The Crystal Field Handbook is based on the modern conceptual understanding of crystal fields. It clarifies several issues that have historically produced confusion, particularly the effects of covalency and ligand polarization on the energy spectra of magnetic ions. Clear instructions and a set of computer programs are provided for the pheno- menological analysis of energy spectra of magnetic ions in solids. Readers are shown how to employ a hierarchy of parametrized models to extract as much information as possible from observed spectra. Special attention is given to the superposition model, the parametrization of transition intensities and methods to be used when the standard (one-electron) crystal field parametrization fails This is the first book on crystal field theory which brings together the whole range of phenomenological models, together with the conceptual and computational tools necessary for their use. It will be of particular interest to graduate students and researchers working on the development of opto-electronic systems and magnetic materials. Berty NG is presently Head of the Air Quality Modelling and Assessment Unit in the Environment Agency. She taught undergraduate courses in quantum mechanics and group theory. She has carried out research work in several areas of computational physics, including crystal field theory, superconducting phase transitions and the dispersion of pollutants in tidal waters. Her doctoral research pioneered the use of many-body techniques in the ab initio calculation of crystal field and correlation crystal field parameters. She is author and co-author of 23 refereed papers on various aspects of crystal field theory. DouG Newman has carried out research on crystal field theory since 1963. His work with Margaret Curtis on the ab initio calculation of lanthanide crystal fields provided the basis for the introduction of several parametrized models into the literature: the superposition model (1967), its application to S-state ions (1972), the correlation crystal field model (1968), the spin-correlated crystal field model (1970) and the vector crystal field model (for transition intensities) in 1975. He is author and co-author of more than 150 refereed papers on crystal field theory and related topics in mathematical physics. He is joint author of How to use Groups, with J. W. Leech, published by Methuen in 1969 and Computational Techniques in Physics, with P.K. McKeown, published by Adam Hilger in 1987.Crystal Field Handbook edited by D. J. Newman University of Southampton and Betty Ng Environment Agency Wales CAMBRIDGE UNIVERSITY PRESSCAMBRIDGE UNIVERSITY PRESS Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, So Paulo 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/9780521591249 © Newman and Ng 2000 ‘This publication is in copyright. Subject to statutory exception ‘and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2000 This digitally printed version 2007 A catalogue record for this publication is available from the British Library Library of Congress Cataloguing in Publication dara Crystal field handbook / edited by D.J. Newman and Betty Ng p. om Includes bibliographical references. ISBN 0 $21 59124 4 1. Crystal field theory. I. Newman, D.J. (Douglas John) Il. Ng, Betty, 1959- QD475.C76. 2000 5387.43-de2199-052178 ISBN 978-0-521-59124-9 hardback ISBN 978-0-521-03936-9 paperback 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 i, of will remain, accurate or appropriate.To our parents. ..Contents List of contributors page ix Preface xi Introduction 1 1 Crystal field splitting mechanisms D. J. Newman and BettyNg 6 2 Empirical crystal fields D. J. Newman and Betty Ng 26 3 Fitting crystal field parameters D. J. Newman and Betty Ng 43 4 Lanthanide and actinide optical spectra G. K. Liu 65 5 Superposition model D. J. Newman and Betty Ng 83 6 Effects of electron correlation on crystal field splittings M. F. Reid and D. J. Newman 120 7 Ground state splittings in S-state ions D. J. Newman and Betty Ng 140 8 Invariants and moments Y. Y. Yeung 160 9 Semiclassical model K. S. Chan 176 10 Transition intensities M. F. Reid 190 Appendic 1 Point symmetry D. J. Newman and Betty Ng 2207 Appendic 2 QBASIC programs D. J. Newman and Betty Ng 239 Appendix 8 Accessible program packages Y. Y. Yeung, M. F. Reid and D. J. Newman 254 Appendie 4 Computer package CST Cz. Rudowicz 259 Bibliography 269 Index 287 viiContributors Chan Kwok Sum Department of Physics and Materials Science, City University of Hong Kong, Kowloon, Hong Kong Guokui Liu Chemistry Division, Argonne National Laboratory, Argonne, USA Douglas J. Newman Department of Physics and Astronomy, University of Southampton, UK Betty K. C. Ng The Environment Agency Wales, Cardiff, UK Michael F. Reid Department of Physics and Astronomy, University of Canterbury, Christchurch, New Zealand Czeslaw Rudowicz Department of Physics and Materials Science, City University of Hong Kong, Kowloon, Hong Kong Yau Yuen Yeung Department of Science, Hong Kong Institute of Education, New Territories, Hong Kong ixPreface Our aim in producing this book is to make a wide range of up-to-date tech- niques for the analysis of crystal field splittings accessible to non-specialists and specialists alike. All of these techniques are based on the phenomeno- logical approach, in which the aims are to (i) parametrize observed crystal field splittings using models which pro- vide the maximum predictive power, (ii) test model assumptions directly by means of the quality of parameter fits and the accuracy of their predictions, (iii) use model parameters as a convenient interface between first princi- ples calculations and experiment. It is expected that readers will be mainly interested in the first two of these aims. Hence relatively little space is devoted to the description of first principles calculations. While most of the book is concerned with practical applications of the phenomenological approach, some space is given to providing a critical appraisal of relationships between the this approach and other approaches to crystal field theory. A series of examples, backed with a set of QBASIC programs, is provided to enable readers to rapidly attain proficiency in the basic techniques of phe- nomenological analysis. Those who wish to embark on more sophisticated analyses are directed to other available program packages. In addition to the procedures covered in the QBASIC programs, the book covers three top- ies which have barely been touched upon elsewhere. These are the method of crystal field invariants (Chapter 8), the semiclassical model (Chapter 9) and the use of parametrized models in the analysis of transition intensities between crystal field split levels (Chapter 10). In the process