Relativistic Effects in Atomic and Molecular Properties

acta physica slovaca vol. 60 No.
3, 259 391
June 2010
RELATIVISTIC EFFECTS IN ATOMIC AND MOLECULAR PROPERTIES

Miroslav Ilias,a,1 Vladimr Kello,b,2 Miroslav Urbanb,c,3
Department of Chemistry, Faculty of Natural Sciences, Matej Bel University, Tajovskeho 40,
SK974 00 Banska Bystrica, Slovakia
b
Department of Physical and Theoretical Chemistry, Faculty of Natural Sciences, Comenius
University, Mlynska dolina, SK842 15 Bratislava, Slovakia
c
Slovak University of Technology in Bratislava, Faculty of Materials Science and Technology
in Trnava, Institute of Materials Science, Bottova 25, SK917 24 Trnava, Slovakia.
a
Received 17 May 2010, accepted 1 June 2010
This paper is dedicated to the memory of our friend, prof. Andrzej J. Sadlej, with whom we had a
privilege and pleasure to collaborate for decades in many areas of theoretical and computational
chemistry. We are particularly indebted to Andrzej for introducing us to the area of relativistic
effects and their importance in atomic and molecular properties. Senior authors will remember
forever days and nights of joint work and discussions on chemistry, physics and life.
We present an overview of basic principles and methods of the relativistic quantum chemistry.
Practical aspects of different methods will be discussed stressing their capability of providing
accurate predictions of molecular properties, particularly in species containing a heavy metal
element. We will present a series of examples showing the importance of relativistic effects
in a variety of molecular properties including electron affinities, ionization potentials, reaction and dissociation energies, electric, spectroscopic and other properties. It is possible to
recognize a link between these properties and behaviour of materials in some cases. Particular attention is paid to relativistic calculations of the nuclear quadrupole moments for which
accurate theoretical electric field gradient is combined with data from the microwave spectra. Important aspect of the present paper is understanding of trends in electronically related
atoms throughout the Mendeleev Periodic Table rather than focusing on highly accurate numbers. We will show that relativistic effects represent an unavoidable instrument for explaining
some unexpected properties of heavy metal containing compounds. We will also discuss an
interplay between the manyelectron correlation and relativistic effects.
DOI: 10.2478/v10155-010-0003-1
PACS: 31.15.A-, 31.15.aj, 31.15.ap, 31.15.bw, 31.15.V-, 31.30.jc, 31.30.jp, 32.10.Hq
Relativistic effects, Electron correlation, Change of picture, Ionization

potentials, Electron affinities, Excitation energies, Dipole moments and
KEYWORDS:
polarizabilities, Electric field gradients, Nuclear quadrupole moments,
Multiple bonds, Intermolecular interactions, NMR.
The
reference [71] was corrected by the authors after the paper had been printed out.
address: [email protected]
2 E-mail address: [email protected]
3 E-mail address: [email protected]; [email protected]
1 E-mail
259
260
Relativistic effects in atomic and molecular properties
Contents
1
Introduction
262
Manifestation of relativistic effects on atomic properties. Basic notes.

264
2.1 Zdependence of the relativistic effects . . . . . . . . . . . . . . . . . . . . . . 269
2.1.1 Semiclassical estimate of spinorbit effects . . . . . . . . . . . . . . . 271
Basic notes on nonrelativistic electronic structure methods for manyelectron atoms

and molecules
274
3.1 Schrodinger equation and the hydrogenlike atoms . . . . . . . . . . . . . . . . 274
3.1.1 Angular momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
3.1.2 Spin momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
3.2 Many-electron systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
3.2.1 The independentelectron model: The HartreeFock method . . . . . . . 279
3.2.2 Algebraic approximation . . . . . . . . . . . . . . . . . . . . . . . . . . 281
3.2.3 Electron correlation in the nonrelativistic theory of many electron systems 282
3.2.4 Configuration interaction . . . . . . . . . . . . . . . . . . . . . . . . . . 283
3.2.5 Coupled Cluster theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
3.2.6 CASSCF, RASSCF, and CASPT2 methods . . . . . . . . . . . . . . . . 293
Relativistic theory of manyelectron atoms and molecules

4.1 Lorentz transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 KleinGordon equation . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3 Dirac equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.1 Nonrelativistic limit of the Dirac equation . . . . . . . . . . . . . .
4.3.2 Solution of the Dirac equation for the hydrogenlike atom . . . . .
4.3.3 Finite nucleus . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.4 Four component relativistic quantum chemistry . . . . . . . . . . . . . . .
4.4.1 The relativistic electronelectron interaction . . . . . . . . . . . . .
4.4.2 The DiracCoulombBreit/Gaunt Hamiltonian . . . . . . . . . . .
4.4.3 The DiracHartreeFock method . . . . . . . . . . . . . . . . . .
4.4.4 Symmetry aspects . . . . . . . . . . . . . . . . . . . . . . . . . .
4.4.5 Classification of relativistic electronic states of atoms and molecules
4.4.6 Fourcomponent relativistic basis set functions . . . . . . . . . . .
4.5 Relativistic correlation methods . . . . . . . . . . . . . . . . . . . . . . .
4.5.1 Secondorder Kramers restricted MP2 method . . . . . . . . . . .
4.5.2 Relativistic Coupled Cluster and Configuration Interaction methods
4.6 Transformations to twocomponent Hamiltonians . . . . . . . . . . . . . .
4.6.1 FoldyWouthuysen transformation . . . . . . . . . . . . . . . . . .
4.6.2 DouglasKroll transformation . . . . . . . . . . . . . . . . . . . .
4.6.3 Exact decoupling . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.4 Infinite order twocomponent Hamiltonian (IOTC) . . . . . . . . .
4.7 Spinorbit effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7.1 Meanfield spinorbit operator . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
296
296
297
297
299
301
303
304
304
305
306
307
308
309
310
311
312
312
313
314
315
318
320
322
CONTENTS
4.8
5
261
4.7.2 RASSISO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324

Fine effects, Lamb shift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
Relativistic effects in atoms and molecules, their properties and interactions

5.1 Ionization potentials, electron affinities and excitation energies . . . . . . . . .
5.1.1 Ionization potentials . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1.2 Electron affinities . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1.3 Excitation energies . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2 Relativistic effects on molecular geometries . . . . . . . . . . . . . . . . . . .
5.3 Electric properties dipole moments and dipole polarizabilities . . . . . . . . .
5.4 Determination of nuclear quadrupole moments . . . . . . . . . . . . . . . . .
5.4.1 Change of picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.4.2 Examples of calculations . . . . . . . . . . . . . . . . . . . . . . . . .
5.5 Dissociation and reaction energies . . . . . . . . . . . . . . . . . . . . . . . .
5.5.1 Quadruple, quintuple or sextuple bonds? . . . . . . . . . . . . . . . . .
5.6 Intermolecular interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.7 NMR properties of molecules . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.7.1 Fourcomponent calculations of NMR shielding constants . . . . . . .
5.7.2 Gauge including atomic orbitals in the fourcomponent theory . . . . .
5.7.3 Electron correlation in NMR calculations . . . . . . . . . . . . . . . .
5.8 From relativistic effects in atoms and molecules to gasphase properties and the
condensed phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Conclusions
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
326
328
328
329
333
336
338
345
346
348
353
358
359
368
368
370
370
. 372
376
Acknowledgment
376
Appendix
377
References
378
262
Introduction
Until the seventies of the 20th century it was generally accepted that for a description of the electronic structure of atoms and molecules and, therefore, for the whole chemistry and for the substantial part of physics, relativistic theory is not needed. According to Sheldon L. Glashow [1],
Nobel Prize Winner for Physics, 1979, Modern elementaryparticle physics is founded upon the
two pillars of quantum mechanics and relativity. I have made little mention of relativity so far
because, while the atom is very much a quantum system, it is not very relativistic at all. Relativity
becomes important only when velocities become comparable to the speed of light. Electrons in
atoms move rather slowly, at a mere of one percent of light speed. Thus it is that a satisfactory description of the atom can be obtained without Einsteins revolutionary theory. Possibly
these beliefs were initiated by Dirac himself [2], who wrote in 1929 that Relativistic effects are
therefore of no importance in the consideration of atomic and molecular structure and ordinary
chemical reactions. In the last thirty years the situation has changed considerably. Presently, it
is generally accepted that relativistic effects must be considered in compounds containing heavy
metal elements. Even understanding of the Mendeleev Periodic Table is impossible without utilizing the ideas of theory of relativity along, of course, with quantum physics. When talking
about the relativistic theory of atoms and molecules we mean, of course, the special theory of
relativity discovered by Einstein in 1905. It was in the same year in which he published the
theory of the photoelectric effect, which is one of pillars on which the quantum mechanics
was founded few years later. Therefore, both fundamental theories of the electronic structure of
atoms and molecules were discovered and developed about at the same time, in first few years
of the twenties century. Quantum physics was accepted as a basic theory for understanding of
atomic and molecular properties almost immediately after its discovery. The first theory of the
chemical bond formulated by Heitler and London in 1927 [3], is considered as the birthday
of quantum chemistry. Theory of relativity, however, was awaiting for recognition of its fundamental importance in molecular physics and chemistry much longer. It took several decades
after Diracs formulation of his fundamental relativistic quantum mechanics in 1928 [4] until his
theory was developed as a tool for the treatment of manyelectron molecular systems. Currently,
it is generally accepted that both, quantum mechanics and the special theory of relativity, are
essential in the description and understanding of molecular properties. Clearly, relativity is increasingly important for molecules containing an element with high atomic number. However,
even accurate description of properties of the lightest hydrogen molecule, as performed by Koos
and Wolniewicz in 1961 [5] showed that careful treatment of relativistic effects is unavoidable.
Solutions of the DiracFock equations [6, 7] for manyelectron systems with atomic numbers up
to 120 was presented by Desclaux in 1973 [8]. An important milestone on a long way towards
recognizing the importance of the relativistic quantum theory in understanding general trends of
molecular properties is a paper entitled Relativity and the Periodic System of elements published by Pyykko and Desclaux in 1979 [9]. This work was followed by many other reviews
on the development of the relativistic quantum theory and its applications in quantum molecular
sciences by Pyykko, Kutzelnigg, Schwerdtfeger, Schwarz and other authors [1022], to name at
least a few. A further development of relativistic quantum chemistry in molecular sciences was
greatly affected by excellent books published recently, particularly Relativistic electronic structure theory, Part 1. Fundamentals and Part 2. Applications, edited by Schwerdtfeger [23, 24]
and books by Dyall and Fgri [25] and by Reiher and Wolf [26].
Introduction
263
The purpose of this review is a short description of basic nonrelativistic theories, including
most important methods for treating the manyelectron correlation problem. An essential part of
this paper is devoted to a transparent overview of relativistic manyelectron theories in which the
electron correlation problem (very difficult to treat accurately even at the nonrelativistic level)
will be considered at the relativistic level. Rigorous fourcomponent relativistic manyelectron
calculations for larger molecules are hardly tractable in the spirit of the fourcomponent Dirac
relativistic quantum mechanics [4,27]. Approximate extensions applicable to manyelectron systems including electron correlation are feasible but very tedious. Methodologies which give rise
to a variety of twocomponent Hamiltonians and allow treating larger molecular systems will be
discussed in more detail in Section 4 and specifically in Section 4.6. For practical purposes it is
important to note that in many applications it is sufficient to consider relativistic effects even at
the nopair onecomponent level. This is frequently denoted as scalar relativistic level. This
is a pragmatic attitude toward the manyelectron correlated relativistic treatment of molecules.
Great advantage is that scalar relativistic effects can be taken into account within common nonrelativistic manyelectron wave function methods, with only a small modification. One and
twocomponent methods are frequently called also as quasirelativistic. Spinorbit (SO) effects are neglected within the onecomponent framework but applications may go to as large
molecules as it is possible in the nonrelativistic case, still using sophisticated manyelectron
wave function methods. We will demonstrate this approach in many applications to problems in
which spinorbit effects are unimportant. A large part of relativistic quantum chemistry calculations is performed using the density functional theory (DFT) which is applicable to truly large
systems, although the control of accuracy is to some extent questionable.
In comparison with scalar relativistic effects, rigorous fourcomponent manyelectron relativistic theories employing sophisticated treatment of the electron correlation, like relativistic
Coupled Cluster (CC) theories, are very tedious in most cases. Nevertheless, there is a large
progress in making these theories (or their theoretically well founded approximations) applicable
to a larger variety of systems of chemical and physical importance. We will pay attention to these
theories and to their transformation to twocomponent treatments which are better applicable to
larger manyelectron calculations still allowing sufficiently accurate treatment of spinorbit effects.
Somewhat out of the main scope of this review are fine structure problems, e.g., spectra
of atoms and molecules, Lamb shift, Quantum Electron Dynamics (QED) effects to which pay
large attention some experts in relativistic many electron theories. Of course, there exists a rich
literature on QED effects in simple systems. Less is known about the importance of these effects
on properties and reactivity of manyelectron atoms and molecules. Presently, some results
indicate that QED effects may eventually lead to interesting chemical and physical consequences
but the knowledge in this area still remains quite limited [28, 29].
264
Historically, a breakdown in understanding of the chemical and physical behaviour of elements

and their compound goes back to the formulation of the periodic law and the Periodic Table
of elements by D. I. Mendeleev in 1869 [30]. The impact of his discovery on chemistry and
molecular physics can be hardly exaggerated. As the greatest (we believe) theoretical chemist
in the history, Mendeleev was able to systematise the accumulated knowledge, to interpret the
facts and to predict chemical properties of elements and their compounds. Yet, he has failed in
placing some elements at proper positions. As we can see in Fig. 2.1, in which we reproduce the
Mendeleev Periodic Table from his 1869 paper published in Germany [31] which was abstracted
from the original Russian paper [30], most elements have proper location. It corresponds to our
present knowledge of the isoelectronic valence electronic structure of elements belonging to the
same group in the Periodic Table. As we can see, incorrect location concerns in fact exclusively
Fig. 2.1. Mendeleev Periodic Table as published in 1869, Ref. [31].
265
elements with a high atomic weight, or, as we are saying nowadays, with a high atomic number.
Most striking examples are mercury, gold, lead, or thallium. Mendeleev was able to recognize
that atomic weights of some elements (one of ingredients on which his Table is founded) are
incorrect. His knowledge about the chemical and physical behaviour of elements and their compounds was essential but what is really admirable, is his fascinating intuition. All this did not
help him in properly placing some heavy elements in the Table. Presently, the reason is well
known. It is relativity which changes chemical and physical behaviour of atoms and molecules
substantially. Throughout the Periodic Table there are some irregularities which were difficult to
explain before the era of an extended exploitation of the relativistic quantum mechanics in the
theory of atoms and molecules. We demonstrate some of these irregularities in Tab. 2.1.
Very transparent example is a behaviour of the coinage elements, Cu, Ag, and Au. In the
seventieth, we could find in textbooks (see, e.g., Ref. [34]) a description of their properties as
...quite different in spite of the same valence electron structure, (n-1)s2 (n-1)p6 (n-1)d10 ns1 . We
have no reasonable explanation of this behaviour.... Concerning copper, it is characterized by
much smaller energy gap between the 3d10 and 4s1 orbitals than is the analogous energy gap
between 4d10 and 5s1 orbitals in Ag. Cu behaves like a first group transition metal and exhibits
larger electron correlation effects than Ag. Different chemical and physical properties of Au and
Ag result mainly from large relativistic effects in Au. Many properties, even the yellow colour
of the metallic gold, are caused by relativistic effects. If relativistic effects were negligible or
absent, a saying a nonrelativistic gold would be a silver is quite appropriate. One example are
large relativistic effects in gold leading to irregularities or Vshaped pattern of ionization
potentials (IPs) in the series Cu, Ag, and Au [33], as shown in Fig. 2.2. It is well known that
atomic ionization potentials vary within the Periodic Table according to their valence electronic
structure. For a group of atoms characterized by the same valence electrons, like (n-1)s2 (n1)p6 (n-1)d10 ns1 in Cu, Ag, and Au one expects that ionization of the ns1 electron would be
easier with increasing atomic number. That means, the expected sequence of IPs should be
Cu<Ag<Au. Experimental data in Tab. 2.1 show that IP of Ag is lower than IP of Cu, as
expected. However, on the contrary to our expectations, IP of gold is the largest within the
family of coinage metals. This experimental finding can be understood by employing three
approximations with gradually improved theoretical levels for calculating ionization potentials,
as demonstrated in Fig. 2.2. The lowest theoretical level is the nonrelativistic Restricted Open
Tab. 2.1. Common anomalies in the periodic system, coinage metal elements. Data from Ref. [32].
Property
Melting point [ C]
Boiling point [ C]
Electron affinity [eV]
Ionization potential [eV]
Standard specific electrical resistivity [108 m]
Polarizabilitya [a.u.]
a
Ref. [33]
Cu
Ag
Au
1085
2562
1.235
7.726
1.712
46.5
962
2162
1.302
7.576
1.617
52.46
1064
2856
2.309
9.225
2.255
36.06
266
Fig. 2.2. Electron correlation and relativistic effects in ionization energies (eV) of the coinage elements Cu,
Ag, and Au. The ROHF nonrelativistic data represent calculations using a single determinant Restricted
Open shell Hartree Fock calculation. Electron correlation is treated by the Coupled Cluster CCSD(T)
method and relativistic effects are calculated by the nopair spinfree DouglasKrollHess method. Data
from Ref. [33].
Shell HartreeFock oneelectron calculation (ROHF), which treats interelectronic interactions

only approximatively. This model is completely insufficient in describing IPs of any of the three
coinage metals. The error is larger than 1 eV. Clearly, electron correlation effects are inevitable.
Sophisticated description of the electron correlation provides the Coupled Cluster CCSD(T)
method in which amplitudes of the single and double excitations with respect to the single determinant ROHF wave function are treated iteratively. Computationally demanding triple excitations employ the resulting CCSD amplitudes perturbatively in a single noniterative step. Normally, this is an excellent manyelectron model [3540] capable of interpreting and predicting
atomic and molecular properties very accurately, see Section 3.2.5. The difference between nonrelativistic ROHF and CCSD(T) results is largest for Cu, followed by Au. Nevertheless electron
correlation effects, as represented by the difference between ROHF and CCSD(T) calculations,
are similar for all three valence isoelectronic coinage metals. IP of Cu calculated using the
nonrelativistic CCSD(T) method agrees with experiment reasonably well. When relativistic effects are neglected for Ag, the CCSD(T) result deviates from experiment considerably, by about
0.5 eV, while for Au is the nonrelativistic result completely misleading. Considering the nopair
onecomponent scalar relativistic approximation, used in calculations presented in Fig. 2.2 is
quite satisfactory for reproducing and predicting ionization potentials of Cu, Ag, Au, and many
other atoms and molecules. This approximation works quite well for states that are not affected
267
by other relativistic effects, particularly spinorbit effects, significantly. The electronic state of
Cu, Ag, and Au and their ions is 2 S and 1 S (a closedshell), respectively, which are not split
into spinorbit components. It is good to remind already now, that SO effects may contribute
even for the closedshell singlet or some doublet states (like 2 S states). Although these states do
not split by SO effects, their corevalence d and f , etc. orbitals do [8]. When working within
true fourcomponent theories, valence sorbitals (which have obviously only one SO component) feel different field from different SO components of orbitals with higher orbital momenta.
This effect may contribute to atomic or molecular properties, when accurate final results are
required. Calculations as presented in Fig. 2.2 represent a scheme which will be used in this
paper quite frequently. In order to recognize the importance of relativistic effects we will start
our treatment with the nonrelativistic Hamiltonian and the nonrelativistic HartreeFock wave
function or, alternatively, using a sophisticated wave function in which electron correlation is
0 Hamiltonian is then supplemented by a relativistic
considered reasonably accurately. This H
relativistic . As we will show later, there are numerous approximations to the relativistic
term, H
Hamiltonian. Occasionally we can consider in the first step scalar relativistic effects corrected
subsequently by SO or higher relativistic effects. We note that the electron correlation, relativistic, and SO contributions are not independent and, therefore, are not additive. We also stress, that
when talking about relativistic effects, atoms and molecules are neither nonrelativistic or relativistic. Relativistic effects are present in all species, irrespective whether participating atoms have
low or very high atomic numbers. The difference is only the magnitude of relativistic effects
related to the accuracy which is required for a property under consideration. Therefore, when
talking about nonrelativistic or relativistic atoms and molecules and their theoretical description, we have in mind just different models representing their Hamiltonian and the corresponding
wave function. In the nonrelativistic representation we refer to the world where the speed of light
c would be infinite. In reality, it is finite and this is the real world of atoms and molecules.
Returning now to Fig. 2.2, the enhancement of IP of gold (and to a lesser extent also of Cu
and Ag) can be rationalized in quite transparent terms. A summary of basic effects encountered
in the real world of the finite speed of light has been formulated by Pyykko [9]. The three
most frequently occurring relativistic effects are the relativistic shrinking and stabilization of s
orbitals (and to a lesser extent also p orbitals), the spinorbit splitting of p, d, etc. orbitals, and
third, the relativistic selfconsistent expansion and destabilization of d and f orbitals. Electrons
in d and f subshells are far from the nucleus and their velocities are much lower than the speed
of light. Therefore, the last effect, relativistic radial expansion and destabilization of d and f
electronic shells is indirect and follows from relativistically affected screening of nucleus by s
and p electrons. Since their radial distribution shrinks, the d and f subshells expand. Now, since
inner shell 1s electrons of Au move near the nucleus with the velocity comparable to the speed of
light, these electrons are relativistically stabilized and corresponding orbitals shrink. In many
electron systems all electrons up to valence electrons feel this effect, being stabilized and
shrunk as well. Consequently, removing an electron from the valence 6s orbital is hindered by
this stabilization and leads to an enhanced ionization potential. The same mechanism explains
a relativistically enhanced electron affinity (EA) of coinage metals and particularly of Au, so
that Au has a larger electron affinity than other coinage metals, see Tab. 2.1. The relativistic
modification of the shape of the valence electrons of Au and particularly shrinking of the 6s
orbital affects also its dipole polarizability. Consequently, the polarizability of this atom is lower
than is the polarizability of Cu or Ag, in spite of a higher atomic number and the expected
268
Fig. 2.3. Nonrelativistic and relativistic HartreeFock energies of outer orbitals of Hg. Data from Ref. [41].
largest volume of Au. This has important consequences in intermolecular interactions with a
participation of a heavy metal element. How relativistic effects affect atomic orbital energies of,
e.g., the mercury atom is schematically shown in Fig. 2.3. This scheme follows from pioneering
DiracFock calculations of Desclaux and Kim [41] and is quite general.
Note that the energy gap between the valence 6s and 5d orbitals diminishes significantly.
At the same time, 5p and 5s orbitals are relativistically stabilized and 4f orbitals destabilized.
The consequence is the reversed sequence of some components of 4f and 5s orbitals. Clearly,
relativistic stabilization/destabilization of orbitals may affect the energy differences between the
ground and excited states of atoms and molecules and alter in this way their spectra and other
properties. The scheme as presented in Fig. 2.3 is quite useful for qualitative discussion of ionization potentials and electron affinities of the coinage elements which have very simple valence
electronic structure. A dominating role in processes which determine IP and EA of Cu, Ag, and
Au, play ns electrons. Stabilization of these orbitals is quite general so that qualitative assessment of relativistic effects in such species is simple.
More complicated is the situation when there are partly occupied valence p, d, etc. shells.
These shells split up due to spinorbit effects and resulting different components behave differently when considering relativistic effects. Note, e.g., p1/2 orbitals shrink and are stabilized
more than p3/2 orbitals. Noteworthy is the Z-dependence of the orbital shifts. The stabilization
269
Fig. 2.4. Nonrelativistic and relativistic valence(n-1)d and ns orbital energies for the mercury element and
the superheavy element 112 derived from HF and DHF calculations. Data from Ref. [41] (Hg) and Ref. [17]
(the 112 element).
of the 7s orbital and the destabilization of the 6d5/2 orbital of the superheavy element with the
atomic number 112 is so strong that the 7s orbital lies between the two 6d components, d5/2
and the lower d3/2 component [17]. A schematic comparison of the valence orbitals stabilization/destabilization of Hg and its valence isoelectronic analogue, the element 112, is shown in
Fig. 2.4. Theoretical methods of relativistic quantum chemistry are capable of treating and predicting electronic properties of superheavy elements. One should take care about the valence
electronic structure since for heavy and superheavy elements the valence electronic structure of
orbitals frequently differs from our expectation based on the experience with lighter elements.
Details of the (valence) electronic structure of all species which participate at processes under
consideration, including the electron correlation and spinorbit effects must be carefully considered in qualitative estimates of relativistic effects in atomic and molecular properties. Since
relativistic and electron correlation effects are not additive, they can not be treated separately.
Particularly demanding are qualitative considerations and accurate theoretical calculations of
systems in which participate transition metal elements.
The orbital pattern of the halogen molecules split due to relativistic effects is schematically
shown in Fig. 2.5.
2.1
Zdependence of the relativistic effects
Relativistic effects can be expressed as the difference between the corresponding relativistic and
the nonrelativistic description of a quantum chemical system. This difference is just the consequence of applying different physical models. In computational practice we asses which theoret-
270
Fig. 2.5. Nonrelativistic and relativistic orbital scheme for halogen dimer molecules.
ical model is more suitable for describing studied chemical system.

Dirac himself at the time of the publishing his famous relativistic wave equation did not foresee the importance of his theory for chemical systems. He wrote [2] The general theory of
quantum mechanics is now almost complete, the imperfections that still remain being in connection with the exact fitting in of the theory with relativity ideas. This give rise to difficulties only
when highspeed particles are involved, and therefore of no importance in the consideration of
atomic and molecular structure and ordinary chemical reactions, in which it is, indeed, usually
sufficiently accurate if one neglects relativity variation of mass with velocity...
Before proceeding to rigorous relativistic quantum theory whose foundations are credited to
Dirac himself, it is worth to present few semiquantitative arguments in order to shed some light
on the importance of relativistic effects for the electronic structure.
First, let us have a closer look at the Bohr radius meaning extension, or characteristic size
of the atomic orbital what is dealt in Bohrs model [42],
a0 =
n2 40 h2
,
mZe2
(2.1)
where n is the principal quantum number, Z is the atomic number, m is the mass of the particle,
and the rest are common constants. Next, as was promoted by Einstein [43], the inert mass of
any fastmoving particle, including electron, is increasing with its speed as
m= q
m0
1
,
v2
c2
(2.2)
271
where m0 is the rest mass, v and c are velocities of the particle and of the light. For small
velocity v remains m very close to m0 , but as v is increasing, the relativistic mass m goes to
infinity. The speed of light, c, may be derived from Maxwells equations [44] and in SI units
has the form of c2 = 1/0 0 with 0 and 0 being the vacuum permeability and permittivity,
respectively. The exact value of the speed of light in vacuum can be found in the latest set of
CODATA recommended values of fundamental physical constants [45]. The relativistic mass
enhancement, Eq. 2.2, has profound influence on the electronic structure. For the simple, semi
quantitative demonstration, let us assume 1s electron in an atom with the nuclear charge Z.
According to the Bohrs simple model [42], by solving mvr = nh, inserting Eq.2.1 for r and by
considering atomic units, a.u., convenient for quantum chemistry (e = m0 = h
= 40 = 1, see
Appendix 6) the speed of the electron is then
v1s = Z.
(2.3)
It is evident from Eq. 2.3 that in heavy elements inner shell electrons move with a speed comparable with the speed of light (c ' 137.036 a.u.), the velocity being proportional to the atomic
number Z and therefore the relativistic description particularly of heavy elements is necessary.
Relativistic mass appearing in the denominator of Eq.2.1 shrinks core orbitals. For instance, the 1s electron of the Rn atom (Z = 86) has the ratio v/c (for v we apply Eq. 2.3)
of 86/137 ' 0.63, implying the radial shrinking approximately by 22%. It is important to realize that relativistic effects do not concern only fast moving inner shell electrons. Real atoms
and molecules are manyelectron systems. The valence electrons feel relativistically altered
average field of inner shell electrons. The complete picture is a bit complicated. Nevertheless a
simplified view tells us that the higher s shells are also contracted as these must be orthogonal to
lower ones.
Note that the above mentioned semiquantitative arguments for importance of relativistic
effects are based upon very primitive atomic model. Also, present considerations concern just
radial distribution of orbitals. The dependence of relativistic effects in bond energies and other
properties [19] on the atomic number Z is more complicated. Following perturbational arguments, Schwarz et al. [19] arrived at the conclusion that for hydrogenlike states relativistic
corrections to valence properties scale with the nuclear charge as Z 2 2 , where
1
(2.4)
= .
c
Relativistic alteration in the electron correlation contribution is also proportional approximately
to Z 2 2 . Clearly, relativistic effects are very complex for manyelectron systems. Note that
methods of quantum mechanics which suppose the finite speed of light are called relativistic
and those in which the infinite speed of light is supposed are called nonrelativistic. Relativistic
effects are defined as the difference between results following from these two approaches. We
stress, again, that atoms and molecules are in fact governed by laws of the relativistic quantum
mechanics. When talking about relativistic effects we have in mind just a model which is applied
to a specific problem.
2.1.1
Semiclassical estimate of spinorbit effects
The spinorbit (SO) coupling is one of the most common manifestation of relativistic effects in
molecular sciences. Individual orbital and spin quantum numbers are not good quantum numbers
272
any more since the orbital and spin operators do not commute with the Hamiltonian (only the
total angular momentum does). We note that the notion of the spin itself is a consequence of
the relativistic character of electrons. Here we present a qualitative introduction of the coupling
between the spin and the orbital angular momentum. A simple demonstration of the SO coupling
can be obtained employing the classical expression for the interaction of the spin of a single
electron and the magnetic field which has the source in its orbital motion. An electron moving
around a nucleus with a speed v is producing the classical magnetic field, B:
B=E
v
,
c2
(2.5)
where the electric field, E, for a central potential V (r) is given by

E=
1 V
r.
r r
(2.6)
The intrinsic magnetic moment of the electron is (in a.u.)

= gs B s,
(2.7)
where s is the spin moment, see Section 3.1.2. The term

B =
e
h
2me
(2.8)
is the Bohr magneton and gs ' 2 is the electron spin gfactor. Its precise value is derived from
quantum electrodynamics. Now, the interaction energy operator of the magnetic moment of the
electron, Eq. 2.7, with the magnetic field, Eq. 2.5, can be expressed as (employing a.u.):
HSO = B =
B V
(l s).
c2 r r
(2.9)
l is the angular momentum of the electron, see Section 3.1.1. The product l s can be written
in the operator form utilizing the definition of the square of the total momentum valid for the
hydrogenlike systems, j2 = (l + s)2 . Therefore, the product l s can be written in the operator
form as
l s = 1 (j 2 l2 s2 ),
2
(2.10)
which allows to estimate SOsplittings by using quantum numbers j, l and s, related to j 2 , l2

and s2 operators in Eq. 2.10, see Sections 3.1.1 and 3.1.2.
The Hamiltonian of a hydrogenlike atom with the central field potential V (r) = Z/r can be
=H
0 + W (r) l s. For the
written as the zeroorder Hamiltonian and the SO perturbation, H
hydrogen atom the first order perturbation contribution due to W (r) l s can be calculated employing analytical expressions for the unperturbed Hatom radial and spherical wave functions
(for more details and a very instructive analysis, see, e.g., [46]). Qualitatively, for hydrogen
like atoms W (r) is proportional to Z/r3 . Since 1/r3 is proportional to Z 3 , the SO splitting in
hydrogenlike atoms scales as Z 4 (see also Eq. 2.9). For the 2p orbital of H, He+ , Li2+ , etc.
273
Fig. 2.6. Spinorbit splitting of the p orbital of hydrogenlike atom into p1/2 and p3/2 states.
is the ratio, HSO /Z 4 constant, 0.3652 cm1 . Considering the eigenvalues of j 2 , l2 and s2
operators in Eq. 2.10, i.e. j(j + 1), l(l + 1), and s(s + 1) in a.u., it is clear that sorbitals do not
split. The 2p orbital splits into 2p1/2 and 2p3/2 states, the ratio of the energy shift with respect
to the unperturbed energy being -2:1, Fig. 2.6.
The SO splitting in manyelectron atoms can be approximated as the sum of the one electron
operators,
X
=
W
(ri )li si ,
(2.11)
i
with (ri ) taken as an analogue of W (r) above. The total angular momentum J for lighter atoms
is frequently considered by a LS coupling, and the first order corrections due to the SO perturbation are calculated as the eigenvalues of the matrix within the degenerate subspace. At this level
of approximation we employ (LS) regarded as parameters to be determined by experiment.
A rigorous treatment of the SO coupling requires relativistic approach and will be described
in Section 4.7.
274
Basic notes on nonrelativistic electronic structure methods for manyelectron atoms

and molecules
Proper description of relativistic effects in atoms and molecules and a quantitative (or frequently
even qualitative) account of their importance in atomic and molecular properties require the treatment of manyelectron correlation methods at the relativistic level. One should realize that the
reasonably accurate quantum mechanical consideration of manyatomic and manyelectron systems is tremendously difficult even at the nonrelativistic level. The analytical solutions of the
Schrodinger equation in the nonrelativistic case, just as the solutions of the Dirac equation in
the relativistic case, are possible only for one electron hydrogenlike systems. For description
of properties of all other atoms and molecules we are faced with the manyparticle problem for
which solutions are necessarily only approximate. Before proceeding into the description of the
relativistic manyelectron quantum theory it may be useful to recall the solutions of the simplest
hydrogen atom at the nonrelativistic level. Quantum numbers which follow from the solutions of
the Dirac equation at the relativistic level are related to the quantum numbers of the solutions of
the Schrodinger equation. We note that the energy spectra in the nonrelativistic and relativistic
case, respectively, are different. Further on, we wish to draw attention to the relation between the
eigenvalues and eigenfunctions of the basic angular and the spin momentum operators for one
electron systems. More difficult is the quantum mechanics of the manyelectron atoms and molecules at both the nonrelativistic and relativistic level. The description of the relativistic methods
for manyelectron systems in which the so called electron correlation is treated approximately
will be a main body of our theoretical overview and should help the reader in understanding the
limits of accuracy attainable by present day electron correlated relativistic approaches.
3.1
Schrodinger equation and the hydrogenlike atoms
The Schrodinger equation, published in 1926, is a fundamental equation of the nonrelativistic

quantum physics [47]. It is a general equation describing the behaviour of not only atoms and
molecules but is fundamental for describing all systems in the microworld as well. The quantum
mechanical (and relativistic) behaviour of atoms and molecules affects also properties of systems
at the macroscopic level and leads eventually to many present days technologies.
For a system exposed to an external potential V (r) the timedependent Schrodinger equation
(from this point we are using a.u., that is h omitted) is
i
n (r, t) = ( 1 + V (r))n (r, t),

n (r, t) = H
t
2
(3.1)
where r = r(x, y, z) is the position of the particle in the 3dimensional space, is the Laplace
operator, V (r) is the potential energy operator of the particle at a given position r and n (r, t) is
the timedependent wave function representing a system in the quantum state n. In the theory of
atoms and molecules in which the Hamiltonian frequently does not depend explicitly on time the
Schrodinger equation can be transformed into the timeindependent form describing stationary
states:
n (r) = En (r).
H
(3.2)
Basic notes on nonrelativistic electronic structure methods
275
Exact solutions of the Schrodinger equation for oneelectron hydrogenlike atoms is notoriously known topic in standard textbooks of quantum physics. The Schrodinger equation is first
transformed into spherical coordinates, which allows its separation into radial and angular parts.
Oneelectron wave functions, the product of the radial Rnl (r) and the Ylm (, ) angular parts,
are referred to as hydrogenlike atomic functions:
nlm (r, , ) = Rnl (r)Ylm (, ).
(3.3)
Integers n, l, m are quantum numbers which characterize the quantum state of the system. The
number n is the principal quantum number (n can have values 1, 2...) and determines the nonrelativistic energy of the hydrogenlike atom with the atomic number Z as En = Z/2n2 a.u. The
angular part of the wave function, Ylm (, ), is represented by the spherical harmonic functions.
Since for the hydrogenlike atoms the nonrelativistic Hamiltonian commutes with the square of
the angular momentum operator l2 and both commute with its zcomponent, the lz operator, the
spherical harmonic functions Ylm (, ) are eigenfunctions of all three fundamental operators.
3.1.1
Angular momentum
Eigenvalues (in a.u.) and eigenfunctions of the square of the angular momentum operator, l2 ,
l2 = l2 + l2 + l2 , are determined by the equation (in a.u.)
z
y
x
l2 Ylm = l(l + 1)Ylm .
(3.4)
The zcomponent l of the angular momentum is quantized as

lz Ylm = mYlm .
(3.5)
The angular quantum number l, with l = 0, 1, 2, ...(n 1) determines the magnitude of the angular momentum. The magnetic quantum number m (more frequently denoted as ml ) can have
values l, (l + 1), ..., (l 1), l and determines the projection of the angular momentum into
the (arbitrarily chosen) zaxis. The nonrelativistic hydrogenlike orbitals nlm are traditional
building stones (serving as basis sets) in the nonrelativistic theory of the chemical bond, particularly in the theory of molecular orbitals and in many other areas (note, nevertheless, that in
modern quantum chemistry we are using modified and more general basis sets, mostly Gaussian
basis sets). The radial and angular distributions of the eigenfunctions of the hydrogen atom, say,
s, p, d, etc. orbitals and their density distributions are very well known to students of physics or
chemistry and need not be reproduced here. We note, however, that not only the energy spectrum of the hydrogenlike systems but also the shape of their wave functions and orbitals in the
nonrelativistic and relativistic theory are different. Later on (Section 4) we will show that nonrelativistic electronic orbitals are frequently unsatisfactory for modelling of chemical bonds for
systems containing heavy elements and that they are to be replaced by relativistically obtained
wave functions.
3.1.2
Spin momentum
The electron, the proton and other particles as well, posses an intrinsic angular momentum.
First direct observation of the electrons intrinsic angular momentum was achieved in the Stern
Gerlach experiment [48]. Another experimental evidence are closely spaced doublets in the
276
hydrogen spectrum examined at a very high resolution, what is known as the finestructure splitting.
The intrinsic momentum of the electron is denoted as the spin 1/2 [49] and is established
as the fourth quantum number in the nonrelativistic framework. With some imagination it could
be approximately compared to a ball spinning around its own axis.
Regarding the pioneering work leading to the discovery of the spin of the electron, nice stories are described in the book by Pisu t and Zajac [50], see also [51]. It was Wolfgang Pauli who
first noticed that there must be a fourth quantum number which is related to the twovaluedness
behaviour of an electron. He has initiated a new theory for the explanation of the doubling of
the spectral lines related to transitions of the valence electron of the sodium atom. The theory of the electron spin was published by two young scientists, George Uhlenbeck and Samuel
Goudsmit, prompted by another distinguished theoretician, Paul Ehrenfest. Initially, Uhlenbeck
and Goudsmit did not trust their theory and wanted to withdraw their German written paper [52].
However, in the meantime, Ehrenfest has forwarded their paper to Naturwissenschaften accompanied by his supporting comment. An English written paper on spin by these two pioneers,
Ref. [49], followed approximately one year later.
If one measures the component of the spin momentum of a single electron along a selected
direction one finds the value of h2 (in a.u. just 21 ). Note that the angular orbital momentum
mentioned before can have only integer quantum numbers in contrast to the spin eigenvalues, see
Eqs. 3.4, 3.5. Further on, spinning distribution of a charge has a magnetic moment.
Similarly as it is expressed for the angular momentum in Eq. 3.4, the square of the spin
operator acts on a spin function as (in a.u.)
s2 = s(s + 1) , = (, ).
The spin of a single electron is associated with Pauli spin matrices [53] :

0 1
0 i
1 0
=
;
;
.
1 0
i 0
0 1
(3.6)
(3.7)
The Pauli matrices,

x,y,z , are complex Hermitian and unitary. Except for a factor of 1/2 they
can be viewed as representations of the sx,y,z spin operators, where and spin functions are
taken as column vectors.

