 is a factor 2 better than that of CID. When triple-excitations are added. The stretched geometries place great demands upon RHF-based single-reference treatments. consequently. there are no bounds on any energy diﬀerences of interest in chemistry and physics. while CISDT is still 22 mEh above full CI on average. As seen from the diagrammatic expansions. and therefore the advantage of a variational method is of lesser consequence than the extensivity property of the many-body methods for most practical applications. The part that remains is small for these small examples. While variational methods provide an upper bound to the total energy.

03 15.23 3.10 −1.4 13.9 5.5 4.76 3.03 0.01 Geometries.0Re Mean abs.80 5.0Re Re H2 O 1.0 27.81 −0.05 0.6 11.8 9.0 12. Musial. Energy errors (in millihartrees).36 0. error Conﬁguration interactionb CID CISD CISDT CISDTQ CISDTQP MBPT(2) MBPT(3) MBPT(4)SDQ MBPT(4) MBPT(5) MBPT(6) 10.75 −1.44 2.40 0.00 18. While MBPT rapidly improves with increasing order relative to CI because of the absence of unlinked diagrams.7 12.98 1.28 0.76 4.70 0.13 5.65 1.03 40.65 0.29 32.01 1.6 14.09 1. d Kucharski and Bartlett (1992). basis sets and FCI energies are from Bauschlicher.84 8.01 0. its errors are substantially greater than those for the comparable CC models.05 0.02 0.6 34.66 Coupled clusterd CCD CCSD CCSD(T) CCSDT-1 CCSDT CC5SDT[Q] CC5SDTQ-1 CCSDTQ CCSDTQP a 5.41 8.11 0.0 7.5.06 0.02 0.29 −0.9 10.4 23.38 7.7 74.17 0.40 0.5 1.01 4.2 21.6 0.3 5.9 10.77 2. c Kucharski and Bartlett (1995).2 −0.06 22.92 0.6 19.02 −0.00 15. many-body perturbation theory and coupled-cluster models for FH and H2 O in a DZP basis with frozen core at three bond lengthsa FH Re 1. for truncated CI calculations.47 −0.02 1.60 0.1 25.2 2. Taylor et al.75 −0. relative to full CI.22 1.49 0.06 0.58 5.12 0.342 Systematic derivation of the coupled-cluster equations Table 10.16 34.49 0.26 0.63 −2.14 0.02 0.5 27.9 10.22 4.5Re 2.06 0.82 53.28 MBPTc 24.08 23.8 75.3 9.13 21.2 14.1 4. b K´ allay and Surj´ an (2000.40 0.5Re 2.11 0.50 1.78 −0.39 0.0 4.53 0.08 7.2 0.10 0.4 4. 2001). For pairs of methods that are accu- .29 13.35 0.72 0. Langhoﬀ.6 60.58 −0.14 0. Kucharski and Bartlett (2002a).05 0.27 0.9 17.3 6.0 1.99 1.92 0.26 0. between these approaches is that inﬁnite order is used for selected terms in CC theory whereas a given ﬁnite order is used for all terms in MBPT.88 0.22 0.9 28.00 13.00 35.30 0.4 12.01 0.65 −2.5 22.13 0.16 10.55 84. (1986) and Bauschlicher and Taylor (1986).9 11.5 30.04 0.3 26.1 0.

6.486 4. Only in sixth order does MBPT(6) narrow the gap with CCSDTQ.002 21.396 72. Energy errors (in millihartrees).134 36.698 −165. the CC models give results that are several millihartrees better for the examples in Table 10.229 0. using RHF and UHF reference functions.4288 1.014 −113.5.10.014 0.011 3.866 1.700 10.926 3.558 2.087 3.182 6.030 −15. Bond length (in ˚ A) Method 1.913 1. all other results are from Chan. The RHF and UHF wave functions are identical at this point. from RHF-based CCSD.021 0.819 0.8 Results of various CC methods compared with full CI 343 Table 10.2700 1.834 0. K´ allay and Gauss (2004) and Musial and Bartlett (2005).469 1. for coupled-cluster calculations in a cc-pVDZ basis and frozen core for N2 at several bond lengths. The iterative form of this approximation. rate to the same order of perturbation theory.123 1.348 0.144 0.839 0.948 3.744 9. CCSD(T) tends to retain the inﬁniteorder advantages of CCSD versus MBPT(4)SDQ.376 3.651 0.1208b 1.503 10. b Equilibrium bond length. CCSD(T) and CCSDT results.866 1.353 0.923 0.5 40.2225 A Krogh and Olsen (2001).202 21. relative to full CI.964 1. The comparative closeness of the results for these two models is the reason why noniterative or perturbative corrections added to an underlying CC model are expected to be eﬀective.548 0. This diﬀerence holds also for the models that are correct through ﬁfth order.035 6.9050 2.572 31.981 12.349 0. as may be seen by comparing MBPT(5) with CCSDTQ-1.839 0.630 6.5875 1.131 UHF reference function 22.368 45. CCSDT-1.188 7.223 −6.141 0. In particular.481 3.295 0.807 5. while the perturbative triples estimate in CCSD(T) is consistent with the improvement of MBPT(4) relative to MBPT(4)SDQ. is a little better for energies but can overcorrect in diﬃcult cases .827 ˚.229 0.088 7.351 20.438 0.002 14.997 6. such as MBPT(3) and CCD.a .038 1.469 1.631 0.310 −1.2225 RHF reference function CCSD CCSD(T) CCSDT CCSDT(Qf ) CCSDTQ CCSDTQP CCSDTQPH CCSD CCSD(T) CCSDT CCSDTQ CCSDTQP CCSDTQPH a 14. MBPT(4)SDQ and CCSD and MBPT(4) and CCSDT. except for R = 2.073 31.968 15.341 2.021 0.

Several examples of this type of behavior are shown elsewhere (Bartlett 1989). For example. on their own they are an inadequate measure of the quality of an approximation.5 reﬂect the increased sophistication of the methods quite well. Also included in this table are RHF-based CCSDT(Qf ) results (Musial and Bartlett 2005). It is seen in both the table and the potential energy plots that the CCSD(T) and CCSDT(Qf ) models are good approximations for CCSDT and CCSDTQ. 2002a). though they still fail eventually as the bonds are stretched further since then the reference function becomes a very poor initial approximation. Higher-order results were .10. Musial and Bartlett 2005). The ﬁrst set of results in this table is based on an RHF reference function and deteriorates badly as the bond is stretched. just beyond the equilibrium bond length. it produces relatively poor results with high spin contamination at intermediate distances and a discontinuity in the energy derivatives at the point at which the RHF and UHF reference functions start to diﬀer from each other. i. Using a UHF reference function eliminates the long-range problems. The (T) approximation is no better but. 10. For CCSDTQ and higher levels the iterative procedures converge very slowly or fail to converge at all beyond the equilibrium bond length. even with approximations like CCSDT-1. however. the correction does not become excessive.6 (Krogh and Olsen 2001. The results of such calculations for the nitrogen molecule in a cc-pVDZ basis (Dunning 1989) at several bond lengths are shown in Table 10. potential-energy surfaces behave diﬀerently at large values of the internuclear distances and in the vicinity of the equilibrium geometry. respectively.e. triple-excitation amplitudes are not allowed to change the T ˆ2 amplitudes. A case that is particularly demanding for single-reference methods is the description of the stretching and breaking of a triple bond. and T Although the energy results in Table 10. 2. K´ allay and Gauss 2004. do much better than MBPT at large R. and perturbation approximations fail for the stretched geometries if the reference function does not separate correctly.118a0 . because it is purely perˆ1 turbative. The inﬁnite-order CC methods.344 Systematic derivation of the coupled-cluster equations since it is fundamentally a linear approximation for the triple-excitations eﬀect. producing energies far below the full-CI values. Potential energy curves for some RHF-based CC results are shown in Fig. Kucharski and Bartlett (2000. which show increasing deviations from the CCSDTQ values as the bond is stretched. along with results for other properties beside the energy that are pertinent to quantum-chemical applications. The coupled-cluster techniques were extended to ﬁve-fold excitations by Musial. at the shorter bond length but deteriorate as the bonds get longer. Chan.

1 1.9 345 −109.8 Results of various CC methods compared with full CI −108.9 2.2 1.9 1. 10.0 2.10.0 E (Eh ) −109.2 CCSD CCSD(T) CCSDT CCSDT(Qf ) CCSDTQ Full CI −109.3 0.5 R (˚ A) 1.0 1.1 Fig.10.6 1.4 1.8 1. .7 1.3 1. Potential energy curves for N2 with RHF reference.1 −109.

Some of these include automated optimization of the computational procedures. including a choice of intermediates.346 Systematic derivation of the coupled-cluster equations obtained by non-diagrammatic methods for some small examples for which the full-CI Hamiltonian matrix was available in a determinantal representation (Hirata and Bartlett. Hirata 2003. 2006). K´ allay and Surj´ an 2000. 2000). Bernholdt et al. . Auer. Li and Paldus 1994. Methods for the automated generation of formulas and computer programs for high-order coupled cluster calculation have also been reported (Janssen and Schaefer 1991. Baumgartner. 2001. 2004.

the use of an expression involving a denominator can be inconvenient for some applications. the expansions for eT and eT in the numerator and denomiˆ operators nator terminate when the total excitation level in the product of T in any term exceeds the number of electrons in the CC wave function. ˆ operators.1) When |Ψ = eT |0 . Thus. such as those in Hartree– ˆ ) of an operator O ˆ is Fock or CI. in contrast with the behavior of the expansions in the CC amplitude equations. this expression gives rise to series expansions in the numerator and denominator: O= = ˆ† ˆ T ˆ |0 0|eT Oe ˆ 0|eT † eT |0 1 1 ˆ 1 + T + 1 T2 + 1 T3 + ··· 0 (T † )2 + 3! (T † )3 + · · · O 0 1 + T † + 2! 2! 3! .2) ˆ† ˆ ˆ Individually. The expectation value O (also written O given by O= ˆ ˆ |Ψ Ψ|O . Ψ|Ψ (11. ˆ2 + 1 T ˆ3 + · · · 0 ˆ† + 1 (T ˆ† )2 + 1 (T ˆ† )3 + · · · 1 + T ˆ+ 1T 0 1+T 2! 3! 2! 3! (11. Furthermore. the number of terms in each expansion and the computational eﬀort required could be very high.1 Expectation value for a CC wave function The exponential form of the coupled-cluster wave function requires an approach to the calculation of properties other than the energy that is diﬀerent from that for more conventional wave functions. 347 . even which always terminate after the product of four T if we evaluated the numerator and denominator separately and then divide the results.11 Calculation of properties in coupled-cluster theory 11.

(11.348 Calculation of properties in coupled-cluster theory In fact. The T ˆ2 + · · · . but products of T ˆ operators produce ˆ1 + T a sum of connected terms T ˆ T disconnected terms and thus the factor e |0 contains both connected and disconnected (but not unlinked) diagrams.7) .6) ˆ† ˆ ˆ ˆ† T 0|T 2 1 T1 T2 |0 = = 1 4 ijab ∗ ab tab ij tij + ∗ c tc k tk + kc ∗ b ∗ ab c tac ij tk tij tk ijkabc + ∗ ac ∗ ab c tb j tik tij tk ijkabc 1 2 ijkabc ∗ c ∗ ab c tab ik tj tij tk .4) ˆ 2 ˆ† 1 (T 0|T 2 2! 1 ) |0 = = 1 2 ijab ∗ a b tab ij ti tj . 0|(T 1 2 i1 i2 k1 !k2 ! · · · l1 !l2 ! · · · (11. ˆ ˆ† T 0|T 2 2 |0 = = 1 4 ijab ∗ ab tab ij tij .5) ˆ† ˆ ˆ† T 0|T 2 1 T3 |0 = = 1 4 ijkabc ∗ c ∗ abc tab ij tk tijk . + − 1 2 − (11. (11. For this purpose we examine each expanˆ operator is sion diagrammatically. To take several examples. beginning with the denominator. division by the denominator can be carried out formally. (11. The products arising from the ˆ† ˆ expansion of 0|eT eT |0 take the general form 1 ˆ† )m T ˆn |0 0|(T m!n! = 0<i1 =i2 =··· 0<j1 =j2 =··· k1 +k2 +···=m l1 +l2 +···=n k1 i1 +k2 i2 +··· =l1 j1 +l2 j2 +··· 1 ˆ† )k1 (T ˆ† )k2 · · · (T ˆj )l1 (T ˆj )l2 · · · |0 . by factorˇ ıˇ ing the numerator (C´ zek 1969).3) ˆ† factors resulting in closed diagrams in which the de-excitations in the T ˆ cancel the excitations in the T factors.

If ˆ= O pq p|o ˆ|q p ˆ† q ˆ= pq opq p ˆ† q ˆ = pq ˆ† q opq {p ˆ} + p ˆ† q ˆ opq {p ˆ† q ˆ} + pq i = ˆ N + 0|O ˆ |0 oii = O (11. Equal labels on lines ˆi vertex produce zero contributions because ˆ † or T connected to the same T i ab.1 Expectation value for a CC wave function 349 Analogs of the above diagrams are obtained for the numerator. its expectation value can be split into a reference value and a correlation correction.. amplitudes in the upper and lower indices.10) (11. of the antisymmetry of the tij. (11. it is easily seen that terms with k = j cancel between the ﬁrst and fourth diagrams and between the second and third diagrams.11) as an example.2) will contain terms like ˆ† T ˆ† ˆ ˆ 0|T 2 1 ON T2 |0 = (a) (i)(j ) (k ) (c) (b) ♦ + (j ) (c) + (a) = 1 4 ijab (b)(a) (i) (k ) (c) ♦ (c) ♦ (i)(j ) (b) (k ) ♦ + (i) (k ) (a)(b) (j ) ∗ ab tab ij tij kc 1 2 ijkabc ∗ tc k ock + ijkabc ∗ ac ∗ ab tb j tik tij ock 1 2 ijkabc ∗ c ∗ ab tab ik tj tij ock . respectively. O = 0|O ˆ N represented diagrammatically by With the normal-ordered part O ˆN = O ♦ + ♦ + ♦ + ♦. ˆ |0 + ON = Oref + ∆O .. There are two types of EPV contribution in these diagrams.11. − ∗ b ∗ ab tac ij tk tij ock − (11.8) is a one-particle operator. Similarly. and thus such terms can be included freely in the summations.9) the numerator of (11..11) Factorization of the numerator requires independent summation over all indices. Taking (11. thus introducing exclusion-principle-violating (EPV) terms. terms with c = b cancel between the ﬁrst and third and between the second and fourth . All other EPV terms can be shown to cancel..

The diﬀerences in the weights of these terms are accounted for by line equivalences in the respective diagrams.3)) in the product of T number of electrons N are necessarily EPV terms. The complete diagrammatic expansion of the numerator will contain both connected and disconnected closed diagrams. Thus we can safely ignore the termination of the expansion by excitation level. Each part containing the O the expansion of the numerator with all possible combinations of parts that do not contain this operator and thus it can be written as a product of the former part times the expansion of the denominator. In the expansion of the numerator. Such terms either vanish (because of the repetition of indices in any amplitude) or mutually cancel.12) from terminating. thus k = i is equivalent to k = j in the ﬁrst diagram. k = i is also equivalent to k = j in the third diagram. producing the ˇ ıˇ result (ﬁrst given in C´ zek 1969) ON = ˆ† ˆ ˆ T 0|eT O N e |0 ˆ ˆ 0|eT † eT |0 ˆ† ˆ ˆ N eT |0 = 0|eT O C. canceled between connected and disconnected terms. terms in which the total excitation ˆj factors exceeds the level l1 j1 + l2 j2 + · · · (see (11. The ﬁrst factor in this expression cancels the denominator. evaluation of the expectation value in this form requires trunca- . providing two identical contributions with weight − 1 2 each to cancel a single contribution with weight 1 in the second diagram. Therefore the entire numerator can be written as the product of the sum of its connected terms multiplied by the denominator. providing two identical contributions with 1 weight 1 4 each to cancel a single contribution with weight − 2 in the fourth term.350 Calculation of properties in coupled-cluster theory diagrams. ˆ† ˆ ˆ† ˆ ˆ T 0|eT eT |0 0|eT O N e |0 C. Similarly.12) The ﬁnal (connected) result retains EPV terms that. Because of the inclusion of EPV terms and the related non-terminating form of the diagrammatic expansion. As a result. since there are no more than N distinct hole states to excite. These uncanceled EPV terms prevent the expansion (11. (11. each disconnected diagram is simply equal to the product of the individual ˆ N operator will appear in connected parts in it. each of which has exactly one ˆ N and any number (zero or more) connected part containing the operator O of parts that do not contain this operator. The resulting nonterminating form of the expansion is necessary for complete factorization of the numerator. in the original numerator. since multiple excitations or de-excitations of the same hole state may now occur.

. holds. such as S ˆ . not just for ˆ N is the normal-product one-electron operators. As an alternative. (11. the CC equations do not always produce a spin eigenfunction. ∆E = 0|H N e |0 . ∆E = = = ˆ ˆ ˆ† ˆ ˆ T 0|eT eT |0 0|e−T H N e |0 ˆ† T 0|eT e ˆ |0 (11. This follows because the CC equations require that Q ˆ The equivalence is exact only when T is the solution of the untruncated CC ˆ2 .14) ˆ H|0 = 0. such as Qe than for a closed-shell system with doubly occupied orbitals. a spin-adapted formulation of CC theory based on the unitary group approach was given by Li and Paldus (1994. When the operator O ˆ − 0|H ˆ |0 .13). (10. (11. thus Q For properties represented by the expectation values of any operators for ˆz . On another level. ˆN = H Hamiltonian H ˆ† ˆ ˆ T 0|eT H N e |0 ˆ† T 0|eT e ˆ |0 ˆ† ˆ ˆ T = 0|eT H N e |0 ∆E = C. The result (11. making it necessary to impose that condition when this aspect is important (Szalay and Gauss 1997. such as T3 . If T ˆ CC equations for higher cluster operators. 1994. the ˆ is truncated to a particular excitation level. this result is consistent with the normally used CC expression ˆ ˆ ˆ ˆ T ˆ N eT |0 C = 0|e−T H (10. 2000). as in the expansion of (1 + A)−1 . Urban and Hubaˇ c 1992. provided that a corresponding analog have the same equations as for H ˆ ˆ2 T ˆ ˆ −T S e |0 = 0. say T ˆ1 + T equations. ˆ ˆ ˆ ˆ )e−T +Q =ˆ 1 We can show the equivalence of these forms by inserting eT (P into the expectation-value form: ˆ† ˆ ˆ T 0|eT H N e |0 ˆ† T 0|eT e ˆ |0 ˆ ˆ ˆ† ˆ ˆ ˆ ˆ )e−T +Q HN eT |0 0|eT eT (P ˆ† T 0|eT e ˆ |0 ˆ ˆ ˆ T = 0|e−T H N e |0 = 0|H|0 . 1997).1 Expectation value for a CC wave function 351 tion at some level and prevents exact determination of the expectation value.11. Other of the second equation in (10. the nonterminating nature of the result may be explained as due to the implicit series expansion of the inverse of the denominator. are not satisﬁed and ˆ H|0 = 0 is not strictly valid.11) for the correlation energy. we formally ˆ2 or S which the wave function is an eigenfunction.13) ˆ operator is the exact solution of the untruncated CC Provided that the T equations.13). Other approaches to the at least approximate imposition of spin conditions have also been used (Neogr´ ady.12) gives the correlation energy.12) is valid for all normal-ordered operators.

x2 . (11. xN )Ψ∗ (x . . xN )dx2 dx3 · · · dxN (11. .2. . γqp = ˆ|Ψ p ˆΨ|q ˆΨ Ψ|p ˆ† q = Ψ|Ψ Ψ|Ψ (11. in the second line. a general route toward their evaluation can be provided by introducing a density matrix. Stanton and Gauss 1994. x3 . (11. x )|x =x dx o ˆ(x)φq (x)γqp φ∗ p (x )|x =x dx = pq (11. . . .18) In terms of the continuous form of the density matrix this result can be obtained as ˆ = O = tr Oγ = pq o ˆ(x)γ (x. .8) is given by O= pq opq 0|eT p ˆ† q ˆeT |0 ˆ† ˆ C = pq opq γqp . For expectation values of other operators. For the particular case of CC theory. Neogr´ ady and Hubaˇ c 1997). where each of x. x2 . x ) in the one-electron (spinorbital) basis. . x3 .2 Reduced density matrices The evaluation of the expectation values of one-electron operators is facilitated by the use of the one-body reduced density matrix γ (x. . . Bartlett 1995.17) The expectation value of the one-electron operator (11. stands for the set of space and spin coordinates of an electron and.16) are the elements of the discrete matrix representation γ of the continuous matrix γ (x.15) φq (x)γqp φ∗ p (x ) . . such as those representing dipole and quadrupole moments. x ) = = pq N Ψ|Ψ Ψ(x. γqp = ˆ† q ˆeT |0 0|eT p ˆ† T 0eT e ˆ |0 ˆ† ˆ ˆ† q ˆeT |0 = 0|eT p ˆ† ˆ C. Urban. . as will now be described in Section 11.352 Calculation of properties in coupled-cluster theory Jayatilaka and Lee 1993. electric ﬁeld gradients or the kinetic energy.19) opq γqp . 11. x2 .

where Γrspq = ˆ† s ˆr ˆ|Ψ Ψ|p ˆ† q q ˆp ˆΨ|s ˆr ˆΨ = Ψ|Ψ Ψ|Ψ (11.11. x2 .2 Reduced density matrices 353 where it is understood that the substitution x = x is carried out after the operation of o ˆ(x) on γ (x. (11. x2 . . x3 . x2 . . x ) and before the integration over x.x2 .20) and the total electron population is obtained by integrating over x .60)) we obtain the expectation value G= 1 4 pqrs pq |g ˆ|rs A Γrspq . For a two-electron operator ˆ= G 1 4 pqrs pq |g ˆ|rs ˆ† q ˆ† s ˆr ˆ Ap (11. . xN )dx3 · · · dxN (11.25) .53).19) this operator gives γ (x ) ≡ γ (x . x )|x =x dx φq (x )γqp φ∗ p (x ) . (11. (3. .23) are the elements of the discrete representation Γ of the continuous matrix Γ(x1 . . Γ(x1 . . . In particular. x1 . xN )Ψ∗ (x1 . x3 . γ (x )dx = pq φp |φq γpq = p γpp = N . the electron density at some space–spin point x is obtained as the expectation value of the Dirac δ -function operator δ (x − x). . x ) = = pq δ (x − x)γ (x.22) ∗ φr (x1 )φs (x2 )Γrspq φ∗ p (x1 )φq (x2 ) . x2 ) in the one-electron (spinorbital) basis. x1 . (11. x2 ) N (N − 1) = 2 Ψ|Ψ = pqrs Ψ(x1 .24) (compare (3.21) For the expectation value of two-electron operators (including the energy) we need the two-body density matrix. When inserted into (11.

∗ ∗ ∗ Γrspq = −Γrsqp = −Γsrpq = Γsrqp = Γ∗ pqrs = −Γpqsr = −Γqprs = Γqpsr (11. (11. Γcdab = (ΓN )cdab . (11. for diﬀerent combinations of particles and holes. Γrspq has the same symmetry properties with respect to permutation of the indices as antisymmetrized two-electron integrals.30) (11.28) The only diﬀerence between the elements of γ N and γ is for the hole–hole elements.32) . with elements ˆ† q ˆ}eT |0 (γ N )qp = 0|eT {p ˆ† ˆ C.27) where γ N is the normal-ordered part of the one-body density matrix.29) When used in the evaluation of the expectation value of a one-electron operˆ . Γcdij = (ΓN )cdij . Γckij = (ΓN )ckij + δkj (γN )ci − δki (γN )cj . For the purposes of the diagrammatic representation it is convenient to limit our attention to the correlation correction of the expectation value. (11. using the generalized Wick’s theorem we obtain Γrspq = (ΓN )rspq + δpi δri (γN )sq + δqi δsi (γN )rp − δpi δsi (γN )rq − δqi δri (γN )sp + δpi δri δqj δsj − δpi δsi δqj δrj . this diﬀerence provides the vacuum-expectation-value part 0|O ˆ |0 = ator O i oii of the total expectation value (11. where i. (11. Γcjai = (ΓN )cjai + δij (γN )ca . j are any hole-state labels and (ΓN )rspq = 0|eT {p ˆ† q ˆ† s ˆr ˆ}eT |0 ˆ† ˆ C. Γklij = (ΓN )klij + δki (γN )lj + δlj (γN )ki − δkj (γN )li − δli (γN )kj + δki δlj − δkj δli .26) and Γrspp = Γrrpq = Γrrpp = 0. ON = pq opq 0|eT {p ˆ† q ˆ}eT |0 ˆ† ˆ C = pq opq (γ N )qp .354 Calculation of properties in coupled-cluster theory As can be seen from (11.31) Speciﬁcally.9). (11.23). For the two-body density matrix. ˆ† † ˆ j −ˆ i†ˆ j eT |0 (γ N )ji = 0|eT ˆ iˆ C = γji − δji . Γcdai = (ΓN )cdai .

34) p . to the three terms G N In the diagrammatic representation of contributions to the one-body density matrix (11. such as ♦. corresponds to multiplication by the corresponding property integral opq and summation over p.173). which apply to ΓN as well as to Γ. (11.28) we need a symbol for the operator {p ˆ† q ˆ}. q (but obviously the density matrix elements need to be calculated only once for any number of one-electron properties). The three sums in this equation correspond ˆ N.11. It is to be understood that the two lines connected to the vertex carry ﬁxed labels that identify the speciﬁc matrix element of γ and are not summed over. in that order. In the evaluation of the expression (11. In a common notation. a gap is left in place of this vertex. The expectation value G then becomes G= 1 4 pqrs pq |g ˆ|rs 1 4 ipq 1 2 ij A (ΓN )rspq + + = 1 4 pi|g ˆ|qi ij |g ˆ|ij pq |g ˆ|rs A A + ip|g ˆ|iq A − ip|g ˆ|qi A − pi|g ˆ|iq A (γN )qp A (ΓN )rspq + pq gpq (γN )qp + 1 2 ij ij |g ˆ|ij A. G ˆ and 0|G ˆ |0 of (3. Here we will represent this operator by the markerless . Adding a marker. The indices on γqp correspond to the labels on the incoming and outgoing lines at the vertex.26). respectively.8).33) pqrs where gpq is given by (3.174). sometimes a dotted line is added across the gap or the vertex is marked with the label 1 → 1.28) we need to contract the operaˆ† ˆ ˆ with operators in the expansions of 0|eT at the top and eT |0 tors p ˆ† and q at the bottom. Symbolically. we can represent this process by the schematic diagram q (γ N )qp = (11. which may be construed as a place-holder for one-electron vertex the operator representing the actual property being calculated. which represents the generic (property-independent) part of the general one-electron operator expression (11.2 Reduced density matrices 355 All other cases can be obtained by index permutations according to the rules in (11.

4. It is convenient to classify the various contributions to the density matrices according to the perturbation order in which they ﬁrst arise. ˆ† operators at the top and the vertex has to appear between the T the ˆ operators at the bottom. (11. More speciﬁcally. In terms of the time sequence. the T . d (ΓN )akij = ˆ† Γidab = i a eT ˆ k j etc. eT (11. As shown in ˆm operators can be resolved into MBPT wave-function Section 9. left–out. j (ΓN )ibaj = eT a i j eT ˆ b . Symbolically. eT ˆ eT eT a i eT ˆ ˆ† ˆ eT b . ˆ† ˆ† ˆ† ˆ† eT γba = eT a b . d (ΓN )klij = eT k i l .35) ˆ eT ˆ The {p ˆ† q ˆ† s ˆr ˆ} operator for the two-body density matrix will be represented . right–out.356 Calculation of properties in coupled-cluster theory ˆ† ˆ where the box represents the 0|eT and eT |0 parts and the slopes of the p and q lines have no signiﬁcance in this case. (γ N )ji = j i eT ˆ . the ﬁrst index T on γqp represents an outgoing line in the diagram relative to the respective box (incoming relative to the vertex) and the second index represents an incoming line relative to the box. the density matrix elements (γ N )qp fall into four types.. depending on whether each of p and q represents a particle or hole label.36) the indices on Γrspq corresponding to the labels on the lines at the vertex in the order left–in. The corresponding schematic diagrams by the double-dashed vertex for the matrix elements of ΓN take the forms ˆ† ˆ† ˆ† Γcdab = eT a c b . As for any one-electron quantity. γia = eT a i eT ˆ eT . right–in. γai = i a .

T3 is second order and so on. 3–7 are third order and 8–19 are fourth order. It is important i a 1 i 2 a ia 3 4 i a 5 ia 6 ai i 7 a 8 ia 9 i a i 10 a ia 11 ai 12 13 i a ia 14 15 i a i 16 a 17 i a ia 18 ai 19 Fig.2. ˆ operators ˆ† and T Diagrams representing contributions from the various T to the particle–hole and particle–particle elements of the one-body density matrix through fourth order in the general (non-HF) case are shown in Figs. which are conjugate (Section 5. Diagram 1 corresponds ˆ† to the leading term (i. Therefore the hole–particle diagrams. T by the lowest PT order in which it arises. 11.e. The hole–hole diagrams are obtained from the particle–particle diagrams by the reversal of all arrows. we need to calculate only Since the density matrix is Hermitian. Looking at Fig. diagrams 20 and 21 are second order. the number 1) in the series expansion of eT . γpq = γqp one of every pair of complex-conjugate oﬀ-diagonal elements. 11. 22 and 23 are third order and 24–41 are fourth order. ∗ . 11. are not shown.1.5) to the particle– hole diagrams. Arrows are shown explicitly only when diﬀerent distinct arrow assignments are possible.1 and 11. . respectively. In the general case diagram 1 is ﬁrst order.2.11. Diagrams representing contributions to the particle–hole element γai of the one-body reduced density matrix through fourth order.2 Reduced density matrices 357 ˆm operator can be characterized components of diﬀerent orders and each T ˆ1 ˆ2 operator is ﬁrst order. diagram 2 is second order. The T ˆ is second order in the HF case and ﬁrst order otherwise.

11. (26. Diagrams representing contributions to the particle–particle element γba of the one-body reduced density matrix through fourth order. 31) in Fig. 28).1 and 11. 23). 11.2 are not conjugate to each other. Thus these diagram pairs represent diﬀerent contributions to the same matrix element. because the ﬁxed labels move with their lines in the conjugation process. 29) and (30.2 (the order of the terms in these expressions corresponds to the order of the diagrams in the ﬁgures): γai = ta i + jb ∗ ba tb j tji − jb 1 4 jkbc 1 2 jkbc ∗ ca b tcb kj tkj ti + jkbc ∗ b∗ c ba tc k tj tj tki jkbc ∗ b a tb j ti tj + jkbc ∗ cba tcb kj tkji − jkbc ∗ b∗ cb a tc k tj tki tj ∗ c ba tcb kj tk tji − 1 2 jkbc ∗ c b a tcb kj tk ti tj + jkbc ∗ b∗ ca b tc k tj tkj ti jkbc ∗ b∗ cba tc k tj tkji ∗ cb a tcb kj tki tj − − 1 2 − − 1 2 + 1 4 jklbcd ∗ dc ba tdcb lkj tlk tji . to note that each of the particle–particle diagram pairs (22.358 a b 20 Calculation of properties in coupled-cluster theory a b 21 a b 26 a b 30 a b 34 a b 38 39 35 a b 31 a b 27 a b 22 a b 28 a b 32 a b 36 a b 40 41 33 a b 37 a b a b 23 a b 24 a b 29 a b a b 25 a b a b Fig. (27.2. The ﬁrst few terms in the algebraic expressions for the density matrix elements are easily obtained from Fig. 11.

(11.38) (ΓN )ji = − + kab 1 2 kab ∗ ba tba kj tki − kab ∗ b a tba kj tk ti − kab ∗ a∗ ba tb k tj tki ∗ c ba tcba lkj tl tki ∗ a∗ a b tb k tj tk ti 1 4 klabc − 1 24 klabc ∗ cba tcba lkj tlki 1 2 klabc − 1 2 klabc − − ∗ cb b tcba lki tlk ti ∗ ba∗ cb a tc l tkj tlk ti − − ∗ ba∗ cba tc l tkj tlki ∗ a∗ c ba tcb lk tj tl tki − − 1 4 klabc ∗ a∗ cba tcb lk tj tlki ∗ ac∗ ba c tb k tjl tki tl klabc klabc ∗ ba∗ b ca tc l tkj tl tki klabc ∗ ba∗ c ba tc l tkj tk tli + klabc + 1 2 klabc + 1 2 klabc ∗ a∗ ca b tcb lk tj tlk ti + + + 1 2 klabc 1 2 ∗ a∗ cb a tcb lk tj tli tk + 1 2 klabc ∗ a∗ cb a tcb lj tk tlk ti ∗ ba∗ db ca tdc ml tkj tml tki klmabcd ∗ ad∗ cb ad tcb lk tjm tli tkm klmabcd + − 1 4 klmabcd ∗ ba∗ dc ba tdc ml tkj tmk tli ∗ ba∗ da cb tdc ml tkj tmi tjk + · · · 1 4 (11.11.39) klmabcd .37) jklbcd γba = i ∗ b ta i ti + 1 2 ijc ∗ cb tca ji tji + ijc ∗ c b tca ji tj ti + ijc ∗ d cb tdca kji tk tji ∗ a∗ cb tc j ti tji − ijc ∗ dc b tdca kji tkj ti ijkcd ∗ ca∗ dc b td k tji tkj ti ijkcd ∗ ca∗ d cb td k tji tj tki ijkcd 1 2 ijkcd 1 4 ijklcde 1 4 ∗ a∗ c b tc j ti ti tj + + + 1 24 ijkcd 1 2 ijkcd ∗ dcb tdca kji tkji + 1 2 ijkcd 1 4 + ∗ ca∗ dcb td k tji tkji ∗ a∗ d cb tdc kj ti tk tji + ∗ a∗ dcb tdc kj ti tkji ijkcd ∗ ad∗ cb d tc j tik tji tk + + ijkcd 1 2 − ijkcd − − − 1 2 ijkcd 1 2 ijkcd 1 4 ∗ ca∗ c db td k tji tk tji − ∗ c∗ dc b tda kj ti tkj ti − ∗ a∗ dc b tdc kj ti tki tj − ijkcd ∗ a∗ db c tdc kj ti tkj ti ∗ ca∗ ec db ted lk tji tlk tji 1 2 ijklcde ∗ ca∗ ed cb ted lk tji tlj tki − ∗ ae∗ db ce tdc kj til tkj til + ijklcde ∗ a ta j ti − a ijklcde ∗ ca∗ eb dc ted lk tji tli tkj + · · · .2 Reduced density matrices 359 − − 1 4 jklbcd 1 2 jklbcd ∗ cb ad tcbd kjl tki tjl + jklbcd ∗ b∗ db ca tdc lk tj tlk tji ∗ cb∗ dc ba td j tkj tlk tji − 1 2 jklbcd 1 2 jklbcd ∗ b∗ dc ba tdc lk tj tlj tki − ∗ d∗ cb ad tcb kj tl tki tjl − 1 2 ∗ d∗ ca bd tcb kj tl tkj til + · · · . (11.

The MBBPT diagrams representing contributions to the particle–hole and particle–particle element of the one-body reduced density matrix through second order.3.3.1 and 11. 11.e. i.93)–(9. CCSD[T] and CCSD(T). 11. this diagram vanishes since fia = fai = 0 in this case.360 Calculation of properties in coupled-cluster theory a i a × i × a i a × i a i i a i × a × a b × × × × i a × a b Fig. and since this 7 requires the third-order component of T component is missing in the CCSDT-1 model (Section 10. the eighth originates from diagram 2 and the last two originate from diagrams 20 and 21. respectively.2 and thus can be evaluated in a CCSD calculation.3 provides the only ﬁrst-order contribution to the density matrix. The ﬁrst seven diagrams originate in the second-order MBPT expansion of diagram 1 of Fig. 11. this model produces a density matrix that is only accurate through third order. As an example. also provide density matrices correct through third order.5).2 except diagram 7 require the second-order comˆ3 operator to achieve fourth-order accuracy.2. All the diagrams that contain a T vertex in Figs. 1 The coeﬃcient 24 for diagram 25 in the latter two equations derives from 1 a factor (3!) for a set of three equivalent lines times a factor 1 2 for each of a pair of equivalent lines. we Using the MBPT expansions (9. All the ˆ2 cluster operators (and ˆ1 and T diagrams in this ﬁgure originate in the T their adjoints) in Figs. Only the second. with many higher-order contributions included. 11. 11. ˆm operators.1. and therefore the CCSDT-derived density ter operator in these ﬁgures is T ˆ3 matrix is correct through fourth order. third and last diagrams survive in the HF case. for the T can generate the MBPT expansion for the density-matrix elements to any order. the diagrams for this expansion for the particle–hole and particle–particle elements to second order are shown in Fig.1 and 11. The only other clusˆ3 . The ﬁrst diagram in Fig. When a Hartree–Fock reference function (canonical or otherwise) is used. . Hence the CCSD-derived density matrix is correct to second order. of Fig. The ﬁrst eight diagrams in this ﬁgure represent γai and the last two represent γba . 11. the noniterative approximations to CCSDT-1. Similarly. Diagram ponent only of the T ˆ3 for this purpose.95) etc. 11.

cannot be represented simply by an expectation value but require a response treatment. λ=0 (11. with H = V . in which the perturbation ˆ =H ˆ 0 + λV the usual perturbation expression H (1) (2) (3) ˆ ˆ ˆ ˆ is linear in λ.40) . The following sections describe the alternative. A Hamiltonian that depends on a parameter λ can be expanded in a Maclaurin series in λ: ˆ ˆ 1 d2 H ˆ (λ) = H ˆ (0) + λ dH + λ2 2 + ··· H dλ λ=0 2 dλ λ=0 ˆ (2) + · · · . .41) ˆ (0) = H ˆ (0) is the usual molecular Here the unperturbed Hamiltonian H Hamiltonian. treatment for property evaluation. The CC wave function and the diagrammatic expansion of the density matrices can be used eﬀectively to evaluate expectation values to any desired order of perturbation theory. Bartlett 1986). which is based on treating properties as responses to perturbations. H = H = · · · = 0. such as derivatives of the energy with respect to displacements of the nuclei (energy gradients. and the parameter λ represents the strength of a perturbation (such as an electric or magnetic ﬁeld or the displacement of a nucleus). including electron correlation.3 The response treatment of properties 361 A diagrammatic treatment of the two-body density matrix will be deferred to Section 11. This expansion can be seen as a generalization of ˆ . response.). ˆ (1) + λ2 H ˆ (0) + λH =H where n ˆ ˆ (n) = 1 d H H n! dλn (11. This approach (Noga and Urban 1988) has only rarely been used but is sometimes a useful alternative to the deﬁnitive method discussed in the following sections. 11. The response treatment has become the standard approach to this problem. Hessians etc.11.3 The response treatment of properties As an alternative to the use of expectation values. In fact. it is possible to evaluate properties in terms of the response of the molecular system to a suitable perturbation (Monkhorst 1977. Jørgensen and Simons 1983. pending the development of more eﬀective tools for its evaluation.7. The nonterminating nature of the expansions and consequent lack of a closed-form expression and need for truncation are seen as serious disadvantages for the expectationvalue approach from a formal viewpoint. many molecular properties.

which is a manifestation of the Hellmann–Feynman theorem. H (11. da = −∂E (E )/∂ Ea |E=0 (a = x. When these expansions are substituted into the Schr¨ odinger equation and terms of ﬁrst order in λ are collected.44) = E (1) = dλ λ=0 Ψ(0) |Ψ(0) ˆ (1) .362 Calculation of properties in coupled-cluster theory The energies corresponding to the Hamiltonian (11. The numerical diﬀerences between the expectation value and the energy derivative at several levels of truncated CC were examined for the dipole moment of BH by Noga and Urban (1988). z ). . Equation (11. can be obtained as the the property operator O derivative of the energy with respect to λ at λ = 0. The ﬁrst derivatives of the perturbed energy E (E ) with respect to the components of the ﬁeld strength provide the components of the permanent dipole mo(0) ment vector.40) can also be expanded in λ: E (λ) = E (0) + λ dE dλ 1 d2 E + λ2 2 2 dλ λ=0 + ··· λ=0 = E (0) + λE (1) + λ2 E (2) + · · · . (11. This result is consistent with the expectation-value form d(0) = Ψ| µ qµ rµ |Ψ / Ψ|Ψ . the perturbation operator is given by the dot product −E · µ qµ rµ . is exact only for a variationally optimized |Ψ(0) and thus holds only for full (untruncated) CC. αab = −∂ 2 E (E )/∂ Ea ∂ Eb |E=0 .43) ˆ (0) = Ψ(0) |E (0) . The conceptual diﬀerence between an expectation value and the equivalent energy derivative is quite important in MBPT and CC theory. This form also allows the calculation of properties using energies that do not necessarily correspond to a wave function. because the derivative form leads to closed-form expressions for properties. where qµ and rµ are the charges and position vectors of the particles in the system. we obtain the ﬁrst-order equation ˆ (0) − E (0) |Ψ(1) + H ˆ (1) − E (1) |Ψ(0) = 0 . (11. When the perturbation is due to an electric ﬁeld E . The second derivatives provide the components of the dipole polarizability tensor. such as those obtained in CCSD(T). and higher derivatives provide the respective hyperpolarizabilities. which is the expectation value of so that the expectation value of H ˆ when H ˆ (λ) = H ˆ (0) + λO ˆ . Properties obtained as nth derivatives of the perturbed energy are referred to as nth-order properties. Projecting this equation onto Ψ(0) | and noting that Ψ(0) |H we ﬁnd ˆ (1) |Ψ(0) dE (λ) Ψ(0) |H .42) with a similar expansion for the wave function. y.44).

3 The response treatment of properties 363 The derivatives can also be evaluated by ﬁnite diﬀerence methods. can be written in the form ˆ=P ˆ H(λ)P ˆ. P ≡ ∆E λ P (11.51). (11. (11. ˆλ H + e−T HN e + H T = −T where ˆ ˆλ T ˆ H[λ] ≡ e−T H Ne . leading to a much more convenient expression for the derivative .49) (11. as described in the following analysis.48) (11. ∆E (λ) = ∆E (0) + λ∆O . T ˆλ ]P ˆ. is more accurate and more eﬃcient particularly when many perturbations have to be considered.51) dλ ˆλ of the perturbed amplitudes can be eliminated from The derivatives T (11. using CC calculations at small positive and negative values of λ.45) ˆ (λ) = H ˆ (0)+ λO ˆ . see (10.50) The superscript λ indicates the derivative with respect to λ. we shall concentrate here on the evaluation of ∆E (λ).46) (11. but analytical evaluation. ˆ ˆ ˆ T Using the CC eﬀective Hamiltonian H = e−T H N e .13). The energy derivative is obtained as d∆E ˆ ˆ=P ˆ H[λ] P ˆ+P ˆ [H. ˆ (λ) = 0|H ˆ (λ)|0 + H ˆ N (λ) = Eref (λ) + H H the corresponding perturbed correlated energy is given by E (λ) = Eref (λ) + ∆E (λ) . ˆ N (λ) . the perturbed CC correlation energy. Splitting the perturbed correlated Hamiltonian into its normal-product part and its vacuum expectation value. using (11.47) The evaluation of Eref (λ) is considered in Section 11. (11. and we have ˆ commute.9) the components of the above equation When H can be written as Eref (λ) = Eref (0) + λOref .8. T ˆλ ] . as in the case of energy gradient calculations. ∆E (λ)P The derivative of H(λ) with respect to λ can be obtained as dH d −T ˆ ˆ ˆ T e H = Ne dλ dλ ˆ ˆλ T ˆ ˆλ = H[λ] + [H. we have H[λ] = e−T O Ne .11. For a ˆλ and T used the fact that the excitation operators in T ˆ ˆ ˆ T ˆ ˆ ˆ linear perturbation. with H (λ) = H (0) + λO.

ˆ=P ˆ H[λ] (0)P ˆ+P ˆ H(0)R(0)H[λ] (0)P ˆ. ∆E λ P (11. T ˆ λ ]P P ˆ+P ˆ HQ ˆT ˆλ P ˆ − ∆E P ˆT ˆλ P ˆ−0=P ˆ HQ ˆT ˆλ P ˆ. ∆E (λ) P When λ = 0 this equation provides the ﬁrst-order energy. ˆT ˆλ P =Q The perturbed amplitudes are then obtained as ˆT ˆλ P ˆ=Q ˆ ∆E 0 − Q ˆ HQ ˆ Q −1 (11.54) ˆ H[λ] P ˆ. Λ(λ) = P (11. T ˆ λ ]P ˆ = 0. ˆ H(λ)R(λ)Q ˆ. the second term of (11.59) (11. taking the derivative of the amplitude equations (10. Trucks and Bartlett 1989). Q (11. Salter. Inserting P 1 and using P ˆ HP ˆ = 0.55) Inserting this expression into the perturbed correlation energy (11. ∆E (λ) − H(λ) (11.53). in which T Deﬁning an eﬀective resolvent operator R(λ) based on H(λ).364 Calculation of properties in coupled-cluster theory ˆ+Q ˆ =ˆ ˆ HP ˆ = ∆E P ˆ.53) Similarly.57) ˆ HRQ ˆ contains no derivatives (and is thus inFinally. ˆ≡ ˆ [∆E (λ) − Q ˆ H(λ)Q ˆ ] −1 Q R(λ) = Q we have ˆ=P ˆ H[λ] P ˆ+P ˆ HRH[λ] P ˆ. recognizing that P dependent of the perturbation at the point λ = 0).58) ˆ Q . the second term becomes ˆ−Q ˆT ˆλ (P ˆ [H.51) becomes Q ˆ−P ˆT ˆλ (P ˆ=P ˆ H(P ˆ+Q ˆ )T ˆλ P ˆ+Q ˆ )HP ˆ ˆ [H. T ˆλ ]P ˆ=Q ˆ H(P ˆ+Q ˆ )T ˆλ P ˆ+Q ˆ )HP ˆ Q ˆ+Q ˆ HQ ˆT ˆλ P ˆ−T ˆλ P ˆ HP ˆ−T ˆλ Q ˆ HP ˆ ˆ HP ˆT ˆλ P =Q ˆ Q ˆ HQ ˆ − ∆E 0 Q ˆ. we have ˆ=P ˆ H[λ] P ˆ+P ˆ HQ ˆ ∆E0 − Q ˆ HQ ˆ ∆E λ P −1 ˆ H[λ] P ˆ. Q (11. ∆E (1) P (11.56) ˆλ has been eliminated. we introduce a new operator (Bartlett 1986.13) gives ˆ H[λ] P ˆ+Q ˆ [H . of the correlation energy. ˆT ˆλ P = ∆E P (11. Q ˆ+Q ˆ=ˆ Inserting P 1.52) and the perturbed correlation energy takes the form ˆ=P ˆ H[λ] P ˆ+P ˆ HQ ˆT ˆλ P ˆ.60) .

which is Integrating this expression over λ we obtain ∆E P the fundamental energy functional of CC theory (Section 11.5 below.53). Thus this procedure is applicable for any of the CC models discussed in Chapter 10 and for any order of perturbation theory. Scuseria. Laidig and Bartlett 1984.3 The response treatment of properties 365 so that we can write ˆ=P ˆ (1 + Λ)H[λ] P ˆ. If one had to compute all 3N T amplitudes.11. Byers-Brown and Epstein 1964). Gauss and Bartlett 1992. these equations can equally well be applied for methods such as CCSDT-1. 1987. this derivation does not depend upon Q for all possible excitations.61) ˆ=P ˆ (1 + Λ)HP ˆ . 1993). since correspondence between a wave function and an energy is not required in this analysis.4). Trucks and Bartlett 1989). as in the case of analytical gradients of potential-energy surfaces when about 3N independent atomic displacements for the N atoms in a ˆλ perturbed molecule need to be considered. it is independent of the perturbation and needs to be solved only once for any number of perturbations. It should be emphasized that Λ(0) needs to be computed once only. Because this system of equations is solved at λ = 0. the ﬁrst-order properties can be obtained in closed form. Unlike the equivalence shown between the transition and expectationˆ HP ˆ=0 value forms for the energy. because it is to be replaced by a solution of a system of simultaneous linear equations for Λ. Once Λ is known. . ∆E λ P (11. The signiﬁcance of this aspect is obvious when there are many perturbations. rather. coupled-cluster theory would not have become the exceptional tool it now is for applications to molecular potential-energy surfaces. but they need some modiﬁcation for two-step methods such as CCSD(T) (Lee and Rendell 1991. this ˆ for any number ˆ H[λ] P is followed by the evaluation of its dot product with Q of properties. the derivation of (11. Furthermore. a manifestation of the interchange theorem familiar from double perturbation theory (Sternheimer and Foley 1953. (11. is what makes analytical derivatives viable in the non-variational CC theory (Adamowicz. This procedure.57) will not be required. Scheiner. rather than evaluating the dot products of a single Λ with the set of Hλ derivatives. Bartlett 1986. Hirschfelder. Inversion of the denominator in (11. Rice et al. Salter. as described in Section 11. Dalgarno and Stewart 1958.55) remains ˆ as long as the same truncation is applied correct for any truncation of T consistently in all equations (including the original calculation of the unperturbed energy and amplitudes). Watts.

the variation of the functional with respect to its arguments is given by ˆ ˆ ˆ ˆ ˆ δE P ˆ=P ˆ δΛ Q ˆ HP ˆ+P ˆ (1 + ΛQ ˆ ) δ (e−T HN eT ) P P ˆ δΛ Q ˆ HP ˆ+P ˆ (1 + ΛQ ˆ )[H. P When the stationarity conditions are satisﬁed we have ˆE P ˆ=P ˆ (1 + Λ)HP ˆ=P ˆ HP ˆ = ∆E P ˆ. including the extension to higher derivatives.64) ˆ ˆ ˆ T where H = e−T H N e . Thus. Λ = ΛQ ˆT ˆ = 0. so that a generalized Hellmann–Feynman theorem is applicable. These results are correct for all values of λ. the second term of (11. δ T ˆ]P ˆ+ P ˆ ΛQ ˆ [H. T (11. P ˆ=Q ˆT ˆ. δ T ˆ]P ˆ. and with ∆E P ˆ ˆ P HP . T (11. T ˆ)P ˆ=P ˆ (1 + Λ)e−T HN eT P .366 Calculation of properties in coupled-cluster theory 11. P ˆ δT ˆP ˆ to zero yields Setting the coeﬃcient of Q ˆ HQ ˆ+P ˆ ΛQ ˆ (H − ∆E )Q ˆ = 0.65) so that requiring that E be stationary with respect to Λ corresponds to ˆ= satisfying the CC amplitude equations. when it is stationary the functional E provides the since Λ = ΛQ perturbed correlation energy. because certain derivatives vanish when the functional is made stationary.63) Noting (11. E (Λ.64) can be transformed to ˆ [H. except that they At this point nothing is said about the nature of Λ and T ˆ N ) depend on λ and that Λ is a de-excitation operator while T ˆ is an (and H excitation operator: ˆ = 0. Q (11. δ T ˆ ]P ˆ=P ˆ H + ΛQ ˆ H − ∆E Λ Q ˆ δT ˆP ˆ.66) ˆ . ˆ deﬁned by Consider a functional of Λ and T ˆ ˆ ˆˆ ˆ) = P ˆ E (Λ.67) (11. based on a stationarity requirement of a functional. Such an approach is convenient for the calculation of derivatives. we set the coeﬃcients of δ Λ and δ T ˆ to zero. . P (11. Requiring that this functional be stationary with ˆ. =P (11.63). ΛP ˆ. For respect to both Λ and T the ﬁrst term we ﬁnd ˆ HP ˆ = 0. With this result.62) ˆ.4 The CC energy functional We shall now derive the results of the previous section in a diﬀerent way.