of constructing the bibliography, a file (cfh-all.bib) contain- ing a far more comprehensive list of references on crystal fields and related xixii Preface topics was generated. This is formatted as an input file to BibTeX, making it useful in the construction of all types of reference list. The editors are grateful to Dr M. F. Reid for his assistance in generating this file, and for making it available on the website http: //www.phys.canterbury.ac.nz/crystalfield It is hoped that readers will assist Dr Reid in keeping this file accurate and up-to-date. One editor (DJN) would like to express his gratitude to all his former col- leagues and students who, over the last 40 years, have participated in the development of the phenomenological approach ot crystal fields. Thanks are also due to the staff and students at the University of Southampton who contributed to the initiation and the development of this book project. We are especially indebted to the contributing authors, who have respected deadlines and shown considerable forbearance with the fussiness of the edi- tors in their attempts to integrate the notation and conceptual approach in different parts of the book. The editors would also like to express their gratitude to the editorial staff at Cambridge University Press for their patient help in eliminating typo- graphical errors and improving the layout. We are particularly grateful to the respondents in ‘texline’ for helping with our TeX problems and adapting the standard ‘cupplain’ format to our requirements. Doug Newman and Betty NgIntroduction D. J. NEWMAN University of Southampton BETTY NG Environment Agency ‘The energy spectrum of an isolated magnetic ion in a crystal carries infor- mation about the magnetic ion itself, its crystalline host and the interaction between these two components of the system. Crystal field theory comprises a range of techniques for extracting as much of this information as possible from observed spectra. The aim is to express the information in a form that can be used to predict the energy spectra of related systems. Magnetic ions in crystals have many useful physical properties. In par- ticular, there is an ongoing search for new types of laser crystal [Kam95, Kam96] and new magnetic materials. In order to design the new systems required for specific applications, it is necessary to be able to predict energy level structures and transition intensities for any magnetic ion in any erys- talline environment. The techniques described in this book go a considerable way towards the achievement of this goal. The term ‘crystal field theory’ has been applied to describe two quite dif- ferent approaches. One of these, the so-called phenomenological approach, which involves the use of linear parametrized operator expressions to fit experimental results, provides a highly successful predictive tool. The al- ternative, so-called ab initio approach, in which energy levels and transition intensities are calculated from first principles, has proved far less useful. This book is mainly concerned with showing how the phenomenological ap- proach can be used to obtain information about the physical properties of magnetic ions in crystals directly from observed spectra. Our aim in this book is to make modern techniques for analysing and predicting spectra of isolated magnetic ions in crystals more widely known and more easily accessible. Physical assumptions inherent in the various parametrized Hamiltonian and transition intensity models are clarified, and conceptual and computational tools are provided for the analysis of experi- mental results. The emphasis is on extracting as much information as possi-ry D. J. Newman and Betty Ng CH2: CF parameters CH3: Single multiplet fitting Ground multiplet spectra CHS: Superposition model CH7: Spin-Hamiltonian Half-filled shell Fig. 0.1 Routes for the analysis of ground multiplet spectra, ble from observed spectra by using a range of tried and tested parametrized models as tools. Many examples from the recent literature on crystal field theory are quoted, and an extensive bibliography is included. Nevertheless, the book is not intended to provide a comprehensive review of the literature. Readers of this book either have access to experimental results that they wish to analyse, or wish to predict the energy level structure or transition in- tensities of a system which has yet to be studied experimentally. The criteria for choosing an appropriate parametrized model are the same in both cases, and depend on the type of spectrum that is to be analysed or predicted. Optical spectra contain information about energy levels and transition in- tensities for many multiplets, while spectra obtained from inelastic neutron scattering, electron spin resonance or far-infrared spectroscopy normally only provide information about the lowest energy or ground multiplets. We distinguish these as the ‘many-multiplet’ and ‘ground multiplet’ problems in the following discussion. The analysis of a given set of experimental results may involve the use of several parametrized models, as shown in Figures 0.1 and 0.2. Figure 0.1 shows three possible routes for handling the ground multiplet problem.Introduction 3 (i) If the experimental spectrum is for a magnetic ion with a half-filled shell, it is appropriate to fit the spectrum to spin-Hamiltonian param- eters as described in Chapter 7. (ii) Complete ground multiplet spectra can be analysed using the param- eter fitting procedure described in Chapter 3. One of the tables in Chapter 2 will often provide a suitable set of starting parameters. (iii) The ‘superposition model’ parametrization, described in Chapter 5, may be employed if relevant starting parameters are not available or when experimental results are insufficient to carry out the procedures described in Chapter 3. A more sophisticated procedure, designed specially for the analysis of neutron scattering results, is mentioned in Appendix 3. (iv) When a ground multiplet spectrum is only partially known, or the lines are broad, it may be possible to use the semiclassical crystal field model described in Chapter 9. This is particularly appropriate for ground multiplets corresponding to large angular momenta. Figure 0.2 (on the following page) shows four possible routes for handling the ‘many-multiplet’ problem. (i) Ifcomplete sets of energy levels are available for only a few multiplets it may be appropriate to analyse them one at a time using the methods described in Chapter 3, possibly using a set of starting parameters provided in Chapter 2. When information about crystal structure is available, further checks and analyses can be carried out using the ‘superposition model’, as described in Chapter 5. Separate