1
1
1
1
1
0
0
sz =
z ; sz = sz
=
; sz = sz
. (3.8)
=
0
0
1
1
2
2
2
The and the are eigenfunctions of the sz operator (for comparison see Eq. 3.5) with eigenvalues of ms = 1/2 a.u. and ms = 1/2 a.u., respectively. The Pauli spin matrices satisfy the
identities
p2 = I ;
p
q +
q
p = 2pq I ;
p
q = Ipq + ipqr r ; (p, q, r) = (x, y, z),
(3.9)
where pqr is the permutation symbol (also called as the Levi-Civita symbol) which is 1 if (p, q, r)
is and even permutation, -1 if it is an odd permutation or 0 if any index is repeated. The I
represents the 2 2 unit matrix

1 0
I =
.
(3.10)
0 1
277
Later on, in Section 4.3.1, we will show that the spin arises naturally as the solution of the Dirac
equation. Let us mention that all operators related to the solutions of the Schrodinger equation
are defined differently within the Dirac equation.
3.2
Many-electron systems
The first step in reducing theoretical and computational demands needed for solving the Schrodinger equation for manyelectron molecules and their clusters is the decoupling of the motion of
the electrons from that of nuclei. The same is desirable also at the relativistic level. This is most
frequently achieved by the BornOppenheimer (BO) approximation [54]. The computationally
simplest way is using the clamped nuclei Hamiltonian. We note that the BO approximation
leads to reasonably accurate results when electronic energy differences between the electronic
ground state and lowlying excited states are much larger than are differences between vibrational states. Details of the interactions between electronic, vibrational and rotational movements
can be found, e.g., in the book by Piela [51]. The Schrodinger equation for the manyelectron
wave function is solved in the field of fixed positions of nuclei. The electronic energy Ee (R)n
for a specific spectroscopic state n plays the role of the potential energy for oscillations. When
treated numerically, electronic energies calculated for different positions of nuclei allow finding
the minima on the potential energy hypersurface and equilibrium geometries, vibrational frequencies and some other spectroscopic characteristics. Having energies, geometries, vibrational
and other spectroscopic data for atoms and molecules participating in chemical reactions, we
can calculate reaction and activation energies, enthalpies and free energies, ionization potentials,
electron affinities and other properties (vibrationally corrected, if desired). Similar approach is
behind the energetics in intermolecular interactions or in energetics of clusters. Geometries and
vibrational frequencies are presently mostly calculated by analytical derivatives of electronic energies with respect to the nuclear coordinates Using appropriate operators we can also calculate
electric properties, like dipole moments, dipole and higher polarizabilities, magnetic properties
etc., by numerical derivatives of electronic energies with respect to an external perturbation (like
the electric field). Modern methods are based on response theories for obtaining molecular properties [39, 40].
The nonrelativistic Hamiltonian for an n-electron system in which electrons are exposed to
the field of N nuclei, I = 1, 2, ...N , at their fixed positions within the BO approximation is
e (1, 2...n) =
H
n
X
i
hi +
N
n
X
X
ZI ZJ
1
+
r
RIJ
ij
i<j
I<J
= 1 i
h(i)
2
N
X
ZI
.
riI
(3.11)
The first term in Eq. 3.11, hi , represents the kinetic energy of an i-th electron plus the interaction
between an electron and all nuclei. This is the oneelectron part of the Hamiltonian expressed
as the sum of oneelectron contributions. The second term represents a sum of twoelectron
contributions, r1ij , describing the inter-electronic interactions. The last term in the Hamiltonian,
ZI ZJ
RIJ
, represents the contribution of electrostatic interactions between nuclei. Within the BornOppenheimer approximation it is a constant, depending only on the fixed geometry of the system.
Now, it is appropriate to present a starting form of the wave function, 0 (1, 2, ...n) which
would be a solution of the stationary Schrodinger equation for a manyelectron system. A com-
278
mon way is expressing the wave function in the form of Slater determinant [55]

1 (1) 1 (2) ... 1 (n)

2 (1) 2 (2) ... 2 (n)

1
.
.
...
.
.
0 (1, 2, ...n) =

.
.
...
.
n!

.
.
...
.

n (1) n (2) ... n (n)
(3.12)
It is a very convenient form. First, it obeys the Pauli principle, i.e. it is antisymmetric with
respect to permutations of any pair of electrons. Second, it is expressed in the form of a product
of (antisymmetrized) oneelectron terms. This reflects the statistical interpretation of quantum
mechanics for manyelectron systems. Each electron in the nonrelativistic realm occupies a
spinorbital , which is the product of the oneelectron spatial function, , and the electron spin
function, :
i (r, ) = i (r); = (, ).
(3.13)
In the manyelectron theory it is advantageous to write the molecular nonrelativistic Hamiltonian in the second quantization form [39]:
=
H
hpq p+ q +
pq
1X
gpqrs p+ r+ sq + hnuc .
2 pqrs
(3.14)
where p+ , r+ , ..., q, s ...., are creation and anihilation operators, respectively; One and two
electron integrals, hpq and gpqrs , respectively, can be expressed over spinorbitals p,q,r, and s,
Eq. 3.13, in the usual notation as
hpq
gpqrs
hnuc
= hp (1)|h(1)|
q (1)i
1
1
= hp (1)q (2)|
|r (1)s (2)i hp (1)q (2)|
|s (1)r (2)i
r12
r12
= hpq||rsi
1 X ZI ZJ
=
.
2
RIJ
(3.15)
(3.16)
(3.17)
I6=J
Note that the second term in gpqrs results from the electron exchange. Employing the standard
rules for integration over spin variables we are left with integrals over spatial coordinates. We will
see later that the twoelectron integrals are real trouble makers in manyelectron calculations.
In sophisticated calculations we go beyond the singledeterminant approximation of the wave
function. In that case we need not only integrals over occupied spinorbitals but also integrals
over frequently enormous number of virtual spinorbitals. The initial wave function 0 (1, 2, ...n)
for a closedshell system in the nonrelativistic case can be expressed in terms of spatial orbitals
only, each one occupied by two electrons having spin and , respectively, and the same spatial
part (in the spinorbital formulation both and electrons are allowed to have different spatial
part). Therefore, the number of orbitals entering into the Slater determinant is one half of the
number of spinorbitals. The number of twoelectron integrals carrying indices of four orbitals
279
which must be calculated and used in the wave function and energy calculations is reduced
considerably. This is much more important in post-Hartree-Fock calculations than in the HF
step.
3.2.1
The independentelectron model: The HartreeFock method
The HartreeFock (HF) method [5658] represents a basic approximation for determining the
wave function of a manyelectron system. It is also called the selfconsistent field (SCF)
method, which reflects the way in which the HF solution is searched. The most frequently used
HF method assumes that the wave function 0 can be approximated by a single determinant,
Eq. 3.12. The energy functional, expressed by a Slater determinant of, say, the ground state of a
system and the nonrelativistic Hamiltonian,
E0HF =
e |0 i
h0 |H
,
h0 |0 i
(3.18)
is minimized, varying occupied oneelectron functions, used for the construction of the Slater
determinant, i.e. spinorbitals or orbitals. An additional condition is that so obtained orbitals
remain orthogonal, i.e. hi (1)|j (1)i = ij , where ij is the Kronecker delta. The procedure
leads to the set of one-electron equations [51], known as the HartreeFock equations:
f(1)i (1) = i i (1) ,
(3.19)
where the general structure of the Fock operator (in the basis of spinorbitals, Eq. 3.13) is
+
f = h
occ
X
i ).
(Ji K
(3.20)
is the oneelectron operator, see Eq. 3.11, Ji and K

i are the Coulomb and exchange
In Eq. 3.20 h
operators, respectively. Summation runs over indices of occupied orbitals, the only orbitals which
determine the HF energy. The coulomb and exchange operators can be formally defined as one
electron operators,
Z
1
dr2 j (1)
(3.21)
Ji (1)j (1) = i (2)i (2)
r12
i (1)j (1) =
K
Z
i (2)j (2)
1
dr2 i (1)
r12
(3.22)
the integration being over the coordinates of electron 2 only. Matrix elements of the J and K
operators are Jij = hi (1)j (2)| r112 |i (1)j (2)i and Kij = hi (1)j (2)| r112 |j (1)i (2)i. By
operators we are able to formally decouple the twoelectron problem
introducing the J and K
into a series of oneelectron problems. The physical picture adopted in the HF method is that
each electron is exposed to the field of all nuclei and a mean field created by all remaining
operators contain
electrons. Since the mean field is initially unknown (note that both J and K
280
spinorbitals which are to be determined) HF solutions are searched iteratively until a threshold
selected for the accuracy of the SCF energy or the HF orbitals is achieved, starting usually from
just a oneelectron part of the Fock operator, Eq. 3.20 . We will see later that in the relativistic
theory the Fock operator is substituted by the DiracFock operator which leads to the Dirac
HartreeFock set of oneelectron solutions of the manyelectron problem, an analogue of the
HartreeFock equations in the nonrelativistic case.
For the closed-shell systems having n electrons, in which we use the doubly occupied molecular orbitals with the same spatial functions for and electrons respectively, the Fock operator
(after the spin integration) can be written as
+
f = h
n/2
X
i ).
(2Ji K
(3.23)
Depending upon restrictions imposed on oneelectron functions entering into the Slater determinant, Eq. 3.12, we define different variants of the HF method. The most natural selection
for closedshell molecules is using doubly occupied orbitals, mentioned above. The advantage
of using the same spatial orbitals for a pair of and electrons can be partly employed for
openshell systems as well, i.e. pair of electrons corresponding to the doubly occupied orbitals
are restricted to have the same spatial part. Remaining orbitals are singly occupied (in so called
high spin openshell systems treated by a single determinant HF method are unpaired electrons
defined as electrons). Clearly, this restriction brings additional approximation into the wave
function, but it allows creating a more efficient computer code for openshell systems than it
is possible within the fully spinorbital formulation. Such defined HF method is the Restricted
Open Shell HF method, or, in short, the ROHF method [59]. If there are no restrictions other than
creating spinorbitals as a product of the spatial and the spin state for a single electron, Eq. 3.13,
allowing different spatial part for a spinorbital having or spin, we talk about the Unrestricted
HF method, UHF [60]. This is conceptually the simplest HF method, used mostly for treating
openshell molecules or for describing the dissociation processes of both closedshell and open
shell molecules. The two drawbacks of the UHF method are the following. First, the UHF wave
function is not a proper wave function of the total spin operator, S2 (it is spin contaminated).
Second, we have to treat twoelectron integrals over spinorbitals. However, there is two times
more spinorbitals than orbitals for closedshell systems (almost the same also applies to open
shells treated by the ROHF method). Treating just occupied and virtual orbitals resulting from
HF or ROHF equations instead of twice as much spinorbitals in UHF is crucial when going beyond the oneelectron HF approximation, i.e. in sophisticated manyelectron methods like CC
methods. Even if the computer code is simpler when the reference wave function is UHF, calculations are about four or three times more time consuming than with the RHF (for closedshell) or
ROHF (for openshell) reference wave functions. Considerations on adopting the oneelectron
functions, orbitals or spinorbitals, in the nonrelativistic theory would not be complete without
mentioning already now that the number of integrals correspondingly increases with using four
spinors in relativistic calculations. In any case, there is much space for improved theoretical
formulations and computer codes focused particularly on an efficient treatment of difficult terms
containing twoelectron integrals at any level of the manyelectron theory. Finally, we should
note that some atomic and molecular states can not be represented by a single determinant wave
function. Typical examples are excited singlet states which need at least two determinants to
281
define a reference wave function.

3.2.2
Algebraic approximation
With just a few exceptions, HF equations of atoms and molecules are solved within the so called
algebraic approximation. This means that the spatial part of spinorbitals or orbitals, i , entering
into the Slater determinant are expanded in terms of properly selected basis sets, :
X
i =
ci ,
(3.24)
In early days of molecular quantum mechanics were mostly Slatertype orbitals, resembling
the orbitals resulting from the analytical solutions of the hydrogenlike atoms. In line with this,
the first orbital theory of the chemical bond, not yet the HF theory, was called the MOLCAO
theory, (Linear Combination of Atomic Orbitals). Note, however, that the first theory of the
chemical bond was the HeitlerLondon valence bond theory [51]. Presently, basis sets are more
general, even though they mostly remain linked with specific atoms which are components of a
molecule or the molecular complex. Molecular calculations are highly dominated by expansions
in terms of Gaussiantype basis functions (GTFs). A general shape of the so called Cartesian
primitive Gaussian basis function located at the point A (usually but not necessarily at the point
nucleus) is
2
j k rA
g(, A, i, j, k) = N,i,j,k xiA yA
zA e
.
(3.25)
The orbital exponents of Gaussian basis sets are optimized in atomic calculations, i, j, k are
positive integers. Gaussians having i=j=k=0 represent stype functions, when i=1, j=k=0 we
talk about px functions etc. The reason why Boys [61] has suggested using Gaussians as basis
functions is that calculations of twoelectron integrals with GTFs are much easier than with
exponential Slatertype basis functions with a radial part erA ( is an optimized orbital exponent). The basic advantage of the Slatertype orbitals is their proper shape in the vicinity of
nuclei. Due to the singularity of the potential at a point nucleus with nuclear charge Z, the wave
function must have a cusp at the nucleus, i.e. its derivative at r = 0 should be Z. Obviously,
Gaussian functions have qualitatively wrong behaviour at the nucleus. They also have inaccurate
behaviour at long distances. This disadvantage can be largely eliminated by expanding i by
more Gaussians than would correspond to Slatertype basis functions. However, having more
basis functions means a need to calculate and to treat much more twoelectron integrals. This
is eliminated by contracting some primitive Gaussian functions into fixed groups with coefficients determined by simpler atomic calculations. Contracted Gaussian basis functions, CGTFs,
are then used in more demanding molecular calculations. The description of this technique can
be found in the literature [39, 62]. Most applications at the relativistic level rely on CGTFs as
well. However, nonrelativistic basis sets are not directly applicable to relativistic calculations.
Basis sets for relativistic calculations must be optimized having in mind a specific relativistic
Hamiltonian. Concerning the wrong behaviour of Gaussian functions in the vicinity of the nucleus, it is interesting to note that this is traditionally considered as a drawback of these basis sets
in nonrelativistic calculations. In the relativistic theory the nucleus is actually not a point charge.
In fact, its shape can be approximated by Gaussians (see Section 4.3.3)
282
Once having basis functions we are ready to calculate the molecular orbitals expansion coefficients ci . Within the HFSCF method coefficients ci are determined variationally. Inserting
the basis set expansion, Eq. 3.24, into HF equations we obtain Roothaan equations [51, 63]:
FC = SC,
(3.26)
where F is the Fock matrix (which depends on the coefficients C due to electronelectron interaction terms, see Section 3.2.1), C is a molecular orbital matrix, is the (diagonal) matrix
of orbital energies, and S is the overlap matrix of the basis functions. The dimension of the C
matrix is determined by the size of the basis set used in the MO expansion, Eq. 3.24. Therefore,
for reasonably large basis sets we obtain much more orbitals than is the number of electrons.
Clearly, the number of occupied orbitals (spinorbitals), which determine the HF energy, unambiguously follows from the number of electrons. Remaining (virtual) orbitals or spinorbitals
are empty in the HF wave function. Their shape in terms of basis set expansion, Eq. 3.24, is
determined by requiring their orthogonality to the space of occupied orbitals. A space of virtual orbitals is important in postHartreeFock correlation calculations. As a rule, the number
of virtual orbitals (Nv ) is much larger than No , the number of occupied orbitals. Very large
space of virtual orbitals leads to enormously large number of two-electron integrals needed in
postHartreeFock calculations. If we forget indices restrictions, the number of twoelectron
integrals over virtual orbitals, hp (1)r (2)| r112 |q (1)s (2)i, scales as Nv4 . Considering that in
modern post HF calculations can Nv be as large as 1000 [64], we realize that not only calculations of the twoelectron integrals (via transforming through integrals over the atomic basis
functions) but also managing large files is a problem. We will return to this issue in the next Section. By imposing the diagonalization of the matrix we create so called canonical HF orbitals.
This is straightforward in closedshell RHF and openshell UHF calculations, but a bit tricky in
ROHF calculations [40, 65, 66]. Most convenient in postHartreeFock correlation calculations
of openshell systems is using so called semicanonical orbitals which means that diagonal are
blocks within the space of occupied and virtual orbitals, respectively.
The HF method is said to be a best independentelectron approximation of a manyelectron
system. It recovers more that 99% of the total electronic energy. Yet, this is not enough for obtaining results with a high and controlled accuracy. The twoelectron part of the Hamiltonian is
just a good approximation, even if it is considered through the Coulomb and exchange operators
as in the HF method. Results are frequently quite satisfactory, but may fail even qualitatively in
predicting some molecular properties, reaction and activation energies and so on. The HF method
is not applicable to calculations of, e.g., van der Waals interactions. For dispersion forces and
many other applications proper treatment of atoms and molecules requires manyelectron methods for considering the electron correlation problem. HF solution is a good starting point for so
called postHartreeFock methods, which use the HF wave function as the zeroorder approximation.
3.2.3
Electron correlation in the nonrelativistic theory of many electron systems
The correlation energy can be well defined [67] for closedshell systems as a difference between
the exact nonrelativistic energy and the Restricted HartreeFock energy of a manyelectron system:
Ecorr = Eexact ERHF .
(3.27)
283
This definition can be extended to openshell molecules with some caution since there are few
variants of the independentelectron models for openshells.
The physical origin of the electron correlation arises from the approximate description of
the mutual repulsion of electrons within the independentelectron model. In the HartreeFock
method is the movement of an electron considered in the average field of other electrons and
this model does not describe their instantaneous interactions accurately. Notable deficiency of
the HF description consists in the fact that two electrons with opposite spins are not prevented
from occupying the same region of space at the same time. In contrast, the probability of finding two electrons with the same spin at a point r is zero, which is a direct consequence of the
Pauli principle which the HF wave function obeys. In most systems, particularly in closedshell
systems with the RHF reference the inaccuracy arising from the approximate treatment of electron interactions is dominated by the dynamical effects related to the instantaneous movement
of electrons. This represents a dynamical electron correlation. When the HF wave function is
not a satisfactory representation of the wave function, that is when a single determinant wave
function is not good enough, we talk about the nondynamical correlation. This is topical in
quasi degenerate systems, when several electronic states of a atom or an molecule have similar
energies. In this case the starting (or zeroorder) wave function should be represented by several determinants (having the same symmetry and spin states in the nonrelativistic theory) which
represent near degenerate states. Examples are lowlying excited states, frequently openshell
systems. In the relativistic theory near degeneracies frequently result from spinorbit effects.
Special is a class of systems, like atoms and molecules in some excited singlet states, which can
by no means be represented by a single determinant wave function. As a rule, a single RHF
determinant wave function is inadequate for a description of a molecule with stretched chemical bonds (far from the equilibrium geometry), with the exception when the molecule and all
fragments have a closedshell electronic structure. In all such cases the reference wave function
for the postHF treatment should be represented by a multiconfigurational SCF (MC SCF) wave
function. Alternatively, the reference HF wave function can be expressed in the UHF form and
all subsequent calculations can be performed within the spinorbital formalism. However, the
UHF wave function has another theoretical disadvantages mentioned in Section 3.2.1.
3.2.4
Configuration interaction
The exact solution of the Schrodinger equation can be expressed in terms of a linear combination of all Slater determinants obeying just the spin and symmetry restrictions which can be
constructed from a selected oneelectron basis in the nelectron Fock space:
F CI = C0 0 +
X
i,a
Cia ai +
ab ab
Cij
ij + . . . ,
(3.28)
i<j,a<b
where ai are monoexcited configurations, ab

ij are biexcited configuration etc. (up to ntuple
excitations according to the number of electrons of the system under consideration) with respect
to a reference wave function, 0 . The i, j... and a, b, ... represent occupied and virtual spinorab
bitals, respectively. Linear expansion coefficients C0 , Cia , Cij
are calculated variationally. The
CI wave function means an expansion in terms of a linear combination of spin and symmetry
adapted Slater determinants, or, more generally, Configuration State Functions, CSFs. F CI ,
284
when no other than the spin and symmetry restrictions are imposed, represents the Full Configuration Interaction (FCI) expansion. It gives the accurate solution corresponding to the selected
basis set within the algebraic approximation. The FCI energy would converge to the exact energy with gradually increasing the size of the basis set. However, the number of configuration
functions in Eq. 3.28 grows drastically with the number of electrons and with the size of the basis
set. Therefore, the FCI expansion is attainable only for relatively small systems and small basis
sets. Yet, FCI solutions are very useful as benchmark results for other methods.
For practical use the FCI expansion must be truncated at some level of the excitation operator
C with respect to the reference 0 , which defines the CI wave function. Truncation at the
single and doubleexcitation level, i.e. C = C1 + C2 , with respect to a reference 0 wave
function, defines the most common variational postHF method, the single determinant reference
CISD:
X
X
X
ab ab
CISD = (1 +
Ci )0 = 0 +
Cia ai +
Cij
ij .
(3.29)
i
i,a
i<j,a<b
We note that the reference function 0 must not necessarily be a HF determinant, even if this
is usually the case. Orbitals which enter the reference function and all ai , ab
ij etc. are known
(resulting from a HF solution) which means that also Slater determinants entering into the CI
expansion are known. The only variables which must be determined are the linear expansion
ab
CI coefficients Cia , Cij
and eventually coefficients of higher excitations, triple, quadruple, etc.
which defines the CISDT, CISDTQ etc. methods. They are determined by the Ritz variational
method requiring that the CI energy is a minimum, solving the eigenvalue problem
HC = EC,
(3.30)
where Hij are CI matrix elements over Slater determinants. The lowest solution represents the
energy and wave function of the ground state, higher solutions represent excited states. One
should realize, however, that at this level all the states refer to different vectors of the C matrix
but with Slater determinants constructed for all states from the same set of orbitals.
Consider now, with some simplification, the physical meaning of a linear expansion represented by Eqs. 3.28 and 3.29. Should the reference function 0 be a single determinant HF
reference means that electrons occupy orbitals (spinorbitals) which define the Slater determinant and interact mutually just through their average field. The HF energy is solely determined
by occupied orbitals i, j.... However, in the CI expansion we allow electrons to occupy more
space exploiting also virtual orbitals a, b, c... through the single and double excitation in CISD
(or higher excitations in a more general CI expansion). Clearly, allowing electrons to occupy
more space means a relief of their mutual repulsion, i.e. their movement is better correlated.
The CI method represents a basic and conceptually straightforward post HartreeFock model
for the treatment of the electron correlation [51, 68]. The problem is the computational tractability. Presently, the number of configurations in CI goes up to billions! This was not possible
until B. Roos [69] has discovered an iterative alternative to the straightforward Ritz variational
method. In a sense this resembles techniques employed in the Many Body Perturbation Theories
(MBPT) and the Coupled Cluster (CC) theory. Even if the excitation space with respect to the reference is restricted to single and double excitations, for manyelectron atoms and molecules are
computational demands unbearable. So we usually do not consider all electrons in the correlated
285
treatment. Electrons in lowestlying orbitals are usually well separated from valence electrons
and so they contribute to the atomic and molecular properties very little. In most processes (for
example in reaction or interaction energies) is their contribution cancelled. Therefore, we leave
these so called inner shell electrons uncorrelated. We talk that about these electrons as about
the frozen core electrons. Similarly, we do not allow electrons to occupy the highest lying
virtual orbitals. These computational tricks are well under control since we can verify the inaccuracy caused by omitting inner shell electrons from the correlation treatment. In molecules with
atoms as C, N, O, ..., the obvious selection of the frozen core are 1s orbitals. Some caution must
be payed to so called semicorevalence correlation. For example, it is insufficient to correlate
just ns2 np1 electrons of Ga, In, or Tl. Definitively, also (n-1)d10 shell must be correlated. For,
say, the gold atom, we usually correlate valence and corevalence 5p6 5d10 6s1 electrons. Special
caution needs selection of frozen orbitals in relativistic calculations. For Au (and, similarly for
Pt, Tl, and other atoms) relativistic shifts of orbital energies alters the order of orbital energies,
see Fig. 2.3. Also, different components of, e.g., f orbitals, may belong to different symmetries,
depending of the symmetry of the molecule. Therefore, we either do not correlate 5s2 and 4f
shells or correlate both (which is demanding).
For more accurate solutions of excited states, for simultaneous solving the problem of the
dynamical and nondynamical electron correlation, in cases when taking just one determinant is
inadequate for representing a reference wave function etc., the Multireference CI (MR CI) is
a method of the choice. In this case the reference wave function 0 is created as a set of determinants, not just as a single determinant. The selection of this wave function is not always
straightforward and needs some experience. Since excitations (again, mostly singles and doubles) refer to all determinants in the MR wave function, the MR CI method becomes quickly
intractable with the size of the MR space (so called model space). In general, all determinants
which are quasidegenerate must be included. The MR CI method is particularly useful in calculations of the bond breaking processes on the hypersurfaces, demanding are problems with
conical intersections which may appear in the treatment of reactions in excited states etc. [70].
3.2.5
Coupled Cluster theory
The CI theory discussed in the previous Section was a leading instrument for the high level treatment of the electron correlation problem in manyelectron atoms and molecules for a long time.
Up to the eighties the variational upper bound property, typical for SCF and CI, was considered
as a principal requirement imposed on molecular manyelectron methods. For this reason the
emerging Manybody Perturbation Theories (MBPT) were considered for the electron correlation calculations as less convenient. This view was quite general even though, for example, the
fourth order MBPT method was that time appreciated as clearly very accurate and computationally much less demanding alternative to CISD [37, 62]. Another bottleneck of truncated CI is the
problem of sizeconsistency, or proper scaling of energy with the size of the system. Size consistency means [40] that if the subsystems A and B are far apart, and if energies of A, B, and AB are
computed in an equivalent way, the computed energies satisfy E(AB) = E(A)+E(B). Related
to the size consistency is the concept of the size extensivity, which has its analogue in thermodynamics. In manyelectron systems was the importance of the size extensivity and the advantage
in using an exponential ansatz first realized by nuclear physicists. A quantum mechanical model
is said to be extensive if the energy of a system computed with this model scales correctly with
286
the size of the system. We recommend the reader to study a detailed analysis of these and related
problems in the literature, see Refs. [40, 71]. One has to be familiar with consequences of the
errors caused by these effects when calculating intermolecular interactions, reaction energies,
bond strength and in many other molecular properties by truncated CI methods, including CISD.
Correction can be achieved by including quadruple excitations, at least approximately .
Unlike truncated CI methods, MBPT and Coupled cluster methods provide properly size
consistent descriptions. Also, with CC methods higher excitations are considered more effectively than in CI methods. Therefore it is worthwhile to pay more attention to CC methods
used frequently also at the relativistic level. Particular attention should be paid to the CCSD(T)
method [35, 36] which is presently accepted as a golden standard of quantum chemistry. Another reason of our attention to the CC theory is that CC methods have been most often used
in our own calculations of molecular properties. The Many Body Perturbation Theory will be
shortly discussed in the correlatedrelativistic Section 4.5.1.
The Coupled Cluster theory was invented in fifties initially as a tool for understanding corre z eks pioneering implementation of the CC theory
lation effects in the electron gas [7275]. C
to manyelectron systems [7678] represents a real breakthrough in the development of methods
for the treatment of the electron correlation in atoms and molecules [3740, 71, 79]. The many
body diagrammatic formulation of MBPT and CC theories as presented in a detailed form by
z ek already in 1975 [80, 81] is up to now a wonderful tool for a transparent order
Paldus and C
by order analysis of the energy and the wave function. The diagrammatic technique also facilitates the formulation of final CC equations that obey the requirements of the size consistency and
an efficient implementation thanks to clever factorization that alleviates the unpleasant scaling
of CC with the number of correlated electrons and the number of virtual orbitals [37, 40, 71, 82].
Presently, the CC theory is extremely useful tool for solving a variety of manybody problems
starting from small systems as atomic nuclei, up to large polyatomic molecules, from weakly
bound atoms and molecules up to strongly bound systems as atomic nuclei. Most recent comprehensive and authoritative account of the CC theory and its applications, including relativistic
arsky, Paldus and Pittner [85].
theory [83, 84] represents a book edited by C
The essence of the CC theory lies in the exponential expansion of the wave function,
CC = exp(T)0 ,
(3.31)
where T is the excitation operator acting on the known reference state 0 , which is in most cases
represented by a singledeterminant HF wave function. The excitation operator can be expressed
using the Taylor expansion, so that CC can be written as
CC = (1 + T + T2 /2 + T3 /3! + . . . )0 .
(3.32)
Expressing the excitation operator T as a sum of all possible excitation operators (up to the
number of electrons),
T = T1 + T2 + T3 + . . . ,
(3.33)
leads to different variants of the CC method according to the truncation of the total excitation
operator T at a specific excitation level. Note that upon including all excitation operators the CC
wave function and the corresponding energy are accurate within a selected oneelectron basis
287
set which defines the reference wave function 0 . This is equivalent to full CI. The effect of the
single, double, etc. excitation operators, T1 , T2 , etc., acting on 0 can be expressed in terms of
the excitation amplitudes tab...
ij... defined as
XX
XX
tai ai
T1 0 =
tai a
+i|0 i =
i
T2 0 = (1/4)
XX
tab
+b+ ji|0 i =
ij a
i,j a,b
XX
ab
tab
ij ij
(3.34)
i>j a>b
(
a+ , b+ , ..., i, j ...., are creation and annihilation
operators, respectively, acting on a Slater de2P
abc... abc...
terminant). More generally, Tn 0 = 1/(n!)
i,j,k...;a,b,c... tijk... ijk.. . Note that in many
electron theories we prefer to deal with the nelectron reference determinant 0 rather than
the true vacuum state. In the above equations indices i, j, k, ... and a, b, c... etc. refer to occupied and virtual spinorbitals in 0 , respectively, and abc...
ijk... are determinants representing single,
double, etc. excitations from occupied to virtual spinorbitals. General indices will be denoted as
p, q, r, s, ....
To get the energy corresponding to the CC wave function |CC i defined in the exponential
form, Eq. 3.31, we have to solve the Schrodinger equation,
exp(T)|0 i = E exp(T)|0 i.
H
(3.35)
Next, we multiply Eq. 3.35 from left by exp(T) and obtain:

exp(T)|0 i = E|0 i.
exp(T)H
(3.36)
= exp(T)H
exp(T) is a similarity transformed Hamiltonian H.
It faThe Hamiltonian H
cilitates calculations of nested commutators of the Hamiltonian H with the cluster operator T
needed for obtaining algebraic expressions for the CC energy and the wave function in terms
of one and two electron integrals. By projecting Eq. 3.36 onto a reference determinant, |0 i,
we obtain the CC energy. The CC excitation amplitudes, tab...
ij... and the corresponding CC wave
function can be obtained by projecting Eq. 3.36 onto all excited determinants, |ab...
ij... i:
exp(T)|0 i = ECC
h0 | exp(T)H

hab...
ij... | exp(T )H exp(T )|0 i = 0.
(3.37)
(3.38)
At this point it is profitable to employ the Hamiltonian in the normalordered second quantization
form as
X
1X
0 i = fN + W
N +h0 |H|
0 i. (3.39)
=
hpq||rsi{
p+ q+ sr}+h0 |H|
H
fpq {
p+ q}+
4
pqrs
pq
Antisymmetrized twoelectron integrals hpq||rsi are defined in Section 3.2, see Eq. 3.16. Curly
brackets {...} denote a normalordered string of the secondquantized operators (see, e.g. [40,
0 i, the energy of the HartreeFock reference state, we have the
71, 81]). Subtracting h0 |H|
normalordered form of the Hamiltonian,
N = fN + W
N = H
h0 |H|
0 i.
H
(3.40)
288
N we obtain the CC correlation energy and the wave function,