. Nooijen and Bartlett 1995). ˆ (1 + ΛQ ˆ )(H − ∆E ) = 0 . In both cases making the functional plays in CI. where |Ψ = (1 + C stationary yields the corresponding wave function and energy. Because Λ(0) is independent of the perturbation it needs to be evaluated only once for any number of properties. diﬀerentiation of the stationary functional with respect to λ yields ˆ=P ˆE λP ˆ=P ˆ (1 + Λ)H[λ] P ˆ. |H|0 of these ij. It can also be interpreted in terms of a constrained optimization process in which E is made stationary subject to the satisfaction of the CC amplitude equations by adding the left-hand sides Φab. Trucks and Bartlett 1989. T ˆ) when the full Hamiltonian H ˆ is written simply as ∆E (Λ. This result represents a generalized Hellmann– Feynman theorem. which in turn requires inversion of the denominator ˆ H − ∆E Q ˆ . The functional E (Arponen 1983. Straightforward evaluation of Λ(0) from (11. Szalay and Bartlett 1992.68) in agreement with (11... equations.61). 11..5 The Λ equations It has been shown in the previous two sections that property values can be expressed as derivatives of the perturbed energy with respect to a strength parameter λ at λ = 0 and that these derivatives can be evaluated in terms ˆ N . ∆E λ P (11. multiplied by Lagrange multipliers λab.69) . ij.. has been called the fundamental energy functional of CC theory. Szalay.11. P (11. Salter. T ˆ used instead of HN ). often ˆ) (or E (Λ. ˆ = C ˆ1 + C treated as a functional of the CI excitation operator C ˆ )|0 .5). Bartlett 1995. to the correlation energy (Koch and Jørgensen 1990).. This functional plays the same role in CC theory as the expectation value of the Hamiltonian. ˆ2 + · · · . Thus of the operator Λ and derivatives of the perturbed Hamiltonian H the evaluation of Λ at the point λ = 0 is a key step in property calculations. In the CC case the stationarity of E also yields the de-excitation operator Λ. but a more eﬃcient procedure is to solve the simultaneous Q equations (11.60) would involve the nondiagonal resolvent R(0)..5 The Λ equations 367 Furthermore. (see Section 11.66) for Λ.. allowing the evaluation of derivatives of the energy in terms of derivatives of the Hamiltonian without diﬀerentiation of the wave function. This approach corresponds to the solution of the left eigenfunction equation for H.

. double. {i aj b · · · } . Projecting it onto Q ˆ (1 + ΛQ ˆ )(H − ∆E )Q ˆ=0 P or ˆ HQ ˆ+P ˆ ΛH Q ˆ − ∆E P ˆ ΛQ ˆ = 0.71) 1 (n!)2 ij ijk in these expressions the de-excitation amplitudes λi a . λabc . deAs in the treatment of T excitation operators.67). Λ = Λ1 + Λ2 + Λ3 + · · · . ab. P (11. . are to be determined. These operators are represented diagrammatically by i a . † † λij. The energy can be eliminated formally by introducing a commutator for the second term. i aj bk c .. H]Q ˆ+P ˆ H(P ˆ+Q ˆ )ΛQ ˆ P ˆ ΛH Q ˆ + ∆E P ˆ ΛQ ˆ+P ˆ HQ ˆ ΛQ ˆ. but on the other hand resemble the CI equations in that the unknown amplitudes appear linearly. ˆ† ˆ} λi a {i a ˆ† ˆˆ λij j †ˆ b} ab {i a ˆ† c ˆ† ˆˆ λijk j †ˆ ˆ} bk abc {i a ˆ merely reproduces the CC energy expression Projecting (11. where Λ1 = ia † λi a {i a} . ˆ. =P C (11.69) onto P ˆ produces the Λ amplitude equation (11.73) . ab..70) Λ2 = Λn = (11. On the one hand these equations resemble the CC amplitude equations in that they are simultaneous equations rather than an eigenvalue problem. we expand Λ in single-. ˆ ΛH Q ˆ=P ˆ [Λ.368 Calculation of properties in coupled-cluster theory ˆ ˆ N eT in which the eigenvalue ∆E and the operator H = H are known from C the solution of the CC amplitude equations. . Thus we have a set of linear simultaneous equations for the components of Λ. etc. λab . the energy is not eliminated and they do not arise from the commutator expansion of an exponential and so are not limited to connected terms. i aj b ... ij.72) Because of the appearance of the energy in these equations the Λ amplitudes could be disconnected... (11. 1 4 ijab † † λij ab {i aj b} ...etc.

11.5 The Λ equations

369

**where the subscript C indicates a restriction to connected terms. For future reference we also note that ˆ ΛH P ˆ=P ˆ [Λ, H]P ˆ+P ˆ H ΛP ˆ=P ˆ ΛH P
**

C

ˆ, P

(11.74)

**ˆ = 0. because ΛP Substituting (11.73) into (11.72) cancels the energy-containing term, leaving ˆ+P ˆ HQ ˆ ΛQ ˆ=0 ˆ HQ ˆ+P ˆ ΛH Q (11.75) P
**

C

or

ˆ ˆ H ˆ N eT P C ˆ ˆ+P ˆ Λ H ˆ N eT Q C C ˆ ˆ+P ˆ H ˆ N eT Q C

ˆ ΛQ ˆ = 0, Q

(11.76)

in which the last term is the only one that is not fully connected, consisting as it does of a product of two terms. Explicitly, this equation can be written in the form ˆ N eT |Φab... 0|H ij... +

k<l<··· c<d<··· ˆ C ˆ ˆ N eT + 0Λ H C

Φab... ij...

C

ˆ Ne 0|H

ˆ T

|Φcd... kl... C

ab... Φcd... kl... |Λ|Φij...

= 0.

(11.77)

The intermediate state |Φcd... kl... in the disconnected term must represent a deexcitation of the initial state |Φab... ij... , and thus its indices must be a subset of the indices of that initial state. Equation (11.77) has to be solved for the Λn amplitudes for all i < j < · · · ··· a and a < b < · · · . When |Φab ij ··· is a singly excited state |Φi the third ˆ -state can be obtained by de-excitation of a singly term vanishes, since no Q excited state, and so the Λ1 equations are fully connected,

ˆ a ˆ N eT |Φi 0|H C ˆ ˆ N eT + 0Λ H C

Φa i

C

=0

(11.78)

ˆ=T ˆ1 + for all i and a. These equations, truncated to the CCSDT case (T ˆ ˆ T2 + T3 and Λ = Λ1 + Λ2 + Λ3 ), are represented by the diagrams in Fig. 11.4. Diagrams 1 and 2 in this ﬁgure represent the ﬁrst term of (11.78), while diagrams 3–83 represent the second term. The latter diagrams all have a ˆ ˆ N vertex but may have as many as three T single Λ vertex and a single H ˆ vertices. Speciﬁcally, diagrams 3–5 represent Λ1 HN terms, 6–11 represent ˆ NT ˆ1 , 12–14 represent Λ1 H ˆ NT ˆ2 , 15–17 represent Λ1 H ˆN 1 T ˆ2 Λ1 H 2 1 , 18 and 19 ˆ N , 20–23 represent Λ2 H ˆ NT ˆ1 , 24–31 represent Λ2 H ˆ NT ˆ2 , 32–35 represent Λ2 H 1 ˆ2 ˆ ˆ ˆ ˆ represent Λ2 HN 2 T1 , 36–43 represent Λ2 HN T1 T2 , and 44 and 45 represent ˆN 1 T ˆ3 ˆ Λ2 H 3! 1 . The remaining diagrams contain T3 and/or Λ3 vertices, and would not be included in a CCSD calculation. The open lines in all these

370 × 1 2

Calculation of properties in coupled-cluster theory × 3 4 × 5 6 × 7 × 8 9 10 11

12

13

14 ×

15 × 25

16

17

18

19

20

21

22

23

24

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52 ×

53 × 60

54

55

56

57

58

59

61

62

63

64

65

66

67

68

69

70

71

72

73

74

75

76

77

78

79

80

81

82

83

Fig. 11.4. Diagrams for the Λ1 equations for CCSDT.

11.5 The Λ equations

371

diagrams have the ﬁxed labels a and i. Diﬀerent values of a and i generate diﬀerent equations in the set of simultaneous equations for Λ. As indicated by the form of the second terms in (11.77) and (11.78), each ˆ T vertex in the diagrams must have a direct connection to the interaction vertex. Diagrams that would become disconnected if the Λ vertex were removed, such as , do not contribute to the Λ equations. Furthermore, diagrams in which no open lines connect directly to the interaction vertex, such as ×,

ˆ

×,

or

,

ˆ (H ˆ N eT )C |0 component. While not individually vancontain an embedded Q ishing, such diagrams add up to zero when the CC amplitude equations are satisﬁed and so have been left out of Fig. 11.4 and subsequent diagram sets for the Λ equations. in (11.77) is a doubly excited state we obtain the Λ2 When |Φab... ij... equations, ˆ N eT |Φab 0|H ij

ˆ C

ˆ N eT |Φab + 0|ΛH ij

ˆ

C

+

c=a,b k=i,j

ˆ N eT |Φc 0|H k

ˆ

C

ab Φc k |Λ|Φij = 0 ,

(11.79) represented diagrammatically for the CCSDT case in Fig. 11.5. In this ﬁgure diagram 1 represents the ﬁrst term in (11.79), the disconnected diagrams 2 and 3 represent the third term, and diagrams 4–71 represent the second term. Only diagrams 1–32 are needed in the CCSD case. The open lines in these diagrams carry the labels a, b, i, j and all permutations of a and b between inequivalent open particle lines and of i and j between inequivalent open hole lines must be included and also a weight factor 1 2 for any pair ˆ of equivalent T vertices, as in diagrams 31 and 32. As indicated in the rules of interpretation, Fig. 10.1, this factor can be canceled with one of the permutation operators. ˆ1 Λ1 , reˆ NT ˆ N Λ1 and H The disconnected diagrams 2 and 3 represent H spectively. Following the sequence of the operators in these terms, these diagrams should be drawn with the time sequences × and ,

372 × 1 2

Calculation of properties in coupled-cluster theory × 3 4 × 10 11 12 13 14 5 × 15 16 17 18 6 7 8 9 ×

19

20

21

22

23

24

25

26

27

28

29

30

31

32 ×

33 × 40

34

35

36

37

38

39

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

61

62

63

64

65

66

67

68

69

70

71

Fig. 11.5. Diagrams for the Λ2 equations for CCSDT.

11.5 The Λ equations

373

respectively, clearly showing the intermediate state. However, the sum over c and k in (11.77) can be expanded as follows,

ˆ c ˆ N eT 0|H |Φk c=a,b k=i,j ab Φc k |Λ|Φij C ˆ b ab ˆ T Φa i |Λ|Φij + 0|HN e |Φi ˆ C C ab Φb i |Λ|Φij C ab Φb j |Λ|Φij

C

ˆ a ˆ N eT = 0|H |Φi ˆ

ˆ N eT |Φa + 0|H j ˆ N eT |Φa = 0|H i

ˆ ˆ C

ab ˆ T b Φa j |Λ|Φij + 0|HN e |Φj ˆ C

ˆ T b 0|Λ|Φb j − 0|HN e |Φi

ˆ C ˆ

0|Λ|Φa j

C

ˆ N eT |Φa − 0|H j

ˆ T b 0|Λ|Φb i + 0|HN e |Φj

C

0|Λ|Φa i (11.80)

ˆ (ij |ab) 0|H ˆ N eT |Φa =P i

0|Λ|Φb j .

It is clear that this expression is represented by diagrams 2 and 3, interpreted as permutation sums of products of their disconnected parts, as drawn without regard for the time sequence in Fig. 11.5. The diagrams for the Λ3 equations for CCSDT are shown in Fig. 11.6. The disconnected diagrams 1–3 in this ﬁgure represent the third term of (11.77), which can be put into the form ˆ N eT |Φde 0|H lm

d,e=a,b,c l,m=i,j,k ˆ abc Φde lm |Λ|Φijk + d=a,b,c k=i,j,k

C

ˆ N eT |Φd 0|H l

ˆ

C

abc Φd l |Λ|Φijk ,

involving both singly and doubly excited intermediate states. When drawn in the time sequence corresponding to the above expression, the disconnected diagrams take the forms × , , ,

respectively, also represented as permutation sums of products, as drawn without regard to time sequence in Fig. 11.6. While the Λ expansion includes disconnected terms, when the Λ diagrams are contracted with H to form the diagrammatic representation of the CC energy functional (11.62) the resulting diagrams are fully connected. Thus all the CC energies and their derivatives are properly linked and connected, as required for extensivity.

374

Calculation of properties in coupled-cluster theory × 1 2 × 7 × 13 14 8 × 15 16 17 9 3 × 10 11 12 4 5 6

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

Fig. 11.6. Diagrams for the Λ3 equations for CCSDT.

ˆ amplitude equations, which are nonlinear, the Λ amplitude Unlike the T equations are linear in Λ but are still solved iteratively because of their ˆ amplitude equations, the diagonal parts size. As in the treatment of the T (involving faa and fii ) of diagrams containing just the one-electron Hamiltonian operator and the Λ operator, i.e. diagrams 3 and 4 of Fig. 11.4 and diagrams 8 and 9 of Figs. 11.5 and 11.6, are moved to the opposite side of the i ab ij abc ijk equation, resulting in expressions for εa i λa , εij λab and εijk λabc , respectively, in terms of the other components of Λ. Comparing the contributions of diﬀerent perturbation-theory orders to the diagrams for the Λ equations with the corresponding contributions to the ˆ(1)† . ˆ amplitude equations, it is easily seen that Λ(1) = T diagrams for the T ˆ† provides a suitable initial approximation to Λ in the iteraTherefore T tive solution of the Λ equations. In second order we ﬁnd a few diﬀerences between the diagrams contributing to the Λ equations and the adjoints of ˆ equations. For Λ1 the diﬀerence is bethe diagrams contributing to the T

11.5 The Λ equations

375

**tween diagram 2 of Fig. 11.4 and the adjoint of diagram S2a of Fig. 10.2, so that
**

(1)

εa i

(2) λi a

−

a(2)∗ ti

=

(1)

− 1 1 − ab b εj εij

×= ij ab fbj = εa i

jb

×−

×

=

jb

ij ab fbj , (11.81) b εab ij εj

**where the thin horizontal lines indicate denominators. As a result we ﬁnd that
**

(2) λi = ti a a(2)∗

+

jb

tij

ab(1)∗ b(1) tj .

(11.82)

**For Λ2 we ﬁnd λab
**

ij (2)

= tij

ab(2)∗

ˆ (ij )ta(1)∗ tb(1)∗ , +P i j

(11.83)

owing to the disconnected diagram 2 of Fig. 11.5, which has no counterpart ˆ2 -equation diagrams of Figs. 9.2 and 10.3. Similarly, in the T λabc

ijk(2)

= tijk

abc(2)∗

ˆ (i/jk|a/bc)ta(1)∗ tbc(1)∗ , +P i jk

(11.84)

due to the disconnected diagrams 1 and 2 of Fig. 11.6. Equations (11.82)– (11.84) can be represented diagrammatically as (2) = (2) = (2) = (2) + (2) + (2) + (1) (1) (1) (1) , (1) , (1) . (11.87) (11.86) (11.85)

ˆ† involve diagrams that All the second-order diﬀerences between Λ and T (2) ˆ(2)† and making T ˆ† vanish in the Hartree–Fock case, resulting in Λ = T a particularly eﬀective initial approximation to Λ in the HF case. The linearized CCSD approximation (LCCSD, analogous to the LCCD approximation of subsection 9.3.1) and MBPT through fourth order in the energy (which depends on the second-order wave function) involve linear terms only, ˆ† exactly and separate consideration of Λ is so that in the HF case Λ = T not required in these approximations.

376

Calculation of properties in coupled-cluster theory

11.6 Eﬀective-Hamiltonian form of the Λ equations Compact forms of the Λ equations can be obtained using the matrix elements of the eﬀective Hamiltonian (10.51), given in Section 10.7. Expressing the Λ1 equations for CCSDT in terms of eﬀective-Hamiltonian matrix elements results in the 22 diagrams given in Fig. 11.7. The numbering of each diagram in this ﬁgure corresponds to the numbering of the leading term of its expansion in terms of the diagrams in Fig. 11.4. The diagrams of Fig. 11.4 accounted for by each diagram of Fig. 11.7 are listed in Table 11.1. Diagram 75 in this ﬁgure uses the three-body matrix element deﬁned in (10.87). Its value does not change if all arrows are reversed; in either form, it expands into diagram 75 of Fig. 11.4. ˆ. All the diagrams in Fig. 11.7 are at most linear or pseudolinear in T However, pseudolinearity is less important in the Λ equations, because the ˆ amplitudes are known before these equations need to be solved. T Expressing the Λ2 equations for CCSDT in terms of eﬀective-Hamiltonian matrix elements results in the 22 diagrams given in Fig. 11.8. As in the case of the Λ1 equations, the numbering of each diagram in this ﬁgure corresponds

1

3

4

5

18

19

26

27

48

49

50

51

52

53

54

55

56

57

58

59

60

75

Fig. 11.7. Diagrams for the Λ1 equations for CCSDT in terms of eﬀectiveHamiltonian vertices.

11.6 Eﬀective-Hamiltonian form of the Λ equations

377

1

2

4

5

8

9

10

11

12

23

24

33

34

45

46

47

48

55

56

57

58

59

Fig. 11.8. Diagrams for the Λ2 equations for CCSDT in terms of eﬀectiveHamiltonian vertices.

1

2

4

5

8

9

10

11

12

23

24

Fig. 11.9. Diagrams for the Λ3 equations for CCSDT in terms of eﬀectiveHamiltonian vertices.

to the numbering of the leading term of its expansion in terms of the diagrams in Fig. 11.5. The diagrams of Fig. 11.5 accounted for by each diagram of Fig. 11.8 are also listed in Table 11.1. Similar results for the Λ3 equations are shown in Fig. 11.9 and Table 11.1. The Λ-equation diagrams do not require the use of the various χ , χ , etc. intermediates of Section 10.7. Permutation factors for labels of open lines

378

Calculation of properties in coupled-cluster theory

Table 11.1. Correspondence between the eﬀective-Hamiltonian-based diagrams for the Λ equations and the original diagrams

H-based diagram Original diagrams H-based diagram Original diagrams

Λ1 equations (Figs. 11.7 and 11.4) 1 3 4 5 18 19 26 27 48 49 1 2 4 5 8 9 10 11 12 23 24 1, 2 3, 7, 8, 14, 17 4, 6, 9, 13, 16 5, 10, 11, 12, 15 18, 20, 23, 25, 28, 35, 37, 39, 41, 45, 19, 21, 22, 24, 29, 34, 36, 38, 40, 44, 26, 42 27, 43 48 49, 66, 77, 81 1 2, 3 4, 6 5, 7 8, 14, 15, 22, 29 9, 13, 16, 21, 28 10, 17, 18, 25, 30 11, 19, 27, 32 12, 20, 26, 31 23 24 50 51 52 53 54 55 56 57 58 59 60 75 33 34 45 46 47 48 55 56 57 58 59 10 11 12 23 24 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 75 33, 52, 34, 51, 45, 46, 47, 48, 55 56 57 58 59 65, 76, 80 61, 64, 78, 82 62, 63, 79, 83 67 68 72 71 73 74 69 70

31, 32, 47 30, 33, 46

Λ2 equations (Figs. 11.8 and 11.5) 35, 54, 36, 53, 66 67 69 68 38, 61, 37, 60, 40, 63, 39, 62, 41, 65, 42, 64, 44, 49, 71 43, 50, 70

Λ3 equations (Figs. 11.9 and 11.6) 1 2 4 5 8 9 1 2, 4, 5, 8, 9, 3 6 7 14, 15, 22, 29 13, 16, 21, 28 10, 17, 18, 25, 30 11, 19, 27, 32 12, 20, 26, 31 23 24

10.10. 11.10 contains a ˆ4 ˆ4 contribution that should be left out of a CCSDT calculation. As indicated in that table. . is represented by all connected diagrams that can be drawn ˆ (ΛH)C Q P containing a single H vertex below a single Λ vertex. 11.2. with n pairs of lines open at the bottom and no lines open at the top. numbered according to their leading terms in Figs. The T is the most direct representation of the Λ equation (11.9 Λ1 equations 26 49 50 53 26. are in one-to-one correspondence Fig. . Correspondence between the T Λ equations in Fig. equations in terms of diagrams in which no T These diagrams are given for CCSDT in Fig. 49. ˆ . The correspondence between the two sets of diagrams is shown in Table 11. 11. ˆ-free form of the Λ diagrams. P and Λ2 diagram sets in Fig. an embedded Q . For the same equation. 58.75).2.) Those connected diagrams in which the H vertex has no open lines connecting directly to it contain ˆ HP ˆ = 0 factor and may be left out.7–11. 57 48.9 Λ2 equations 45. 11. 50. 59 + one T ˆ4 diagram 54 + one T 45 47 48 Fig. 11. 48 ˆ4 diagram 52.7–11. we can obtain even more compact representations of the Λ ˆ vertices appear explicitly.9.7. 53. including all Λ3 diagrams. 11. 55. 27. 55 47. 50 and 53 of the Λ1 equation in Fig. 11. each of diagrams 49.10 and the diagrams in Figs. These T T contributions are described by the diagrams . 60 + one T ˆ4 diagram 51. 46. Using the three-body and four-body matrix elements of H presented in Section 10. 11. 11. Diagrams not listed.10 Figs. 11. (A pair of open lines consists of one particle line and one hole line.11. For the Λn equation.10 Figs.7–11. 11. as exempliﬁed for CCSDT in Fig. 56 that become closed lines when the diagrams expressing the H vertices are embedded in the Λ-equation diagrams are automatically accounted for by the unrestricted summations over internal-line labels.9. the second term. The ﬁrst term ˆ HQ ˆ . 57. in order.10. is represented by the ﬁrst diagram in each of the Λ1 of that equation.7–11.6 Eﬀective-Hamiltonian form of the Λ equations 379 ˆ-free diagrams for the Table 11. 56.

The Λ-equation diagrams for CCSDT in terms of eﬀective-Hamiltonian ˆ vertices. The last three diagrams of the Λ1 equation vertices without explicit T ˆ each contain a T4 contribution.10.380 Calculation of properties in coupled-cluster theory Λ1 -equation diagrams 1 3 4 5 18 19 26 49 50 Λ2 -equation diagrams 53 1 2 4 5 8 9 10 11 12 23 24 33 34 45 47 48 Λ3 -equation diagrams 58 59 1 2 4 5 8 9 10 11 12 23 24 Fig. . 11.

(11.28).74). The diagrams for the hole–particle.62) rather than the derivatives of the correlation-energy expectation value (11. because they are higher than fourth order in MBPT. to emphasize their origin in the CC energy derivatives treatment.1 and 11. respectively. noting (11. Diagrams 14a and 14b have no counterparts in Fig. Thus.). for the correlation component of the energy derivatives we need the derivatives of the CC energy functional (11. we need to use H vertices with n∨ − n∧ = 2(m − n).37)–(11. resulting in ˆ† q ˆ}eT |0 = 0 (1 + Λ) {p ˆ† q ˆ}eT (γ N )qp = 0|(1 + Λ)e−T {p ˆ† q ˆ† s ˆr ˆ}eT (ΓN )rspq = 0 (1 + Λ) {p ˆ C ˆ ˆ ˆ C 0 C . grams in Figs. and particularly for geometric derivatives of the potential-energy surface (gradients.11. Additional intermediates would also need to be deﬁned to facilitate eﬃcient implementation of the calculations. for which n∧ = 4 is also possible.1. with a total of n pairs of lines open at the bottom and no lines open at the top. it is important to maintain consistency between the energy calculations and the derivative calculations. to generate diagrams describing the contributions of Λm vertices to the Λn equation. These matrices are referred to as response density matrices. 11. As noted previously. is represented by all diagrams containing one H the third term. n∧ < 4 except for the two-body vertex. they are a subset of the diaˆ† vertex replaced by a Λ vertex. (11.2. particle–hole and particle–particle elements of the one-body response density matrix for the CCSDT case are shown in Fig.11. with the top T and are numbered correspondingly.13). they have the important beneﬁt that they lead to ﬁnite expansions for any level of CC rather than the inﬁnite expansions for the expectation-value forms seen in (11.39).7 Response treatment of the density matrices 381 ˆ HQ ˆ ΛQ ˆ .88) (11. Because of their linearity in Λ. Using this analysis. In general. 11. 11. . where n∨ and n∧ are the number of lines connecting to the H vertex from above and from below. the treatment can be extended to higher CC levels. The density matrices appropriate for the determination of ˆ† these derivatives are obtained by replacing the factor eT of (11.89) 0 C . 11. P vertex and one Λ vertex disconnected from each other. Extension to CCSDTQ requires adding diagrams with Λ4 vertices and possibly higher-order H vertices to the Λ1 –Λ3 equations and generating the corresponding Λ4 -equation diagrams.31) ˆ by the factor (1 + Λ)e−T of the CC energy functional.7 Response treatment of the density matrices For coupled-cluster energy derivatives. Except for diagram 1 . Hessians etc.

89). Diagrams with more than than one T because of the linearity in Λ of the density matrix formula.88). Also missing from the Λ-based diagrams are any diagrams that would become disconnected if the Λ vertex were removed. (11. respectively. the particle–hole γai (1–14b) and the particle–particle γba (20–25) elements of the one-body response density matrix for CCSDT. Because of this linearity.382 i a i 1 a 1 Calculation of properties in coupled-cluster theory i 2 a ia 3 ia 5 a ai 6 a b 21 i 7 a a b 25 ia 14 ia 14a 14b ai b 20 Fig. Diagrams representing the hole–particle γia (1 ). γai = i a i a jb jb (2) (1) (11.90) (2) i a (2) a(2)∗ ba(1)∗ b(1) (2) = λi = ti + tji tj . ˆ† vertex do not have a Λ counterpart. b by hole labels j . the response density matrices are not exactly Hermitian. Diagrams for the hole–hole matrix elements (γ N )ji can be obtained by reversing all arrows in the particle–particle diagrams in Fig.11 and replacing the particle labels a. Diagram 1 represents the hole–particle element γia = λi a and is the only surviving diagram arising from the adjoints of the diagrams of Fig.1. such as the counterparts of diagrams 4. 11. At second order we ﬁnd (1) (2) a(2) ba(1) j (1) a(2) ba(1) b(1)∗ + = ti + tji λb = ti + tji tj . As a result. (11. 11. 11. The asymmetry with respect to time inversion in the expansions obtained for the hole–particle (diagram 1 ) and particle–hole (diagrams 1–14b) elements is a manifestation of the non-Hermiticity of the CC energy functional and the corresponding Λ-based expressions for the density matrices.1. Examination of the contributions of diﬀerent MBPT orders to the ∗ in ﬁrst hole–particle and particle–hole elements readily shows that γia = γai order.88). 8 and 13 of Fig.91) γia = a jb . 11. because of the connectedness restriction in (11. i. the diagrams in Figs.11.11 present untruncated expressions for the corresponding density-matrix elements for CCSDT. (11. 11.

phph and ˆ(1) in second order and. b = λab = tij +P i j (2) (2)∗ (11.92). the pphp and hppp blocks.83).9. 11.92) In general. +P i j (11.82). the particle–particle and hole–hole blocks of γ do not have ﬁrst-order contributions. The remaining matrix-element types can be obtained from matrix elements of the above diagrams using (11. respectively. hphh and hhhh can be obtained by reversing all arrows on the diagrams for ppph. Only one diagram contributes at second order to each ppph and phpp block. pphh. and the hhph and phhh blocks. They also are Hermitian conjugates of each other in second order.12) that some of the index-permutation relationships (11. The diagrams for the two-body response density matrix for CCSDT are shown in Fig. All other blocks involve only Λ(1) and ˆ(1)† . where p and h stand for particle and holes. and these blocks are also Hermitian conjugates of each other at second order. 11.12 for the pppp. Λ(1) = T For the two-body density matrix we ﬁnd from (11. the anti-Hermitian components generate imaginary contributions to the CC energy derivatives used in property calculations. so that γia = γai . hhhh.11.93) i j (2) ij (2) ab(2)∗ ˆ (ij )ta(1)∗ tb(1)∗ . ppph. (11. The deviation from Hermiticity of the response density matrices is an artefact of the truncated CC treatment and disappears in full (untruncated) CC. As seen from diagrams 20 and 21 of Fig. Γabij = a Γijab = a (2) (2) i j b +a (2) i j (1) b = tij (1) ab(2) ˆ (ij )ta(1) tb(1) . Γrspq = Γ∗ pqrs . phpp. the pppp.26) still hold: Γrspq = −Γrsqp = −Γsrpq = Γsrqp .7 Response treatment of the density matrices (2) (2)∗ 383 by (11. Thus we see that γ is Hermitian at second order also.12. Therefore Γ is Hermitian in second order. Diagrams for hhhp. and they are Hermitian in second order because ˆ(1)† . and the hpph and phhp blocks are Hermitian conjugates of each other at second order. 11. The pphh and hhpp blocks of Γ are the only blocks that have ﬁrst-order contributions (from diagrams 21 and 66 of Fig.94) by (11. and clearly are Hermitian conjugates of each other in that order. A similar result is obtained for the hhhp and hphh blocks. respectively. respectively). hpph and hhpp elements. so that Γijab = Γabij . As will be seen in Section 11.11. because Λ(1) = T T hphp blocks of Γ are clearly Hermitian at second order. These contributions are mathematically correct components . phpp and pppp.

Diagrams for the pppp (1–4). pphh (21–59).12. phpp (19. ppph (5–18). 20).384 Calculation of properties in coupled-cluster theory 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 Fig. hpph (60–65) and hhpp (66) elements of the two-body response density matrix for CCSDT. 11. .

but use of the CC energy functional redresses this defect. lack of Hermiticity is not a problem. When the orbitals and matrices are real. the results ﬁnite-order perturbation theory or CC for any truncation of T are not invariant under arbitrary transformations of the orbitals. even if there is no corresponding wave function. On the one hand. because these treatments depend critically on the choice of reference function (or reference space in the case of multireference treatments) and this function .27) or (11. But for this small inconvenience in the CC methods we gain extensivity and the ability to generalize to any model that provides a well-deﬁned energy expression. in a full-CI calculation these integrals fully determine the results. the above treatment.8 The perturbed reference function When we are dealing with properties expressed in terms of expectation values of one-electron or two-electron operators. additional factors accounting for orbital relaxation must be considered. consistent with the Q pression.11. However. ˆ. as is the case in almost all calculations. The response density matrices can be deﬁned for any system that oﬀers an energy expression. when the property also involves variation in the orbitals. but they are clearly unphysical and are usually ignored. regardless of how the basis functions are transformed into orbitals.8 The perturbed reference function 385 ˆ HP ˆ energy exof the derivatives of the CC energy.33). On the other hand. such as truncated CI. 11. Because the response density matrix in CC property calculations appears only in dot products with sets of symmetric property integrals. since the wave function and energy are invariant under any linear transformation of the orbitals. is suﬃcient for the determination of the correlation contribution to the property values through (11. in any truncated correlation treatment. the contributions from the anti-Hermitian components cancel. involving the solution of the Λ equations and evaluation of the corresponding response density matrices. These advantages are shared by all MBPT approximations and by CC models such as CCSD(T) and are another reason why CC theory is the most frequently used high-level method in quantum chemistry today. the electronic structure problem is fully speciﬁed by the set of one-electron and two-electron integrals over the basis functions (plus the internuclear-repulsion terms). much less the solution of auxiliary equations. The need to solve the auxiliary Λ amplitude equations is the price paid for the failure of CC methods to produce a variational or stationary energy. The calculation of properties in CI does not require knowledge of the λ-derivative of the CI coeﬃcients. Within a given basis set.

In all these cases the results are fully determined by the one-electron and two-electron integrals over the orbitals and by the truncation scheme. For the consistency of the treatment for diﬀerent values of the perturbation strength. Under the eﬀect of a one-electron perturbation. the treatment will be in terms of the spinorbitals. the operators themselves remain formally unchanged and may be considered invariant under the perturbation as long as the orbital occupations do not change. once we know the dependence of the orbitals on the perturbation. so that the relevant equation is the UHF equation. we cannot ignore the impact of the perturbation on this equation or condition and its resulting eﬀect on the reference function and orbitals. The dependence on the orbital integrals is obvious from the form of the second-quantized Hamiltonian. Other orbital choices can be treated by analogous methods. this equation is the Hartree–Fock equation in any of its forms. It will also be assumed here that the unperturbed HF function is in canonical form. Therefore. the usual UHF Fock operator ˆ= h ˆ +J ˆ− K ˆ. As throughout most of this book. The creation and annihilation operators that appear in this Hamiltonian aﬀect only the occupation numbers of the orbitals in the functions on which they operate. because this operator has the eﬀect of transforming the orbitals. The orbitals must change smoothly under the eﬀects of a perturbation. but it does not eliminate it. presumably. on the choice of the orbitals. but their occupation numbers cannot be allowed to change if the wave function and energy are to vary smoothly.95) .3.386 Calculation of properties in coupled-cluster theory ˆ1 in a CC depends. and thus the reference function. f (11. are usually chosen to satisfy some equation or condition that serves to provide a satisfactory zero-order function and. in turn. to produce a relaxed density matrix as will be shown in Section 11. and thus for obtaining meaningful values of the derivatives.9. In the present section we shall conﬁne our analysis to the case in which the Hartree–Fock equation is used to determine the reference function. The inclusion of T treatment reduces this dependence. The common case of closed-shell restricted HF is easily accommodated by the spinsummation techniques described in Section 7. satisfactory convergence of the correlation treatment. The resulting orbital relaxation leads to a perturbation of the response density matrix γ . Most commonly. The orbitals. while the creation and annihilation operators will create and annihilate the modiﬁed orbitals.

In the electric ﬁeld case such a dependence is not common. J ˆ = K i ˆi K (11. without any operator λO pq p|o ˆ higher-order terms in h. since atomic-orbital basis functions move with the atoms. (11.99) The elements Cνp are known as the MO (molecular orbital) coeﬃcients. f (11. The orbitals {φp (λ)} are expressed in terms of the basis functions {χν (λ)} by means of a transformation matrix C(λ): φp (λ) = ν χν (λ)Cνp (λ) .98) (11. The and K perturbed Fock operator can be expanded in a strength parameter λ as ˆ dˆ (J − K ) ˆ(λ) = f ˆ(0) + λ dh +λ f dλ λ=0 dλ λ=0 2 2 ˆ ˆ 1 d h 1 d (J − K ) + λ2 2 − λ2 + ··· 2 dλ λ=0 2 dλ2 λ=0 ˆ(0) + λ(h ˆλ + J ˆ(0) + λf ˆλ + · · · ˆλ − K ˆ λ) + · · · = f =f where ˆλ + J ˆλ = h ˆλ − K ˆλ . The basis functions used in the calculation may also be functions of λ. Sadlej and Bartlett 1988) or magnetic ﬁeld (London 1937.11. Gauss 1993).8 The perturbed reference function 387 with coulomb and exchange operators ˆ= J i ˆi . Ditchﬁeld 1974. . in a complete basis set or one that does not follow the atoms. but for a magnetic ﬁeld the gauge dependence makes it a necessity. In some other cases the basis functions can be made functions of an electric ﬁeld (Pluta. Of course.97) In the common case in which the perturbation consists just of a one-electron ˆλ = o ˆ = λ ˆ|q {p ˆ† q ˆ} we simply have h ˆ. This dependence is absolutely essential when the property of interest is the analytical gradient. such variations should not arise.3).96) (Section 1. is modiﬁed not only by changes in the one-electron operator ˆ but also by changes induced in the Coulomb and exchange operators J ˆ h ˆ by changes in the orbitals in terms of which they are deﬁned. such as a plane-wave basis.

Cµp (λ) = q (0) Cµq Uqp (λ) . S (λ) = C(0)† S(λ)C(0) . (11.103) or. we obtain the usual form of the HF equations in terms of the perturbed basis functions. or C(λ) = C(0) U(λ) .107) U† (λ)S (λ)U(λ) = 1 . (11.101) Expanding φp (λ) according to (11. (11. Sµν (λ) = µ(λ)|ν (λ) and ε(λ) is the where Fµν (λ) = µ(λ)|f block-diagonal matrix of Lagrange multipliers εqp (λ) satisfying (11.105) to the perturbed HF equation (11.99) and operating from the left with χµ (λ)|. (11.388 Calculation of properties in coupled-cluster theory Without assuming that the perturbed HF orbitals are canonical. so that the unperturbed equation is ˆ(0) φ(0) = φ(0) ε(0) f p p p (for all p) .102) ˆ(λ)|ν (λ) . We shall assume that the unperturbed HF function is canonical.106) All matrices. F(λ)C(λ) = S(λ)C(λ)ε(λ) . (11. orbitals and operators are expanded in powers of λ. S (0) = 1 .102) and multiplying on the left by C(0)† . (11.101).100) with εia (λ) = εai (λ) = 0 (for all i.105) Applying the transformation (11. (11.108) . C† (λ)S(λ)C(λ) = 1 . in the unperturbed basis-function representation. where F (λ) = C(0)† F(λ)C(0) . U(0) = 1 . We now expand the perturbed MO coeﬃcients in terms of the unperturbed coeﬃcients by means of a transformation matrix U(λ). (11.104) (0) where ε(0) is the diagonal matrix of eigenvalues (orbital energies) {εp }. F(0) C(0) = S(0) C(0) ε(0) (11. we obtain F (λ)U(λ) = S (λ)U(λ)ε(λ) . the HF equations can be written in the form ˆ(λ)φp (λ) = f q φq (λ)εqp (λ) (for all p) . a). For the zero-order matrices we have F (0) = ε(0) . φp (λ)|φq (λ) = δpq .

q } = {i.109) (11.111) but can be obThe diagonal elements Upp tained from the orthonormality condition in (11. ε(0) = F λ − S λ ε(0) − ελ . we obtain (0) λ (0) λ 1 (0) (ε(0) q − εp )Upq = 2 (εp − εq )Spq or λ λ Upq = −1 2 Spq ({p. For the hole–hole and particle–particle blocks of Uλ we note that the oﬀ-diagonal Lagrange multipliers ελ pq ({p.110) (11.8 The perturbed reference function 389 The ﬁrst-order perturbed HF equation is obtained from (11. (11. providing an equation for the diagonal elements of ελ . λ λ (0) ελ pp = Fpp − Spp εp . for which ελ ai = 0.106). this condition is clearly satisﬁed by (11.111).113) and (11. (11. j } or {a. Substituting this choice into (11.112) For the particle–hole block of Uλ .106) as F λ + ε(0) Uλ = S λ ε(0) + Uλ ε(0) + ελ or Uλ . (0) (0) (11.113).114) which satisﬁes the Hermiticity condition ελ† = ελ .115) λ are not determined by (11. (11. (11. q } = {i. j } or {a.113) λ is obtained by interchanging The equation for the hole–particle elements Uia a and i in (11.115).11. (11. Thus we choose a generalization of (11. b}) .111) becomes λ λ λ (εi − ε(0) a )Uai = Fai − Sai εi .111) For p = q the left-hand side of (11. b}) . j } or {a.111) vanishes. q } = {i. so that the latter may also be used to determine the λ. which takes the ﬁrstorder form Uλ† + S λ + Uλ = 0 . which corresponds to (0) λ λ λ (0) λ (ε(0) q − εp )Upq = Fpq − Spq εq − εpq . (11.112) in the form λ (0) λ 1 (0) ελ pq = Fpq − 2 (εp + εq )Spq ({p.116) Because F λ and S λ are Hermitian. diagonal elements Upp . b}) are at our disposal since they reﬂect separate unitary transformations among the perturbed hole orbitals and among the perturbed particle orbitals.

resulting in the simpliﬁcation of most of the preceding λ = −U λ∗ .390 Calculation of properties in coupled-cluster theory Equations (11. are obtained as follows. in such cases Uλ is skew-Hermitian. Gerratt and Mills 1968). but equations (11. and equations. (11.117) vanishes. Pitzer and Lipscomb 1963. and all (0) superscripts are omitted to simplify the notation: [λ] φp λ = Sai µν λ = Fai µν ∗ ˆ(λ)|ν (λ) λ Cνi Cµa µ(λ)|f ∗ Cµa µ|ν λ Cνi = a[λ] |i + a|i[λ] = a|i [λ] . S(λ) = S(0) . S (λ) = 1 and S λ = 0. (11.105) as φλ p = ν λ λ (0) χ(0) ν Cνp + χν Cνp = q λ [λ] φ(0) q Uqp + φp . For such perturbations.111) are known as the coupled-perturbed Hartree–Fock (CPHF) equations (Stevens.118) For perturbations which do not involve changes in the basis functions the λ part of (11. The equations (11.113) for the particle–hole block of Uλ are not explicit because they are coupled by the dependence of F λ on Uλ through the derivatives of the Coulomb and exchange integrals. i.115) provide expressions for all elements of Uλ and ελ . needed for the solution of the CPHF The matrix elements Sai ai equations (11. Also. and their solution will be derived in the following analysis. |µ = |χµ etc. where we set |a = |φa .113).111)–(11.99) and (11.e.98).117) where λ] φ[ p = ν (0) χλ ν Cνp .119) ˆλ |i + a[λ] |f ˆ|i + a|f ˆ|i[λ] = a|f ˆ λ |i + a|J ˆλ − K ˆ λ |i + a[λ] |J ˆ− K ˆ |i + a|J ˆ− K ˆ |i[λ] = a|h ˆ λ |i + = a|h j aj λ ij + aj ij λ + a[λ] j ij + aj i[λ] j λ∗ λ Upj ap ij + aj ip Upj jp ˆ λ |i + = a|h . (11. The ﬁrst-order perturbed orbitals are obtained from (11. λ and F λ . (11. Uia ai λ λ Uij = Uab = 0. and we need only consider the changes Uqp in the orbital coeﬃcients.

Equation (11. straightforward computation of the second term would require a four-index transformation of the basis-function derivative integrals to orbital derivative integrals.11.122) are independent of the particle– hole block of Uλ and vanish if the perturbation is not accompanied by a change in the basis functions.8 The perturbed reference function 391 + j aj [λ] ij + aj ij [λ] + a[λ] j ij + aj i[λ] j λ∗ λ Ubj ab ij + aj ib Ubj jb λ + ak ij + aj ik Ukj j ˆ λ |i + = a|h + jk λ∗ Ukj aj ij [λ] . this sum may be written as λ∗ λ Ujk aj ik = − + Ukj jk jk λ Skj aj ik = − jk aj ik k |j [λ] . Pνµ = j ∗ Cµj Cνj . In other cases they can be computed from the basis-set derivative integrals.120) then takes the form λ ˆ λ |i + Fai = a|h jb λ∗ λ Ubj ab ij + aj ib Ubj [λ] − jk aj ik k |j + j aj ij [λ] . using the orthogonality condition (11. (11.122) The three terms in the last line of (11.113) can now be written as λ λ λ =Fai − Sai εi (εi − εa )Uai ˆ λ |i + = a|h jb λ∗ λ Ubj ab ij + aj ib Ubj [λ] − jk aj ik k |j + j aj ij [λ] − a|i [λ] εi .123) . (11. a relatively lengthy process that would have to be repeated for each perturbation.120) Since j and k are dummy summation indices in the ﬁrst sum in the last line and may be interchanged. however.121) The CPHF equations (11. This transformation can be avoided by use of the reference-function density matrix in the basis-function representation.116). (11. (11.

105).bj ≡ Φa i |H |Φj = ( a − i )δij δab − aj bi .392 Calculation of properties in coupled-cluster theory or in matrix form P = Cref C† ref . = a|h Xai (11.bj ≡ Φab ij |H |0 = ab ij . (11. ˆ Bai. Using this density matrix (at zero order).129) We next introduce the A and B matrices.113). (11. the relevant term of (11. µν µσ ντ µσ ντ λ j λ ∗ Cσj Cτ j = = where Pτ σ (11.125) λ Yµν = στ µσ ντ λ Pτ σ (11.130) .126) represents a two-index contraction of the two-electron basis-set derivative integrals with the unperturbed reference-function density matrix.127) and collecting the Uλ -dependent terms on the left-hand side of the CPHF equation (11. we obtain λ (εi − εa )Uai − jb λ∗ λ λ Ubj = Xai ab ij + aj ib Ubj . ˆ b Aai.127) simpliﬁes to λ ˆ λ |i . Deﬁning λ ˆ λ |i − a|i = a|h Xai [λ] εi − jk aj ik k |j [λ] + µν ∗ λ Cµa Cνi Yµν (11.122) can be written as aj ij j [λ] = µν ∗ Cµa Cνi στ ∗ Cµa Cνi µν στ ∗ λ Cµa Cνi Yµν .124) where it is understood that the matrix Cref contains only those columns of C. (11. (11.128) If the perturbation is not accompanied by a change in the basis functions then (11. that correspond to the occupied (hole) orbitals.

115). B and Xλ are real this equation simpliﬁes to (A + B)Uλ = −Xλ . the derivative of which is easily calculated.11. we shall consider the derivatives of the reference energy. (11. respectively. with A = AR + iAI etc. the coeﬃcients Uai λ coeﬃcients would require a separate soluActual evaluation of the Uai tion of the CPHF equations for each perturbation.8 The perturbed reference function 393 in which ai and bj are each treated as a single composite index.9.bj Ubj + Bai.133) (11. and using (3. The λ describing the changes in the virtual orbitals are related to coeﬃcients Uia λ by the orthogonality relationship (11. (11.131) or AUλ + BUλ∗ = −Xλ . In the complex case. AR + BR −AI + BI A I + BI A R − BR Uλ R Uλ I =− Xλ R Xλ I .bj Ubj ) = −Xai bj (11. as shown in detail in Section 11.116). (11. of the corresponding matrix or vector. When A. by Gerratt and Mills (1968) and by Pulay (1969). Leaving out the nuclear repulsion term. we write the perturbed reference energy in the form Eref (λ) = i ˆ (λ)|i(λ) + i(λ)|h 1 2 ij i(λ)j (λ) i(λ)j (λ) .134) where the subscripts R and I indicate the real and imaginary parts.111) can then be rewritten as λ λ∗ λ (Aai.132) which is the equation used in most applications. (11. Practical procedures for evaluating these derivatives were ﬁrst developed. primarily for the calculation of SCF energy gradients and force constants. Before proceeding to discuss the correlation-energy derivatives in Section 11.132) can be transformed into a set of coupled equations for the real and imaginary parts of Uλ . but this multiple solving can be avoided by use of the interchange theorem. The CPHF equations (11.9 below.135) .

= i ˆ λ |i + i|h i ˆ|i + i|f ˆ|i[λ] i[λ] |f Using (11. we expand the orbital derivatives in terms of the basis functions and their derivatives.117). where the Coulomb and exchange operators J Even though we are assuming that the reference function is canonical.115) this becomes λ Eref = i ˆ λ |i + i|h i ˆ|i + i|f ˆ|i[λ] i[λ] |f − i λ∗ Sii i.394 Calculation of properties in coupled-cluster theory The derivative of this expression is given by λ = Eref i ˆ |i + i|h ˆ λ |i + i|h ˆ |iλ iλ |h 1 2 ij + = iλ j ij + ij λ ij + ij iλ j + ij ij λ ˆ |i + i|h ˆ λ |i + i|h ˆ |iλ iλ |h + ij iλ j ij + ij iλ j ˆj − K ˆ j iλ J j i = i ˆ+ ˆ λ |i + iλ h i|h j ˆ+ ˆj − K ˆj i + i h J . The ﬁrst sum in this expression is referred to as the Hellmann–Feynman term. Instead.136) ˆ and K ˆ are given by (11. (11.107) and its derivatives are computed from the derivatives of the basis-set orbital integrals. because the unperturbed HF equation is not valid in the space spanned by the perturbed orbitals. The last sum can be expressed in terms of an energy-weighted density matrix Π: λ∗ Sii i i = µν i ∗ Cµi Cνi i µ|ν λ = µν Πνµ µ|ν λ .137) + i λ∗ λ Uii + Uii i. (11. = i ˆ λ |i + iλ |f ˆ|i + i|f ˆ|iλ i|h (11. it ˆ|i by i |i in the last form would be incorrect at this point to replace f of (11. (11. and obtain λ = Eref i ˆ λ |i + i|h i ˆ|i + i|f ˆ|i[λ] i[λ] |f i + i p λ∗ λ Upi p|i + i|p Upi (11.138) where the transformed overlap matrix S (λ) is given by (11.96).136).139) .

in which the orbitals are obtained from an SCF optimization for a diﬀerent electronic state). dλ ∂λ ∂ P dλ (11.142) that the solution of the CPHF equations is not needed for the determination of the reference-energy derivative. (11. In this case it may be seen from (11. Eref and the reference-function density matrix P: λ = Eref µν ˆ |ν Pνµ µ|h λ − Πνµ µ|ν λ ) + 1 2 µνστ Pνµ Pτ σ µσ ντ λ . only the Hellmann–Feynman term. the energy derivative becomes λ = Eref i ˆ λ |i + i[λ] |h ˆ |i + i|h ˆ |i[λ] i|h i[λ] j ij + ij i[λ] j ij + = i − µν [λ] Πνµ µ|ν − µν λ ˆ |i i|h [λ] + 1 2 ij ij ij Πνµ µ|ν λ . Furthermore.141) In order to avoid a full transformation of the derivative two-electron inteλ is usually calculated directly from the basis-set derivative integrals grals. (11. λ = ∆P Eref µν ˆ |ν P λ + µ|h νµ 1 2 µνστ λ λ µσ ντ Pνµ Pτ σ + Pνµ Pτ σ . in more general cases in which the orbitals are not variationally optimized for the state in question (such as for quasi-HF.142) This derivation assumes that the reference function has been variationally optimized.8 The perturbed reference function 395 where Π = Cref εC† ref or Πνµ = i ∗ Cµi Cνi i . for perturbations that do not require changes in the basis functions. (11. ∂Eref ∂Eref dP dEref = + .11. However. the ﬁrst term in (11. as is the case for SCF wave functions.144) .138) in terms of two-electron inteExpanding f grals.138).143) The second term vanishes in the SCF case because the orbitals have been variationally optimized.140) ˆ in the middle sum of (11. In other cases we have an additional contribution to the reference-energy derivative. there is an added term involving the derivative of the reference-function density matrix with respect to the perturbation. survives. (11.