multiplet analyses can also be used to provide information about correlation effects, as described in Chapter 7. (ii) When the observed spectrum provides energy levels for many multi- plets, the most productive method of analysis is provided by specially developed computer packages. One of these is described in detail in Chapter 4, and another is described briefly in Appendix 3. (iii) When quantitative information about transition intensities is available, the parametrization techniques described in Chapter 10 should be em- ployed. This parametrization is particularly relevant for predicting the properties of crystal lasers. (iv) In cases when it is difficult to carry out the fitting procedures described in Chapter 4, e.g. for reasons such as low site symmetry, multiplet moments can be used to provide a starting point for the techniques described in Chapter 8.4 D. J. Newman and Betty Ng CH3: Single multiplet fitting CH2: CF parameters CH4 & APP3: Many multiplet fitting CH8: Moment method Many multiplet spectra CH6: Correlation CHI1O: CHS: Superposition Intensities model Fig. 0.2 Routes for the analysis of many-multiplet spectra. It is important to start with a clear idea of the purpose of any proposed analysis of experimental results. While this is generally not relevant to the choice of route, it will determine just how far it is sensible to go with a specific analysis. The intention may be to: (i) verify the interpretation of an observed set of energy levels; (ii) determine the single-ion part of a parametrized Hamiltonian as the starting point of a more general investigation of the dynamic interac- tions between magnetic ions, or between magnetic ions and phonons; (iii) provide information which will assist in predicting properties of related systems, e.g. as the first step in designing systems which are intended to have specific opto-electronic properties;Introduction 5 (iv) investigate the various types of parametrized model, e.g. study their reliability and predictive power; (v) use parametrized models as a route to obtain an improved understand- ing of the electronic structure of magnetic ions in crystals. The theoretical justification of simple parametrized models, given in Chap- ter 1, is particularly relevant to the last two intentions. ‘The parametrized models used in this book for the analysis of energy lev- els and transition intensities are all based on ‘effective operators’ which act solely on the many-electron states of the open-shell electrons of the mag- netic ions. The structure of these operators depends upon a few general, but nevertheless clearly defined, physical assumptions. None of the param- eters has predetermined values. Given sufficient experimental results, an attempt can be made to fit all the model parameters. This phenomenologi- cal approach is quite rare in physics because it is unusual to have the large amounts of information that are provided by energy spectra [New78]. ‘The most important magnetic ions are all within the transition metal se- ries (with partially filled 3d or 4d shells), the lanthanide (or rare-earth) series (with partially filled 4f shells) and the actinide series (with partially filled 5f shells). Most of the examples in this book are concerned with the analy- sis of observed energy spectra of ions in the lanthanide and actinide series. Some excellent treatises are already available on the transition metal series, notably the books by Gerloch and Slade [GS73] and Ballhausen [Bal62). While there are some differences in detail arising from the different relative importance of the crystal field and other interactions involved in generating the spectra, the phenomenological models discussed in this book, and their justifications, are essentially the same for all magnetic ions. It is worth noting the close parallels that exist between the phenomeno- logical models used for the analysis of energy level structures and those used in the analysis of transition intensities. Similar parallels also relate the con- ceptual analysis of contributions to the transition intensities, developed in Chapter 10, and the analysis contributions to the crystal field, developed in Chapter 1. A package of QBASIC programs, described in Appendix 2, has been de- signed to provide a simple ‘tool kit’, designed to illustrate the various tech- niques of analysing spectra which are discussed in this book. The use of these programs is described in Chapters 3 and 5. More specialized computer packages, which the reader can access, are briefly described in Chapter 4, and Appendices 3 and 4.1 Crystal field splitting mechanisms D. J. NEWMAN University of Southampton BETTY NG Environment Agency In order to interpret the information obtained from magnetic ion spectra using parametrized crystal field models it is necessary to have a qualitative understanding of the physical mechanisms through which the crystalline environment induces energy level splittings. The discussion of mechanisms given in this chapter should be seen in this light. It is not intended to provide a practical basis for quantitative ab initio calculations of crystal field splittings. Instead, it provides a conceptual description of the various mechanisms that contribute to crystal field splittings. The first three sections give a qualitative description of the most impor- tant mechanisms which contribute to crystal field splittings. These comprise the electrostatic, charge penetration, screening, exchange, overlap and co- valency contributions. Sections 1.4 and 1.5 express these contributions in terms of a simple algebraic formalism related to the ‘tight-binding’ model in solid state physics. Several other formal approaches to crystal field theory, including the phenomenological approach, are described in the subsequent sections. Numerical results for a particular system, based on the formalism developed in Sections 1.4 and 1.5, are given in Section 1.9. The signifi- cance of these results to the phenomenological methods used in this book is summarized in Section 1.10. 