Using H
4ECC
0
N exp(T)|0 iC
= h0 |H
ab...
= h
|HN exp(T)|0 iC ,
(3.41)
(3.42)
ij...
N and T1 ,
Eqs. 3.41 and 3.42 may be expressed using the diagrammatic representation of H
T2 , ... Tn , see e.g., Refs. [40, 71, 80, 81]. The subscript C in Eqs. 3.41 and 3.42 indicates that
only those terms in which the Hamiltonian is connected to the every cluster operator on its right
N involves merely
should be included in the algebraic representation of the expression. Since H
one and twoelectron parts, the expansion of the excitation operator exp(T ) is truncated. That
means, no higher than double excitations, i.e. no higher excitation operators than T1 , T2 and the
product T1 T1 (this would correspond to CC with singles and doubles excitations, i.e. CCSD)
may appear in any CC energy expression:
N (T1 + T2 + 1/2T1 T1 )|0 iC .
4ECC = h0 |H
(3.43)
The algebraic representation of Eq. 3.43 in spinorbitals is

4ECC =
X
ia
fia tai +
1X
1X
hij||abitab
hij||abitai tbj .
ij +
4
2
ijab
(3.44)
ijab
We stress that above expressions for the CC energy are general. The expression for the CC
energy remains the same when along with T1 and T2 also T3 , T4 , and so on would be included
in the cluster expansion of T, Eq. 3.33. Of course, the energy itself would be different, since for
any truncation of T = T1 + T2 + T3 + . . . different amplitudes enter into Eq. 3.44.
Eqs. 3.42, which determine the tab...
ij... excitation amplitudes are truncated by the excitation
level as well. This time, however, we have to consider which operators are included in the
CC exponential expansion, Eq. 3.33. The most common is the CCSD method [86] (Coupled
Cluster Singles and Doubles), approximating T as T = T1 + T2 . This means, that we have
to determine the single and double excitation amplitudes, tai and tab
ij , respectively. Again, due
to the existence of solely one and twoelectron parts in HN the exponential expansion of the
CCSD wave function contains no higher contributions than those from the triple and quadruple
excitations

1 2
1 3
|CCSD i =
1 + T1 + T2 + T1 + T1 T2 + T1
2
6

1 2 1 2
1
+
T + T2 T1 + T14 |0 i.
(3.45)
2 2
2
24
By the symbol = we are giving a notice that terms listed in Eq. 3.45 contribute to the CCSD
energy and the wave function. The exponential expansion of the wave function is in fact infinite
and terminates only upon projecting on 0 , singles and doubles, respectively, see Eqs. 3.41 and
3.41. Note that the expansion, Eq. 3.45, contains not only terms easily recognized as single and
double excitations which arise from the so called connected clusters (terms T1 and T2 ) but also
their products, disconnected clusters T1 T1 , T1 T2 , T2 T2 , up to T14 . T1 T2 represents a part of
contributions due to triple excitations, T2 T2 represents a dominating disconnected contribution
289
due to quadruple excitations. For their calculations we need, however, only amplitudes of the
single and double excitations. This explains the advantage of expressing the wave function as
an exponential expansion in CC as compared with the linear expansion in CI (along with the
sizeextensivity of CC). The CI coefficients and CC amplitudes can be symbolically related by
equations,
C1
T1
C2
T2 + (1/2)T1 T1
C3
T3 + T1 T2 + (1/6)T1 T1 T1
C4
T4 + (1/2)T2 T2 + T1 T3 + (1/24)T1 T1 T1 T1
(3.46)
...
Theoretical and computational implementation of T3 , T4 and even higher excitation operators
(which introduce connected contributions of triple and quadruple excitations) is available [82,87
89], but iterative CCSDT, CCSDTQ and higher level methods are prohibitively demanding for
other than benchmark calculations of smaller molecules. The algebraic representation of Eq. 3.42
is quite complicated even for the CCSD method. The reader is referred to the literature, see
e.g., [37, 40, 71, 80]. We just mention that the single and double excitation amplitudes in CCSD,
tai and tab
ij , have to be determined in an iterative process starting from the simplest amplitudes,
(1)
ab
which are the firstorder wave function tab
). Let us assume that we work with
ij amplitudes (tij
(1)
canonical HF orbitals, for which hp|f |qi = p pq . In the spinorbital form tab
amplitudes are
ij
represented by a simple formula,
tab
ij
(1)
hij||abi
.
i + j a b
(3.47)
ab...
The denominator in Eq. 3.47 , (denoted in the following as Dij...
)
ab...
Dij...
= (fii + fjj + . . . ) (faa + fbb + . . . ) = i + j + a b . . .
(3.48)
N . For the reference determinant with canonical HF

results from the oneelectron part of H
orbitals, Eq. 3.19, the only nonzero terms are diagonal Fock matrix elements, i.e. orbital energies.
We should add that the single excitation amplitudes tai appear for the first time as the second
order terms (however, they are the first order terms for openshells with the ROHF reference).
(1)
It is easily recognized that with tab
defined in Eq. 3.47 we obtain the second order MBPT
ij
correlation energy (MBPT2),
(2)
4EM BP T =
X
1X
(1)
hij||abitab
=
ij
4
ijab
i<j,a<b
hij||abi
.
i + j a b
(3.49)
The expression is pair additive, that means the correlation energy can be expressed in terms over
pairs of occupied (virtual) orbitals. Higher orders of MBPT do not exhibit such additivity. Note
(2)
that the expression for 4EM BP T is formally the same as is the second term in the 4ECC energy.
The only difference is that the amplitudes in 4ECC are the infiniteorder tab
ij CC amplitudes
290
(2)
while in 4EM BP T there are the firstorder double excitation amplitudes (terms with tai are zero
in the MBPT2 energy, since they appear for the first time in the secondorder wave function).
We will use this simplest nonrelativistic expression for the correlation energy in Section 4.5.1 as
a starting point for discussing the theoretical and computational demands when going from the
nonrelativistic to the relativistic correlated treatment of atoms and molecules. This instructive
example may give the reader some idea of what can be expected with higher electron correlation
methods at the fourcomponent relativistic level.
Realizing that all iterative methods lying higher in the hierarchy of the CC methods are computationally much more demanding than CCSD, we have to ask a question of including connected
triples and quadruples in a more approximate way. CCSD contains disconnected contributions of
triples and quadruples, but for accurate results are important connected triples and for quasidegenerate systems also connected quadruples. The importance of triples was for the first time
z ek [78]. The first systematic inclusion of triple excitadocumented by Paldus, Shavitt and C
tions into the CC theory was studied by Lee, Kucharski and Bartlett [90]. Their approximate
iterative CCSDT1 method was largely extended to a series of the CCSDTn approximations
of the full CCSDT by Urban et al. [35] based on the careful orderbyorder CC wave function
analysis. Full iterative CCSDT was implemented by Noga and Bartlett [87]. Higher level CC
calculations are also available [40, 91], albeit obtaining an algebraic representation of CC equations beyond CCSDT is extremely tedious. The problem was alleviated by using diagrammatic
techniques and the second quantization formalism, leading to the string-based coupled cluster
theory and automated generation of computer codes [89], as pioneered by Li and Paldus [92].
The critical need for including triples in an economical fashion culminated by introducing the
famous CCSD(T) method. The idea was that the main contributions from triple excitations (4
th order energy terms in the perturbation theory), that are missing in CCSD can be effectively
added via perturbationbased technique noniteratively, using all the correlation information
available from the converged T1 and T2 CCSD amplitudes. The CCSD(T) method [3537], represents perhaps the best compromise between accuracy and efficiency for treating highlevel
electron correlation problem for small and medium size molecules represented by a good single
determinant reference. As many comparisons with experiment and benchmark FCI and higher
level CC methods showed, it is accurate enough and still allowing the usage of sufficiently large
basis sets. For its safe use it is recommended checking the suitability of a single determinant
reference by inspecting the tai and tab
ij CCSD amplitudes which should not be larger than 0.20 or
to use the diagnostics for single excitation amplitudes (the T1 diagnostics, [71]). We note, however, that even CCSD(T) remains computationally very demanding for large scale calculations.
Its importance, besides being an excellent tool for predicting molecular properties lies also in its
role as a source of benchmark results for DFT methods, which are missing a systematic test of
accuracy.
In CCSD(T) calculations with large basis sets the most demanding step remains the triples
step, even though it is noniterative (in contrast to the iterative CCSD). It should be interesting
to realize that CCSD(T) for openshells and for fourcomponent relativistic calculations is even
more demanding than are nonrelativistic closedshell calculations. Therefore we will show a few
equations in order to demonstrate the scaling of triples with the number of occupied (No ) and
virtual (Nv ) spinorbitals.
The CCSD(T) energy is represented by a sum of the converged CCSD energy plus contributions from connected triples calculated noniteratively using the perturbative MBPT arguments
291
(for more details see [36, 37, 40, 71]:

(CCSD)
ECCSD(T ) = ECCSD + ET
(CCSD)
where ET
fashion,
(CCSD)
ET
(CCSD)
+ EST
(3.50)
is the correction from triple excitations calculated in the fourth order MBPT
=
abc abc
tabc
ijk Dijk tijk .
(3.51)
i<j<k a<b<c
abc
Dijk
is the denominator defined by Eq. 3.48. The triple excitation amplitudes are calculated
using the expression (see, e.g. [37, 40, 71])
X
X
abc abc
Dijk
tijk = P (i/jk)P (a/bc)
hbc||diitad
hla||jkitbc
(3.52)
jk P (i/jk)P (a/bc)
il .
d
All these equations are presented in the simplest spinorbital form. Eqs. 3.51 and 3.52 look like
the MBPT equations which define the fourth order energy contribution from the triple excitation amplitudes. However, triples in CCSD(T) are calculated using the converged CCSD tab
ij
ab (1)
infinite order amplitudes instead of the first order t

double excitation amplitudes. P is the
ij
(CCSD)
permutation operator. The ET

contribution which supplements the 4ECCSD correlation
energy (or ECCSD in a shorter notation) defines the CCSD + T(CCSD) method [35] which is a
forerunner of the CCSD(T) method. The difference is that Pople et al. [36] have incorporated in
(CCSD)
CCSD(T) a single fifth order contribution, EST
(all other fifth order energy contributions
were neglected), which represents an interaction of singles and triples:
(CCSD)
EST
X X
a
hbc||jkitabc
ijk ti .
(3.53)
i,j<k a,b<c
(CCSD)
The EST
contribution is an analogue of the fifth order contribution to the MBPT energy,
(CCSD)
(5)
calculated, again,
EST , comprising the single and triple excitations. In CCSD(T) is EST
employing the converged CCSD tai infinite order single excitation amplitudes instead of the second order wave function single excitation MBPT amplitudes. Formally the second order MBPT
wave function triples amplitudes tabc
ijk are calculated using Eq. 3.52 from the infinite order CCSD
a
tab
doubles.
We
note
that
the
t
singles
are first order wave function amplitudes for openshells
ij
i
with the ROHF (or any other nonHF) reference. For details see [38, 40, 65, 66, 71].
No doubt, the most frequently used CC method, CCSD(T), is computationally much more
demanding than are the SCF or the Density Functional Theory (DFT) calculations. Even more
demanding are CCSDT, CCSDTQ etc. methods and their variants. The role of CC methods
and particularly the CCSD(T) method as a source of benchmark results for e.g., tuning less
demanding DFT methods can be hardly overestimated. Enormous importance of CC methods
lies in their potential for obtaining results with high and controlled accuracy, i.e. good results
for a good reason. This follows from the fact that CC methods form a natural perturbativelike
hierarchy [35, 3740, 71] employing gradually more accurate wave functions in more efficient
292
way than in other (for example variational) methods. The efficiency follows from utilizing the
exponential ansatz for the wave function, Eq. 3.31. At the same time this ansatz guarantees the
sizeextensivity of CC energies [40, 71]. We note that an instrument for controlling the accuracy
and reliability of frequently used DFT methods is missing. Another problem with DFT is plenty
of functionals used in calculations of a specific property and difficulties of DFT in representing,
e.g., dispersion forces in Van der Waals interactions, diffuse parts of the electron distribution
needed for proper treatment of electron affinities, etc.
Should CCSD(T) results for smaller molecules serve as a benchmark for other methods,
reasonably large basis sets for accurate description of occupied orbitals are required. The number
of occupied orbitals (spinorbitals) is strictly determined by the number of electrons. Large basis
sets create large space of virtual orbitals and, therefore, enhanced computational demands in
the treatment of the correlated calculations. With the UHF reference the most demanding steps
for obtaining the single and double excitation amplitudes in CCSD scale with the number of
occupied (No ) and virtual spinorbitals Nv as No2 Nv4 in an iterative process. Resulting single
and double excitation CCSD amplitudes are afterwards used in the noniterative approximate
determination of the triple excitations. Calculations of the triple excitation amplitudes, Eq. 3.52,
is the most demanding part in CCSD(T) with large basis sets (for large basis sets is Nv No ).
The tabc
ijk amplitudes are carrying three target indices of occupied orbitals and three indices of
virtuals. Adding the summation over index of the occupied or virtual orbital leads to unfavorable
scaling No3 Nv4 and No4 Nv3 . The final calculation of the energy resulting from triples is slightly
less demanding, scaling as No3 Nv3 , see Eqs. 3.51 and 3.53. Worst steps in the rigorous iterative
CCSDT would need arithmetic operations which scale as No3 Nv5 in each iteration. Approximate
CCSDTn methods are a bit simpler but still very demanding. Another problem with CCSDT
and CCSDTn is also storing of No3 Nv3 triples amplitudes.
For closedshell systems the computational cost of CCSD(T) calculations can be reduced
by about a factor of 4 when working with orbitals instead of spinorbitals (the number of occupied/virtual orbitals for closedshells is one half of the number of spinorbitals). Substantial
reduction of computer demands for openshell systems with respect to the UHF reference can
be achieved using the ROHF reference [65]. Theoretical background of the ROHF CCSD(T)
method with the spin adapted amplitudes is more complicated [66, 93] than with the UHF reference or methods for the closedshells with the RHF reference. Theoretical reduction of the
computer time for closedshells or for openshells with the ROHF reference does not correspond to the expected factor of reducing the number of orbitals by one half since it is hampered
by more complicated structure of the final equations after applying the spin integration. Clearly,
working with fourcomponent spinors makes relativistic CCSD(T) calculations even more demanding than are nonrelativistic calculations utilizing doubly occupied orbitals for closedshells
or onecomponent spinorbitals with the UHF reference.
Modern formulations of the CC theory, sophisticated factorization of complicated equations
for obtaining CC amplitudes, combined with a quickly enhanced computer power accompanied
with a high level of parallelism allow calculations of gradually larger and larger molecules with
a high and controlled accuracy [64]. Techniques exploited in modern approaches utilize many
mathematical tricks and physically well founded approximations like Choleski decomposition
for more efficient treatment of the twoelectron integrals (which is a bottleneck in all sophisticated manyelectron methods), optimization and subsequent reduction of normally very large
space of virtual orbitals (OVOS, Optimized Virtual Orbitals Space), the symmetry constrains
293
etc. [64]. Using of CC methods based on the DouglasKrollHess Hamiltonian [18, 94, 95], in
which the scalarrelativistic effects are included needs modified integrals over basis functions
(forgetting for a need of modified contractions of basis sets) but otherwise it is quite straightforward. Very efficient are approaches relying on the idea of localized orbitals and also explicitly
correlated CC methods [96] working with the r12 variables in the wave function. Eventually, CC
methods approach in some cases an almost linear scaling with the size of a molecule, which is a
dream goal of the quantum chemistry methods for manyelectron systems.
Large progress is observed in developments of the Multireference CC methods (MR CC) and
approaches capable of treating quasidegenerate systems [38, 88, 9799]. MR CC approaches are
very important in relativistic calculations particularly for the SO coupling or QED effects [84].
Relativistic CC methods will be discussed in more detail in Section 4.5.2.
3.2.6
CASSCF, RASSCF, and CASPT2 methods
The physical origin of the electron correlation was summarized in Section 3.2.3. We have already introduced a concept of the dynamical and nondynamical correlation. Clearly, both effects
can not be entirely separated and mean just a possible way for understanding a difficult problem
of the treatment of interelectronic interactions in many electron systems. The problem can be
somewhat simplified if the starting wave function for high level calculations can be just a single
Slater determinant which is a solution of the HF equations. The single determinant methods for
treating the dynamic correlation were discussed in previous sections. In this part we will summarize some methods applicable to problems which require two or more Slater determinants which
represent a reasonable starting wave function of a system under consideration. As mentioned in
Section 3.2.3 this occurs in investigations of the bond breaking processes, in many openshell
systems, particularly in lowspin excited states, and other. First indication that the single determinant reference is insufficient is a small energy gap between the SCF occupied and virtual
orbitals in a single Slater determinant wave function (if used as a reference). In the CISD wave
function the leading determinant must be strictly dominating (by more than 90%) and the singly
and doubly excitation amplitudes in CCSD must be not larger than 0.2 for calculations with a
single determinant reference. A trivial example of problems with a single determinant reference
is the bond breaking in the hydrogen molecule. It is perfectly well represented by a single determinant around the equilibrium bond distances Re . Upon stretching the bond length the potential
energy curve (calculated within the BornOppenheimer approximation) has a reasonable shape
up to distance of about 1.5 Re . For longer distances the energy gap between the bonding g and
the antibonding u orbitals gets narrow and the potential energy curve deteriorates. At the full
separation is the situation simple - dissociation products are just two hydrogen atoms. In medium
distances, however, we need at least two determinants, corresponding to the electronic configurations g2 and u2 (having the same spin and the same state symmetry). Analogous orbitals
participate also in the bond breaking in the F2 molecule. More complicated is the situation when
breaking (or forming) the double or triple bonds in which participate orbitals. Examples are the
dissociation of the ethylene molecule into two methylene radicals or bond breaking of the triple
bonded N2 molecule. The fact that the electronic potential energy curves around minima are for
many molecules well represented by energies obtained from the correlated single determinant
reference wave functions and methods, like CCSD(T), can be documented numerically: In most
cases the polynomial fit over energies around the minima leads to reasonably accurate harmonic
294
vibrational frequencies (alternatively they can be obtained by analytical derivatives at Re ). Anharmonicity is occasionally less accurate with a single determinant reference, since it depends
on energies more apart from the equilibrium. Careful analysis of participating determinants is
also needed for many excited states and radicals.
Basic method for obtaining a linear combination of Slater determinants needed for a physically correct description of the reference wave function is the MultiConfiguration SCF method
(MC SCF). Several or many Slater determinants are chosen to guarantee a qualitatively correct
description of all states and electronic configurations. So selected set of Slater determinants leads
to the description of the nondynamical electron correlation. The dynamical correlation must be
added by the CI method in which the reference is the MC SCF wave function. So defined MR
CI is mostly computationally very demanding.
The MC SCF wave function for the electronic state I is a linear expansion in terms of a set
of preselected Slater determinants P ,
I =
CP P .
(3.54)
The variational parameters CP are optimized simultaneously with the expansion coefficients cp
of the spatial part of molecular orbitals p which define the Slater determinants P . Molecular
orbitals
(either occupied or virtual) are expressed within the algebraic approximation, p =
P
c
p
, Eq. 3.24. Resulting molecular orbitals are the same for the ground state (the lowest
solution of the eigenvalue problem) and for excited states. Their shape is, so to say a compromise
between the accuracy and effectivity. Obviously, selecting a set of determinants for a MC SCF
wave function, Eq. 3.54, may be in some cases difficult and to some extent arbitrary. Selecting
too many Slater determinants as a reference makes the subsequent MR CISD calculation quickly
intractable.
Presently, MC SCFtype methods are dominated by the CASSCF (Complete Active Space
SCF) approach. In CASSCF the user selects the set of active orbitals and the number of electrons
in active orbitals instead of preselected Slater determinants. Clearly, for a regular user of the
computational chemistry programs thinking in terms of orbitals is easier than in terms of determinants. Selected orbitals are those which actively participate in the process under consideration
or which participate in electronic states which are to be considered (for example, the ground state
and several lowlying excited states). The CASSCF wave function is defined [100] by selecting
a set of active orbitals and active electrons and is constructed as a linear expansion in the set of
configuration state functions (CSFs) that can be generated by occupying the active orbitals in all
possible ways consistent with a spin and state symmetry. Therefore, they form full CI within
active orbitals. This is very advantageous since the wave function is then fully variational, both
in orbitals and in CSFs and satisfies some important requirements, as the HellmannFeynman
theorem, which allows, e.g., calculations of electric properties as expectation values. In some
difficult cases we may need for a proper representation of the process or a state under consideration so many CSFs in the CASSCF wave function that calculations become computationally too
demanding. To alleviate the problem, the active space is split into three subspaces, listed below.
Some excitations from/to orbitals in the RAS1 and RAS3 space, respectively, are somewhat restricted. Methods has an acronym RASSCF. When the RAS1 and RAS3 spaces are empty, the
RASSCF method is equivalent to CASSCF.
295
Partitioning of the occupied molecular orbitals in RASSCF and the participation of orbitals
in CASPT2 (which will be defined later in this Section) can be summarized as follows:
Frozen orbitals: Orbitals that are doubly occupied in all configurations in CASSCF and
PT2.
Inactive orbitals: Orbitals that are doubly occupied in all configurations in CASSCF.
Allowed are excitations in the PT2 step.
Active orbitals: These orbitals are subdivided into three separate groups:
RAS1 orbitals: Orbitals that are doubly occupied except for a maximum number of
holes allowed in this orbital subspace.
RAS2 orbitals: In these orbitals all possible occupations are allowed (former Complete Active Space orbitals).
RAS3 orbitals: Orbitals that are unoccupied except for a maximum number of electrons allowed in this subspace.
Secondary orbitals: Virtual orbitals that are inactive in CASSCF. Excitations to these
orbitals are allowed in the PT2 step.
The CASSCF method can handle the neardegeneracies very effectively and is particularly suitable as a starting point for applications to processes on the energy hypersurfaces and in the electronic spectroscopy [100]. The extension of CASSCF to the relativistic domain and particularly
to the spinorbit interactions will be treated in Section 4.7.2.
The CASSCF wave function is not capable of treating the dynamic correlation effects satisfactorily (unless the active space would be unrealistically large). Therefore, CASSCF is used as
a zeroorder wave function (a root function) for the multiconfigurational secondorder perturbation theory, which defines the CASPT2 method [64, 100, 101]. CASPT2 uses orbital excitation
operators applied to the root wave function to express the firstorder perturbation function. With
respect to the structure of the active space we should add that within the PT2 perturbation treatment also excitations from inactive orbitals are allowed. They do not participate, however, in
optimization of the CI coefficients in the expansion, Eq. 3.54. The same holds for so called
secondary orbitals which can be employed in the excitation process in the PT2 step. Finally, as
with CI and CC methods there are frozen orbitals that are not used in any excitation process and
remain doubly occupied in both the CASSCF and in the PT2 step. Analogously, some highest
lying virtual orbitals are neglected in the PT2 dynamic electron correlation calculations
The problem which had to be solved [100,101] is the definition of the zeroorder Hamiltonian
0 . A good choice in defining the simple enough H
0 is a generalization of the zeroorder HamH
iltonian based on the HF operator, usual in a single determinant reference perturbation theory.
We refer the reader to the original literature [100, 101].
CASSCF, CASPT2, RASPT2 and their generalizations to the relativistic domain, RASSI
[102, 103] have been successfully applied to a large variety of problems in chemistry and molecular physics. These methods are implemented into the famous MOLCAS computer package [64]. When the active space is selected properly, the method is sufficiently accurate and
reliable. Possible problems may occur due to an inadequate choice of active orbitals [104].
Alternative selection/extension of the active space is also a tool for controlling the accuracy.
Relativistic treatments include the scalarrelativistic methods based on the DouglasKrollHess
Hamiltonian [18, 94, 95]. The RASSI method for the treatment of spinorbit effects will be discussed in Section 4.7.2. Important is also availability of basis sets for both nonrelativistic and
relativistic [105, 106] calculations.
296
While the Schrodinger nonrelativistic description, Section 3.1, is sufficient for light elements,
here we focus ourselves on the derivation of a more general equation fulfilling the principles of
the theory of relativity.
4.1
Lorentz transformations
When Maxwell published his four equations for electromagnetism [44], in contrast to Newtonian physics these were not invariant with respect to the Galilean transformation which serves
for transformation between the coordinates of two reference frames differing only by constant
relative motion. The Galilean transformation, acting in the realm of the Newtonian physics, relates two different reference systems with coordinates r and r0 , moving with a relative velocity
v,
r0 = r vt.
The spacetime Galilean transformation between two systems can be written as follows
0

r
1 v
r
=
.
t0
0 1
t
(4.1)
(4.2)
Lorentz proposed the following transformation [107] to make Maxwells equations invariant:

0

r
r
1
v
.
(4.3)
=
1
t
t0
cv2
is the relativistic scaling factor
r
v2
1 2,
=
c
(4.4)
which approaches the value of 1 in the nonrelativistic limit when velocities are negligible in
comparison to the speed of light.
The Lorentz transformation illustrates how, in line with the theory of the special relativity,
two observers varying measurements of space and time can be turned into each others frame of
reference. Though the Lorentz transformations were invented in a somehow ad hoc manner, they
may be also introduced if two basic postulates are taken into account:
1. The Principle of Relativity: The physical laws are the same in all inertial reference frames.
2. The Principle of Invariant Light Speed: Light in free space is propagated with a constant
velocity, regardless of any relative motion of observer and of the source.
These postulates were first suggested by Henri Poincare [108], but it is Albert Einstein [43]
who is credited with discovering and promoting the full impact of the new theory of relativity.
Note, that the second postulate claiming the invariance of the speed of light, c, can also be derived
from Maxwells equations. The constant speed of light in all inertial frames can be viewed as the
central theme in relativity. By applying these presuppositions it can be shown that physical laws
should be invariant under the Lorentz transformation, Eq. 4.3, rather than under the old Galileo
transformation, Eq. 4.2.
4.2
297
KleinGordon equation
The Einstein theory of relativity [43, 109] gives the total energy expression in terms of the momentum and of the rest mass, m0 :
q
(4.5)
E 2 = p2 c2 + m20 c4 ;
E = p2 c2 + m20 c4 .
This equation may be utilized to form the classical free particle Hamiltonian, Section 4.3. The
corresponding quantum mechanical expression is derived by substituting the classical momen = ih. The same holds for
tum, p, by its (three component) quantum mechanical operator, p
.
the total energy which is replaced by the energy operator, ih t
We insert these operators into the square root of energy expression, Eq. 4.5, and get:
i
h
q
(r, t)
h2 2 + m20 c2 )(r, t).
= c (
t
(4.6)
The resulting wave equation is known as the KleinGordon equation [110,111] and is manifestly
Lorentz covariant. However, other aspects of the equation are problematic. It is difficult to
interpret the square root operator. The Taylor expansion is possible, but it would lead to all
powers of the derivative operator and thus to a nonlocal theory. Negative energy solutions have
negative probability densities. Neither does the KleinGordon equation handle the spin nature
of the electron. These facts lead to rejection of the equation, but in 1934 it found its place as the
relativistic wave equation for spinless charged particles [112].
4.3
Dirac equation
Above mentioned weaknesses motivated Dirac to take the right track and derive a single particle
equation overcoming all these problems. Though there were several attempts to formulate the
theory that would couple principles of relativity with those of quantum mechanics, it was Dirac
who invented relativistic quantum mechanics for oneelectron systems.
Dirac tried to derive an Lorentz invariant equation in the general, timedependent form (in
a.u.)
i
D
= h
t
(4.7)
which can be turned, as in the nonrelativistic case (see Section 3.1), into the timeindependent
form:
D = E.
h
(4.8)
Eq. 4.7 is linear in the first derivative with respect to time. The Lorentz invariance requires the
Hamiltonian to be linear in the first derivatives with respect to space as well. Such equation is
expected to have the form (in a.u.)
i
= ic(
x
+
y
+
z
) + c2 ,
t
x
y
z
(4.9)
298
and terms must be determined by additional criteria. One of them

where newly introduced
is that solutions of the new equation must still fulfil the KleinGordon equation, Eq. 4.6. Paul
Dirac commented on the topic: It took me quite a while...before I suddenly realized that there
was no need to stick to quantities ...with

just two rows and columns. Why not go to four rows
and columns? [113]. In this way Dirac found missing terms,

0
I 0
=
; =
,
(4.10)
0
0 I
with mutual relations
[
p ,
q ]+ = 2pq , [
q , ]+ = 0 ,
q2 = 2 = 1 ,
(p, q) = (x, y, z),
(4.11)
where expression in []+ parenthesis is the anticommutator, [a, b]+ = ab + ba, and pq is the
are 2 2 Pauli matrices, mentioned in Section 3.1.2, and I is the
Kronecker delta symbol.
2 2 unit matrix, Eq. 3.10.
Dirac equation posses the form of a 4 4 matrix. Further usage of the notation four
component Hamiltonian means the Dirac based energy operator. For the freeelectron it can
be written as (in a.u.):

p
c2 I
c
2
f e = c
p
+
c
=
.
(4.12)
D
p
c2 I
c
The Dirac relativistic wave function has four degrees of freedom besides the spacetime
variable: spinup () and spindown () both for electron and its positively charged conjugated
partner. It is called the fourcomponent spinor, or shortly fourspinor.
L
L
L
=
=
(4.13)
S .
S
S
The L and S in the Dirac fourspinor are twocomponent spinors containing large (L) and
small component functions (S), respectively. The probability density is given as the product of
fourcomponent vectors:
= (L ) L + (L ) L + (S ) S + (S ) S ,
(4.14)
where is a onerow vector:

= ((L ) , (L ) , (S ) , (S ) ).
(4.15)
The current density vector, j, can be expressed as (in a.u.)
j = c .
(4.16)
Contrary to the Dirac equation, its nonrelativistic Schrodinger counterpart, Section 3.1, is of
the second order in momentum. One could say that the Schrodinger equation is suitable for the
description of a universe where - hypothetically speaking - the speed of light is infinitely high.
4.3.1
299
Nonrelativistic limit of the Dirac equation
In general it is important to understand relations between relativistic and nonrelativistic theories.