The two-electron derivative integrals in (11.145) ˆ Substituting the response density matrices (11.147). but the interchange-theorem technique to be described in the next section can be used to avoid explicit evaluation of Uλ for each perturbation (Gauss. (11. we obtain ∆E λ = pq λ fpq (γN )qp + 1 4 pqrs λ λ pq rs λ (ΓN )rspq = ∆E1 + ∆E2 .145).105). Considering the two-electron inte- .117).89) into (11. (11. (11. (11.68).148) Here and in the rest of the chapter we omit the subscript N from the normalordered density matrices to simplify the notation.88). and thus requires the solution of the CPHF equations.148) can be separated into contributions from the two terms in the expression for the ﬁrstorder perturbed orbitals {φλ p }.396 Calculation of properties in coupled-cluster theory Evaluation of the density-matrix derivatives requires the knowledge of Uλ . Stanton and Bartlett 1991).146) where λ = ∆E1 pq λ fpq γqp = pq ˆλ |q + pλ |f ˆ|q + p|f ˆ|q λ γqp p|f pk λ qk + pk qk λ + pλ i qk + pk q λ k k = pq p|h |q + ˆ λ |q + p|h pq k ˆλ γqp = pk qk λ γqp . (11.147) λ ∆E2 = 1 4 pqrs pq rs λ Γrspq . as seen from (11.9 The CC correlation-energy derivative The coupled-cluster correlation-energy derivative. (11. 11. (11. see (11. can be written in the form ˆ ˆλ T ˆ ∆E λ = 0|(1 + Λ)e−T H N e |0 = 0|(1 + Λ)e−T ˆ pq λ fpq {p ˆ† q ˆ} + 1 4 pqrs pq rs λ {p ˆ† q ˆ† s ˆr ˆ} eT |0 .

116) to eliminate all but the particle–hole block of Uλ : λ fpr Urq γqp = pqr pia λ fpa Uai γip − pai λ fpb Sba γap − pab 1 2 pij λ∗ fpi Uai γap − pai λ fpj Sji γip λ fpi Sia γap − = 1 2 λ λ∗ fpa γip Uai − − fpi γap Uai pia pai fpi i|a [λ] [λ] γap (11. we have pk qk k λ = kr λ∗ λ λ∗ λ + pk rk Urq + pr qk Urk + pk qr Urk rk qk Urp + k pk qk [λ] = r λ∗ λ frq Urp + + fpr Urq kr λ∗ λ pr qk Urk + pk qr Urk + k pk qk [λ] .151) − 1 2 pab fpb b|a [λ] γap − 1 2 pij fpj j |i γip .152) fbp a|b pab [λ] γpa − 1 2 pij fjp i|j γpi .s of (11. (11.11. and separating it into contributions from the diﬀerent blocks of Uλ .150) We next use (11.149) produces λ∗ frq Urp γqp = pqr qpr λ∗ frp Urq γpq λ∗ λ fap γpi Uai − − fip γpa Uai ipa pai = − 1 2 fip a|i [λ] [λ] γpa (11.h. (11. we obtain λ fpr Urq γqp = pqr pia λ fpa Uai γip + pai λ fpi Uia γap + pab λ fpb Uba γap + pij λ fpj Uji γip .9 The CC correlation-energy derivative 397 gral in (11.149) Taking the second term in the ﬁrst sum on the r.h.s of (11. (11. Similar treatment for the ﬁrst term in the ﬁrst sum on the r.147) ﬁrst. .115).149) as an example.

respectively. (11.147) can now be written as λ = ∆E1 pq ˆ λ |q γqp + ∆E U + ∆E . by the interchange of dummy summation U and indices. (11. . pq + p (fap γpi − fpi γap ) + (11. that depend on Denoting the sum of the contributions to ∆E1 λ U the particle–hole block of U by ∆E1 . Both ∆E1 [λ] ∆E1 are simpliﬁed for canonical HF orbitals by the relation fpq = εp δpq . we collect all such terms in (11.157) The sums involving b|a [λ] and j |i [λ] were combined with those involving a|b [λ] and i|j [λ] .155) Equation (11.151)– (11.149) result in λ pk qr Urk γqp = pqrk qpri λ pi qr Uri γqp λ pi qa γqp Uai − pqai λ∗ pr qk Urk γqp = pqrk pqri λ∗ qr pi Uri γpq λ∗ qa pi γpq Uai − pqai 1 2 pqji 1 2 pqji = pi qj j |i [λ] γqp .154) and obtain U = ∆E1 ia p (fpa γip − fip γpa ) + pq λ pi qa γqp Uai λ∗ qa pi γpq Uai .154).153) = qj pi i|j [λ] γpq .398 Calculation of properties in coupled-cluster theory The terms in the second sum of (11. including the interchange of p and q in (11. p|h 1 1 [λ] (11. (11. (11.154) λ .147).156) where ∆E1 = − ia p [λ] fpi γap 1 2 ab p 1 2 ij p i|a [λ] + p fip γpa [λ] a|i [λ] − − + fpa γbp + fbp γpa a|b fpi γjp + fjp γpi + pj qi γqp i|j pq [λ] pi qi pq i [λ] γqp .

∗ 1 ΓH pqrs = 2 (Γpqrs + Γrspq ) .11.159) ∆E2 = − 1 2 as pqr [λ] pq rs Γrapq s|a [λ] + pqr [λ] rs pq Γpqra + pqrs a|s [λ] [λ] − 1 2 pqrij rj pq Γpqri + pq ri Γrjpq i|j pq rs Γrspq . given by ∆E λ = pq p|h 1 2 A common feature of almost all terms in the expansion of the ﬁrst-order energy is that if we ignore the non-Hermiticity of the response density matrices then they come in complex-conjugate pairs.9 The CC correlation-energy derivative 399 A similar treatment of the two-electron contribution to ∆E λ results in λ U = ∆E2 + ∆E 2 . pqr + (11.164) + . ∗ 1 ΓA pqrs = 2 (Γpqrs − Γrspq ) . It is convenient to split the response density matrices into Hermitian and anti-Hermitian components: H A H∗ A∗ + γpq = γqp − γqp .163) Deﬁning the perturbation-independent vectors M and N. pqr H pi qa γqp (11. ∆E2 [λ] (11.158) U =1 ∆E2 2 ai pqr λ pq ra Γripq − ri pq Γpqra Uai λ∗ ra pq Γpqri − pq ri Γrapq Uai . (11. (11. and then the entire ﬁrst-order perturbed energy is ˆ λ |q + ∆E U + ∆E U . A ∗ =1 γpq 2 (γpq − γqp ) .161) where H ∗ γpq =1 2 (γpq + γqp ) . γpq = γpq A H∗ A∗ Γpqrs = ΓH pqrs + Γpqrs = Γrspq − Γrspq . with components Mai = p H H fpa γip − fip γpa + q H pq ra ΓH ripq − ri pq Γpqra .160) For perturbations that are not accompanied by a change in the basis functions we have ∆E [λ] = 0.162) (11. (11.

133) for each perturbation we solve a single perturbation-independent equation. (11. Hirschfelder. the quantity needed for the evaluation of ∆E U is the contraction 2MT Uλ . this is sometimes referred to in the CPHF context as the Z -vector method. (A more general use of the interchange theorem for the correlation problem is represented by the introduction of the Λ equations in Section 11. (11. can be avoided by use of the interchange theorem (Sternheimer and Foley 1953. Trucks and Bartlett (1989)).8.167) and the vector N and the anti-Hermitian components of the response density matrices are not needed for the calculation of ∆E U . Byers-Brown and Epstein 1964. see Adamowicz.168) Then the contractions of the solution vector Z with the right-hand-side vectors −Xλ for the various perturbations. the explicit calculation of Uλ . −ZT Xλ = −2MT (A + B)−1 Xλ = 2MT Uλ . Considering ﬁrst the case in which all quantities are real. M = MR + iMI etc.400 Calculation of properties in coupled-cluster theory Nai = p A A fpa γip − fip γpa + q A pi qa γqp + pqr A pq ra ΓA ripq − ri pq Γpqra (11. Laidig and Bartlett (1984) and Salter. we can write the Uλ -dependent part of the ﬁrst-order energy as U U ∆E U = ∆E1 + ∆E2 λ λ∗ (Mai + Nai )Uai + (Mai − Nai )∗ Uai ai = (11. Dalgarno and Stewart 1958. which would require repeated solution of the CPHF equations for diﬀerent perturbations.165) (with ai treated as a single composite index).166) T λ † λ∗ = (M + N) U + (M − N) U λ T λ T λ T λ = 2(MT R UR − MI UI ) + 2i(NR UI + NI UR ) . As stated in Section 11.169) . Handy and Schaefer 1984). To obtain this result. this result simpliﬁes to ∆E U = 2 ai λ Mai Uai = 2MT Uλ . (A + B)T Z = 2M . instead of solving (11.167).5. (11. (11. When all quantities are real. where the superscript T indicates the transpose and subscripts R and I indicate the real and imaginary parts of the respective vectors.

being an artefact of the non-Hermiticity of the response density matrices. This form can be used to deﬁne a relaxed density matrix D that incorporates the eﬀects of orbital relaxation. described in connection with the treatment of the Uλ equations. such a result is not normally ergy. In any case. replacing them by their Hermitian average is sometimes useful. Though the non-Hermitian forms can be used.129). the truncated-CC unperturbed enˆ HP ˆ . The same techniques involving the interchange theorem. when the perturbation does not involve changes in ˆ λ |i = hλ . at least when real orbitals are used for ground-state molecules near their equilibrium geometry.170) where hλ is treated as a vector with components identiﬁed by the composite index ai. (11. For this case ∆E λ is the basis functions we have Xai = a|H ai given completely by ∆E U . . and when all quantities are real we have ∆E U = 2MT U λ = −2MT (A − B)−1 hλ . As given in (11. + pqr pq rb Γrjpq − rj pq Γpqrb bj. are used in its calculation to avoid the explicit inversion of A − B. As will be discussed in Section 12.172) ˆ can then be obtained as ˆλ = O One-electron properties given by H ON = ai Dia oai . (11. for which Dia = −2 bj p fpb γjp − fjp γpb + pq pj qb γqp (A − B)−1 .171) so that ∆E U = ai Dia hλ ai .8. may be complex. and it is reasonable to ignore it. Generalization of this treatment to the case of complex functions is straightforward. computed as P expected.9 The CC correlation-energy derivative 401 provide the required answers. however.173) The relaxed density matrix is perturbation independent and needs to be computed once only for any number of perturbations. (11.11.ai (11. the imaginary component of any CC energy-derivative result is unphysical.

represented by the CPHF equations (11. λ T λ MT Uλ + MT∗ Uλ∗ = 2(MT R UR − MI UI ) (11. Such a situation can arise when the HF solution is unstaˇ ıˇ ble with respect to a lower-energy symmetry-broken HF solution (C´ zek and Paldus 1967. As seen in (11. AR + BR −AI + BI A I + BI and AR + BR −AI + BI A I + BI A R − BR T T ZR ZI A R − BR = MR −MI (11. the right-hand-side vector −Xλ of the CPHF equations (11.402 Calculation of properties in coupled-cluster theory Even if the A and B matrices are real. will now be given. with zero ﬁrst-order contributions. also referred to as gauge invariant atomic orbitals ) are used as basis functions (London 1937. and thus Xλ and Uλ . Sekino and Bartlett 1986). Another example arises when gauge-including atomic orbitals (GIAOs. such as those due to magnetic susceptibility or nuclear magnetic shielding (used in the evaluation of NMR chemical shifts). the use of complex orbitals may be useful for spatial symmetry adaptation and may result in complex A and B matrices. including the anti-Hermitian contributions. In calculations on atoms. Cnh and Snh (n > 2).176) ZR ZI = NI NR . Ditchﬁeld 1974. In such cases the basis-set-integral derivatives.175) These contributions to the correlation-energy derivative can be obtained by solving two sets of coupled equations. (11. and the second-order (and all even-order) energy derivatives are real (Gauss 1993). linear molecules and molecules belonging to degenerate irreducible representations of axial point groups such as Cn . For the sake of completeness an analysis of the complex case.166). leading to complex Uλ vectors. (11. in the complex case we need two contractions.134). Ruud and Helgaker 1996) in calculations involving some magnetic perturbations.177) . Gauss.174) and λ T λ NT Uλ − NT∗ Uλ∗ = 2i(NT I UR + NR UI ) . are pure imaginary but these properties are of second order.132) may still be complex.

11. its contribution to ∆E U is pure imaginary and. (11. while mathematically it constitutes a correct part of the derivative of the CC energy.9 The CC correlation-energy derivative 403 Then −2 ZT R ZT I Xλ R Xλ I = −2 MT R −MT I AR + BR −AI + BI A I + BI Uλ R Uλ I AR + BR −AI + BI A I + BI Uλ R Uλ I A R − BR A R − BR −1 Xλ R Xλ I T = 2 MT R −MI (11. The anti-Hermitian part of the response density matrices is of at least third order in MBPT (Section 11.183) . (A + B)−1 Xλ R (A − B)−1 Xλ I Uλ R Uλ I .177) become A+B and A−B T T ZR = M (11.7) and would normally be very small.178) and −2i T ZR ZIT Xλ R Xλ I −1 = −2i = 2i NT I NT R NT R Xλ R Xλ I NT I (11. (11.180) ZI = N . as in the case of magnetic perturbations with GIAO orbitals.181). In this case (11. are complex. it is unphysical and therefore is usually ignored: the vector N is not calculated and the corresponding equations (11.177) are not solved. and (11. we would only have (11.182) then becomes T −1 λ T λ −2iZI T Xλ I = −2iN (A − B) XI = 2iN UI . and thus the Xλ vector.181) The contribution to the correlation-energy derivative is then given by −2 ZT R iZIT Xλ R Xλ I = −2 MT iNT (11. (11.179) provide the required answer. It is also instructive to look at the case in which all unperturbed quantities are real but the perturbed orbitals. As previously stated.176).182) = 2 MT iNT If Xλ is pure imaginary.

181). µpq = rst ηpq = r γpr frq . rst the ∆E [λ] contribution is obtained as (ξia + 1 2 µia ) a|i ai 1 2 ab 1 2 ij [λ] ∆E [λ] = − − − + pq + (ηai + 1 2 νai ) i|a [λ] [λ] (ξba + ηba + µba + νba ) a|b (ξji + ηji + µji + νji + 2ζji ) i|j pi qi i [λ] [λ] γqp + pqrs pq rs [λ] Γrspq . (11. For this purpose it is useful to note that η H = ξ H† . (11. To avoid repeated four-index transformations of two-electron derivative integrals from basis-function to orbital representations in the last two sums of (11. If the basis functions are modiﬁed by the perturbation. (11. the corresponding sums are best evaluated in the basis-function representation. the correct result for the ﬁrst-order energy being zero. the resulting imaginary energy derivative is unphysical.185). plays a role in the second-order calculation. by transforming the response density matrices from the orbital representation to the basis-set representation: γνµ = pq ∗ Cµp Cνq γqp . ζ H = ζ H† . we have to determine the ∆E [λ] contributions. Γtprs rs tq . and the ﬁrst true nonzero contribution is of second order. the anti-Hermitian terms will generate imaginary contributions to ∆E [λ] and would not normally be calculated. In this case we do not need to solve (11.184) tp rs Γrstq .185) This expression can also be separated into contributions from the Hermitian and anti-Hermitian parts of the response density matrices. Collecting the terms from (11. and deﬁning the perturbation-independent contractions ξpq = r fpr γrq . but Uλ . so that N is not needed. which is pure imaginary. (11. η A = −ξ A† .160). When all quantities are real the anti-Hermitian contributions cancel.404 Calculation of properties in coupled-cluster theory Again. ν H = µH† .187) .157) and (11. νpq = ζji = rs rj si γsr . ν A = −µA† . ζ A = −ζ A† .186) As in the case of ∆E U .

are described in Section 13.9 The CC correlation-energy derivative 405 Γστ µν = pqrs ∗ ∗ Cµp Cνq Cσr Cτ s Γrspq . (11. the total correlation-energy derivative is made up of the diﬀerent contributions: ∆E λ = pq ˆ λ |q γqp + ∆E U + ∆E [λ] p|h ˆ λ |ν γνµ + ∆E U + ∆E [λ] . µ|h µν = (11. Such methods for CC higher-order properties involving external ﬁelds. (11. = Dτ σ γνµ = µν (11. Then pi qi pqi [λ] γqp = µνστ µσ ντ µσ ντ µνστ λ i λ ∗ Cσi Cτ i pq ∗ Cµp Cνq γqp λ Yµν γνµ .5.189) where Yλ is given by (11. including second and higher derivatives.126) and pq rs pqrs [λ] Γrspq = µνστ µν στ µν στ µνστ λ pqrs λ ∗ ∗ ∗ Cµp Cνq Cσr Cτ s Γrspq = Γστ µν .190) This approach is particularly important in the calculation of energy gradients.188) where we should remember that. Extensions of the treatment described here can be used to obtain higherorder properties.171) can be used to compute the ∆E U term (11. respectively. even though the subscript N has been omitted. the response density matrices γ and Γ are the normal-ordered matrices γ N and ΓN . using EOM–CC methodology. each in up to three directions) need to be considered. Finally.191) The relaxed density matrix (11. .11. for which many diﬀerent perturbations (displacements of the diﬀerent atoms.172).

For spin summation we 406 . quantum number.12 Additional aspects of coupled-cluster theory This chapter addresses several more subtle but nevertheless important aspects of coupled-cluster MBPT theory. integration over spin is absolutely necessary to eliminate many integrals and to allow the introduction of constraints over the summation indices. The incorporation of spin integration can be done algebraically or. Even in the case of unrestricted Hartree–Fock (UHF) reference functions. Furthermore. In the restricted closed-shell Hartree–Fock (RHF) case. deriving expressions in terms of spatial orbitals only and thus reducing the range of all indices by about a factor 2 (see Section 7. without regard to integration over spin coordinates. spatial-orbital.1 Spin summations and computational considerations The formalism described in the previous sections was presented in terms of spinorbitals.3). because the most eﬀective algorithms are still of the spin-integrated. but not the S . The diagrammatic representation of this equation in a spinorbital basis was given in Fig. 12. in which the spatial orbitals for α and β spin are diﬀerent. all amplitudes in which the number of α and β spinorbitals is diﬀerent for the hole and particle indices vanish. preserving the MS . diagrammatically. despite double occupancy. in some cases. form. Restricted open-shell Hartree–Fock (ROHF) calculations are usually performed as UHF. As an example of the diagrammatic treatment of spin summation in coupled-cluster calculations we shall consider the case of the CCD equation with an RHF reference function.2 in terms of antisymmetrized Goldstone diagrams. spin integration is used to combine contributions from α and β spinorbitals. achieving a computational eﬀort of no more than three times that of comparable RHF calculations. The double occupancy cannot be exploited further without special eﬀort. 9.

The labels in this ﬁgure reﬂect the origin of each diagram in terms of a diagram of Fig. e. . as for MBPT Goldstone diagrams. just before (9. J. 12. indicated by capital letters I. diagrams D3b − D3bx5 all originate in the antisymmetrized Goldstone diagram D3b . Diagrams for the spin-summed CCD amplitude equations for an RHF reference function. This correspondence can be veriﬁed by the fact that all six of these diagrams collapse to the same Hugenholtz diagram (labeled Qb in Section 9.5. 8). but the numerical weight factors and the permutation factors are diﬀerent. symmetry with respect to reﬂection in a vertical line). as shown in Fig. For example. for holes and A. an x in the label indicates exchange. each spatial orbital reﬂects the existence of two spinorbitals having the same spatial part. There is a weight factor 2 for each closed loop (reﬂecting summation over two spin assignments) and a weight factor 1 2 for diagrams with left–right symmetry (i. . The sign rule is the same. Iα and Iβ .e.1.2. 9. diagram D3b has two closed loops and × D1 D2a D2b × D2c D2d D2e D2ex1 D2ex2 D2ex3 D3a D3b D3bx1 D3bx2 D3bx3 D3bx4 D3bx5 D3c D3cx D3d D3dx Fig.1 Spin summations and computational considerations 407 expand these diagrams in terms of ordinary Goldstone diagrams representing ˇ ıˇ spatial orbitals (C´ zek 1969).12. The interpretation of these diagrams diﬀers in some respects from that of the antisymmetrized diagrams.121)) when each vertex is collapsed to a dot. 12. B. The particle and hole lines in the diagrams of Fig. . for particles. . 12.1 (Bartlett and Musial 2007. . .g.1 represent spatial orbitals. In the RHF case considered here. Fig. For example. .

and therefore has weight 1 2. For the permutation factor (which does not appear in MBPT diagrams). for a total weight equal to 2 × 2 × 1 2 = 2. is AB |v ˆ|IJ ˆ ((IA)(JB )) +P C fBC tAC IJ − K fKJ tAB IK 1 2 KL|v ˆ|IJ tAB KL KL 1 + 2 + AB |v ˆ|CD tCD IJ + CD ˆ|CJ tAC 2 KB |v ˆ|CJ tAC IK − KB |v KI KC − KB |v ˆ|IC tAC ˆ|JC tAC KJ − KB |v IK .1. BA BA AB tAB IJ = tJI = tIJ = tJI . 12.2) The ﬁnal form of the algebraic amplitude equations for CCD. we note that the spin-free amplitudes do not have the same antisymmetry property as in the spinorbital form. P (12. Special attention needs to be paid to diagram D3a .408 Additional aspects of coupled-cluster theory left–right symmetry. while as drawn in Fig. it can also be drawn in a symmetric form. 12. while diagram D3bx3 has one closed loop and symmetry. Therefore we may only permute (IA) as a unit with (JB ). with the terms ordered as in Fig.1 this diagram does not show left–right symmetry. using the permutation factor ˆ ((IA)(JB )) = 1 + PIJ PAB . giving a total weight 1. . (12.1) and so for the tAB IJ diagram I and A must remain on the same continuous path and so must J and B .

3) The summations over internal lines are over spatial orbitals.6) BA The two independent amplitudes tAB IJ and tIJ serve respectively as coeﬃcients for two sums of four Slater determinants each.3) for the case in which the reference determinant is constructed solely from doubly occupied spatial orbitals shows the following relationships between the corresponding amplitudes: AαBβ AβBα BA tAB IJ = tIαJβ = tIβJα = tJI . the CCD correlation energy is given by ∆ECCD = = IJAB + 2 IJ |v ˆ|AB − IJ |v ˆ|BA tAB IJ . ≡ Bα Aα JαIα|0 + Bα Aβ Jβ Iα|0 |ΦBA IJ + Bβ Aα JαIβ |0 + Bβ Aβ Jβ Iβ |0 . (12.4) Careful comparison of the spinorbital CCD equation (9. For diagrams D2c and D2d the permutation factor can be canceled with the weight factor. B ).8) where |0 is the reference determinant constructed from doubly occupied orbitals only. AβBβ AB BA tAαBα IαJα = tIβJβ = tIJ − tIJ .7) (12. Using the same interpretation rules. ≡ Aα Bα JαIα|0 + Aα Bβ Jβ Iα|0 |ΦAB IJ + Aβ Bα JαIβ |0 + Aβ Bβ Jβ Iβ |0 . because the permuted form is equal to the original.5) (12. Each of these two linear combinations of Slater determinants .126) and the spinfree CCD equation (12. † † † † † † † † † † † † † † † † (12. (12.12. (12.1 Spin summations and computational considerations 409 + KLCD KL|v ˆ|CD 1 CD AB 2 tIJ tKL DB AC DB + 2tAC IK tLJ − 2tIK tJL CA BD AD CB AD CB +1 2 tIK tLJ − tIK tLJ + tKI tLJ CB AD CD AB CD AB +1 2 tIL tKJ − 2tKI tLJ + tIK tLJ DB AC DB − 2tCA KL tIJ + tKL tIJ =0 (for all I ≥ J and all A.

These types of state require more sophisticated spin-adapted theories.4. because the ﬁrst determinant in (12. resulting in an overlap BA = −2 ΦAB IJ |ΦIJ (12. However. such as are found in quasi-restricted Hartree–Fock (QRHF) (Rittby and Bartlett 1988) or the Brueckner orbital representation to be described in Section 12. because they require a multideterminantal zero-order function. Spin integration in UHF-based CCD calculations is described in terms of antisymmetrized diagrams by Bartlett and Musial (2007.11) (12. only algebraic procedures are easily applicable. including the treatment of open-shell systems. 1992. such as the . II |ΦAA ≡ Aα Aβ Iβ Iα|0 . |ΦAB ≡ Aα Bβ Iβ Iα|0 + Aβ Bα IαIβ |0 . II † † † † † † † † † † (12. For more general CC treatments using diﬀerent orbitals for diﬀerent spins. Li and Paldus 1997). this is the approach used for all cases in the ACES program system (Stanton.9) AB = ΦBA |ΦBA = (note that these functions are not normalized. Watts et al. these two spin eigenfunctions are not orthogonal. A fully spin-adapted and orthogonal coupled-cluster formalism. ΦAB IJ |ΦIJ IJ IJ 4). Fig. Gauss. including open-shell UHF cases. which explains the exclusion of amplitudes such as tAβBα IαJβ and justiﬁes the requirement that spin be conserved along each continuous path in the diagrams. Low-spin open-shell states are not accessible in this way. For ROHF cases the approach described here may lead to functions that are not spin eigenfunctions.10) (12. Any other doubly excited singlet-spin eigenfunction that can be constructed involving the same four orbitals would be linearly dependent on these two functions. This approach readily permits the treatment of high-spin open-shell molecules with either UHF or ROHF reference functions as well as the use of even more exotic reference functions. 9).410 Additional aspects of coupled-cluster theory is a spin eigenfunction with S = 0. The cases A = B and/or I = J are simpler.12) The situation is more complicated for higher excitations and for openshell states.8) is equal to the negative of the ﬁrst determinant in (12. and the same is true of the last determinant in the two equations. is provided by the orthogonally spin-adapted coupled-cluster method (Paldus 1977. resulting in one spin-free amplitude each and providing coeﬃcients for the following singlet-spin eigenfunctions: |ΦAA IJ ≡ Aα Aβ Jβ Iα|0 + Aβ Aα JαIβ |0 . Adams and Paldus 1979.7). Bartlett and Watts 1998).

Szalay and Bartlett 1994). with its computational simpliﬁcations. In all cases.g. 1992) for CCSDTQ.12. generalizations to arbitrary reference functions often may not be given due consideration. In this and the next sections we shall also show that using CC theory with an arbitrary reference function can lead to a generalized MBPT treatment (GMBPT) that provides the ﬁnite-order perturbation-theory approximations as well as the CC solution. Because the use of a canonical HF reference function. (12. careful organization of the calculations. Open-shell singlets can be treated within a spinorbital context by a two-determinant-based CC theory (Balkov´ a and Bartlett 1992.2 Arbitrary single-determinant reference function 411 orthogonally spin-adapted approach mentioned above. which reﬂect the non-HF nature of the reference function. as well as the oﬀ-diagonal fij and . for which fai = 0.7. is so pervasive in CC and MBPT applications. localized HF orbitals. which require a block-diagonal Fock matrix. 12. ˆN = H pq fpq {p ˆ† q ˆ} + 1 4 pqrs pq rs {p ˆ† q ˆ† s ˆr ˆ} . as would be the case for e. fai = 0. Examples of the use of intermediates in CC calculations were presented by Noga and Bartlett (1987) for CCSDT and by Kucharski and Bartlett (1991a. These techniques are discussed further in Section 10. Open-shell states can also be treated by the various EOM–CC techniques discussed in Chapter 13. without requiring the individual hole–hole and particle–particle blocks to be diagonal. the deﬁnition of appropriate intermediate sums and other summation techniques are important in keeping the computational work within reasonable bounds.13) In the general case we have to retain the fai elements. or they can be more arbitrary non-HF single determinants. The normal-ordered Hamiltonian retains the same general form for any choice of orbitals.2 Coupled-cluster theory with an arbitrary single-determinant reference function The maximum ﬂexibility and range of applicability of the coupled-cluster method often derives from the use of reference functions that are not canonical Hartree–Fock determinants. In the discussions of MBPT and CC methods in this book we have always tried to include the general case. Such reference functions may be noncanonical HF solutions.

(12. CCSDT-3 or CCSDT calculations this . in which the zero-order part includes the oﬀ-diagonal elements of the hole–hole and particle–particle blocks of the Fock matrix: ˆ 0 )N + V ˆN .412 Additional aspects of coupled-cluster theory fab elements. the oﬀ-diagonal × interaction does not vanish. ˆ N = (H H where ˆ 0 )N = (H p (12. the ﬁrst term on the r.17) In earlier descriptions of MBPT procedures and of the iterative solutions ˆ d )N . it has three weaknesses. For non-HF and noncanon- 3n 4 ical HF orbitals the evaluation of this diagram would require an ∼ nh p procedure. (12. 10. was used as the zero-order normal-product Hamiltonian.5) can cause computational diﬃculties because.e.16) ˆN = V ai i} + fia {ˆ fai {a ˆ†ˆ i† a ˆ} + pq rs {p ˆ† q ˆ† s ˆr ˆ} .h. in the CCSDT equations (diagram T2a of Fig.17) but not by a separation using (H 0 0 increases the likelihood that large oﬀ-diagonal Fock-matrix elements will be included in the perturbation and will contribute to poor convergence of the iterative process. only the diagonal part (H 0 i. using (H (12. First. this leads to the detailed form ˆN = H p fpp {p ˆ† p ˆ} + i=j j} + fij {ˆ i†ˆ a=b fab {a ˆ †ˆ b} 1 pq rs {p ˆ† q ˆ† s ˆr ˆ} . Second. of the CC equations (see Chapters 9 and 10). This requirement is satisﬁed by the separation in ˆ d )N ˆ d )N . In iterative CCSDT-1.16). While this approach is practical and common.14) + ai i} + fia {ˆ fai {a ˆ†ˆ i† a ˆ} + We can separate this form into a zero-order part and a perturbation.s. Third.15) fpp {p ˆ† p ˆ} + i=j j} + fij {ˆ i† ˆ a=b 1 4 pqrs fab {a ˆ †ˆ b} . all other terms were treated as part of the perturbation.15)–(12. a diagram like ×. of (12. in many-body theory we prefer the results to be invariant under linear transformations of the occupied orbitals or the unoccupied orbitals separately. 4 pqrs (12. unlike the situation in the canonical HF case.

and thereby the CC approach oﬀers a far more eﬀective evaluation of these contributions than individual MBPT diagrams. fij = fii δij . of the iterative CC equations are those that represent fia and fai interactions.h. One interesting possible choice is the Kohn–Sham determinant of densityfunctional theory (Kohn and Sham 1965). CC results at the CCSD level and beyond are much less sensitive than the corresponding CI results to the choice of reference determinant. deﬁned by the requirement that the Fock matrix takes the form . the hole–hole and particle–particle blocks of the Fock matrix are diagonal. Gauss and Bartlett 1993).s. The hole–particle and particle–hole blocks are changed by the semicanonical transformation but do not vanish. fia = 0. . but for a noniterative CCSD(T) calculation its oﬀ-diagonal part cannot be ignored and would become a serious limitation (Watts. Another possible choice is a determinant composed of the N highest-occupancy natural spinorbitals (the so-called ﬁrst natural conﬁguration). and thus the diﬃculties of dealing with the corresponding oﬀ-diagonal elements are avoided. which has the eﬀect of mixing the occupied and virtual the perturbation V spaces. and remain as a part of ˆN . fab = faa δab .2 Arbitrary single-determinant reference function 413 diagram is simply incorporated in the generalized intermediates without undue concern. With the semicanonical transformation. All three weaknesses can be readily removed by exploiting the invariance of CC theory under separate unitary transformations of the occupied orbitals and of the unoccupied orbitals. The iterations of the CC equations indirectly introduce all the noncanonical HF diagrams that would proliferate excessively in standard MBPT. This is the basis for the GMBPT method to be discussed in the next section.12. which are easy to deal with in CC theory (unlike in MBPT) because they add very few diagrams. the only one-electron vertices that need to be included in the diagrams on the r. fai = 0. which states that any Slater determinant in a given Hilbert space can be generated from any other Slater determinant in that space by the application of an operator of the ˆ form eT1 . For the semicanonical orbitals. This property of the CC method opens the door for the eﬀective use of many convenient single-determinant choices as bases for CC expansions. Because of the Thouless theorem (Thouless 1960). for which the density matrix has maximum overlap with the exact one-particle density matrix (L¨ owdin 1955). This objective is accomplished by a unitary transformation of the spinorbitals to semicanonical form. which is formulated to provide the exact electron density for the system.

Another case of interest is the use of a so-called quasi-RHF (QRHF) reference function (Rittby and Bartlett 1988). as well as for ROHF–CCSD gradients (Gauss. The choice of an ROHF reference function can be beneﬁcial in many cases. For example. A particular case of substantial interest is the use of a restricted openshell Hartree–Fock (ROHF) reference function (Rittby and Bartlett 1988). 1991). this approach to open-shell problems employs an HF calculation that enforces all spin conditions. Lauderdale. For ROHF–CCSD(T) (Watts. 1991). as the full-CI level is approached. their occupancy in the reference determinant being adjusted to describe the desired hole state. SCF orbitals from a related system are used to treat the system of interest. the ROHF determinant has the maximum number of doubly occupied spatial orbitals consistent with the required multiplicity. In particular. this is followed by a semicanonical transformation to diagonalize the hole–hole and particle–particle blocks. Gauss et al. the dependence of the results on the choice of orbitals disappears. Unlike unrestricted HF (UHF). including treatments of the excited states of radicals by the EOM–CC method described in Chapter 13 (Hirata. or the the HF orbitals of any of these systems can be used to study the others. because of the Thouless theorem. and often all or some point-group symmetry requirements. In this approach. QRHF orbitals are very useful when studying hole states of ions created by the excitation of an electron from an orbital other than the highest occupied. Nooijen and Bartlett 2000a). In particular. This QRHF reference function is in no way variationally optimal for the problem addressed. As a result of this transformation. the HF orbitals of CH+ 3 (a closed-shell system) can be used to study the q CH3 radical or the CH− 3 ion. If the orbitals of the radical are to be used then they would be taken from either the spin-α set or the spin-β set). The HF orbitals of the neutral species can be used in the calculation. Stanton. but that deﬁciency is much less important in CCSD and higher CC levels than in CI and MBPT calculations.414 Additional aspects of coupled-cluster theory Yet another is a determinant composed of Brueckner orbitals (the reference determinant for which all single-excitation contributions to the exact wave function vanish). Gauss and Bartlett 1993) and ROHF– MBPT (Lauderdale. the ROHF orbitals are used to construct the (noncanonical) spinorbital Fock matrix. Stanton et al. Of course. of the exact solution. which has maximum overlap with the exact wave function. the spatial parts of the α and β spinorbitals are no longer equal in pairs and so the CC calculations are carried out in a spinorbital-based formalism in the same way as in a calculation based on a UHF reference function. thus avoiding the potential for variational .

3 Generalized many-body perturbation theory As discussed in Section 9.20) The ﬁrst-order amplitudes are obtained explicitly from the ﬁrst-order CCSD equations (noting that the diagonal fii and faa interactions are of zero order) × + (1) × + (1) = 0. simplify to (faa − fii )ti a(1) + fai = 0 .12.16) of the previous section and a semicanonical transformation.22) (12.23) (12. For example. (12. iterations of the CC equations can easily be used to generate MBPT solutions for the usual case of an HF reference function with a diagonal zero-order Hamiltonian.19) = (1) D1 (2) + (1) D2 (2) . .3 Generalized many-body perturbation theory 415 collapse that would be present if the reference function were to be optimized for the ion. × E (2) D2 D3 (12. (12.21) × + × + = 0. . 12.5) from the corresponding CC iterations for arbitrary reference determinants. (12.5 and in Chapter 10.24) (faa + fbb − fii − fjj )tij ab(1) . for example.18) ∆E = D1 can be expanded in orders of MBPT: ∆E = E (2) + E (3) + E (4) + . × (12. × + + . with the semicanonical transformation. the CC correlation energy. + ab ij = 0 .4. where. order-by-order generalized MBPT (GMBPT) solutions can be obtained for the non-diagonal case (subsection 2. . Using the non-diagonal zero-order Hamiltonian (12. (1) (1) which.

Continuing to third order (compare with the diagonal CCD case in Fig. (1) (12.28) where each skeleton represents the sum of the corresponding diagrams with all distinct arrow assignments.416 Additional aspects of coupled-cluster theory Noting that fii = εi . we obtain × E (3) = (2) D1 (3) + (2) D2 (3) + (1) D3 (3) .10). the GMBPT(3) energy is obtained as × E (3) = + × + .3).9). Substituting the resulting ﬁrst-order and second-order amplitudes. (12. × × 1 = 2 = × ×+× × + + × + × × 1 ab ij fia fjb 2 εab ij 1 1 + b a εi εj = 1 ab ij fia fjb . 9. fii − faa = εa i etc.29) × × × Note the introduction of factorized-denominator MBPT diagrams.30) . (1) (12. (5.. (12. b 2 εa i εj (12. the second-order amplitudes in this equation are obtained explicitly from × + (2) + (1) × + (2) (1) + (1) + × = 0. (5.25) E (2) = × having the same components as the diagonal-case MBPT(2) energy. × + . the insertion of these amplitudes into the second-order energy expression (12. (1) = 0.20) provides the GMBPT(2) approximation.27) (12.26) Knowing the ﬁrst-order amplitudes.

2. with a similar result for = × × 1 2 ×+× . As seen in (12. ◦ ROHF. Their sum is equal to twice the real part of either: × × + × × = Re ab ij fia fjb b εa i εj . unlike the CC energy expression. the above procedure also regains the MBPT . but there is little reason to do so since the accumulated CC amplitudes naturally account for all higher-order diagrams including the proliferating non-HF (fai -dependent) terms. 12. Obviously the latter terms do not appear when canonical HF orbitals are used and. Extensions to higher orders of the explicit expression of the GMBPT energies in terms of MBPT diagrams can be made.30).12. since the semicanonical transformation also eliminates the oﬀdiagonal fij and fab terms. • UHF. × (12.29). GMBPT electron aﬃnity of the CN radical.32) These denominator factorizations and summations are automatically incorporated in CC diagrams.5) to those of (12. and their values are complex conjugates of each other.31) are conjugate (Section 5.3 Generalized many-body perturbation theory 440 Electron aﬃnity (kJ/mol) 420 400 380 360 340 320 GMBPT(2) SDQ-GMBPT(4) GMBPT(3) GMBPT(4) CCSD 417 Fig. (12.31) × The diagrams of (12. the GMBPT expansion for the energy is necessarily Hermitian.

33) The second approach is more general. A comparison of ROHF-based and UHF-based GMBPT results for the electron aﬃnity of the CN radical. In this example. the CCSD results for the two reference functions are almost identical. ˆ1 12. ˆm operators commute with each other. 0|W ES = 0|f N T 1 |0 C + 2 1 (12.7. from those of double-excitation amplitudes. Stanton. 12. making the expresThe number of diagrams involving T sions for the various CC approximations appear unnecessarily formidable.. we may rewrite the Because all T CC equations in the form ˆ2 −T ˆ3−. ES . is shown in Fig. there are some advantages in ˆ1 from the equations by a transformation to orbitals in terms of removing T ˆ1 = 0. The result is a seamless treatment of MBPT for any choice of single-determinant reference functions and orbitals.35) (12. ROHF-based GMBPT provides a satisfactory result even at second order.36) separating the energy contributions of single-excitation amplitudes.2 ((Lauderdale. and we shall consider it ﬁrst. As expected. which T ˆ1 in an eﬀective Hamiltonian L¨ owdin 1962b). Gauss et al. ∆E = 0|H |0 + 0|e−T2 W 1 ˆ ˆ ˆT ˆ 2 |0 C . Expanding the exponentials in . (12. Such orbitals are known as Brueckner orbitals (Nesbet 1958. because of the low sensitivity of CC theory to the choice of orbitals. P ˆ ˆ T ˆ e 2 |0 = E S + E D . However.4 Brueckner orbitals and alternative treatments of T ˆ1 can be quite large..34) (12. while the UHF-based calculation (which is the same as MBPT) does not approach the inﬁnite-order CCSD result even at fourth order. computed as the energy diﬀerence beq tween CN and CN− . ˆ ˆ ˆ e−T ˆ = 0.. Another approach is to hide T ˆ ˆ ˆ1 T H = e −T 1 H Ne . 1991). excessive spin contamination in the UHF reference function causes very slow convergence of the corresponding GMBPT values in comparison with the ROHF case. Q HeT2 +T3+.418 Additional aspects of coupled-cluster theory diagrams for the canonical HF case.. ED . Most of this apparent complexity is removed in the pseudolinear form of the equations through the use of the intermediates (eﬀective-Hamiltonian diagrams) introduced in Section 10.

+ 1 (W 2 3! 4! (12. and the exact wave function |Ψ .33). Fig. Kobayashi et al. This transformation involves an n5 spond to the updated T 2 n 4 algorithm computational process.ˆ1 12. H can also be represented conveniently in terms of the basis functions (Koch. Brueckner orbitals can be deﬁned. recomputing the T using H instead of H singles equation.e. that. so none of these contribute to shows no terms with T ˆ N . 1994).37) The one-electron part of this operator is represented diagrammatically by those one-electron components of the eﬀective Hamiltonian H. just like H evaluation is equivalent to a transformation of the original one-electron and ˆ N ) into the elements of H except two-electron integrals (the elements of H ˆ N . that do not contain T ˆ T1 . (10. ˆ1 vertices exclusively. Each iteration requires re-solving the CCD (and CCDT) equations ˆ1 amplitudes from the CCSD ˆ N . 10. χia . Similarly.63)–(10. Christiansen. The equivalence of these conditions can be seen by considering the overlap between |ΦB . by either of two equivalent conditions on the orbitals occupied in the reference determinant: (a) the requirement that single-excitation contributions to the exact wave function vanish (Nesbet 1958) or (b) the requirement that the reference determinant has maximum overlap with the exact wave function (L¨ owdin 1962b). and its H . within a given one-electron Hilbert space.2. ˆ vertices other than χij and χai . For algorithms that use the untransformed basis-set integrals directly. (10. and transforming the elements of H to correˆ1 amplitudes.53)–(10.96). the two-electron part of H is represented by those terms ˆ vertices other than T ˆ1 . H is not Hermitian. the Brueckner reference determinant. Since any reference determinant can be converted to any other determinant constructed in the same one-electron . omitting erator ﬁrst and then use it instead of H ˆ1 and thus H depend on the solution of the ˆ1 -containing terms. T T CC equations and so an iterative procedure is required for the construction of H . which is less demanding than the nh p needed for the solution of the CCD equations and much less than that for CC equations that retain triple and higher excitations.80) that do not contain T of the three-electron and four-electron components of H. χab .4 Brueckner orbitals and alternative treatments of T 419 (12. However. we obtain 1 ˆ ˆ2 ˆ ˆ N + (f ˆ ˆ ˆ H=H N T1 )C + 2 (fN T1 )C + (W T1 )C ˆT ˆ12 )C + 1 (W ˆT ˆ13 )C + 1 (W ˆT ˆ14 )C . i.56). Thus H is a normal-product two-electron operator. Inspection of (10.81)–(10. unlike H A possible strategy for solving the CC equations is to build the H opˆ N in the CC equations.

ˆ2a and S with the ﬁrst-order S Fig.5.31). (10. Cµa = (0) Cµa − i Cµi ta i . 10. the ˆ1 amplitudes from the current T termine a new set of T normal procedure uses the leading terms of the CCSD single-excitation equation.39) 0 = a Ψ|eT1 |0 = Ψ|a i . so that T (b) are equivalent. we can write the unnormalized Brueckner determinant in the form |ΦB = eT1 |0 .40). and the process is repeated by .2: × + + + × + × = 0. 10.40). and comparing them ˆ2b diagrams of the CCSD singles equation. (12. ˆ (12. using the second-order T ˆ1a and T ˆ1b ˆ3 can be deduced from T of T diagrams of the CCSDT triples equations. (11.2. according to Cµi = Cµi + a (0) (0) a Cµa ti . that the ﬁrst approximation ˆ2 anyway. ∂ti showing that single excitations from the Brueckner determinant do not conˆ1 = 0. It should be recognized.40) ˆ2 will be unimportant compared with the The higher-order terms involving T ˆ two T2 diagrams in (12. Fig.99).41) where C is a new transformation matrix (Chiles and Dykstra 1981). The transformed orbitals are orthonormalized. hence conditions (a) and tribute to the exact wave function. To deˆ2 amplitudes. (0) (12. ˆ Once a new set {ta i } of T1 amplitudes has been obtained from (12. If explicit consideration of triples is of interest ˆ3 diagram S7 .420 Additional aspects of coupled-cluster theory ˆ Hilbert space by a one-electron exponential operator of the form eT1 (Thouless 1960). Fig.38) ˆ1 Requiring stationarity of the overlap with respect to the components of T leads to ∂ ˆ ˆ ieT1 |0 = Ψ|(ΦB )a ˆ†ˆ (12. can in the generation of Brueckner orbitals then the T be added. 10. it is used to deﬁne a new set of orbitals by modifying the transformation matrix (0) C(0) deﬁning the current orbitals {φp } in terms of the basis functions {χµ }. Determining the Brueckner orbitals requires an iterative process. however.

for symmetry-broken cases. the eﬀects of many nonlinear terms that are not included in with T ˆ3 and T ˆ4 products. A comparative study of several diﬃcult. the comparative energy of the C2v and D3h forms of the open shell NO3 molecule (Stanton. trices. An advantage of Brueckner-orbital-based CC (B-CC) calculations is that. this procedure can be made nearly as eﬃcient as using predetermined HF or other orbitals in CC calculations. Then (12.ˆ1 12. Fig. unlike the case for UHF or other choices of orbitals that may vary discontinuously as the symmetry of the molecular geometry changes. Obviously.g. potentially multireference. et al. and the importance of the T1 contributions in CCSD (and higher) equations diminishes until they can be neglected altogether. the exact vanishing of the T in B-CC calculations of one-body reduced density matrices and electronic properties (Chapter 11). (1990).4 Brueckner orbitals and alternative treatments of T 421 ˆ2 using just the leading T ˆ1 re-solving the CCSD doubles equations for T contributions from Fig. As this process is iterated. Replogle. problems at the CCSDT level using B and HF orbitals was reported by Watts and Bartlett (1994). all EOM-CC calculations (see Chapter 13) can be performed with B or any other orbitals. e. such as those due to T B-CC calculations and thus do not constrain the accuracy of the B-CCD or ˆ1 contributions B-CCDT results. Grabowski. the T again for new T ˆ tudes get progressively smaller. (12. 2005). Head-Gordon et al. Pople. Pople. With proper organization.40) is solved plus all the T ˆ1 ampliˆ1 amplitudes. + . Furthermore. 10. (1989) and by Raghavachari. . were reported by Handy. Extensive applications of B-CCD (referred to as BD) and B-CCD(T). 9.42) ˆ2 terms of the CCD equations. may contribute to the accuracy of these density maˆ1 terms often play a signiﬁcant role (Bartlett. At this point we have a set of Brueckner orbitals. B-CC has been felt to have some advantages. ˆ1 = 0. in which T Hirata et al. Using B orbitals the two geometries can be described equivalently without the use of broken-symmetry orbitals. including analytical gradients. do not contribute in ˆ1 T ˆ1 T CCSD.2. Gauss and Bartlett 1992).3. Finally. as can all property calculations. besides greatly reducing the number of diagrams that need to be included.

ROHF. KS (Kohn– Sham). An untruncated. ≡ 0|S (12. pure spinorbitals and to be S Single-reference CC theory with an arbitrary single-determinant reference function treats spin passively. S (12.44) Approximate solutions Ψapprox . B (Brueckner).43) (12. The deviation of the computed expectation value from the correct eigenvalue is a useful diagnostic for the accuracy of the solution. in which the spin-eigenstate requirement is enforced in every term (conﬁguration state function) of the CI expansion. H ˆ2 Ψ = S (S + 1)Ψ .45) constitutes a quadratic equation for S in terms of S with the solution (Purvis. satisfying ˆ Ψ = EΨ .46) Satisfying the condition S = S does not generally guarantee an exact wave function. . For such solutions we can derive an approximate value S of the spin quantum number S using the projected ˆ2 proj . corresponding to the spin multiplicity 2S + 1 of the desired electronic state. all determinants contributing to the CC expansion are expected to be composed of ˆz eigenfunctions having the same value of Sz . Sekino and Bartlett 1988) 2S + 1 = ˆ2 1+4 S proj . NO (natural orbitals) or any other single-determinant reference function. Thus an indicator deviate to some extent from being an S ˆ2 of the quality of an open-shell CC solution is provided by computing its S expectation value. For an accurate solution this value should equal S (S +1). However. This deﬁciency is particularly evident for UHF reference functions. Of course. but even for ROHF the truncated CC wave function can ˆ2 eigenfunction.422 Additional aspects of coupled-cluster theory 12. provides a very useful ﬂexibility in the application of CC methods to diverse problems. do not necessarily satisfy the spin eigenvalue equation (12.44). QRHF. UHF. but for open-shell states the amount by which S diﬀers from the exact quantum number S is normally a useful measure of the quality of the solution. applications of these methods to open-shell states do not always result in pure-spin solutions. deﬁned by value S ˆ2 S proj ˆ2 |Ψapprox = S (S + 1) . (12.5 Monitoring multiplicities in open-shell coupled-cluster calculations The ability to use RHF. ˆ2 Equation (12.45) proj .2 for example. as discussed in Section 12. converged CC wave function is a spin eigenfunction. such as in truncated CC. as opposed to most CI calculations.