1.1 The crystal field as a perturbation of free-ion open-shell states Crystals can be regarded as assemblies of free ions, bound together in several possible ways. Ionic crystals are held together by a balance between long- range electrostatic attractive forces and short-range repulsions, the latter being largely attributable to the exclusion principle acting between the outer electrons on neighbouring ions. Most crystals also have some degree of 61 Crystal field splitting mechanisms 7 covalent bonding, in which the total energy is reduced by the sharing of electrons between orbitals on neighbouring ions. In conductors, bonding is effected through the coupling between the ions and a sea of conducting electrons. Free magnetic ions are characterized by a Hamiltonian with spherical symmetry. In a crystalline environment, the site symmetry of magnetic ions is lowered, thus breaking the degeneracy of the free-ion J or L multiplets. The crystal field can therefore be thought of as a perturbation, generated by the same mechanisms that produce crystal bonding, acting on ‘free’ magnetic ions. The qualitative form of the splitting (ie. the effect of the crystal field on the degeneracies) can be predicted from a knowledge of the site symmetry of the magnetic ion, as will be explained in Chapter 2 and Appendix 1. An electrostatic potential at points r near to a magnetic ion, due to any external charge distribution, can be expressed as the multipolar expansion centred on that ion, viz. Vor(r) = S03 V(r), (1.1) K where the individual contributions V)(r) can be expressed as sums (over 4g) of numerical factors times the functions TY ¢q(0,4) Here r is expressed in terms of spherical polar coordinates (r,,$) centred on the magnetic ion, and Yi,g are spherical harmonic functions. Whatever the magnitudes of the terms in (1.1), their r* radial dependence is determined by the requirement that the Laplace equation is satisfied. Given that the one-electron wavefunctions can be factorized as a product of radial and angular parts, it is possible to replace (1.1) by an expansion in which the radial integration has already been carried out, making the potential a function of the angular coordinates @ and ¢ alone. Equation (1.1) can then be rewritten as Ver (0,4) = VQ, ¢4). (1.2) k The individual contributions V) (9,6) are now written in terms of (*)Ving(8, 4), where (r*) is the ‘expectation’ of r* for an open-shell electron, and can be expressed as a radial integral (e.g. see equation (16.3) of [AB70]). Free-ion8 D. J. Newman and Betty Ng radial wavefunctions necessary for evaluating these integrals can be obtained by Hartree-Fock free-ion calculations (e.g. [FW62]). Given the uncertainties in the precise form of the radial wavefunctions, the calculated values of the (r*) are necessarily inaccurate. In the phenomenological approach used in this book, the radial integrals (r*) are absorbed into the numerical coefficients of equation (1.2). As shown subsequently in this chapter, the resulting expansion is not restricted to the representation of electrostatic potentials and can take into account the full range of mechanisms that contribute to the crystal field splitting. The value of using a multipolar expansion for Vor is that only a few of the multipole moments, i.e. those with even k < 21, have non-zero angular factors in the matrix elements between open-shell single-electron states with angular mo- mentum /. A discussion of these selection rules is given in Appendix 1. The first term, with k = 0, does not produce any crystal field splitting. The second term, with k = 2, is frequently referred to as the ‘quadrupolar’ contribution, while the k = 4 term is occasionally referred to as the ‘hex- adecapolar’ contribution. Happily, this nomenclature is seldom extended to the case k = 6. It is more convenient to refer to contributions by their rank ke The first step in formulating an ab initio calculation of crystal field split- tings is to construct the open-shell states for a ‘free’ magnetic ion. Consid- erable simplifications will result if these are taken to be the many-electron wavefunctions corresponding to the free-ion configurations /”, where n is the number of open-shell electrons. These many-electron functions can be constructed as determinants of one-electron wavefunctions, which are de- rived from a Hartree-Fock self-consistent mean field calculation for the ion [LM86]. In addition to the one-electron wavefunctions, the Hartree-Fock calculation also produces an isotropic (or spherically symmetric) electro- static potential. The crystal field can be regarded as an anisotropic pertur- bation of this potential. It must be noted that free-ion interelectronic coulomb interactions (that are not included in the mean field) and spin-orbit interactions produce even greater splittings of the degenerate configurational energies than does the crystal field. The resulting free-ion many-electron wavefunctions are usually expressed as linear combinations of the aforementioned determi- nantal wavefunctions constructed using the one-electron wavefunctions of a Hartree-Fock calculation. However, considerable progress in calculating crystal field splittings from first principles can be made simply by regarding these interactions as ‘other perturbations’, which act independently on the Hartree-Fock single-electron states.1 Crystal field splitting mechanisms 9 1.2 The electrostatic crystal field model Historically, the first approximation to the crystal field in ionic crystals was obtained by assuming that, apart from the magnetic ion under considera- tion, all the other ions in the crystal may be treated simply as sources of an electrostatic field. In this, so-called electrostatic crystal field model, each of the electrons in the open shell of a magnetic ion is supposed to move independently in an electrostatic potential, the anisotropic component of which is produced by the surrounding crystal. It turns out that this model is incorrect, and much of this chapter is devoted to showing why this is the case, However, given that the electrostatic model is so entrenched in the literature of the subject, it provides a useful starting point for the present discussion. 1.2.1 Quantitative development of the electrostatic model The simplest form of electrostatic model is the ‘point charge model’, in which the electrostatic potential at a magnetic ion is obtained by summing the contributions of all the other ions (regarded as point charges) in the lattice. The physical justification for doing this is that the electrostatic potential external to any spherically symmetric charge distribution can be shown to be the same as that generated by the same net charge concentrated at a point at the centre of that distribution. Hence an approximation inherent in the point charge model is that the open-shell electrons on the magnetic ion lie outside the charge distributions on the neighbouring ions. Although this approximation is poor, the simple point charge model is successful in the sense that, at least in the case of ionic crystals, it almost always predicts crystal field parameters of the correct sign. This corresponds to a repulsion of the open-shell electrons by the negative charges carried by the neighbouring ions. The rank k point charge contributions to the electrostatic potential can be shown to have the simple functional dependence 1/R*+!, where R is the distance between the point charge and the centre of the magnetic