For that reason we establish the connection towards the nonrelativistic limit of the Dirac equation.
First, we write the freeparticle Dirac equation in the shortened form of twocomponent spinors:
fe
h
D
L
S

= E
L
S

.
(4.17)
In the next step we express the Dirac equation for the free electron, Eq. 4.12, as a simple set of
two (2 2)dimensional matrix equations (in a.u.):
p
)S = 0,
(c2 E)L + c(
L
2
p
) (c + E)S = 0.
c(
(4.18)
The small component can be obtained from the second expression of Eq. 4.18
S =
p
) L
c(
,
2
(c + E)
(4.19)
and inserted it into the first equation:

[(c2 E) +
p
)(
p
)
c2 (
)]L = 0.
(c2 + E)
(4.20)
We make now the key approximation to approach the nonrelativistic framework: the nonrelativistic energy, , is positive and very small compared to c2 (in a.u.):
E = c2 + c2 .
(4.21)
After neglecting the energy contribution in the denominator, Eq. 4.20, we get
[ +
p
)(
p
) L
(
] = 0.
2
(4.22)
Utilizing the identity

p
)(
p
) = p2 ,
(
(4.23)
we get a twocomponent equation for L , which is in fact the Schrodinger equation, Section
3.1, for a free electron (in a.u.):
[ +
2 L
p
] = 0.
2
(4.24)
Thus, the large component L can be interpreted as the nonrelativistic limit (c ) of the
fourcomponent solution. Moreover, its twocomponent solutions
L = c L + c L ,
(4.25)
300
where and are defined in Eq. 3.8, and are carrying information about the spinorientation
of a particle. In this way we have shown that the Dirac equation contains inherently the spin
quantum number. This demonstration can also be carried out for a particle in an external field.
In the Schrodinger description of the electronic structure the spin quantum number (see Section
2 /2, is mathematically
3.1.2) must be added ad hoc. Note that since the kinetic energy operator, p
2
p
) /2, it may be stated that the electron spin is actually
equivalent, according to Eq. 4.23, to (
present in the nonrelativistic case. Therefore, if the kinetic energy term is present in the form of
p
)2 /2 in the Schrodinger equation, than the electron spin is involved in the nonrelativistic
(
case. This would have consequences only in the presence of a magnetic field.
Regarding the small component, Eq. 4.19, by employing the approximation of Eq. 4.21, we
get (in a.u.)
S
p
) L
(
.
2c
(4.26)
Values of small component, S , are 1c times smaller (in a.u.) than those of large component,
L .
Negative energy solutions According to Eq. 4.5 solutions for a free particle are symmetric
with respect to zero. They are of continuum type and energies span the range below mc2 and
above mc2 (or just c2 and c2 in a.u.). The spectrum is thus not bounded from below and from
above.
Negative energy solutions worried Dirac: One gets over the difficulty on the classical theory
by arbitrarily excluding those solutions that have a negative E. One cannot do this in the quantum
theory, since in general a perturbation will cause transitions from states with E positive to states
with E negative. [4].
Negative solutions seemed to be problematic, since it was assumed that a particle has a positive energy. Dirac introduced that the vacuum is the manybody quantum state in which all the
negativeenergy electron eigenstates are occupied. Negative states are occupied with particles
(i.e. they form the physical vacuum), so that electrons of positive energy do not spontaneously
fall into negative states with the release of energy. This description of the vacuum as a sea
of electrons is called the Dirac sea. Transitions down into the Dirac sea are forbidden by
the Pauli exclusion principle. Any additional electron therefore would be forced to occupy a
positiveenergy eigenstate. However, there is no need to assume infinite number of particles in
the vacuum since the modern relativistic theory describes both particles and antiparticles in a way
that only the total charge is conserved, but not the total number of all particles and antiparticles.
Dirac stated that if energy higher than 2c2 a.u. is applied on an electron in a negative
energy state this electron could jump into a positive energy state where it could be observed as
an ordinary electron. The remaining hole, i.e. positively charged electron (positron), in the
negative energy state could also be observed. If an electron with positive energy eventually falls
into this hole, both would disappear with releasing of the energy higher that 2c2 a.u. Diracs
relativistic theory predicted the existence of the positron, the particle of the weight of an electron
but with the positive charge. This brought a great success for his theory when this particle was
discovered in the cosmic radiation by Carl Anderson in 1932 [114].
301
Fig. 4.1. Scheme of the Dirac energy spectrum for a free particle (left) and for an electron in the nuclear
field (centre). On the right hand side of the figure there is an energy spectrum with the scale shifted with
respect to the zero energy reference, Eqs. 4.34,4.35.
4.3.2
Solution of the Dirac equation for the hydrogenlike atom
The explicit solution of the Dirac equation for a hydrogenlike atom (in a.u),
D = c2 + c
p
+ V I,
h
(4.27)
was first published by Darwin [115] in 1928. He honoured Diracs outstanding works, Refs.
[4, 27], with words ...one of the great papers in physics in this century. Excellent modern
presentations can be found in later monographs, for instance Bethe and Salpeter [116], Moss
[117], Rose [118] or Messiah [119].
Finding fourcomponent solutions, Eq. 4.13, of the Dirac equation is similar to that for solving the Schrodinger analogue, Section 3.1. Eigenfunctions are written as products of radial and
angular terms what leads to the separation into individual angular and radial equations (for comparison see Eq. 3.3).
The eigenspectrum of the Dirac hydrogenlike operator, Eq. 4.27 (see Fig. 4.1), has three
separated parts: i) the continuum of negative energy states from to mc2 (c2 in a.u.), ii)
another continuum of positive energy states from +mc2 (+c2 in a.u.) to +, and iii) a discrete
spectrum of states between both continuum, leaned towards the bottom of the positive continuum.
In the case of the Dirac free particle (Eq. 4.12), the part of the energy spectrum iii) is empty.
Dirac quantum numbers There are four quantum numbers required to specify the state of a
Dirac hydrogenlike atom. These are similar, but not exactly the same as those of the Schrodinger
302
oneelectron atom, Section 3.1, where only three parameters are needed to determine the state
of the electron. The Dirac quantum numbers are:
i) The principal quantum number, n, which can have values of 1, 2, 3....
ii) The azimuthal quantum number, l, with values of 0, 1, 2, ..., (n 1). This parameter is usually
denoted by the alphabetical symbols s, p, d, f... However, contrary to the Schrodinger description
of the atom, the l number does not represent the orbital angular momentum.
iii) The angular quantum number, j. It can carry at most two values, |l 1/2|. This parameter is
written as a subscript of the alphabetical symbol of the azimuthal quantum number l.
iv) The last parameter is the magnetic quantum number, m, taking all halfodd integer values
from j to +j.
Regarding pictures of the electron distribution of the Dirac atom, they were published shortly
after the Diracs paper. In 1931 White [120] presented, as continuations to his famous electron
cloud pictures of a nonrelativistic Schrodinger atom, the electroncloud illustrations of the corresponding Dirac atom. For more recent pictures of spinors of the Dirac hydrogenlike atom the
reader is referred to works like that of Szabo [121].
Shapes of fourcomponent Dirac electronic wave functions are determined by quantum numbers j and m. Orbitals with identical values of j and m have the same shape, or identical angular
distribution. The s1/2 spinor is spherically symmetrical like the p1/2 one. The pair of d3/2
spinors has the same angular distribution as the pair of p3/2 . Furthermore, the p3/2,1/2 (j =
3/2, m = 1/2) resembles a dogbone rather than two touching ellipsoids. Some chemists
from the old school might be a little upset to learn that nonrelativistic perpendicular p orbitals
are no longer degenerate and look like a sphere p1/2 , a doughnut p3/2,3/2 , and a dogbone
p3/2,1/2 . All Dirac spinors have electron densities symmetrical about the zaxis and all have a
plane of symmetry perpendicular to the zaxis. However, none have nodal planes [121,122]. The
mathematical reason for the absence of nodes is due to the fact that all four components of the
wave function (Eq. 4.13) never have nodes all in the same place, because the probability density, Eq. 4.14, is never zero. From a physical point of view this feature is a natural consequence
of the correct handling of the angular dependence for an electron with spin. Schrodinger angular
distributions with nodes are caused by a nonexistent spinless electron. Likewise no nodes can
be observed already in the simplest case of an electron in a onedimensional box [116], because
whenever the large component has a node, the small component remains nonzero. Thus, the
answer to a wouldbe question of a student - How does a p electron pass from one lobe to another? - is that the physically more general Dirac theory is without nodes. Note, that the correct
nodeless results would be obtained already through the Pauli twocomponent theory. And it is
even possible to construct Schrodingertype orbitals from certain linear combination of the Dirac
fourspinors.
For the determination of nonrelativistic spectroscopic states of chemical systems one needs
eigenvalues of the angular and spin operators. However, the usual angular momentum operator
of the nonrelativistic theory, Section 3.1.1, does not commute with the Dirac Hamiltonian. It is
the operator, called the total angular momentum for a relativistic particle
j = l + 1 ,
(4.28)
303
which does. It has the spincomponent containing the Pauli matrices, Eq. 3.7, of the form:

0
.
(4.29)
0
Components of the operator of Eq. 4.28 satisfy all commutation rules for the angular momentum
operators:
[jp , jq ] = ijr ,
(4.30)
where (p, q, r) labels correspond to cyclic permutations of (x, y, z) components. Analogous to

the relations of the nonrelativistic Hamiltonian, and the angular operators, the square of the total
angular momentum and the zcomponent of the joperator commutes with the Dirac Hamiltonian, Eq. 4.27 :
D , jz ] = 0 , [h
D , j 2 ] = 0.
[h
(4.31)
As for the nonrelativistic counterpart, l2 , there is a similar eigen equation for the j2 :
j2 Y4 (, ) = j(j + 1)Y4 (, ),
(4.32)
where the eigenfunction Y4 (, ) is the fourcomponent spinor function depending on spherical

angles and .
The Dirac equation is rigorous only for a one particle system since it obeys all the requirements of the special relativity and quantum mechanics. With the hydrogen like atom as the analytically solvable model serves as the conceptual background for relativistic theories of many
electron systems.
4.3.3
Finite nucleus
In the nonrelativistic theory the electronnuclei interaction is usually described by the simple
Coulomb interaction, V = Zr (in a.u.), where r is the distance of the electron from the point
nucleus with the charge Z. Nonrelativistic orbitals have zero amplitude at the nuclei (centre)
except for sorbitals involving small exponential cusp of the form exp(r).
In relativistic calculations, however, the s1/2 and p1/2 spinors have a weak singularity at the
nucleus [123]. Thus it is customary to employ a physically more realistic finitenucleus model
which also improves the basis set convergence [124]. Perhaps the most popular nucleus model is
based on Gaussian distribution, as was proposed by Visser et al. [125]. The nuclear distribution,
, and the potential, V , are given by (in a.u.)
3
= Z( ) 2 exp(r2 ) ,
Z
V = erf ( r),
r
(4.33)
where erf (x) is the error function. The nuclear exponent is chosen in such a way that the
nuclear model has the same rootmeansquare radius as the experimental value for the nucleus.
Recommended values for nuclear parameters were published by Visscher and Dyall [126].
The choice of the nuclear model has negligible effect in cases where only valence spinors
are important for chemical properties. However, there are cases of properties where spinors
304
close to the nuclei are important and the nuclear model may have large influence, and one should
avoid unphysical point nucleus. Note that the computational implementation of nuclear attraction
integrals is easy as these are only modified twoelectron integrals expressed in a Gaussian basis
set.
4.4
Four component relativistic quantum chemistry
The relativistic theory of chemical systems appears to be more challenging with respect to
established nonrelativistic methods based on the Schrodinger equation. Corresponding four
component computational methods are not as good developed as their nonrelativistic counterparts. The reward for the development of more complex relativistic apparatus is its generality,
which fully recovers the nonrelativistic realm in the limit of c , providing in this way the test
of the correctness of the implementation. For this reason it is worth to have both nonrelativistic
and relativistic energy scales aligned.
The rest energy of the free electron, Eq. 4.21, mc2 , is equal to 0.511 MeV. This value, which
is much higher that the binding energy of the hydrogen atom - 13.6 eV, is defined as zero in the
nonrelativistic framework. Accordingly, the zero point in the relativistic energy scale is obtained
by subtracting the free electron rest energy. At the operator level this is achieved by subtracting
the diagonal 4 4 matrix from the term, Eq. 4.10, and using the modified 0 instead of in the
Dirac Hamiltonian

0
0
0
=
.
(4.34)
0 2I
Now we can write the Dirac equation for one electron in a molecular field as

L L
p
)
V
c (
=
E,
S
S
p
) V 2c2
c (
(4.35)
where V represents the electrostatic interaction of the electron with the clamped nuclei. Here
we wish to shed more light on the distinction between electron and positron solutions of the
Dirac equation, as it is not often precisely clarified in the literature. All solutions of the Dirac
equation, i.e. both of positive and negative energies, are truly electronic because the introduction
of potential by the principle of minimal electromagnetic coupling [127] requires specification
of charge [128]. For chemical purposes one sets the charge of the electron e = 1 a.u. and
not the charge of the positron, e = +1 a.u. This assignment was used by Dirac himself [129].
The observable positron solutions of the same potential can be obtained by choosing by the
charge conjugation of the negativeenergy solutions of the electronic problem, or are calculated
by explicitly choosing the positron charge in the potential.
4.4.1
The relativistic electronelectron interaction
The Dirac equation is valid only for the oneelectron system. To extend it for manyelectron
systems the proper relativistic electronelectron interaction is to be taken into account. The
straightforward approach is to augment the oneelectron Dirac operator with a twoelectron
interaction term. When we consider the Coulomb interaction, what is the simplest case, this term
305
is not Lorentz invariant. Although an invariant interaction potential can be derived from quantum
electrodynamics (QED), the fully Lorentz invariant form of the electronelectron interaction term
can not be written in a closed form [123]. Therefore, in practise one prefers truncation to the so
called Breit operator [130],
Breit
gij
=
(i rij )(j rij ) i

1 h
,
i j +
2
2rij
rij
(4.36)
which is added to the Coulomb interaction term, r1ij . The Breit operator accounts for magnetic
interactions and retardation effects up to the order of c12 .
The first term in brackets on the righthand side of Eq. 4.36 is the Gaunt interaction term
[131], often used as an approximation of the Breit operator since it includes the largest portion of
the full Breit form. The Gaunt operator contains spinspin, orbitorbit and spinorbit contributions, while the second term in Eq. 4.36 is the gauge interaction representing retardation effects
induced by the finite speed of light. Though the Gaunt term itself is not gauge invariant, it is easy
to implement within twoelectron integral generating codes.
4.4.2
The DiracCoulombBreit/Gaunt Hamiltonian
The DiracCoulombBreit/Gaunt (DCB/DCG) operator represents nowadays the most rigorous

representation for manyelectron relativistic effects. It is formed by selecting the Dirac operator
in a molecular field as the oneelectron operator and the Coulomb term, possibly extended by
the Gaunt or the full Breit twoelectron term, Eq. 4.36. Keeping the lowest order Coulomb term
is the often used approximation of the electronelectron interaction because the twoelectron
magnetic interaction terms are typically small.
For the purpose of relativistic quantum chemical calculations one works within the Born
Oppenheimer approximation [54] as in the nonrelativistic framework, Section 3.2. This approximation introduces very small errors in the relativistic quantum mechanical treatment, but
corrections can be applied in the same manner as in the nonrelativistic theory [132].
The aim is to solve the molecular Dirac equation for n electrons localized in the static field
of N nuclei. The DiracCoulombBreit/Gaunt Hamiltonian for this general system is
DCB/DCG =
H
n
X
i
n
N
X
X
ZI ZJ
Breit/Gaunt
D (i) + 1
gij
+
.
h
2
RIJ
i6=j
(4.37)
I<J
D (i), is identical with the operator in Eq. 4.35 where the external
The oneparticle operator, h
potential is created by static atomic nuclei. Note that the classical potential of the electron in the
point nuclear field is usually replaced by the finite nuclear distribution, Section 4.3.3.
In comparison to the nonrelativistic manyelectron Hamiltonian, Eq. 3.11, it is the one
electron operator that makes the difference which is sometimes accompanied with the gij electronelectron interaction term in its relativistically specified form, see Section 4.4.1.
Brown and Ravenhall problem. A discussion arose about the validity of the DCB/DCG operator. The oneelectron Dirac equation is not bounded from below and the same holds for the
306
manyelectron case. Brown and Ravenhall [133] argued that one can not find stable stationary
states for this Hamiltonian (what is called as the BrownRavenhall disease). This is due to the
fact that when the electronelectron interaction is treated as a perturbation, one can construct an
infinite number of states for electrons as well as for positrons, causing final states to be completely delocalized. For the case of a twoelectron system (like the helium atom) this can be
demonstrated as follows: Let us assume a double excitation from an orbital, where one electron
ends in the negative energy continuum and the other lands in the positive energy continuum.
The sum of their orbital energies is the same (or very close to) as the initial unperturbed energy.
Theoretically, when the Coulomb electronelectron interaction is turned on, what would happen
is transition to these doubly excited states and subsequent decay of former twoelectron bound
state into the continuum. However, the real helium atom does not disintegrate into electrons in
the positive and negative continuum bands. As a remedy Brown and Ravenhall suggested to
restrict the Hamiltonian only for positive energy states by applying projection operators. Once
negative energy (positron) solutions are abandoned, the corresponding theory becomes what
is sometimes called as the fourcomponent theory for electrons only, or the no pair approach.
However, later it has been found that for practical fourcomponent calculations it is sufficient to
have a proper basis set (see Section 4.4.6) and keep only positive energy solutions [134].
4.4.3
The DiracHartreeFock method
The relativistic analogue of the HartreeFock self-consistent field method (see Section 3.2.1), the
DiracFockBreit method, was first formulated by Bertha Swirles [6] in 1935. The story goes
that suggestion came to her by Douglas R. Hartree, who is associated with the HartreeFock
theory [57, 58], when they were waiting at the London train station [135].
The resulting theory has been implemented first to atoms [7]. Molecular closedshell formalism where fourcomponent spinors were used as basis functions were proposed by Malli and
Oreg [136] in 1975. Later more flexible basis sets were employed for both components coupled
by the kinetic balance condition [137139], see Section 4.4.6.
Both HartreeFock and DiracHartreeFock methods are independent particle models where
each electron feels the field of nuclei and the average field of other electrons. The derivation
of DiracHartreeFock equations is the same as in the nonrelativistic framework. Shortly, one
seeks the wave function 0 , which makes the energy functional (for comparison see Eq. 3.18),
stationary
EDHF =
DCB/DCG |0 i
h0 |H
.
h0 |0 i
(4.38)
The 0 wave function is the antisymmetrized product (the Slater determinant, Eq. 3.12) of relativistic oneparticle functions, in this case built of fourcomponent spinors, Eq. 4.13. Here we
will label the (i) as the fourcomponent spinor representation of the oneelectron state. The
expectation value of the energy can be written as
EDHF =
N
occ
X
D |i i + 1
hi |h
(hi j |
gij |i j i hi j |
gij |j i i).
2 i,j
i=1
N
occ
X
(4.39)
307
The DiracFock operator has the same structure as its nonrelativistic analogue, Eq.3.20:
D + u
fDHF (i) = (h
DHF )(i) = i (i),
(4.40)
where the Coulomb and exchange operators involved in the u

DHF term depend on fourspinors,
and are determined in a selfconsistent manner.
Since we are seeking electronic states solutions, orthogonal spinors for the Fock operator,
Eq. 4.40, must be taken from the positive part of the spectrum. And as we do not allow electrons
to enter negative (positron) states, the fourcomponent energy operator remains bounded from
below (also thanks to the kinetic balance condition, see Section 4.4.6), and clear separation
between negative and positive part of the energy spectrum is retained. The so called Brown
Ravenhall problem, Section 4.4.2, is not relevant for DiracHartreeFock calculations, although
this so called continuum dissolution was a discussed topic in the early eighties.
For the single determinant openshell system one may employ, for instance, the average
configurations formalism [140, 141]. Available is also the fourcomponent multiconfiguration
SCF version [142, 143], see also Ref. [144].
Solutions of the DHFequations couple one-electron spin s and orbital l angular momenta
(2)
(2)
(4)
(4)
through the jj or coupling. Atomic jj split shells are as follows: s1/2 , p1/2 , p3/2 , d3/2 ,
(6)
(6)
(8)
d5/2 , f5/2 , f7/2 etc., for comparison see Figs. 2.3 and 2.4. Example of molecular coupled
(2)
(2)
(2)
shells (1/2 , 1/2 , 3/2 ) is given in Fig. 2.5.

Relativistic effects (see Section 2), which are observable already in solutions of the hydrogen
Dirac equation, Section 4.3.2, are for chemical systems fully manifested within the DHF method.
Besides the spinorbit splitting of l > 0 shells these can be briefly summarized as i) contraction
and stabilization of s and p shells which leads to the strengthening of the chemical bond, and ii)
relativistic expansion of d and f spinors causing the weakening of the bond.
4.4.4
Symmetry aspects
Computational costs of calculating the energy and properties of any chemical system can be
greatly reduced by considering the symmetry of the molecule in question.
Kramers symmetry Since we are in the relativistic domain, the spinsymmetry, Section 3.1.2,
can not be used for this type of relativistic operator. Note, however, that by removing spinorbit
interaction terms from the relativistic Hamiltonian, i.e. keeping the spinfree form, one can
employ the spinsymmetry and singlepoint group symmetry.
The relativistic replacement of the spinsymmetry is called the time reversal symmetry [119],
and is represented by the antiunitary timereversal operator which has the form
0
y
= i
0,
K
K
0
y
(4.41)
0 is the complex conjugation operator. The timereversal operator commutes, in the

where K
absence of the magnetic field, with fourcomponent energy operator, Eqs. 4.37 and 4.40.
Due to the timereversal symmetry DiracHartreeFock solutions come out in degenerate
pairs, known as Kramers pairs [145,146]. This is the consequence of the Kramers theorem [145]
308
claiming that energy levels of a system with halfinteger spin are at least doubly degenerate,
and any degeneracy is evenfold. The theorem does not hold in the presence of magnetic field.
is
Relation between two degenerate spinors, representing oneelectron states, and ,
K
= ,
K(a)
= a ,
(4.42)
where a is an arbitrary complex number.

Doublegroup symmetry In the nonrelativistic theory one deals with point group symmetry
operations acting only on spatial parts of the chemical system. However, with the Dirac Hamiltonian invariant under symmetry operations, these must be a product of spatial and spin operations,
as this fourcomponent operator is coupling both. Therefore, a more general point group comprising both spatial and spin coordinates has to be introduced. To found the ground for the new
group theory, rotation by 2 is treated not as an identity operation but as a symmetry operation.
Any ordinary rotation group is expanded by taking the product of this new operation. The number
of symmetry operations is thus doubled and these groups are called double groups. Besides the
irreducible representations (irreps) used in the nonrelativistic theory, called boson irreps, there
are also irreps that describe oneelectron functions of the DiracCoulomb Hamiltonian, known
as fermion irreps. The number of fermion irreps is lower than the number of boson irreps. For
example, a molecule of the D2h symmetry has eight boson irreps but only two twodimensional
fermion irreps, E1/2g and E1/2u . Likewise the symmetry group of water, C2v , has four different (onedimensional) boson irreps, but only one (twodimensional) fermion irrep, E1/2 . This
means that fewer twoelectron integrals are zero due to the double group symmetry. For that
reason the rigorous fourcomponent symmetry apparatus has to be applied appropriately. Let us
consider a molecule that has only light elements and its energy is computed by using the Dirac
Coulomb Hamiltonian in the double group symmetry. Calculations are much more demanding
due to many twoelectron integrals which need to be computed, while most of them are nearly
zero. In this case it is sufficient to resort to onecomponent quantum chemical methods with
adapted single point group symmetry.
The theory for the double group symmetry was first presented by Bethe [147]. There are
several books [148, 149] of this subject available to the reader, together with many works on
the usage of double groups for constructing relativistic orbitals like Refs. [150157]. Peng et
al. [158] published recently scheme for constructing Kramers paired double group symmetry
functions.
4.4.5
Classification of relativistic electronic states of atoms and molecules
For atomic systems one takes into consideration also the so called jj coupling convention as the
Dirac Hamiltonian commutes with the operator of the total angular momentum, Eq. 4.31. One
adds the third number, J, representing the size of the total angular momentum, to the spin (S)
and angular (L) quantum numbers. For example, the ground state of the thallium atom is X2 P1/2 ,
meaning that the valence 6p(1) electron occupies the 6p1/2 (l = 1, s = 1/2, j = 1/2) relativistic
spinor. The lowest excited state is 2 P3/2 (l = 1, s = 1/2, j = 3/2). Note that the 2 P1/2
2
P3/2 spinorbit interaction splitting serves as a testing ground for a variety of quasirelativistic
methods, see Ref. [159].
309
Another case is the lead atom with the valence configuration of 6p(2) . While in the spinfree
framework we have X3 P, 1 D, and 1 S states, the spinorbit coupling gives rise of more states (in
ascending order): X3 P0 , 3 P1 , 3 P2 , 1 D2 , and 1 S0 .
For diatomic molecules one employs the so called coupling where the total angular
momentum is
= | + S|.
(4.43)
The number is assigned by the group theory as a direct product between the irreducible representation of the spin and the irreducible representation of the total angular momentum. First, let
us have a look at the correspondence between nonrelativistic and relativistic symmetry states. In
the notation the + irrep is designed as 0+ , as 0 , and , , ... as 1, 2, 3... The
E1/2 double group representation is labelled as 21 , the E3/2 as one 32 in the notation. As
an example consider a 3 state of a diatomic. The spatial symmetry (angular momentum) is
, the triplet spinsymmetry (S=1) spans and irreps. The direct product of the spin and
spatial symmetries is (spin) (spatial) = + and (spin) (spatial) = . So
the spinorbit interaction splits a 3 state into 0+ (+ ) and 1 () states.
In the 2 nonrelativistic state the doublet spin corresponds to E1/2 double group representation. Multiplying the spin and spatial irreducible representations, E1/2 (spin) (spatial) =
E1/2 + E3/2 , gives = 12 and 23 electronic states, respectively.
And in the last example, shortly, the 3 state of some diatomic molecule leads to these
split states (in ascending order): 0+ , 0 , 1, 2. The reader can found more in Ref. [160].
4.4.6
Fourcomponent relativistic basis set functions
A common way of solving the molecular DiracFock problem, Section 4.4.3, lies in the expansion of molecular spinors in a set of analytic basis set functions. This is borrowed from the
nonrelativistic realm where the basis set expansion of the HFSCF oneelectron wave functions,
Eq. 3.24, was introduced by Roothaan [63].
In the relativistic domain the oneparticle fourspinors are expanded in a separate scalar basis
functions set for large (L) and for small (S) components:
L
P L L
c
p
Pi iL ip
L
L
p
ip
= Pi iS cS
p =
(4.44)
S
c
p
i i ip
P
S S
S
p
i i cip
The four sets of coefficients, {cL } , {cL }, {cS }, and {cS } are in general complex. The
upper (L) and lower (S) components of a fourcomponent molecular spinor are expanded as a
linear combination of primitive Cartesian Gaussian type of orbitals,
L
L k i m i oi
L
z exp(iL r2 ) , Si = NiS xki y mi z oi exp(iS r2 ),
i = Ni x y
(4.45)
where N is a normalization constant and k, m, o are parameters whose sum gives the angular
momentum of the atomic orbital.
310
The large {L } and small {S } component basis functions are related through the so called
kinetic balance condition [137139]:
{S } =
1
p
){L }.
(
2c
(4.46)
This condition is nowadays, starting from a given (decontracted) large component set, L , used
for constructing the relativistic small component basis set. Note that Eq. 4.46 is derived in the
same way as Eq. 4.26 with employing the shifted energy scale, Eqs. 4.34 and 4.35. In pioneering
years of the relativistic electronic structure theory the negative energy continuum was sometimes
considered to be guilty for the failure to get upper bounds of the total energy what was known
as so called variational collapse. However, this was caused by a poor basis set causing bad
finite matrix representations of relativistic operators [139]. By employing the balanced basis
the unphysical positronic solutions of the DHF method are easily identified according to their
eigenvalues which are about (in absolute value) 2c2 a.u.
Thanks to the basis set expansion, Eq. 4.44, the DiracHartreeFock problem, Eq. 4.40, can
be reformulated into the matrix representation (analogous with Eq. 3.26)
FDHF C = SC,
(4.47)
which is suitable for an efficient computer implementation. Dyall and coworkers [161165] are
nowadays intensively developing basis sets for allelectron fourcomponent relativistic calculations.
p
, acting as a partial derivative with respect
The small component depends on the operator
to x, y, z coordinates. By applying Eq. 4.46 on large component of Eq. 4.45 with angular momentum l one obtains two small component functions with angular momentum l 1 and l + 1. The
kinetic balance prescription generates the small component basis set which is about twotimes
of the size of the corresponding large component basis set. This already indicates that solving
a DHF problem can be quite expensive when compared to the nonrelativistic counterpart. For
the construction of the twoelectron Fock(Dirac) matrix twoelectron atomic integrals (over the
Coulomb interaction operator) are sorted into three classes, (LL|LL, (LL|SS), and (SS|SS).
Dyall and Fgri stated [25] that in addition to approximately equal number of (LL|LL) integrals comparing to the nonrelativistic framework, the number of (LL|SS) integrals is about 13
times larger than (LL|LL) integrals and the number of (SS|SS) integrals is approximately 39
times larger. The overall computational cost for a DHF calculation in an uncontracted basis set
is about 53 times larger than in the corresponding nonrelativistic domain. However, the effect of
(SS|SS) integrals is often negligible and their number can be reduced substantially by adopting
simple Coulombic correction [166]. In the nonrelativistic limit (c ) small component basis
functions, S , according to Eq. 4.46, disappear and one works with the large component basis
set only.
4.5
Relativistic correlation methods
The starting point in fourcomponent relativistic correlation methods is the DiracCoulomb

Breit/Gaunt Hamiltonian, Eq. 4.37, in the nopair approximation, i.e., neglecting positron
electron pair creation. Only electronic states are treated in relativistic correlation calculations.
311
By using Kramers unrestricted spinors which correspond to spinorbitals in the nonrelativistic

theory, the second quantized form of the Hamiltonian can be expressed by Eq. 3.40. In the
relativistic case the fpq are matrix elements over the FockDirac operator, Eq. 4.40. Regarding
the twoelectron integrals, the obvious step in correlated calculations is their transformation from
atomic orbital basis to molecular orbitals (spinors) basis. Expression for twoelectron integrals
in the basis of the fourcomponent Kramers unrestricted spinors, based on the expansion of
Eq. 4.44, can be written as
XX X
X Y X Y
Y
X Y
hp q |r s i =
Cip
Cjq Ckr Cls hX
(4.48)
i j |k l i,
X,Y , i,j,k,l
where X and Y run over large (L) and small (S) components, and over and components. The p , q ... are fourspinors of arbitrary electronic states, either Kramers restricted or
Y
X Y
not, coming out of the DiracHartreeFock procedure, Section 4.4.3, and hX
i j |k l i are
twoelectron integrals over (L/S)atomic basis functions, described by Eq. 4.45. Due to the
presence of atomic basis functions over small component basis functions obtained by Eq. 4.46,
the transformation of fourcomponent spinors, Eq. 4.48, is usually computationally most expensive part of correlated relativistic calculations. Various speedups have been reported, like that
of Abe et al. [167].
4.5.1
Secondorder Kramers restricted MP2 method
The simplest and relatively tractable correlation correction to the HartreeFock theory is the
second order MllerPlesset (MP2) energy [168], given by the general expression for the nonrelativistic Hamiltonian
E (2) =
Nocc N
virt
X
|hijkabi2 |
1 X
.
4 ij
i + j a b
(4.49)
ab
In connection with fourcomponent relativistic methods we refer the implementation for the
Kramers restricted closedshell DHF wave function as was done by Lrdahl et al. [169]. Being
in the relativistic domain, it employs the relation between degenerate Kramers partners, Eq. 4.42,
connecting twoelectron integrals with barred and unbarred spinors:
hijkabi
hijkabi
hijkabi
hijkabi
= hijkabi,
= hijkabi,
= hijkabi,
= hijkabi,
hijkabi
hijkabi
hijkabi
hijkabi
= hijkabi,
= hijkabi,
= hijkabi,
= hijkabi.
(4.50)
Inserting these formulas into Eq. 4.49 and carrying out algebraic manipulations we get the simplified expression for the Kramers restricted MP2 correlation energy
E
(2)
No /2 Nv /2
1
1 X X
(|hijkabi|2 + |hijkabi|2 +
=
2 ij
i + j a b
ab
+|hijkabi|2 + |hijkabi|2 + |hijkabi|2 +

+|hijkabi|2 + |hijkabi|2 + |hijkabi|2 ).
(4.51)
312
The formula Eq. 4.51 is implemented in a direct and parallelized way in the DIRAC [170] program suite. For that reason an average cost for calculation of the relativistic Kramers restricted
MP2 energy is negligible with respect to the cost of the full fourindex transformation, Eq. 4.48,
which is necessary for higher level correlation methods, Section 4.5.2, even when employing the
Kramers restriction.
By replacing the Kramers restricted spinors, Eq. 4.42, with pure spin restricted spinorbitals,
Eq. 3.13, one can further reduce the MP2 expression to the form of the nonrelativistic limit. This
is due to the spinorthogonality, where densities (pq) = (pq) with an odd numbers of bars
(equivalent to spin) have no contributions. This leads to the simplified expression
E (2) =
No /2 Nv /2
1 X X
1
(|hijkabi|2 +
2 ij
i + j a b
(4.52)
ab
+|hijkabi|2 + 2|hij|abi|2 ),
and, after the spinintegration we finally get the familiar nonrelativistic expression:
(2)
EN R =
Nocc /2 Nvirt /2
1 X X hij|abi(2hij|abi hij|bai)
.
2 ij
i + j a b
(4.53)
ab
To compare the effectivity of nonrelativistic and fourcomponent relativistic MP2 calculations one has to take into consideration the transformation over small component basis functions
as the main computational bottleneck, both at the DiracHartreeFock and the correlated MP2
levels, together with higher number of terms in Eq. 4.51 in comparison to Eq. 4.53. The relativistic complete active space second order perturbation theory with the fourcomponent Dirac
Hamiltonian has been reported by Abe et al. [171].
4.5.2
Relativistic Coupled Cluster and Configuration Interaction methods
The Kramers unrestricted Coupled Cluster method with singles, doubles and noniterative connected triples CCSD(T), applicable for a wide scale of atomic and molecular systems, was implemented by Visscher and coauthors [172]. Authors utilized the spinorbital form of the CC
equations (see Section 3.2.5) in the framework of the effectively implemented doublegroup
symmetry. Parallelization of the code was carried out by Pernpointer and Visscher [173]. The
Kramers restricted CC version was also reported [174], but this program is much less used.
There is another group of relativistic CC methods based on the Fockspace Kramers unrestricted CC expansion, which give a variety of excited states, see Refs. [175179].
Relativistic Kramers restricted CI methods are discussed in Refs. [180183]. Note, that the
comprehensive theory on the generalized Kramers restriction of spinors applicable for openshell
systems was published by Fleig [184].
4.6
Transformations to twocomponent Hamiltonians
Fourcomponent Diracbased theories are connecting positronlike and electronlike degrees of freedom. In addition, fourcomponent description of the shell structure is always associated with small component basis functions, as mentioned in Section 4.4.6, what increase
313
computational time. Without the coupling between the states with negative energy (positrons)
and the states of positive energy (electrons) one would have a possibility to focus on the latter and work solely with the electronic twocomponent wave function. A high computational
cost of fourcomponent methods is the stimulus towards the development of less demanding
twocomponent Hamiltonians dealing with electronic states only, and without the need of small
component basis functions. Methodologies for decoupling charge conjugated degrees of freedom
leads to the appearance of a variety of twocomponent Hamiltonians accurate to various orders
of some expansion parameters.
The following attributes are required for approximate relativistic Hamiltonians: i) They
should be accurate enough to give close enough results to corresponding fourcomponent operators. ii) They are to be well balanced for treating chemical systems containing a wide variety
of atoms, i.e. both heavy and light. iii) Variational stability together with effectiveness to apply
them to moderate and large molecules containing heavy elements.
For twocomponent methods which include spinorbit coupling at the orbital level formation
the Kramers restriction and the doublegroup symmetry, Section 4.4.4, is applicable. Hence one
can utilize the very same (four)spinors based correlation apparatus, like MP2, Section 4.5.1, CI
or CC methods, Section 4.5.2. Let us stress that the main computational saving is due to abandoning small component basis functions, both at the HartreeFock level and at the transformation
level, Eq. 4.48, what makes them attractive for everyday relativistic calculations.
Twocomponent methods can be divided into allelectron and pseudopotential methods.
We focus here on the former group, regarding the latter, we refer the reader to the literature like
Ref. [185189].
In subsequent section we present two known approximate decoupling schemes, the Foldy
Wouthuysen transformation and the DouglasKroll transformation. Afterwards we demonstrate
how to perform the decoupling transformation up to the infinite order.
4.6.1
FoldyWouthuysen transformation
The idea of Foldy and Wouthuysen (FW) [190] was to separate the electronic and positronic
solutions of the Dirac equation by the Van Vleck type unitary decoupling transformation of the
Dirac Hamiltonian in the form
4comp U .
H 0 2comp = U H
(4.54)
The Foldy-Wouthuysen transformation can be specified in a number of different forms. The most
common is the exponential form. Moreover, it can be also expressed as a product of a series of
unitary transformations, U = ...U2 U1 U0 .
For one particle in an external potential Vext and within the single transformation, U =
U0 = exp(iS) one gets, after applying the Taylor expansion in the fine structure constant 1/c,
the following oneelectron terms:
4
2
1
BP = (mc2 + Vext + p ) p + 1 (
Vext p)
+ ... (4.55)
p2 Vext ) +
(
h
3
2
2
2
2m
8m c
8m c
4m2 c2

The definition of S in the exponential above, U0 = exp p
p (p) with a yet unknown parameter (p), as used by Foldy and Wouthuysen [190] in their original work is described in
314
the literature, see Ref. [15, 26]. It is the mandatory starting point for all decoupling schemes
employing unitary transformations of the Dirac Hamiltonian. The first terms in parenthesis in
Eq. 4.55 represent the usual nonrelativistic Hamiltonian, then the second one is the so called
massvelocity term, the third is called the Darwin term and the fourth operator describes the
spinorbit interaction. It can be analytically proved that the scalar massvelocity and Darwin
terms are unbounded from below [15]. Carrying out the FW transformation on the entire Dirac
CoulombBreit Hamiltonian, Eq. 4.37, one obtains the full twocomponent BreitPauli (BP)
operator, which in short notation has the form
X
X
BP
BP (i) +
BP =
H
h
gij
.
(4.56)
i
i<j
BP
The transformed electronelectron interaction operator, gij
contains the Coulomb, spinorbit
and also other interaction terms, like spinspin and orbitorbit terms along with numerous high
order terms.
The main disadvantage of the resulting BreitPauli Hamiltonian, also known as the first order
relativistic Hamiltonian, is that its terms are highly singular and variationally unstable. Therefore
this operator is suitable to be used in the first order perturbation theory.
4.6.2
DouglasKroll transformation
Douglas and Kroll (DK) pointed out that the FW transformation in an external field yields highly
singular operator which is not suitable for practical calculations. Their approach decouples the
large and small components of the Dirac fourspinors in the presence of an external potential
through successive unitary transformations. The DK transformation is adopting the external
potential, Vext , as an expansion parameter rather than the speed of light, c, in the FW transformation.
The very first step in the DK transformation is the freeparticle FW transformation of the
Dirac Hamiltonian with the external potential, Eq. 4.27, in the momentum space. Such a term
is often referred as the first order DK Hamiltonian, but is unsuitable for practical computations.
The DKproposed operator is of the type
n = (1 + Wn2 )1/2 + Wn ,
U
(4.57)
which is unitary if Wn is antiHermitian. We shall not go into details of the derivation, but
rather give a final form of the DKtransformed operator. The twocomponent oneelectron
Hamiltonian is then
DK = Ei +V ef f (i) = Ei +Ai (Vext (i)+Qi Vext (i)Qi )Ai 1 [[Ei , W1 (i)]+ , W1 (i)]+ , (4.58)
h
2
where [a, b]+ is the anticommutator. Exact form of the Ei , Ai , Qi , W1 (i) terms is beyond the
scope of the present work and the reader can found them in the literature, like Ref. [15].
Applying the freeelectron transformation on a twoelectron operator, including the Breit
interaction, Eq. 4.36, we get the twoelectron DouglasKroll operator:
X
X ef f
DK =
H
(Ei + V ef f (i)) +
Vij .
(4.59)
i
i<j
315
We point to the Dirac relation enabling separation of the spin and spatial operator terms
= a
+ i
a
)(
b)
b
(
(
a b),
(4.60)
which gives individual spin and scalar (spatial) parts of the DouglasKroll transformed operator:
DK =
H
X
X ef f sf
sf
DK
DK
(Ei + Vef
(Vij
+ gSO
ij ),
f (i) + hSO (i)) +
i
(4.61)
i6=j
DK (i) and gDK are one and twoelectron spinorbit terms, respectively. For a dewhere h
SO
SO ij
tailed meaning of the terms we refer the reader to the specialized literature, like Ref. [191].
The DK transformation to the second order in the external potential has been extensively
studied by Hess and coworkers [95, 192]. The third order (DK3) was implemented by Nakajima
and Hirao [193], who applied the DK3 transformation onto the twoelectron Coulomb term
[194]. Previously Park and Almlof [195] applied only the first order spinfree DK transformation
on the Coulomb interaction operator.
In recent years the DouglasKrollHess (DKH) transformation has been enhanced up to
higher orders. Van Wullen reported the generalized transformation through the sixth order [196].
Reiher and Wolf invented a generalized transformation scheme up to the infinite order [197,198].
4.6.3
Exact decoupling
We shall give here alternative derivation of an exact twocomponent Hamiltonian and show
afterwards its truncation to known approximate forms. First we show equivalence of two approaches for the exact decoupling of negative and positive energy solutions of the DIRAC equation: i) elimination of small components and ii) unitary decoupling transformation.
Similarly as was done for the free particle, Eq. 4.18, we express the Dirac equation for a
charged particle in the potential field, Eq. 4.35, as a system of coupled equations for large and
small components:
11 L + h
12 S = EL
h
L
21 + h
22 S = ES .
h
(4.62)
representing the exact coupling between small and large compoLet us introduce the operator R
nents:
L.
S = R
(4.63)
First we eliminate small components in Eq. 4.62 by using Eq. 4.63. We get
11 + h
12 R)
L = EL .
(h
(4.64)
Than one has to introduce the normalization of individual largecomponents since these are not
normalized, although enveloping fourcomponent wave functions are normalized to themselves.
+ , can be extracted from normalization condition of the four
The normalization operator, N
component wave function
L
R)h
h|i = hL |L i + hS |S i = (1 + R
|L i = 1,
(4.65)
316
and is equal to
q
+ = (1 + R
R).
(4.66)
+ L
The resulting exact Hamiltonian for the normalized twocomponent wave function = N
thereby becomes (the subscript + is representing positive energy solutions, negative solutions):
q
h
i
h
i
11 + h
12 R
+ = N
11 + h
12 R
R
h
p
+ h
N
1 = 1 + R
h
+
1
R
1+R
(4.67)
to decouple solutions of positive and of negative