2 ∆ˆ (12. In any wave function which is an eigenfunction of S ˆ determinants have the same values of nα and of nβ . Sekino and Bartlett 1988. including that of gradients and second derivatives. with s ˆ+ α = 0 . as shown in the following analysis (Purvis. ∆ˆ n=n ˆα − n for spinorbitals with spin α and spin β . ∆pq is the overlap integral between the spatial parts of the ˆ β and n ˆ α and n ˆ β are the number operators spinorbitals p and q . Stanton 1994). βq p|s ˆz |q p ˆ† q ˆ= 1 2 pα ˆ+ = S r α sβ † ∆ r α sβ r ˆα s ˆβ . respectively.7). Sekino and Barlett 1988) ˆ− = S pβ qα ∆ pβ qα p ˆ† ˆα . the operators can be represented in second-quantized form as (Purvis. ˆz = S µ s ˆµz (12. ˆ− = S µ s ˆµ− . ˆ2 we use the well-known relationship To evaluate the matrix elements of S ˆ2 = S ˆ+ + S ˆ− S ˆz2 + S ˆz . ˆ+ and S where S Here ˆ+ = S µ s ˆµ+ . as would be required for any property calculation. (12. respectively. Of course.50). s ˆ+ β = α . s ˆz α = 1 2α . s ˆz β = −1 2β . (12. returning the numbers nα and nβ of the corresponding spinorbitals in the determinants on which ˆz all they operate.49) (the sums are over the electrons). computation of the response density matrices requires the solution of the Λ equations.5 Monitoring multiplicities in open-shell CC calculations 423 A more appropriate measure of the spin property for CC wave functions (Stanton 1994) is provided by an expectation-value form based on the response treatment of properties (Chapter 11): ˆ2 S resp ˆ ˆ2 T ˆ = 0|(1 + Λ)e−T S e |0 .12. (12. S (12.52) where the sums over pα and qβ are over spinorbitals with spin α and spin β .47) This expectation value can be computed easily using the response density matrices (Section 11. respectively. s ˆ− α = β .50) Using (12. When S+ is applied .48) ˆ− are the spin raising and lowering operators.51) ˆz = S pq p ˆ† ˆα − αp 1 2 rβ † r ˆβ r ˆβ = 1 n. s ˆ− β = 0 .

56) In an ROHF-based high-spin calculation the overlap terms vanish and the last term clearly gives S (S +1). ˆ2 is obtained by collecting the fully The vacuum expectation value of S contracted terms of (12. Similarly. we have ˆ+ = ˆ− S S pβ qα r α sβ † ∆ pβ qα ∆ r α sβ p ˆ† ˆα r ˆα s ˆβ βq † ∆ p β q α ∆ r α s β {p ˆ† ˆα r ˆα s ˆβ } + βq pβ qα r α sβ a α pβ sβ † ∆iβ qα ∆rα iβ {r ˆα q ˆα } + iβ qα rα iβ aα = − ∆pβ aα ∆aα sβ {p ˆ† ˆβ } βs 2 ∆iβ aα . (12.53) where {· · · } indicates the normal-ordered product.424 Additional aspects of coupled-cluster theory to a “high-spin” determinant (a determinant in which all singly occupied orbitals have spin α) the result is zero. ˆz2 = S 1 4 {∆ˆ n} + ∆n 2 2 =1 n}2 + 1 n}∆n + 1 4 {∆ˆ 2 {∆ˆ 4 (∆n) .54) =1 n} + 1 2 {∆ˆ 2 ∆n . the second and third terms result from single contractions and the last term results from a double contraction. ˆz = S 1 2 pα p ˆ† ˆα − αp pβ p ˆ† ˆβ βp {p ˆ† ˆβ } + βp pβ 1 2 iα = 1 2 pα {p ˆ† ˆα } − αp 1− iβ 1 (12.55): ˆ2 |0 = 0|S iβ aα ∆iβ aα 2 +1 2 ∆n 1 2 ∆n +1 .53)–(12.48) can be ignored in such cases. so the ﬁrst term in (12. (12. The remaining terms constitute the normal- . Using Wick’s theorem.55) where {∆ˆ n } = {n ˆ α }−{n ˆ β } and ∆n = nα − nβ (note the distinction between the operator ∆ˆ n and the number ∆n). (12.

47) and comparing the result with the response density matrices (11. (12. product part of S 2 ˆ2 − 0|S ˆ2 |0 ˆN =S S † ∆pβ qα ∆rα sβ {p ˆ† ˆα r ˆα s ˆβ } βq pβ qα r α sβ = + ∆pβ aα ∆aα sβ {p ˆ† ˆβ } − βs a α pβ sβ iβ qα rα 1 n}2 4 {∆ˆ † ∆iβ qα ∆rα iβ {r ˆα q ˆα } + + 1 n} 2 {∆ˆ ∆n + 1 .58) 12. An alternative way to view such orbital relaxation can be developed using the tools of CC theory (Paldus 1990). (12.57).88) ˆ2 can be comand (11.57) may usually be left out. we see that the response expectation value of S N puted easily from these density matrices and the overlaps between the spatial parts of the α and β spinorbitals.59) . (12.8 we derived the CPHF equations for the determination of orbital changes due to a perturbation. ˆ ˆ T ˆ1 |0 E = 0|eT1 He † C. respectively.6 The A and B response matrices from the viewpoint of CCS 425 ˆ2 .12.6 The A and B response matrices from the viewpoint of CCS In the response treatment of properties in Section 11. In normal CC calculations this number is zero for all determinants with nonzero amplitudes.57) (without the contributions from the last line) into (12. so that the terms in the last line of (12. Substituting (12. relative to the Fermi vacuum state. in the determinant on which they operate. 2 ˆN S resp = pβ qα r α sβ ∆pβ qα ∆rα sβ (ΓN )qα sβ pβ rα + a α pβ sβ ∆pβ aα ∆aα sβ (γ N )sβ pβ − iβ qα rα ∆iβ qα ∆rα iβ (γ N )qα rα .57) ˆ β } provide the excess The normal-product number operators {n ˆ α } and {n number of α and β spinorbitals. Consider the (connected) expectation value for the CC-singles (CCS) wave function. In EOM–CC (Chapter 13) or Fock-space MRCC (Chapter 14) these excess-occupancy numbers may be nonzero but are known numbers that are the same for all determinants in the computed wave function and can simply replace the corresponding operators in the last line of (12.89).

bj = a i j b = ab ij . (12. which is contrary to some claims that the B matrix represents or T a double-excitation.61) with a similar result for the complex conjugate. This equation is represented diagrammatically by × + + = 0. Diagrammatic representations of these maˆ† vertices from the ˆ1 and T trix elements can be obtained by stripping the T 1 diagrams.62) ˆ1 components in the ﬁrst two diagrams constitute The coeﬃcients of the T the elements of the singly excited CI matrix A of (11. Using the non-standard notation introduced in Chapter 4.e. the matrix elements can be represented as a Aai.426 Additional aspects of coupled-cluster theory Expansion of the exponentials up to quadratic terms gives 1 ˆ† 2 ˆ 1 ˆ ˆ2 ˆ +T ˆ† H ˆ† H ˆ +H ˆT ˆ1 + T ˆˆ E = 0|H 1 1 T1 + 2 T1 H + 2 H T1 |0 C + · · · .130).64) This analysis demonstrates that both matrices arise from orbital rotation. . ˆ1 terms. setOptimization of this expression with respect to the T 1 i ∗ ting δE/δta = 0.60) ˆ† amplitudes. results in ˆ ˆˆ ˆ† ˆ Φa i |H + H T 1 + T 1 H |0 = 0 . Bai.bj = δij i b × + δab a i j ×+ a i j b (12. It emphasizes the role ˆ ˆ1 of eT1 in orbital relaxation and helps to explain why the inclusion of T in CCSD and higher CC levels results in a lowered sensitivity to orbital choices. (12.63) = (εa − εi )δij δab + aj ib .130). (12. It also shows that ˆ there is some overlap between the role of eT1 in orbital rotation and explicit imposition of the CPHF conditions in the relaxed density matrix. contribution. allowing the use of QRHF and other reference functions that are not variationally optimum for the system being studied. while the coefˆ† components in the third diagram constitute the elements ﬁcients of the T 1 of the B matrix in (11. i. (12. or correlation.

6. it is useful to reconsider the noniterative triple.7 Noniterative approximations based on the CC energy functional 427 12. see also Table 10.67) can be written as ˆ ˆ N eT 0 E = 0 (1 + Λ)H 0 C = 0 (1 + Λ) H C 2 1 ˆ + ···) ˆN + W ˆ )(1 + T ˆ+ T = 0 (1 + Λ) (F 2 C C 0 C .5. given in Sections 10. of (12.66) cancel through fourth order. T1b and the diagonal part of diagrams T2a . C (12.s. The CC energy functional (11. Crawford and Stanton 1998) as an alternative to the treatment in terms of the CC expectation value.65) ˆ has been obtained. [4] d ˆ [2] ˆN T3 E T = 0 Λ3 F C [2] ˆ ˆ Λ1 W T3 C ˆT ˆ[2] + Λ3 W 2 + + C o ˆ [2] ˆ Λ2 FN T3 C ˆT ˆ[2] + Λ2 W 3 0 . 10.66) ˆ o stand for the diagonal and oﬀ-diagonal parts of F ˆN . ˆ[2] amplitudes are obtained from the second-order diagrams in the The T 3 CCSDT triple-excitation equation (diagrams T1a .12. This condition means that the where T ﬁrst two terms on the r.68) which is represented diagrammatically as follows: ×.and quadruple-excitation corrections to CC on the basis of this functional (Kucharski and Bartlett 1998a. this approximation is very close to the original orbitally invariant CCSD(T) form ET = [4] . and we are left with [4] ˆT ˆ[2] ET = 0 Λ2 W 3 C ˆT ˆ[2] + Λ1 W 3 C o ˆ [2] ˆN T3 )C 0 .3): ˆT ˆ2 W C d ˆ [2] ˆN |0 + F T3 C 0 = 0. we focus Assuming that a CCSD solution for Λ and T on the terms in the CCSDT correlation-energy expression that depend upon ˆ3 and/or Λ3 and isolate those terms that can contribute to this energy in T fourth order. T2b of Fig.62). Since Λ = T † at ﬁrst order (and at second order in the HF case). reˆ d and F where F N N spectively.5 and 10.b. (12. (12.67) ˆ2 is taken from the CCSD solution.69) + + (2) (2) (2) the third diagram vanishes in the Hartree–Fock case.h. + Λ2 (F (12. (12.7 Noniterative approximations based on the CC energy functional Following on from the introduction of the CC energy functional (11.

we start from the CCSDT solution and focus on the terms involving Λ4 and ˆ[3] in the CCSDTQ equation. . is proportional to nh p 3 n 4 for the triples correction. it requires ˆ. which is the rate-determining step in the calcu4 n 5 . which recommends it for the treatment of transition states that do not diﬀer too much from equilibrium geometry. which contains only (generalized) fourth-order corrections. Nevertheless it is better for describing bond solutions for both Λ and T breaking. in diagrammatic form. (12. (10. (10. which may be compared lation of this correction.50). (12. the T ˆ4 correction is where T obtained as ˆT ˆ[3] EQ = 0 Λ 3 W 4 or.71) (3) (3) this deﬁnies the ΛCCSDT(Q) model.70) ˆ2 and T ˆ3 are taken from the CCSDT solution. as in (10. If we limit ourselves to a ﬁfth-order with nh p correction. This Λ-based ΛCCSD(T) model was shown to have the prospect of being a better approximation than the usual CCSD(T) (Kucharski and Bartlett 1998a.48). depending upon the choice of reference function.72)) and sixth order (the ﬁrst and third diagrams). To apply the Λ-based approach to connected quadruple excitations.7). Crawford and Stanton 1998). we obtain EQ = [5] . The computational eﬀort for the ﬁrst diagram.41) except that. EQ = + + × . Unlike ΛCCSD(T).73) (3) 4 n 4 .39). because of the time-inversion asymmetry beˆ (Section 11. this model includes terms of both ﬁfth order (the second diagram in (12.49).428 Additional aspects of coupled-cluster theory (10. the perturbation theory that underlies the noniterative correction can fail in such cases. (12. Unlike the latter. (12.72) (3) C ˆT ˆ[3] + Λ2 W 4 C o ˆ [3] ˆN + Λ3 F T4 C 0 . however.39)–(10. Furthermore.6. if and the resulting computational eﬀort is of order nh p we use a factorization similar to that used in Section 10. we cannot simplify further the ﬁrst diagram tween Λ and T as in (10. Proceeding in the same way as for the triples T 4 ˆ[3] satisﬁes the third-order CCSDTQ equation correction. and noting that T 4 d ˆ [3] ˆN T4 F C + 1 2 ˆT ˆ22 W C ˆT ˆ3 + W C 0 = 0. No noniterative approximation can be fully satisfactory for large deviations from the equilibrium geometry since.

(12. (2004). In almost all ground-state calculations at near-equilibrium geometries. at least when the orbitals are real the . Grabowski et al. A critical comparison of various quadruples corrections and their performance in studies of bond breaking was presented by Musial and Bartlett (2005). propor3 n 5 . tend to be less stable than those based on even orders of perturbation theory.8 The nature of the solutions of CC equations The CC amplitude equations are nonlinear.12. and barring pathological cases such as near degeneracies. Zivkowiˇ c and Monkhorst 1978. as are the Hartree–Fock equations. As long as that reference function is a good approximation. and In this ﬁfth-order approximation we avoid the need for computing T 4 the computational eﬀort remains that of the underlying CCSDT. Piecuch and Kowalski 2000). Auer et al. For (12. In practice. Jankowski and Kowalski 1999a–d.and quadruple-excitation contributions to CCSD and their applications were described by Kucharski and Bartlett (1998a. the usual iterative procedures for the solution of the CC equations converge to the one desired solution. The number and character of the solutions of the CC equations have been ˇ ˇ investigated by several researchers (Zivkowiˇ c 1977. These questions are of no practical concern for most CC applications but are of theoretical interest for the understanding of the CC method in all its aspects. the so-called standard solution. (2005). Odd-order contributions. and therefore have multiple solutions (Kowalski and Jankowski 1998). Stanton. however. K´ allay et al. in single-reference CC we are usually interested in a speciﬁc electronic state (most often the ground state) and choose the reference function to be a reasonable approximation to the exact wave function for that state.74) ˆ[3] . 12. while applications of the sixth-order correction were considered by Bomble. Paldus. Jankowski.b) and by Gwaltney and Head-Gordon (2001). Applications of the unfactorized ﬁfth-order quadruples correction were given by Hirata.74) itself the computational scaling is of order n n 5 tional to nh h p p 2 2 5 ˆ ˆ for the T2 term and nh np for the T3 term. 1999. Fan. Several noniterative approximations for triple. Takahashi and Cho 1984. we obtain a ΛCCSDT(Qf ) analog of the CCSDT(Qf ) correction (Kucharski and Bartlett 1998a): ∆EQ = f [5] 1 2 ˆ 0 Λ2 T 2 †(1) ˆ2 ˆ (T ˆ3 + 1 T W 2 2 ) C 0 .8 The nature of the solutions of CC equations 429 though this factorization is not quite rigorous in this case because of the use of Λ. Kowalski.

the standard solution may produce a complex energy even for real orbitals. Grabowski et al. being an artefact of the non-Hermiticity of the CC eﬀective Hamiltonian. with |0 as the corresponding right eigenfunction) are necessarily real. was summarized in the extensive review by Piecuch and Kowalski (2000).) Furthermore the wave functions obtained for the various solutions are orthogonal to each other. Takahashi and Cho used complex orbitals.430 Additional aspects of coupled-cluster theory standard solution is well behaved. as is necessary for practical reasons in virtually all CC calculations. which can happen in cases of near degeneracy or of many excited states. FCI) equations on the other hand. and all the corresponding energies (which are eigenvalues of their respective eﬀective Hamiltonians. with amplitudes that are small relative to unity. and results in a real energy. However. Kowalski. for example. For the full ˆ is not coupled-cluster (FCC) model. but their results should be invariant under a transformation to real orbitals. when the CC cluster operator is truncated. where they determine the width of resonances. and with those of the FCC (or equivalently. and it is no longer possible to associate a speciﬁc nonstandard solution unambiguously with each FCC solution.) Research on the number and nature of all the solutions of the CC amplitude equations. However. the number of solutions is equal to the number of solutions of the full-CI (FCI) model that are not orthogonal to the reference function. this was demonstrated. show many branch points and other singularities. and particularly the work of Jankowski and Kowalski (1999a–d) and Jankowski. . but are not proper components of stationary-state energies. Analytic continuation methods connecting the solutions of truncated CC equations with those of the corresponding CI equations on one hand. (Paldus. the corresponding wave functions can no longer be orthogonal to each other and many solutions become complex. in which the CC cluster operator T truncated. when the reference function is not the dominant component of the exact solution. Many of the solutions appear to lack physical meaning. by Paldus. (Imaginary energy components are a proper result in some scattering calculations. the number of solutions increases substantially. and it is reasonable to ignore it. (Solutions that are orthogonal to the reference function cannot be represented in intermediately normalized form. Takahashi and Cho (1984) in model calculations using semiempirical π -electron Hamiltonians for some cyclic aromatic hydrocarbons.) The imaginary component of any CC energy result is unphysical. (1999).

ionized and electron-attached states are open-shell states. Koch and Jørgensen 1990). An eﬀective alternative in many cases is provided by the equation-of-motion coupled-cluster (EOM-CC ) method (Emrich 1981. ionized and electron-attached states 13.13 The equation-of-motion coupled-cluster method for excited. Furthermore. such as those described in Chapters 8 and 14. One solution to these problems is to resort to multireference methods. Such states are predominantly closed-shell singlet states. such as most molecular ground states near their equilibrium geometry. single-reference. Most excited. A third related approach is the symmetry-adapted cluster CI (SAC-CI) method (Nakatsuji 1979a. 431 . A closely related approach is the coupled-cluster linear response (CCLR) method (Monkhorst 1977. While such problems can be partially treated by using UHF reference functions. and CC calculations on them using UHF or ROHF orbitals do not usually result in pure-spin wave functions. so that single-reference CC based on RHF orbitals is then not usually appropriate for the calculation of entire potential-energy surfaces. but such treatments are still quite diﬃcult to apply at a high enough level. But even these states become dominated by more than one determinant when one or more bonds are stretched close to breaking. coupled-cluster method is very eﬀective for electronic states dominated by a single determinant. which usually separate correctly.b. and CC calculations on them produce pure singlet wave functions. Dalgaard and Monkhorst 1983. Stanton and Bartlett 1993a).1 Introduction The conventional. the UHF approach makes use of symmetry breaking and is poor in the spin-recoupling region. Sekino and Bartlett 1984. Comeau and Bartlett 1993. such states often involve large contributions from more than one determinant and thus do not respond well to conventional single-reference treatments.

The basic idea of EOM-CC is to start with a conventional CC calculation on some initial state.3) (13. is more like CI than CC.432 The equation-of-motion coupled-cluster method Nakatsuji. H (13.5) since the same reference energy has been subtracted from both E0 and Ek . Most commonly. H (13. because of approximations made.2 The EOM-CC Ansatz In the EOM-CC method we consider two Schr¨ odinger-equation eigenstates simultaneously. as shown by applications to ionization and electron-attachment processes. Ohta and Hirao 1981). as well as extensions to processes such as double ionization and double electron attachment. (13. 13. but we shall not use this term in order to avoid confusion with the reference determinant |0 . the initial state is the ground state.4) ˆ |0 . H ˆ Ψk = Ek Ψk . they need not have the same number of electrons. an initial state Ψ0 and a target state Ψk . The aim of the method is to determine the energy diﬀerence ωk = Ek − E0 (13. ionization potential EOM-CC (IP-EOM-CC) and electron-attachment EOM-CC (EAEOM-CC). with Eref = 0|H we have ωk = ∆Ek − ∆E0 . ˆ Ψ0 = E0 Ψ0 .1) become ˆ N Ψ0 = ∆E0 Ψ0 . Then where ∆E0 = E0 − Eref and ∆Ek = Ek − Eref .1) The initial state is often referred to as the reference state. equations (13. while the target state is an excited or ionic state. which is conceptually similar but. H ˆ N Ψk = ∆Ek Ψk . . If we use the normal-product form of the Hamiltonian. usually a conveniently chosen closed-shall state. Thus we distinguish excitation-energy EOM-CC (EE-EOM-CC). Although the calculations for the two states must use the same set of nuclei in the same geometrical arrangement and the same set of spinorbitals deﬁning a common Fermi state |0 .2) and related properties of the target state eﬃciently by canceling common terms in the solutions for the two states before the actual calculation. and obtain the desired target state by application of a CI-like linear operator acting on the initial-state CC wave function.

An operator R state.a (13.8) In the EE-EOM-CC case. ˆ (13.6). using (13. ˆ k |Ψ0 . the target-state Schr¨ odinger equation (13. ˆ Multiplying (13. and its exponential eT The operator T ˆ and its disconnected corresponds to a sum of linked diagrams arising from T ˆ k is used to generate the target state from the initial products. Unlike the initialˆ state wave operator Ω0 = eT . this operator is linear and thus CI-like.10) ˆ k |0 is a right eigenfunction of H with eigenvalue ∆Ek .13.a a †ˆ ri {a ˆ i} + i<j. if R ˆ k is not truncated cluster operator T then it will produce the full-CI result for the target state. if all possible excitations from the initial state are included we have ˆ k = r0 + R i. it commutes with the CC ˆ and all its components.7) so that. a<b iˆ b† ˆ j} + · · · . (13. ˆ NC ˆk |0 = ∆Ek C ˆk |0 H where ˆk = 1 + C i. |Ψk = R (13. Clearly. It is showing that R similar to the CI eigenvalue equation in normal-product form.4) can be written in the form ˆ ˆ ˆ k eT ˆ k eT ˆ NR H |0 = ∆Ek R |0 .9) The constant term r0 is required for the description of states of the same symmetry as the initial state. we get ˆ k and T tween R ˆ k |0 . The objective then is to determine the amplitudes in the R ˆ operator. a<b ab †ˆˆ† ˆ rij {a ˆ ib j } + · · · . (13. cab ˆ†ˆ ij {a (13.12) . ˆ k |0 = ∆Ek R HR (13. Since Rk is an excitation operator. if the two states are of diﬀerent symmetry ˆk then r0 = 0. |Ψ0 = Ω0 |0 = eT |0 .11) i} + ca ˆ†ˆ i {a i<j.2 The EOM-CC Ansatz 433 The initial-state coupled-cluster wave function is represented by the action ˆ of an exponential wave operator Ω0 = eT on a single-determinant reference function |0 .6) ˆ ˆ consists solely of connected terms.8) on the left with e−T and using the commutation beˆ.

R or ˆ k |0 HR C ˆ k |0 .20) . ˆ 1= k ˆ k |0 0|L ˆk . but L 1. Multiplying this equation on the left by R we obtain the EOM-CC equation in the form ˆ k |0 ˆ k ]|0 = (∆Ek − ∆E0 )R [H .14) and therefore satisﬁes ˆkP ˆ = 0.434 The equation-of-motion coupled-cluster method except that the operator H is non-Hermitian and therefore also has left ˆ k . = ωk R (13.18) As mentioned above. ˆkR (13.15) ˆ0 = ˆ ˆ0 = ˆ For the initial state (k = 0) we have R 1. ˆkQ L (13.16) 0|L They provide a resolution of the identity. because R ˆ k |0 = δ k 0 . δlk δkm = δlm = 0|L ˆ0 = ˆ 1 we have Also. The operator L ˆ k = l0 + L i. with the same eigenvalues ∆Ek as the corresponding eigenfunctions 0|L ˆ k |0 . the initial-state version of (13. Since ˆ 0 = 1. 0|L (13. (13. the essence of the EOM-CC method is to eliminate common terms from the target-state and initial-state treatments before the actual calculations and thus obtain the energy diﬀerence directly.13) (13. L ˆk = L ˆ.a i ˆ† la {i a ˆ} + i<j.10) is R H|0 = ∆E0 |0 . R (13. a<b ij ˆ† ˆ†ˆ lab {i a ˆj b} + · · · (13. ˆ k H = 0|L 0|L ˆ k is a de-excitation operator.17) as can be seen by placing this operator between any two left and right eigenfunctions. satisfying right eigenfunctions R ˆ k ∆Ek . ˆlR ˆkR ˆ k |0 0|L ˆ m |0 = 0|L k k ˆlR ˆ m |0 .10). The two sets of eigenfunctions are biorthogonal and can be normalized to satisfy ˆ l |0 = δkl .19) ˆ k and subtracting it from (13.

As an eigenvalue equation. the constant eigenvalue equation for R term r0 can be obtained from (13. (13.21). is components of R ˆ HP ˆ P ˆ HQ ˆ P ˆ ˆ ˆ ˆ QHP QHQ giving ˆ HP ˆ P C ˆ P 0 = C 0 0 .e.25) ˆR ˆk P ˆ. QRk P (13. for which r0 = 1. Diagrammatic techniques are used in EOM-CC to calculate the elements of this product. ˆ and Q ˆ The EOM-CC equation can be separated into blocks by the P projection operators. based on the Davidson algorithm (Davidson 1975) or any of its variations.21) The corresponding initial-state (k = 0) equation. = ωk Q Equation (13. ˆ HP ˆ P ˆ HQ ˆ P ˆ ˆ ˆ ˆ QHP QHQ ˆ r0 P ˆ ˆ ˆ QRk P = ωk C ˆ r0 P ˆ ˆ ˆ .24). the elements of ˆ . using the current estimate of the amplitudes in R ˆk .22) = 0. all other ˆ 0 vanish and ω0 = 0. (13. ˆ HQ ˆR ˆk P Q C Recognizing that the energy functional deﬁned in the treatment of properties in Chapter 11 can be written in the form ˆ 0 HR ˆ 0 |0 . in each iteration.13. using the current estimate of the solution vector. developed for the direct CI approach.25) is the key equation of the EOM-CC method.2 The EOM-CC Ansatz 435 Comparing this equation with (13.26) . = ωk r0 P (13. i.24) (13. ˆ HP ˆ Q C = 0. This iterative method avoids explicit evaluation of the matrix and requires instead direct calculation of the matrix–vector product. It is an ˆ k and ωk and.10) we see that the restriction to connected terms eliminates the ground-state correlation energy ∆E0 and the associated diagrams from the calculation and provides the excitation energy ωk directly.23) the same as the initial-state CC equations (the initial-state correlation energy ∆E0 has been eliminated from the ﬁrst of these equations by the connectedness condition). obtaining ˆ ˆ HQ ˆR ˆk P P ˆ ˆ HQ ˆR ˆk P Q C C ˆ. Therefore we can eliminate the ﬁrst column of (13. E0 = 0|(1 + Λ)H|0 = 0|L (13. once it has been solved. its solution beneﬁts from the techniques.

29) does not contribute to the matrix element Ψk |D|Ψl .32) ˆ n collects the n-tuple excitation terms of (13. and write ˆ = r0 + R ˆ1 + R ˆ2 + · · · . The application of these methods to a time-dependent perturbation theory treatment of excited states is discussed in Section 13.28) The dipole strength for a transition between states k and l is proportional to the absolute square | Ψk |D|Ψl |2 of the matrix element of the dipole operator D = 0|D|0 + DN = 0|D|0 + pq dpq {p ˆ† q ˆ} .27) we ﬁnd that EOM-CC provides a natural generalization for excited states (Stanton and Bartlett 1993a).31) The application of EOM-CC methods to the calculation of higher-order properties of the initial state is described below in Section 13. (13.9). ˆ k and In the following we shall drop the index k from the notation for R ωk . whose elements are ˆ † ˆˆ ˆ −T {p ˆq ˆ}eT R (γN )lk l |0 . R (13. qp = 0|Lk e (13.5. In particular.6. in analogy with where R ˆ2 + · · · . (13. R (13. .29) where d = µ qµ rµ and qµ and rµ are the electric charges and position vectors of the particles in the system.436 The equation-of-motion coupled-cluster method where ˆ0 = 1 + Λ . The ˆ manifests itself in the structure of the eﬀective Hamiltotruncation of T nian H. ˆ † ˆˆ ˆ −T {p ˆq ˆ}eT R (γN )k k |0 . which can be obtained from the transition density matrix γ lk N . Because of the orthogonality of Ψk and Ψl the vacuum-expectation-value term in (13. L ˆ0 = 1 .30) as Ψk |D|Ψl = Ψk |DN |Ψl = pq (γN )lk qp dpq . qp = 0|Lk e (13. In applications to excitation-energy ˆ = T ˆ1 + T the notation for T ˆ ˆ calculations both R and T are truncated at the same excitation level. excited-state properties are usually obtained in EOM-CC using excited-state generalizations of the groundstate response density matrices.

(R These diagrams are given in terms of eﬀective-Hamiltonian vertices.25) can be written in the form ˆˆ Φa i |HQR|0 C a ˆ = ω Φa i |R|0 = ωri . actual calculations use intermediates and factorization techniques along the lines of Section 10.7.7 (Kucharski.34) The algebraic forms of the EE-EOM-CCSDT equations are shown in Fig.3 Diagrammatic treatment of the EE-EOM-CC equations The application of diagrammatic methods to the solution of the EE-EOMCC equation beneﬁts greatly from the diagrammatic representation of H ˆ are represented described in Section 10. as in a i . The left-eigenstate operator L i a . Kucharski and Bartlett 2003. Wloch. The eigenvalue ω (the excitation energy) multiplies the left-hand-side diagram of the equation.3 Diagrammatic treatment of the EE-EOM-CC equations 437 13. particle–hole or ph) component of the eigenThe single-excitation (R value equation (13.2.63). a i b j . The diagrammatic ˆ 2 or pphh) and triple-excitation representation of the double-excitation (R ˆ 3 or ppphhh) equations for EE-EOM-CCSDT are also shown in Fig.1. However.. the constant term r0 can be obtained from (13. a {a i} ri ˆ†ˆ ab {a iˆ b† ˆ j} rij ˆ†ˆ abc {a ˆ} iˆ b† ˆ jc ˆ† k rijk ˆ†ˆ ˆ is represented similarly.33) This equation is represented diagrammatically for EE-EOM-CCSDT by the ﬁrst equation in Fig. this is represented diagrammatically as follows: ωr0 = + . Musial and Bartlett 2003).24). . (13. i aj bk c etc.13. 2001. For the last diagram we have taken advantage of (10. Musial. Musial et al. i {ˆ la i† ˆ l} ij ˆ† ˆ†ˆ lab {i a ˆj b} ijk ˆ† ˆ†ˆˆ † labc {i a ˆj bk c ˆ} consistently with the representation of Λ in Chapter 11. The ﬁxed labels a and i are understood to be associated with the open lines in these diagrams. Once the eigenvalue problem has been solved and the excitation energy ω is available. it should be noted that this notation masks the fact that these two sets of operators are not adjoint to each other. ˆ 1 . 13.1. a i b j c k etc. 13. i aj b . (13. The components of the operator R ˆ by diagrams similar to the corresponding T diagrams but with a heavy horizontal line to distinguish them from the former. 13.

is manifested by the limitation n∧ < 4 (except for the two-body vertex. This asymmetry. Diagrammatic representation of the EE–EOM–CCSDT equations.6.438 The equation-of-motion coupled-cluster method = + + + + + + = + + + + + + + + + + + + = + + + + + + + + + + + + + + + + Fig. where n and n∧ are the number of lines connecting to the H vertex from above and from below. 13.1. 11. respectively. in order to ˆn ˆ m vertices to the R generate diagrams describing the contributions of R ∨ ∧ ∨ equation we need to use H vertices with n − n = 2(n − m).10. because of the unsymmetrical nature of the H vertices with respect to up–down reﬂection. . In analogy with the case of the Λ equations in Section 11. without such limitation for n∨ . for which n∧ = 4 is also possible). which reﬂects the non-Hermiticity of H. ˆ equations are not identical to the upside-down The diagrams for the R images of the Λ-equation diagrams seen in Fig.

de lm de rlm . a > b).2. ωr0 = dl d χld rl + 1 4 delm Fig.13. a).3 Diagrammatic treatment of the EE-EOM-CC equations d χad ri − d l 1 2 dlm ad χlmid rlm a χli rl + dl d χladi rl + dl ad χld ril + 1 2 del a ωri de χalde ril 439 − + 1 4 delm lm de ade rilm = (for all i. a χlbcijk rl + l dl ab χlcjk ril l de χabcdje rik de de χalbcdejk ril del abd χcd rijk d abd χlcdk rijl dl abc χlmjk rilm lm abe χlmcjdk rilm dlm d χlabcdijk rl ˆ (i/jk ) P d d ˆ (a/bc) χabcejk ri −P ad χbcdk rij d ˆ (a/bc|k/ij ) +P ˆ (a/bc|i/jk ) +P dl ˆ (i/jk |c/ab) −P ad ˆ χlbcdjk ril +1 2 P (j/ik ) ac χlbmijk rlm + − 1 ˆ 2 P (b/ac) lm 1 ˆ 2 P (a/bc) dlm + 1 ˆ 2 P (i/jk ) ae χlmbcidjk rlm abc χlk rijl l ˆ (c/ab) +P ˆ (k/ij ) −P + + 1 ˆ 2 P (a/bc) ˆ (c/ab|k/ij ) +P + 1 ˆ 2 P (i/jk ) ade χbcde rijk de 1 ˆ 2 P (a/bc|k/ij ) del abe χblcdek rijl − 1 ˆ 2 P (c/ab|i/jk ) abc = ωrijk (for all i > j > k. ˆ (ij ) P d d ˆ (ab) χabej ri −P l ab χlj ril l a χlbij rl + dl de χabde rij d ˆ (ab) χlabdij rl +P d ab χlmij rlm lm ad χbd rij ˆ (ij ) −P ˆ (ab|ij ) +P + 1 2 de + 1 2 ad χlbdj ril dl + 1 ˆ 2 P (ij ) del deb χalde rilj de ˆ χalbdej ril −1 2 P (ab) dlm ae χlmbidj rlm + dl abd χld rijl + 1 ˆ 2 P (ab) del ab = ωrij − 1 ˆ 2 P (ij ) dlm adb χlmid rlmj (for all i > j. followed by the equation for the constant term r0 . The EE–EOM–CCSDT eigenvalue equations. . a > b > c). 13.

A single-determinant-based CC description of the open-shell singlet state would require using one of the two determinants in (13. Since most electronic states reached by dipole-allowed optical excitation from closed-shell ground states are open-shell singlets. The diagrams for the L are identical to those for the Λ equations in Fig. because there is no connectedness condition for the left-eigenfunction equation (13. particularly open-shell singlets obtained by a single excitation from a closed-shell state. The zero-order approximation for an open-shell singlet state is a twodeterminant wave function. (13. The solutions for these equations are diﬀerent for each target state and diﬀerent from the initial-state Λ-equation solutions.440 The equation-of-motion coupled-cluster method ˆ equations. Of course. the Q space) in the CC expansion. while a¯ b+a ¯b describes the MS = 0 component of the triplet. obtained from the solution of ˆ eigenvalue problem. with the zero-order descriptions ab and a ¯¯ b. For this wave function. such states have to be treatable in any reasonable theoretical approach for excited states. are diﬀerent for each state. if we simply anticipate this equivalence then we can accelerate the process but even so the . the two spinorbital determinants. described symbolically as a¯ b and a ¯b. respectively. 11. but no such alternative is available for the open-shell singlet.35) as the reference determinant. Since the reference determinant always has coeﬃcient 1 because of intermediate normalization. At convergence both determinants should have coeﬃcients of the same magnitude (but opposite sign). appear with equal-magnitude weights. ΦOS = 1 √ 2 A{(core)[φA (N − 1)α(N − 1)φB (N )β (N ) − φA (N − 1)β (N − 1)φB (N )α(N )]} . the other determinant being part of the external space (i.35) where “(core)” is a product of the spinorbitals occupied by the other N − 2 electrons.13). because the energy diﬀerences ω .10. including the disconnected diagrams. making the basic perturbative structure for solving the CC equations poorly convergent. the R A particularly important application of EE-EOM-CC is to a class comprising problems that are diﬃcult to treat otherwise by single-reference methods: the low-spin open-shell states.e. The a¯ b−a ¯b combination describes the singlet. the coeﬃcient of the second determinant has to grow to that same magnitude. representing the left eigenfunctions. The triplet state can easily be handled in a single-reference treatment by using the MS = 1 or MS = −1 component.

computed excitation energies at several levels of EE-EOMCC for three states of the CH+ ion (Hirata.1. if the EOM-CC treatment is based on a closed-shell initial state then all eigenstates are pure spin eigenfunctions. in which several quasidegenerate determinants need to be treated equivalently. Stanton 1994). the evaluation of S culation. in its various forms. can also provide an alternative to multireference methods for the treatment of other states that are not well described by a truncated single-reference expansion. the spin-ﬂip CC method (Krylov 2001). as measured by the value of 0|L k |0 . 2000). EOM-CCS is equivalent to CIS because the contributions of single excitations to the ground state vanish in the absence of double excitations. for example. CCSDTQ would be equivalent to full CI for this example. this includes states involving bond breaking. The initial state in these calculations is the closed-shell ground 11 Σ+ state. which may require more than two determinants for their zero-order description. a lowto R spin doublet state. Sekino and Bartlett 1988. Since only the four valence electrons are correlated. For a closed-shell HF reference function. . and constraints on the amplitudes can be imposed to guarantee pure-spin eigenstates for open-shell-based calculations (Szalay and Gauss 1997. achieves even more ﬂexibility for applications to open-shell states and bond breaking. Furˆ2 can be made a part of the EOM-CC calthermore. such as a high-spin state. the EOM-CC approach.13. In principle. In practice higher excitations. whether one is using ROHF (spin eigenfunctions) or UHF to deﬁne the spinorbitals and the Fermi state in the underlying CC calculation the initial-state CC wave function does not deviate greatly from a pure spin eigenfunction (Rittby and Bartlett 1988. and therefore there are ˆ k ) in the no restrictions on the coeﬃcient magnitudes (the amplitudes of R solutions. Often.2. Even when the EOM-CC calculation is based on an open-shell initial state. A generalization of EOM-CC. Nooijen and Bartlett 2000a) are given. Purvis. Clearly.3 Diagrammatic treatment of the EE-EOM-CC equations 441 treatment would be unbalanced since the excitation levels included relative to the second determinant would be fewer than those relative to the ﬁrst. the EOM-CC method provides an eﬀective solution to this problem. the EE-EOM-CC method is just as eﬀective for other lowspin open-shell states. should be included for the correct description of. As an example. In particular. in Table 13. with full-CI values for comparison. like those due ˆ 3 . most EOM-CC eigenstates are found to be close to ˆ ˆ2 T ˆˆ ˆ k e−T S e R spin eigenfunctions. the excited states are obtained in EOM-CC from a matrix eigenvalue problem as eigenstates of H. As seen in Section 13.

1 represent an approximation of full EOM-CCSDT that uses the CCSDT-3 wave function (Table 10. for states containing signiﬁcant contributions of double excitations EOM-CCSD is less accurate. we expect satisfactory excitation energies from EOM-CCSD. Nooijen and Bartlett 2000a).1. Excitation energies relative to the full CI results (in eV) at several levels of EOM-CC for three states of the CH+ iona State Full CI CCSb 0. However.544 0. b CCS is equivalent to CIS. The treatments of the initial and target states remains consistent.442 The equation-of-motion coupled-cluster method Table 13.1). RCH = 1. most of this inaccuracy is resolved.304 a Frozen core.231 0. this is not the case for an EOM-CCSD treatment of an excited state that contains a signiﬁcant doubleexcitation component.2087 21 Σ+ 8. ˆ operators are used in both.5/9. 6-31G** basis set.Comparisons with full CI.062 CCSDT-3 — 0.282 — 0.23/95. as measured by the weight of single excitations in the target-state wave function. This approximation neglects the contribution of T ˆ3 to the T R 3n 5 equations.2). When the target state is dominated by single excitations relative to the initial state. While the initial-state CCSD wave function accounts for almost all the correlation energy for the initial state.028 0. which can usually .5 11 Π 3. The various CC results are given as diﬀerences from the full-CI values.073 0.1 0. though not as dramatically.548 CCSD 0.131˚ A (Hirata.7/3.002 0. c Percentage weights of single and double excitations in the excited-state wave function. This simple example illustrates an important aspect of EOM-CC. reducing the scaling of the ground-state calculation from nh p 3 4 to nh np . Nooijen and Bartlett 2000a). such an extension also improves the results for singly excited states. since the same approximate T Another instructive example is provided by the methylene (CH2 ) molecule (Hirata. sometimes by one or more eV.5304 31 Σ+ 14.022 CCSDT 0.003 Weightsc 94. Once the treatment is extended to EOM-CCSDT. when excitation energies for various states at several EOM-CC levels are compared with full CI results in a 6-31G* basis set (Table 13. At the same time. The EOM-CCSDT-3 results included in Table 13. because of an imbalance in the treatment of the initial and target states.6 87. and this expectation is conﬁrmed for the open-shell singlet states 11 Π and 31 Σ+ (Table 13.4) for the initial state but does not truncate the EOM-CCSDT ˆ3 ˆ equations.

32 6.91 8.7 0.15 1.90 8.35 9.3 2.08 −0.310 5. The lowest singlet state. b CCS is equivalent to CIS. c Percentage weights of single and double excitations (relative to 11 A1 ) in the wave function.69 9.55 — 6.310 5.a CCSDT-3 CCSDT CCSDTQ CCSDTQP full CI Weightsc State CCSb CCSD 1 B1 A1 1 A2 1 B2 1 A1 3 B1 3 A2 3 B2 3 A1 3 B2 1 1.12 −0.1/6.59 8.67 4.7 deg.322 5.102 A 11 A1 . Frozen core.8/92. is used as the initial state (Hirata.45 — 9.09 8.09 8.31 7.312 5.95 8.67 5.38 8.28 — 8.52 6.6 94.25 9.2/92.69 −0.25 9. Nooijen and Bartlett 2000a).30 8.2/5.79 1. RCH = 1.15 94.05 −0.47 8.76 5.84 6.68 4.52 6.5 a ˚.6 89.310 5.1 2.68 4.31 6.32 6.0/4.6 92. 6-31G* basis set.09 8.97 6.30 1.25 9.Table 13.68 4.33 9.3/7.10 8.19 1. .39 9. relative to the lowest singlet state.06 −0.6 91.9/2.15 1.33 9.05 −0. αHCH = 104.344 5.15 1.90 8. for several levels of EOM-CC ( eV).5/93.28 9.2/5.33 9.3 93. Energies for various states of CH2 .05 −0.52 6.2.6/2.33 9.10 9.32 6.68 4.56 6.09 8.0 90.75 9.

Hirata (2004) and Musial and Bartlett (2004). (1995). but full EOM-CCSDT is needed to obtain a satisfactory description of these states. respectively. Koch. In this example the closed-shell lowest singlet (1 A1 ) state is the initial state. Several comparisons of EOM-CC results with experimental data can be found in the extensive review by Bartlett and Musial (2007).444 The equation-of-motion coupled-cluster method be done only with relatively small basis sets.23) would no longer hold. Hald. Wloch. Nooijen and Bartlett 2000a). to allow very-high-order excitation levels as in EOM-CCSDTQP. Furthermore.24) from the rest of the EOM-CC eigenvalue problem. The calculation of the ﬁrst and second derivatives of EOM-CC potentialenergy surfaces was described by Stanton (1993) and by Stanton and Gauss (1995). an algorithm based on the availability of the full CI Hamiltonian matrix in a determinantal representation was used instead (Hirata. Jørgensen et al. (2001). these calculations were similar to the high-order benchmark calculations for the coupled-cluster method quoted in Chapter 10 (Hirata and Bartlett 2000). because the full-CI comparisons separate basis-set eﬀects from correlation-treatment eﬀects. Larsen. Meissner (1998). particularly when one is accounting for zero-level vibrational contributions or extracting accurate vertical excitation energies from high-resolution spectra. Had the open-shell ground state been used as the initial state. Christiansen. Olsen et al. Musial et al. The EOM-CCSDT-3 approximation corrects this deﬁciency to some extent. comparisons with experiment can involve many uncertainties. including questions of interpretation. . Nonetheless. 2001) but. so that the ground state 3 B1 appears as a de-excitation. Once again. (1996). the results would have been very similar but would not have produced purespin eigenfunctions. preventing the separation of the r0 equation (13. ˆ and in R ˆ has been explored The use of diﬀerent levels of excitation in T but in such a case (13. an accurate description of a full spectrum of states is readily obtained. Other comparison of calculated EOM-CC (or CCLR) excitation energies with full-CI results can be found in Koch. Christiansen. large errors are encountered at the EOMCCSD level for states with a high percentage of double excitations. Jørgensen et al. Such an unbalanced approach would also cause diﬃculties in derivative calculations for properties. The CH+ and CH2 results quoted here could have been obtained by the diagrammatic methods normally used for larger basis sets and molecules (Kucharski. are more instructive than those with experimental data.

4 EOM-CC treatment of ionization and electron attachment Because second-quantized operators are Fock-space operators that are not restricted to a speciﬁc number of electrons. j baˆ† ˆ † rj b ja ˆ + ··· (13. the eigenvalues and eigenvectors of EE-EOM-CC smoothly approach those of IP-EOM-CC.13. pph. sectors. pphh. The IP-EOM-CCSD ˆ . but this procedure does not translate into an optimal factorized CC program. These techniques are related to Green’s function methods (Nooijen and Snijders 1992. The analogous treatment of electron attachment (EA-EOM-CC) is repreˆ (Musial sented in EOM-CC by using the p. Kucharski and Bartlett 2003). while IP-EOM-CCSDT (Musial. ˆ operator that reduces The ionization process can be treated by using an R the number of electrons by one. 13. simply zeroing out the appropriate matrix elements in the EE-EOM-CC equations.3. Instead of the ph (particle– reﬂects the altered structure of the R ˆ used in the EE-EOM-CC treatment. In fact. pphhh etc. 13. The latter requires the use of the IP-EOM-CC structure from the outset (Musial. but Green’s functions are normally Hermitian while EOM-CC is not. ppphhh etc. such as ionization and electron attachment.4 EOM-CC treatment of ionization and electron attachment 445 13. The IP-EOM-CCSDT equations are represented diagrammatically in Fig. the formalisms described here can be used for processes involving changes in the number of electrons. phh.4. Since all ionized states are preceded by a series of Rydberg excitations that approach the ionization continuum. is ˆ |0 formally the same as the EE-EOM-CC equation except that HR now C ˆ operator. j>k>i bc ˆ† ˆ † ˆˆ ˆ ki + · · · . hole). ˆ= R a ra a ˆ† + a>b. ˆ= R i i+ riˆ b. model uses the h and phh sectors of R Kucharski and Bartlett 2003) adds the pphhh sector. rjki b jc (13. will provide the IP-EOM-CC solutions (Stanton and Gauss 1999). the IP-EOM-CC equation.36) The eigenvalue equation for this operator. sectors of R and Bartlett 2003). 1993). sectors of R in IP-EOM-CC we use the h. corresponding to the removal of an excited electron into the continuum. j>i b ˆ† ˆˆ rji b ji + b>c. It is clear that this representation is identical to the corresponding representation for the IP-EOM-CC equations except that some line directions are reversed. this change in . However. ppphh etc.37) The diagrammatic representation of these equations in terms of eﬀectiveHamiltonian vertices is shown in Fig.

Like the EE-EOM-CC examples described in Section 13. for IP-EOM-CC calculations for two ionization potentials of C2 and EA-EOM-CC calculations for two electron aﬃnities of CH+ are given in Tables 13. Diagrammatic representation of the IP-EOM-CCSDT equations.4. nh h p p 2 3 CCSD and nh np for EA-EOM-CCSDT. Nooijen and Bartlett 2000b). 13.3.446 The equation-of-motion coupled-cluster method direction introduces diﬀerent intermediates because of the asymmetry with respect to time inversion of the diagrammatic expansion of H (Section 10. Analogous treatments can be applied to double-ionization (DIPEOM-CC) and double-electron-attachment (DEA-EOM-PP) processes by natural extensions of the IP-EOM-CC and EA-EOM-CC methods.3 and 13. n n 2 for EA-EOMnh np for IP-EOM-CCSD. respectively (Hirata. Sample results. . Double ionization has an obvious application to Auger experiments. The dimensions of the matrices in the eigenvalue equations are of order 2 3 n 2 for IP-EOM-CCSDT. these results were obtained by utilizing the full-CI Hamiltonian matrix in a determinantal representation.3. with full-CI results for comparison.7). but both = + + + = + + + + + + + + = + + + + + + + + + + + + Fig.

Wave functions describing the stretching and breaking of bonds in molecules. Diagrammatic representation of the EA–EOM–CCSDT equations. double-ionization and double electron attachment can be useful in treating bond breaking and some other multireference problems. RCC = 1.4 EOM-CC treatment of ionization and electron attachment 447 = + + + = + + + + + + + + = + + + + + + + + + + + + Fig.662 15. usually have more than one dominant conﬁguration. b The CCS approximation is equivalent to Koopmans’ theorem.130 14.3.721 Frozen core. 13. Ionization potentials of C2 at several levels of IP-EOM-CC with the full CI results (in eV) for comparisona State 1πu → ∞ − →∞ 2σu a CCSb 12. Table 13. 6-31G basis set.151 14.195 13.262 ˚ A (Hirata.749 CCSDTQP 12. Nooijen and Bartlett 2000b).134 14.942 CCSD 12.4. In the simplest . such as required in calculations of potential-energy curves and surfaces.180 CCSDT 12.131 14.724 Full CI 12.13.803 CCSDTQ 12.

The automated generation . but EOM-CC uses a linear CI-like operator and is more sensitive. An important consideration in such applications is the suitability of the orbitals of the underlying initial-state calculation for the description of the target state.117 1. When the orbitals are far from optimal for the target state.740 Full CI 10. In addition to the (core)a2 and (core)b2 conﬁgurations.109 1. b). using the (core)a2 b2 conﬁguration as the initial state (Nooijen and Bartlett 1997b). the breaking of a single bond in a closed-shell molecule.150 1. Since the a and b orbitals are both in the Q-space in such a calculation. b The CCS approximation is equivalent to Koopmans’ theorem. The calculation of derivatives of potential-energy surfaces in IP-EOM-CC has been described by Stanton and Gauss (1994). and are treated equivalently. RCH = 1. The IP-EOM-CC and EA-EOM-CC methods also occur as steps in the calculation of excitation energies by the similarity-transformed EOM-CC (STOEM-CC) approach (Nooijen and Bartlett 1997a. example. This approach produces results very close to those of Fock-space (multireference) CC.159 CCSD 10. described symbolically as (core)a2 and (core)b2 . Nooijen and Bartlett 2000b).734 CCSDTQ 10. a proper description of the dissociation process can be obtained. this problem can also be treated by double-ionization EOM-CC. Such situations can be treated by doubleelectron-attachment EOM-CC. Alternatively. using the doubly ionized “core” state as the initial state.922 1.120 ˚ A (Hirata. two conﬁgurations. such a calculation would automatically include the (core)a¯ b and (core)¯ ab conﬁgurations (assuming that a and b have the same spatial symmetry). Electron aﬃnities of CH+ at several levels of EA-EOM-CC. The initial-state CC calculation is relatively insensitive to the ˆ orbital choice because of the inclusion of the eT1 operator. the EOM results are bound to reﬂect this deﬁciency (Tobita.109 1.741 Frozen core. respectively. Perera. Musial et al. where a and b describe the bonding and antibonding orbitals. to be discussed in Chapter 14.701 CCSDT 10. approach degeneracy as the bond is stretched. 2003).4. compared with full CI results (in eV)a State ∞ → 1π ∞ → 3σ a CCSb 8.448 The equation-of-motion coupled-cluster method Table 13. 6-31G* basis set.