ion. This distance dependence, especially in the case k = 2, leads to the expectation that distant ions in the crystal can make a significant contribution to the electrostatic potential. Because there are strongly cancelling positive and negative contributions in the point charge lattice sums, quite sophisticated methods have to be used to obtain reliable results. While it is feasible to overcome the difficulties involved in making accu- rate calculations of wavefunctions and lattice sums, the results obtained still10 D. J. Newman and Betty Ng depend on the doubtful assumptions inherent in the point charge model. Ex- perience has shown that, however much effort is spent on improving their accuracy, such calculations never even come close to reproducing the ex- perimentally observed crystal field splittings. Typically, in the case of the lanthanide ions, they overestimate the rank 2 contributions by a factor of 10 and underestimate the rank 6 contributions by a similar factor [New71]. Early attempts to improve the point charge model (e.g. see [HR63]) were focussed on taking account of the lack of exact: spherical symmetry of ionic charge distributions in crystals. It was suggested that, apart from point charge contributions, electrostatic contributions from induced (point) dipole and quadrupole moments on the ions should be included in the lattice sums. However, these improvements did not enhance agreement with experiment, the gross errors mentioned above remaining unchanged. The overestimation of the rank 2 crystal field in electrostatic models of lanthanide and actinide crystal fields can be largely attributed to the neglect of the screening effects of the filled (N + 1)s%p® shell, which lies outside the partially filled N f shell. Virtual excitations of the outer-shell electrons mod- ify the externally generated electrostatic field experienced by an open-shell electron. This screening effect can be incorporated into the electrostatic point charge model by means of multiplicative screening factors, denoted (1—0,). Several attempts have been made to calculate the screening factors which, in accord with intuition, are found to be less than unity. For the trivalent lanthanide ions, 04 and o¢ are small, but 02 has been calculated to be about 0.7 [SBP68]. This clearly has a significant effect on calculated rank 2 electrostatic contributions. There are, however, considerable difficulties in carrying out calculations of screening factors which are both realistic and accurate. For example, the calculation by Sternheimer et al. [SBP68] was carried out for free ions, rather than ions in a crystalline environment. More recent computational {NBCT71] and phenomenological [NP75] analyses suggest that the value of 09 is in fact close to 0.9, allowing only about 10% of an externally generated rank 2 point charge contribution to ‘get through’ to the 4f electrons in the trivalent lanthanides. Although the point charge approximation may provide an adequate rep- resentation of distant ion contributions, it cannot possibly give an accurate description of the contributions from the immediate neighbours (or ligands) of a magnetic ion. One reason for this is that the charge distributions of the ligands significantly overlap the open-shell electronic wavefunctions of a magnetic ion. Allowing for the effect this has on the electrostatic field makes the results much worse, in the sense that the net rank 4 electrostatic1 Crystal field splitting mechanisms ul contributions become too small and the lanthanide and actinide rank 6 con- tributions change sign. Moreover, no purely electrostatic contribution has been found which could correct these discrepancies. In summary, every attempt to improve the results obtained with the sim- ple point charge model introduces further computational difficulties and a widening of the gap between theoretical predictions and experimental ob- servations. This exposes the fundamental deficiencies of the electrostatic model. The failure of the more sophisticated forms of the electrostatic model has led some authors to attempt to exploit the fact that a simple point charge model usually gives crystal field contributions of the correct sign by postulating different ‘effective’ magnitudes of the point charges for the contributions of each rank k. However, while this fudge may work for a given system, the ‘effective’ charges have no physical reality, so there is no basis for making comparisons between the effective charges for different systems. In consequence, effective point charge models have no predictive capability. 1.2.2 Qualitative features of the electrostatic model Although the electrostatic model does not provide an adequate quantitative description of the crystal field, several of its qualitative features carry over to more realistic models. The first of these features is the interaction of the electrostatic potential separately with each of the open-shell electrons. In other words, crystal field splittings can be described in terms of one-electron matrix elements. This feature makes the number of parameters required to describe observed crystal field splittings independent of the number of electrons in the shell. The second feature is that the electrostatic potential is spin independent: the one electron matrix elements do not depend on spin. Thirdly, the electrostatic potential can be expressed as a finite multipolar expansion, with terms spanning very few ranks. These three properties of the electrostatic crystal field model are all expressed by equation (1.2). A fourth feature of the electrostatic point charge model, not represented in (1.2), is that the electrostatic potentials in the neighbourhood of a magnetic ion can be calculated as a simple sum, or superposition, of contributions from all the different sources. It is shown in Chapter 5 that this property provides a useful means for the analysis and interpretation of empirically determined crystal fields.BR D. J. Newman and Betty Ng 1.3 Other contributions to the effective potential ‘Ligand field’ models put the emphasis on the construction of an appropriate representation for the open-shell wavefunctions, rather than on determining the form of an effective crystal field potential which acts upon the open-shell electrons. ‘Molecular orbitals’ are constructed from an appropriate linear combination of open-shell and ligand wavefunctions determined by overlap and covalency (see Section 1.7). The alternative ‘crystal field’ approach used in the present analysis, in which an effective potential acts upon the open-shell electrons, must certainly take account of overlap and covalency. Overlap contributions arise from the non-orthogonality of the one-electron free-ion wavefunctions on neighbouring