We are looking for unitary transformation U
energy. The general form (for comparison see also Eq. 4.54 ) comes from Heully et al. [199]:
11
h
21
h
12
h
22
h
=
U
+
h
0

.
(4.68)
Kutzelnigg [200] found the advantage of writing the unitary operator matrix as the product of
two transformations
=W
1W
2 :
U
1=
W
R
1

;
2 =
W
1
N
+
0
0
1
N
+
N
1+R
.
R
1+R
(4.69)
1 leads to the decoupling:

The first transformation with W

11 + h
12 R
21 + R
22 R
+R
h
h
Rh11 Rh12 R + h21 + h22 R
11 R
12 R
21 R
22
+ h
h
+R
h
h
Rh11 R R h12 h21 R + h22

.
(4.70)
It is clear for the exact decoupling the offdiagonal elements, which are Hermitian conjugated,
have to be equal to zero. This gives relation which is valid only for the exact R:
21 + h
22 R
11 + R
21 R.
=R
h
h
(4.71)
2 introduces the renormalization, thus ensuring the unitarity of the

The second transformation W
transformation. In this final step we obtain the Hamiltonian for the positiveenergy solutions
h

i
11 + h
12 R
21 + h
22 R
+ = p 1
+R
h
p 1
h
,
(4.72)
h
R
R
1+R
1+R
which is identical to Eq. 4.67, when we use the coupling equation, Eq. 4.71.
We have shown the equivalence between the elimination of the small components and the
unitary decoupling transformation for the case of the exact coupling between the large and small
components represented by R.
317
Regarding the twocomponent wave functions, we consider the decoupling transformation,

Eq. 4.68, of the matrix of fourcomponent eigenvectors,
L

L
,
(4.73)
U = W2 W1
S
S
+
which gives
"
L
e+
S
e
+
L
e
S
e
=W
2
L
S
+
+R
+
S
L
+
+ R
L
S
+R
S
L

R

.
(4.74)
After the exact decoupling small component wave functions for positive energies and large comS
L
ponent wave functions for negative energies are zero, e+
= 0 and e
= 0. This implies (after
making off-diagonal blocks of Eq. 4.74 equal to zero):
S
L
+
+
= R
;
L
S

= R
,
(4.75)
couples the large and small positiveenergy eigenvectors. The final renormalized
One sees that R
nonzero twocomponent functions for separated positive and negative states therefore are :
p
p
L
S
S
L
R
+
R

(4.76)
= 1+R
e
= 1+R
.
e+
The exact coupling between the large and small components of the Dirac equation, Eq. 4.35, can
be expressed as

1
EV
= 1 B(E)
b
b
p
) ; B(E)
R
(
= 1+
.
(4.77)
2mc
2mc2
This coupling is, however, energydependent and thus statespecific. Heully et al. [199] developed a stateuniversal quadratic equation. The coupling equation obtained in a kinetically
balanced basis set (Eq. 4.46) forms the basis of the iterative XQR scheme proposed by Kutzel in Eq. 4.63 will refer to the state universal coupling operator, in
nigg and Liu [201]. The R
agreement with the notation used in the paper by Heully et al. [199]. Finally, one may note that
the resulting working equation for electronic states,
+ eL = E eL ,
h
+
+
can be rearranged to
h

i

11 + h
12 R
21 + h
22 R
+R
h
L = E 1 + R
R
L .
h
+
+
(4.78)
(4.79)
This corresponds to the Normalized Elimination of Small Components (NESC) [202204].

Truncation to approximate schemes Several twocomponent relativistic Hamiltonians can
be obtained using approximate couplings, for which Eq. 4.71 is no longer valid. The previously
mentioned BreitPauli Hamiltonian, Eq. 4.55, is extracted using the approximation
1
p
) ,
(
2mc
(4.80)
318
and keeping terms to the order of O(c2 ). Difficulties due to the unboundedness and singularity
of the Pauli operator are avoided in regular approximations (RA). These are constructed upon the
approximation
c
2mc2 V
p
) .
(
(4.81)
The ZORA (ZeroOrder Relativistic Approximation) Hamiltonian is obtained from decoupling

with Eq. 4.81 and no renormalization [205207]. Insertion of the approximate coupling, Eq. 4.81,
into the NESC equation, Eq. 4.79, gives the IORA (Infinite Order Regular Approximation) equation [208]. A precursor to IORA was the scaled ZORA approach [209], which is an approxiR
in Eq. 4.79 by its expectation
mation to IORA obtained by replacing the operator product R
value. Comparing the exact coupling, Eq. 4.77, with the regular approximation, Eq. 4.81, one
sees that the latter is obtained from the former simply by setting the energy to zero. Note that
most recently ZORA was implemented in a new relativistic treatment in the Quantum Monte
Carlo technique [210].
There is another decoupling strategy: performing first the freeparticle FoldyWouthuysen
transformation [190] which gives a twocomponent kinetic energy operator on square root form
what ensures the variational stability of the resulting operator. Subsequent decoupling in orders
of the external potential V (or in the inverse speed of light in c1 ) defines the previously mentioned DouglasKrollHess [94, 95, 192], Section 4.6.2, and BaryszSadlejSnijders [211216]
transformations, respectively. Alternatively, the small components may be eliminated and the
exact coupling obtained through iterative procedures [217219].
4.6.4
Infinite order twocomponent Hamiltonian (IOTC)
One may obtain the exact coupling [211, 213, 214, 216] in the finite basis approximation by solving the Dirac equation, Eq. 4.35, in some suitable basis set. The exact decoupling can be carried
out in a single step starting from a matrix representation of the Dirac operator in a molecular
field.
The coupling relations, Eq. 4.75, are in a finite basis set transformed into two systems of
linear equations
+L ;
Y+S = RY
YS ,
YL = R
(4.82)
L(S)
where Y+() is the block of the eigenvector matrix corresponding to the large (small) compo
nents of the positive (negative) energy solutions of the Dirac equation. The coupling operator R,
Eqs. 4.63, 4.77, now in a matrix form (without the hat symbol), can be obtained directly by
solving the first of system of equations, but in practice it turns out to be better to take the Hermite
conjugate of the second system and premultiply it by YS giving
h
i
h
i
AR = B; A = YS YS ; B = YS YL .
(4.83)
The A matrix is positivedefinite and Hermitian. The system can accordingly be solved by
a Choleski decomposition. The unitary transformation can now be constructed and the Dirac
Hamiltonian decoupled in matrix form.
319
Another important point is that in a rigorous construction of a twocomponent relativistic

Hamiltonian the decoupling transformation should be defined with respect to the complete electronic Hamiltonian, Eq. 4.37, not only to the oneelectron part, as is usually done in the literature.
Evidently, not only the oneelectron part, but the twoelectron part as well should be subject to
the change of picture transformation. However, this would bring the computational cost back
to the fourcomponent level [194, 195]. Therefore, in the twocomponent electronic Hamiltonian one prefers untransformed twoelectron Coulomb term, which implies that twoelectron
spinorbit contributions (see Section 4.7) are neglected.
Calculation of properties. Change of picture. Till now various transformations to the two
component form have been demonstrated solely on the Dirac energy operator. In the relativistic quantum mechanics, however, one works with a variety of fourcomponent property operators. For this reason, to be physically fully consistent at the twocomponent level, all relativistic
property operators have to undergo the very same transformation as the parental Dirac Hamiltonian. Clearly, when descending to the twocomponent framework, one has to adapt any
fourcomponent operator accordingly. The adaptation of the operator with respect to the wave
function is known as the change of picture [220, 221].
For the IOTC case the picture change transformation matrix U , Eq. 4.69, can be applied
not only for the Dirac operator, but also for any fourcomponent property operators in matrix
representation as well. Thus the picture change transformation of a fourcomponent property
operator in the basis set, X 4c , can be done easily:
2c
U X 4c U X++
.
(4.84)
It should be stressed, however, that the decoupling transformation U is constructed to block

diagonalize a specific oneelectron Hamiltonian, defined by the external scalar potential V appearing in Eq. 4.35. Hence, in general, the decoupling transformation does not blockdiagonalize
2c
, are obtained by
an arbitrary operator. Therefore, elements of the twocomponent operator, X++
selecting the (++)matrix block, see Eq. 4.73, of the transformed operator. This corresponds to
projecting the fourcomponent operator onto the space spanned by the positiveenergy solutions
of the particular Hamiltonian defining the decoupling transformation. At the algebraic IOTC
level the importance of the change of picture for expectation values was reported by Kedziera
et al. [222]. Note that relativistic picture change transformations of the twoelectron interaction
term have been reported in Refs. [194, 195, 223].
Implementation of the IOTC method The implementation was described by one of us (MI)
[224] and is distributed within the DIRAC08 program suite [170].
The Dirac equation, Eq. 4.35, is solved in the orthonormal (MO) basis [225] and subsequently
the IOTC Hamiltonian with positive energy solutions only is then projected back onto the large
component AObasis. This algorithm runs as follows:
1. Setting up the Dirac equation on matrix form in the AObasis
LL

S
0
AO AO
AO AO
AO
h ci = S ci i ; S
=
,
0
S SS
(4.85)
320
where S AO is the overlap matrix. Uncontracted Gaussian basis functions are used in the
implementation.
2. Transforming to an orthonormal basis using the nonunitary transformation
LL

V
0
O
MO
hM O cM
=
c
;
V
=
; V S AO V = I.
i
i
i
0
V SS
(4.86)
In our implementation the so called restricted kinetic balance (RKB) is enforced in this
step by embedding the transformation to Dyalls modified Dirac equation [225, 226] in the
transformation V , as described in Refs. [128, 225].
3. The eigenvalue problem, Eq. 4.86, is solved by the diagonalization and the coupling matrix
R is extracted using Eq. 4.83. The transformation matrix U , Eq. 4.69, is constructed, and
O
subsequently the positiveenergy twocomponent relativistic Hamiltonian hM
+ , Eq. 4.72,
is obtained in the matrix form.
4. In the final step the Hamiltonian is projected back onto the large component AObasis
LL; 1 M O LL 1
hAO
h+ V
;
+ = V
V LL
1
= V LL; S LL;AO .
(4.87)
The matrix representation of the IOTC Hamiltonian, hAO

+ , is then written to file with whatever transformed property operators are needed and all the subsequent calculations continue in the twocomponent framework.
In comparison to the scheme proposed by Jensen and Ilias [227] the preliminary freeparticle
FW transformation is completely avoided. Since this transformation was used by Barysz and
coworkers [211214], we will refer to the twostep approach as BSS. The direct construction
of the coupling matrix emphasised here was already proposed by Dyall [204] as well as Kutzelnigg and Liu [201, 228], although the latter authors seem to prefer an iterative approach. The
noniterative algebraic approach to the IOTC relativistic theory was published by Kedziera and
Barysz [229]. In the same group it has been shown recently how to recover fourcomponent
solutions by the inverse transformation of the IOTC wave function [230].
Tab. 4.1 displays ionization energies and electron affinities of the radon atom computed with
the Fockspace CoupledCluster method, see Section 4.5.2. We utilized fourcomponent DC,
and twocomponent BSS and IOTC Hamiltonians, all of them including both scalar and spin
orbit effects. The IOTC and BSS Hamiltonians give results closely matching the DC data.
Tab. 4.2 shows NMR shielding constants (see also Section 5.7), magnetizabilities and polarizabilities of the Rn atom calculated with the DC, IOTC and BSS Hamiltonians. These properties were obtained upon the DC/IOTC/BSS HartreeFock wave function with the second order
linearresponse module [232]. Twocomponent IOTC and BSS results agree with each other and
follow the corresponding fourcomponent DC values.
4.7
Spinorbit effects
The fourcomponent Dirac Hamiltonian contains by default both scalar and spinorbit effects.
Note that its modification is possible in order to distinguish between those relativistic effects
321
according to K. Dyall [226]. Dyalls adaptation of the Dirac Hamiltonian was enhanced by
adopting the so called quaternion symmetry by Visscher and Saue [225]. Although leaving out
spinorbit effects creates some computational speed up in the fourcomponent framework due
to the use of the single point group symmetry, small component basis set functions still remain
and cause a computational bottleneck.
Therefore a better way is transforming the fourcomponent Hamiltonian to a twocomponent
form, Section 4.6, and afterwards separate spinorbit terms according to Eq. 4.60. Here we give
three examples of SO Hamiltonians. Note that in following formulas for spinorbit operators we
prefer to use Pauli matrices (see Section 3.1.2) instead of spin operators, utilized in Eq. 2.11.
i) Applying the FW transformation on the DCB Hamiltonian, Section 4.6.1, one can extract
(from Eq. 4.56) the full spinorbit BreitPauli Hamiltonian [53] of the form
BP =
H
SO

X rij p
i
i
e2
h X X riI p
+
2
)
.
Z
i
j
I
i
3
3
4m2 c2
riI
rij
i
I
(4.88)
i6=j
Tab. 4.1. Ionization energies (IE) of the Rn atom and excitation energies (EE) of the Rn+ cation calculated
by the Fockspace CCSD method [in eV]. Absolute differences from DC values in parentheses. Data from
Ref. [224].
DC
IOTC
BSS
IE of Rna
1
S0 2 P3/2
10.5614
10.5304
(-0.0310)
10.5304
(-0.0310)
EE of Rn+
2
P3/2 2 P1/2
3.7706
3.8967
(0.1261)
14.8427
(-0.0281)
3.8953
(0.1247)
14.8420
(-0.0288)
P3/2 2 S1/2
14.8708
Experimental value of the first IE (1 S0 2 P3/2 ) is 10.7485 eV, Ref. [231].
Tab. 4.2. Magnetic and electric properties of the Rn atom. Absolute differences from DC values are in
parentheses. Data from Ref. [224].
Property
NMR shieldinga
Magnetizabilityb
Polarizabilityc
a
DC
IOTC
BSS
23348.4
-12.7160
27.066
23373.6 (25.2)
-12.7230 (-0.0070)
27.119 (0.053)
23372.4 (24.0)
-12.7236 (-0.0076)
27.122 (0.056)
NMR shielding constants in ppm.

Magnetizabilities in a.u.
c
Polarizabilities in a.u.
b
322
ii) Through the DouglasKrollHess (DKH) transformation, Section 4.6.2, one may separate (by
adopting Eq. 4.61) the variationally stable spinorbit Hamiltonian
XX Z
X
i
rij p
I
DK = e2
i B i Aj
i (ri p
i )Bi
(Bi Aj
H
h
Bi 3
SO
3
r
r
ij
iI
i
I
i6=j

i
rij p
j Bj Ai ) .
(4.89)
+2Bi Aj
3
rij
The DouglasKroll SO Hamiltonian, Eq. 4.89, contains projector operators and the BreitPauli
spinorbit Hamiltonian, Eq. 4.88, can be obtained by expanding and truncating the kinematic
factors Ai and Bi in both one and twoelectron operators, see, e.g., Ref. [15]. Note that the
twoelectron SO terms appearing in Eqs. 4.88,4.89 are known as the spinsameorbit (SSO)
and spinotherorbit (SOO) terms, respectively. Both of them are an approximation of the full
CoulombBreit/Gaunt interaction operator, see Section 4.4.1.
iii) Since it is difficult to compute different types of twoelectron spinorbit integrals, as
a cheap alternative one can utilize an effective oneelectron spinorbit operator of the general
form:
ef f =
H
SO
i
e2
h X X ef f riI p
i,
ZI
3
2
4m c2 i
riI
(4.90)
where Z ef f is a parameter. However, since the twoelectron contribution does not scale as the
oneelectron spinorbit term, this SOoperator is state dependent and therefore less accurate.
4.7.1
Meanfield spinorbit operator
Calculation of manycentre twoelectron spinorbit integrals is costly [159, 191, 233] and is
approximately equal to the cost of twoelectron integrals over smallcomponent basis functions,
Section 4.4.6. This disadvantage has been overcome by introducing a pseudo oneelectron Fock
type spinorbit operator, called as the meanfield spinorbit operator, MFSO. It is an effective
onecentre operator obtained from the full DKH atomic spinorbit Hamiltonian [234], Eq. 4.89.
Atomic matrix elements of the MFSO operator have the form:
DK
M F SO |jic = hi|h
DK (1)|jic + 1 Pmf orbs nM (2hiM |
hi|h
(1, 2)|jM ic
gSO
SO
M
2
(4.91)
DK
DK
hiM |
gSO
(1, 2)|M jic hM i|
gSO
(1, 2)|jM ic ) , c = (x, y, z),
where one and twoelectron spinorbit terms are given in Eq. 4.61. The twoelectron part of this
operator is represented as the summation running over occupied (meanfield) spatial M (l, ml )
atomic orbitals; nM are occupation numbers. The orbitals M can be taken either from contractions of the basis set, or, what was preferred here, from an interfaced internal atomic SCF code
producing scalar relativistic orbitals for a given set of orbital exponents.
A twocomponent Hamiltonian accounting for both scalar and spinorbit relativistic effects
can be simply constructed by combining available operators. In the previous work [235] we
have employed the nopair Hamiltonian containing the oneelectron spinfree DouglasKroll
Hess term, DKH2sf, which is accurate to the second order in the external potential, together
323
with the MFSO operator, Eq. 4.91. Note that from Section 5 onward the acronym DKH is used
solely for the DKH2sf operator. The twoelectron interaction was represented by the standard
nonrelativistic Coulomb interaction term.
DKH2sf +M F SO =
H
DKH2sf (i) + h
M F SO (i)) +
(h
X 1
.
r
i<j ij
(4.92)
The meanfield spinorbit operator can be expressed as:

M F SO = h
M F SO(x)
M F SO(y)
M F SO(z)
h
x + h
y + h
z .
(4.93)
The spinorbit term in the Hamiltonian is the reason why one can not utilize the spin symmetry and work with spinorbitals. One has to resort to general twocomponent spinors, L of
Eq. 4.13. The platform for the implementation was another fourcomponent relativistic program
suite, MOLFDIR [236], which fully utilizes the relativistic doublegroup symmetry, Section
4.4.4. MOLFDIR performs Kramers restricted four and twocomponent single determinant
HFSCF calculation on closed or openshell system (see Section 4.4.3), followed by electron
correlation methods, Section 4.5. MOLFDIR was slightly modified in such a way that its two
DKH2sf and h
M F SO operators.
component mode handles external oneelectron integrals of the h
The effectiveness of this operator was demonstrated on FO and ClO molecules in Tabs. 4.3
and 4.4. For these light systems DKH2sf+MFSO results follow closely fourcomponent DCG
counterparts. However, for treating heavier elements one has to resort to the Hamiltonian infinite
in both scalar and spinorbit terms, Section 4.6.3.
Tab. 4.3. Spin-orbit splitting of the 2 state of the FO molecule [cm1 ]. Calculated in decontracted basis
sets, 1s spinors of fluorine and 1s spinors of oxygen were frozen. Data from Ref. [235].
Method/Hamiltonian
SCF
CCSD(T)
Experiment
DKH2sf+MFSO
DCG
183.5
184.9
194.1
194.4
193.81
Tab. 4.4. Spinorbit splitting of the 2 ground state of the ClO molecule [cm1 ]. Calculated in decontracted
basis sets, two sets of spinors, 1s spinors of oxygen and 1s, 2s, 2p spinors of chlorine were frozen. Data
from Ref. [235].
Method/Hamiltonian
SCF
CCSD(T)
Experiment
DKH2sf+MFSO
DCG
223.3
310.7
225.2
312.6
318
324
4.7.2
RASSISO
The so called RASSISO [102] method belongs to the group where spinorbit relativistic effects
are treated a posteriori in the so called LS or RussellSaunders coupling scheme [237], see also
Ref. [233], which is based on the combination of manyelectron states with the resultant angular
and spin momentum numbers, L and S, respectively, leading to a new wave function of a total
angular number J. The spinorbit interaction in accounted for through a multiconfigurational
interaction approach, which employs the Complete Active Space (CASSCF) or Restricted Active
Space (RASSCF) wave functions as a reference.
Relativistic effects are treated in two steps as follows: First, the CASSCF/RASSCF wave
function is calculated for a specified total spin S, and can be viewed as a common representative of all possible spin states within the given multiplet. Its advantage is that this spinfree
basis can involve many wave functions, and these may use individually optimized molecular orbitals. Such determined scalar relativistic wave functions include both static (through the use of
the multideterminantal RASSCF/CASSCF method) and dynamic correlation effects (using the
multiconfigurational perturbation theory, CASPT2).
In the second step, one computes eigenstates of a twocomponent quasirelativistic Hamiltonian in a basis set of spinfree (scalar) wave functions. The matrix over spinfree states,
where the Hamiltonian operator contains the DouglasKrollHess scalar term together with the
AMFI operator, Eq. 4.91, is calculated by an extension of the Restricted Active Space Interaction
(RASSI) method, and is utilizing the WignerEckart theorem due to the spinorbit operator.
To include the dynamic correlation within, one can employ the multi state (MS) CASPT2
method in the RASSISO procedure for calculation of spinorbit matrix elements. The MS
CASPT2 method is mixing CASSCF states and the resulting SOC wave functions are mixtures
of CASSCF functions of different space and spin symmetry.
Matrix elements over the original basis as well as over the eigenstates can involve also other
property operators, and can be used to calculate, for example, transition strengths.
4.8
Fine effects, Lamb shift
Although the Dirac theory predicts the entire spectrum of the hydrogenlike atom, some cases of
the fine splitting are not covered. Lamb and Retheford [238] discovered (1947) in a microwave
experiment that the 2s1/2 level was higher than the 2p1/2 level by 1.060 MHz. According to the
Dirac theory, both states, 2 S1/2 and 2 P1/2 should have equal energy. Bethe [239] was the first
to explain the very small shift, called the Lamb shift, in the hydrogen spectrum. This was the
beginning of modern Quantum Electrodynamics, QED.
Diracs theory does not deal with the possibility of creating and destroying particles, which
is one of the basic consequences of the theory of relativity. An extension using the quantized
electromagnetic field leads to the theory of quantum electrodynamics, which fixes this difficulty.
According to the QED theory electrons and positrons appear as quanta of the quantized Dirac
field. Vacuum fluctuations are perceived as zero point oscillations of this field. The interaction
of an electron with zeropoint oscillations of the electromagnetic field creates its selfenergy,
which is, together with effect of vacuum polarization, observed experimentally. The selfenergy
is the largest QED correction reducing the binding energy of an electron due to the exchange
of virtual photons with the nucleus. The vacuum polarization contribution causes shift in the
325
binding energy of an electron due to the creation and annihilation of virtual e e+ pairs in the
vacuum. It increases the binding energy mostly of the time and is the second largest correction.
In the two excellent reviews published recently by Lindgren, Salomonson a Hedendahl [83] and
Eliav and Kaldor [84] discussed physical aspects of QED effects in physics and chemistry and
present computational tools for incorporating of QED within the Coupled Cluster formalism for
electron correlation effects.
The value of the Lamb shift scales as the fourth power with respect to the nuclear charge Z,
thus is rather important for heavy systems. For instance, a good example is the U 91+ cation,
where the shift is of the order of 105 cm1 , what was confirmed experimentally [240].
It appears that QED corrections may be chemically important in the electron binding energy
of the ekaradon (Z=118) amounting to a 9% reduction of this property [241]. Ekaradon is a
superheavy element for which a relatively large QED effect is not so much surprising. However,
in compounds with rather light atoms as BF3 , AlF3 , and GaF3 it appears that the Lamb shift to
the atomization energies may be in the range of 35% of the scalar relativistic contribution [242]
and may be important for higher accuracy calculations particularly for molecules containing
heavy elements. Hirao et al. [243] calculated ionization energies of Zn and Mo12+ and conclude
that unprecedented accuracy can be achieved by using the fully relativistic manybody treatment
including the Lamb shift.
When working in relativistic molecular calculations with the no pair approximation (i.e. one
works with electronic states only), all QED effects are thus neglected.
326
Accurate calculations of atomic and molecular properties require sophisticated manyelectron

theories for treatment of the electron correlation problem. This is computationally very demanding even at the nonrelativistic level for medium or larger molecules. In so called wave function
methods the starting point is the HF wave function expressed most frequently in the form of the
Slater determinant, Eq. 3.12, as a reference. A singledeterminant HF reference wave function
is usually sufficient for closedshell atoms and molecules. For highspin openshell systems the
simplest reference is the singledeterminant ROHF wave function. In quasidegenerate situations
the reference is a combination of Slater determinants. Many problems require a reference consisting of more than a singledeterminant reference. This is the case of quasidegenerate systems,
like calculations on the hypersurface far from the equilibrium geometry or many excited states.
A multideterminant reference is also needed for treatment of the SO coupling.
Let us remind (see Section 3.2.5) that at the simplest singledeterminant level most demanding steps in sophisticated manyelectron methods, like CCSD, scale with the number of occupied
(No ) and virtual orbitals Nv as No2 Nv4 in an iterative process. Number of operations required for
triples scales as No3 Nv4 . When using the spinorbital basis for openshell systems with the UHF
(Unrestricted Hartree Fock) reference, we have to treat twice as much spinorbitals when compared to the number of closedshell or ROHF openshell orbitals. The number of operations
further increases for calculations in which we work with 4spinors (see Section 4). In spite of an
enormous progress in technology of CC calculations (see, e.g., our implementation of CCSD(T)
in the MOLCAS program [64]), they still remain computationally demanding, particularly, when
combined with specific requirements on the construction of basis sets with some relativistic Hamiltonian. This explains that rigorous computational applications of ideas of the relativistic theory
for manyelectron atoms and molecules is much more complicated than it is in the nonrelativistic
case.
Orbitals or spinorbitals in the Slater determinant, Eq. 3.12, are in practical nonrelativistic
calculations expressed in the form of the linear combination of atomic basis functions, mostly
contracted Gaussians, [39, 51, 62, 244]. Selection of basis sets for practitioners of relativistic
manyelectron calculations is rather delicate. In Section 4.4.6 we learned that there are specific
requirements on basis sets which are capable of representing the large and small components of
the Dirac spinors. Nonrelativistic basis sets are not applicable for such calculations. A small
component set of exponents is obtained from the large component set and leads to basis sets
which are too large for practical use. Also, in most calculations of atomic and molecular properties we need just a large component. The remedy is the FoldyWouthuysen transformation to a
twocomponent Hamiltonian. Theoretical background of these methods is described in Section
4.6. Most frequently used relativistic calculations employ nopair DKH methods which provide
excellent results in situations when the spinorbit coupling can be safely avoided. Such calculations are very appealing since they require nothing more than standard methods and computer
programs of nonrelativistic manyelectron quantum chemistry, just supplemented with relativistic integrals. Yet, using of standard nonrelativistic contracted Gaussian basis sets is not recommended. In fact, to be rigorous, we should say that any specific Hamiltonian requires a specific
basis set. This is clearly not practical for extensive applications of relativistic methods. A good
compromise is using standard exponents of the primitive Gaussians optimized at the nonrelativistic level. When left uncontracted, the flexibility is sufficient for quite different Hamiltonians.
327
This, however, would be of little help, since uncontracted basis sets are too large. Sadlej, in collaboration also with our group has developed a counterpart of popular nonrelativistic polarized
basis sets. The family of Pol basis sets [245256] was constructed having in mind nonrelativistic
calculations of molecular electric properties, which represent a response of the wave function
to the external electric field and thus suitable, especially, for calculations of molecular dipole
and quadrupole moments, dipole polarizabilities, higher polarizabilities, etc. Due to the presence of diffuse Gaussians these basis sets are suitable also for calculations of electron affinities,
weak intermolecular interactions, etc. Their relativistic counterparts, the PolDK family of basis sets [255262], share the same exponents as nonrelativistic Pol basis sets, but are contracted
differently. Contraction coefficients are obtained from atomic DKH calculations and therefore
feel the relativistic change of corresponding orbitals. Review of basis sets for relativistic calculations can be found in [263]. Another newer series of relativistic basis sets which are generally
available are augccpVTZDK [264, 265], ANORCC [105, 106]. Efficient are bases connected with pseudopotentials. Frequently used are the Stuttgart group effective core potentials
and corresponding basis sets, see e.g., Refs. [186188]. These basis sets and corresponding methods are designed to simulate cores of heavy atoms in valenceonly calculations with a formally
nonrelativistic Hamiltonian. Direct relativistic effects, including spinorbit interactions, on the
valence electron systems as well as indirect effects due to relativistic changes within the core, are
implicitly incorporated into the pseudopotential (PP). Twocomponent energyconsisted pseudopotentials in atomic Fockspace Coupled Cluster calculations [189] were successfully applied
in calculations of accurate (to within a few hundred wavenumbers) ionization potentials, electron affinities, and excitation energies of the group 11 and 12 as well as the 13 and 14 postd
main group elements. Treatment of just valence electrons within relativistic pseudopotentials
leads to substantial savings of computer time with usually small loss of accuracy with respect to
allelectron correlated calculations.
The density functional theory (DFT), originally formulated by Hohenberg, Kohn and Sham
[266, 267], was later extended also to highZ systems within the framework of quantum electrodynamics [268271]. DFT covers important part of the relativistic treatment of atomic and
molecular properties of systems with heavy atoms. This is due to well known advantages of DFT,
its simple mathematical formalism and less demanding computational implementation (e.g., very
favourable scaling with the number of electrons). Both these aspects make DFT attractive for
large manyelectron atoms, molecules, clusters and/or solids. In similarity with the HFSCF
method, Section 3.2.1, the KohnSham DFT paves the way for the electron density of the ground
state of system of independent particles moving in the effective oneparticle external potential.
The key point and the stumbling block in DFT is the unknown exchangecorrelation functional that describes the electronelectron interactions. There is a large spectrum of functionals
ranging from local density approximation, gradient corrected functionals to hybrid ones. The latter class can be classified as having a semiempirical flavour, since all hybrid functionals are products of some kind of calibration or fitting with respect to accurate data obtained from wave function based methods (mostly CCSD(T)). They provide quite reasonable properties like geometries,
harmonic frequencies, proton affinities, ionization potentials and work well in thermochemistry.
Their performance in more sensitive properties like electron affinities, dipole moments and/or
polarizabilities is less satisfactory [272]. Although there is no well defined systematic hierarchy
of improved approximations in DFT (compared to wave function based methods), it seems that
for medium and large molecular systems (including heavy atoms) the DFT is and will still remain
328
the only method of choice in the near future. We are convinced that combining the advantages
of both approaches in a smart way one can arrive to reasonable molecular relativistic properties.
Wavefunction based data used as benchmarks for small systems can provide valuable numerical
control for DFT applied in extended systems. Implementation of the DFT method at the relativistic fourcomponent level together with applications for molecular properties are reported in
Refs. [273276].
An important idea of this paper is to show how molecular properties are affected by relativistic effects. Working with models including nonrelativistic and relativistic Hamiltonians allows
us to assess the difference of both results and estimate the importance of specific relativistic contributions. Clearly such model is approximate. One reason is that we can not use the same basis
set for different models. We just believe that our basis sets lead to results which are not too far
from the basis set limit and the basis set effects are small.
5.1
Ionization potentials, electron affinities and excitation energies
Ionization potentials, together with electron affinities and other properties are important quantities in calculating chemical potentials, electronegativity and other values useful in chemistry or
physics. How relativistic effects affect ionization potentials, electron affinities, and excitation
energies will be discussed in this Section.
5.1.1
Ionization potentials
An interesting V shaped dependence of ionization potentials of Cu, Ag, and Au on the atomic
number Z, Fig. 2.2, was used in Section 2 as an introductory demonstration of relativistic effects
on atomic property. Similar behaviour as seen in Fig. 5.1 is quite common, at least for atoms
with similar valence electronic structure.
In Tab. 5.1 are presented calculations of ionization potentials of alkali metal elements, which
are based on three different sources [277279]. The nrCCSD(T) values and the first line DKH
CCSD(T) values are calculated in equivalent basis sets and thus their comparison show us how
big is the scalar relativistic effect. This effect is growing from 0.017 eV for K up to 0.435 eV for
Fr and for the 119 element it is even 1.226 eV. Qualitatively, the pattern of IPs is similar as for
elements in the group 11. Note, however, that the difference of IP between Cs and Fr is much
smaller than is the analogous difference for Ag and Au. The second series of DKHCCSD(T)
values are calculated in a better basis set and this improvement accounts for about 0.1 eV. The
spinorbit effect is much smaller, than is the basis set effect. Even though it is small, its inclusion via the twocomponent pseudopotential ARPPCCSD(T) method and particularly via
the relativistic Fockspace Coupled Cluster singles and doubles approximation starting from the
DiracCoulomb Hamiltonian RCC(C) method, decently improves agreement with experiment.
Tab. 5.2 with calculations of ionization potentials of an extended series of atoms gives us
another paradigm for relativistic effects. We can follow several trends. The first one is increasing
magnitude of relativistic effects with the increasing Z number within the groups, which varies
from 0.3% for Ca up to 30% for Au. The second trend is that while nonrelativistic ionization
potentials are monotonously, mostly, decreasing with Z, relativistic values exhibit extremes in
the series and the last members, i.e. IPs of Ra, Au, and Hg are larger than the previous ones.
However, this is not true for elements in the group 15, where IPs are going monotonously down.
329
Fig. 5.1. Electron correlation and relativistic effects on electron affinities [eV] of the coinage elements Cu,
Ag, and Au. The ROHF nonrelativistic data represent calculation using a single determinant Restricted
Hartree Fock calculation. Electron correlation is treated by the CCSD(T) method and relativistic effects are
calculated by the spinfree DouglasKrollHess method.
This group differs from other presented groups by the fact that outermost occupied orbitals are
of ptype with the p3 occupancy. In general, p orbitals exhibit smaller stabilization and contraction than s orbitals which are outermost ones in the other groups. In the particular case of Bi,
in the group 15, the electronic structure means that the electron is occupying the relativistically
destabilized 6p3/2 orbital, similar as in Section 5.5, is the situation schematically shown for Hg
in Fig. 2.3 in Section 2. Of course, the important role is also played by the interplay between
the electron correlation and relativistic effects in obtaining quantitative values of IPs. With high
level correlated relativistic methods there is mostly a reasonable agreement between relativistically calculated ionization potentials and experiment. This gives a confidence in predicted value
of IP of the superheavy element 119, as shown in Tab. 5.1.
5.1.2
Electron affinities
Now, let us analyse relativistic effects on electron affinities. Before going into details, let us stress
that obtaining EA accurately is more difficult than obtaining accurate values of IPs. Frequently,
the electron configuration of an anion is more complicated than that of the mother molecule.
Second, to accommodate an extra electron in the originally unoccupied orbital requires specific
diffuse basis sets, that means more extended bases than are normally required. One has to be
careful since many states of an anion represent a situation when the electron is in fact unbound.
330
In practice, obtaining AE of seemingly simple (but extremely important) molecule like the O2
molecule by quantum chemistry methods with accuracy better than, say. 0.005 eV is a torture
[284]. Interestingly, even if oxygen is a light element, both scalar [284] and spinorbit [172]
relativistic effects matter in EA of the O2 molecule (together with many other effects) if the
above mentioned accuracy should be achieved. The relief for theoreticians is that experimental
determination of the electron affinity is very difficult as well.
In spite of a previous general note, obtaining reasonably accurate electron affinities for Cu,
Ag, and Au (as a counterpart to their IPs) is relatively simple. The DKHCCSD(T) data are
presented in Fig. 5.1. We note that EAs obtained by the independent HartreeFock model are
completely meaningless. At the nonrelativistic HF level all three EAs are close to zero, completely out of experiment. Electron correlation treated by CCSD(T) helps a lot for Cu, but the
true remedy for Au is CCSD(T) with at least scalar relativistic effects at the DKH level. This
treatment leads to reasonable agreement of theoretical and experimental data and represent the
trend with respect to the atomic number excellently.
Supplementary insight into electron affinities offer data in Tab. 5.3. Here we compare the
pattern of EAs for elements of the group 11 and 14. In contrast to the Cu, Ag, and Au series,
EAs of the group 14 elements go down monotonically (data for Si and Ge can be found in
the original literature [285]). The decrease is rather slow when going from Si to Ge and Sn,
but is enhanced for Pb. This is attributed to the relativistic destabilization of the 6p3/2 orbital
and is analogous to the effect commented already in IP of bismuth, Tab. 5.2. Extrapolating
to larger atomic numbers Borschewski et al. [285] predict that no electron binding will occur
Tab. 5.1. Ionization potentials of alkali atoms [eV].
Atom
nrCCSD(T)a
Group 1 [ns]
K
4.267
Rb
4.068
Cs
3.681
Fr
3.542
119
3.312
DKHCCSD(T)
4.284a
4.336b
4.137a
4.167b
3.821a
3.881b
3.977a
4.038b
4.538a
4.713b
ARPPCCSD(T)c
RCC(C)d
experimente
4.333
4.343
4.347
4.161
4.181
4.171
3.884
3.898
3.899
4.013
4.072
4.073
4.333
4.181
Ref. [277].
Ref. [278].
c
Twocomponent pseudopotential method [278].
d
Relativistic Fockspace coupled cluster singles and doubles approximation starting from
DiracCoulomb Hamiltonian [279].
e
All experimental data from Ref. [280].
b
331
in the element 114. They predict that this element is expected to be highly inert and volatile.
This is related to an adsorption process on gold (and other) surfaces of chromatography column
detectors, using the gasphase chromatography techniques.
Our next example of relativistic effects in calculations of EA of the PbO molecule demonstrates essential importance of the spinorbit effects. Data in Tab. 5.4 represent the adiabatic
detachment energy. The PbO anion is more stable than is the mother molecule. The adiabatic
Tab. 5.2. Ionization potentials of selected atoms [eV].
Atom
relativistic
experimenta
Group 2 [ns2 ]
Cab
Srb
Bab
Rab
nr-CCSD(T)
6.074
5.607
5.054
4.815
DKH-CCSD(T)
6.093
5.678
5.194
5.238
6.1132
5.6949
5.2117
5.2784
Group 10 [(n-1)d8 ns2 ]