H (13. H = e−T He = H + 0|H (13. Using (13. which in this context we shall denote by H (“H-bar”) to distinguish it from H.16).13. As shown in Section 13.10).13) and (13.3.5 EOM-CC treatment of higher-order properties 449 of computer programs for high-order EOM-CC calculations was described by Hirata (2003. 13.40) with perturbation-strength parameter λ. (13.41) (13. In the perrelative to the Fermi-vacuum reference energy Eref = 0|H turbation treatment of properties it is convenient not to separate the reference energy from the total energy and therefore we deﬁne a modiﬁed CC eﬀective Hamiltonian. The corresponding eigenvalues are the energies ∆Ek = Ek − Eref 0|L ˆ |0 . The ﬁrst-order perturbed energy can be used to obtain the static dipole . the EOM-CC wave functions are obtained. as |Ψk = e Rk |0 and Ψk | = ˆ ˆ k e−T . 2004). using (13.18) and (13.38) This operator has the same biorthogonal sets of eigenfunctions as H but its eigenvalues are the total energies Ek . Ek = 0|L (13. where qµ and rµ are the charges and position vectors of the particles in the system. it provides a path for the calculation of higher-order properties of the initial state by a closed-form equivalent of the usual perturbation theory that manifests a sum-over-states (SOS) form. by ˆ ˆ T ˆ ˆ |0 . An example. is a perturbation due to an electric ﬁeld λE .19). the initial-state energy is given by E0 = 0|H |0 . ﬁrst discussed in Section 11. As a result. for which the perturbation operator is given by the dot product Θ = −E · µ qµ rµ . ˆ k of the ˆ k |0 and 0|L in terms of two biorthogonal sets of eigenfunctions R ˆ ˆ ˆ ˆˆ − T T T CC eﬀective Hamiltonian H = e HN e .5 EOM-CC treatment of higher-order properties The EOM-CC method generates energies and wave functions for a range of states up to a speciﬁed excitation level relative to a given initial state. Now consider a perturbed Hamiltonian ˆ (λ) = H ˆ 0 + λΘ .2.39) In particular. we ﬁnd that ˆkHR ˆ k |0 . The energy diﬀerences ωk = Ek − E0 = ∆Ek − ∆E0 are not aﬀected.

ˆ The Schr¨ odinger equation for the perturbed initial state Ψ0 (λ) = eT (λ) |0 is ˆ(λ) ˆ ˆ (λ)eT |0 = E0 (λ)eT (λ) |0 . ˆ 0 + λΘ)eT0 |Υ(λ) = E0 (λ)eT0 |Υ(λ) .43) (13. the second order will yield the static dipole polarizability and higher-order terms yield static hyperpolarizabilities (Sekino and Bartlett 1993).41) and (13.51) . 3 = T ˆ(2) + T (13. this aspect was used previously to distinguish the relaxed and response density matrices in Section 11.42) As is common in most applications. ˆ (13.45).46) ˆ|0 plus any products Each order of |Υ equals the corresponding order of T ( n ) ˆ terms that combine to give the same order. we have assumed that the reference function |0 is not aﬀected by the perturbation (but relaxed-orbital choices are possible.47) ˆ(1) 2 |0 .45) (13. ˆ(λ) can be separated into an unperturbed part The CC cluster operator T ˆ0 and a perturbation (which commute with each other). The perturbed initial-state wave function is |Ψ0 (λ) = eT (λ) |0 . (13.9). |Υ(1) = T |Υ (2) (13. ˆ ˆ ˆ ˆ (13. ˆ(λ) = λT ˆ(1) + λ2 T ∆T The perturbed wave function can then be written as |Ψ0 (λ) = eT0 e∆T (λ) |0 = eT0 |Υ(λ) where |Υ(λ) = e∆T (λ) |0 = |0 + λ|Υ(1) + λ2 |Υ(2) + · · · . T where ˆ(2) + · · · .48) |0 (13. using (13.49) ˆ(2) + T ˆ(3) + T ˆ(1) T ˆ(1) |Υ(3) = T etc. T ˆ(λ) = T ˆ0 + ∆T ˆ(λ) .450 The equation-of-motion coupled-cluster method moment of the system.44) (13.50) H or. of T ˆ(1) |0 . (H ˆ ˆ (13.

0 1 + Λ(λ e−∆T (λ) H 0 + λΘ e∆T (λ) 0 = E0 (λ) .60) ˆ(2) part in the second term in (13.60) are ˆ(1) term in (13.52) (13.53) (13.13.56) which is equal to E0 (λ) when the functional is stationary. ˆ0 ˆ in (13.s. (13. .59) E (2) = 0|(1 + Λ)(Θ − E (1) )|Υ(1) − 0|(1 + Λ)(E0 − H 0 )|Υ(2) ˆ(1) |0 − 0|(1 + Λ)(E0 − H 0 )[T ˆ(2) − (T ˆ(1) )2 ]|0 = 0|(1 + Λ)(Θ − E (1) )|T ˆ(1) |0 − 0|(1 + Λ)(E0 − H 0 )(T ˆ(1) )2 |0 . The prime is inˆ) tended to distinguish this functional from the CC energy functional E (Λ. T H 0 + λΘ e∆T (λ) 0 ) . we obtain E (1) = 0|(1 + Λ)Θ|0 − 0|(1 + Λ)(E0 − H 0 )|Υ(1) ˆ(1) |0 = 0|(1 + Λ)Θ|0 − 0|(1 + Λ)(E0 − H 0 )|T = 0|(1 + Λ)Θ|0 = Θ . ˆ ˆ (13.2) to (13. = 0|(1 + Λ)(Θ − E (1) )|T (13. ˆ (13. of this equation deﬁnes a corresponding CC energy functional.60) remains.57) (13. T ˆ of Section 11. (E0 − H 0 )|Υ(2) = (Θ − E (1) )|Υ(1) − E (2) |0 . as does −E (1) .58) etc.46).h. ˆ(λ) ˆ ˆ) = 0 [1 + Λ(λ)]e−∆T E (Λ. (13.59) and the T The T eliminated by intermediate normalization.43) allowed us to absorb eT Thus the separation of T into the operaˆ ˆ tors and to incorporate e∆T into Υ.55) The l. (E0 − H 0 )|Υ(1) = (Θ − E (1) )|0 .54) Θ=e ˆ0 −T Θe ˆ0 T . reﬂecting the use of H instead of H in the present context. (13. The quadratic term in (13.52) results in inhomogeneous equations for the various orders of Υ and E0 .4 which becomes equal to ∆E = E − 0|H |0 when stationary. Applying standard Rayleigh–Schr¨ odinger perturbation theory (Section 2.52) onto 0|(1+Λ)e−∆T (λ) and noting (13. where ˆ ˆ T ˆ0 H 0 = e−T0 H 0e . we obtain the perturbed energy. since 0| is not an eigenfunction of H 0 . subject to intermediate normalization. Extracting the energies requires left-multiplication by 0|L0 = 0|(1 + Λ). Projecting (13.5 EOM-CC treatment of higher-order properties 451 multiplying on the left by e−T0 results in H 0 + λΘ |Υ(λ) = E0 (λ)|Υ(λ) .

60) becomes ˆ(1) |0 ˆ(1) )2 |0 − 0|(1 + Λ) H 0 (T E (2) = 0|(1 + Λ) ΘT C C ˆ(1) |0 . linked connected expressions are the most desirable. but not identical. but the presence of −E (1) does not allow that.4) we ﬁnd that |Υ(1) = RΘ|0 = k=0 ˆ k |0 0|L ˆ k Θ|0 R ωk (13. T (13. T ˆ(1) . This provides a linear approximation. allow a separation of the Hilbert space into two complementary subspaces. = ωk k=0 (2) (13.63) and a related resolvent (similar. to the resolvent (11. deﬁned by the projectors ˆ 0 = |0 0|(1 + Λ) ˆ 0 |0 0|L P=R and Q=1−P = k=0 (13. Elin = 0|(1 + Λ)Θ|Υ(1) = 0|(1 + Λ)ΘRΘ|0 ˆ k |0 0|L ˆ k Θ|0 0|(1 + Λ)ΘR . the preferred representation of the resolvent operator is by means of a set of excitations |h that are orthogonal to the reference function . T = 0|(1 + Λ) Θ. Using the diagrammatic techniques of Chapter 5. R (13.65) and the E (1) term is removed. By providing a resolution of the identity.62) ˆ k |0 0|L ˆk .64) Using this resolvent (compare subsection 2. we ﬁnd that all the expressions we have obtained for the energy and wave function can be written in a connected and linked form that eliminates all renormalization terms.55)) R = Q E0 − QH 0 Q −1 Q= k=0 ˆ k |0 0|L ˆk R . which constitute the basis for the sum-over-states perturbation treatment.4.452 The equation-of-motion coupled-cluster method As in all many-body applications. It is also more laborious to compute because of the quadratic term. producing extensive and (when appropriate) intensive results. the second-order energy (13. ωk (13. the biorthogonal sets of eigenfunctions of H . though Λ itself is necessarily disconnected.61) This fully connected form is the most satisfactory expression for the secondorder energy but does not correspond to an SOS form. In particular.66) In practice. ˆ(1) |0 − 0|(1 + Λ) H 0 .

(13. 1994). Then the second-order energy is obtained in either of two forms: ECI = 0|(1 + Λ) Θ − E (1) |h t(1) (2) Elin (2) (13. (13. form (13.68) h|Θ|0 . as in CI. 0|(1 + Λ)|Υ = 0. causing this approximation to behave more like CI (Koch. where ˆ(1) amplitudes and is obtained as the solution t(1) is the column vector of T to a set of linear equations h|E0 − H 0 |h t(1) = θ . when truncated retains the Θ term.70) with θ = h|Θ|0 . (13.60) and (13. Such a resolvent involves the usual projector ˆ ˆ = 1 − |0 0| and takes the form Q=1−P ˆ ˆ (E0 − H 0 )−1 Q R=Q = |h h|E0 − H 0 |h −1 h| .71) (13. Russ and Crawford 2004). 0|(1 + Λ)|h = 0. (13.13. to avoid matrix inversion. but must be included here because of the change in orthogonality. These equations are solved using iterative strategies.69) (Stanton and Bartlett 1993b).71) or (13.72) provides a closed-form evaluation of the ﬁrst-order wave function that is equivalent to an SOS expression like (13. Kobayashi. linear and quadratic. The second.69) can ˆ of |h spans fully the original Q space of the be recovered if the space Q EOM-CC excited states.67) (Here h| and |h are column and row vectors.65). The various approximations. A connected form of (13. which is still linear but cannot be expressed in SOS form.61) because of the intermediate normalization condition. de Mer´ as et al. linear.5 EOM-CC treatment of higher-order properties 453 0| instead of to 0|(1 + Λ). Iterative solution of the linear equations (13.69) Note the appearance of Θ = E (1) in the ECI equation. The ﬁrst-order function can be expressed as |Υ(1) = |h t(1) .) The equations for the perturbed wave function and energy become |Υ(1) = |h h|E0 − H 0 |h providing a CI-like approximation. Tam. respectively.72) = 0|(1 + Λ)Θ|h t(1) . ECI = 0|(1 + Λ) Θ − Θ h h|E0 − H 0 |h (2) (2) −1 −1 h|Θ|0 . this term was left out of (13. The form (13. Including the quadratic correction adds a double-commutator term. CI-like.72) is usually a very good approximation with reasonable choices of |h (Sekino and Bartlett 1999. have . referred to as the eﬀective Hamiltonian approximation.

13. This procedure (called Model III by Sekino and Bartlett) gives a fully linked extensive expression. we can consider the time-dependent Schr¨ odinger equation and ask how a molecule responds to the imposition of a perturbing oscillating electric ﬁeld. Nooijen and Bartlett 1996) and polarizabilities (Rozyczko. aimed at determining the energy diﬀerence between two such states. as would be used in a spectroscopy experiment.454 The equation-of-motion coupled-cluster method been compared for nuclear magnetic resonance coupling constants (Perera. 1997) and little numerical diﬀerence observed. including states with diﬀerent numbers of electrons. λ) = H ˆ 0 + λΘ eiωt + e−iωt ) H ˆ 0 + λV ˆ (t) . ˆ (t)Ψ(t) = i ∂ Ψ(t) . Perera. though these are small for HF reference functions. One other form has been suggested (Sekino and Bartlett 1999). Instead of a purely stationary-state description. Hence. this approach follows time-dependent perturbation theory to obtain dynamic (frequency-dependent) properties. Nooijen et al. and their associated wave functions. now termed EOM-CCL (Crawford and Sekino 2009). H ∂t with the perturbed Hamiltonian ˆ (t.72) recommends it for most routine calculations (Sekino and Bartlett 1999). =H (13. Dalgaard and Monkhorst 1983).6 EOM-CC treatment of frequency-dependent properties The EOM-CC approach is a time-independent method built upon stationary states. at a saving of ≈ 30% in computation. such as frequency-dependent polarizabilities (Monkhorst 1977.73) . Consider the time-dependent Schr¨ odinger equation. This model has been found to provide highly accurate approximations to the quadratic expression for circular dichroism.74) (13. so the comparative ease of evaluation of the connected form of (13. The molecule will undergo an excitation when the imposed frequency is in resonance with the transition frequency to another stationary state of the molecule. The reason the linear form is not exactly extensive is that Λ itself has disconnected parts. a slightly modiﬁed Λ equation can be solved that removes the disconnected terms. Properties of the stationary states can then be obtained using the corresponding EOM-CC left and right eigenfunctions. Alternatively.

λ) = ψ (x. E0 = 0|(1 + Λ)e−T0 H ˆ ˆ ˆ (13.t)t . (13. t)|b(x. are periodic in t. λ). t) = 1 τ τ 0 a(x. λ) = ψ (x. Ψ(x. λ)e−iE (λ.81) ˆ (x) = Eψ (x). (13. H (13. t. ψ0 (x. ˆ The unperturbed solution is |Ψ0 = eT0 |0 .t)t = eT (t.5.76) and projection onto 0|(1 + Λ)e−T0 results in ˆ 0 eT0 |0 = 0(1 + Λ)|H 0 |0 . in which ψ0 (x) = .79) where x represents all the spatial and spin coordinates of the particles. the time-dependent wave function undergoes periodic oscillations. t) .13. The functions ψ (x. t) over one period. t) = ψ0 ground state. thus (13. Ψ(x. ±1. (13. ˆ (13. t)|b(x. t. t)b(x. we have ˆ0 ˆ ˆ 0 eT |0 = E0 eT0 |0 .75) where τ = 2π/ω is the period. satisfying H 0 |0 = E0 |0 .77) ˆ ˆ T ˆ0 where H 0 = e−T0 H is the form of the CC eﬀective Hamiltonian used in 0e Section 13.80) is called the quasienergy .82) is the stationary Also. Thus. t) = ψ (x)e−iEt .79) becomes analogous to the timedependent wave function for a stationary state of energy E . .. . t) dt = 1 τ τ dt 0 ˆ eT0 |0 dx a∗ (x.6 EOM-CC treatment of frequency-dependent properties 455 The periodicity of the perturbation is given by ˆ (t) = V ˆ (t + mτ ) V (m = 0. i. for the unperturbed state. the analysis can be limited to the ﬁrst period. t) are deﬁned in an extended Hilbert space x ⊕ t with scalar product a(x. (x)e−E0 t . t) = T (λ. The time average of E (λ. (13. where Hψ Since all quantities are periodic in t.e. This wave function and its components.) . . ±2. T (λ. t + mτ. t)dt . (13.78) Under the eﬀect of a periodic perturbation. where |0 is the time-independent Fermi-vacuum reference function. which is analogous to a Brillouin zone for a solid.λ) |0 e−iE (λ. t + mτ ) etc. E (λ) = 1 τ τ 0 E (λ.t)t . ψ (x. Static properties emerge in the limit ω → 0. t. including the phase factor e−iE (λ.

λ) . Jørgensen and H¨ attig 1998). −n + 2.85) The energy and wave function are expanded in orders of the external perturbation. leads to ˆ (t) ψ (x. E (λ. n) . . To separate the time dependence from the spatial dependence. The limits −n ≤ k ≤ n are a consequence of the fact that no terms higher than . t) = E0 + λE (1) (t) + λ2 E (2) (t) + · · · . ∂t ∂t ∂t (13. after elimination of the phase factor. ˆ (13. t)t + i ψ (x.86) (13. t. . t) + · · · . ψ (n) (x. t.73) with the perturbed Hamiltonian (13. ∂t ∂t (13.88) (13. λ) .74) results in ∂ ∂ E (λ.79) is i ∂ ∂ ∂ Ψ(x. Epstein and Karplus 1972. t) = k ψk (x)eikωt (n) (k = −n. The time derivative of the time-dependent wave function (13. Sambe 1973.t) + λ2 ψ (2) (x. Perhaps the simplest approach is to recognize that all relevant quantities have an oscillatory behavior. (13. λ) = DE (λ.t)t E (λ. λ)e−iE (t. ψ (x.90) The spacing of k by 2 reﬂects a cosine choice for the Fourier expansion.λ)t = e−iE (λ. .456 The equation-of-motion coupled-cluster method Various approaches can be used for time-dependent perturbation theory (Langhoﬀ. ∂t (13. λ) . t. we invoke a Fourier expansion in the former. t. t. which enables them to be described by a ﬁnite Fourier series. the terms of which can simply be collected in multiples of eikωt and in which the phase factor can then be eliminated to provide equations largely analogous to those of time-independent perturbation theory for each order. λ) = ψ (0) (x) + λψ (1) (x. λ) = e−iE (λ. where D =1+t and ˆ0 − i H0 = H ∂ .t)t ˆ 0 + λV H [H0 + λV (t)]ψ (x. .83) Substituting in the time-dependent Schr¨ odinger equation (13.89) where ψ (0) (x) = eT0 |0 . t)t + i ψ (x. Christiansen. t. t.84) which.87) ∂ ∂t (13. t) ψ (x.

(1) (2) (1) (13. t) = V (13. t) = T (13. ψ (1) (x.99) Just as in the time-independent case.100) 2 ˆ(2) (t) + T ˆ(1) (t) |ψ (2) (x.s. t) = T eT0 |0 ˆ (13.98) Inserting (13.h.97) where Θ = eT0 ΘeT0 . ˆ ˆ ˆ ˆ (13.96) Left-multiplying by ψ0 (x)| = 0|(1 + Λ)e−T0 eliminates the l. t) − DE (1) (t) ψ (0) (x) .13. |ψ (1) (x.6 EOM-CC treatment of frequency-dependent properties 457 order n contribute to the nth order wave function.91) (13. (because of intermediate normalization) and results in DE (1) (t) = ψ (0) |Θ|ψ (0) (eiωt + e−iωt ) = 0|(1 + Λ)Θ|0 (eiωt + e−iωt ) = E (1) (eiωt + e−iωt ) . t) = ψ0 (x) + ψ+2 (x)e2iωt + ψ−2 (x)e−2iωt (2) (2) etc. t) = ψ+1 (x)eiωt + ψ−1 (x)e−iωt .92) ψ (2) (x. DE (1) (t) = 1 τ τ 0 E (1) (eiωt + e−iωt )dt = 0 . t) − DE (2) (t) ψ (0) (x) (E0 − H0 )ψ (2) (x. Here E (1) is the usual static ﬁrst-order term.101) . t) = V (13.97) over one period is zero. (E0 − H ˆ (x. Application of the standard perturbation-theory methods leads to inhomogeneous equations analogous to those of the time-independent case. (13.93) (13. The time average of (13.94) results in ˆ (x. the perturbed CC solutions are ˆ0 ˆ(1) (t)eT |0 . In particular. t) − DE (1) (t) ψ (1) (x. for the ﬁrst and second orders. t) − E (1) (eiωt + e−iωt ) ψ (0) (x) .97) into the ﬁrst-order equation (13. Inserting the Fourier expansion of the ﬁrst-order wave function into the ﬁrst-order equation results in ˆ 0 − ω )|ψ (x) e+iωt + (E0 − H ˆ 0 + ω )|ψ (x) e−iωt (E 0 − H +1 −1 (1) (1) = Θ(eiωt + e−iωt ) − DE (1) (t) eT0 |0 .94) ˆ (x.95) etc. (E0 − H0 )ψ (1) (x. (E0 − H0 )ψ (1) (x. (13. t) = V (13. ˆ 0 )ψ (0) (x) = 0 .

107) with H0 = h|H 0 |h and θ = h|Θ|0 .99) into the form ˆ(1) |0 e±iωt = Θ − Θ |0 e±iωt . and their Fourier components are given by ˆ(1) eiωt + T ˆ(1) e−iωt |0 ˆ(1) (t)|0 = T T +1 −1 ˆ(2) (t)|0 = T ˆ(2) + T ˆ(2) e2iωt + T ˆ(2) e−2iωt |0 T 0 +2 −2 ˆ 0 eT0 and etc. using H 0 = e−T0 H ˆ (t)eT0 = Θ(eiωt + e−iωt ) . E0 − H 0 ∓ ω T ± (13. Left-multiplication by e−T0 .458 The equation-of-motion coupled-cluster method etc. Using ωk to specify the position of the k th pole relative to E0 . V (t) = e−T0 V puts the ﬁrst-order perturbed equations (13. we can deﬁne the frequency-dependent (dynamic) polarizability. Since this equation is true for all values of ω the factor e±iωt can be eliminated. This result is equivalent to the EOM-CC expression (13. (13. The homogeneous part of this linear equation provides the poles. This reˆ (E0 − H 0 ± ω )−1 Q where R(±ω ) = Q solvent is nonsingular provided that the reference function is nondegenerate and ω = 0. we have H0 t k = ω k t k (1) (1) (13. which are modiﬁed only by the presence of ω : ˆ(1) |0 = Θ − Θ |0 .108) for each transition with which the radiation can be in resonance. and R For the perturbed ﬁrst-order wave function we ﬁnd T±1 |0 = R(±ω )Θ|0 (1) (13. with Q ˆ = 1−P ˆ = 1 − |0 0|. projection by a set of determinants h| suﬃcient to determine ˆ(1) .20). respectively.104) where the designation ± indicates the sign of ω . In second order we need to .103) (13.105) ˆ ˆ ˆ ˆ ˆ (13. i. providing the analogs of the time-independent equations. H0 and tk representing H ˆ k |0 .e.109) ˆ .102) (13. E0 − H 0 ∓ ω T ± (13. Continuing to the next order of perturbation theory. gives E0 1 − H0 + ω 1 t(1) = θ .106) Finally. t(1) (ω ) = t(1) (−ω ) = t(1) = h|T ˆ(1) |0 the vector of coeﬃcients in T ± + − + (note that these are the same for positive and negative frequencies).

V (13. (13.113) ˆ 0 eliminates the T ˆ(2) operators beLeft-multiplication by 0|(1 + Λ) = 0|L cause of the biorthogonality of the EOM-CC eigenfunctions.114) The second-order frequency-dependent energy can be obtained by timeaveraging the energy derivative over one period. after collecting factors of e±2iωt and e0 = 1. but it does ˆ(1) terms. The terms involving e±2iωt . ˆ(1) T not eliminate the T ˆ 0 is a left eigenfunction of H 0 .) Hence. E 0 − H 0 − 2ω ˆ(2) + T ˆ(1) T +2 +1 2 |0 e+2iωt − Θ+ − Θ+ T+1 |0 e+2iωt = −DE (2) (t)|0 . T ˆ(2) . (13.110) even though Θ+ = Θ− .13.111) (1) E0 − H 0 + 2ω ˆ(2) + T ˆ(1) T −2 −1 2 |0 e−2iωt − Θ− − Θ− T−1 |0 e−2iωt (1) = −DE (2) (t)|0 . For consider all three Fourier components leading to T 0 +2 −2 ˆ (t) operator as this purpose it is convenient to express the V ˆ (t) = Θ+ eiωt + Θ− e−iωt . (13. Therefore the quadratic term should be retained in the expression for the energy derivative. (13. (More general situations can be accommodated by keeping these two terms separate from the outset. as previously discussed for the static case.6 EOM-CC treatment of frequency-dependent properties 459 ˆ(2) . ˆ 0 Θ+ − Θ+ T ˆ(1) |0 + 0|L ˆ 0 Θ− − Θ− T ˆ(1) |0 DE (2) (t) = 0|L −1 +1 ˆ(1) ˆ 0 (E0 − H 0 )T ˆ(1) T − 2 0|L +1 −1 |0 .112) ˆ(1) ˆ(2) +2T ˆ(1) T ˆ(1) ˆ(1) E0 − H 0 T 0 +1 −1 |0 − Θ+ − Θ+ T−1 |0 − Θ− − Θ− T+1 |0 = −DE (2) (t)|0 . and T ˆ(2) . if 0|Λ is limited to categories of While 0|L ˆ(1) T ˆ(1) excitations in h| such as singles and doubles then the products T can introduce triple and quadruple excitations that are not involved in the determination of Λ.

α(ω. Rozyczko.118) The numerator in this expression contains the left and right transition moments. whose products (the dipole strengths) are the residues at the poles. The spectral expansion of this resolvent is R (ω ) + R (−ω ) = k ˆ k |0 0|L ˆ k |0 0|L ˆk ˆk R R + ωk + ω ωk − ω .460 The equation-of-motion coupled-cluster method integrate to zero and leave just the time-independent terms: E (2) (ω.1 k ˆ k |0 0|L ˆ 0 |0 ˆ 0 rR ˆ k rR 0|L . If instead we ﬁrst use the fully connected form. is obtained when Θ = µ qµ rµ . and leads to the textbook sum-overstates expression. −ω ) = DE (2) ˆ 0 Θ+ − Θ+ T ˆ(1) |0 + 0|L ˆ 0 Θ− − Θ− T ˆ(1) |0 = 0|L −1 +1 ˆ 0 (E0 − H 0 )T ˆ(1) T ˆ(1) − 2 0|L +1 −1 |0 ˆ 0 Θ+ − Θ+ R(−ω )Θ− 0 + 0 L ˆ 0 Θ− − Θ− R(+ω )Θ+ 0 = 0L ˆ 0 (E0 − H 0 )T ˆ(1) R(−ω )Θ− 0 − 0L +1 ˆ(1) R(+ω )Θ− 0 . (13.66) (Stanton and Bartlett 1993b. ωk + (−1)l ω (13. −ω ).116) Then ˆ 0 |ΘR (ω )Θ|R ˆ0 + L ˆ 0 |ΘR (−ω )Θ|R ˆ0 α(ω. instead of explicitly introducing the expansion of eT0 into the second-order perturbed wave function. we regain the linearized connected .61). −ω ) = 2 l=0. 1997) discussed in Section 13.115) The dynamic polarizability. This approximation treats H 0 as completely deﬁned by its left and right ˆ eigenvectors. ˆ 0 E0 − H 0 )T − 0L −1 (13. analogous to (13. − L +1 (13.5. −ω ) = −2E (2) (ω. in which all terms correspond to connected diagrams and then make the same approximation as in the static case. −ω ) = 2 L ˆ 0 |(E0 − H 0 )T ˆ(1) R (+ω )Θ− |R ˆ0 − L −1 ˆ 0 |(E0 − H 0 )T ˆ(1) R (−ω )Θ− |R ˆ0 . αr r (ω.115). Perera. An alternative choice for the resolvent R (±ω ) is in terms of the EOMCC left and right eigenfunctions of the non-Hermitian Hamiltonian H 0 .117) If we neglect the last two terms in (13. Nooijen et al. we obtain the time-dependent analog of the EOM CI-like approximation (13.

extensive and intensive (Sekino and Bartlett 1999). Kobayashi.13. 1994.6 EOM-CC treatment of frequency-dependent properties 461 approximation that retains the SOS form for the dynamic polarizability.117) gives the beneﬁt that the full exponential expansion can be used. though the right-hand transition moment in this approximation is linked. As also discussed in Section 13. 1997). de Mer´ as et al. The result is a time-dependent analog of the connected approximation (13.72) (Sekino and Bartlett 1999). Nooijen et al. but this is at the cost of a more diﬃcult evaluation. The numerical diﬀerences between these approximations are very small for most applications. including the frequency-dependent polarizabilities (Rozyczko. When the quadratic terms are included there are also slight changes to the values of the dipole strengths (Koch. retaining the last two terms of (13. Sekino and Bartlett 1999).5. . Perera.

Although. they arise usually because of the degeneracy or quasidegeneracy of the reference determinants. including a single excitation from a closedshell state to an open-shell singlet state. As already discussed in Chapters 8 and 13. since then the only choices that need to be made are of the basis set and the level of correlation treatment. they require an SRCC solution for an initial state (not necessarily the ground state) to initiate the procedure. once initiated multireference target states are available by the diagonalization of an eﬀective Hamiltonian matrix in 462 .1 Introduction As in the case of quasidegenerate perturbation theory (Chapter 8). these situations arise primarily for certain open-shell systems that are not adequately described by a high-spin single determinant (such as transitionmetal atoms). These methods have the advantage of being operationally of single-reference form. as well as spin-ﬂip CC (Krylov 2001). the conditions discussed here involve nondynamic correlation eﬀects that are not described well by truncated SRCC at practical levels of treatment. allow the treatment of many inherently multireference target states. many open-shell and multireference states can be treated by EOM-CC methods. for excited states in general and for studies of bond breaking on potential-energy surfaces.4. As shown in Section 13.14 Multireference coupled-cluster methods 14. multireference coupled-cluster (MRCC) theory is designed to deal with electronic states for which a zero-order description in terms of a single Slater determinant does not provide an adequate starting point for calculating the electron correlation eﬀects. While single-reference coupled-cluster (SRCC) methods are very eﬀective in treating dynamic electron correlation. which normally requires two equally weighted determinants in its zero-order description. doubleionization and double-electron-attachment EOM-CC. Furthermore.

The diﬀerences between these approaches will be discussed in the following sections. intermediate-Hamiltonian MRCC (IH-CC). The methods employed in MRCC are a combination of the techniques of QDPT (Chapter 8) and SRCC (Chapters 9 and 10).1 Introduction 463 a determinantal representation. In each case. when operating on the model and Ω = ΩP space. H (14. Ψα = ΩΦα = ΩP With intermediate normalization. Hilbert-space MRCC also has a state-speciﬁc (SS-CC) version designed so as to obtain a single solution based on a multireference zero-order function. They fall into two main classes: Hilbert-space MRCC (HS-CC. produces the space spanned by the perturbed wave functions. Q ˆ I = |ΦI ΦI | . Ω is split into two components: ˆ+Q ˆΩ . ˆ Φα . Fock-space MRCC also has an additional version. the projection operator P by a set of model functions Φα . P (14. ˆ 0 ]P [Ω. and Fock-space CC (FS-CC.14. or SU-CC). which. are based on the generalized Bloch equation ˆ=V ˆ ΩP ˆ − ΩP ˆV ˆ ΩP ˆ.e. Ω=P where ˆ=ˆ ˆ= Q 1−P I (14. ˆ= P α |Φα Φα | = α ˆα .4) |ΦI ΦI | = I ˆI . or VU-CC). usually in the form of state-universal CC . the multireference approach is the most general and natural way to extend the reach of high-accuracy electronic-structure calculations. Q (14. this is analogous to multireference CI (MRCI) but includes proper factorization of the disconnected clusters and leads to extensive solutions. the methods can be based either on complete model spaces (which are often too large for practical applications and also suﬀer from intruder-state problems) or on incomplete model spaces (which introduce some complications and may result in some loss of extensivity). All multireference methods.2) ˆ is the wave operator. P ˆα = |Φα Φα | . the space . also called valence-universal MRCC.5) is the projector onto the orthogonal (or complementary) space. including QDPT and MRCC. i.3) (14. As in the case of perturbation theory.1) ˆ projects onto a model space spanned As in QDPT.

6) Another common form of the generalized Bloch equation can be obtained ˆ =H ˆ . As in all coupled-cluster methods. From (14. P (14. the functions Ψα . are not individually ˆ but span the space of eigenfunctions Ψα for which the eigenfunctions of H model space forms a zero-order approximation. In analogy with the case of QDPT. (14.7) This form is more convenient for MRCC because. i. as shown by substituting these expansions into the Schr¨ odinger equation ˆ Ψα = Eα Ψα .8) ˆ into diﬀerent excitation levels.4) it follows that ˆ Ψα = Φα . Ω = eT .3) and (14. instead of an orderby-order expansion of the wave operator Ω we express it in exponential form.10) are the bonnes fonctions (Bloch 1958) already discussed in Section 8.1) and using H ˆΩ = P ˆ as can be seen from (14.11) . The details of this expansion and expand T and of the treatment of the model space will diﬀer between the two classes of MRCC methods. Ψα = β Ψβ Cβα (14.e.3). H (14.1. at the same time ˆ0 + V by rearranging the terms in (14. HS-CC and FS-CC. the weights (transformation coeﬃcients) Cβα are obtained by diagonalization of an eﬀective Hamiltonian. The result is noting that P ˆΩ = Ω H ˆ 0P ˆ+V ˆΩ = Ω H ˆ0 + V ˆ Ω H or ˆ Ω = ΩH ˆΩ. ˆ (14. H (14.464 Multireference coupled-cluster methods spanned by all the functions ΦI that are not in the model space. unlike QDPT. it does not involve an order-by-order expansion but is generally formulated in terms of the full Hamiltonian.4).9) ΩΦβ Cβα = ΩΦα . β = where Φα = β Φβ Cβα (14.1. As discussed in Section 8.

the generalized Bloch equation (14. we get the matrix eigenvalue equation Applying P ˆH ˆ ΩΦα = Eα Φα . Ω= α Ωα = α ˆα ˆ eT P α.14). ij (α)ˆ (14. P from which the transformation matrix C can be determined.7) can be written in the form ˆ Ω = ΩH ˆ eﬀ .14) Once the Bloch equation is satisﬁed. respectively. ˆ eﬀ is used here.12) (14. the wave operator is separated into individual wave operators for the diﬀerent model states.2 Hilbert-space state-universal MRCC The Hilbert-space approach to multireference coupled-cluster (MRCC) theory assumes a separate Fermi-vacuum deﬁnition. and thus a separate partition of the spinorbitals into hole and particle states.16) where α α ˆ1 ˆ2 ˆα = T +T + ··· T = ia ta a†ˆ i+ i (α)ˆ ijab tab a†ˆ b† ˆ jˆ i + ··· . we can obtain another expression for ˆ eﬀ by operating with Ω−1 on (14. Following the Ansatz of Jeziorski and Monkhorst. (14. ˆ eﬀ = Ω−1 H H (14. 8.2 Hilbert-space state-universal MRCC 465 ˆ from the left.14.17) In the last expression the primes on the summation signs indicate the exclusion of internal excitations (excitations that result in model-space . H (14. In this respect it resembles the Hose–Kaldor approach to incomplete-model-space QDPT (Section 8. for each model-space determinant (Jeziorski and Monkhorst 1981).8) but can be applied to both complete and incomplete model spaces.13) ˆ -space and whose eigenfunctions and eigenvalwhich operates entirely in P ues are Φα and Eα . With this notation. is called the eﬀective Hamiltonian operator . H (14.7 and Fig.15) 14. to distinguish it from the The notation H coupled-cluster eﬀective Hamiltonian H of Chapters 10–13. as in Chapter 8. H ˆΩ. The operator ˆ eﬀ = P ˆH ˆΩ.

In this case ˆα is equivalent to their exclusion the exclusion of internal excitations from T from Ωα .2) and Ω = ΩP α . the amplitude of the function is given by a b a b tab ij (α) = −[ti (α)tj (α) − tj (α)ti (α)] . expanding eT ˆn from T in its power series. For example.19). These requirements. α α ˆ1 ˆ2 ˆα +C + ··· P (14. If we expand the wave operator Ωα in a CI-like series (compare (1. But when the model space is incomplete some lower excitations ˆ -space functions. instead of excluding all internal exα such operators must be retained. any term in T space function is accompanied in (14.18) Ωα = 1 + C α is a linear combination of n-fold excitation operators on Φ .466 Multireference coupled-cluster methods ab determinants). as discussed in Chapter 8. However. provided that the model space is complete. also called general model spaces (GMSs).19) α that generates a modelˆn When the model space is complete. and ta i (α). the exclusion ˆα is not valid for incomof internal excitations from the cluster operators T plete model spaces.19) by disconnected-cluster terms generating the same model function.20) which would be nonzero if a ˆ†ˆ i and ˆ b† ˆ j and/or a ˆ† ˆ j and ˆ b†ˆ i are external excitations. The exclusion of internal excitations in (14. But this exclusion is not equivalent to the exclusion of such excitations ˆn C ˆα α because. their amplitudes being ˆn citations from T α have ˆn determined by the requirement that the corresponding terms in C jˆ i generates a b† ˆ zero amplitude in (14. then ˆn where C α ˆ require the exclusion of internal excitations from each (14. (14. As shown by Li and Paldus (2003a). expressed as products of lower-order excitations each of which also generates a model-state function. are much more desirable. complete model spaces are usually impractical because of their size and because of intruder-state problems and so treatments using incomplete model spaces (IMS).17) is not an approximation. are amplitudes. (14. which were termed the connectivity conditions . we ﬁnd that the exact (full CI) expansion must satisfy α ˆ2 ˆα 2 ˆα C =1 2 (T1 ) + T2 .8)). Thus. α 1 ˆα 3 ˆα T ˆα ˆα ˆ3 = 3! (T1 ) + T C 1 2 + T3 etc. if the operator a ˆ†ˆ model-space function when operating on Φα . so in the above disconnected-cluster terms can generate Q that the relevant connected-cluster amplitudes cannot be set to zero without introducing approximations. it is to be remembered that the deﬁnitions of particle and hole states are diﬀerent for each model state. tij (α) etc.

excitation. can be represented compactly in diagrammatic form. its amplitude must satisfy β = 0.g. The limitation on the spinorbital labels is indicated by the bars and double arrows in these diagrams. where the diagrams are in the Φα representation and the box at the top indicates the ﬁnal state Φβ .24) ˆ† u ˆy ˆ† v ˆ|Φα . enforcing the C-conditions ensures the extensivity of the MRCC energy when ˆ operators are truncated consistently at any excitation level.22) β + β + β =0 (14. but not both. they are deﬁned by just the two model states involved in the matrix element. Such spinorbitals were called local active spinorbitals by Li and Paldus. double. . For an excitation from internal state Φα to an internal state Φβ that diﬀers from it by a single. respectively. Section 8.1). of the two internal states. e. in analogy with the notation in Section 8. β = u x v y −u y v x. in analogy with such contributions in incomplete-model-space MRPT (subsection 8. (14. 8. the T .23) etc. triple.21) β + β = 0. The C-conditions indirectly introduce some disconnected-cluster contributions into the MRCC equations. (14..7. (14. .7 (see Fig.7. in analogy with the valence orbitals in a “local” version of the classiﬁcation of the spinorbital space used in the incomplete-model-space perturbation theory. Most importantly. . however.14. All these diagrams involve only the where |Φβ = |Φxy uv (α) = x spinorbitals that are occupied in one.2 Hilbert-space state-universal MRCC 467 (or C-conditions ) by Li and Paldus.8). The disconnected diagrams in these equations implicitly include all distinct permutations of their labels.

as eﬀ ˆ eﬀ |Φα = Φβ |H ˆ Ω|Φα = Φβ |He ˆ T |Φα .29) where normal order is deﬁned relative to Φα as the Fermi vacuum..27) ˆ eﬀ appearing in this The matrix elements of the eﬀective Hamiltonian H equation are obtained. (α)| = a left and the model function |Φα on the right.. Φab. = Φβ |H Hβα ˆα (14.468 Multireference coupled-cluster methods Substituting the deﬁnition of the wave operator (14...16)..31) The diagrammatic techniques of single-reference CC (Chapter 10) can be used for the left-hand side of these equations.14) may be written in the form ˆβ ˆ ˆ T He Pβ = β β ˆα −T ˆ β ˆ ˆ eﬀ ˆ eT P βH P . (α)|e β ˆα T ˆβ −T ˆ eﬀ |Φα ...13) and (14. (14. e |Φβ Φβ |H ij... but the equations differ from those of single-reference CC in two respects. of For the use of diagrammatic techniques. (α)|e (14.16). β ˆα ˆ T ˆα ˆ Pα = e−T He (14. (α)|H |Φα = β −T eﬀ Φab..... (14. (α) contained in the operators T : ˆα ˆ T ˆα −T He |Φα = Φab. ij. the generalized Bloch equation (14. A minor diﬀerence is the distinction between the treatment of internal and external excitations (relative to the model space)..27) can be replaced by ˆα = H ˆ − Φ α |H ˆ |Φ α ..h.... ij. eT |Φβ Hβα . we obtain equations for the ˆα external-excitation amplitudes tab.28) ˆ on the l. We can then use the CC eﬀective Hamiltonian (the similarity-transformed Hamiltonian) for model state Φα . the Hamiltonian H (14. ˆα ˆ T ˆα ˆα ˆ T = He Hα = e−T He C ..26) Applying an external-space determinant Φab. ˆ†ˆ iˆ b† ˆ j · · · Φα | on the ij. H N (14.30) ˆα are determined by the C-conditions. The internal-excitation amplitudes in T while the equations for the external-excitation amplitudes take the form α Φab..25) Projecting on the left with e ˆα we obtain and on the right with P ˆα ˆ β ˆ ˆ eﬀ ˆ e−T eT P β H Pα . ij.. A major diﬀerence is the appearance . ij. since the the constant term does not contribute to the matrix element. using (14.s.. (α)|e ˆα ˆβ (14...

28). When the model state Φβ represents a single excitation relative to Φα we obtain β eﬀ = Hβα β + eﬀ . cycling through the diﬀerent model-state equations using current approximate values for the amplitudes on the r. Φβ |T necessarily vanish unless the model space is complete and therefore the inˆα sertion of e−T in (14. For typical incomplete model spaces the number of such ˆα excitations is quite small and a series expansion of eT will then include just a few contributing terms. vanishes because it involves amplitudes for internal single excitations.31). until self-consistency is achieved between the input and output amplitudes.33) Thus we can use the diagrammatic formulas of single-reference CC to calculate these elements.34) ˆα ˆα that can contribute to Φβ |eT The only excitations in T involve local active spinorbitals. × Hαα (14. ˆα ˆα Instead. Hβα ˆα (14.h. . we insert eT e−T = 1 and obtain eﬀ = Φβ |eT Hα |Φα . (14. For the diagonal elements of H so that ˆα ˆα + · · · ) = Φα | Φα |e−T = Φα |(1 − T (14. which is unlinked (since Hαα part). allowing easy diagrammatic evaluation of a matrix element. with the minor diﬀerence that the internal excitations ˆα . We now consider in more detail the evaluation of the various matrix elements that appear in (14.2 Hilbert-space state-universal MRCC 469 of the coupling term on the right-hand side.s. are omitted from T ˆα does not Considering now the oﬀ-diagonal elements of (14.35) eﬀ is a closed disconnected The second term.14.32) and therefore ˆα ˆ T ˆα eﬀ Hαα = Φα |e−T He |Φα = Φα |Hα |Φα . ˆ eﬀ we note that Φα |T Paldus (2003a).28) and (14. Because of this coupling the equations are solved iteratively. following the analysis of Li and ˆα = 0.28) may not be justiﬁed for incomplete model spaces. and updating these amplitudes with new values as they become available.

The single-excitation amplitudes in the second and third term are not necessarily zero.... ˆα ˆβ (14. ij.22).. −T Sij... (14..470 Multireference coupled-cluster methods (14. (α)|e −T eT |Φxy... eliminating the unlinked term. except in the completemodel-space case.. I −T Φab. because they may correspond to external excitations individually and only lead to the internal state Φβ in the indicated combinations.. Furthermore.21)–(14. = Φab... of (14. these are analogous to the expressions T (14....21). (α) . we obtain ab..23) involving the C-conditions but they also contain excitations . β ) = Sij. ˆα ˆβ ˆα ˆβ (14. involving CI-like amplitudes. eT |Φβ ij.. Next we consider the ﬁrst factor in the sum on the r.31). (α.... corresponding to linear combinations of ˆ amplitudes and their products.s.. when the model space ˆ vertices vanish.36) The sum in the parentheses vanishes when the C-conditions are satisﬁed. (α.. (α)|e uv.....37) where the unlinked terms vanish when the C-conditions are satisﬁed. (α) . for a triple excitation we have β eﬀ Hβα β + + β + β = + β β + + β eﬀ × Hαα . (14. β ) = Φab. leaving only the ﬁrst is complete all diagrams that include T diagram for each matrix element. (α)|e uv.. As ˆ eﬀ |Φα are stated above. For a double excitation we get β β β eﬀ + + Hβα = β + eﬀ × Hαα . Similarly. |ΦI ΦI |eT |Φxy.39) The series expansions of the exponentials in this equation result in expressions. (14. ab. ij...h. uv.38) where |Φβ = Φxy.. the labels on the lines in the diagrams for Φβ |H restricted to local active spinorbitals. (α) . Inserting a resolution of the identity between the two exponentials..

928 26.854 A1 A1 1 A1 3 A1 1 A1 1 A1 3 A1 1 A1 1 A1 1 A1 a b −0. The diagrammatic representation and the evaluation of this expansion are described in detail by Paldus. Additional results and comparisons with those obtained using various different methods can be found in the review by Bartlett and Musial (2007).152 26. we are usually interested in just one electronic state.363 15. Li and Petraco (2004). 2004).550 17.456 10.669 10 −0.459 24. statec 3a1 → 4a1 3a1 → 4a1 1b1 → 2b1 1b1 → 2b1 2 1b2 2 → 4a1 3a1 → 5a1 3a1 → 5a1 2 3a2 1 → 4a1 2 2 1b2 → 2b1 CI 3 −0.431 10. 14.253 24. Several applications of Hilbert-space SU-MRCCSD were discussed by Li and Paldus (2003c. But.590 25.467 23. for most problems in electronic structure theory for which a multireference treatment is indicated. As an example we show in Table 14. for states of the water moleculea SU model-space dimension State 1 3 b Excitation gd. which will not be discussed further here.3 Hilbert-space state-speciﬁc MRCC 471 1.1 some of their results for several 1. see also Kucharski and Bartlett (1991b) for the explicit equations.588 23. This approach is most appropriate for cases of exact degeneracy of the model functions. as in the case of many open-shell states.414 24.364 6 −0.238149 9.241833 9. who compared model spaces of diﬀerent dimensions with high-excitation single-reference CI. An example of such a case is the . c Ground-state energies as E + 76 in Eh .3 A1 states of the water molecule. and to some extent to cases of close quasidegeneracy.356 15. Single-reference CISDTQP (single to ﬁve-fold excitations).964 20. SU-MRCCSD in a cc-pVDZ basis.425 10.937 20.3 Hilbert-space state-speciﬁc MRCC The salient feature of the state-universal form of Hilbert-space MRCC is the simultaneous treatment of several electronic states (equal in number to the dimension of the model space) on an equal footing.884 20. outside the model space.14. State-universal MRCC excitation energies.554 17.3 A 1 Table 14.238073 9.475 17. and the simultaneous computation of a number of states on an equal footing is an undesirable burden. in eV.367 15.756 Li and Paldus (2004).457 10.1.238252 9.

41) ˆ is the projector onto the model space. which often involve near degeneracy in some geometries but are far from quasidegenerate in most other geometries. called conﬁguration state functions ) similar to the model space in MRCC. the eigenvalue problem is replaced by a set of simultaneous equations. Chattopadhyay. In this section we shall focus on state-speciﬁc MRCC. their number may be smaller than the desired dimension of the model space. and the desired eigenstate obtained. ˆ |Ψα . Datta and Mukherjee 1998. the reference where P ˆ P -space will consist of several determinants {Φβ } containing all the inactive core orbitals plus various sets of active orbitals. it beneﬁts from the linked nature of the CC theory and. owing to the exponential nature of the wave operator. as in . Mahapatra and Mukherjee 1999). 1999. as usual. In CC theory. A state-universal treatment of such potential-energy surfaces is bound to introduce intruder-state problems for some geometries. and the Φα are linear combinations (the bonnes fonctions ) of these model determinants. State-speciﬁc MRCC is intended to provide a CC equivalent of multireference CI in generating a single electronic-state solution from a multideterminantal model space but. in which just one electronic state is computed using a multidimensional model space. usually implicitly (“direct CI”). Even when we are interested in a few electronic states at a time. unlike MRCC. as the determinantal basis states (both model states and external states) change their relative energy positions and order in diﬀerent regions of the PES. Typically. The desired SS-CC wave function Ψα for the electronic state α is obtained by the operation of a wave operator Ω on a linear combination Φα of model functions Φα .40) (14. Li and Paldus (2003b) proposed a generalization of SU-CC for this type of situation. |Φα = P (14. to which are added all excitations of speciﬁed levels (usually just single and double excitations. We shall follow the general approach of Mukherjee and co-workers (Mahapatra. includes higher excitations obtained as disconnected clusters of lower excitations. MRCISD) from the reference determinants. The Hamiltonian matrix in the basis of all these determinants is constructed.472 Multireference coupled-cluster methods calculation of potential-energy surfaces (PESs). |Ψα = Ω|Φα . The MRCI wave-function expansion is generated from a set of reference determinants (or spin-adapted linear combinations of determinants.