ions. In particular, overlap between the electronic wavefunctions in the open-shell and the outer-shell wave- functions on the ligand results in Pauli exclusion. This has the effect of repelling the open-shell electrons from the ligand directions, producing a crystal field contribution of the same sign as that produced by their nega- tive ionic charges. Covalency arises from a mixing of open-shell and ligand wavefunctions, which contributes to crystal bonding. Because different open-shell elec- tronic states mix with the ligand states differently, covalency also produces a contribution to the crystal field splittings. As will be demonstrated below, the signs of the overlap and covalency contributions to the crystal field are invariably the same, although their origin is different. The non-local nature of covalency and overlap makes it apparent that their contributions to the crystal field cannot be expressed simply as a function of electron position, as in (1.1). Nevertheless, it can be represented by a one-electron operator. It turns out that this makes no difference to the number of parameters required to characterize a given one-electron crystal field. 1.4 Crystal field energies in terms of many-electron matrix elements In this section, we formulate a crystal field model in which the open-shell electrons of a magnetic ion interact individually with their crystalline envi- ronment. The aim is to demonstrate that the crystal field potential energy of the magnetic ion can be expressed as a sum of these individual elec- tron contributions. Furthermore, as all the electrons other than those in the open shell are supposed to be in filled orbitals, these contributions are independent of the spin of the open-shell electron1 Crystal field splitting mechanisms B ‘The formulation includes one electron in an open-shell orbital on the mag- netic ion and all the outer sp°-shell electrons on the ligands. The wave- function of the open-shell electron is denoted by $a. The single-electron wavefunctions corresponding to electrons in the outer (closed) shells of all the ligands, are denoted by x,. The ground state of the complex is approx- imated as a single determinant including all N = 8M +1 single-electron wavefunctions, where M is the number of ligands. It is written Wao = (1/VN)|ballv xv, (1.3) where the Greek subscripts differentiate between both orbitals and spins. Wavefunctions for strongly bound, or ‘core’, electrons are included with the atomic nucleus as contributing only to the electrostatic field. Hence they do not appear explicitly in the determinantal wavefunction. To first order, the energy Eno of the approximate ground state Wao is given by (VaolH|Vao) (Wao|Wao) * Here H is the Hamiltonian containing the coulomb interactions e?/rj; be- tween electrons and the (one-electron) contributions arising from the charges on the atomic cores of the magnetic ion and ligands. The Hamiltonian H also contains the kinetic energy operator, but spin-orbit coupling is ignored. The effects of covalency are included in the formalism by including an admixture of many excited states into the ground state. The excited states used in the construction of this admixture are related to the ground state by supposing that ligand electrons are excited into the open shell. They take the form Eqo = aa) Vase = (1/VN)ibabalegrXel, (1.5) where 7 denotes the ligand state vacated by the excited electron. The admixture of the excited states into the ground state can be deter- mined using perturbation theory or through the variational principle. It takes the general form Va = Yoo + WraarVanr, (16) ae where the admixture coefficients are given by Nasr apr = —F (1.7) Oar ,14 D. J. Newman and Betty Ng with numerator Nasr =[(VasrlH|Veo) — (Wasr| a0) Zao] % (Waar|Vase)/? (Wal Yao)” (1.8) and denominator (YoprlH|apr) De — Foo. 1.9) = aorlWaar) ‘The energy corresponding to Wa is given by is no 25 Mea oy Meet, aan Dew The last term in (1.10) appears because the open-shell orbital ¢q under consideration is already occupied, and is therefore not available for electron transfer. The non-orthogonality of the open-shell and ligand orbitals makes it necessary to include overlap integrals, such as (Wagr|Vao), in (1.8). When the magnetic ion is at a site of sufficiently high symmetry (e.g. cubic or axial), open-shell wavefunctions can be chosen so as to make the off-diagonal energy matrix elements, (Vq|H|Wg) (with 9 # a) identically zero. The crystal field is then determined entirely in terms of the energy differences Eq, — Eg. While there are many real systems with cubic symme- try, axial symmetry (Coop) occurs only when there is one ligand. Although it does not correspond to any real (crystalline) system, one-ligand systems provide a convenient way to determine the relative importance of the various crystal field contributions. Further simplifications result from noting that the energy denominators Dagr are effectively independent of the suffices a and f, and the many- electron matrix elements Nag, are effectively independent of the suffix a (see [New71]). We can therefore write Dagr = D;(all a, B) (1.11) and Nogr = Nar(all a). (1.12) Cancellation between terms then produces a remarkably simple expression for the crystal field splitting, viz. 2B = (Boot el )- (Go+E Meh). (1.13)1 Crystal field splitting mechanisms 15 Alternatively, the energy of an electron in orbital @ can be written as |Narl? Eq = Foo + 3) 5 + Bos (1.14) where the energy Eo is independent of a. It is now clear that the approxi- mations made in (1.11) and (1.12) correspond to making the covalency, or charge transfer, contributions dependent only on a single open-shell orbital. In other words, these approximations produce covalency contributions to a one-electron crystal field. 