Nic
nr-CCSD(T)
7.676
DKH-CCSD(T)
8.302
7.6398
Group 11 [(n-1)d10 ns]

Cud
Agd
Aud
nr-CCSD(T)
7.555
6.934
7.035
DKH-CCSD(T)
7.733
7.461
9.123
7.7264
7.5762
9.2255
Group 12 [(n-1)d10 ns2 ]

Zne
nr-CCSD(T)
9.167
DKH-CCSD(T)
9.363
(9.393)
8.851
(8.941)
10.285
(10.362)
Cde
8.351
Hge
8.326
Group 15 [ns2 np3 ]

Asf
Sbf
Bif
a
nonrelativistic
MSIH
9.660
8.487
7.284
9.3942
8.9938
10.4375
9.7284
8.6084
7.2855

Ref. [281].
c
Ref. [282].
d
Ref. [33].
e
Data in parentheses are with estimated contribution from correlating of inner shell electrons on
top of the correlation of 18/17 electrons, [283].
f
Relativistic mixed sector intermediate Hamiltonian (MSIH) Fockspace coupled cluster calculations [175].
b
332
Tab. 5.3. Electron affinities of selected atoms [eV].
Atom
Group 11 [(n-1)d10 ns]
Cub
Agb
Aub
Group 14 [ns2 np2 ]
Snc
Pbc
114c
relativistic
experimenta
DKH-CCSD(T)
1.236
1.254
2.229
1.235
1.302
2.309
MSIH
1.085
0.383
no EAd
1.112
0.364

Ref. [33].
c
Relativistic calculations in the framework of the projected DiracCoulomb-Breit Hamiltonian.
Mixed sector intermediate Hamiltonian (MSIH) Fockspace coupled cluster calculations with
spinorbit effect included [285].
d
The 114 element does not bind an extra electron.
b
detachment process is denoted in Fig. 5.2 by the solid line connecting the electronic minima of
both species. For comparison with experiment the energy difference must be corrected by vibrational contributions. The vertical EA is denoted by the dotted line. The PbO molecule is not split
due to SO effects. The SO states of the anion and both separated atoms are inserted in the ovals.
Both electron correlation and relativistic effects are important in calculations of EA of PbO.
Our data show that at the SCF level one obtains no better EA than is 0.31 eV, less than half of the
experimental value. Our best result follows from the BSSCCSD(T) calculation with the large
decontracted ANORCC basis. Calculated EA, 0.696 eV, agrees reasonably well with the experimental value, 0.714 eV. We note that the experimental electron affinity for the ground vibrational
state of PbO, 0.722 eV [287], was corrected considering the zero point energies, 0.045 eV and
0.037 eV for the PbO ground state and the negative ion, PbO , respectively. Harmonic vibrational frequencies were taken from experiment. The reliability of the BSSCCSD(T) result is
supported by the good agreement of BSSCCSD(T) and DCCCSD(T) results with the PolDK
basis set. Both EAs agree to within 0.01 eV. It is fair to mention that our best BSSCCSD(T) results with ANORCC basis sets were obtained after deleting virtual spinors with energies higher
than 10 a.u., which introduces some uncertainty in the final EA. Spinfree DKHCCSD(T) data
in Tab. 5.4 show, however, that deleting virtuals affects EA very little. We believe that the same
is valid with methods treating spinorbit effects as well. The estimate of the basis set effect is another issue that has to be considered. Spinfree CCSD(T) results with the two basis sets, PolDK
and ANORCC, differ by 0.01 and 0.005 eV with the DC and BSS Hamiltonians, respectively.
When considering the basis set effect with the SO coupling within BSSCCSD(T), that is the
difference of results with the PolDK and ANORCC basis sets, it is estimated to be 0.1 eV.
Proper consideration of the relativistic effects including the spinorbit coupling is a must for
333
calculations of EA for PbO. Results presented in this work show that due to the SO coupling the
magnitude of EA increases by 0.17 - 0.20 eV with the PolDK basis set with BSSCCSD(T) and
DCCCSD(T) methods. With larger decontracted ANORCC basis set the SO effect increases to
0.23 eV. Lower SO effect was obtained when comparing the spinfree and the spinorbit coupling
CASSCF/CASPT2/RASSISO results, 0.607 eV and 0.659 eV, respectively. In this comparison
the CASPT2/RASSISO calculations for PbO were performed using the contracted ANORCC
Pb[9s8p6d4f 3g] and O[5s4p3d2f 1g] basis set for the two components of the 2 state. The spin
orbit calculation included only these states. The ground state level is 2 1/2 , which is separated
from the 2 3/2 state by 0.42 eV. It is possible that a more elaborate study would increase the
spinorbit coupling somewhat and would subsequently increase the SO effect on the electron
affinity. In any case, it is important to realize that EA of PbO is affected dominantly by SO states
of the anion and the Pb atom. The oxygen atom contributes much less, as expected. Reasonable
agreement of EA following from the RASSISO method (which uses DKH orbitals instead of
spinors) and BSS or DC CCSD(T) methods shows that SO splitting of reference orbitals do
not deteriorate the accuracy significantly. One should not forget that the electron correlation
treatment is different in CASPT2/RASSISO and CCSD(T) methods.
5.1.3
Excitation energies
There are many examples of relativistic effects on excitation energies (EE) available in the literature, see at least results obtained by the RASSISO, the fourcomponent Fockspace CC, ZORA
Fig. 5.2. Spinorbit effects on the dissociation energy and the electron affinity of PbO. Data from
Ref. [286].
334
and other methods [102, 103, 105, 189, 290, 291]. In this Section we will restrict ourselves just
to an interesting example of the excitation energy between the ground state of the Fe atom and
its first excited state. Another example, lowlying states of InN, will be discussed in the next
Section together with SO effects affecting the potential energy curves of the two states of this
molecule.
We tried to calculate EE of the Fe atom some years ago [292] using a simple MVD (Mass
VelocityDarwin) approach, i.e. we have omitted SO effects. Considering that Fe is relatively
light element, we were surprised to observe that approximately considered relativistic effect enhanced EE of Fe by about 0.3 eV. Now, we present newly obtained more accurate results which
confirm our previous experience. Nonrelativistic, scalar relativistic and relativistic results including the SO contribution are compared with experimental energy splitting between the 5 D ground
state and the first excited 5 F state of the Fe atom in Fig. 5.3. The NR value was calculated
by the MRCISD method using the ANOLVQZP basis set [293]. REL represents the scalar
relativistic DKH MRCISD value and the SO column represents the spinorbit effect included
via the RASSISO method [102]. Both REL and SO values were calculated with the relativistic ANORCCVQZP basis set [106]. Clearly, the scalar relativistic effect alters the excitation
energy of Fe significantly. Neglecting SO effects exaggerates the relativistic correction with re-
Tab. 5.4. Electron affinity of the PbO molecule [eV]. Data from Ref. [286].
Method
PolDK basis
SCF
CCSD(T)a (700)
ANO-RCC basis
SCF
CCSD(T)a (10)
CCSD(T)a (700)
CCSD(T)a (10)b
DKHsf
BSSsf
DCsf
BSS
DC
0.182
0.377
0.180
0.382
0.178
0.380
0.312
0.594
0.307
0.584
0.160
0.463
0.466
0.157
0.470
0.156
0.291
0.696
0.285
contracted ANO-RCC basis

CASPT2/RASSI-SOc 0.607d
Experimente
a
0.472
0.696
0.659
0.714
Values in parentheses denote energy limit [a.u.] of deleted virtual orbitals.

PbO and PbO molecules are in the CCSD(T) optimized distances. Otherwise experimental
geometries are used.
c
Used the spinfree CASPT2/SOCCASPT2 optimized distances.
d
Result neglecting the spinorbit coupling.
e
The given EAe value is recalculated from the adiabatic EA0 value of 0.722 eV from Ref. [287],
using experimental values of e .
b
335
Excitation energy [eV]
0.5
0.674
0.914
0.890
0.859
0
NR
REL
SO
Experiment
Fig. 5.3. Comparison of calculated nonrelativistic, spinfree relativistic, spinorbit relativistic and experimental values of excitations between the two lowest electronic terms of the Fe atom. Data from Ref. [288].
spect to experiment. Proper treatment of both scalar relativistic and SO effects leads to very
good agreement of theoretical values with experiment. More quantitative account is presented in
Tab. 5.5 in which we collect excitation energies of individual spectroscopic levels of these two
Tab. 5.5. Spinorbit splitting for the two lowest electronic terms of the Fe atom [cm1 ]. Data from Ref.
[288].
Configuration
Term
level
J
RASSI-SO
Ref. [289].
experimenta
3p6 3d6 4s2
4
3
2
1
0
0
382
669
860
956
0
416
704
888
978
3p6 3d7 (4 F)4s
5
4
3
2
1
7 180
7 657
8 040
8 326
8 517
6 928
7 376
7 728
7 986
8 155
336
Fig. 5.4. Contractions of bond lengths in coinage metal fluorides due to relativistic effects.
electronic terms. The calculated total spinorbit splitting for the ground state term 5 D differs
from experiment by 22 cm1 and for the 5 F term by 110 cm1 .
5.2
Relativistic effects on molecular geometries
One can easily understand that due to the relativistic shrinking and expanding of orbitals participating in molecular bonds, due to the shift of orbital energies and other effects, as mentioned
already in Section 1, molecular geometries must also undergo changes due to relativistic effects.
A simple example (which, nevertheless reflects the situation in many other cases) is presented
in Fig. 5.4. Here are depicted contractions of bond lengths in coinage metal fluorides due to
relativistic effects calculated by the IOTCCCSD(T) method with 18 electrons correlated. There
were used completely uncontracted Pol basis sets for coinage metals [254] and for fluorine [245].
for AuF it is already 0.17 A.
This leads to the fact that
While for CuF the contraction is 0.02 A,
the bond length of AuF is much shorter than that of AgF, which is clearly against trivial nonrelativistic expectations and fully in line with other effects (like relativistic effects in IP and EA
discussed in the preceding Section), which follow from the relativistic contraction and stabiliza-
337
Fig. 5.5. Potential energy curves of the InN molecule calculated by the RASSISO method. Taken from
Ref. [294].
tion of valence 6s electrons of gold. Of course, relativistically altered potential energy curves
alter vibrational frequencies as well.
More complicated are the potential energy curves of the InN molecule with relatively light
In atom. In Fig. 5.5 we present potential energy curves of the ground X3 state and the low
lying 3 state of InN. Six different relativistic SO states arise from the 3 and 3 states of
InN: 2, 1, 0+ and 0 (corresponding to the 3 state ) and 1 ,0+ (corresponding to the 3
state). In Tab. 5.6 we also present basic spectroscopic constants of this molecule, the electronic
equilibrium bond length Re , the harmonic vibrational frequency e , and the excitation energy Te .
The relativistic ground state is 0+ , with almost the same Re and Te as for the spinfree ground
state 3 . Two avoided crossings, corresponding to 1 and 0+ states can be seen in Fig. 5.5.
Without these crossings the spinorbit effects on the InN molecule would be negligible. Instead,
three pairs of similar states are produced. The first pair includes two lowest states 0+ and 1,
with spectroscopic properties being very close to the 3 state. The second pair (2, 0 ) is not
affected by the avoided crossings, leading to almost the same Re and e as in the 3 state. The
third pair includes second 0+ and 1 states, which have avoided crossings in the vicinity of their
their minima. Their properties are thus strongly affected by these crossings. Bond lengths of
338
these states are closer to the 3 state than to 3 , but the respective harmonic frequencies are
higher by 220330 cm1 when compared to both spinfree lowlying states. This feature can be
clearly seen in Fig. 5.5, avoided crossing leads to the narrower potential energy curves.
5.3
Electric properties dipole moments and dipole polarizabilities
Calculations of electric properties in general can be performed in two different ways, either as
an expectation value of the given operator or by so called derivatives methods. Conditio sine qua
non to apply the first method is to have a wave function which satisfies the HellmannFyenman
theorem [51]. This is fulfilled for, e.g., variational wave functions in which all parameters are
fully optimized (like in SCF or CASSCF methods). When using correlated methods, only the
full CI or CC with all excitation operators guarantees that the HellmannFyenman theorem is
satisfied. This is possible only for really small molecules. For more realistic applications we
are very rarely dealing with wave functions fulfilling this theorem. Therefore, even when using
highly sophisticated correlated methods we are usually relying on energy derivatives methods
(numerical or analytical) with respect to the external perturbation which defines the pertinent
property. The analytical derivatives methods need derivatives of atomic and molecular one and
twoelectron integrals and excitation amplitudes with respect to the external field. There are
many well elaborated methods of this category, see, e.g., Refs. [39, 40]. Analytical derivatives
may be a problem for some relativistic operators. Generally applicable and straightforward is
the so called finite field method [37] which exploits the expansion of the energy with respect
to an electric external field, Fa (or, for some properties, also other external fields) along the
adirection,
1
1
1
E = E 0 i Fi ij Fi Fj ijk Fi Fj Fk ijkl Fi Fj Fk Fl .
2
6
24
(5.1)
Tab. 5.6. Spectroscopic properties of the relativistic states of the InN molecule. CASPT2/RASSI
SO/ANORCC calculationsa . Data from Ref. [294].
State
Re (bohr)
e (cm1 )
Te (eV)
Spin-free relativistic results

X3
4.163
447.3
13
3.884
522.3
0.111
SO results
0+
1
2
0
1
0+
0.001
0.100
0.133
0.133
0.154
4.157
4.157
3.882
3.887
3.940
3.942
441.1
441.6
522.6
521.6
789.4
705.6
Inactive orbitals 12a1 /5b1 /5b2 /2a2 , active orbitals 7a1 /3b1 /3b2 /1a2 ,18 correlated electrons in
CASPT2.
339
The first order properties like dipole and quadrupole moments, electric field gradient, etc. are
calculated as the first order derivatives,

E(+Fa ) E(Fa )
E
=
.
(5.2)
a =
Fa Fa =0
2Fa
The second order properties like dipole, quadrupole polarizabilities, force constants, geometry
derivatives of the first order properties, etc. are calculated as the second order derivatives, e.g.,
2
E(+Fa ) + E(Fa ) 2E(0)
E
=
.
(5.3)
aa =
Fa2 Fa =0
2Fa2
In this way we can calculate also first and second hyperpolarizabilities. We would like to stress
the importance of theoretical and computational methods in this area since particularly nonlinear
optical properties are frequently not available from experiment. One problem with the finite
field method is a numerical accuracy which depends on the careful selection of the intensity of
an external field. It should be small enough to avoid contamination by higher order properties, so
that the accuracy with which the energy is calculated must be very high. For accurate calculations
the series of fields with different strength and more complicated numerical derivative methods
must be used.
In this Section we will document the impact of relativistic effects on dipole moments and
polarizabilities. First, we should mention that when calculating atomic and molecular electric
properties, two and onecomponent relativistic methods suffer from the change of picture effect,
see Section 4.6.4. Tab. 5.7 gives example of the magnitude of this effect in the case of dipole
moments of coinage metal hydrides. There is a growing relativistic contribution going from
copper to gold. While the CCSD(T) value, with the change of picture effect included, for Cu
Tab. 5.7. Change of picture effect on dipole moments of CuH, AgH, and AuH [a.u.]. Data from Ref. [257].
Method
a
b
CuH
AgH
AuH
Nonrelativistic results
SCF
CCSD(T)
1.593
1.136
1.808
1.362
1.660
1.223
Relativistic results
DKHSCF
DKHSCF
no-CPa
CPb
1.519
1.526
1.588
1.618
0.830
0.980
DKHCCSD(T)
DKHCCSD(T)
no-CPa
CPb
1.042
1.052
1.102
1.135
0.400
0.522
DouglasKrollHess calculations with the change of picture not included.

DouglasKrollHess calculations with the change of picture included.
340
was diminished by 7%, for Au it was 57% already. The electron correlation effect is significantly
more uniform here and accounts for 30% to 47%. The change of picture effect is rather small
here, what is a typical behaviour of valencedetermined properties. Compare it with electric field
gradients, Section 5.4, where it is a serious issue.
Dipole moments of GeO, SnO, and PbO are collected in Tab. 5.8. Before we will concentrate
on relativistic effects throughout the group we note that the lowering of dipole moments due to
the electron correlation is more important than are relativistic effects. Quite small relativistic
change of the dipole moment of PbO is particularly surprising. As it is usually the case, the
electron correlation and the relativistic effects are not additive [295]. In general, the agreement
of theoretical and experimental results is good, especially for GeO. Dipole moments for SnO and
PbO are systematically lower than experimental ones by about 5.9% and 3.8%, respectively. The
spinorbit effect is rather small even for PbO, as it follows from results obtained by Roos and
Malmqvist [297]. Theoretical values presented in Tab. 5.8 are uncorrected for vibrational effects,
but their accuracy is still sufficient for discussion of trends within the group of molecules.
Good agreement of theoretical dipole moments with experiment gives some confidence to
theoretical polarizabilities presented in Tab. 5.9. For GeO and SnO experimental data are missing. The only experimental values are available for PbO from the optical measurements for the
Tab. 5.8. Dipole moments of GeO, SnO, and PbO [a.u.]. Data from Ref. [21].
GeO
e
SnO
PbO
nrCCSD(T)
DKHSCF
DKHCCSD(T)
1.263
1.681
1.280
1.565
2.114
1.598
1.669
2.374
1.764
Experimentb
1.291
1.70
1.83
Methoda
The HyPolf basis set, which is the HyPol set extended by two f functions, was used, see
Ref. [295]. In DKH calculations bases with the DKH contractions were used.
b
Ref. [296].
Tab. 5.9. Dipole polarizabilities of GeO, SnO, and PbO [a.u.]. Data from Ref. [21].
zz
Methoda
nrCCSD(T)
DKHSCF
DKHCCSD(T)
a
See footnote in Tab. 5.8.
xx
GeO
SnO
PbO
GeO
SnO
PbO
40.52
35.02
40.33
57.18
47.89
56.29
65.76
53.39
62.55
28.23
28.19
27.62
40.34
37.19
37.75
45.85
34.97
37.46
341
55
[a.u.]
50
Cu,Ag,Au
45
Ge,Sn,Pb
40
Zn,Cd,Hg
35
Rb,Cs,Fr
(the alkali metal series scaled by the factor 0.1)
30
20
30
40
50
60
70
80
90
Fig. 5.6. Theoretical atomic dipole polarizabilities, relativistic effects included. Data from Refs. [16, 33,
258, 259, 277].
solid state [298]. As with dipole moments of all the group 14 oxide molecules, parallel component of the polarizability is less affected by relativistic effects than by the electron correlation.
The electron correlation for zz of PbO (9.16 a.u. or 14% of the final DKHCCSD(T) value) is
almost three times larger than is the scalar relativistic effect for zz of PbO (3.21 a.u. or 5.1%
of the final DKHCCSD(T) value). Again, as with dipole moments, the signs of electron correlation and relativistic effects, respectively, are opposite. Similar compensation of relativistic
and electron correlation effects is observed in calculations of xx . This time, however, the xx
component of the PbO polarizability is more affected by scalar relativistic effects than by the
electron correlation. Indeed, results in Tab. 5.9 show, that relativistic effects lower xx of PbO
by 8.39 a.u. (22.3% of the final value) at the CCSD(T) level. With the electron correlation xx
increases by 2.49 a.u. Both relativistic and electron correlation effects in xx of GeO and SnO
are much smaller than in PbO but relativistic effects remain more important than are correlation
effects even for the molecules containing relatively lighter atoms.
The appearance of the non monotonic pattern of, e.g., atomic dipole polarizabilities, is quite
general. Fig. 5.6 shows polarizabilities for the group 1, 11, 12, and 14 elements (we restrict ourselves to the three heaviest elements in each group). We note that the pattern of polarizabilities
is the same for all four groups. Since all elements in the Zn, Cd, and Hg group and in the Rb,
Cs, and Fr groups have similar valence electronic structure (valence selectrons) as the group 11
elements, the interpretation of the pattern remains the same. The group 14 elements are char-
342
acterized by the valence np2 electronic structure. For these elements, especially for properties
of Pb, the spinorbit interaction can not be ignored. Polarizabilities of Ge, Sn, and Pb were
taken from the review of Schwerdtfeger [16]. In this review the discussion on polarizabilities
in the spinorbit split states can be found. The SO contribution to the dipole polarizability is
known also for the group 1 elements, even if estimated at the uncorrelated level [277]. Dirac
HartreeFock calculations revealed that the spinorbit contribution to the polarizability of Rb is
negligible. The SO coupling lowers the polarizability of Cs very slightly, by 2.7 a.u., while for
Fr its magnitude is 13.0 a.u. It further supports the tendency to the non monotonic pattern. In
any case these values are too small to affect our discussion of the general trends within the group
11 elements. To proceed further, Lim et al. [277] calculated the polarizability of the superheavy
element 119. It is as low (185 a.u.) as the polarizability of the Na atom (165 a.u.). The SO
contribution is not negligible but still no larger than 16 a.u. Electric dipole polarizabilities of the
halogen atom in the 2 P1/2 and 2 P3/2 spinorbit split states have been investigated by Fleig and
Sadlej [299]. They used the twocomponent variational treatment and showed that for heaviest
homologues it is necessary to replace averaged values by individual component polarizabilities.
Calculations of electric properties for separated atomic SO states are rather unique. Full account
of atomic polarizabilities throughout the Periodic Table was reviewed by Schwerdtfeger [16]. His
analysis shows that the group 1 elements have the largest polarizabilities, followed by the group
2 elements polarizabilities. The polarizability then decreases within one period of the Periodic
Table.
The trend for polarizabilities of the group of the coinage metal cations is shown in Fig. 5.7.
After ionization, the valence electronic structure in all these cations is (n-1)d10 . Valence d
electrons are relativistically destabilized and, consequently, their polarizabilities increase [254,
258]. The same trend is observed in the group of CuF, AgF and AuF molecules. Analogous arguments are used in the explanation of trends in the group RbF, CsF, and FrF. However, the valence
electronic structure of their metal element cations corresponds to the noble gas element and is
(n-1)p6 . Valence pelectrons are relativistically stabilized, and the polarizability of Rb+ , Cs+ ,
and Fr+ ions relativistically decreases. The same trend is, again, observed in polarizabilities of
RbF, CsF, and FrF molecules. Analogous interpretation, with some reservation, is also valid for
GeO, SnO, and PbO molecules. Polarizabilities of these molecules also increase monotonically
with an increasing atomic number of the metal element, Tab. 5.9. If we could interpret these
molecules as ionic species with the charge of the metal Me2+ , the polarizability of PbO should
decrease by relativistic effects much more than it is observed in Tab. 5.9. The valence electronic
structure of the Ge2+ , Sn2+ , and Pb2+ ions is ns2 np0 . These electrons are relativistically stabilized and their polarizabilities relativistically decrease [295] (like in isoelectronic elements Zn,
Cd, and Hg, Fig. 5.6). This is qualitatively in line with relativistic effects in GeO, SnO, and PbO,
but the effect in ions is larger and the Z dependence is not monotonic. The change of zz polarizability due to relativistic effects in PbO is 3.21 a.u., while in Pb2+ it is 7 a.u. This quantitative
discrepancy is explained by the fact, that PbO can not be completely represented as Pb2+ O2 ;
the charge on participating atoms is actually lower. This can be supported by the Mulliken population analysis (with all the reservation to any interpretation based on the HF orbital picture)
which shows that there is the charge transfer of about 1.5 electrons from the metal atom to the
oxygen atom. We note that the polarizability of the O2 anion is estimated by DKHCCSD(T)
as 75.3 a.u.
To demonstrate the trends of dipole moments for a series of a heavy metal element contain-
343
25
[a.u.]
20
15
Ge2+,Sn2+,Pb2+
(d10,s2,p0)
10
+
Rb ,Cs ,Fr
Cu ,Ag ,Au
(d10,s0)
5
20
30
40
50
60
70
80
90
Fig. 5.7. Nonrelativistic (dashed lines) and DKH (full lines) dipole polarizabilities of ions calculated in the
CCSD(T) approximation using the Pol family bases. Data from Ref. [21].
ing molecules we will discuss data in Fig. 5.8 for CuF, AgF, AuF, and the series of inter metallic
compounds, CuAl, AgAl, AuAl. Relativistically calculated dipole moments in the highly polar
MeF series [300] (polarity is Me(+) F() ) are much smaller than are their nonrelativistic counterparts. This should be related to the relativistic increase of IP particularly for the gold element
which hinders the charge transfer from the metal element to fluorine, in spite of high electron
affinity of the F element. The AuF molecule becomes less ionic due to relativistic effects which,
in turn, reinforce the covalent character of the bond in AuF. Opposite situation is encountered in
the MeAl series [255]. The polarity at the CCSD(T) level is Me() Al(+) in this case. Consequently, huge relativistic increase of the electron affinity supports enhanced polarity of all MeAl
molecules, particularly of AuAl. We note that the convention used within this review is that if
the left atom of a diatomic molecule has a positive charge and the righthand atom is negative,
the dipole moment has a positive sign. It should be mentioned here, that experimentalists have
problems to determine the sign of the dipole moment. Recently, Gijsbertsen et al. [301] reported
a novel experimental direct approach for determining the sign of permanent dipole moments. Of
course, in calculations of theoreticians there are no problems to get the sign.
The number of chemical elements is increasing considerably. The heaviest, manmade elements at the farend of the Periodic Table are located in the area of the longawaited superheavy
elements [303]. From the theory follows that these heavy elements should exhibit largest relativistic effects. It would be worthwhile to look at it. In Tab. 5.10 we presented CCSD(T)
344
[a.u.]
AgF
CuF
polarity Me
AuF
(+) (-)
polarity Me(-)Al(+)
AuAl
CuAl
AgAl
0
20
30
40
50
60
70
80
90
Fig. 5.8. Nonrelativistic (dashed lines) and DKH (full lines) dipole moments of the MeF and MeAl (Me
= Cu, Ag, Au) series calculated in the CCSD(T) approximation using the Pol family bases. Data from
Ref. [22].
calculations of dipole moments for superheavyelements hydrides. There are manifested two
interesting features. The total relativistic effect for sblock element hydrides 111H and 119H
is considerable larger than for pblock element hydrides 113H and 117H, while the spinorbit
effect is larger for pblock element hydrides. The explanation of these two opposite effects is
rather simple. The largest relativistic contraction is for sorbitals and, naturally, the spinorbit
coupling should be larger for p than for sorbitals.
Tab. 5.10. Dipole moments of superheavy elements hydrides [a.u.] icalculated at the CCSD(T) level of
approximation. Data from Ref. [302].
Method
Nonrelativistic
Scalar relativistic
DiracCoulomb
e
111H
113H
117H
119H
1.658
0.055
0.005
0.322
0.971
-0.026
-0.028
0.243
0.765
3.633
2.159
2.205
5.4
345
Determination of nuclear quadrupole moments
All nuclei with the nuclear spin I 1 have nonzero nuclear quadrupole moments Q, which
means that the electric charge is not spherically symmetric. The knowledge of reliable nuclear
quadrupole moments is of considerable interest in nuclear physics for testing nuclear models for
stable isotopes and also in chemical and solid state spectroscopy [304, 305].
The electric field at the given nucleus X produced by all electrons and other nuclei of the
molecule is not uniform and has a nonvanishing electric field gradient q. For quadrupolar nuclei this leads to the interaction between nuclear quadrupole moment Q of the nucleus X and the
electric field gradient q resulting from the distribution of other charges. This interaction might
be observed as splittings of lines in atomic spectra [306], microwave rotational spectra of molecules [307], and nuclear quadrupole resonance or Mossbauer spectra of solid state systems [308].
The measured value of the Q q interaction is often called the nuclear quadrupole coupling constant Q . The electric field gradient q, which from the point of view of the given nucleus is a
measure of the inhomogeneity of the external electric field of all other charges, is therefore a
molecular property of the first order and can be determined from the knowledge of the electronic
wave function and positions of nuclei. Thus combination of experimental nuclear quadrupole
coupling constant and theoretical electric field gradient provides us with nuclear quadrupole moment values. This is the currently best way [304,305, 309], at least for light elements. According
to the source of the experimental data, the nuclear quadrupole moment values are referred to as:
atomic, molecular or solidstate. Another approach was the mesonic method, based on measuring the hyperstructure of essentially Coulombic energy levels of or mesons near the nucleus.
No such experiments have been published for more than two decades, however, these mesonic
nuclear quadrupole moment values still stand as the benchmarks for heavier elements [304,305].
Lowprecision determinations of Q are also available from nuclear Coulomb scattering, nuclear
rotational energy levels and from nuclear theory [304, 305].
To obtain the nuclear quadrupole moment of the given nucleus we can either calculate the
electric field gradient for the given atom and to combine it with experimental data of the nuclear
quadrupole coupling constant of given atom or calculate the electric field gradient at the given
nucleus in a molecule, preferably diatomic, for which the nuclear quadrupole coupling constant
is available. From the computational point of view the highest accuracy of the electric field
gradients can be achieved in atomic calculations. The spherical symmetry greatly simplifies the
computational problems and permits to go much beyond the present accuracy limits in molecular
calculations. Of course, a special computer code for atoms is required. On the other hand, the
atomic nuclear quadrupole coupling constants are usually known with rather low accuracy due
to the significant line broadening and the high accuracy of the electric field gradient may not be
very helpful.
In this review we will concentrate ourselves on molecular values of the nuclear quadrupole
moment since we have been involved in these calculations for two decades [220, 310324]. In
contrast to the atomic experimental data the nuclear quadrupole coupling constant of the nucleus
determined from molecular rotational spectra are usually of very high accuracy. In order to profit
from this advantage of the molecular microwave spectra it is necessary to calculate electric field
gradient values with a very high accuracy also, which might be very tough task, even for diatomic
molecules.
The nuclear quadrupole moment Q(X) of the quadrupolar nucleus X in a diatomic mole-
346
cule in the vibrational state v with the zaxis along the bond becomes (in milibarns; 1mb =
1031 m2 )
Q(X) =
X (v)
,
0.2349647q X (v)
(5.4)
where X (v) is the nuclear quadrupole coupling constant for the nucleus X and the vibration
state v given in MHz, and q X (v) is the axial (zz) component of the electric field gradient tensor
at the nucleus X and the vibration state v given in atomic units. The vibrational corrections to
the equilibrium value of the electric field gradient are usually done via the approximate formula
by Buckingham [325]