ˆβ (14. to obtain eﬀ eT |Φγ Hγβ Cβα + βγ β ˆβ ˆβ ˆ eT Q Hβ |Φβ Cβα − Eα β (14. Inserting the Jeziorksi–Monkhorst Ansatz (14. β ˆβ ˆ T ˆβ In order to introduce the connected-cluster operator e−T He = Hβ .9). ˆ Ω|Φ = Heﬀ = Φ|H ˆ ΦH β eT Φ ˆβ (14. (14. the wave operator Ω operates directly on a linear combination of ˆ -space functions from its model determinants and may generate the same Q operation on diﬀerent model determinants. Exchanging the dummy summation variables β and γ in the ﬁrst term of (14.10). Datta and Mukherjee 1999).42) is substituted into the Schr¨ odinger equation we obtain ˆβ ˆβ ˆ T He |Φβ Cβα − Eα eT |Φβ Cβα = 0 . β γ ˆγ ˆβ ˆβ (14. noting that P ˆ = insert the identity 1 = eT e−T = eT (P γ |Φγ Φγ |. unlike the case of SU-CC.46) The redundancy in the determinants is taken into account for each β by setting the corresponding terms in the β -sum to zero.40).42) A complication arises in this procedure due to the fact that. in matrix form.14. .44) ˆβ where.3 Hilbert-space state-speciﬁc MRCC 473 (14. the coeﬃcients of the linear combination are obtained as the components Cβα of the appropriate eigenvector cα of the Heﬀ matrix. we obtain |Ψα = Ω|Φα = β eT |Φβ Φβ |Φα = β ˆβ eT |Φβ Cβα . When the wave function (14. giving rise to redundancy and non-uniqueness in the formalism.16) for Ω into (14. we β β β β ˆ ˆ ˆ ˆ ˆ +Q ˆ )e−T .45) and Φ is a vector composed of the model-space determinants {Φβ }.44) leads to eﬀ eT |Φβ Hβγ Cγα + eT QHβ |Φβ Cβα − Eα eT |Φβ Cβα = 0 . (14.43) eT |Φβ Cβα = 0 . It is therefore necessary to omit the redunˆ -space functions from the expansions or to invoke a set of suﬃciency dant Q conditions (Mahapatra.

this equation is similar to a form of the SU-CC equation partitioned to separate one state from all the others.. Brown et al. These equations for SSˇ arksy (2000) and by Pittner.48) Hence. (β )|e |Φβ − Φij.4. CC were presented by Hubaˇ c... they are coupled only through Eα . several a posteriori corrections for restoring the extensivity have been developed. Except for the cα vector. the H4. Many of the currently applicable MRCC methods have been compared with high-level single-reference CC for several prototype multireference problems in a detailed paper by Evangelista. The equations that determine the amplitudes have the form ab. who also presented several applications demonstrating the eﬃcacy of the approach. it does not have renormalization terms and one can derive therefore quite straightforwardly the relevant equations for BWCC. unlike RSPT. Datta and Mukherjee (1999). 1983). Φab. (β )|e ij. Allen and Schaefer (2006). P4 and H8 quasidegenerate systems (Jankowski and Paldus 1980) and cyclobutadiene and methylene (Balkov´ a and Bartlett 1994. the main disadvantage of BWPT is that its energy is not extensive.. Another version of SS-CC is based upon Brillouin–Wigner PT (subsection 2.47) Thus we obtain a single solution for state α. As pointed out there.. Detailed discussions of extensivity considerations and the use of suﬃciency conditions for dealing with the redundancies were given by Mahapatra.474 Multireference coupled-cluster methods To obtain the amplitudes we multiply this Schr¨ odinger equation on the ˆβ − T ˆ and project onto the Q-space determinants. (β )|H |Φβ Cβα = 0 (for all β ). However. Shepard.. βγ γ (=β ) (14. the lack of extensivity defeats the purpose of a CC theory. However.. e |Φβ H ij. obtaining left by e ˆβ T ˆγ −T β ˆ eﬀ Cγα + Φab. The coeﬃcients Cβα are obtained as the components of the appropriate eigenvector cα of Heﬀ . (β )|H |e |Φβ = 0 ˆβ ˆβ (for all β ).. (14. (14.. T ˆ T Eα Φab. as well as corrections to the iterative procedures. ij. this approach would appear to be inferior to the SS-CC approach of Mukherjee and co-workers. also considered have been the insertion of Be into H2 (Purvis.. 1995).. making them virtually the same as the SRCC equations for each β once the contributions from the other active-orbital determinants are excluded from the β -equation... Nevertheless. To account for this deﬁciency in an approximate manner.. Pittner and C´ ˇ arsky and Hubaˇ C´ c (2002).2). .45). because its equations retain a dependence on the exact energy Eα (as does a truncated-CI calculation).

. It ﬁnds its major applications in the determination of energy diﬀerences. y. The model space is regarded as consisting ˆ onto the model space is composed of of diﬀerent sectors.) Fermi level active holes (u. for which many multiplets lie close in energy. wave operator for all states. . f. j. such as n = 0. .) Fig. . . . nmax .) (i. can deal with a set of electronic states with diﬀerent numbers of valence electrons. v. inactive particles (e.) particles (a. active space . n. 2. . . 14. repeated here as Fig. .1 The Fock-space approach The Hilbert-space state-universal approach treats several states with the same number of electrons and can directly compute excitation energies (as shown in Table 14. .8. This approach refers all equations to a single Fermi vacuum and uses a single.14. inactive holes (m. . . Kutzelnigg and Koch 1983). The projector P . A Fock space (Kutzelnigg 1982. valence-universal.1. . . .4 Fock-space valence-universal MRCC 14. Notation for the one-electron states in Fock-space MRCC. . active particles (x. Examples of the type of quasidegeneracy it can address include complicated open-shell atoms.4 Fock-space valence-universal MRCC 475 14. . 14. . . . The Fock-space approach starts formally with the same Bloch equations as used in Hilbert-space methods.1 to introduce an extended notation. .) holes . which is the union of Hilbert spaces for a range of numbers of electrons. 1. except that the meaning of the model functions and their projectors has to be extended to accommodate diﬀerent numbers of electrons. . .1). . . 8. .) . 1984.4. . b. . Fig. . . such as ionization potentials and electron aﬃnities as well as electronic-state excitation energies. . The classiﬁcation of the one-electron states is the same as in the Hose–Kaldor approach.

55) . The total model-space projector P np nh ˆ= P k=0 l=0 ˆ (k.1) = P xu x ˆ† u ˆ |0 0| u ˆ† x ˆ= xu x |Φx u Φu | . components we also need the P ˆ (2. which is useful in the treatment of ionization potentials: ˆ (0. Other sectors have a role in Auger processes etc.0) = |0 0| .476 Multireference coupled-cluster methods ˆ (p.1) projector removes one active orbital (creates an active Similarly.53) ˆ (1. ˆ (1.50) ˆ (0.1) = P u u ˆ|0 0|u ˆ† = u |Φu Φu | .2) = P 1 4 xyuv x ˆ† u ˆy ˆ† v ˆ|0 0|v ˆ† y ˆu ˆ† x ˆ= 1 4 xyuv xy |Φxy uv Φuv | .l) . The P x † ˆ |0 for all active particle labels x.2) = P 1 2 xuv x ˆ† u ˆv ˆ|0 0|v ˆ† u ˆ† x ˆ= 1 2 xuv x |Φx uv Φuv | . (14. ˆ is the sum of the sector projectors.h) producing p active particles and h active diﬀerent sector projectors P holes in the common Fermi vacuum (see Fig. The simplest sector projector is ˆ (0. for (n + 1)space determinants |Φ = x electron states: ˆ (1.1) projector creates all possible single excitations from active holes The P to active particles.0) = P x x ˆ† |0 0|x ˆ= x |Φx Φx | . (14.49) ˆ (1. (14. the P ˆ|0 for all active hole labels hole) to generate (n − 1)-electron states |Φu = u u.1).0) projector produces modelwhich projects onto the Fermi vacuum. 14.2) projector. In a similar manner.52) This sector projector is useful for the treatment of states dominated by single excitations from the Fermi vacuum. For states with strong double-excitation ˆ (2. P (14.2) projector.54) corresponds to shake-up eﬀects in photoelectron spectroscopy. (14.51) ˆ (1. P (14. (14. the P ˆ (1.

h) and includcluding model determinants belonging to sectors other than P ( k. Ψ(p. Rittby. Unlike the P ( p.0) ˆ (0.h) .0) HP (0. ˆ (p. and is obtained as ˆ ˆ (0.0) ˆ (0. through an exponential wave operator Ω: ˆ (p.4 Fock-space valence-universal MRCC 477 where np and nh are the numbers of particle and hole states. Chaudhuri. In complete-model-space (CMS) calculations. which consist of Sinha and Mukherjee 1989).0) .0) (1 + T ˆ + ···) = P ˆ (0.h) = 1 − P Q (14.h) projectors.k ) ˆ projectors. the second term of (14. Once the correlated wave function for the (0.56) ˆ (p. ing |0 for Q |0 cannot contribute because no connected diagrams can be constructed with it in the Bloch equations (Pal.4. diﬀerent Q electrons have most terms in common. 1) sector has been evaluated.1) + · · · . 14.0) H P P ˆ ˆ T ˆ −T ˆ ˆ (0. corresponding to single-reference CC. in the active space.h) P (14.0) . so that complete model spaces (IMSs).0) HP ˆ (0. Partitioning the model-space determinants by their numbers of active particles and holes leads to a partitioning of the eﬀective Hamiltonian: ˆ = P (0. Bartlett et al.0) H ˆ (1. +P (14.14.0) + P ˆ eﬀ P ˆ (1.0) eT = P because P operator. inwhich is restricted to terms with the same number of electrons as P ˆ (p.h) . P =P e He =P (14.57) The ﬁrst term on the right describes the vacuum-sector contribution. while the fourth term determines electronic excitation energies. The third term provides electron aﬃnities.h ) ˆ projectors with the same number of mutually exclusive sets.0) eT ˆ (0.0) He ˆ eﬀ P ˆ (0. ˆ (p.h) = ΩP ˆ (p.h) .0) + P ˆ eﬀ P ˆ (0.1) ˆH ˆ eﬀ P ˆ (0. respectively. 1988).57) provides the ionization energies.1) H P ˆ (1. For inˆ eﬀ . each based on the corresponding sector of the model space.2 The valence-universal wave operator The correlated wave function is also constructed in sectors.0) = P ˆ T ˆ (0.58) ˆ ˆ (0. For the projector onto the complementary space for each sector we have ˆ (p.1) H ˆ eﬀ P ˆ (1.59) . since T ˆ is an excitation ˆ (0. |0 may contribute to Ω and to H intermediate normalization is no longer valid (Mukherjee 1986.

0) = T S We also deﬁne the partial sums p h (14. ˆ ˆ (0.0) space has one more electron (active Each model determinant in the P particle) than |0 .h ) ˆ P sector.h) ˆ(k. this added electron. ˆ (1.63) S (p.h) |S |Φα .64) ˆ(0.61) where the braces indicate normal order. Ψ(0. This operator is also partitioned by sectors of Fock space. The objective of the valence-universal (VU) treatment reference T is to add the excitation amplitudes that can arise with the multidimensional model space and that can be used to correlate the new states available with this model space. although it can produce a nonvanishing result when operating on any model determinant that has at least p occupied active particle states ˆ(0.0) P P (14.l) S (p ≤ np . To proceed.h) operates on a model determinant in the ˆ (p. The S on |0 and provides the usual deﬁnition of the t amplitudes of the singleˆ operator.0) + S S=S with ˆ.0) + S ˆ(0.nh ) .1) + · · · + S ˆ(np .0) .60) but a more general treatment is needed for the other sectors. l ≤ h.h) sector they generate determinants belonging on a determinant in the P ˆ (p.h) = S (p.65) The components of S operate on model determinants. Usually S P ( p.h) ˆ (p.0) operator operates and at least h unoccupied active hole states. respectively.h) = k=0 l=0 ˆ(k.0) sector this process simply reduces to the usual single-reference CC solution.h) and and Φα are determinants belonging to Q amplitudes.h) . Sinha. (14. (14.0) = eT . we express the wave operator in terms of an auxiliary operator S (Lindgren 1978.62) (14. and new terms are needed to provide correlation eﬀects for ˆ(1. thus the only nonvanishing matrix elements of S components are to Q ∗(p. ˆ(0.0) operator. Here ΦI ˆ(p. Mukhopadhyay and Mukherjee 1986): Ω = {eS } .h) . ˆ(1.1) + S ˆ(1.h) − S (14.l) (p.h) (p.0) ˆ (0. This task is accomplished by the new S . h ≤ nh ) . ∆S (p. When operating ˆ (p. ∆S = S − S ˆ(0. ˆ(0. and furnish the s of the form ΦI ∗(p.478 Multireference coupled-cluster methods For the (0. with k ≤ p.0) .

sau ˆ† u ˆ†ˆ ix ˆ}P xi {a aiux (14. 14.1) = ˆ(1. Operators for any other sector of Fock space can be diagrammed as P in Fig.67) ˆ(1. because that information is evident from the type S of subscripts and superscripts on the amplitude symbol. except that ˆ(0. possibly accompanied by additional excitations.1) = S 2 ˆ(1.0) = S 2 ˆ (1.14.1) = ˆ(0.0) the requirements are minant belonging to P 2 more relaxed: any x → a excitation may be accompanied by any i → b ˆ operators ˆ(1. ˆ ˆ j replace e ˆ† . 14. su ˆ† m ˆ }P m {u 1 2 ijau (14. The thick bars in this ﬁgure represent the correˆ operators and have the value of the corresponding s amplitude. se ˆ† x ˆ}P x {e 1 2 ixab (14. The S ˆ (1. ˆ(1.1) P ˆ (1.1) .1) sua ˆ† a ˆ† ˆ jˆ i}P ij {u ˆ(1.2.0) ix ˆ}P sab ˆ †ˆ b†ˆ xi {a ˆ(1.69) ˆ (0. Our notation for the s amplitudes does not specify the sector of ˆ to which they belong.68) (14. For S ˆ(1.4 Fock-space valence-universal MRCC 479 ˆ (1.0) .1.1) .0) and other S excitation.1) P S 2 ˆ (0.0) P ˆ (1.0) = ˆ(1.1) = ˆ(0.0) and is excluded. Such terms do not contribute at all for complete model spaces .1) P S 1 mu ˆ (0.1) ). sponding S ˆ(1.1) operator is given by etc. A diagrammatic representation of S is shown in Table 14.0) to an inactive particle which excites the active particle produced by P state.1) P S 1 ux ˆ (1.1) . su ˆ† x ˆ}P x {u ˆ (1. Using the designation of Fig.1) ˆ.71) ˆ(1.1) will be of the same form as S It is easy to see that S † ˆ (0. ˆ (1.70) (14. back to the vacuum tor taking the |Φu x model function.66) (14.0) P S 1 ex ˆ (1.1) . generated by P state |0 . An x → y excitation would result in a model deterˆ (1. because it can ﬁll the vacant hole state created by P u ˆ will replace x while m. The excitation due to S 1 to inactive particle).2.0) has to be of the form x → e (active particle etc.1) The prime on the summation sign in the expression involving the S 2 operator indicates that the a and i labels cannot both refer to active orbitals at the same time (if they did then the resulting state would belong to ˆ (1.0) . ˆ b† : ˆ (0.1) operator contains only active lines and is a de-excitation operaThe S 1 ˆ (1.

0) = T S 2 1 2 a ˆ†ˆ i} ia ti {a ab ˆ†ˆ b† ˆ jˆ i} ijab tij {a i a i a j ˆ(0. Meissner and Bartlett 1991).2.1) = S 1 ˆ(1.68)–(14.1) S 2 a j i b ˆ(0. The heavy bar represents the ˆ operator. Model spaces comprising complete sectors are called quasicomplete. Fig.S and S operators at the CCSD level.1) = S 2 1 2 u ˆ† m ˆ} mu sm {u ua ˆ† a ˆ† ˆ jˆ i} ijau sij {u i u ˆ(1. (CMSs). and has the value of the corresponding s amplitude.1) ˆ ˆ S . Diagrammatic representation and algebraic expressions for the S (1. In the deﬁnition S ˆ(1. 14. ˆ (1.1) is the highest sector in particular. and they do not contribute to the evaluation of excitation energies in important special cases of incomplete model spaces (IMSs) (Sinha.480 Multireference coupled-cluster methods ˆ(0. to MRCCSD calculations.0) S 2 a x ˆ(1.1) S 2 ˆ(1. A critical feature of the VU approach is the need to insure normal ordering in products of S operators.1) S 1 b ˆ(1.1) S 1 ˆ(0. though they can contribute to the Fock-space wave function. in which P and deﬁnes the model space. Mukhopadhyay and Mukherjee 1986.0) S 2 i u ˆ(0.0) = S 2 1 2 e ˆ† x ˆ} ex sx {e ab ˆ†ˆ b†ˆ ix ˆ} ixab sxi {a a ˆ(0.0) = S 1 ˆ(1. This property applies. as indicated by the braces in the exponential Ansatz .1) = S 1 ˆ(0.1) = S 2 u ˆ† x ˆ} ux sx {u au ˆ† u ˆ†ˆ ix ˆ} aiux sxi {a x u x ˆ(1.0) = T ˆ1 = S 1 ˆ2 = ˆ(0.0) ˆ(0.1) the prime on the summation sign indicates exclusion of the case in which of S 2 both upper labels (a and i) refer to active orbitals (see (14.71)).0) S 1 e x ˆ(1.1) (1.0) .0) S 1 m u ˆ(0. such as those arising from the expansion of the exponential eS .

Normal ordering is ˆ operators easily imposed in the single-reference theory. . In particular. In fact. ˆ = {eS }P ˆ . the T and their products are always in normal order because they consist of operators (for particle creation and hole annihilation) all of the same kind with ˆ respect to normal ordering. As an example we examine the factorization shown in Fig. .72) . requires explicit product normal ordering in VU theory (Lindgren 1978). but going beyond ﬁrst order introduces the requirement for product normal order.3. which is trivial in the singlereference theory.4 Fock-space valence-universal MRCC 481 x a i + a x i = a × y x i z y D1 u y z D2 u z u Fig. has the value D1 = xu sa y szi ˆ† x ˆ† u ˆ†ˆ iz ˆy ˆ} . to diagrams 56 and 57 of Fig. except for the reversed direction of two of the lines. A consequence of this situation is that the factorization of sums of diﬀerent time orderings of disconnected diagrams. 14. The disconnected diagram D1 . and evaluation of their contractions via the generalized ˆ1 . 14. Normal ordering of the Hamiltonian and of the cluster ΩP ˆ2 .14.61). in ﬁrst order.101). in ﬁrst order the S operators can be repAs is the case for T resented by the one-body and two-body Hamiltonian operators of MBPT and QDPT theory. and therefore the exponential eT is also in normal order. the disconnected diagrams in Fig. as shown for example in (9.97) and (9. 8. But the products of S operators in VU theory do not have this property and require explicit normal ordering of the exponential in (14.3.61). axu {a εa ε y yzi (14.5. The braces on the right-hand side of the equation denote normal order. Illustration of the factorization of disconnected diagrams in the openshell case (Lindgren and Morrison 1986). 14. ˆ operators. interpreted as a single diagram. Such ﬁrst-order approximations do not need explicit normal ordering. .3 are equivalent. T operators T Wick’s theorem have been used throughout this book. (14.

Similarly. which involves a model operator {eS }P ˆ (0.s. we have (S ˆ only one vacant active hole available for the two S (0.h. While S ˆ (0. ˆ† x ˆ† u ˆ†ˆ axu {a εxu ε zi yzi (14. A useful consequence of product normal ordering is that there can be no contractions among products of S operators. its value is xu sa sa sxu y y szi † † †ˆ zi { a ˆ y ˆ } × { x ˆ u ˆ {a ˆ† y ˆ}{x ˆ† u ˆ†ˆ i z ˆ } = iz ˆ} . In the multireference case. Such problems cannot arise in a closed-shell case involving products of ˆ operators.1) )2 P associated with s amplitudes. of the equation in Fig. 14.73) Using the factorization theorem and adding the two diagrams gives a factored denominator product.3). however.1) is ˆ(0. 14. Several simpliﬁcations are possible in expressions containing the VU wave ˆ .1) = 0 since there is ˆ(0. disregarding the braces around it for now.5. Only by imposing product normal order (indicated by the braces in Fig. xu a εxu εa ε ε y y zi zi (14. the active orbital operators do not anticommute.h.1) P space of determinants with one active hole vacant. consider the (0. 1) sector.482 Multireference coupled-cluster methods (For the determination of the denominators see Section 8. since all such products are automatically in normal order and T all creation and annihilation operators anticommute. ˆ† x ˆ† u ˆ†ˆ xu {a εa ε y zi (14. iz ˆ}} = {a ˆ† x iz ˆy ˆ} . The correct result is restored by imposing normal order explicitly. as is done on the r. even though these operators may have open active hole lines that could have been contracted in the absence of normal ordering. and cannot simply be reordered. as seen by the fact that contraction is possible between y ˆ † and x ˆ .3 to agree.s.75) The product of creation and annihilation operators in this equation is not in normal order. the r. of Fig. (14. showing that the correct products will be obtained in a Fock-space theory once Ω = {eS } is product normal ordered. 14. First.76) {{a ˆ† y ˆ}{x ˆ† u ˆ†ˆ ˆ† u ˆ†ˆ do we get the two sides of the equation in Fig. 14. unless they are inside a single normal product. with value D2 = xu sa y szi iz ˆy ˆ} . .74) However.) Reversing the time order gives diagram D2 .3 is a product of two diagrams and. eliminating the six-index denominator factor: D 1 + D2 = xu sa y szi iz ˆy ˆ} .1) operators.3. Lindgren (1978) oﬀered this approach as a simple resolution of this problem.

0) Since S 14.h) ˆ (p.4 Fock-space valence-universal MRCC 483 ˆ(1.h) has to terminate.j ) = 0 if k > i ˆ(1. the VU wave operator can be factored in the form =T ˆ ∆S ˆ ˆ = eT {e }P .0) S ˆ (1.0) } + 1 {S ˆ(0. S ˆ(0. when operating VU operator {e∆S }P (1 .1) {e∆S }P ˆ(1. while formally an exponential.79) Our remaining objective is to derive the Fock-space VU equations.h) H H{e∆S }P (14.80) Left-multiplying by e−T gives ˆ ˆ T ˆ ∆S ˆ (p.h) ˆ eﬀ P ˆ (p. Substituting (14. +1 2 {S 2 (14. ˆ (p.0) P ˆ (1. A consequence of this property is that the expansion of the ˆ (p. h) sector.1) .2) or S ˆ(2.1) + S ˆ(1. in general.l) P ˆ S .79) in the Bloch equation (14.1) } P ˆ(0. ˆ there is no contribution from S on P As a result.2) .1) + S ˆ(1.77) (14. S ˆ (i. Thus the VU ˆ(0.1) = S ˆ(1.h) ˆ eﬀ P ˆ (p. which will allow us to determine the amplitudes of ∆S . we get ˆ ∆S ˆ (p.81) or ˆ eﬀ P ˆ (p.4.1) vanish and.h) .1) P ˆ(k. Another key simpliﬁcation derives from the observation that terms like (1 ˆ (1.1) )2 P ˆ(p.h) H ˆ T {e }P = eT {e∆S }P He (14.1) .0) . S ˆ(2.14) for the (p.1) = S ˆ(1.0) + S ˆ(0. a nonzero (S ˆ(1.h) = 0. the amplitudes in such a product derive from sectors below the sector of interest.3 The Fock-space Bloch equations ˆ. However. ΩP (14. As we shall see presently.2) .82) .1) .0) + S ˆ(0. ˆ (1. 1) and (1. S ˆ(1.0) and S ˆ (0.h) = {e∆S }P ˆ (p.h) . ˆ (p.78) ˆ(0.h) H e−T He {e }P = {e∆S }P ˆ (14. In particular. ˆ (1.1) .1) = 0 and in general (S ˆ (p.1) P ∆S P and ˆ (1.1) S result can be obtained for products such as S wave operator. 0) sectors and only has nonlinear terms involving products of s amplitudes in higher sectors.1) S ˆ(1.14.1) P and/or l > j .h) )2 P ˆ (1. reduces to a linear form in the (0. 1) ˆ(2.

1) (0. Such diagrams are familiar r.1) }H ˆ eﬀ(0.4(A)) are distinguished by two labels.85) (14.1) = P H ˆ (0. which requires S 1 (0.h) provides the amplitude equations. Chapter 8.4–14. For S1 and their number is nh nh . Mukhopadhyay and Mukherjee 1989).1) }P ˆ (0. 1) ˆ ˆ requires knowledge of the S and S amplitudes. This is a general ˆ(p. and noting that P ˆ (p. 14.h) requires the feature of the VU equations: solution of the equations for S ( k. ˆ(0.1) }P ˆ (0.6 for VU-CCSD. short bars across a line indicate limitation to active orbitals. The main diﬀerence from the single-reference theory is the appearance of “folded” diagrams.h) {1 + Now projecting by P ˆ (p.5 and 14. ˆ (0.s. that depend upon H from MR-MBPT.2).1) and double excitations from these model determinants.83) ˆ (p.h) {e∆S }P ˆ (p. The energies for each sector are obtained by diagonalizing the Heﬀ matrix for the corresponding sector.84) The simplest example is provided by the equations for the (0.6.1) = 0 . 14.1) As can be seen from Figs. which requires a model space of one-active-hole determinants. projection of (14.h) . ˆ(1.82) by Q ˆ (p. Chaudhuri.h) = P ˆ (p.h) {e∆S }H ˆ eﬀ(p.h) H{e∆S }P ˆ (p.h) ∆S = 0. S2 .484 Multireference coupled-cluster methods ˆ (p. Q (14.1) H{1 + S ˆ(0. ˆ eﬀ(0.1) . we get an equation for the ˆ (p.h) = P ˆ (p.1) H{1 + S ˆ(0. Finally. Considering ﬁrst the (0. As in Chapter 8.l ) ˆ for all k ≤ p and all l ≤ h (obviously prior solution of the equations for S not including k = p and l = h simultaneously). The various S 1 . the number of active hole states.1) − {1 + S ˆ(0.h) .1) equations (Fig. the VU amplitude ˆ eﬀ can be diagrammed as shown in skeleequations and the associated H ton form in Figs.h) . 14. the number of such determinants is nh . solving the equations for S (1 . H (14. 1) sector. of the equations. 1) sector. 14. where nh is the number of inactive hole states. An alternative demonstration of the subsystem embedding conditions can be obtained by comparing the number of amplitudes to be determined by the equations for each sector with the number of such equations.h.h) H{e∆S }P ˆ (p. from the ˆ eﬀ . For an MRCCSD calculation we need to consider all single and ˆ(0.h) = P ∆S + · · · }P eﬀ ˆ : sectors of H ˆ eﬀ P ˆ (p. This requirement for the hierarchical solution for all sectors of Fock space lower than the sector of interest has been called the “subsystem embedding condition” (Haque and Mukherjee 1984.h) = Q ˆ (p.h) = P ˆ eﬀ(p. 0) (0 .1) u ˆ ˆ the single-excitation amplitudes are sm (see Fig. Q (14.h) because P ˆ (p.86) Using the standard diagram-generation techniques.h) H ˆ (p.

Similarly. Diagrammatic representation of the VU-CCSD equations for the (1. as speciﬁed by the subsystem embedding condition. 1) ˆ(1.6 that we also need the amˆ(0. 1) sector ˆ(1.5 and 14. (C) = + + + .1) . 14. but most also apply to a truncated MRCCSD calculation with (1. 1) (1 .2.4. the same as the number of amplitudes to be determined. 0) ˆ1 .0) and S plitudes for S 1 1 independently without ﬁrst solving the lower (1. a similar result will be obtained for the S 2 the corresponding numbers will be np np and 1 2 np nh (np ) . negating ˆ eﬀ and thus intermediate normalization. Obis equal to the number of S 2 ˆ(1. 14. (A) is the equation for S eﬀ ˆ (C) is for the one-electron part of H (represented by the rectangle). 14. (B) is for S ˆ2 and and (0. 0) and (0.14. allowing unique determination of all ˆ(0.4(B)).4 Fock-space valence-universal MRCC 485 (A) + + + − = 0. Fig. For VU-CCSD This situation is to be contrasted with the case of S (1 . But there is no contribution to H 1 2 2 nh np (nh ) . u) and the ﬁnal inactive hole line (m). as can be seen from Fig.1) the amplitudes. 1) sectors in skeleton form. In this latter case the S 1 thus to the wave function.1) equations (Fig. 14. where it introduces a |0 contribution. 1) ˆ ˆ and S but. the relevant operators are S 1 2 (1 . the number of possible amplitudes sub for S ij 2 ˆ(0. respectively. ˆ is not needed and we have enough equations to determine S S 1 2 Howover. 1) sectors. The equations and diagrams given here are appropriate for the CMS case. 1) as the ˆ(1. . (B) + + + + − = 0.0) operator except that viously. we can see from Figs.) The number of these equations is thus nh nh .1) and therefore cannot solve the (1. (The internal lines are summed over and have dummy summation indices.1) operator contributes to Ω and top-level sector.1) . those for the initial active hole line (e.g. ˆ(1.

Consider.6. 14.1) operator can be used to demonstrate another aspect of the VU The S formalism. the contribution of ˆ(1. the following instance of the second and fourth diagrams of Fig. It is possible to have the unusual situation in which. for energy calculations only.7. Diagrammatic representation in skeletal form of the VU-CCSD equation ˆ eﬀ is shown in Fig. 14. 14. ˆ(1.1) and the lack of intermediate normalization can be ignored. ˆ(1. The two-electron part of H for S 2 to the energy. after an acˆ (1. each lower sector must include all model states having as orbital indices all allowed subsets of the orbital indices of the selected model states of the sectors above it. in which the second diagram includes an .1) operator can be calculated from wave function is also of interest. shown diagrammatically in Fig. it can be reoccupied by the tive hole state has become unoccupied by P next operator.1) .5. (14.1) . Once the model states have been speciﬁed for the top sector. Fig. the S 1 the corresponding Bloch equation. Its contribution to the wave function is given by (x) (u) = ux u tx u sx |0 . 14.87) The IMS case imposes consistency requirements on the selection of the model states for the various sectors.5.486 Multireference coupled-cluster methods + + + + + + + + + + − − − − =0. If the S 1 ˆ(1. for example. Therefore.

14. Fig.7. Fig. of H − − − = 0. Diagrammatic representation in skeletal form of the two-electron part ˆ eﬀ .14.4 Fock-space valence-universal MRCC 487 = + + + + + + + + + + .6. Diagrammatic representation of the VU-CCSD equation for S 1 emptying and reoccupation of the active hole state u at the internal hole line (no other active hole label can replace the u on this line. Both diagrams have the same initial and ﬁnal state and have the matrix . although inactive hole lines would also produce valid contributions): a (b) u x x u i a + (b) i u .1) . 14. ˆ(1.

where such eﬀects are known to be quite important.488 Multireference coupled-cluster methods element χaubi of H (see Eq. 10.89) identify the hole state m as most strongly involved in the ionization. The second form in (14. The required presence of these spectator terms in the VU theory limits the orbital relaxation found naturally in an FS calculation in comparison with a ˆ model that beneﬁts fully from the presence of an exponential operator eT1 . }|Φm = (1 + S (14. To improve upon the accuracy. R electron from the vacuum state and generates a CI-like expansion for the reˆ sulting ionized state and then eT introduces ground-state correlation terms ˆ m . An illustration would be the poor accuracy of core-ionization states. with and include S xuv 3 similar contributions at higher levels. since they are fundamentally simple. Bandyopadhyay and Mukherjee 1999).2) we would also have sbuv as a spectator amplitude. In the second. In the ﬁrst. because the full power of the exponential is not there.90) . eT |0 creates a correlated ground ˆ m removes an electron from this CC state and introduces state and then R ˆ m removes an the correlation changes due to ionization. While operationally R the commutation of T ˆ. result in the correct description of the ionized-state that. together with R correlation.4 Relationship to EOM-CC Another interesting aspect is the similarity of VU-CC to EOM-CC (Chapter 13). 14.89) ˆm = R i riˆ i+ 1 2 ija a rij {a ˆ† ˆ jˆ i} The dominant terms in (14.88) + ··· .4. In the VU-CC case. (14. selective inﬁnite summations of spectator terms have been recommended and.88) takes advantage of ˆ m cannot be deterˆ and R ˆ m . such summations can in fact be accomplished (Jana. The second amplitude in this sum is called a spectator amplitude . (14.1) )|Φm . the Ansatz is |Ψm = {eS ˆ(0. The lower parts of bu the two diagrams contribute the sum su x + sxu .1) ˆ(0.70) as a common factor. If we were to go beyond VU-CCSD ˆ(1. The IP-EOM-CC Ansatz for this state is ˆ ˆˆ ˆ m eT |0 = eT R |Ψm = R m |0 . Consider an ionized state dominated by the removal of an electron from hole state m. the two forms of the Ansatz in (14.88) allow two diﬀerent mined before T ˆ interpretations of the result.

2) are column vectors consisting of elements of types ri a . (14. Normally. Speciﬁcally.1) )2 P because (S active-hole determinant and Φ(0.4 Fock-space valence-universal MRCC 489 where |Φm = m ˆ |0 is a one-vacant-hole-state model-space determinant and the eﬀective Hamiltonian matrix is ˆ eﬀ |Φ(0. Also.1) is a row vector of the one-vacantˆ(0. (14.1) |H (14.1) | is its adjoint column vector.1) Φ(1. Φi |riˆ a a jˆ i}|0 = rij Φa ˆ† ˆ .96) (14.14. r= r(0. to bring out the analogy between EOM-CC and VU-CC.1) |H|Φ(1.1) r(1.95) For the remaining equations we deﬁne the vectors |Φ = |Φ(0.1) |H(1 + S ˆ(0.94) −1 we have now multiplying on the right by rm Φm |H|Φm + n=m −1 Φm |H|Φn rn rm + ija a −1 Φm |H|Φa ij rij rm = ωm .1) Φ(0. Using (13.1) ωm . respectively. ij |rij {a (14.1) = Φ(0.92) the IP-EOM-CCSD eigenvalue equation for hole state m can be written in partitioned form as Φ(0.1) = 0.93) r(1.1) Φ(1. if we were to use sector designators such as (0.1) and r(1.2) C where ωm is the ionization energy of hole state m. we may designate a subset of orbitals as active and restrict the sector deﬁnitions to this subset.91) and the IP-EOM-CCSD matrix implicit in (13.2) Φ(1.1) .25) is that the latter has a dimension of one-hole single and double excitations while the former has the smaller dimension nh of just the one-vacant-active-hole determinants.1) |H|Φ(0.2) |H|Φ(0.1) r(1. 1) and (1.97) . Thus we expect to have a matrix partitioning relating the two matrices. |Φ(0.2) are row vectors consisting of all determinants of types Φi and Φa ij .2) |H|Φ(1.1) )|Φ(0.2) r(0. 2) in EOM-CC then they would refer to all orbitals since there is no separation between active and inactive orbitals in this model. respectively. The main diﬀerence between the matrix (14. (14. Heﬀ(0. Here |Φ(0.1) = Φ(0.2) .91) ˆ (0.2) (14. and rij Separating the equation for rm gives Φm |H|Φn rn + n ija a Φm |H|Φa ij rij = rm ωm .2) = r(0.25) and noting that i|0 = ri .1) and |Φ(1. r(0.

Each principal ionization potential for any orbital in IP-EOM-CC can be obtained and. however.98) From (14. treating active hole states similarly to state m. Meissner and Malinowski 2000).100) With minor modiﬁcations we can identify any number of orbitals as active.1) − S(0.1) |Φ(0. Heully and Malrieu 1993). unlike a true multireference theory. Here we shall focus on the approach in Meissner (1998). because such spaces tend to include highenergy multiple-excitation model states that contribute little to the wave function but are included merely for completeness. It does not matter which orbitals are actually active. These equations now take the form −1 −1 = rωm rm . but it was also used to provide “dressings” for extensivity-corrected multireference conﬁguration interaction (Daudey. the amplitude equations for S 2 14. we regain the valence-universal Heﬀ . Meissner and Nooijen 1995.1) . ˆ eﬀ |Φm = Φm |H(1 + S ˆ(0. They are particularly common in complete-model-space treatments. The prime on (0.91).1) + Φ|H|Φ (S(0. ˆ -space states with energies close to or within the Intruder states are Q range of the energies of the model states. Applications of this approach to Fock-space multireference coupled-cluster theory mostly use a similaritytransformation technique (Stolarczyk and Monkhorst 1985. Its original primary application was to quasidegenerate perturbation theory. 1) indicates the omission of Φm and rm .98) becomes Φ|H|Φ(0. the VU amplitude equation (14.1) )|Φm .490 Multireference coupled-cluster methods and the adjoint column vector Φ|.5 Intermediate-Hamiltonian Fock-space MRCC The intermediate Hamiltonian (IH) approach was introduced by Malrieu. ˆ(0. 1) −1 . Φm |H (14. (14. we also have the simpliﬁcation that there are no single excitation contributions to ˆ(0.99) ˆ eﬀ |Φm with ωm (which it equals when Heﬀ is digonalized) Identifying Φm |H (0 . Φ|H|Φm + Φ|H|Φ rrm (14. Durand and Daudey (1985) in order to overcome intruder-state problems in perturbative and other methods. In the event that all orbitals are active. Because .1) Heﬀ ) = 0 .1) = rrm with Φ|S and S Furthermore. The intruder states tend to be single (or very low) excitations into the lowest part of the inactive particle space or from the highest part of the inactive hole space. expanding the number of active orbitals does not provide any better an approximation. Meissner 1998.

of dimension n.8): the Q ˆ (this 1. where they often cause discontinuities in potential-energy curves and surfaces (Malrieu. the rest of the Q ˆ int operates in the augmented model space The intermediate Hamiltonian H MI = M ∪ M . the union of the main and intermediate spaces. they often result in divergence or severe convergence diﬃculties. 14. with projector Q ˆI = Q 3. Intruder states also tend to occur at regions of avoided crossings. the main model space M . ˆ -space. ˆ−P ˆ. The total space is divided into three subspaces (Fig. The intermediate-Hamiltonian approach overcomes intruder-state problems by interposing a buﬀer space between the model space and the rest of ˆ -space. whose projector is P is the original model space of the problem). MI has ˆI = P ˆ+P ˆ .8.5 Intermediate-Hamiltonian Fock-space MRCC 491 they generate very small or negative energy denominators in perturbative or iterative solution procedures. while the eﬀective dimension nI = n + n and projector P ˆI Q ˆ Q ˆ P ˆI P ˆ P Fig. . usually 2. the intermediate space M . Projectors for the various subspaces in the intermediate-Hamiltonian approach. 14. of dimension n and projector P ˆ consisting of a part of the original Q-space closest in energy to M and including the states responsible for the intruder problems. Durand and Daudey 1985). projector Q ˆ .14. Its complement has the ˆ.

101) In the usual Fock-space MRCC the solution of the amplitude equations ˆ eﬀ . tonian H = e−T He ˆ )H(1 + X ˆ) H = e−X HeX = (1 − X and require this transformation to satisfy the equation ˆ (1 − X ˆ )H(1 + X ˆ )P ˆ=0 Q (14. We have Hamiltonian H ˆ+Q ˆ=P ˆI + Q ˆI = ˆ P 1. eigensolutions of H ˆI -space. using diﬀerent ˆ int . X (14.492 Multireference coupled-cluster methods ˆ eﬀ is deﬁned in the main model space M . this procedure can be used to protect the main model space ˆ -space.104) in successive sectors of Fock space. The intermediate Hamiltonian H ˆ int . of dimension nI .106) . such that deﬁne a new operator X ˆ = {e∆S − 1} . have been proposed and applied. Writing this equation in the form ˆ H(1 + X ˆ )P ˆ−Q ˆX ˆP ˆ H(1 + X ˆ )P ˆ=0 Q (14.103) Next we introduce a similarity transformation of the CC-transformed Hamilˆ ˆ T ˆ (Meissner 1998). eX = 1 + X ˆ ˆ. e−X = 1 − X ˆ =Q ˆX ˆP ˆ. (14. X and for which ˆ2 = 0 . from intruder states in the low-lying part of the Q ˆ int . without wave functions. is the similarity-transformation method of Meissner (1998).105) ˆ ˆ (14. We ˆ note that the valence-universal wave operator is Ω = {eS } = eT {e∆S } and ˆ . Diﬀerent variants of the intermediate-Hamiltonian approach. The remaining nI − n eigenvalues to the corresponding eigenvalues of H are essentially arbitrary. X ˆ ˆ. We ﬁrst consider a general similarity-transformation treatment that is analogous to the Fock-space MRCC approach of the previous section. the Because of the arbitrariness of the extra solutions produced by H intermediate Hamiltonian and the equations determining it are not unique. which can be diprovides an eﬀective Hamiltonian H agonalized to produce n eigensolutions that are equal to a subset of the ˆ . They may (but need not) be rough approximations ˆ . Because no requirements are imposed on this extra to the eigenvalues of H set of solutions.102) (14. and n of its eigenvalues are required to be equal is deﬁned on the P ˆ eﬀ . according to the subsystem embedding condition. of dimension n. auxiliary conditions to deﬁne a unique H A particularly convenient and eﬀective version of this approach when the objective is limited to obtaining energies.

P we have ˆ H(1 + X ˆ )P ˆ−Q ˆX ˆP ˆH ˆ eﬀ P ˆ = 0. Q (14.14. ˆ and thus has multiple soluEquation (14. Also.82). .1) + S ˆ(1. the (0.0) S ˆ(1.4.111) (14. The principal eigenvalues obtained by these two methods should be the same. of single excitations.112) ˆ=P ˆX ˆ =P ˆZ ˆP ˆ Z (14. This step is followed by solution of the (1.109) ˆZ ˆ=Z ˆY ˆ = 0).0) sector) and construction of the CC transformed Hamiltonian H.1) (14.0) S ˆ(1.1) + S ˆ(0.1) P ˆ= S ˆ(1. ˆ(0.0) S ˆ(1. this procedure suﬀers from the intruder-state problem.4.0) + S (14. (the last line follows because Y ˆ subspace for the (1. 1) sector consists For an IH-CCSD calculation the P ˆ I -subspace.0) S Y 2 2 1 2 2 1 2 2 (Bartlett and Musial 2007). The intermediate-Hamiltonian treatment starts with a conventional CC calculation for the Fermi-vacuum state (i.110) ˆ ˆ ˆ ˆ (14.1) P ˆ (1. or by EA-EOM and IP-EOM (Chapter 13).5 Intermediate-Hamiltonian Fock-space MRCC 493 and noting that (14.113) Z 1 1 1 1 2 ˆ = S ˆ(1.114) ˆ(0.83) for the FS eﬀective Hamiltonian can be written as ˆ H(1 + X ˆ )P ˆ=H ˆ eﬀ . as described in Section 14. H = e−Z e−Y HeY eZ = (1 − Z where ˆ =Q ˆI X ˆ =Q ˆI Y ˆP ˆ. while double excitations are put into the Q ˆ and Y ˆ operators are given Explicitly.0) + S ˆ(0. either by VU-CC.114).1) + S ˆ(0. as discussed in subsection 14. To overcome these diﬃculties we introduce the intermediate-Hamiltonian approach by splitting the transformation (14.4. This ˆ procedure provides all the s amplitudes needed for the determination of Y using (14. as follows: ˆ )(1 − Y ˆ )H(1 + Y ˆ )(1 + Z ˆ) . X ˆ = (1 + Y ˆ )(1 + Z ˆ) .e.105) is a quadratic equation in X tions.0) and (0. while the eigenvectors would diﬀer in their normalization. the Z by ˆ(0.102) and (14.1) . Y and ˆ =Y ˆ +Z ˆ.104) into two successive transformations.108) (14. ˆ = (1 − Z ˆ )(1 − Y ˆ) 1+X 1−X (14.107) which is equivalent to the Bloch equation (14.1) + S ˆ(1.1) + S ˆ (1.1) sectors. noting (14.111).

The representation includes disconnected diagrams but ex- ˆ int = H + + + + + + + + + + + Fig. A diagrammatic representation of the intermediate Hamiltonian for IHCCSD.109). H (14. we do not need to evaluate Z Z ˆI (1 − Y ˆ )H(1 + Y ˆ )P ˆI are related by a similarity transformation via e ˆ .115). As a result. we ˆ(1.494 Multireference coupled-cluster methods Assuming that our objective is limited to obtaining energies. and thus have the same eigenvalues. The initial and ﬁnal states ˆI -space. The diagonalˆ int then provides the desired n eigenvalues.1) sector. 14. thus avoiding the do not need to evaluate S 2 need to solve the (1. 14.115) which is analogous to (14. derived from (14. not wave ˆ )H(1 + X ˆ )P ˆI ˆ because P ˆI (1 − X functions.1) .113).107) for the eﬀective Hamiltonian.9. Diagrammatic representation in skeletal form of the intermediate Hamiltonian for IH-FS-MRCCSD (Bartlett and Musial 2007).114). (14. is shown in Fig.9 (Bartlett and Musial 2007). and P as seen in (14. plus n additional ization of H ˆ eigenvalues deriving from the contributions of the original buﬀer space P and likely to contain any intruder-state eﬀects. the last term in (14. The next step is then the evaluation of the intermediate Hamiltonian ˆ int = P ˆ int P ˆI = P ˆI H ˆI (1 − Y ˆ )H(1 + Y ˆ )P ˆI = P ˆI H(1 + Y ˆ )P ˆI . in all diagrams are restricted to be in the P .

197 −0.004 0.250 −0. a tensivity is preserved because the disconnected terms cancel in the diagonalization process.072 −0. n) indicates an active space of m spatial-orbital particles and n holes (which equals 2m spinorbital particles and 2n holes).608 1. which include a description of the geometries and basis sets.022 −0. (1996) for N2 . 4)c −0.2. 4)c 1 EOM-CC CCSD CCSDT Πg Σ− u 1 ∆u 1 Πu 1 1 9.211 9.312 0.095 0. (1995) for Ne.008 0.009 0.447 9. in eV.028 0. and of Koch. 4)c −0. They provide a number of sample applications at both the MRCCSD and MRCCSDT levels and give comparisons with EOM-CC and the full-CI results.14.089 0.034 0. An extension of the intermediate-Hamiltonian treatment to the IH-CCSDT case has been described by Musial and Bartlett (2008).162 C2 (8. b The calculations of Christiansen. Their results are shown here in Table 14.385 5.245 −0.029 −0.015 −0.089 −0.009 −0.220 −0.005 0. Koch.102 −0.028 −0.081 0.584 10.033 −0.485 −0.309 0.206 −0.602 7.718 13.029 −0. The inner-shell electrons are frozen.004 0. c The notation (m.398 18.5 Intermediate-Hamiltonian Fock-space MRCC 495 Table 14.047 0.317 0.005 0.026 −0.035 1 Πu Σ+ u 1 B1 A2 1 A1 1 1 1 P0 D 1 P 1 S Musial and Bartlett (2008).189 0.017 0.237 0.251 −0.2.329 10.053 0. C2 and H2 O. for the N2 . .111 −0.240 −0. Vertical excitation energies. Jørgensen et al.256 18. Christiansen.213 18. The IH-CC and EOM-CC values are deviations from FCI (in eV).010 0.138 0.069 H2 O (4.005 Ne (7.136 0. Jørgensen et al. The full-CI values are given for comparisona IH-CC State FCIb CCSD CCSDT N2 (9.000 −0.028 −0.068 −0. C2 and H2 O molecules and the neon atom calculated using the IH-CC and EOM-CC methods.401 0.112 −0. 3)c −0.009 0.333 −0.874 16.

F. R. Lischka. Auer. Quantum Chem. Baumgartner. (1979). (New York) 207. (1992). Int. Coupled-cluster method for openshell singlet states. Baker. et al. Chem. Introduction to their holomorphic representations. (New York) 151. Chem.. Phys. S. 104. Adamowicz. Arponen. and Bartlett. 311–82. Phys. Symp. Alternants and continuous groups. Independent-cluster parametrizations of wave functions in model ﬁeld theories. Ann. A. D. 496 .. B. 1225–34. Phys. 144–8. I. A. A. Adams. 211–28. Ser. and Paldus. B.. Phys. Bernholdt D. Phys. and Kutzelnigg. and Bishop. J. (2006). 245–54. A 20. (1983). (1991). Proc. E. Staemmler.. 193. H. Phys. J. J. Quantum Chem. Rev. Lett. Arponen. J.. Phys. J. 18. Variational principles and linked-cluster exp S expansions for static and dynamic many-body problems. and Bartlett. Lett. 2 3. Ann. W. and Paldus. J. G. PNOCI (pair natural orbital conﬁguration interaction) and CEPA-PNO (coupled electron pair approximation with pair natural orbitals) calculations of molecular systems. 24–47. Jankowski. H. J.. (1979). F. G. (1984). London Math. R. 364–72. Ahlrichs.. V. 62. (1905). Automatic code generation for many-body electronic structure methods: the tensor contraction engine. Soc. Analytical gradients for the coupled-cluster method.References Adams. Chem. Mol. G. L. Balkov´ a. W.. Outline of the method for closedshell states. 67. Orthogonally-spin-adapted coupledcluster theory for closed-shell systems including triexcited clusters. Laidig. 1–17. 171– 217. Quasi-degeneracy and coupled-pair theories. R. (1975). I. J. K. R.

pp. I. Dordrecht). (1981). Bartlett.References 497 Balkov´ a. Ground states.. Cole. Rev. The quartic force ﬁeld of H2 O determined by many-body methods. Some consideration of alternative ans¨ atze in coupledcluster theory. J. Bartlett. J. 87. Chem. The exchange-correlation potential in ab initio density functional theory. Grabowski. 1047–131.. 102. Frenking. I. (1994). J. How and why coupled-cluster theory became the preeminent method in ab initio quantum chemistry. R. 101. (1986). Yarkony (World Scientiﬁc. S.. A multireference coupled-cluster study of the ground and lowest excited states of cyclobutadiene. and Br¨ andas. Chem. Bartlett. 1191–221. J. J. Bartlett. Phys. Chem. 125–49. Coupled-cluster theory: an overview of recent developments. Bartlett. Purvis. Simons (Reidel. J. Kaldor (Springer. J. 1697–708. S. K. Phys. Dykstra. (1995). 6579–91. Singapore). Part I. Chem. and Trucks. in Many-Body Methods in Quantum Chemistry (Lecture Notes in Chemistry. (2005). eds. J. eds. pp. and Shavitt. J. ed. W. Phys. Scuseria (Elsevier. R. 034104-1–12. (2005). Ermler. Phys.. R. Berlin). R. and Ivanov. On the singlet–triplet separation in methylene.. and Bartlett. A. R. 5467–77. Phys. H. A. A critical comparison of single. J. 8972-87. G. Bartlett. Bartlett. Noga. Phys. J. Hirata. 7116-23. Ann. Kim and G. pp. Kucharski. Amsterdam). R. Jørgensen and J. Analytical evaluation of gradients in coupled-cluster and many-body perturbation theory. 56. D. pp. Phys. Watts. Coupled-cluster approach to molecular structure and spectra: a step toward predictive quantum chemistry. (1987). Hsieh. Chem. The First Forty Years. S. Balkov´ a. J. R. in Modern Electronic Structure Theory. J. and Bartlett. 35–61. C. D. 359–401. Many-body perturbation theory and coupled cluster theory for electron correlation in molecules. G..versus two-determinant (generalized valence bond) coupled-cluster theory. Reduced partitioning procedure in conﬁguration interaction studies. R. R. 52). G. 32. R. W. (1995). S. J. 93. J.. J. in Theory and Applications of Computational Chemistry. . J. R.. J. U. D. (1972). No. C. I. R. ed.. Bartlett. Bartlett. II. (1989). A. J. P. J. Chem. J. Chem. C. (1989). in Geometrical Derivatives of Energy Surfaces and Molecular Properties. E. 122. Eﬀects of triple excitations.