1.5 Crystal field energies in terms of one-electron matrix elements The formulae derived in Section 1.4 were expressed in terms of matrix ele- ments between many-electron states. The analysis required to express the previous results explicitly in terms of one-electron wavefunctions is rather tedious and can be found elsewhere [New71]: only results are given here. For simplicity, the system is taken to have an axial symmetry, with only one ligand in the +2 coordinate direction. In this case the one-electron orbitals on the magnetic ion which give only diagonal energy matrix ele- ments are the spherical harmonic functions, labelled I,m. In the following discussion, therefore, magnetic quantum number labels (m) will replace the Greek labels previously used for the one electron orbitals on the magnetic ion. The energy Em of an electron in the orbital 1m is composed of con- tributions from four distinct processes: electrostatic, including point and distributed sources; overlap; eachange and covalency. Symbolically Em = Em(electrostatic) + Ey, (overlap) + Em (exchange) + E,n(covalency). (1.15) In the present formulation, screening effects are omitted from the consid- eration but will be included in the numerical example given in Section 1.9. Electrostatic contributions to the crystal field energies are given by En (electrostatic) = (m|1/rr|m) + (2) (mr||mr)—8(m|1/rr|m)). (1.16) Here (mr||mr) stands for the coulomb matrix element (nXr|1/rizlémXz) between electrons 1 and 2 in the open shell and on the ligand, respectively. The term (m|1/rr|m) abbreviates the matrix element describing the po- tential energy of an open-shell electron (in orbital ¢m) in the electrostatic field produced by a point charge at ligand T. The distance between the16 D. J. Newman and Betty Ng electron and ligand T is denoted by rr. Similar abbreviations are used in the following equations. The bracketed expression on the right hand side of (1.16) is the differ- ence in electrostatic energy for eight electronic point charges at the origin of ligand T and that for the same eight charges in the outer ligand s and p orbitals. This expression measures the so-called charge penetration con- tribution arising from the distributed charge of the eight ligand outer-shell electrons. In evaluating the full determinantal expressions for Bao in equa- tion (1.14) an exchange contribution also appears, viz. Em(exchange) = — }~{mr||rm). (1.17) The fact that the ligand and open-shell orbitals are non-orthogonal produces the overlap contribution in (1.15), given by Ey (overlap) = — Sth) (Nine +2Nmz2); (1.18) where the expressions for Nnri are given in (1.21) and (1.22). The covalency contribution to the energy matrix has the closely related form Em(covalency) = “EE Nmnril ? De, (1.19) where the denominator is given by D, = €0— & +U* + U~ + ($6116) — (¢7ll6r) (1.20) Here ¢ refers to an arbitrary open-shell orbital; ¢9 and ¢, are, respectively, the Hartree-Fock energies of the open-shell and ligand electrons. The nu- merators in (1.19) have the three contributions Nmri = (m|r)(€0 — & +U* +U7 +1/R), (1.21) News = (23(mne"|[r7') — 8(m1/rr|r)) — (mrlirr), (1.22) and Nira = (n6||76) — (mir) @rllér)- (1.23) In these expressions 7’ and 7 are ligand orbitals. Note that ¢ without a suffix is used to denote the radial part of open-shell electronic wavefunctions in all cases where the inclusion of the angular part has a negligible effect on the value of the integrals. In the above equations R is the distance between the magnetic ion and ligand centres, U+, U~ are, respectively, the Madelung1 Crystal field splitting mechanisms 7 potentials at the magnetic ion and the ligand. The values chosen for these quantities must, of course, reflect a specific crystalline environment. Equations (1.16)-(1.23) involve only the outer-shell orbitals labelled 7 on a single ligand. What happens if several ligands contribute to the crys- tal field? According to the superposition principle for electrostatic poten- tials, both the point charge and the charge penetration contributions in Em(electrostatic) generalize to a sum over the ligands. Ey(exchange) in- volves only one ligand orbital and again generalizes to a sum over individual ligand contributions. The ligand orbital + which appears in E,, (overlap) and Ey,(covalency) refers to a specific ligand. Hence the only possibility for involving more than one ligand in these contributions arises from the bracketed expression for Ninrz. In this expression it is possible for both T and 7’ to refer to a ligand different to that which the orbitals 7 belong, giving rise to three-centre contributions involving inter-ligand charge pen- etration. However, such contributions are expected to be small because of the strong electrostatic repulsion between ligands. It is found, in practice, that the most important three-centre contributions occur in the next order of perturbation theory, and these are an order of magnitude smaller than the corresponding two-centre contributions [CN70]. Hence, it is a reason- able approximation to assume that all the crystal field contributions can be built from a superposition of contributions from the individual ligands. This provides the theoretical underpinning of the superposition model, which expresses the crystal field as a sum of single ligand contributions. Experi- mental evidence for the effectiveness of this model, and its development as a practical tool for the analysis of crystal field splittings will be given in Chapter 5. 1.6 Many-body approach to crystal field calculations The aim of the simple ab initio formalism described in Sections 1.4 and 1.5 was to identify the dominant contributions to the one-electron crystal field. If it is desired to obtain more accurate results, together with magnitudes of the two-electron and spin-dependent contributions, both the neglected terms in the Hamiltonian (e.g. spin-orbit coupling) and the unfilled higher energy states located on both the magnetic ion and the ligands must be taken into account. Diagrammatic many-body theory [LM86] provides a comprehensive ap- proach for including the contributions of all perturbations and excited states in an ab initio calculation of the energy matrix. This approach also pro- vides an appropriate formalism for the calculation of transition intensities18 D. J. Newman and Betty Ng (discussed in Chapter 10). However, the multicentre nature of the crystal field problem introduces special difficulties in applying many-body theory to the calculation of crystal field parameters. In particular, the standard for- malism needs to be generalized because realistic orbital basis sets contain occupied states on each of the centres, which are non-orthogonal. There are also problems in determining the appropriate excitation energies when electrons are transferred between ions, and in carrying out the summations of energy contributions over higher order perturbations. These issues have been discussed, and some calculations for the system Pr*+-CI~ have been carried out [NN87b, NN87a]. 