1
X
,
(5.5)
q X (v) = qeX + qvib
qeX + q1X v +
2
where qeX = q X (Re ) and
"
X
2 X #
Be
q
q
e e
X
q1 =
.
)
+
3(1 +
2
e
6Be
=0
2 =0
(5.6)
The symbols e , Be , and e denote the usual spectroscopic constants in cm1 taken from experiment [296] and = (R Re )/Re . This formula requires only the knowledge of the first and
second order derivatives of q with respect to the interatomic distance R evaluated at Re .
The methodology of calculating electric field gradient at the nucleus X in a diatomic molecule
at the equilibrium bond distance Re is based on partitioning of the total value to individual
X
contributions. First, it is expressed as a sum of the electronic contribution, qe,el
and the nuclear
X
one, qe,nucl
X
X
qeX = qe,el
+ qe,nucl
.
(5.7)
X
X
and electron correlation qe,corr,M
Further, the electronic term is divided into HartreeFock, qe,HF
contributions
X
X
X
qe,el
= qe,HF
+ qe,corr,M
.
(5.8)
In order to reach the required accuracy, highly sophisticated electron correlation methods have to
be used. The CCSD(T) method what is a coupled cluster approximation with iterative solution
for T1 and T2 amplitudes (CCSD) with noniterative T3 amplitudes [40] is the method of the
X
choice in this area [305]. Later we will discussed also a role of the T3 contribution qe,corr,T
3
X
X
X
qe,corr,CCSD(T)
= qe,corr,CCSD
+ qe,corr,T
.
3
5.4.1
(5.9)
Change of picture
Inclusion of relativistic effects in calculation of electric field gradients is inevitable. If one

or twocomponent relativistic methods (e.g. DKH, IOTC) are exploited, the usual method of
347
Fig. 5.9. The point charge nuclear quadrupole moment (PCNQM) model [329] and shifted nucleus (SN)
model [321].
computing expectation values, which is just a counterpart of the nonrelativistic scheme leads to
the change of picture effect [220, 221] manifested by significant inaccuracies of the calculated
property values. The physical background of this effect was mentioned in Section 4.6.4 already.
The way how to overcome this obstacle is to properly transform also the electric field gradient
operator. Reiher et al. recently published such calculations [326328].
Of course, computing expectation values of properties is relevant only if our wave function
satisfies HellmannFyenman theorem. If this is not the case we can use the finite field method,
see Section 5.3. Pernpointer et al. [329] proposed a numerical procedure, named the point charge
nuclear quadrupole moment (PCNQM) model, which might be easily used within finite field
calculations. The PCNQM model is based on placing a finite nuclear quadrupole moment built of
6 point charges in the octahedral arrangement in the nearest vicinity of the nucleus, see Fig. 5.9.
Such artificial quadrupole moment Q placed at the nucleus typically is of the order 106 a.u.,
and its interaction energy with the electric field gradient at the nucleus represented by the axial
X
component qzz
is
Uint =
3 X
3
X
Qqzz = d2 qzz
.
2
2
(5.10)
In nonrelativistic calculations the PCNQM model does not seem to offer any particular advantages over the direct use of the perturbation. On the contrary, it brings a disadvantage of the
determination of its two parameters, the value of point charges and the distance of these point
charges from the nucleus d, which requires a considerable amount of numerical testing. However, since the PCNQM model introduces a perturbation operator defined in terms of the usual
nuclearattraction operators, it becomes very useful in approximate relativistic calculations as a
simple way to avoid the explicit change of picture transformation [330]. This appears to be the
most important feature of the PCNQM model. However, the problem of the determination of the
most suitable values of the two parameters, which might be in an interval 102 1011 a.u. for
and 104 106 a.u. for d remains.
348
Kello and Sadlej proposed an alternative shifted nucleus (SN) model [321], which also removes the change of picture effect in one and twocomponent relativistic methods. The idea of
the SN model is to build up a small quadrupole moment by splitting the original nuclear charge Z
into two halves and shifted them by a small distance, see Fig. 5.9. If E (d2 ) denotes the energy
calculated with V (d2 ), = x, y, z, i.e., with the shifts of halfcharge nuclei along the th
direction, the value, qzz , of the field gradient operator will be given by:

(d2 )
4
X
,
(5.11)
qzz
=
3ZX
(d2 ) d=0
where
1
(d2 ) = Ez (d2 ) [Ex (d2 ) + Ey (d2 )] = a1 d2 + a2 d4 + . . .
2
(5.12)
can be approximated by fitting to a polynomial in d2 . In the case of diatomic molecules Ex (d2 ) =

Ey (d2 ) and in order to obtain the first derivative we have to perform two calculations, one for
Ez (d2 ) and other for Ex (d2 ). Similarly in the PCNQM model, two calculations for two Q differing by the sign have to be carried out. The important difference between these two approaches
is that while the PCNQM model requires two parameters ( and d), the SN model needs just one
(d).
5.4.2
Examples of calculations
In the next paragraphs we will show and discuss a magnitude of the change of picture effect in
approximative relativistic calculations. While the change of picture effect in valencedetermined
properties is rather small, see Tab. 5.7 in Section 5.3, this effect becomes very large for operators
which assume large values in the vicinity of nuclei, e.g., electric field gradient. In Tab. 5.11 are
collected calculated values of the qzz component of electric field gradient of hydrogenic ions of
rare gas atoms. In the column qnoCP are the expectation values without the change of picture
and in the column qCP are values calculated using the PCNQM model, which takes into account
the change of picture. While for Z = 9 the difference between the noCP and CP value is only
0.5%, we can see that going to Z = 85, this difference is almost 80%. Here we present values
Tab. 5.11. Change of picture effect on the electric field gradients in one-electron ions as calculated in
DKH-SCF approximation [a.u.]. Data from Ref. [220].
Ion
Ne+9
Ar+17
Kr+35
Xe+53
Rn+85
Term
qnoCP
qCP
2p1 ,2 P
2p1 ,2 P
2p1 ,2 P
2p1 ,2 P
2p1 ,2 P
-33.614
-199.73
-1735.05
-6777.89
-44728.2
-33.449
-196.58
-1624.69
-5760.44
-25171.8
349
just for the lowest 2 P electronic state, where is the largest effect. In the original paper [220] are
presented also values for higher electronic states.
Well, one can argue that hydrogenic ions are little bit artificial systems, so let us have a more
chemical example, hydrogen halides. In Tab. 5.12 are presented electronic contributions to
electric field gradient at the halogen nucleus in hydrogen halides calculated within the DKH
SCF approximation. We can see that while for HCl the change of picture effect accounts for
less than 1%, in the case of astatine it is 30% already. Of course, if we would calculate it
nonrelativistically, there should be no differences between these two ways of the calculation.
In Tab. 5.13 are presented values of the electric field gradient for Br and I nuclei calculated
within the CCSD(T) approximation with correlated 8 and 18 electrons. We can see that the
change of picture amounts for 4% in HBr and for 9% for HI. This is precisely the same trend as
for the SCF data presented in Tab. 5.12. Increase of the number of correlated electrons from 8 to
18 electrons has a marginal effect which is an indication of properly chosen active space. In the
Tab. 5.12. Change of picture effect on the electronic contribution to electric field gradients at the halogen
nucleus in hydrogen halides as calculated in DKHSCF approximation [a.u.]. Data from Ref. [220].
Molecule
HCl
HBr
HI
HAt
qnoCP
qCP
3.540
7.806
12.657
34.621
3.511
7.520
11.683
26.656
Tab. 5.13. Values for the nuclear quadrupole moments of the 79 Br and 127 I nuclei. Data from Refs. [311,
320].
q [a.u.]
Q [mb]
H Br, (v=0)
no-CP CCSD(T) 8 el.
CP
CCSD(T) 8 el.
CP
CCSD(T) 18 el.
4component CCSD(T)a
Recommended value this workb
7.579
7.308
7.313
7.035
299
310
310
322
3133
H127 I, (v=0)
no-CP CCSD(T) 8 el.
CP
CCSD(T) 8 el.
CP
CCSD(T) 18 el.
4component CCSD(T)a
Recommended value this workc
11.935
11.013
11.038
10.751
-651
-707
-705
-724
-71010
System
1
79
Ref. [331].
Considering the atomic value of 314.
c
Considering the atomic value of -717.
b
350
paper by Bieron et al. [320], which is the source of data for Tab. 5.13, are presented also atomic
multiconfigurational DiracFock calculations, which lead to the nuclear quadrupole moment data
314 mb for bromine and -717 mb for iodine. Taking into account both molecular and atomic
values the recommended value for 79 Br is 3133 mb and for 127 I it is -71010 mb. While the
bromine value is still the recommended one [305], for 127 I the present recommended value is
-69612 mb by van Stralen and Visscher [332].
In Tab. 5.14 we want to show the effect of the electron correlation (20 electrons correlated),
which amounts about 20%. There is also a rather large effect of triply excited configurations
which is about 5%. Concerning this table it should be stressed that GeO and GeS represent
completely independent sources of the data and thus the good agreement between Q obtained
from these two diatomics supports a reliability of the new recommended value, -1961 mb. The
new value falls in the error bars of the older reference atomic value, however, it is shifted towards
the upper bar.
Tab. 5.15 is a nice example how theoretical data on the nuclear quadrupole moment are
improving during the years. In 1982 there was a muonic value of 1506 mb. In 1984 Sundholm
et al. [335] published atomic value 140.31.0 mb, which was rather off the muonic one. In 1999
we published a paper [313] in which molecular data calculated on systems AlF and AlCl were in
a perfect agreement with new atomic calculations. The present recommended value is 146.61.0
mb.
In Tab. 5.16 are combined DKHCCSD(T) with fully relativistic 4component CCSD(T)
calculations for ZrO. Using DiracFock and DKHSCF values calculated in the same basis set
we estimated the spinorbit contribution to the electric field gradient as -0.0056 a.u. For the
nuclear quadrupole moment we got the value of 179 mb from DKHCCSD(T) calculation on
ZrO, while from 4component CCSD(T) it was 176 mb. DKHCCSD(T) value from ZrS is 173
mb. The reference atomic values are significantly higher. Our present recommended value is set
to 1763 mb.
Tab. 5.17 represents another interesting time dependence of nuclear quadrupole moment
values. The old 121 Sb atomic value of Murokawa from 1955 was -530100 mb [344], the 2001
Tab. 5.14. Values for the nuclear quadrupole moment of the 73 Ge nucleus. Data from Ref. [314].
System
73
73
Ge16 O, (v = 0)
Ge32 S, (v = 0)
Q a [MHz]
208.33
186.72
electronicb contributions to q [a.u.]

DKH-SCF DKH-CCSD DKH-(T)c
-6.332
-5.742
0.947
0.861
0.335
0.289
-195
-197
Reference atomic valued
-17326
Present recommended value
-1961
Data from Ref. [333].

Nuclear contributions to q are 0.533 and 0.582 for GeO and GeS respectively.
c
Contribution of noniterative triples.
d
Ref. [334].
b
Q [mb]
351
Tab. 5.15. Values for the nuclear quadrupole moment of the 27 Al nucleus. Data from Ref. [313].
System
Molecular values
27
Al19 F, Re
27
Al19 F, (v = 0)
27
Al35 Cl, (v = 0)
27 37
Al Cl, (v = 0)
Atomic value
27
Al(2 P3/2 )
Q a [MHz]
q b [a.u.]
Q [mb]
-37.750.08
-37.530.12
-30.40810.0027
-30.41120.0028
-1.099
-1.091
-0.879
-0.879
146.2
146.4
147.2
147.2
18.915260.0007
-0.5493
146.6
Reference atomic value

Reference muonic valued
140.31.0
1506
146.61.0
The molecular data from Refs. [333, 336] and the atomic data from Refs. [337, 338].
Molecular values are of DKHCCSD(T) quality, atomic ones are relativistically corrected numerical MCHF calculations.
c
Ref. [335].
d
Ref. [339].
b
Tab. 5.16. Values for the nuclear quadrupole moment of the 91 Zr nucleus. Data from Ref. [316].
System
Q a [MHz]
91
Zr16 O, (v = 0)
130.54990.0046
91
Zr32 S, (v = 0)
116.46090.0047
q [a.u.]
-3.11b
-3.16c (SO = 0.0056)
-2.86b
-179
-176
-173
Reference atomic values
-20610d
-257e
-23020f
-1763
Data from Ref. [340].

The DKHCCSD(T) values, 28 respectively 34 electrons correlated.
c
The 4component CCSD(T) values.
d
Ref. [341].
e
Ref. [342].
f
Ref. [343].
b
Q [mb]
352
year standard [304] value is 36040 mb, and in 2003 has appeared a solid state value of -66915
mb [345]. In 2005 Steve Cooke sent us new experimental antimony nuclear quadrupole coupling
constant values for diatomics SbN, SbP, SbF, SbCl [346, 347]. So we did CCSD(T) calculations
of electric field gradient with scalar relativistic effects covered by the spinfree IOTC method,
which led to our recommended value with rather conservative error bars of -55624 mb. Just
few months later Haiduke et al. published paper [348] based on the same experimental nuclear
quadrupole constants [346, 347] and DCCCSDT electric field gradient calculations. Their
recommended value is -54311 mb in a perfect agreement with our result and considerably
different from 2001 and 2003 values. Since Haiduke et al. used the fully relativistic method and
have a lower error limit, their value was adopted by Pyykko as the 2008year standard [305].
It is amusing that old Murakawas value [344], of course within its very large error bars, agrees
with it.
So far we did not discussed questions concerning the basis set choice, which is a crucial
issue in electric field gradients calculations. Usually completely uncontracted Gaussian function
basis sets are exploited. In calculations presented in this review extended even tempered basis
sets based on Huzinaga and Klobukowski paper [349] were used. In Tab. 5.18 is presented a
basis set dependence study for calculation of the electric field gradient at the nitrogen nucleus
in the NP molecule [324]. We can say that the phosphorous atom is a spectator atom here
and nitrogen is an actor atom. The spectator atom basis is rather fast saturated, there is
very little effect going from the B set to C set on phosphorous, while extending the C set of
nitrogen is significant. Let us discuss behaviour of the actor basis set. Going from the E
set to H set of the actor brings the q contribution of -0.0026 a.u., while using the B or C set
on the spectator brings no change. It has also been found that the atomic basis set for the
quadrupolar atomic centre needs to be carefully augmented with highexponent polarization
Tab. 5.17. Values for the nuclear quadrupole moment of the 121 Sb nucleus. Data from Ref. [323].
System
SbN
SbP
SbF
SbCl
Q a [MHz]
q [a.u.]
Q [mb]
649.669
620.350
-586.802
-515.124
-5.199
-4.931
4.407
3.788
-532
-535
-567
-579
1955 atomic valueb

2001 recommended valuec
2003 solid state valued
2006 IOTC-CCSD(T) value, this work
2006 DC-CCSD-T valuee
a
Refs. [346, 347].

Ref. [344].
c
Ref. [304].
d
Ref. [345].
e
Ref. [348].
b
-530100
-36040
-66915
-55624
-54311
353
functions. The rules established for valence properties do not apply in the case of electric field
gradients. The consequence of this finding is that one needs to use very large basis sets on the
quadrupolar centre. However, it has been simultaneously found that the diffuse part of the
polarization subset used for the quadrupolar centre and the spectator atom is less significant.
These findings give some guiding rules for the generation of basis sets for accurate calculations
of electric field gradients in molecules.
The presented results clearly proved that accurate ab initio calculations of electric field gradients at nuclei in combination with nuclear quadrupole coupling constants obtained from microwave spectra are reliable source of nuclear quadrupole moments. This is an outstanding example of the synergy effect of experiment and theory in contemporary science.
5.5
Dissociation and reaction energies
Relativistic effects on dissociation and reaction energies are described in several reviews, see,
e.g., Refs. [9, 10, 19, 20, 350]. Relativistic effects may affect the structure of compounds, they
lead to unusual thermochemistry and to relativistically controlled reactivity patterns, relativistically affected properties of atoms explain unexpected catalytic effects of heavy metals, they
explain synthesis of interesting new compounds, including large inorganic complexes and nanostructures [351]. Particularly important and interesting is the role of relativistic effect in catalysis [20]. Remarkable is the gold chemistry, (particularly metalmetal bonding and other phenomTab. 5.18. Basis set dependence of electric field gradient q calculated at N in the NP molecule [a.u.]. Data
from Ref. [324].
Basis set on N
Basis set on P
Study of the spectator atom basis set

A-16s11p5d
A-19s14p7d
qcorr,CCSD(T)
qa
-2.7312
0.3103
-1.0793
C-16s12p7d6f
C-16s12p7d6f
C-16s12p7d6f
A-19s14p7d
B-19s14p9d8f
C-19s15p10d9f
-2.7385
-2.7415
-2.7416
0.3104
0.2973
0.2973
-1.0865
-1.1026
-1.1027
E-16s14p9d8f
E-16s14p9d8f
E-16s14p9d8f
B-19s14p9d8f
C-19s15p10d9f
E-19s17p12d11f
-2.7114
-2.7115
-2.7115
0.2885
0.2885
0.2885
-1.0813
-1.0814
-1.0814
-2.7415
-2.7114
-2.7079
0.2973
0.2885
0.2876
-1.1026
-1.0813
-1.0787
-2.7416
-2.7115
-2.7079
0.2973
0.2885
0.2876
-1.1027
-1.0814
-1.0787
Study of the actor atom basis set

C-16s12p7d6f
B-19s14p9d8f
E-16s14p9d8f
B-19s14p9d8f
H-16s15p12d11f B-19s14p9d8f
C-16s12p7d6f
E-16s14p9d8f
H-16s15p12d11f
a
qHF
C-19s15p10d9f
C-19s15p10d9f
C-19s15p10d9f
The total value of q which includes the nuclear contribution 1.3416 a.u.
354
ena frequently described as the aurophilicity, Au is considered sometimes as a pseudohalogen) [12, 13, 351353]. Strong tendency of gold to form metalmetal interactions, particularly
with electropositive ligands, which donate electron density to the relativistically contracted and
stabilized the 6s orbital of gold, may lead to creation of remarkable complexes and nanostructures [351].
In the first part of this Section we will show the importance of scalar relativistic effects using
an example of a series of bond dissociation energies (BDE) of simple coinage metal fluorides, MF
and coinage metal aluminium molecules, MAl, with M=Cu, Ag, and Au. The interpretation of
the magnitude of BDE and the effect of relativity on this basic molecular property will be based
on previous information of relativistic effects on ionization potentials and electron affinities of the
metal. We also employ the knowledge of the polarity of MF and MAl molecules as discussed in
Section 5.3. Our selection of the two series combines molecules composed of a common coinage
metal atom and atoms having very different electronegativities, namely the most electronegative
atom, fluorine, with electron affinity as high as 3.40 eV, and aluminium with much lower EA,
0.441 eV. MAl molecules also serve as a group of intermetalics of great interest in materials
sciences. Results of DKHCCSD(T) calculations are collected in Tab. 5.19. For comparison, we
present also BDEs of analogous diatomics, as mercury fluorides and oxides [354]. These results
were obtained differently, by DFT calculations, with scalar relativistic effects calculated by the
Normalized Elimination of the Small Component method (NESC) which is related to the DKH
method. For AuF both methods give sufficiently similar results that allow to study the trends of
BDEs of the whole series as presented in Tab. 5.19. Good agreement of DFT with CCSD(T)
results also confirms the applicability of DFT in similar calculations which gives an argument
for using these methods for interpretation and prediction of bonding situation in large molecules
which are not amenable to sophisticated CC or other wave function calculations.
Understanding of the trends [21] in calculated De in the CuF, AgF, and AuF series [300],
and another group, CuAl, AgAl, and AuAl [255], is facilitated by the analysis of relativistic
contribution to BDEs, as shown in the sixth column of Tab. 5.19. We can follow essentially
the same physical arguing as was used in our discussion on dipole moments and dipole polarizabilities. CuF, AgF, and AuF molecules have large dipole moment and the polarity is M+ F .
CuAl, AgAl, and AuAl have small dipole moments and the polarity is opposite, M Al+ . The
charge transfer between the metal and the ligand, like the polarity, is mainly determined by the
electronegativity of participating elements. Some other aspects, particularly relativistic effects
in the bond length should be considered as well. Using simple arguments, relativistic changes
in De can be understood easily. Ionization potential of the coinage element, which determines
the charge transfer to the fluorine relativistically increases, especially in gold, and hinders the
charge transfer to the ligand. In accord with lower electron affinity of Al and the sign of the
dipole moment of MAl, the charge transfer from aluminium to the coinage element in MAl molecules is supported by the relativistically increased electron affinity of M, especially gold. Thus,
relativistic effects act in the two groups differently, destabilizing MF bonds and stabilizing MAl
bonds. The relativistic destabilization of the AuF bond is very large, 0.46 eV. Enormous, however, is the relativistic stabilization in AuAl, 1.7 eV. This means that De is twice as large at the
relativistic level in comparison with the nonrelativistic CCSD(T) result. Due to relativity, De of
AuAl is even larger than that of AuF. Without considering relativistic effects, De of AuAl would
be about half of De of AuF.
Up to now our main interest was focused on relativistic effects in this Section. Obviously,
355
electron correlation must not be forgotten in the discussion of bond energies. Like as relativistic
effects are unavoidable in understanding the underlying coinage metal properties, the one electron HartreeFock picture is completely useless in interpreting EAs and IPs of these species.
Therefore, it is not surprising that the electron correlation plays so important role in BDEs in
MeF and MeAl molecules. Electron correlation is essentially equally important in all molecules
presented in Tab. 5.19. It behaves monotonically, always increasing the bond dissociation energy
by about 1.35 1.55 eV in MF molecules and by 1.2 1.5 eV in MAl molecules. It is imperative
that both, the electron correlation and relativistic effects must be considered at highly sophisticated level together, to avoid inaccuracies caused by the nonadditivity of both contributions.
Quantitative explanation of the pattern of dissociation energies of the coinage diatomic molecules can not be achieved solely by arguments used so far. Other effects, including the exchange
repulsion which is connected with the relativistic shrinking of valence orbitals plays a role as
well. Also, there is the difference in the bonding character in MeF and MeAl molecules. All
are characterized by the bonding valence (sp) orbital. MeF bods are more ionic, which is
demonstrated not only by larger dipole moments, but also by the dipole moment curves, which
are almost linear [300] and correspond to what can be expected from the Coulomb law. Just AuF
becomes more covalent and less polar due to the relativistically enhanced ionization potential of
Au. The MeAl bonds are more covalent. Due to the stabilization of the metal valence s electrons,
the valence MeAl orbital can be relativistically stabilized as well. This further contributes to
the enhancement of De of MAl molecules, with M being a coinage metal.
Tab. 5.19. Electron correlation and scalar relativistic effects in dissociation energies Ee [eV] of diatomic
molecules. Data from Refs. [255, 300, 354].
Molecule
nr-SCF
nr-CCSD(T)
Ecorrel
DKH-CCSD(T)
CuF (1 + )
AgF (1 + )
AuF (1 + )
2.53
2.17
2.04
4.08
3.55
3.39
1.55
1.38
1.35
4.04
3.32
2.93
-0.04
-0.17
-0.46
4.4 a
3.6 a
3.01b
CuAl (1 + )
AgAl (1 + )
AuAl (1 + )
0.59
0.41
0.51
2.08
1.58
1.68
1.49
1.17
1.17
2.25
1.95
3.41
0.17
0.37
1.73
2.32 c
2.22 c
3.34 c
AuI (1 + )d
AuO (2 )d
HgO (3 )d
HgF (2 + )d
a
nr-DFT
2.53
1.86
1.16
2.70
NESC-DFT
2.47
2.27
0.41
1.44
Erelat
Erelat
-0.06
0.31
-0.75
-1.26
Exper.
2.31
Ref. [296].
Ref. [355].
c
See data in Ref. [255].
d
Scalar relativistic effects calculated by the Normalized Elimination of the Small Component
(NESC) B3LYP DFT data, Ref. [354].
b
356
The overall pattern for De of all molecules presented in Tab. 5.19 corresponds to available
experimental data only when relativistic effects are considered. In general, the agreement between theory and experiment is very good even for heaviest members of both groups. The
thermodynamic stability of AuF, Au2 F2 and related compounds has been subjected to many
discussions [20, 356]. Reliable experimental evidence and accurate experimental De of AuF was
not available until 2002 when Okabayashi et al. [355] have observed AuF in the X1 + ground
state by employing a sourcemodulated microwave spectrometer. In fact, they were motivated
by theoretical papers cited above. Their De , 3.01 eV, agrees with theoretical data (compare to
our value 2.93 eV) very well.
Pioneering analysis of the relativistic bond stabilization and destabilization of diatomic gold
molecules was published by Schwerdtfeger et al. [357] who pointed out that it depends on the
electronegativity of the ligand. They also showed that the largest relativistic destabilization of
gold containing diatomics holds for AuF, like in our gold compounds and the largest stabilization
was calculated for AuLi (-174 kJ/mol).
Comparison of the gold bonding with the mercury bonding in diatomic molecules was published recently by Kraka et al. [354]. They analysed the scalar relativistic effects at the DFT level
considering different bonding situation in the series of molecules, residual bonding, lone pair
repulsion, and the dblock effect. The interplay of the various electronic effects leads to strongly
differing trends in calculated BDEs, which can be rationalized with a simple MO model based on
electronegativity differences, atomic orbital energies and their change due to scalar relativity. A
relativistic increase or decrease in BDE is directly related to relativistic changes in the 6s orbital
energy and density. An increase in BDE upon including scalar relativistic effects is found when
charge transfer occurs from X to M due to the low electronegativity of X, like it was described
in our comparison of the MF and MAl series for coinage metals above.
Valence electronic structure of molecules treated in the MF and MAl series, as presented in
Tab. 5.19, is relatively simple. More complicated picture exhibit molecules in which the valence
electronic structure is determined by, e.g., orbitals. A good example is the PbO molecule and its
anion, PbO . Recently both species were studied [286] utilizing CCSD(T), CASSCF/CASPT2,
and the Fock space FSCCSD methods. Relativistic effects were treated by the spinfree, two
component infinite-order DouglasKrollHess, DiracCoulomb and the Restricted Active Space
State Interaction RASSISO methods. The potential energy curve for the ground X1 + /0+ state
is presented in Fig. 5.2 in Section 5.1 together with the potential energy curve of the 2 1/2 state
of the PbO anion (see Section 5.1 on electron affinities). The anion splits due to the spinorbit
effect into the 2 1/2 and 2 3/2 states, respectively. The PbO molecule in its ground state exhibit
no split spinorbit states. Nevertheless dissociation energies of both species are significantly
affected by spinorbit effects. As it is clearly demonstrated in Fig. 5.2, the large SO contribution
in both PbO and PbO arises from the SO splitting of the product, the Pb atom. The lowest
energy state of the neutral lead element, j1 = 1/2, j2 = 1/2 (X3 P0 ), was obtained at the
CCSD level by inserting two 6p electrons into two shells with quantum numbers mj = 1/2 and
mj = 1/2. It lies by 8526 cm1 lower than the spinaveraged state of Pb. This contributes to
De of both PbO and PbO significantly. The oxygen 3 P2 ground state is affected by the SO effect
much less (the same is valid for the O anion). Data collected in Tab. 5.20 demonstrate that spin
orbit effects reduce De by 1.06 1.36 eV (see also Refs. [207, 297, 358, 359]). The magnitude of
SO effects as obtained utilizing the RASSISO method [102] method (see Section 4.7.2) in which
spinorbit relativistic effects are treated a posteriori, do not differ significantly from results in
357
which SO effects are treated inherently within two and fourcomponent calculation working
with spinors. Theoretical dissociation energies De of PbO and PbO (3.91 eV and 3.20 eV)
obtained by the BSSCCSD(T) method with the ANORCC basis set agree with experiment
perfectly, within 0.04 and 0.07 eV, respectively. Referring to BSS and BSSsfCCSD(T) results
[286] for Re and e of PbO one concludes that the bonding situation of PbO is affected by
SO effects very little at the correlated level. Consequently, the lowering of De is due almost
exclusively by SO effects of the lead element. The electron correlation, relativistic scalar and
spinorbit effects are crucial for obtaining good agreement of theoretical and experimental data
for EA and De . Contributions to the spinorbit effect on De of a series of diatomics were
discussed e.g., in Ref. [207]. The contribution to De arising from SO effects in the molecule,
in the reaction product, or as the effect arising from an inherent SO interaction within spinors
depend on a specific bonding situation in each molecule.
Challenging are scalar relativistic and fourcomponent treatments of spinorbit effects in
closedshell superheavy element monohydrides or their cations, AH or AH+ , with A= 111
120. Thierfelder et al. [302] presented dissociation energies, electron affinities, and spectroscopic
constants of these properties and trends within the series of these molecules.
Tab. 5.20. Dissociation energies De of the PbO molecule and the PbO anion [eV].
Method
PbO molecule
PolDK basis set
ZORAa
CASPT2b
BSSCCSD(T)c
DCCCSD(T)c
ANORCC basis set
BSSCCSD(T)c
Experimentd
PbO anion
ANORCC basis set
BSSCCSD(T)c
Experimente
a
spinfree
spinorbit included
5.51
5.10
4.85
4.82
4.15
4.04
3.63
3.69
5.20
3.91
3.87
4.31
3.20
3.13
Ref. [207], GGA functional.

Spinorbit effect included using RASSISO method, De recalculated from Do using e , [297].
c
Ref. [286].
d
Using experimental Do (3.83 eV) and experimental e , [296].
e
Recalculated from Do (3.09 eV 0.07) and experimental e , [287].
358
5.5.1
Quadruple, quintuple or sextuple bonds?
Our discussion of unusual chemical bonds resulting from relativistic effects can be introduced
by a few examples describing a strong multiple bond in gold compounds. Two main effects
leading to strong Au bonds are the stabilization of the 6s shell and the destabilization of the
5d shell. Barysz and Pyykko [360] have reported in 1998 about AuBe+ , AuC+ , and AuSi+ ,
and AuMg+ cations, that time experimentally unknown. First three species have dissociation
energies larger than any other De known for gold. At the same time, the strikingly short bond
length and population analysis of AuC+ suggest that this diatomic cation may approach the
first known triple bond to gold with dissociation energy 3.815 eV and a bond length 1.766 A.
+
Similarly, PtSi and PtTh also form triple bonds, while AuTh does not show any evidence of a
multiple bond. In triple bonds of Au and Pt compounds a decisive role is played by their 6s and
5d orbitals. Particularly strong AuC and AuF bonds (De 387 and 390 kJ/mol, respectively)
were reported [361] also in XAuC molecules (X is an halogen atom, F, Cl, Br, I).
The existence of unprecedented quadruple bonds between two metal atoms dates from 1965,
when Cotton and Harris [362] have reported on the crystal structure of K2 Re2 Cl8 2H2 O. This discovery affected inorganic chemistry considerably. Inorganic quadruple bond is, however, not the
end of the story, in which, as stressed by Ritter [363], theory plays an important role. Theoreticians, particularly Roos and coworkers [364, 365] have investigated Cr2 compounds which can
form quintuple or even sextuple bonds. In 2005, Power and coworkers [366] reported evidence
for the first quintuple bond between two chromium atoms in RCrCrR compounds. Theoretical calculations [367] on the model compound, phenylCrCrphenyl essentially confirms the
picture which describes a quintuple bond between the transition metal atoms. As analysed by
Gagliardi and Roos in their review [368], the chromium atom, which has the ground state with
six unpaired electrons, 3d5 4s1 (7 S), can offer altogether 12 electrons in the dimer, which have a
potential of creating even sextuple bonds. Nevertheless, an effective bonding order is 4.46 and
the dissociation energy D0 of Cr2 is relatively low, 1.65 eV. The problem is that there is large difference in size between the 3d and 4s orbitals of Cr. When 3d orbitals reach an effective bonding
distance, the orbital constructed from 4s electrons is already far up on the repulsive part of the
potential energy curve [368]. The difference in size decreases for heavier atoms. For this reason
compounds with participation of Mo or W where the relativistic contraction of the 6s orbital
and the relativistic expansion of 5d orbitals makes their bonds much stronger, the effective bond
order is about 5.2.
The joint effort of experimentalists and theoreticians working in a fascinating chapter of quintuple and possibly sextuple bonds is described by Ritter [363] and Power [366]. Particularly interesting is the prediction of a quintuple bond in the U2 molecule by Gagliardi and Roos [368,369].
The ground quintet state of the uranium atom has the 5f 3 6d1 7s2 valence electronic structure.
All valence electrons of the U atom are energetically close to one another. The energy cost of
unpairing the 7s electrons is low and consequently, uranium has in fact six electrons available
for creation of the sextuple bond. Involved are , , , and orbitals. Calculations [369] involve
electron correlation and scalar relativistic effects at the DKHCASPT2 level, spinorbit effects
coupling was included by allowing the 16 CASSCF wave functions of the gerade symmetry to
mix under the influence of the SO Hamiltonian. The analysis of valence orbitals and the bond
order analysis indicate that the molecule exhibits a quintuple bond. Similar quantum chemistry
methods predict multiple bonded stable diuranium compounds as diuranium chlorides and three
359
different carboxylates [365]. Multiple bonds exhibit also Th2 and Pa2 diatomics [370]. Unusual bonding situation and stabilities of all these novel compounds clearly result from striking
relativistic effects.
5.6
Intermolecular interactions
The literature on interaction energies in which participate atoms with high atomic numbers is
very rich. It includes both strong closedshell and van der Waals type interactions in inorganic
chemistry [11, 12], various metalligand interactions up to interactions which represent the nonconventional hydrogen bonds in which gold participates as an electronegative atom [371, 372].
Clearly, this is related to the relativistic enhancement of the Au electron affinity. Much attention
is focused on properties of model metallic complexes [352, 373, 374] since the knowledge of the
bonding properties of such species can be used in materials sciences with possible technological applications. One area is the construction of SelfAssembled Monolayers (SAMs) which
are typically formed from a coinage element interacting with a ligand, like alkylthiolates with
different alkyl chains adsorbed on a metal surface [375379]. Among important characteristics
of the functionality of SAMs belongs the bond strength between the metal and the alkylthiolate
ligand. Much attention is paid to the relativistic origin of the difference between Cu, Ag, and Au.
Relativistic effects in model systems related to the construction of SAMs were studied by Hirao,
Marx, Stich
et al. [376, 377, 380]. Potential applications of SAMs based on the metalorganic
ligand interaction include nanofabrication, corrosion prevention, and the development of biocompatible materials. Interactions of heavy metals with different ligand molecules are also important
in many biological processes or in developments of drugs in human medicine. [381384]. Famous are platinum complexes (cisplatinum) with applications as anticancer drugs. There is an
indication about possible role of relativistic effects in the anticancer activity of these drugs [383].
Realistic theoretical models relevant for construction of SAMs and other metalligand organic or inorganic complexes are usually too large to be accessible to accurate highly correlated methods. Therefore, most frequently used methods for research in this area are DFT techniques [375, 378]. Nevertheless, for the basic knowledge underlying the bonding mechanism of
such species and their geometric features accurate high level ab initio data may be very useful.
In this Section we present several examples in which differences between nonrelativistic and
relativistic interaction energies of small metalligand complexes are related to relativistically affected properties of the heavy metal. It is clear that physically relevant are only results considering both electron correlation and relativistic effects. However, our aim is not only presenting data
on structural and energetic features of interacting species. We will rather try to demonstrate that
relativistic effects may contribute to the understanding of some trends of stabilities of different
complexes involving a heavy metal element. In this sense comparing the relativistic and nonrelativistic models may serve as a useful tool for achieving our goals. We will focus our attention
primarily on complexes with Cu, Ag, and Au employing data presented in previous Sections of
this paper, particularly ionization potentials, electron affinities and polarizabilities of these atoms
(see Section 5.1 and 5.3). Nevertheless we hope that data presented in Tab. 5.21 [385387] are
sufficiently instructive and can be generalized to other complexes as well (at least to those which
have analogous valence electronic structure).
Initially we were prompted to undergo a comparative study of Ag interacting with H2 O and
NH3 by experimentalists [389, 390] who could indicate the existence of the AgNH3 complex in
360
the Resonantly Enhanced Multiphoton Ionization (REMPI) spectra. In contrast to the data for
AgNH3 , the REMPI spectrum for the similar complex of Ag with the water molecule does not
exist. According to the dipole moment data (1.85 Debye for H2 O and 1.47 Debye for NH3 )
[32] the leading dipoleinduceddipole contribution to the interaction energy should favour the
AgOH2 complex over the AgNH3 complex. This contradicts experimental observations. Also,
Tab. 5.21. BSSE corrected DKHCCSD(T) relativistic and nonrelativistic interaction energies [mhartree]
of M OH2 , M SH2 , M NH3 , and M PH3 complexesa (doublet states) and the Cu SCH3 complex in triplet and singlet states. PolDK and Pol basis sets are used for DKH and nonrelativistic calculations, respectively. Cu SCH3 calculated with the augccpVTZDK basis. Data from Refs. [385388].
EDKH
Complex
ENR
rel b
Cu OH2
Ag OH2
Au OH2
-3.78
-1.81
-1.77
-3.79
-2.13
-1.62
0.01
0.32
-0.15
Cu SH2
Ag SH2
Au SH2
-5.99
-1.99
-9.08
-4.53
-1.53
-1.40
-1.46
-0.46
-7.68
Cu NH3
Ag NH3
Au NH3
-16.68 (18.9, -21.4c )