The quartic force ﬁeld of H2 O determined by methods that include quadruple excitation eﬀects. Bartlett. coupled-pair many-electron theory. G. J. 12. 190–8. (1988). R. 543–4. Phys. and the importance of quadruple excitations for the correlation problem. and Shavitt. Erratum (1978).. Many-body perturbation theory. Chem. D. 271–6. J. Phys. ACES II program system. Phys. Rev. Molecular applications of coupled cluster and many-body perturbation methods. 8. Symp.. 281–91. Bartlett. Bartlett. 185–6. R. F. 165–73. and HF with many-body perturbation theory using Slater-type atomic orbitals. . J. 50. R. Chem. 255–65. P. Correlation energy in LiH. Schleyer. Quantum Chem. R. 29–36. Lett. D. and Silver. I. J. in Encyclopedia of Computational Chemistry. and Purvis III. J. M. Lett. Bartlett. Comparison of MBPT and coupled-cluster methods with full CI. I. Lett. G. J. Chem. Erratum (1976). and Stanton. R. Int. 29. 10. D. I. Coupled-cluster theory in quantum chemistry. R. Shavitt. (1983). D. R. 198.498 References Bartlett. Quantum Chem. Chem. Phys. J. 65–169. Symp. R.. 11. J. (1998). Bartlett. (1977a). Lett. Bartlett. J. and Noga. R. J. Quantum Chem. 561–81. 199–203. and Shavitt. Importance of triple excitations and inﬁnite summations. Chem. Bartlett. R. M. Chem. and Silver. D. The expectation value coupled-cluster method and analytical energy derivatives. Many-body perturbation theory applied to hydrogen ﬂuoride. R. 66–71. 79. J. J. J. G. Phys. 14. J. (1974a). 98. Int. Phys. Chem. Quantum Chem. M. (1977b). and Purvis III. Applications of post-Hartree– Fock methods: a tutorial. 37. Sekino. Lett. v. D. Rev. R. Erratum (1978). Bartlett. Symp. (1980). eds. Phys. 5.. (2007). Comput. J. (1994). 157–8. (1974b). H. Chem. Determination of the size-consistency error in the single and double excitation conﬁguration interaction model. and Watts. Int. Bartlett. and Purvis III. R. Bartlett. J. Lett. Mod. 71. Bartlett. and Purvis III. G. D. Quantum Chem. J. J. Quantum Chem.. (1978). J. Int. Quantum Chem. (1979). Erratum (1976). Phys. J. 291–352. BH. Int. and Musial. Phys. 150. J. Quantum Chem. Comparison of high-order manybody perturbation theory and conﬁguration interaction for H2 O. 57. Scripta 21.

Clark et al. B. Phys. and Horowitz. Phys. pp. Phys. E. K´ allay. H. H. L. R. J. Vol. Benchmark full conﬁguration-interaction calculations on H2 O. Watts. A. 496–512. Bauschlicher Jr. Quantum Chem. R. Linked-cluster perturbation theory for closed. (1986). and Noga. Brandow. and Knowles P. E.. J. J. C. (1956). (1986). Rev. 1. C. J. Brillouin. 1469–74. Phys.. 7 4. Brillouin.. J. L. Nuc. Quantum Chem. Rev. 10. 1–9. 054101-1–8. N. Noniterative ﬁfth-order triple and quadruple excitation energy corrections in correlated methods. Handy. Møller–Plesset theory for atomic ground state energies. Proc. Phys. Bloch (Academic Press. F. S. 552–60. C. in Proceedings of the International School of Physics “Enrico Fermi”. Quantum Mechanics of One. (1958). (1967). Soc. B. Variation-perturbation expansions and Pad´ e approximants to the energy. J. J. 329–47. T. (1975). Phys. Eﬀect of a repulsive core in the theory of complex nuclei. 6. 8. and Gauss. Ser. and Taylor P. Bauschlicher Jr. J. J. A 1. and Goscinski. Taylor. (2005). Rev. Lett. J. and Pople. 103. Linked-cluster expansion for open-shell nuclei. Radium. L. Benchmark full conﬁguration-interaction calculations on HF and NH2 . and Salpeter. 7 3. Brandow. Chem. B. (1990). Bomble. Bartlett. J. 513–22. 39. J. (1932). C. Adv. New York). C. 373–89. Int. E. Nuclear many-body problem. Bethe H. J. S. Phys. 165. H. Mod. Bethe. Coupledcluster methods including noniterative corrections for quadruple excitations. Chichester. Phys. Sur la d´ etermination des premiers ´ etats d’un syst´ eme de fermions dans le cas d´ eg´ en´ er´ e. Binkley. 771–828. UK). and F− . Les probl` emes de perturbations et le champs selfconsistents.References 499 N. 9. 1353– 90. (John Wiley & Sons.. Phys. Bethe H. W. 85. Champs self-consistents et ´ electron m´ etalliques (III). 1–7. Kucharski. Chem. M. F.and open-shell systems. Nuc. 91–105. Chem. E. Radium. W. ed C. (1970). 551–67. D. J. (1958). Y. O. (1933a). Linked-cluster expansions for the nuclear manybody problem. Br¨ andas. (1977). Phys. Bloch. 85. Sur la th´ eorie des perturbations des ´ etats li´ es.and Two-Electron Atoms (Springer. Stanton. J. Langhoﬀ. A. (1957). pp. R. 187–249.. J. Bloch. R. (1957).. E. Berlin). Brandow. 229–36. . Allinger. Ser. (London) A 238. 2779–83. pp. and Goldstone. 123. R.. J. S. Phys. (1966). 122–3. Chem. Course 36. E. P.

500

References

Brillouin, L. (1933b). La m´ ethode du champ self-consistent, Actual. Sci. Ind., No. 71. Brillouin, L. (1934). Les champs “self-consistents” de Hartree et de Fock, Actual. Sci. Ind., No. 159. Brueckner, K. A. (1955). Many-body problem for strongly interacting particles. II. Linked cluster expansion, Phys. Rev. 100, 36–45. Brueckner, K. A. and Levinson, C. A. (1955). Approximate reduction of the many-body problem for strongly interacting particles to a problem of self-consistent ﬁelds, Phys. Rev. 97, 1344–52. Burton, P. G., Buenker, R. J., Bruna P. J. and Peyerimhoﬀ, S. D. (1983). Comparison of perturbatively corrected MRD CI results with a full CI treatment of the BH ground state, Chem. Phys. Lett. 95, 379–85. Campbell, J. E. (1897). On the law of combination of operators bearing on the theory of continuous transformation groups, Proc. London Math. Soc. 28, 381–90. Chan, G. K.-L., K´ allay, M. and Gauss, J. (2004). State-of-the-art density matrix renormalization group and coupled cluster theory studies of the nitrogen binding curve, J. Chem. Phys. 121, 6110–6. Chattopadhyay, S., Mahapatra, U. S. and Mukherjee, D. (1999). Property calculations using perturbed orbitals via state-speciﬁc multireference coupled cluster and perturbation theories, J. Chem. Phys. 111, 3820–31. Chaudhuri, R., Mukhopadhyay Jr, D. and Mukherjee, D. (1989). Applications of open-shell coupled cluster theory using an eigenvalueindependent partitioning technique: approximate inclusion of triples in IP calculations, Chem. Phys. Lett. 162, 393–8. Chaudhuri, R., Sinha, D. and Mukherjee, D. (1989). On the extensivity of the roots of eﬀective Hamiltonians in many-body formalisms employing incomplete model spaces, Chem. Phys. Lett. 163, 165–70. Chiles, R. A. and Dykstra, C. E. (1981). An electron pair operator approach to coupled cluster wave functions. Application to He2 , Be2 , and Mg2 and comparison with CEPA methods, J. Chem. Phys. 74, 4544–56. Christiansen, O., Jørgensen, P. and H¨ attig, C. (1998). Response functions from Fourier component variational perturbation theory applied to a time-averaged quasienergy, Int. J. Quantum Chem. 68, 1–52. Christiansen, O., Koch, H., Jørgensen, P. and Olsen, J. (1996). Excitation energies of H2 O, N2 and C2 in full conﬁguration interaction and coupled cluster theory, Chem. Phys. Lett. 256, 185–94.

References

501

ˇ ıˇ C´ zek, J. (1966). On the correlation problem in atomic and molecular systems. Calculation of wavefunction components in Ursell-type expansion using quantum-ﬁeld theoretical methods, J. Chem. Phys. 45, 4256–66. ˇ C´ ıˇ zek, J. (1969). On the use of the cluster expansion and the technique of diagrams in calculations of correlation eﬀects in atoms and molecules, Adv. Chem. Phys. 14, 35–89. ˇ ıˇ C´ zek, J.and Paldus, J. (1967). Stability conditions for the solution of the Hartree–Fock equations for atomic and molecular systems. Application to the pi-electron model of cyclic polyenes, J. Chem. Phys. 47, 3976–85. ˇ ıˇ C´ zek, J. and Paldus, J. (1971). Correlation problems in atomic and molecular systems III. Rederivation of the coupled-pair many-electron theory using the traditional quantum chemical methods, Int. J. Quantum Chem. 5, 359–79. Coester, F. (1958). Bound states of a many-particle system, Nucl. Phys. 7, 421–4. Coester, F. and K¨ ummel, H. (1960). Short-range correlation in nuclear wave functions, Nucl. Phys. 17, 477–85. Comeau, D. C. and Bartlett, R. J. (1993). The equation-of-motion coupledcluster method. Applications to open- and closed-shell reference states, Chem. Phys. Lett. 207, 414–23. Crawford, T. D. and Sekino, H. (2009). On the performance of a sizeextensive variant of equation-of-motion coupled cluster theory for optical rotation in chiral molecules, Prog. Theor. Chem. Phys., in press. Crawford, T. D. and Stanton, J. F. (1998). Investigation of an asymmetric triple-excitation correction for coupled-cluster energies, Int. J. Quantum Chem. 70, 601–11. Dalgaard, E. and Monkhorst, H. J. (1983). Some aspects of the timedependent coupled-cluster approach to dynamic response functions, Phys. Rev. A 28, 1217–22. Dalgarno, A. and Stewart, A. L. (1958). A perturbation calculation of properties of the helium iso-electronic sequence, Proc. R. Soc. (London) A 247, 245–59. Daudey, J.-P., Heully, J.-L. and Malrieu, J.-P. (1993). Size-consistent selfconsistent truncated or selected conﬁguration interaction, J. Chem. Phys. 99, 1240–54.

502

References

Davidson, E. R. (1974). Conﬁguration interaction description of electron correlation, in The World of Quantum Chemistry, eds. R. Daudel and B. Pullman (Reidel, Dordrecht), pp. 17–30. Davidson, E. R. (1975). The iterative calculation of a few of the lowest eigenvalues and corresponding eigenvectors of large real-symmetric matrices, J. Comput. Phys. 17, 87–94. Davidson, E. R. and Silver, D. W. (1977). Size consistency in the dilute helium gas electronic structure, Chem. Phys. Lett. 52, 403–6. Dirac, P. A. M. (1927). The quantum theory of the emission and absorption of radiation, Proc. R. Soc. (London) A 114, 243–65. Ditchﬁeld, R. (1974). Self-consistent perturbation theory of diamagnetism. I. A gauge-invariant LCAO method for N.M.R. chemical shifts, Mol. Phys. 27, 789–807. Dunning Jr, T. H. (1989). Gaussian basis sets for use in correlated molecular calculations. I. The atoms boron through neon and hydrogen, J. Chem. Phys. 90, 1007–23. Dyson, F. J. (1949a). The radiation theories of Tomonaga, Schwinger, and Feynman, Phys. Rev. 75, 486–502. Dyson, F. J. (1949b). The S matrix in quantum electrodynamics, Phys. Rev. 75, 1736–55. Emrich, K. (1981). An extension of the coupled cluster formalism to excited states, Nucl. Phys. A 351, 379–96. Epstein, P. S. (1926). The Stark eﬀect from the point of view of Schroedinger’s quantum theory, Phys. Rev. 28, 695–710. Evangelista, F. A., Allen, W. D. and Schaefer III, H. F. (2006). High-order excitations in state-universal and state-speciﬁc multireference coupled cluster theories: model systems, J. Chem. Phys. 125, 1541131–16. Feynman, R. P. (1949a). The theory of positrons, Phys. Rev. 76, 749–59. Feynman, R. P. (1949b). Space–time approach to quantum electrodynamics, Phys. Rev. 76, 769–89. Frantz, L. M. and Mills, R. L. (1960). Many-body basis for the optical model, Nucl. Phys. 15, 16–32. Gauss, J. (1993). Eﬀects of electron correlation in the calculation of nuclear magnetic resonance chemical shifts, J. Chem. Phys. 99, 3629–43. Gauss, J., Lauderdale, W. J., Stanton, J. F., Watts, J. D. and Bartlett, R. J. (1991). Analytic energy gradients for open-shell coupled-cluster singles and doubles (CCSD) calculations using restricted open-shell Hartree–Fock (ROHF) reference functions, Chem. Phys. Lett. 182, 207–15.

References

503

Gauss, J., Ruud, K. and Helgaker, T. (1996). Perturbation-dependent atomic orbitals for the calculation of spin-rotation constants and rotational g tensors, J. Chem. Phys. 105, 2804–12. Gauss, J. and Stanton, J. F. (1995). Gauge-invariant calculation of nuclear magnetic shielding constants at the coupled-cluster singles and doubles level, J. Chem. Phys. 102, 251–3. Gauss, J. and Stanton, J. F. (2000). Analytic ﬁrst and second derivatives for the CCSDT-n (n = 1–3) models: a ﬁrst step toward the eﬃcient calculation of CCSDT properties, Phys. Chem. Chem. Phys. 2, 2047– 60. Gauss, J., Stanton, J. F. and Bartlett, R. J. (1991). Analytic evaluation of energy gradients at the coupled-cluster singles and doubles level using quasi-restricted Hartree–Fock open-shell reference functions, J. Chem. Phys. 95, 2639–45. Gerratt, J. and Mills, I. M. (1968). Force constants and dipole-moment derivatives of molecules from perturbed Hartree–Fock calculations. I, J. Chem. Phys. 49, 1719–29. Goldstone, J. (1957). Derivation of the Brueckner many-body theory, Proc. R. Soc. (London) A 239, 267–79. Goscinski, O. (1967). Continued fractions and upper and lower bounds in the Brillouin–Wigner perturbation scheme, Int. J. Quantum Chem. 1, 769–80. Gwaltney, S. R. and Head-Gordon, M. (2001). A second-order perturbative correction to the coupled-cluster singles and doubles method: CCSD(2), J. Chem. Phys. 115, 2014–21. Handy, N. C., Pople, J. A., Head-Gordon, M., Raghavachari, K. and Trucks, G. W. (1989). Size-consistent Brueckner theory limited to double substitutions, Chem. Phys. Lett. 164, 185–92. Handy, N. C. and Schaefer III, H. F. (1984). On the evaluation of analytic energy derivatives for correlated wave functions, J. Chem. Phys. 81, 5031–3. Haque, M. A. and Mukherjee, D. (1984). Application of cluster expansion techniques to open-shells: calculation of diﬀerence energies, J. Chem. Phys. 80, 5058–69. Harris, F. E., Monkhorst, H. J. and Freeman, D. L. (1992). Algebraic and Diagrammatic Methods in Many-Fermion Theory (Oxford University Press, New York). Hausdorﬀ, F. (1906). Die symbolische Exponentialformel in der Gruppentheorie, Ber. Verhandl. S¨ achs. Akad. Wiss. Leipzig, Math.Naturw. Kl. 58, 19–48.

504

References

Hirao, K. and Nakatsuji, H. (1978a). Cluster expansion of the wave function. Structure of the closed-shell orbital theory, J. Chem. Phys. 69, 4535–47. Hirao, K. and Nakatsuji, H. (1978b). Cluster expansion of the wave function. The open-shell orbital theory including electron correlation, J. Chem. Phys. 69, 4548–63. Hirata, S. (2003). Tensor contraction engine: abstraction and automated parallel implementation of conﬁguration-interaction, coupled-cluster, and many-body perturbation theory, J. Phys. Chem. A 107, 9887–97. Hirata, S. (2004). Higher-order equation-of-motion coupled-cluster methods, J. Chem. Phys. 121, 51–9. Hirata, S. and Bartlett, R. J. (2000). High-order coupled-cluster calculations through connected octuple excitations, Chem. Phys. Lett. 321, 216–24. Hirata, S., Fan, P.-D., Auer, A. A., Nooijen, M. and Piecuch, P. (2004). Combined coupled-cluster and many-body perturbation theories, J. Chem. Phys. 121, 12187–207. Hirata, S., Nooijen, M. and Bartlett, R. J. (2000a). High-order determinantal equation-of-motion coupled-cluster calculations for electronic excited states, Chem. Phys. Lett. 326, 255–62. Hirata, S., Nooijen, M. and Bartlett, R. J. (2000b). High-order determinantal equation-of-motion coupled-cluster calculations for ionized and electron-attached states, Chem. Phys. Lett. 328, 459–68. Hirschfelder, J. O., Byers-Brown, W. and Epstein, S. T. (1964). Recent developments in perturbation theory, Adv. Quantum Chem. 1, 255–374. Hose, G. and Kaldor, U. (1979). Diagrammatic many-body perturbation theory for general model spaces, J. Phys. B 12, 3827–55. Hose, G. and Kaldor, U. (1980). A general model space diagrammatic perturbation theory, Phys. Scripta 21, 357–61. Hose, G. and Kaldor, U. (1982). Quasidegenerate perturbation theory, J. Phys. Chem. 86, 2133–40. Hose, G. and Kaldor, U. (1984). General-model-space many-body perturbation theory: the (2s 3p)1,3 P states in the Be isoelectronic sequence, Phys. Rev. A 30, 2932–5. ˇ arsky, P. (2000). Size-extensivity correction Hubaˇ c, I., Pittner, J. and C´ for the state-speciﬁc multireference Brillouin–Wigner coupled-cluster theory, J. Chem. Phys. 112, 8779–84. Hubbard, J. (1957). The description of collective motion in terms of manybody perturbation theory, Proc. R. Soc. (London) A 240, 539–60.

References

505

Hubbard, J. (1958a). The description of collective motion in terms of many-body perturbation theory. II. The correlation energy of a freeelectron gas, Proc. R. Soc. (London) A 243, 336–52. Hubbard, J. (1958b). The description of collective motion in terms of many-body perturbation theory. III. The extension of the theory to the non-uniform gas, Proc. R. Soc. (London) A 244, 199–211. Huby, R. (1961). Formulae for non-degenerate Rayleigh–Schr¨ odinger perturbation theory in any order, Proc. Phys. Soc. (London) 78, 529–36. Hugenholtz, N. M. (1957). Perturbation theory of large quantum systems, Physica 23, 481–532. ¨ Hylleraas, E. A. (1930). Uber den Grundterm der Zweielektronenprobleme − + ++ usw. Z. Phys. 65, 209–25. von H , He, Li , Be Jana, D., Bandyopadhyay, D. and Mukherjee, D. (1999). Development and applications of relaxation-inducing cluster expansion theory for treating strong relaxation and diﬀerential correlation eﬀects, Theor. Chem. Acc. 102, 317–27. Jankowski, K. and Kowalski, K. (1999a). Physical and mathematical content of coupled-cluster equations: correspondence between coupledcluster and conﬁguration-interaction solutions, J. Chem. Phys. 110, 3714–29. Jankowski, K. and Kowalski, K. (1999b). Physical and mathematical content of coupled-cluster equations. II. On the origin of irregular solutions and their elimination via symmetry adaptations, J. Chem. Phys. 110, 9345–52. Jankowski, K. and Kowalski, K. (1999c). Physical and mathematical content of coupled-cluster equations. III. Model studies of dissociation processes for various reference states, J. Chem. Phys. 111, 2940–51. Jankowski, K. and Kowalski, K. (1999d). Physical and mathematical content of coupled-cluster equations. IV. Impact of approximations to the cluster operator on the structure of solutions, J. Chem. Phys. 111, 2952–59. Jankowski, K., Kowalski, K., Grabowski, I. and Monkhorst, H. J. (1999). Correspondence between physical states and solutions to the coupledcluster equations, Int. J. Quantum Chem. 75, 483–96. Jankowski, K. and Paldus, J. (1980). Applicability of coupled-pair theories to quasi-degenerate electronic states: a model study, Int. J. Quantum Chem. 18, 1243–69. Janssen, C. L. and Schaefer III, H. F. (1991). The automated solution of second quantization equations with application to the coupled cluster approach, Theor. Chim. Acta 79, 1–42.

506

References

Jayatilaka, D. and Lee, T. J. (1993). Open-shell coupled-cluster theory, J. Chem. Phys. 98, 9734–47. Jeziorski, B. and Monkhorst, H. J. (1981). Coupled-cluster methods for multidimensional reference states, Phys. Rev. A 24, 1668–81. Jordan, P. and Klein, O. (1927). Zum Mehrk¨ orperproblem der Quantentheorie, Z. Phys. 45, 751–65. ¨ ¨ Jordan, P. and Wigner, E. P. (1928). Uber das Paulische Aquivalenzverbot, Z. Phys. 47, 631–51. Jørgensen, P. and Simons, J. (1983). Ab initio analytical molecular gradients and Hessians, J. Chem. Phys. 79, 334–57. Kaldor, U. (1973a). Many-body perturbation-theory calculations with ﬁnite, bound basis sets, Phys. Rev. A 7, 427–34. Kaldor, U. (1973b). Many-body perturbation-theory calculations for excited molecular states, Phys. Rev. Lett 31, 1338–40. Kaldor, U. (1984). General model space perturbation theory: excitation and ionization of N2 , J. Chem. Phys. 81, 2406–10. K´ allay, M. and Gauss, J. (2003). Analytic ﬁrst derivatives for general coupled-cluster and conﬁguration interaction models, J. Chem. Phys. 119, 2991–3004. K´ allay, M. and Gauss, J. (2004). Analytic second derivatives for general coupled-cluster and conﬁguration interaction models, J. Chem. Phys. 120, 6841–8. K´ allay, M. and Surj´ an P. R. (2000). Computing coupled-cluster wave functions with arbitrary excitations, J. Chem. Phys. 113, 1359–65. K´ allay, M. and Surj´ an, P. R. (2001). Higher excitations in coupled-cluster theory, J. Chem. Phys. 115, 2945–54. Kelly, H. P. (1963). Correlation eﬀects in atoms, Phys. Rev. 131, 684–99. Kelly, H. P. (1964a). Correlation eﬀects in many fermion systems. II. Linked clusters, Phys. Rev. 134, A1450–3. Kelly, H. P. (1964b). Many-body perturbation theory applied to atoms, Phys. Rev. 136, B896–B912. Kelly, H. P. (1968). Correlation structure in atoms, Adv. Theor. Phys. 2, 75–169. Kelly, H. P. (1969). Application of many-body diagram techniques in atomic physics, Adv. Chem. Phys. 14, 129–90. Koch, H., Christiansen, O., Jørgensen, P. and Olsen, J. (1995). Excitation energies of BH, CH2 and Ne in full conﬁguration interaction and the hierarchy CCS, CC2, CCSD and CC3 of coupled cluster models, Chem. Phys. Lett. 244, 75–82.

References

507

Koch, H., Christiansen, O., Kobayashi, R., Jørgensen, P. and Helgaker, T. (1994). A direct atomic orbital driven implementation of the coupled cluster singles and doubles (CCSD) model, Chem. Phys. Lett. 228, 233–8. Koch, H. and Jørgensen, P. (1990). Coupled cluster response functions, J. Chem. Phys. 93, 3333–44. Koch, H., Kobayashi, R., de Mer´ as, A. S. and Jørgensen, P. (1994). Calculation of size-intensive transition moments from the coupled cluster singles and doubles linear response function, J. Chem. Phys. 100, 4393–400. Koch, S. and Kutzelnigg, W. (1981). Comparison of CEPA and CP-MET methods, Theor. Chim. Acta 59, 387–411. Kohn, W. and Sham, L. J. (1965). Self-consistent equations including exchange and correlation eﬀects, Phys. Rev. 140, A1133–8. Kowalski, K. and Jankowski, K. (1998). Towards complete solutions to systems of nonlinear equations of many-electron theories, Phys. Rev. Lett. 81, 1195–98. Krogh, J. W. and Olsen, J. (2001). A general coupled cluster study of the N2 molecule, Chem. Phys. Lett. 344, 578–86. Krylov, A. I. (2001). Size-consistent wave functions for bond-breaking: the equation-of-motion spin-ﬂip model, Chem. Phys. Lett. 338, 375–84. Kucharski, S. A. and Bartlett, R. J. (1988). Multireference many-body perturbation theory, Int. J. Quantum Chem., Quantum Chem. Symp. 22, 383–405. Kucharski, S. A. and Bartlett, R. J. (1989). Coupled-cluster methods that include connected quadruple excitations, T4 : CCSDTQ-1 and Q(CCSDT), Chem. Phys. Lett. 158, 550–5. Kucharski, S. A. and Bartlett, R. J. (1991a). Recursive intermediate factorization and complete computational linearization of the coupledcluster single, double, triple, and quadruple excitation equations, Theor. Chim. Acta 80, 387–405. Kucharski, S. A. and Bartlett, R. J. (1991b). Hilbert space multireference coupled-cluster methods. I. The single and double excitation model, J. Chem. Phys. 95, 8227–38. Kucharski, S. A. and Bartlett, R. J. (1992). The coupled-cluster single, double, triple, and quadruple excitation method, J. Chem. Phys. 97, 4282–8. Kucharski, S. A. and Bartlett, R. J. (1993). Coupled-cluster methods correct through sixth order, Chem. Phys. Lett. 206, 574–83.

508

References

Kucharski, S. A. and Bartlett, R. J. (1995). Sixth-order many-body perturbation theory for molecular calculations, Chem. Phys. Lett. 237, 264–72. Kucharski, S. A. and Bartlett, R. J. (1998a). Noniterative energy corrections through ﬁfth-order to the coupled cluster singles and doubles method, J. Chem. Phys. 108, 5243–54. Kucharski, S. A. and Bartlett, R. J. (1998b). Sixth-order energy corrections with converged coupled cluster singles and doubles amplitudes, J. Chem. Phys. 108, 5255–64. Kucharski, S. A. and Bartlett, R. J. (1998c). An eﬃcient way to include connected quadruple contributions into the coupled cluster method, J. Chem. Phys. 108, 9221–6. Kucharski, S. A. and Bartlett, R. J. (1999). Connected quadruples for the frequencies of O3 , J. Chem. Phys. 110, 8233–5. Kucharski, S. A., Wloch, M., Musial, M. and Bartlett, R. J. (2001). Coupledcluster theory for excited states: the full equation-of-motion coupledcluster single, double and triple excitation method, J. Chem. Phys. 115, 8263–6. Kutzelnigg, W. (1977). Pair correlation theories, in Methods of Electronic Structure Theory, ed. H. F. Schaefer III (Plenum, New York), pp. 129–88. Kutzelnigg, W. (1982). Quantum chemistry in Fock space. I. The universal wave and energy operators, J. Chem. Phys. 77, 3081–97. Kutzelnigg, W. (1984). Quantum chemistry in Fock space. III. Particle– hole formalism, J. Chem. Phys. 80, 822–30. Kutzelnigg, W. (2003). Theory of electron correlation, in Explicitly Correlated Wave Functions in Chemistry and Physics. Theory and Applications, ed. J. Rychlewski (Kluwer, Dordrecht), pp. 3–90. Kutzelnigg, W. and Koch, S. (1983). Quantum chemistry in Fock space. II. Eﬀective Hamiltonians in Fock space, J. Chem. Phys. 79, 4315–35. Kvasniˇ cka, V. (1974). Construction of model Hamiltonians in framework of Rayleigh–Schr¨ odinger perturbation theory, Czech. J. Phys. B 24, 605–15. Kvasniˇ cka, V. (1977a). A diagrammatic construction of formal E independent model Hamiltonians, Czech. J. Phys. B 27, 599–628. Kvasniˇ cka, V. (1977b). Application of diagrammatic quasidegenerate RSPT in quantum molecular physics, Adv. Chem. Phys. 36, 345–412. Langhoﬀ, P. W., Epstein, S. T. and Karplus, M. (1972). Aspects of timedependent perturbation theory, Rev. Mod. Phys. 44, 602–44.

References

509

Langhoﬀ, S. R. and Davidson, E. R. (1974). Conﬁguration interaction calculations on the nitrogen molecule, Int. J. Quantum Chem. 8, 61–72. Larsen, H., Hald, K., Olsen, J. and Jørgensen, P. (2001). Triplet excitation energies in full conﬁguration interaction and coupled cluster theory, J. Chem. Phys. 115, 3015–20. Lauderdale, W. J., Stanton, J. F., Gauss, J., Watts, J. D. and Bartlett, R. J. (1991). Many-body perturbation theory with a restricted openshell Hartree–Fock reference, Chem. Phys. Lett. 187, 21–8. Lee, T. J. and Rendell, A. P. (1991). Analytic gradients for coupled-cluster energies that include noniterative triple excitations: application to cis- and trans-HONO, J. Chem. Phys. 94, 6229–36. Lee, Y. S. and Bartlett, R. J. (1984). A study of Be2 with many-body perturbation theory and a coupled-cluster method including triple excitations, J. Chem. Phys. 80, 4371–7. Lee, Y. S., Kucharski, S. A. and Bartlett, R. J. (1984). A coupled cluster approach with triple excitations, J. Chem. Phys. 81, 5906–12. Erratum (1985). J. Chem. Phys. 82, 5761. Li, X. and Paldus, J. (1994). Automation of the implementation of spinadapted open-shell coupled cluster theories relying on the unitary group formalism, J. Chem. Phys. 101, 8812–26. Li, X. and Paldus, J. (1997). Unitary group based coupled cluster methods and calculation of molecular properties, in Recent Advances in Coupled-Cluster Methods, ed. R. J. Bartlett (World Scientiﬁc, Singapore), pp. 183–219. Li, X. and Paldus, J. (2003a). General-model-space state-universal coupledcluster theory: connectivity conditions and explicit equations, J. Chem. Phys. 119, 5320–33. Li, X. and Paldus, J. (2003b). N -reference, M -state coupled-cluster method: merging the state-universal and reduced multireference coupledcluster theories, J. Chem. Phys. 119, 5334–45. Li, X. and Paldus, J. (2003c). The general-model-space state-universal coupled-cluster method exempliﬁed by the LiH molecule, J. Chem. Phys. 119, 5346–57. Li, X. and Paldus, J. (2004). Performance of the general-model-space stateuniversal coupled-cluster method, J. Chem. Phys. 120, 5890–5902. Lindgren, I. (1974). The Rayleigh–Schr¨ odinger perturbation and the linkeddiagram theorem for a multiconﬁgurational model space, J. Phys. B 7, 2441–70.

F. Lower bounds to energy eigenvalues in perturbation-theory ground state. 611. . 1. 6171–88. P. 2. P. L¨ owdin. Rev. Atomic Many-Body Theory. Lindgren. 197–266. P. 3. I. P. X.-O. (1985a). 3.-O. (1985b). and Mukherjee. Datta. Datta. Chem. 357–72. A note on the linked-diagram and coupled-cluster expansions for complete and incomplete model spaces. Pindzola (Cambridge University Press. 94. V. Cambridge). A 139. Physical interpretation by means of density matrices. Phys. (1977). second edition (Springer. D. I. 8. B. Phys. Suppl. J. U. L¨ owdin. Phys. Math. IV. Phys. 7. Linked-diagram and coupled-cluster expansions for multi-conﬁgurational complete and incomplete model spaces. Quantum theory of many-particle systems. 12. J. (1962b). I. Radium. S. Lindgren. Symp. (1962a).-O. 97.. Some aspects on the exact self-consistent ﬁeld theory. U. J. Mahapatra. 397–409. eds. Ser. and convergence problems in the method of conﬁguration interaction. Quantum Chem. Studies in perturbation theory. Quantum Chem. Lindgren. I. 3–38. London.-O. Phys. S. L¨ owdin. Phys. L¨ owdin. Review of diﬀerent approaches and discussion of some current ideas. A state-speciﬁc multi-reference coupled cluster formalism with molecular applications. Phys. I. B. (1937). (1978). (1986). D. Chem. pp. (1959). Lindgren. P. Phys. Quantum theory as a trace algebra. Development of atomic many-body theory. P. Mahapatra. Th´ eorie quantique des courants interatomiques dans les combinaison aromatiques. S. Boyle and M. Studies in perturbation theory. 1171–84. (1965). Phys.-O. and Morrison. 1474–89. J. 157–71. in ManyBody Atomic Physics.. L¨ owdin.. (1999). Quantum Chem. Correlation problem in many-electron quantum mechanics. Phys. natural spin-orbitals. 12. Scripta 32. Int. Adv. J. Berlin). J. I. 110. Mol. (1998). L¨ owdin. 969–82. J. I. 33–58.-O. (1955). 207–322. (1998). Scripta 32. J. A size-consistent state-speciﬁc multireference coupled cluster theory: formal developments and molecular applications. 291–302. J. Int. Studies in perturbation theory. Solution of eigenvalue problem by projection operator formalism. Rev.510 References Lindgren. and Mukherjee. Math. A coupled-cluster approach to the many-body perturbation theory for open-shell systems.

Size-consistency corrections for conﬁguration interaction calculations. (1946). and Berthier. Young. Manne. K. Rev. 18. J. G. Phys. C. Plesset (1934). and Bartlett. J. 346–53. Quantum Chem.. 9227–35. J. Ph. Durand. Potential curves and dipole moment functions of the OH radical. Meissner. Meissner. Eﬀective and intermediate Hamiltonians obtained by similarity transformation. Møller. (1998). Lett. and Malinowski. L. L. The linked-diagram expansion of the ground state of a many-electron system: a time-independent derivation. C. 46. (1977). Fran¸ cois. Møller. Mat. P. J. General properties of the characteristic matrix in the theory of elementary particles II. and Nooijen. Selsk. (1967).-Fys.-Fys. 9604–14. 204–10. Medd. Phys. L. A Math. Phys. B. (1976). Chem. R. Levy. Møller. Dan. J. II. 1061–70. Intermediate Hamiltonian as a new class of eﬀective Hamiltonians. No. Chem. 172. Selsk. J. Int. Dan. 277–92. Lett. H. Phys. Phys. J.. W. and M. Chem. . Medd. Rev. 19. 102. C. and Daudey. R. S. Phys. 10. 618–22. 23. General properties of the characteristic matrix in the theory of elementary particles I.. 6670–6. N. Int. Note on an approximation treatment for many-electron systems. M. 062510-1–14.References 511 Malrieu. 809–26. (1945). P. (1990). Chim. Meyer. Vidensk. 11. London). Meissner. (1985). H. and Gijbels. W. Quantum Chem. (1988).. S. J. (1974). 146. Meunier. Chem. L. Symp. PNO-CI and CEPA studies of electron correlation effects. A. Intermediate Hamiltonian formulation of the valence-universal coupled-cluster method for atoms. Fock-space coupled-cluster method in the intermediate Hamiltonian formulation: model with singles and doubles. J. M. 94. 1–46. A 61. The N 2 problem in molecular CI calculations. R. Quantum Chem. 1–48. Meissner. Phys. J. Martin. P. Theor. 1.. (1995). J. Acta 35. (2000). K. Gen. 175–92. 108. The ManyBody Problem in Quantum Mechanics (Cambridge University Press. March. Chem. Meissner. (1991). L. 22. L. P. and Sampanthar. Phys. Mat. On size-consistency corrections for limited conﬁguration interaction calculations. Vidensk. No. Transformation of the Hamiltonian in excitation energy calculations: comparison between Fock-space multireference coupled-cluster and equation-of-motion coupledcluster methods.

. Chem. Musial. Diagrammatic structure of the general coupled-cluster equations. Chem.. A. Phys. and Bartlett. 125. Musial. 542–8.. and Bartlett. Calculation of properties with the coupledcluster method. Chem. Symp. Mukherjee. Phys. J. Kucharski. J.. (1977). Musial. S. Lett. R. Phys. Lett. J. and Bartlett. (1991). Chem. Size-extensive eﬀective Hamiltonian formalisms using quasi-Hilbert and quasi-Fock space strategies with incomplete model spaces.512 References Monkhorst. Molecular applications of size-extensive quasi-Hilbert. S. T5 operator in coupled-cluster calculations. D. R. (2005). and Mukherjee. J. J. 177. Lett. (1986).and quasi-Fock-space coupled-cluster formalisms using incomplete model spaces. Phys. J. and Bartlett. J. D. Chem. Critical comparison of various connected quadruple excitation approximations in the coupled-cluster treatment of bond breaking. M. H. D. J. 1867–72. 457. (2002a). and Mukherjee. M. R. 121. M. J. Chem. Musial. Mukhopadhyay Jr. (2008).. Quantum Chem. Lett. 119. Phys. J. R. Phys. 267–70. 100. Phys. Musial. and Bartlett. R. and Bartlett. Chem. Phys. 122. 421–32. M. Phys. S. R. Chem. R. M. Equation-ofmotion coupled cluster method with full inclusion of the connected . Musial. (1989). Phys. A. 4382–8. D. S. 171–7. 2241021–9. (2003). 320. Lett. double. J. Mol. and Bartlett. (2003). The linked-cluster theorem in the open-shell coupledcluster theory for incomplete model spaces. 207–12. M. Chem. D. J. Formulation and implementation of the full coupled-cluster method through pentuple excitations. Int. 1901–8. Fock-space multireference coupled cluster method with full inclusion of connected triples for excitation energies. R. Musial. A. J. and triple excitation method in the intermediate Hamiltonian formulation for electronic excitation energies. 1670–5. Benchmark calculations of the Fockspace coupled cluster single. Kucharski. Musial. M. (2004). 11. Kucharski. A. Mukhopadhyay Jr. 441–6. 116. M. J. Kucharski. (2002b). and Bartlett. (2000). Quantum Chem. Equation-of-motion coupled cluster method with full inclusion of connected triple excitations for electron-attached states: EA-EOM-CCSDT. 163.

J. 100. Phys. Cluster expansion of the wave function. ionized states and electron-attached states of formaldehyde by SAC and SAC-CI theories. and Hirao. Urban. 2952–8. K. Spin adapted restricted open shell coupled cluster theory. and Hirao. Symmetry-adapted-cluster expansion. (1987). Acta 73. 3706–16. Noga. Cluster expansion of the wave function. Bartlett. Phys. Nakatsuji. Chem. H. I. Calculation of electron correlations in ground and excited states by SAC and SAC CI theories. 86. J. On expectation value calculations of oneelectron properties using the coupled cluster wave function. Chem. 291–306. Neogr´ ady. Nakatsuji. Ohta. Noga. Nesbet. K. Cluster expansion of the wave function. 68. Cluster expansion of the wave function. J. and Urban. H. (1978a). R. Noga. Phys. 97. its variational determination. Phys. 67. and Hubaˇ c. Proc.. (1981). Conﬁguration interaction in orbital theories. P. 118. Spin adapted restricted Hartree–Fock reference coupled cluster theory for open shell systems. (London) A 230. (1955). (1992). H. Nesbet. The full CCSDT model for molecular electronic structure. Chem. Lett.References 513 triple excitations for ionized states: IP-EOM-CCSDT. Chem. J. Cluster expansion of the wave function. . Electron correlation in ground and excited states by SAC (symmetry-adaptedcluster) and SAC CI theories. Nakatsuji. J. R. 109. CCSDT-n models. (1978b). Towards a full CCSDT model for electron correlation. Phys. R. valence and Rydberg excited states. Chim. Phys. M. Pseudo-orbital theory based on the SAC expansion and its application to the spin density of open-shell systems. Soc. M. 312–21. K. H. J. H. Nakatsuji. 5074–80. Chem. Chem. and Bartlett. Chem. 2053–65. Electron correlation in ground state. and Hirao. and Hubaˇ c. 4279–91. J. Phys. 75. 126–32. Urban. 3401. 1632–8. Brueckner’s theory and the method of superposition of conﬁgurations. and extension of open-shell orbital theory. Linear version. (1987). Rev. (1994). M. Phys. J. M. K. Chem. J. 134. Phys. R. P. 334–42.. J. (1979b). Neogr´ ady. 89. 68.. Chem. Phys. K. 7041–50. R. Chem. Erratum (1988). Nakatsuji. Phys. (1979a). 329–33. K. Lett. Phys. and Urban. Chem. J. Theor. I. J. (1958). Lett. (1988). J.. 1128–36. 67.

Quantum Chem.. Coupled cluster approach or quadratic conﬁguration interaction? J. ˇ ıˇ Paldus. 6812–30. Pal. J. J. Spin-adapted coupled cluster many-electron theory. D. (1992). D. Chem. (1977). 1–45. Phys. ˇ ıˇ Paldus. and Jeziorski. 9. (1974). Int. and Mukherjee. Sinha. M. The First Forty Years. ˇ ıˇ Paldus. Mol. (1997b). J. J. Phys. Chem. examples. (1975). M. J. 26. and Snijders. Similarity transformed equationof-motion coupled-cluster theory: Details. C´ zek. S. and Bartlett. 60. The beginnings of coupled-cluster theory: an eyewitness account. 55–83. eds. 105–97. J. (1990). Chem. J. 4356–62. G. Chem. 67. Chem. (2005). Paldus. Nooijen. 90. G. Green’s function approach to the direct perturbation calculation of the excitation energies of closed shell fermion systems. B. Frenking. Hartree–Fock stability and symmetry breaking. Refelections on size-extensivity. and Mukherjee.. B. Quantum Chem. Bartlett. Molecular applications of multirefrence coupled-cluster methods using an incomplete model space: direct calculation of excitation energies. 15–48. J. Coupled cluster Green’s function method: working equations and applications. J. J. R. 48. J. Paldus. K. in Theory and Applications of Computational Chemistry. J. (1988). and C´ zek. Head-Gordon and Raghavachari. A new method for excited states: Similarity transformed equation-of-motion coupled-cluster theory. J. R. K. (1989). Quantum Chem. Coupled cluster approach to the single-particle Green’s function. J. 2277–98. Coupled cluster approach or quadratic conﬁguration interaction?: Reply to comment by Pople. Adv. J. . ˇ ıˇ Paldus. Chem. Kim and G. Nooijen. and Bartlett. C´ zek. Shamasundar. Scuseria (Elsevier. and Snijders. eds. R. J.. in SelfConsistent Field: Theory and Applications. Phys. G. Paldus. (1993).. pp. Amsterdam). J. Phys. 4357–66. Phys. Phys. Correlation problems in atomic and molecular systems. J. M. 106. 88. Carb` o and Klobukowski (Elsevier. M. Dykstra.. R. Quantum Chem.. 103. and Jeziorski. 149–63. Phys. 93. 1485–6. Time-independent diagrammatic approach to perturbation theory of fermion systems. J. Symp. 107. J. R. C. M.. D. (2005). 6441–8. J. J. J. size-consistency and generalized extensivity in many-body theory. and comparisons. V. 115–47. Int. 303–18. Nooijen. (1990). Phys. (1997a).514 References Nooijen. J. Nooijen. Chem. pp. M. Rittby. Amsterdam). and C´ zek.

Theoretical models incorporating electron correlation. 1910–8. General-model-space stateuniversal coupled-cluster method: diagrammatic approach. In search of the relationship between multiple solutions characterizing coupled-cluster theories. A. Phys. Chem. 4267–91. Takahashi. Seeger. Phys.. R. (2004). Nooijen and Bartlett. Theory. in Computational Chemistry: reviews of Current Trends.. (1982). A. and Bartlett. S. A full coupled-cluster singles and doubles model: the inclusion of disconnected triples. Paldus. 35. J. Int. Chem. Pulay. Quantum Chem. Quadratic conﬁguration interaction. Quantum Chem. P. Pople. N. Pople.. H. I. Head-Gordon. J. J. IV. Ab initio calculation of force constants and equilibrium geometries in polyatomic molecules. ˇ arsky. W. 90. Symp. 1031–7. J. Int. . J. A. (1996). 1–105. J. Coupled-cluster approach to electron correlation in one dimension: cyclic polyene model in delocalized basis. J. 17. Quantum Chem. Correlation problems in atomic and molecular systems. Li. Extended coupled-pair many-electron theory and application to the BH3 molecule. M. Pople.. and Seeger. G. I. R. (1984). Perera. Chem. A. 143. Paldus. Phys. 91–6. 10. and Bartlett. B 30. J.. Int. K. Lett. and Cho..References 515 ˇ ıˇ Paldus. Singapore). Electron correlation eﬀects on the theoretical calculation of nuclear magnetic resonance spin–spin coupling constants. A. Quantum Chem. (1989). Chem. Head-Gordon. R. (1972). 11. (1977). J. J. K. A. T. and Krishnan. J. J. pp.. 149–63. 87.and 8-reference statespeciﬁc Brillouin–Wigner coupled-cluster method: study of the singlet oxygen. (1988). J. Pople. Phys. Leszczynski (World Scientiﬁc Press. Symp. R. Binkley. A 5. and Kowalski. Rev. Quantum Chem. 76. D. and Shavitt. 90. J. (2002). Quadratic ˇ ıˇ conﬁguration interaction: reply to comment by Paldus.. J. (1969). K. C´ c. X. Polarizability of OH− . 215–51. 5968–75. M. J. (2000). 197– 204. (1987). Phys. Four. Rev. J. A general technique for determining electron correlation energies. J. Phys. M. and Raghavachari. Piecuch. 3290–305. 5. Sadlej. and Hubaˇ Pittner.. Mol. K. Chem. P. Vol. Phys. and Petraco. J. Purvis III. S. (1976). D. R. J. Pluta. Variational conﬁguration interaction methods and comparison with perturbation theory. Chem. R.. C´ zek.. J. 104. 1–19. I. J. R. C´ zek and Jeziorski. ed. Math. 50–67. 4635–6. and Raghavachari. J.. Phys. P.

J... J. and Bartlett. Chem. R.. S. J. . J. (1983). 835–45. 53. Head-Gordon. Rozyczko. (1975). and Handy. J. (1997). G. A ﬁfth-order perturbation comparison of electron correlation theories. Sekino. (1989). Trucks. S. (1987). B. Head-Gordon. Phys. 3033–6. Amsterdam). P. D. Symp. C. Reid. R. Ab initio SCF and CI studies of the ground state of the water molecule.. Phys. Chem. 157. Chem. 23. An open-shell spin-restricted coupled cluster method: application to ionization potentials in N2 .. 1. (1988). Ab initio SCF and CI studies of the ground state of the water molecule. Czech. J. Quantum Chem. Rosenberg. Lett. Chem. Comparison of CGTO and STO basis sets near the Hartree–Fock limit. J. with application to an anharmonic oscillator. Int. Raghavachari. S.. Chem. 4072–80. K. A. G. J. Phys. Perera. Salter. and Bartlett. H. Brown.. J. E. R. J. F.. R. Raghavachari. M. C2v insertion pathway for beryllium dihydride (BeH2 ): a test for the coupled-cluster single and double excitation model. N. and Bartlett. 502–9. Replogle. Comm. 1945. 63. C. 115–21. Nooijen. Phys. and Shavitt. and Pople. E. Theory of Sound. W. Reply to comment on: coupled cluster approach or quadratic conﬁguration interaction?. and Head-Gordon. J. B. Pople. E. Dover. Chem. and Bartlett. 479–83. H. W. R. Int. Chem. J. J. 87. Quantum Chem. I. Phys. Rayleigh. 1 (Macmillan. Raghavachari. second edition. 6736–47. M. New York). Many-Electron Theory (North-Holland. Purvis III. Ermler. Vol. J. (1990).. (1990). Phys. 1486–7. I. 93. Lett.. (1988). J. Rosenberg. (1976). reprinted. Baron (1894).516 References Purvis III. 92. II. M. A. Shepard. D. W. I. Chem. 167. Size-consistent Brueckner theory limited to double and triple substitutions. Quantum Chem. Strutt. Property evaluation and orbital relaxation in coupled cluster methods. J. 2203–13. Pople. Potential energy and property surfaces. J. J. B. A.. Phys. 2162–74. Sekino. Correlated calculations of molecular dynamic polarizabilities. Multiplicity of many-body wavefunctions using unrestricted Hartree–Fock reference functions. Col. A. 65. R. Transformation of perturbation series into continued fractions.. K. (1967).. 107. (1972). 521–34.. C. K. B. Chem. A. M. G. M. Rittby. J. Raimes. Phys. London. and Bartlett. and Shavitt.

Steady states and quasienergies of a quantum-mechanical system in an oscillating ﬁeld. Physik 80. Phys. Phys. coupled-cluster theory for excitation energy. H. G. Phys. (1987). J. Chem. Phys. D. E. 255–68. J. Multiple substitution eﬀects in conﬁguration interaction calculations. (1984). J. I. Sambe. W. F. R. J. J. M.. Chem. Shavitt. R. J. Phys. A linear response. G. (1977). 3022–37. 101. Sekino. E. Lett. 55. Scheiner. Rice. A note on the direct calculation of excitation energies by quasi-degenerate MBPT and coupled-cluster theory. (1986). Siegbahn. C. Mukhopadhyay.References 517 Salter. (1986). H. F. Chem. Quantum Chem. I. H. E. Chem. Quantisierung als Eigenwertproblem. Chem. Phys. . I. H. Rev. Sekino. R. 87. 2203–13. 18.. F. 99. (1978). Sekino. Chem. Quantum Chem. and Mukherjee. A. Sinha. 365–419. Chem. S. 98. 5361–73. J. E. 5711–7. J. R. and Schaefer III. J. Quantum Chem. and Bartlett. D. P. Int. (1994). 90. J.. Molecular hyperpolarizabilities. Eﬀectiveness of conﬁguration interaction calculations for large molecules. Lett. 8840–7. 3945–9. and Redmon. (1999). 386–94. R.. Sandars. Phys. Stanton. J. H. and Bartlett. Ann.. 14. Chem... and Bartlett. Scuseria. Quasidegenerate perturbation theories. A. Symp. 437–90. G. Lee. 371–4. J. 1752–66. (1973). F. (1926). J. (1993). Phys. A 7. and Bartlett. 125–30. H. 129. and Bartlett. J. E. Analytic evaluation of energy gradients for the single and double excitation coupled cluster (CCSD) wave function: theory and application. (1980). Sekino. 149– 73. Chem. J. Quantum Chem. On the extensivity problem in coupled-cluster property evaluation. 11. On the extent of spin contamination in open-shell coupled-cluster wave functions. 73. H. A linked diagram treatment of conﬁguration interaction in open-shell atoms. First derivatives. (1993). Sasaki. J. P. T. T. Nuclear spin–spin coupling constants evaluated using many body methods. Int. J. J. Chem. Many-body methods for excited state potential energy surfaces. 35. Quantum Chem. General theory of energy gradients for the equationof-motion coupled-cluster method. Trucks. 369–74. Phys. Adv. Schr¨ odinger. Stanton. 85. Phys. Analytic energy derivatives in many-body methods. A canonical van Vleck formalism and its relationship to other approaches. L. Phys. (1989). (1969). Symp. Adv.