1.7 Relation to other formalisms The analysis in Section 1.4 showed that the mechanisms considered all con- tribute predominantly to a one-electron crystal field. ‘This is in broad agree- ment with the experimental evidence. It is exploited in the phenomenologi- cal approach by combining all the contributions into a single effective energy operator, denoted Wor, which acts on one-electron open-shell states. The energies Em defined in equation (1.15) can then be expressed in terms of the diagonal one-electron matrix elements of this operator as em = Em = (m|Wor|m). (1.24) The lower case ‘e’ is introduced in this equation to emphasize the fact that Wer operates on one-electron states and defines one-electron energies. The exchange, overlap and covalency contributions to Wor have a simi- lar form to the ‘pseudo-potential’ operator, which has been employed in the theory of metals [Har66]. Attempts have been made to calculate the crys- tal field pseudo-potential directly from density functional theory. However, such calculations have required the introduction of a new type of (fitted) pa- rameters. We take the view that ‘hybrid’ theories of this type, which mix ab initio calculations and fitted parameters, do not advance our understanding of crystal field splitting mechanisms. In the simplified theory of crystal field splittings developed in Sections 1.4 and 1.5 it has been assumed that the symmetry of the system is sufficiently high that the open-shell orbitals can be chosen to give off-diagonal matrix elements of Wor which are identically zero. When this is not the case, the orbital energies alone cannot provide a complete description of the crystal field. It is then necessary to introduce a parametrization of the complete (21+ 1) by (27 + 1) energy matrix.1 Crystal field splitting mechanisms al 1.7.1 Crystal field parameters The numerical coefficients of the terms (r*)¥j,q(0, 4) in equation (1.2) are termed ‘erystal field parameters’. In order to obtain a clear understanding of the relationship between the energy matrix, the energy level splittings and the crystal field parameters it is best to start with the simplest model system: a single open-shell electron for which the spin-orbit coupling is ne- glected. Let us suppose that this electron has an orbital angular momentum 1. The standard approach is to use a linear expansion of the operator Wer, which generalizes the multipolar expansion of (1.2) by removing its explicit functional dependence on angular coordinates, viz. (amy |Wer|lma) = S(t |e ime) BE. (1.25) kq Here the BE are the crystal field parameters, m; are the magnetic quantum. numbers with values between —I and +1, a Mima? ma) are matrix elements of the one-electron tensor operators t\) (defined in Appendix 1). The crystal field parameters B} are, in general, complex quantities. ‘Tensor operators are employed in the crystal field expansion because of their orthogonality properties and the factorizability of their matrix ele- ments, viz. t kool Ck) Saas se scaeees Cia} = (yr™ (FT ) cv. (1.26) Here the 2x3 array is a 3/ symbol and (i||||!) denotes a so-called ‘reduced’ matrix element, which defines the normalization of the tensor operator t{"), As the notation implies, the reduced matrix elements are independent of the labels m1, ms and q. The definition and properties of tensor operators, together with the normalization adopted in this work and the allowed values of k and q, are given in Section A1.1.1 (also see [Jud63] and Chapter 4). Using (1.25) and (1.26), one-electron energy matrix elements of the crystal field may be written, in terms of crystal field parameters, as _ loko (im, |Wee|lms) = (-1)" Ea (. tn CI). 2.27) my, Given values of the crystal field parameters, the crystal field energies for a single open-shell electron, with no spin-orbit coupling, can be obtained from the eigenvalues of this energy matrix. The orthogonality properties of 3j symbols (e.g. see [CO80], p. 180) make it possible to invert equation (1.27) analytically, to obtain an expression for20 D. J. Newman and Betty Ng the crystal field parameters in terms of matrix elements, viz. eptim (bok oT ) -1 7 we, dy c= ——rC—s——esS—ee (1.28) In the case of axial symmetry, as was assumed in the theoretical discussion of Section 1.5, a very simple relationship exists between the orbital ener- gies and the azial crystal field parameters Bf. This is because the energy matrix is diagonal, reducing the relationships between the axial parameters and the | + 1 distinct single-electron energies ém(= €—m) to simple linear expressions. In order to obtain explicit numerical relationships it is neces- sary to adopt a specific normalization of the tensor operators. Using the Wybourne normalization, defined by equation (A1.2), the axial parameters B§ for f electrons can be shown to be given by BS = (1/7)(eo + 2e1 + 2€2 + 2€3), Bz = (5/14)(2e0 + 3e1 — 5es), a (1.29) BS (3/7)(3e0 + e1 — Teo + 3e3), B§ = (13/70)(10e9 — 15e; + 6e2 — €3). The inverse relations are 9 = BS + (4/15) B2 + (2/11) Bg + (100/429) BS, e1 = BY + (1/5)B3 + (1/33) Bg — (25/143) BS, i) 2 = B? — (7/33) Ba + (10/143) B, = BS — (1/3) B3 + (1/11) Bg — (5/429) Be. The corresponding relations between d-electron energies and axial crystal field parameters are BS = (1/5)(eo + 2€1 + 2e2), BB = (eo + €1 — 2€2), (1.31) Bé = (3/5)(3eo — 4er + 2), and eo = Bh + (2/7) BG + (2/7)B5, e1 = BS + (1/7) BB + (4/21) BS, (1.32) ex = Bh — (2/7) Bg + (1/21) BS. In both cases the number of (necessarily real) axial parameters is equal to. the number of diagonal one-electron energy matrix elements. As the mean1 Crystal field splitting mechanisms 21 energy, corresponding to B, is unobservable, the crystal field is represented by the energy differences e; — ¢9, etc. It can, in fact, be shown that there are always exactly the same number of crystal field parameters as there are independent one-electron matrix elements for any site symmetry. 1.7.2 Ligand field approach In the ‘ligand field’ approach crystal field splittings are seen as the result of overlap and covalent bonding between the open-shell electrons on the magnetic ion and the outer-shell electrons on the ligands. This can be expressed mathematically by constructing molecular orbitals from a linear combination of atomic orbitals on the two ions. Molecular orbitals take a particularly simple form for the axially symmet- ric one-ligand system. The modified ‘open-shell’ orbital is written Gm = (bm — AmrXz)/V1 — Amr (1.33) In this expression the ‘covalent mixing parameter’ \mr is given by Amr = mr + (mIr), with Nint mr = the numerator Nir = 30; Nmri and denominator D, being defined in equa- tions (1.20)-(1.23). The corresponding normalized molecular orbitals for the ligand outer-shell electrons are orthogonal to ¢j,,, and can be written X= (Xe + Imrdm)/ 1+ Zimrdmlr) + Be (1.34) Molecular orbitals provide a good approximation to the real one-electron states of the system. In particular, the energy expressions em = (Ginlh|¢m) can be brought into accord with Ey, (equation (1.15)) if an appropriate choice of the one-electron Hamiltonian h is made. Molecular orbitals can also be used to calculate the effects of covalent bonding on other observ- ables, such as g-factors. If, as is usually the case, an ab initio calculation is impracticable, molecular orbitals can be estimated from the phenomeno- logical parameters together with calculated overlap integrals, as described in Chapter 5.