(1953). Phys. J. A: Math. A coupled-cluster based eﬀective Hamiltonian method for dynamic electric polarizabilities. and Gauss. . 38. R. J. R. Why CCSD(T) works: a diﬀerent perspective. Symp. Lauderdale. Phys. I. F. (1963). J. The equation of motion coupledcluster method.. J. Watts. Rev. General formalism. F. and Bartlett. 99. R. Analytic second derivatives in highorder many-body perturbation and coupled-cluster theories: computational considerations and applications. Perturbed Hartree–Fock calculations. (1993a). W. R. Swain. Quantum Chem. 111. 98. J. and Bartlett. 7029–39. Gauss. Phys. Stanton. R. 5554–9. J. and Gauss. Chem. Z. A systematic biorthogonal approach to molecular excitation energies.. Many-body methods for excited state potential energy surfaces. and Bartlett. J. Phys. Phys. and Bartlett. S. 8931–43. J. Chem. N. J. Gen. 19. F. 550–60. J.518 References Stanton. Stanton. D. On the choice of orbitals for symmetry breaking problems with application to NO3 . and Gauss. Stanton. 879–94. 5178–83. Phys. J.. Phys. Int. (1992). J. Chem. J. J. Int. (1995). M. J. J. H. Sternheimer. Chem. II. Phys. J. Stanton. Chem. (1977). 8938–44. (1985). (1992). 130–4. Stanton. J. Continued fraction solutions in degenerate perturbation theory. 155–65. and Gauss. Stanton. Analytic energy derivatives for ionized states described by the equation-of-motion coupled cluster method. Lett. M. M. H.. Stanton. Phys. Coupled-cluster method in Fock space. Analytic second derivatives for excited state energies in the equation-of-motion coupled cluster method. Chem. J. R. (2000). Gauss. J. 10. F. M. J. Nuclear quadrupole coupling in the Li2 molecule. 8785–8. and excited state properties. Pitzer. A 32. transition probabilities. F. and Lipscomb. W. Rev. J. 97. J. Phys. A simple scheme for the direct calculation of ionization potentials with coupled-cluster theory that exploits established excitation energy methods. 725–42. J. J. J. J. Chem. Quantum Chem. 103. Phys. (1994).. F. (1999). 61–95. 26. Magnetic susceptibility and shielding in the LiH molecule. and Foley. Phys. 281. F. I. J. J. 92. 1460–8. Chem. Stolarczyk. F. Stevens. R. and Monkhorst. (1997). L. Stanton. Rev. The ACES II program system. J. J. (1993b).. F. 101. Chem.

Modern Quantum Chemistry: introduction to Advanced Electronic Structure Theory (Free Press. P. and Bartlett. 103. S. . J. D. 119. Phys. J. R. Critical comparison of single-reference and multireference coupled-cluster methods: geometry. J.. T. Phys. 101. 3550–7. N. 225–32. J. 85–106. Towards a full CCSDT model for electron correlation. Phys. Szalay. P. G. and Bartlett. and excitation energies of N2 O2 . J. Singapore). G. J. in Recent Advances in CoupledCluster Methods. Rev. R. Lett. (1985). New York). Szalay. J. 281–98. P. J.. 4936–44. J. and Gauss. A. R. Van Vleck. (2004). J.References 519 Szabo.. and Bartlett. C. G. Noga. Symp. Comparison for the two-electron problem and the role of the exclusion principle violating (EPV) terms. J. 83. 548–54. Szalay. 145. Trucks. J. Spin-restricted open-shell coupledcluster theory for excited states. J. and Crawford. Chem. G. (2000). Phys. M. 121. Neogr´ ady.. 4027–36. Phys. Chem. and Bartlett. M. Nooijen. Bartlett (World Scientiﬁc. 275–306. ed. 85. On σ -type doubling and electron spin in the spectra of diatomic molecules. Convergence of the coupled-cluster singles. Musial. Phys. J. 4041–6. Chem. 112. J. G. Bartlett. Chem. P.. (1995). Chem. Phys. Chem. G. and Lee. Quantum Chem. Alternative ans¨ atze in coupledcluster theory. 10713–23. Thouless. Chem.. Cole. R. (2003). Stability conditions and nuclear rotations in the Hartree–Fock theory. harmonic frequencies. (1994). 26. Erratum (1986). Perera. Quantum Chem. M. Nooijen. and Bartlett. Phys. Analytical energy gradients for the two-determinant coupled cluster method with application to singlet excited states of butadiene and ozone. J. (1929). J.. pp. (1988). Nucl. D. and Gauss. J. 9028–38.. H. M. Phys. M. Chem. I. 5383. (1997). S. W. A critical analysis of diﬀerent methods. Noga. and Ostlund. (1997). Tobita. R. Phys. Tam. Urban.. A. M. P. Spin adaptations in the open shell restricted coupled cluster theory with a single determinant restricted Hartree–Fock reference. IV. 21. J. J.. J. J. P. J. Int. doubles and triples method. and Hubaˇ c. S. N.. S. J. Urban. M. 107. Szalay. Russ. R. Alternative ans¨ atze in single reference coupled-cluster theory. (1992). Phys. III. J. (1982). Chem. (1960). Coupled cluster calculations of optical rotatory dispersion of (S )-methyloxirane. Spin-restricted open-shell coupledcluster theory. 467–506. R. Szalay. 33.

413–20. P and Monkhorst. Lett. R. Quantum Chem. (1979).. (1990). J. Chem. Chem. 1007–22. Phys. J. Watts. Phys. 6104–5. T. (1993).. T. and Bartlett. ˇ Zivkowiˇ c. J. D. CCSD+T(CCSD). Accurate electrical and spectroscopic properties of X 1 Σ+ BeO from coupled-cluster methods. J. 195–203. (1950). Open-shell analytical energy gradients for triple excitation many-body. Anz. D. (1935). and Bartlett. 11. and Head-Gordon. J. Phys. J. Existence and reality of solutions of the coupledcluster equations. Chem. 90.. Erratum (1980). Watts. C. Phys. 53. and Bartlett. D. Urban. T. Quantum Chem. coupled-cluster methods: MBPT(4). Coupled-cluster singles. L657–60. J. 268–72. 1505. 98. J. 113. R. Wick. Phys. ˇ Zivkowiˇ c. Gauss. Phys. double. Naturwiss. 341–55. J. (1995). G. 200. Symp. 80. R. J. On a modiﬁcation of the Rayleigh–Schr¨ odinger perturbation theory. 8718–33. Rev. Diagrammatic perturbation theory: the contribution of triply excited states to correlation energies.. CCSD(T). Math. Int. and QCISD(T). 28. Wigner. Chem. Watts. Accts. Akad. J. (2000). J. M.520 References Van Voorhis. Quantum Chem. 8873–9. . S. Phys. M. H. J. Symp. R. R. (1977). Watts. and triple excitation model for open-shell single-reference functions. Gauss. (1992). Analytical connection between conﬁguration-interaction and coupled-cluster solutions. J. and triples calculations with Hartree–Fock and Brueckner orbital reference determinants: a comparative study. J. 477–82. D. D. J. Math. Theor. Chem. J. (1994). Coupled-cluster methods with noniterative triple excitations for restricted open-shell Hartree– Fock and other general single determinant reference functions. and Bartlett. Energies and analytical gradients. 93. Watts. B 13. and Bartlett. Wiss. Ungar. E. Int. P. The coupled-cluster single. J. J. (1978). Wilson.. 19. The evaluation of the collision matrix. Benchmark variational coupled-cluster doubles results. Quantum Chem. 1–7. doubles. B 12. J. J. Phys.

504 Baker. 346. 301. E. 293.. 365. 400. 461. 365. 324–328. 496.. 41. H. 464. 293. 499 Brandow.. 396. 80. 496 Christiansen. 134. 519 Comeau.. 336. 499 Bruna. 501 Datta. 162.. C. 186. 497 Bandyopadhyay. M.. 426. 504 Ermler.. 495. 499 Bomble. J. 491. 500 ˇ arsky. 53. R. R. 502 Emrich. 499 Bernholdt. J.. 80. 327. 348. L. 142. 1. M. 419. 410–414. 187. R. 367. A. A. R... 437. W. 12. 80. S. 501–504. L. 435. T. 501 Cole. 420. J. J. F.. 202. 318–321. 402. 156. 320. 500 Cho. A... 316. 496 Balkov´ a. 504 Campbell. 387. 414. E. 501. 175. 499 Bibireata. A. 420... 431. 159. 474. 12. 496 Coester.. W. 510 Frantz. P. S. 499. 508 de Mer´ as. 164. 315. 421. 501 Crawford. 500. 497.. 503 Feynman. 496 Berthier. Ph. 400. 121.. 352. 162. 429. D. 361.. 344. J.. K. 474.. 496 Bloch. 421–423.. 471. 154. C. 477.Author index Adamowicz. E. 33. 402. 496 Auer. H. 327. 480. 153.. P.. 162. 365. 346. 342–344. 502 Freeman.. T. 160. M. D. 514 Choppella. F. 500 Byers-Brown. 91. 400. P. 367. 493–498. X. 407. S. 400. 402. 488... J. 161. 496 Ahlrichs. 426–429.. 444.-P.. 263. T. 165. P. 163..-D. 429.. 510 Daudey. 343. 460. 41.-L. 165. 511 Bethe. 162. 351. H.. 427. 431. 501. 506 Dirac. 162. D. 429. P. 162. 400. 499 Br¨ andas. 491. 417. 129. 454. 514 C´ Chan. E. C. 507. 154. F. J. 342.. 502 Foley.. 346. A. 504 Bartlett. H. 500 Chattopadhyay. P. 431... H. 496 Gauss... 1. R.. L. 515 Bishop.. 516 Evangelista. 499 Brillouin. 501. 450. K. 343. B. 407. 474. 474. 518 Dalgaard.. 474. 185. 365. B. A. C. 490.. 514 Cociorva. 510–519 Baumgartner. 501 Dalgarno. 417. L. A. 344. 154. 1. J. 500 Chiles. H. 508. 318. 346. 11. 521 . G... 501 Durand. 1. 263. 2. 502 Epstein. 496 Adams.. R. 410. 456. 498. 185.. 365. 346. R. B. D. J. 503 Gao.. 1. 461. 412. 319.. 346. 499 Brown. D. P.... 500 Buenker. O. 163. 496 Binkley. 352. 474. 506 ˇ ıˇ C´ zek. 344. 350. 184. R. W. D. 485. J. 227. 167. G. 400.. 502 Fan. S. 387. V. S. 429. 396.. R. G.. 2. 500. 496. 452–454. 52. 504. 496 Bauschlicher Jr. 364. W. 410.. J. 411. A. 456. 153. 441–448. 172. 164. 500 Burton. 346. D. 163. K. 477.... 365. 316.. 263. 518 Fran¸ cois. J. 182. G.. 25.. 301. 501. 501 Ditchﬁeld. C. W. 1. J.. 162–164.. R. 410. 387. F... 365. 501 Dunning Jr. G. 515 Brueckner. 472. E.. 500 Dyson. 453. 346. 431. 496 Allen. F. P. 402. 198. 453. 502 Epstein. A. 472–474. 153. B. 321. E. 490.. S. 510 Davidson. 500 Chaudhuri. S. 315. F.. 436. 167. 502 Arponen. 497. 106. A. 497. 510 Dykstra. Y.. J.

293..-C. 495.. 509 Mukherjee.. 496 Mahapatra. 419. S. 501 Hirao. 429.. F. 502. L. 400. 498. 263. 448. 501 Kutzelnigg. R. 508 Laidig. 513 Noga.. 2.. 402. 508 Kohn. D.. 506 Koch.. 429. 321... 505 Jeziorski... 1. J. Y. 431.. 167. 505 Gwaltney. 172.. 437. 510 Lu.. 490. 508. 456. H. 437. H. 510 Meissner. C. 187. 367.. 504. 187. 342. 167. 497 Hubaˇ c. 512 Hirata. 432. 497 Jana. C. 475. 129. 502... 343. A. D. 519 Nesbet. S. 8. 500. N.-O. 414. 492. 315. P. 301. 456.. 480. F. R. D. 508 Lauderdale. 430. 510 Manne. 490.. 511.. 431. 466. 505. D. 1. 342. 474. 519 Helgaker. N. 496 London. 499.. 503. 393. 444. 465. J. 505.. 324–328. 508. J.. 162. 509. 11. 346. 485.. 508 Handy. 478.. R. 508 Larsen. 351. 505... 411. 500. 444. 519 Hubbard. 511 Meyer. J.. 421. 518.. 365. Q. 336. 315.. 499. 507. 419. 490. 508.. 430. 414. 165. 316. 419. 446–449. 25.. H. 361. 361. 1.. 390. H. 352. 498. 506–508 Kvasniˇ cka. 344. 502. 511 Mukhopadhyay. 506. P. 510.. 414. 502 Gijbels. 41. 52. W. 504 Huby. 506 Karplus. S. J. 2. 52. K. 456.. N.. 351.. S. J. 167. I. 509. 162. 444. 411. 504 Jankowski. 505. 505 Jayatilaka. G. 517 Mukhopadhyay Jr... 346. 508. 429–431. 152. T. K. O. H. 448. 478. 445. V. 186. 161. R.. 319. M. 515 Jordan. 497. A. H. 499. 419. 426–428. 175. 514 Krishnamoorthy. L. K. 444.. C. 342–344. 500. 518 Nakatsuji. O..522 Author index Krogh. 480. 515. 343. 175. 175. 508 Langhoﬀ. 498... 496 Langhoﬀ.. A. E. 503. 503 Head-Gordon. 503 Hald. 495. 510. 227. 421.. 508 Kaldor. 503 Goscinski. L. 1. J. M. 475.. 328. 498 Kobayashi.. S.. H. 497. J. 315. 80. 402.. 471. 505 Knowles. Krishnan. 503.. 504 Hsieh. 510 Malinowski. 301. U. 321. 8. 510 Martin. 505 Jørgensen. 352. 53.. J. W. 505. S. 255. 513. 421. C. 502 Møller. 506. 490. 162. 185. 493–495. M... 471. 490. 419. J. 511 Malrieu. 1. 316. 503 Harrison. 413. 418.. P. M. P..-L. S. 519 Nooijen. 467. 518 Lee. P.. 361. 504 Horowitz. 453. 503. 512 K¨ ummel. R. 320. 503. 513. 515 Haque. 346. 154. C. 510 Goldstone. J. 485. 517 Musial. 504.. 11.. W. 346. 503 Grabowski. 162. 429. 410. 501. 511 Li. I. 429. R. 462. 346. 351. 472–474.. 129. S. 461.. 469. 508. 346. 318.. 502 Mills. 491. 514. 165. 162. 472–474. 400. 344. P. 504 Hugenholtz. 511 Mills. 227. 255. 472. 500. M.. 497.. 444. 511. 520 Morrison. M.. 513. 497. 507 Kucharski.. 346. 497. 227. P. 453. 503 Harris. 496. 1. T. 163. R. W. 509 Lipscomb. F.-P. 496. R... 506. 198. 2.. 162.. 1. A. 511 Monkhorst. 315.. 512. 413. 516 Lee.. see Raghavachari. 498. 346. 514. 1. X. 511. 501. B. I.. 499 Hose. N.. 186. 518 Lee.. 429. 509 L¨ owdin. 432. 316. I.. 511 Meunier. 417. S. 461.. 342. 80. 507. 418. 465. 503. 441–444. 344. C. 444. 506 Klein. 400. J. 429. 518 Lischka. 327. 500. 471. 352. S. 52. 500 Hausdorﬀ. 336. K. M. 514. 346. 482. 367. A. W. L. 362.. 445. 505.. 445. 444. 153. 508 Levinson. R. 419. 504. 517–519 Gerratt. 512 Neogr´ ady. 441. 504 Hylleraas. A. 506.. H. 365. D. 202. 499. J. P. 429. 504 Ivanov. 514 Lindgren. 69. W... 441. 428. 510 March. 481. 496 Lam. 315. 461. 421. 485. 167. 118. 54. M. 346.. 172. 227. 506 Janssen. 327. 500. 481. 503. 163. J. 54. 429. 480. 513.. 504 Hirschfelder.. 444. 506 Heully. 488.. J.. 352. 500. J. 407.. W. 448. 390. 367.. K. 390. 327. 506 Kowalski.. E. 496 Krishnan. 410. 52. 456. 453. W. 421. I.. 393.. C. 173. 365. 505 K´ allay. 164. 342. 496 H¨ attig. O. 387. U. 507 Krylov. 163. 163. 506 Koch. 499 Levy. 477.. H. 410. 227. B. 496 . 488. K. 429. S.. 430. 478. 53. 474. M. 342. R. 162. 431. 508 Kelly.. S.

T. S. 129. 327. 164. R. 414. R. 516 Sampanthar. T. 162. 506–508 Ostlund. 461.. 460. 477. 499 Salter. 516 Sambe. 460. 367. 134.. 162.. 367. 474. 520 . J. 496 Rayleigh. 255.. 37. 164. 192. 448. 2. 500. J. N. 365. 159. 198. 365. 344. 364.. 514.. 53. 508. 154.. P. 410. 508 Replogle. 263. 12.. H. 365. M. 346. H. 517 Reid. 514. 514 Tam. 518 Taylor.. 499. 436. 342. 162. 365. A. 400. P.. 387. 508. 514 Peyerimhoﬀ. 453. 365.. M. 516. P. E. G.. 515 Sibiryakov. 80. 498.. 402. 69. 501. 454.. H. 319. 516 Scuseria. 519 Wigner. 515 Redmon. 365. 417. 318.. 46. 505 Sinha. 2. C. 497. 514 Pople. 423. G. T. 454. 346. 393. S. 444. 162. 160. 504. 410. 514.. 453. J. 516 Scheiner.. 496 Stanton. D. 362. 477. 437. 412.. R. 467. 518 Ruud. 346. 365. 153. 466. 517 Silver. 490. 414. 471. J. 365. 519. 53. K.. 516 523 Seeger. M. 500. 429. 68. E.. 509. 425.. 515. 315. 400. K. 153.. 301. 518 Szabo. 518 Tobita. 163. 316. 498. 519 Wick. 319. 517 Snijders. 502.. 24. 319.. S. 343... J. 471. M. 518 Plesset.. 365.. 12. P. P. T. D.. 495. 518 Sternheimer. 502.. 510 ˇ Zivkowiˇ c.... 263.. 80. E. A. C. D. 503. R. 498.. J. 413. 426. 41.. S. 448. G. 497. 499 Thouless. 515 Ramanujam. 518 Takahashi. M. 361. 516. R. 507 Young.. 422... 515 Rendell. 503. 502 Sadayappan. 441. 429. 185. 515 Raimes. 505. 69. G. 519 Urban. 516 Rosenberg. S. 430. 497. Baron. 431. 300. 80. 502. 414. 518 Trucks. 503. 506 Swain. 517 Shepard. 410. K. 461. 263. 421.. 410. 511. 518 Stevens. 163. L.. 504. 12. 423. 1. 517.. 421. 510 Sandars.. J.. P. 153. 515 Perera. Z. 46. 514 Pittner. 519 Wloch. M. D. W. 501 Simons. 400. D. 417.. 516 Russ. 519 Van Voorhis. 351. J. 513. 402. 351. 516 Schr¨ odinger.. 519 Watts. J. G. 3. P. 351. 506 Shamasundar. 46. M. 396. 501. 421. 162. 445.. 518 Szalay. 453. 320. 400. 461. C. 478.. A.. 516 Rozyczko.. D. 25. 505. 3. 316. 501 Stolarczyk.Author index 441–448. 513. P. 441.. 480. 461. 33. 162–164. 516. 500 Piecuch. 499.. H.. 496.. 513. 514. 496. 410. 450. 129. 498. 195. 422. N. 516 Rittby. 1. 419. C. 315.. 346. 469. 516. J.. 519 Wilson.. 503. 387. A. E. M. K. 472. 513 Shavitt. 456. G. 431. 444. 184. W. S. V. 496 Siegbahn. 315.. R. 430. 1. 186. A. 444.. 505. M. 496 Sadlej. 429. 513 Paldus. 160... D. 441. S. J. 422. E. L. G. M. 511 Pluta.. 514. F. 515 Sekino. 513. 441. 520 Van Vleck. 429. 518 Pal... A. 301. E. 515 Purvis III. 390.. S.. B. 80. 257. 129. 162. 154. 342. 154. 430.. 327. 432. 156. 490. 421. K. 364... 515 Pulay. 503. 454. 518 Ohta. 515 Raghavachari. 12. 185. 301. 498 Silver. 316. P. A.. 512 Olsen. 37. 441. 498. 515–517 Sham.. 301. 427. 518 Surj´ an. P. S. 106. 423.. 460. 516 Schaefer III. 163. J. 515 Rice. 321. 2. I. 346. M. E. F. E. P. 503. J. 497. 11. 474. P. 498.. 500. 445. 53. 413. W. R.. 516 Sasaki. 156. S.. L. J. 454. M. 346. 518.. 346. 474. 352. 11. D. 421. 518 Stewart. 514. 460. 501. N. 365. R.. 448. L. 402. 361.. 474. 513 Staemmler. 400. 352. 390. H. 429. 421.. F. 327. 514 Pitzer. 367. 477.. 508. A. 518 Petraco. E. 514 Salpeter. 319. J. J. 474. 182. H. 496. 301. 91. 316. B. A.

299–308. 393. 200 with folded resolvent lines. 419 Brueckner-orbitals CC (B-CC). 489 active particles. third-order canonical HF case. 387 complex. 300–301 role of T role of the A and B matrices. see Brueckner-orbitals CC Baker–Campbell–Hausdorﬀ expansion. 476. 9. 4 ASG. 489 anticommutation relations. 142 double-excitation diagrams. 447–448. 132. 400–404. 138 antisymmetrizer. 342 algebraic derivation. 33–34 non-extensivity. 189–191 normal-product form. 301 linearized. 224 QDPT wave operator second-order. 420–421 conditions. 425–426 complex case. 237 inapplicable to ROHF. 404–405 Bloch equation. 300 Brillouin–Wigner perturbation theory (BWPT). 279–291 linearized. 163. 208. 418–421 advantages. 462 bonnes fonctions. 144 QDPT level-shift operator second-order. 463–465 alternative forms. 402 derivative integrals. 425 ˆ1 equations. 476. 258–271. 186. 392. 186 generalized.Subject index A and B matrices. 421 broken-symmetry cases. 413. 52. 491 B-CC. 137 non-HF case. 429. 132. see insertion Brillouin theorem. 419 iterative process. 142 triple-excitation diagrams. fourth-order single-excitation diagrams. 133 non-HF case. 483–488 matrix form. 187–188 in Fock-space MRCC. 293 basis set complete. 221 folded. 402–404 annihilation operator. 421 BWPT. 34–36 Brueckner orbitals. 121–123 energy. 466–467 CC. 189 schematic representation. 163. 465 derivation. 206 third-order. see antisymmetrized Goldstone diagrams atomic units. 394–399. 187. 57–59 antisymmetrized Goldstone diagrams (ASGs). 427. second-order canonical HF case. 425 computational cost. 143 quadruple-excitation diagrams. 72 active holes. see coupled cluster CCD (CC doubles). 198–203 secondary term. 263–271 amplitude equations. 185. 290 conﬁguration-space derivation. 263 CCSD (CC singles and doubles). 194–195 order-by-order expansion. 472 bracket insertion. 64 avoided crossings. 392. 343. 52. 464. 202 bond breaking. 258–263 diagrammatic derivation. 387. 342. 10. 2. 209 wave function. 441. see Brillouin–Wigner perturbation theory C-conditions. 390. 308 equations. 464. see derivative integrals perturbed. 419–420 stationarity condition. 55–67 pseudo-. 375 ˆ1 . 133 energy. 304 T 524 . 288–289 iterative solution.

342. 10 CISD (CI singles and doubles). 9–10 extensivity correction dressing. 471 cluster operators. 343 CCSD(T). 285 in state-universal MRCC. 381 CCSDT(Qf ). 427 ˆ3 . 447. 159–164 H2 O in SU-CC. 321–328. 495 N2 potential-energy curves. 444 FH and H2 O by CI. 13. 342. 343 CCSD[T]. 342 CIS (CI singles). 495 C2 ionization potentials by IP-EOM-CC. 448 CH+ excitation energies by EE-EOM-CC. 345 Ne excitation energies by IH-CC and EOM-CC. 135. 342 quadratic. 310 ASG diagrams for T ˆ3 equations. 472 conjugate diagrams. 185 contractions. 307 T ˆ1 contributions. complete commutator simpliﬁcation. 495 advantages. 342. 316 iterative procedure. 465 ˆn contributions. 343 CCSDT[Q]. 446. 159. 427–428 ˆ2 equations T ˆ4 contributions. 495 H2 O in 39-STO basis. 163 CISDT. 340–346. 142. 442 CCSDT (CC singles. 342 H2 O excitation energies by IH-CC and EOM-CC. 495 using full-CI Hamiltonian matrix. see conﬁguration interaction CID (CI doubles). 342 computational cost. 317 CCSDT-2. 163. 316. 321 CCSDT-n. 309 ΛCCSD(T). see model space. 8. 159. 159. 9. 342–344 CCSDTQPH. 310 T ˆ3 contributions. 443 CN electron aﬃnity by GMBPT and CCSD. 416 connectivity conditions. 323 CCSDTQP. 342. 153–156. comparison of included terms. 161 quadratic CISD. 488 . 322 ASG diagram for T ˆ4 equations T ASG diagram. limitations. 317–319 CCSDT-1. 305 ASG diagrams for T ˆ1 generation of ASG diagrams for T contributions. 293 vacuum expectation value. 325 computational cost. 343. 446. 74–75 core ionization. 314 CCSDTQ. 301 relationship to RSPT. 311 T ˆ3 contributions. 426. 313 T ASG diagrams. 320 computational cost. 321 computational considerations. 159. 273–275. relative importance of T 308–309 CMS. 343–344 N2 excitation energies by IH-CC and EOM-CC. 310 ASG diagram for T ˆ2 equations. 162 full CI. 342 525 CISDTQ. 325 CCSDTQ-2 approximation. 490 MRCISD. doubles and triples). 344 CI. 306 CCS (CC singles). 303 ˆ2 equations. 447 CH+ electron aﬃnities by EA-EOM-CC.Subject index ASG diagrams. 444 comparison with full CI. 302 generation of ASG diagrams. 319–320. 417 comparison with experiment. 12. 308–309 role of T ˆ1 equations. 255–258 amplitudes. 163. 472 non-extensivity. 315–318 computational cost. 312 perturbation-theory order of terms. 321–322 factorized quadruples (Qf ) approximations. 340–346. 322 ASG diagram for T ˆ3 equations. see C-conditions continued fractions. 441–442 CH2 excitation energies by EE-EOM-CC. 342 CCSDTQ-1. 315–321. 430 multireference. 101 computational results C2 excitation energies by IH-CC and EOM-CC. 444 conﬁguration interaction (CI). 342 computational cost. 322–325. 490 quadruples. 159. 92–95 relative to Fermi vacuum. 442–444. 255–258 diagrammatic representation. 342 CISDTQP. 325–328 ΛCCSDT(Qf ) approximation. 322 T ˆ4 contributions. 159–164 conﬁguration state functions. 163. 448. 343. 301 with Davidson correction. 308–321. 316–317 equations. 320–321 computational cost. 321 CCSDT-3. 471 N2 dissociation with RHF and UHF references. 134. 325 equations. MBPT and CC. 10. 68–69 diagrammatic representation.

328 EOM-CC. 294 diagrams and intermediates. 404–405 destruction operator. 461 direct CI. see equation-of-motion CC. 187. 292–295 nature of solutions. 340. 362. 2 Λ-based noniterative approximations. 193. 352–361 diagrammatic representation. 394 one-body. 431 coupled-perturbed Hartree–Fock (CPHF). 8. 300–301 wave function. 396–405 imaginary component. 468 ˆ eﬀ ). 401 equations. 302 complex case. 188. 366–367 as basis for noniterative approximations. 7–9. 7 Coulomb operator. 401 energy derivative. 362. 396 avoiding transformation. 254–255 foundations. 411–414 coupled-cluster linear response. 272–279 factorization of disconnected diagrams. 395. 410 relationship to MBPT. 65. 90–129 CC. 85. 401. 72 . 182–184 core states. 8. see equation-of-motion CC. 275–279 ˆn contributions. double electron attachment degenerate perturbation theory. 426–429 stationarity condition. 352 diagrams for ph terms. CC energy functional EA-EOM-CC. 357 diagrams for pp terms. 429–430 number of solutions. 131. 292 connected form. 429 connected form. 358 MBPT diagrams. see coupled cluster. 258–328 characteristics. see resolvent density matrix. 293–294 CC energy functional (E ). 402 CPHF. 195–198 reduced density matrix. 472 disconnected clusters. 413 derivative integrals. 296 systematic generation. 429 systematic derivation. 355–361 DIP-EOM-CC. see coupled-perturbed Hartree–Fock creation operator. see equation-of-motion CC. 182–184 spin-free. 355–361 energy weighted. 477 electron aﬃnities. 254 complex solutions. 391. 111 antisymmetric. excitation energies eﬀective Hamiltonian ˆ1 (H ). 161–163 DEA-EOM-CC. 295 rules of interpretation. 16 dynamic. 177–180 diagrammatic methods. 136. 192. 418 incorporating T of CC (H). 254–258 with arbitrary reference function. 450 dipole operator. 430 standard solution. 392. 251–430 CC eﬀective Hamiltonian (H). 430 relationship of solutions to full CI. 153 derivative. 450 dipole strength. 391. 437 Λ operator. 7. 65. 426–429 orthogonally spin-adapted. 396–405 imaginary component. 386. 436 two-body. electron attachment electron correlation. 426 response. 8. 90–164 history. 436. 465 in valence-universal MRCC. 401 Coulomb integral. 418–419 iterative process. 186 denominator. 1 diagrammatic notation. 462 nondynamic. 462 electron-repulsion integral. 387–393 complex case. of QDPT and MRCC (H 194. 15 disconnected diagrams. 353–355 density-functional theory. 63. 328–340 simpliﬁcation. 51. 411 diagrams general form. 258 ˆ1 (H ). 430 history. 366–367 computational considerations. double ionization potentials dipole moment. 251–291 full CC. 419 energy. 55–67 pseudo-. eﬀective Hamiltonian incorporating T 418–419 diagrammatic representation. see response density matrix response treatment. 429. 228 correlation energy. 111. 436 dipole polarizability. 273–275 CC eﬀective Hamilton. relative importance of T 308–309 role of single excitations. 292–328 exponential Ansatz and extensivity. see annihilation operator diagram summation techniques. 292 for model state α (Hα ). 76 coupled cluster. 360 relaxed. 368 QDPT. 137 E .526 Subject index Davidson correction. 381–385 transition. 297–299 doubles model. see equation-of-motion CC. see equation-of-motion CC. electron attachment EE-EOM-CC.

72 Fermi vacuum. 444 EE-EOM-CCSDT. 227 in QDPT. 433. 76 excitation energies. 443 inadequacy for doubly excited states. 7. 445 IP-EOM-CCS. excitation energies exclusion-principle-violating (EPV) terms. see equation-of-motion CC Epstein–Nesbet partitioning. 443 EE-EOM-CCSDTQ. 443 EE-EOM-CCSDT-3. 156 disjoint. 347–352 one-body operator. 442. 448 target state. 445–449 diagrammatic equations for EA-EOM-CCSDT. 443 EE-EOM-CCSDTQP. 448 EE-EOM-CCS. 386–387 Fock space. see gauge-including atomic orbitals GMBPT. 432 ionization potentials. 254–255 lack of in truncated CI. 402. 445. 444 VU-CC analogy. 11–15. 215–220 folded resolvent lines. full CI Fermi level. 445 electron-attachment operator (R energy derivatives. 196 double electron attachment. see equation-of-motion CC. 350 expectation value CC wave function. 436 frequency-dependent properties. 155. 156–159 extrapolation techniques. 447 similarity-transformed. 437–444 algebraic EE-EOM-CCSDT equations. generalized Goldstone diagrams. see coupled cluster. 201. 435 computational considerations. 432 unbalanced truncation. 155. 475–495 intermediate Hamiltonian. 119 antisymmetrized. 437–440 left-eigenfunction equation. see Fock-space MRCC gauge-including atomic orbitals (GIAOs). 112. 80 EPV terms. 151. 433 excitation operator (R right eigenfunctions of H. 446–448 double ionization. 476 Fock-space MRCC. see valence-universal MRCC folded diagrams. 442. 15–17 exponential Ansatz. 448 EA-EOM-CCSDT-3. 431–461 biorthogonality relationship. 443 EE-EOM-CCSD. 440 ˆ k ). 445 diagrammatic equations for IP-EOM-CCSDT. 4. 475 sectors. 447 IP-EOM-CCSDTQP. 45–46 unlinked diagrams. 448 EA-EOM-CCSD.Subject index EOM-CC. 12 N He atoms. 456–458 FS-CC. 445. 433 excited-state properties. 442. 202 Fourier series. 7. 353–355 extensivity. 442. 151. 463. 165–172 factorized-denominator diagrams. 439 diagrammatic equations for EE-EOM-CCSDT. 442. 447 IP-EOM-CCSDTQ. 443 electron attachment. 449–454 initial state. 488–490 equivalent diagrams. 454–461 higher-order properties. 446 EA-EOM-CCS. 200. 434 CI-like character. 434 diagrammatic notation. 349. 180–182 factorization theorem. 185 527 factorization of quadruple-excitation diagrams. 437–440 double arrows. 192–194 Fock operator. 34–36. 475 in Hose–Kaldor approach. 115 exchange integral. see coupled cluster. 7 exchange operator. 12–15. 444 EOM-CC Ansatz. 445. 76 normal-product form. 447 IP-EOM-CCSDT-3. 445 H. 448 EA-EOM-CCSDT. 156–157 disconnected clusters. 416 FCC. 432–436 EOM-CC equations. 445–449 as limits of Rydberg series. full CC FCI. 446 ionization operator. see exclusion-principle violating terms equation-of-motion CC. 403 GIAO. 434 de-excitation operator (L left eigenfunctions of H. 435 excitation energies. 72 in Fock-space MRCC. see antisymmetrized Goldstone diagrams Green’s function methods. 446 ˆ k ). 447 ˆ ). 202. 280 perturbed. 156 in expectation values. CC eﬀective Hamiltonian H . see coupled cluster. 54. see conﬁguration interaction. see intermediate-Hamiltonian MRCC valence universal. 150–156 conjoint. 447 IP-EOM-CCSDT. 85. 352–354 two-body operator. 210–211 denominator. 158. 111–113. eﬀective Hamiltonian ˆ1 incorporating T . 447 IP-EOM-CCSD. see many-body perturbation theory.

491 diagrammatic representation. 125–128 CC. 495 function-space partitioning. 374 disconnected diagrams. 140 non-HF case. 447. 243–245 unlinked Hugenholtz skeletons. 54 Hilbert-space MRCC. 238 representative classes. 378 diagrams for Λ1 equations. 5. 139 conversion to ASG skeletons. 480 intruder states. 120. 495 diagrammatic representation. see Hose–Kaldor approach hole lines. 228–230 wave operator. 450 IH. 5. 223 Hugenholtz skeletons. 491 projectors on subspaces. see model space. 477. 377 ˆ-free form. 239 model states. 316. 242–250 linked ASG skeletons. 431 high-spin case. 83–85. 490–491 ionization potentials. 139. 386. 10. 6 noncanonical. 142. 3–7 insertion. see equation-of-motion CC. second-order ASG skeletons. 406. 18 normal-product form. 18. 235 Fermi-vacuum states. 421–425. 4. 238–242 fourth-order canonical HF case. see Hartree–Fock Hilbert space. 5. 376–379 correspondence table. 183. see eﬀective Hamiltonian. 132. 193. 393. 380 T correspondence table. 494 excitation energies comparison of results. 235–236 classiﬁcation of one-electron states. 343. 465. 232. 490. 177 Λ equations. 236–250 contribution status of diagrams. 494 augmented model space. 417–418. 172 interaction line. 465–474 state-speciﬁc. 410. 366. 186. 473 Kohn–Sham reference state. 280–281 partitioning. derivative Hartree–Fock (HF) theory. second-order ASG skeletons for principal term. 472. 72 Hose–Kaldor approach. 494 intermediate space. 9. 406. 371–373 disconnected term. 491 main model space. 197 hole states. 394. 411 restricted. 128. 421–425. 370 diagrams for Λ2 equations. 375 diagrams for Λ1 equations. 448 ladder diagrams. 257.528 Subject index second-order. 133. 362. 227 level-shift operator. 367. 285 fourth-order energy canonical HF case. 413–414. 372 diagrams for Λ3 equations. 395 HF. 166. 228 diagrams. 491 IH-CCSD. 4 average-interaction approach. 158. 413 Koopmans’ theorem. 20. 239–242 linked Hugenholtz skeletons. 133 multiconﬁgurational. see vertex interchange theorem. 185. 417–418. 424 unrestricted. second-order. see state universal MRCC HK. see intermediate Hamiltonian H half-vertices. 118–123. incomplete independent particle approximation. third-order. 466. 6–7 canonical. 75–80 zero-order. 490 intermediate-Hamiltonian MRCC. of QDPT and H MRCC ˆ int . 362. 491 similarity transformation method. 233 level-shift operator. 139 connection patterns. 227–250 cancellations. 242–250 contribution status of skeletons. see equation-of-motion CC. 463 in valence-universal MRCC. 146 QDPT level-shift operator . 228 one-electron interaction. 234 for renormalization term. see Hilbert-space MRCC Hugenholtz diagrams. 116. 148. 25–27 hyperpolarizability. ﬁrst-order. 257. see intermediate-Hamiltonian MRCC IMS. 343 open-shell. 431 open-shell high-spin case. 463. see reference function. 379. 145 Hylleraas variation principle. ionization potentials IP-EOM-CC. 365. 75 Hartree–Fock gradients. see state-speciﬁc MRCC state-universal. 401 intermediate Hamiltonian (IH). 185. 494 IH-CCSDT. 492–494 intermediate normalization. 141 non-HF case. 153. 234 HS-CC. 257. 237 unlinked ASG skeletons. 379 ˆ eﬀ . 410. 300. 463. see intermediate Hamiltonian IH-CC. 111 Hamiltonian. 376 diagrams for Λ2 equations. 369 eﬀective-Hamiltonian form. 396. 377 diagrams for Λ3 equations. 410 Hellmann–Feynman theorem. 230–232 diagrams. ionization potentials Jeziorski–Monkhorst Ansatz. 120 classiﬁcation. 400. 129. 236 disconnected diagrams. 206 third-order.

466 disadvantages. 80 Møller–Plesset perturbation theory. fourth-order without triples MBPT(n) (n = 2. 154 renormalization term. multireference multiplicity. 131–132. 227–250 in valence-universal MRCC. 342 comparison with CI. 154 529 role of EPV terms. 403 magnetic susceptibility. see spin multiplicity multireference coupled cluster. 165–176 London orbitals.Subject index Λ equations. 374–375 relationship to T ˆ ). 3. 160 ASG diagrams for principal term. 140 in canonical HF case. see gauge including atomic orbitals magnetic perturbation. 157–159 in RSPT. 132 EPV terms. 368 ˆ† . 465–467 in valence-universal MRCC. 1. 157–159 . 203 schematic. 154. 86–89 of operator products. 144 cancellation of unlinked diagrams. see conﬁguration interaction. 186 Møller–Plesset partitioning. 163. see conﬁguration interaction. 139–144 linked quadruples contribution. 154 doubles contribution. 342 sixth-order. see self-consistent ﬁeld. . 134. 341–344 computational aspects. see Hartree–Fock theory. 135–136 second-order wave function. 196 second-order. 186 complete. 80 MRCC. 177–184 diagram summation techniques. 203–227 ﬁrst order. 139. 192–194 in state-universal MRCC. 152–153. 159–164 many-body perturbation theory energies ﬁrst-order. 480 model states. multiconﬁgurational MCSCF. 156 singles contribution. 251–254 in CI. 129. see state-universal MRCC linked diagrams. third-order. 418 many-body perturbation theory (MBPT). 49. 46 N -dependence of diagrams. 6. 34–36 in CC. 1 motivation. . 205–207 third-order. 51. 159 fourth-order. see model states model space. 140. 475 incomplete. 2. 153. 462–495 multireference methods. 342 ﬁfth-order. 118–123 MBPT. 158–159 proof. 414 history. 163. see multireference coupled cluster MRCI. see many-body perturbation theory energies. 402 many-body methods history. 136–138 matrix elements evaluation. 194 Li–Paldus approach. 154 without triples. 146. 477 in Hose–Kaldor approach. 10–11 many-body perturbation theory energies fourth-order principal term. 367 Λ operator. 102–105. 411. 163. 477 quasicomplete. 228 selection. 1 relationship to CI. 138. 136 in QDPT.). see many-body perturbation theory MBPT(4)SDQ. 131 second-order. see many-body perturbation theory energies MCHF. 159 third-order. see quasidegenerate perturbation theory N He atoms example in BWPT. 138 renormalization term. 146–150 Hugenholtz diagrams for principal term. 154 non-HF case. see transformation matrix model functions. 228 in QDPT. 51. 144–146 principal term. 154. 342 comparison with CI. 365. 138–152. 193. 342 many-body PT wave functions ﬁrst-order. 431 conﬁguration interaction. multiconﬁgurational MO coeﬃcients. 193. 11. . 177–180 energy. 188 level-shift operator (W diagrams. 154. 466 in QDPT. 141 Hugenholtz skeletons for principal term. 163. 156 comparison with CC. 490–491 in valence-universal MRCC. 198 linked-diagram theorem. 402. see multireference coupled cluster perturbation theory. multireference coupled cluster. 12–15 in MBPT. 466. 139 generalized (GMBPT). 154 triples contribution. 366 diagrams. 150. 154–156 generalized (GMBPT). 220–225 modiﬁed. 342 approximation of CI.

5. 92–100 product. 197. 75 notation. 344. 123–128 particle–particle. 385–396. 69 relative to Fermi vacuum. 329. 347–405 response treatment. 410. 19. 5 spin-multiplicity monitoring. 344. 51. 345 potential-energy surface. 362–363 excited states. 491 gradient. 361. 193 virtual. 475 low-spin. 185–250 classiﬁcation of one-electron states. 361–365 pseudolinearity. see degenerate perturbation theory open-shell states. 185. 186 sector projectors. 412 normal-product form. 76 orbital relaxation. 463 on model space. 448 Hessian. 129. 462 high-spin. 80–83 one-body. 194 diagrammatic expansion. 27–29 on complementary space. 92 particle–hole. 454–461 phase factor. 488 orbitals including spin. 197 particle states. 421–425. 402 nuclear magnetic shielding. 3 polarizability frequency-dependent. 410 EOM-CC. 431. see spinorbitals localized. 80–81 two-body. 463 on orthogonal space. 94 hole–particle. 81–83 not in normal-product form representation. 381.530 Subject index particle lines. 410 in state-universal MRCC. 5 operators hole–hole. 193 valence. 410 diﬃculties for single-reference methods. 72 passive lines. 425. see wave operator one-body basis. 462 appropriateness of SS-CC. 476 properties as energy derivatives. 83–85. 2–3 excitation level. 431–432. 193 open-shell perturbation theory. 98 normal-product form. 100–105 one-body states core. 362 using EOM-CC. 361–365 vs. 67–68 relative to Fermi vacuum. 136. see normal product normal product. 162 quasidegenerate perturbation theory (QDPT). 3 nuclear magnetic resonance (NMR) chemical shifts. 193 convergence. 185. 441 UHF. see quasi-Hartree–Fock theory. 449–454 in CC theory. 71–75 . 376 QCISD. 440–441 zero-order approximation. 5. 363. 441 open-shell singlets. 39 projection operator. 411 spatial. 67. 185. 365. see many-body PT perturbation theory (PT) convergence. 387–405. 185 particle–hole formalism. 182–184. 440 ROHF. 128–129 perturbation theory many-body. 194 natural orbitals. 138. 436 frequency-dependent using EOM-CC. 46–53 time-dependent. 472 discontinuities. 407–410 orthogonal space. 97 representation. as expectation values. 74–75 with contractions. 186. 471 in valence-universal MRCC. 75. 381 principal term of mth-order energy. see conﬁguration interaction. quadratic QDPT. 440 EOM-CC. 75–78 GMBPT. 204 perturbation operator. 5. 454–461 higher-order. restricted quadruples corrections. see quasidegenerate perturbation theory QHF. 106–111 in QDPT. 59–65 invariance under unitary transformation. 72 Planck’s constant. 362 potential-energy curve. 130–164 formal. 413 non-HF case. 211–215 physical vacuum. 455 RSPT. 18–53 general derivation. 53 similarity transformation. 29–33 other approaches. 10. 194–195 periodic. 5. 11. 459–461 tensor. 65–67 orbital energy. 185. 361. 132. see quasi-Hartree–Fock theory QRHF. 402 Ω. 186 Pad´ e approximants. 5. 144–146 normal order. 54 one-body operators diagrammatic representation. 133.

396 diagrams. 410 spinorbital basis. 422 computation. 39. 295 normal-product form. see state-speciﬁc MRCC state-speciﬁc MRCC. 441. 436 Hermitian average. 414 ROHF. 422 raising and lowering operators. 473 suﬃciency conditions. open-shell RSPT. 471–474 Brillouin–Wigner form. see coupled cluster SS-CC. 202 eﬀective (R). 146–150 representation by insertion. 381–385 anti-Hermitian part. 40 quasidegenerate case. 413 non-diagonal case. 54–89 history. 52. 415 QRHF. 486. 383. 410 diagrams for CCD. 37–40 bracketing procedure. auxiliary operator SCF. 45–46 fourth-order energy renormalization term. see valence-universal MRCC. 422 expectation value. 455 quasi-equivalent external lines. 18–27. 364 EOM-CC states. 215–220. 4 multiconﬁgurational. 460 line. 91–92 SOS form. 198–203. see Rayleigh–Schr¨ odinger perturbation theory Rydberg states. 39 and folded resolvent lines. 12 Slater determinants. see one-body basis spinorbitals. 149 folded. 315. 130 denominator. 421 monitoring. 395 restricted. 120. 225. 207 similarity transformation. 422–425 projected value. space. 4 diagrammatic representation. 202 non-diagonal case. 414 quasiloop. see self-consistent ﬁeld Schr¨ odinger equation connected form. 403 asymmetry to time inversion. 43–44 reaction operator. 417. 205. 385–396 reference state. 6 semicanonical transformation. 46–53 quasienergy. 185–191 incomplete model space. 40–41 summary of formal results. 463. 5 localized. 72 Rayleigh–Schr¨ odinger perturbation theory. 85. 473–474 extensivity. 280 time-dependent. 4. 452 frequency-dependent. 474 state-universal MRCC. 411 noncanonical HF. 410 QRHF case. 40–41. 383 RHF. 130–131. 413–415 size extensivity. 407 low-spin open-shell case. 5 semicanonical. 445. 226 schematic representations. 399–400 one-body. 396 diagrams. 410. 34. 406–411 Brueckner orbital case. 462 spin multiplicity. 410 orthogonally spin-adapted case. 182–184 in CC. 423 spin contamination. 39 non-diagonal case.Subject index diagrammatic notation. 421 spin-ﬂip CC. 5. 382 excited states. 411–415 Kohn–Sham determinant. see level-shift operator reference energy. 139 resolvent. 78 arbitrary. 129. 458 response density matrix. restricted ring diagrams. 71–73. 382–385. see sum over states form spin eigenvalues. 139. 135 quasi-operator. 421–425 spin summations. see Hartree–Fock theory. 493 self-consistent ﬁeld. 189 resonance. 465–471 coupling term. 31–32. 413–414 renormalization term. 283 quasi-Hartree–Fock (QHF) theory. 445 S operator. 474 equations. see Hartree–Fock theory. 381–382 two-body. 393–396 reference function perturbed. 384 permutational symmetry. 79 derivative. 410 ROHF case. 473. 41–42 extensivity. 73. 474 redundancy. 195–198 folded diagrams. 469 . 226 formal. 129. 474 comparison of results. 1 sectors of rock. restricted. 315 SRCC. 401 non-Hermiticity. 117. 142 531 ROHF. 182–184 canonical HF. 473–477. 215–217. 463. 381–383. 410 UHF case. 455 second quantization. see Hose–Kaldor approach schematic representations. 225. see extensivity size-consistency.

486 incomplete. 480. 5 ˆ1 operator T alternative treatment. 115 vacuum expectation value. see state-universal MRCC sum-over-states (SCS) form. 425–426 terminology. 125. 480–482 Bloch equations. 5. 280 in QDPT and MRCC. 300–301. 477 diagrammatic representation. 187. 192. 95–97 two-body. 465 local active spinorbitals. 111–113. 118 interpretation. 193. 469–471 STEOM-CC. 467 equations. 195 folded. 485. 300. 465 in valence-universal MRCC. 24–25. 402. 480–482 intermediate normalization validity. 327 Z -vector method. 483–488 diagrammatic representation. 38 wave operator (Ω). 387. 257. 467 matrix elements. 477–483 weight factor. 482–483 vertex one-body. 475–490 auxiliary operator (S ). 207–211 third-order. see Hartree–Fock theory.532 Subject index EOM-CC analogy. 431 symmetry restriction. 477. 172 and extensivity. 48. 476 spectator amplitude. 139 unrestricted summation. 480 sector projectors. unrestricted unlinked diagrams. 454. 69–70 generalized. 112. 6 in QDPT. 486 consistency requirement. 192. 487 hierarchical solution. 468–469 extensivity. 90–91 transformation matrix. see equation-of-motion CC. 186. 475. 51. 125. 115 in QDPT. 111. see interchange theorem zero-order approximation. 487 wave operator. 211–215 Wick’s theorem. 198 unlinked skeletons. 475 complete. 93 interpretation. 146–150. 111–129 product. 460 symmetry-adapted cluster CI. 156–159 in QDPT. 119. see valence-universal MRCC ˆ operator W in MBPT and CC. 128. 477–483 simpliﬁcations. 156–159. 203–227 ﬁrst-order. 485 eﬀective Hamiltonian. 67. 121 virtual states. 6 zero-order energy. 449. 85–86 proof. 487 model space. 70–71 relative to Fermi vacuum. 472 in state-universal MRCC. 193. 228 VU-CC. 254. see level-shift operator in single-reference RSPT. 199 valence states. see reference state diagrammatic notation. 486 . 3 Thouless theorem. 467 disconnected clusters. 113–118 vacuum gap. 419 time axis. 467 internal excitations exclusion. 431 symmetry breaking. 90. 203 multiconﬁgurational. similarity-transformed SU-CC. 74–75 Wigner’s 2n + 1 rule. 473 perturbed. 484. 53. 136. 485. 104. 53. 463–464 diagrams. 433. 413. 91 time ordering. 84. 488 selective summation. 418–421 role. 436 two-body operators diagrammatic representation. 478–483 diagrammatic representation. 488 subsystem embedding condition. 225–227 in state-speciﬁc MRCC. 480 normal order of products. 487 quasicomplete. 464. 228 valence-universal MRCC. 488–490 factorization of disconnected diagrams. 197 second-order. 140 valence lines. 51. 70. 204–205 schematic. 388 transition moment. 139. 79 zero-order function. 113–118 UHF. 414.

Sign up to vote on this title

UsefulNot useful