-6.87 (-8.7, -10.7)
-14.64 (17.9, -20.4)
-14.86
-6.52
-5.23
-1.82
-0.35
-9.41
Cu PH3
Ag PH3
Au PH3
-9.89 (-12.53c )
-2.24 (-4.40)
-18.31 (-25.42)
-7.25
-1.15
-1.05
-2.64
-1.09
-17.26
Cu SCH3 , 3 A
Ag SCH3
Au SCH3
00
-20.5c
-6.9
-28.4
Cu SCH3 , 1 A
Ag SCH3
Au SCH3
-94.4c
-75.1
-87.2
Qualitative description of the main structural features of M-L complexes: Optimized structures
of M OH2 and M SH2 complexes are nonplanar, with the outofplane angle 58, 55, and
67 deg for M OH2 and 80, 89, and 81 deg (almost) perpendicular structure of Cs symmetry
for M SH2 . M NH3 , and M NH3 complexes have linear M N (or M P) bonds in the
main axis of the complex with the C3v symmetry.
b
Relativistic contribution to the interaction energy rel = EDKH - ENR .
c
Results with relativistic augiccpVTZDK basis sets [387] and complete basis set limit [388]
in parentheses.
361
the model based on the longrange expression for the induction energy,
1
E I = M 2L (3 cos2 + 1)R6
2
(5.13)
(M and L are the polarizability of the metal atom and the dipole moment of the ligand, respectively), would lead to the planar M OH2 structures with M interacting with C2v molecules in
the direction of their principal axis. Theoretical calculations [391] suggest that it is in fact non
planar. Prompted by the puzzling character of the REMPI experimental data we decided [385] to
perform DKHCCSD(T) calculations for all three coinage atoms and their complexes with H2 O
and NH3 . Rigorous theoretical information about the structure and energetics was until recently
scarce, see e.g., Refs. [391,392] and any systematic comparison of all these complexes was missing. Calculations were performed by the so called supermolecule method. For the weakly bound
dimer AB we can assume that the deformation of geometries of participating subsystems, A and
B, would be small. The interaction potential, VAB depends, under this assumptions, only on
intermonomer coordinates (that is, the M. . . O or M. . . N bond distance and the orientation of
M with respect to the ligand L) and is calculated as:
VAB = EAB (AB) EA (AB) EB (AB),
(5.14)
where (AB) denotes that the calculated energies refer to the basis set of the complex. In other
words, the complex (supersystem) and both subsystems are calculated with the same basis set.
This approach [393] eliminates the basis set superposition error (BSSE) and leads to balanced
energies of all participating species. Results are collected in Tab. 5.21. Along with interaction
energies presented in former studies on M OH2 and M NH3 complexes [385] we also summarize interaction energies in M SH2 , M PH3 and M SCH3 complexes [386, 387, 394].
The structures of the complexes are schematically presented in Fig. 5.10 (for details see the original literature [385387]). M OH2 and M SH2 complexes are nonplanar, with M approaching the O or S atom in the direction of the lone pair of the ligand. It is interesting that nonplanar
structures were found not only for interactions of neutral atoms with lonepair ligands like H2 O
but also with cations of M [395398] in which one expects a dominating charge dipole interaction, with planar structure of M+ XH2 complexes. Analogously in M SCH3 complexes
a metal element approaches the SCH3 molecule in the direction of the lonepair on the sulphur
atom, leading to the MSC angle of 76 up to 82 deg, depending on the metal and the spin state
of the resulting complex. These structural features point out on the inadequacy of the induction
model of interactions to correctly describe differences in the structure and energetics of metal
atom complexes with lonepair ligands. M NH3 and M PH3 complexes preserve the C3v
structure of isolated molecules. Again, the metal element is oriented towards the lone electron
pair of the ligand.
Returning back to the missing evidence of the existence of the Ag OH2 complex by
the REMPI spectroscopy, as contrasted to the Ag NH3 complex, the interaction energies in
Tab. 5.21 seem to offer the explanation of this observation. Nonrelativistic CCSD(T) interaction
energy in Ag NH3 is 3.1 times larger than in Ag OH2 , the DKH relativistic ratio of the two
energies is slightly higher, 3.8. This is just a computational observation of higher stability of
NH3 complexes, not yet the explanation. More interesting is similar ratio of interaction energies
in Au NH3 and Au OH2 complexes. The nonrelativistic ratio of interaction energies of the
362
Fig. 5.10. Schematic structures of complexes M OH2 (and M SH2 ), M NH3 (and M PH3 ) and
M SCH3 .
two Au complexes is 3.2, about the same as with Ag NH3 and Ag OH2 . However, when
considering scalar relativistic effects the ratio enhanced to 8.2. Also, relativistic interaction energy is much higher for Au NH3 than for Ag NH3 (for complexes with water both values are
almost the same). One could think about a dispersion interaction as another contributing energy
component to the stability of complexes under consideration. With the same ligand interaction
energies should be proportional to the magnitude of atomic polarizabilities of the metal atom.
Clearly, dispersion forces, which depend on the product of polarizabilities of the metal and the
ligand, respectively, do not explain observed order of stabilities for Ag and Au. Nonrelativistic
polarizabilities are similar for Ag and Au. Correspondingly, interaction energies for the Ag and
Au L complexes at the nonrelativistic level, see Tab. 5.21, are also similar. Relativistic effects
reduce the polarizability of gold from 64 a.u. to about 36 a.u., so that the relativistic polarizability of Ag is the largest from among all coinage elements, (see Section 5.3). However, interaction
energies of Au with all ligands relativistically increase (albeit only a little for Au OH2 ). Clear
indication of different mechanisms of interactions in which participate water and ammonia and
other ligands provide different relativistic contributions to the interaction energy, last column of
Tab. 5.21. For all M SH2 , M NH3 and M PH3 complexes is the relativistic contribution
large and negative. For complexes with Au is the relativistic effect highly stabilizing. The stability of Au PH3 is practically completely caused by relativistic effects. For this reason we call
the bond in this complex as a relativistic bond [387]. For Cu OH2 and Ag OH2 complexes,
363
Fig. 5.11. Relativistic contributions to the interaction energies of Cu (middle line), Ag (upper line), and
Au (the lowest line) as a function of the IP of the ligand L, PH3 (IP= 9.87 eV), NH3 (IP= 10.07 eV), H2 S
(IP= 10.46 eV), and H2 O (IP= 12.61 eV). Interaction energies from Refs. [385387], ionization potentials
of ligands from the NIST database [399].
in contrast, is the relativistic contribution destabilizing. Au OH2 is relativistically stabilized

very little.
Our hypothesis on the interpretation of the interaction energies pattern for complexes of the
coinage metals with different ligands is based on the idea that the basic mechanism behind interaction energies in M L complexes is the charge transfer form the ligand to the coinage metal.
It is mediated by the lone pair of the ligand directed towards the metal, which explains linear
M X bonds in M XH3 complexes and the nonplanarity of M OH2 complexes. Hence, we
expect that relativistic contributions to interaction energies should be more important for ligands
which have low ionization potentials, as is shown in Fig. 5.11. Charge transfer is substantially
enhanced by the relativistic contribution to the electron affinity of the metal. This expectation
should hold particularly for complexes with Au, which has the far highest electron affinity out
of the three coinage metals. High stability of Au SH2 , Au NH3 and particularly Au PH3
complexes is therefore linked with a large relativistic enhancement of the electron affinity of Au.
364
Fig. 5.12. Interaction energies of Cu, Ag, and Au complexes with PH3 . Comparison of nonrelativistic
(empty symbols, dotted lines) and DKHCCSD(T) relativistic (full lines) potential energy curves. Relativistic contribution Erelativistic at the equilibrium M. . . P bond distance is shown by full vertical lines.
Inserted is a longrange part of the DKHCCSD(T) potential energy curves. See Ref. [387].
Clearly, the charge transfer from the ligand to the metal is hindered in M OH2 complexes since
the ionization potential of water is too large. This is compensated by large relativistic EA of gold
in Au OH2 so that relativistic contribution to E becomes negative, but very small.
The high stability of the Au PH3 complex (Tab. 5.21) is considerable. Interaction energy
in Au PH3 is much higher than in Au NH3 . To ensure that relative stabilities are not affected by basis set effects relativistic DKHCCSD(T) calculations with PolDK basis set [385]
were supplemented [387] with calculations with much larger augccpVTZDK basis sets. The
picture remains the same, just interaction energies are higher upon extending the basis set. The
crucial observation that the bond strength in Au PH3 is about twice as large as in Cu PH3
holds with both relativistic basis sets. Also, the interaction energy in Au PH3 is higher than
in Au NH3 (note, however, that interaction energies in Cu NH3 and Ag NH3 are higher
than in their Cu PH3 and Ag PH3 counterparts).
The bond strength in Au PH3 with augccpVTZDK basis set is as high as 67 kJ/mol,
nearing the bond strength of a regular chemical bond. We stress that with the DKHHartreeFock
relativistic approximation all M PH3 complexes are repulsive. Upon including the electron
correlation by the highlevel CCSD(T) method but neglecting relativistic effects the Au PH3
complex exhibits low attractive interaction energies corresponding just to the longrange van der
Waals weak interaction (with a shallow minimum at the M P distance of 6.9 bohr). As it is
365
demonstrated in Fig. 5.12, considerable bond strength results only upon considering (scalar) relativistic effects on top of the electron correlation. Hence we characterize the interaction between
Au and PH3 as a relativistic bond. The enhancement of the bond strength is accompanied
by considerable shortening of the M P bond distance, see Fig. 5.12. In Au PH3 is the
Au P bond length shortened by about 2 bohr due to relativistic effects. This is related to relativistic shrinking of the valence 6s orbital of Au. Experimental data for interaction energies
in M PH3 complexes are missing. For analogous Cu NH3 and Ag NH3 complexes Li
et al. [388] present DKHCCSD(T) interaction energies extrapolated to the CBS limit. Both
our calculations with larger augccpVTZDK basis set and the CBS limits (50 56 kJ/mol for
Cu NH3 and 18 28 kJ/mol for Ag NH3 ) agree fairly well with experimental values, 47
15 and 8 13 kJ/mol for the two complexes, respectively, as presented in Ref. [388].
Another interesting relativistic effect is shown in the inset in Fig. 5.12. It represents a long
range region of the potential energy curves of M PH3 complexes. The longrange van der
Waals bonding region in M. . . PH3 is governed mainly by electric properties of participating
species. We remind that the highest polarizability out of the three coinage metals was calculated
for the silver atom, see Section 5.3. The potential energy curves and the inset displayed in
Fig. 5.12 show that beyond R=6.0 bohr is the Ag. . . PH3 complex more stable than Cu. . . PH3
and beyond of about R=10 bohr it tends to be the most stable M. . . PH3 species, in line with the
highest polarizability of Ag. At R=10.0 bohr are relativistic binding energies of all M. . . PH3
complexes smaller than nonrelativistic ones, which is attributed to the fact that relativistic effects
reduce the dipole polarizability of all coinage metals. Longrange induction and dispersion
forces based on relativistic alterations of dipole polarizabilities and ionization potentials of the
metal elements and electric properties of ligands were discussed in detail in Ref. [385].
The analysis of energy contributions due to polarization forces and partial charge transfer
from the lone pair of the ligand molecule to the ns valence orbital of M provides a uniform
interpretation of the interaction energy data and explains some irregularities in the energetics and
structures of the investigated systems. Similar interpretation can be also forwarded to explain
the bonding and structure of complexes between ligands and heavy metal cations [395, 397, 398,
400, 401]. Analogous mechanism applies to interesting X(AuPH3 )2 complexes [402] with X
representing an electronegative atom.
Straightforward insight into the interplay between the electron correlation and relativistic
effects can be demonstrated by excitation amplitudes obtained by the nonrelativistic and relativistic treatment of the electron correlation, respectively. CCSD single excitation amplitudes of Ag
and Au, see Fig. 5.13, are very similar in the nonrelativistic case, confirming once again that the
nonrelativistic gold would behave like silver. Enhancement of the CCSD excitation amplitudes
resulting from the relativistic DKH calculations is attributed to a large mixing of electron correlation and relativistic effects. Note that relativistically treated Au amplitudes behave similarly
as amplitudes of Cu. Relativistic effects in Au diminish the energy gap between valence 5d and
6s orbitals. Therefore, we may say that from the point of view of the electron correlation the
relativistically treated Au becomes similar to Cu. In a sense, both behave like transition metal
elements. Relatively large excitation CC amplitudes occurring in Cu. . . NH3 are close to the
limit at which the single reference CCSD(T) can still be safely used. In general, large amplitudes
are characteristic for quasidegenerate systems in which the energy gap between different states
(typically between the ground state and a lowlying excited state) are not sufficiently separated.
For such systems iterative calculation of higher than single and double excitation amplitudes
366
Fig. 5.13. Nonrelativistic and relativistic CCSD single excitation amplitudes of Cu, Ag, and Au (M)
complexes with NH3 as the function of the MN bond length. Equilibrium M. . . P bond distances of
M. . . NH3 are shown by arrows. Based on data from Ref. [385].
(i.e. triples and particularly quadruples) is needed when we stay within the single determinant
reference scheme, or we can use multireference MRCC or MRCI approach, but all these alternatives are very expensive. This topic [38, 40] is beyond the scope of the present paper. More
practical alternative is using the RASSI and CASPT2 method mentioned in Section 4.6.3.
Larger metalligand complexes are usually tractable only with DFT methods. Therefore,
it may be instructive to compare DFT and CCSD(T) results, when available. In the case of
Au. . . PH3 , the present interaction energy is quite different from the Local Density Functional
(LDF) value calculated by Haberlen et al. [403] (124 kJ/mol 47.2 mhartree). Similarly, Bernhardt et al. [404] found that all DFT predicted binding energies of CO and O2 with small anionic
gold clusters are much larger than energies obtained from the gasphase kinetic data for the oxygen and CO adsorption on small massselected coinage metal cluster anions. Solely, the relativistic CCSD(T) result by Schwerdtfeger et al. [405] coincides favourably with binding energies
deduced from the temperature dependent rate constants. Both underestimated and overestimated
binding energies were found in an elaborated investigation of the performance of DFT with various exchangecorrelation functionals by Li et al. [388] for M. . . NH3 complexes (M=Na, Al,
Ga, In, Cu, Ag, Au and their cations). The analysis of the performance of DFT methods employing different functionals [406] shows that simple local density functionals perform quite well in
molecules containing the AuAu aurophilic bonds.
An example of the different bonding mechanisms related to the relativistic effects represent
367
M. . . SCH3 complexes (M= Cu, Ag, and Au) [394]. These complexes may serve as the simplest
representatives of compounds participating in the creation of SAMs. Optimized structures of
M. . . SCH3 complexes (the Cs symmetry) are schematically shown in Fig. 5.10. The methylthio
radical SCH3 is an openshell species having the two lonepairs at the sulphur atom, one of
which is doubly occupied, another one is singly occupied. The first ionization potential calculated by the CCSD(T) method is 9.18 eV (lower than of other ligands discussed previously). It
agrees fairly well with the experimental value of 9.262 eV obtained by the photoionization mass
spectrometry [407]. Applying the same symmetry as considered in M. . . SCH3 complexes it rep00
0
resents the process 2 A 3 A . The second IP representing ionization of the electron from the
2 0
1 0
singly occupied (the A A process) is 10.57 eV. The structure of SCH3 is not strictly C3v
since it is deformed by the JahnTeller effect. This effect (95 cm1 ) is much smaller than the
spinorbit splitting of the two states, the X2 E3/2 ground state and the 2 E1/2 state, respectively,
which differ by 356 cm1 [408]. We were able to reproduce this value by the RASSI method
within 10 cm1 .
Two different states of the M. . . SCH3 complex have similar structures (Fig. 5.10). Interaction energies for both states are compared in Tab. 5.21 with M. . . L interactions discussed above.
A weaker bonding results from the metal element approaching the doubly occupied lonepair
orbital of SCH3 . The second lonepair remains singly occupied and thus the complex is the
00
triplet 3 A . The interaction energy is higher than in Au PH3 . In accord with the dominating
charge transfer mechanism this corresponds to lower IP of SCH3 than is IP of PH3 and other
closedshell ligands. Again, as with Au PH3 complexes, the M. . . SCH3 interaction energy
is the highest for Au. . . SCH3 . Preliminary calculations [394] indicate, that relativistic effects
00
stabilize the 3 A state of the Au. . . SCH3 complex even more than Au. . . PH3 , as is should be if
0
our interpretation of the bonding mechanism is correct. The bonding character of the 1 A state
3 00
of M. . . SCH3 is different than for its A state. It is essentially the bond created by the ns
electron of the coinage metal and the singly occupied lonepair valence orbital of SCH3 . For
all M. . . SCH3 it is quite strong, 248 kJ/mol for Cu. . . SCH3 and slightly less, 229 kJ/mol for
Au. . . SCH3 . Structural and energy characteristics of these complexes are relevant for constructing SAMs [375379].
To summarize, it is appropriate to note that the interpretation of the structural and energy
pattern of metalligand complexes based on the charge transfer mechanism represents just a
principal driving force of their interaction. The basic properties of heavy metal properties participating in their interactions are the enhancement of EA and IP, the relativistic stabilization and
contraction of valence s and pelectrons, destabilization and expansion of delectrons, lowering of the dipole polarizability, and the overlap of valence orbitals of the metal element and the
orbitals of the ligand. In a particular case of the binding energy in Au. . . PH3 as compared with
Au. . . NH3 it may increase also due to the participation of the bond reinforced in M. . . PH3
complexes (back donation). This idea goes back to Mulliken and Person [409] who call such
mechanism as a twoway donoracceptor action. The bonding model of M. . . L neutral and
charged clusters focuses on the donoracceptor donation and acceptordonor back donation [374, 410, 411]. Lim et al. [410] have compared the geometries and interaction energies
of the series of M NH3 (M=K, Rb, Cs, and Fr and their cations) with coinage metal NH3
complexes, analysed the character of highest doubly occupied bonding orbitals (HOMO1) and
the highest singly occupied antibonding orbitals within the two series. They pointed out a significant 5d admixture in the HOMO1 orbital in Au. . . NH3 , which appears to be even more
368
important in Au. . . PH3 . The donation and acceptordonor back donation mechanism was
used by Li et al. [374] in their DFT relativistic calculations and the analysis of relative stabilities
of metalphosphorus WAu12 PX3 (with X=H, F, Cl, Br, I, Me, and OMe) and various metalAu
cationic MAu12 PH3 complexes. Similar clusters are powerful catalysts and have potential application in nanoelectronics, nanosensors, and as biological markers [351]. Complexes of adenine
and adenosine monophosphate with gold were investigated experimentally and theoretically by
Kundu et al. [412]. The central focus was on understanding of these interactions in relation to
the development of functional biomedical nanodevices and accurate biosensors.
An interesting comparison of molecular gold halides with molecular mercury halides and
their complexes was published by Hargittai [413]. Differences between dimers, like Au2 X2 and
Hg2 X4 (X is a halide atom) are attributed to relativistic effects. The peculiar structure of the
Hg2 X4 dimers and the crystal structure of mercury halides is also a consequence of relativistic
effects [414]. Kryachko and Remacle [415] studied different neutral and charged AuZ
n (NH3 )m
complexes (Z = 0, 1). They have proposed such complexes as simple logic gates based on
goldammonia bonding patterns using the conformational manifold of these complexes with
conventional Au. . . N and the hydrogen NH. . . Au bonds.
So far we did not mention explicitly spinorbit interactions in describing the metalligand
complexes. As an example of the importance of spinorbit contributions may serve interactions
of the superheavy element 112 (E112, eka-Hg) interacting with Au, Au4 and Au6 (and comparison of these complexes with HgAu, HgAu4 and HgAu6 ) [416]. These complexes are considered
as simple models of adsorption of E112 and Hg on the Au(111) surface. Comparative analysis
also indicates a strong underestimation of scalar relativistic binding energies by conventional
relativistic DFT methods. In complexes involving a group of metals, like lanthanides, the consideration of spinorbit interactions is unavoidable.
5.7
NMR properties of molecules
Relativistic effects have a profound influence on a wide scale of molecular properties. Here
we focus on nuclear magnetic resonance (NMR) shielding constants. Review papers on NMR
properties discussing also relativistic effects, let us name a few, are by Helgaker et al. [417], or
by Vaara [418].
We present the relativistic methodology for computing NMR properties and will briefly show
how to remove the gaugeorigin dependence and will report few demonstrative results. Likewise
we will mention the importance of the electronic correlation on NMR properties.
5.7.1
Fourcomponent calculations of NMR shielding constants
Individual components of the NMR shielding tensor for the nucleus K are calculated within the
BornOppenheimer approximation as the second derivative of the electronic energy at the zero
perturbation strength:

d2 E
K

,
(pq) = xx, xy, xz, yy, yz, zz.
(5.15)
pq =
dmK,p dBq {mK }=0, B=0
In our implementation in Ref. [419] we are using the DiracHartreeFock wave function. To
evaluate the second derivative, Eq. 5.15, one has to provide the oneelectron Dirac Hamiltonian,
369
Eq. 4.35, with perturbations created by both, a uniform external magnetic field, B, and internal nuclear magnetic moments, mK . These are represented through the vector potential A in
accordance with the principle of a minimal electromagnetic coupling [127, 420]:
X
A = AB +
AmK .
(5.16)
K
The vector potential of a uniform external magnetic field B is usually chosen in the gauge
AB (r) =
1
B rG ;
2
r G = r RG ,
(5.17)
where RG is the selected gauge origin. The vector potential associated with a point magnetic
dipole at nucleus K is conveniently written as
AmK (r) =
1 mK rK
.
3
c2
rK
(5.18)
The perturbed oneelectron Hamiltonian has the form as follows:

D (A) = mc2 + c(
D + c(
)
+ Vnuc = h
A).
h
With the vector potentials defined above, the Dirac Hamiltonian takes the form
X
D (B, {mK }) = h
D + B h
B +
m ,
h
mK h
K
(5.19)
(5.20)
B , appears as
where the Zeeman operator, h
B = 1 (rG c),
h
2
(5.21)
m , takes the form

and the hyperfine operator, h
K
m = 1 rK c .
h
K
3
c2
rK
(5.22)
The resulting perturbed Dirac Hamiltonian, Eq. 5.19, is linear in both, the magnetic field, B,
and the magnetic dipoles, mK . It contrasts to its nonrelativistic Schrodinger counterpart, con )
2 (or (
p)2 without the perturbation). Likewise in the relativistic
taining the term (
fourcomponent theory there is no explicit diamagnetic term, although such a contribution can
be obtained from the positive energynegative energy coupling terms of the linear response function as demonstrated in Ref. [421].
The formula in Eq. 5.15 together with the Hamiltonian carrying perturbations, Eq. 5.20, is
the prescription for the so called linear response method, Ref. [232]. It is used for calculating
a wide scale of second order molecular properties. Without going into more details, for NMR
shielding constants it is written in the short form of
DD
EE
m ; h
B
A = h
.
(5.23)
A
0
370
5.7.2
Gauge including atomic orbitals in the fourcomponent theory
The term of Eq. 5.21 contains the gauge origin reference, rG , which complicates calculations
of properties depending on external magnetic fields. This is removed by the introduction of
physically motivated London atomic orbitals (LAOs) [422], also called as gauge including atomic
orbitals (GIAOs), what ensures that these properties are independent of the choice of the gauge
origin [423]. At the fourcomponent level the gauge origin problem was solved by implementing
London atomic orbitals as was reported by one of us (MI) in Ref. [419]. The definition of the
magnetic field dependent LAO employed within the fourcomponent framework is
M,X (rM , B) = exp {iAB (RM ) r} M,X
(rM ); X = L, L, S, S ,
(5.24)
where the extended superscript M, X designates that M,X

(rM ) is an atomic scalar basis
function of type X centred at position RM , see also Section 4.4.6. Operating with the Dirac
operator on a London atomic orbital of any type L, L, S, or S, Eq. 5.24, gives
i
h
M
D + 1 B (rG c) M (B)
h(B)
h
(B) =
2
h
i
(5.25)
D + 1 B (rM c) M ,
= exp {iAB (RM ) r} h
2
showing that the reference to the gauge origin, rG , in the operator is replaced with the reference
to the centre of the orbital, rM . All formulas for computing NMR shielding constants therefore
became gauge origin free.
As an example of usefulness of London atomic orbitals, in Tab. 5.22 we collected a series of
computed 127 I iodine NMR shielding constants in six increasing basis sets (these are explained
in Tab. I of Ref. [419]). Note that the preferential approach is the unrestricted kinetic balance
(UKB) [424] in combination with LAO, where the convergence of NMR shielding constants is
smoother than in the common gauge origin (CGO) approach, where the gauge origin has to be
chosen arbitrarily. The UKB method generates small component basis function according to
Eq. 4.46 without any restrictions and the number of small component basis functions is approximately twice as much as the number of large component basis functions. On the contrary, in
order to make number of positive and negative states in the DHF solution equal, the restricted
kinetic balance (RKB) [128,225] is giving 1:1 ratio between the large and small component functions. This choice make the basis set magnetically unbalanced. To overcome this drawback the
so called restricted magnetic balance (RMB) approach was adopted by Komorovsky and coauthors [425], which leads to the introduction of perturbationdependent basis functions, what is
a different problem than the gaugeorigin dependence. Combination of both treatments, i.e. of
RMB and LAO, was reported recently by Cheng et al. [426] and very recently by Komorovsky et
al. [427]. Note that the relativistic LAOUKB approach significantly improves values of NMR
shielding constants with respect to nonrelativistic LAOLL values. In the best basis set VI, the
relativistic increase makes about 1368 ppm.
Another stream of fourcomponent relativistic property oriented methods can treat, for instance, electronic gtensors, as is reported by Repisky et al. [428].
5.7.3
Electron correlation in NMR calculations
The HartreeFock method is oneelectron approximation and NMR properties calculated from
the fourcomponent DHF wave function, Section 4.4.3, are missing an important contribution
371
of the electron correlation, which is essential to obtain results in agreement with experiment.
However, correlation methods for NMR properties based on the DHF wave function as reference
are not yet implemented, moreover, the calculations would be computationally intractable for
larger systems due to a prohibitive scaling.
In the wave function domain of quantum chemistry, there were formulated perturbation approaches for MCSCF correlation method to incorporate relativistic effects for NMR shielding
constants [429]. Comparison of perturbation relativistic contributions calculated at SCF and
MCSCF levels shows that the relativistic correction at the correlated MCSCF level differs from
the relativistic correction at the SCF level by very small values [429, 430]. Of course, one should
take into account that the MCSCF method insufficiently incorporate the dynamic part of correlation, but at least for weakly correlated systems containing lighter atoms, correlation effects
and relativistic contribution to the NMR shielding can be treated as additive. This fact opens
the possibility to incorporate relativistic effects for NMR shielding constants in an approximate
way, where relativistic corrections are estimated as difference between the DHF value and the
corresponding nonrelativistic HartreeFock value of the NMR shielding. This correction is then
Tab. 5.22. Isotropic 127 I iodine NMR shielding constants [ppm] in hydrogen iodide calculated with nonrelativistic (LL), RKBrelativistic, and UKBrelativistic methods; each both with common gauge origin
(CGO) on GO and with London atomic orbitals (LAO). The difference of each value to the corresponding
value in the largest basis set VI is given in parenthesis. Data and basis sets are from Ref. [419].
UKBCGO
GO=I
GO=(0,0,7)
UKBLAO
RKBCGO
RKBCGO a
GO=I
GO=(0,0,7)
GO=I
GO=(0,0,7)
RKBLAO
LLCGO a
LLLAO a
GO=I
II
III
IV
VI
5673.0
(-187.4)
6306.1
5905.2
(-2.0)
5808.9
(-51.5)
5839.5
5961.4
(+54.2)
5817.8
(-42.6)
6184.6
5951.6
(+44.4)
5852.0
(-8.4)
5889.4
5921.7
(+14.5)
5854.4
(-6.0)
5851.7
5960.3
(+53.1)
5860.4
4753.3
5388.8
6556.9
(-217.9)
7185.3
4975.3
4890.6
4920.4
6690.1
(-84.7)
6716.3
5027.3
5239.4
5608.9
6732.9
(-41.9)
7095.2
5367.1
5278.7
5317.7
6765.1
(-9.7)
6798.2
5343.9
5322.8
5320.2
6769.5
(-5.3)
6762.4
5422.1
4390.4
(-148.6)
4536.5
(-2.5)
4538.6
(-0.4)
4573.3
(+43.3)
4506.4
(-32.6)
4549.8
(+10.8)
4546.8
(+7.8)
4548.9
(+9.9)
4541.3
(+2.3)
4541.5
(+2.5)
5857.2
5907.2
5378.2
5375.4
6774.8
6767.3
5423.2
4539.0
4539.0
Diamagnetic term calculated as expectation value of the nonrelativistic diamagnetic operator.

Otherwise obtained from the -+ linear response function. For details see Ref. [419].
372
added to the nonrelativistic value of the NMR shielding obtained from a highlevel correlation
method like the coupled cluster method [431]. This approach was used for relativistic corrections
of NMR shielding constants for third row hydrides [432] and the results differ by fraction of ppm
from MCSCF perturbation corrections [430]. Although, spinspin coupling constants (defined
as the second derivative of the energy with respect to magnetic dipoles) correspond to principally different perturbations in the Hamiltonian, this approach is successfully applicable for these
NMR properties as well. The test was performed on spinspin coupling constants of the group
14 XH4 hydrides [433], where authors showed that adding relativistic corrections calculated using the described approximate method to vibrationally corrected nonrelativistic coupled cluster
values of spinspin coupling constants, J1 (XH) (X=C, Si, Ge), leads to an excellent agreement
with gasphase experimental results. The relativistic correction brought theoretical prediction of
the spinspin coupling constant in SnH4 closer to experimental value but the value was overestimated by about 10%. Although the approach based on additivity of correlation and relativistic
effects gives in some cases better results for NMR properties than relativistic DFT approaches
even for relatively heavy compounds (see, e.g., NMR properties of xenon fluorides [434]), the
usage is obviously limited to lighter systems. Moreover, the calculations are quite expensive due
to a slow convergence of the relativistic correction to the basis set limit.
Apart from the mentioned ab initio methods for computing NMR properties there is a large
group of methods dealing with the electronic correlation and effects of relativity within the two
component DFT approach, see Refs. [435440].
5.8
From relativistic effects in atoms and molecules to gasphase properties and the
condensed phase
Relativistic effects, as demonstrated in previous Sections, concern primarily properties of isolated atoms or molecules. However, real chemistry and physics proceeds in the gas phase, or,
mostly, in the condensed phase. Clearly, the understanding of basic trends in molecular structures, in the bonding character, bonding energies and atomic and molecular properties is unavoidable for understanding the processes in the gas phase, in solutions and in the solid state,
too.
An example of relativistic effects beyond isolated molecules is the investigation of sublimation enthalpies of metals in groups 10, 11, and 14 linked with different molecular properties by
Pershina et al. [441, 442]. Particularly interesting is clear correlation between binding energies
of homonuclear and heteronuclear MM0 diatomics of group 10, 11, and 14 elements with their
sublimation enthalpies. In previous Sections we have presented several examples of a typical
Vshaped pattern of some molecular properties in the series of elements down in the Periodic
Table. The Vshaped pattern characterizes binding energies of the group 10 elements (Ni, Pd,
and Pt) and coinage metal elements (Cu, Ag, Au, the group 11). Experimental sublimation enthalpies Hsub precisely copy the same pattern. Different picture exhibit the group 14 elements.
Their binding energies in the Ge, Sn, Pb, and the 114 element series smoothly diminish with
increasing atomic number and precisely the same trend is observed for experimental sublimation enthalpies. The behaviour of a series of binding energies of homonuclear and heteronuclear
diatomics and relation of these energies to Hsub is analysed using relativistic effects in participating valence orbitals and their relativistic stabilization (particularly s and p1/2 orbitals) or
destabilization (particularly (n-1)dorbitals in Hg and the element 112 containing molecules).
373
Comparison of relativistic and nonrelativistic properties of compounds in atomic properties and

volatility of the element 112 and Cd and Hg containing molecules and the Vshaped pattern of
their properties were treated in detail also in former papers of Pershina et al. [442, 443]. Relativistic effects also affect the adsorption of inert gases, Ne up to Xe, Rn and the element 118 on
noble metal and other surfaces, including quartz, ice, graphite and teflon [444].
A typical Vshaped trend exhibit transfer Gibbs energies of the coinage and alkali metal
cations in the combined theoretical and experimental (Secondary Ion Mass Spectrometry, SIMS)
research [445] on ion transfer properties in water and acetonitrile. The theoretical G0298K were
based on the binding energies of the coinage metal Cu+ , Ag+ , and Au+ cations with water
and acetonitrile obtained from CCSD(T) and DKHCCSD(T) calculations. Relativistic data do
nicely correlate with experimental transfer Gibbs energies while the nonrelativistic data are significantly off this trend. The highest binding energy was found for the Au+ interaction with
acetonitrile. This results from the dative sigmabonding accompanied by the charge transfer
from N to Au+ , like it was discussed in detail in Section 5.6 and a significant backdonation of
the electron density from the Au+ 5d to N 2p shell suggesting the existence of a multiple bond
between the nitrogen atom and the gold cation. Ilcin et al. [446] performed relativistic CCSD(T)
and perturbational calculations of interatomic and intermolecular interactions in complexes of
CdZn, HgZn, and HgCd, and used these data for the temperature dependence of the viscosity
for lowdensity binary mixtures of the group 12 elements. CCSD(T) spectroscopic constants of
Zn2 , Cd2 , and Hg2 calculated by the DC Hamiltonian give support for an acceptable agreement
of theoretical and experimental dynamic viscosities of the Zn, Cd, and Hg vapours [447]. Similar approach has been used for calculations of diffusion coefficients and stability of Hg N2
complexes [448].
Vogt et al. [449] synthesised and studied experimentally and theoretically the low and high
temperature modifications of Sc3 RuC4 and Sc3 OsC4 carbides. They analysed the local electronic
structure of the T(C2 )2 moieties and discussed an interplay between chemical bonding and relativistic effects which is responsible for the Vshaped pattern of the CC bond length in crystals
with T=Fe, Ru, and Os. Their interpretation includes the T (CC) back donation in line
with model M(C2 H2 ) benchmark systems (M=Ni, Pd, and Pt) studied by Ziegler et al. and
Eickerling et al. [450, 451].
Another new species, the icosahedral WAu12 nanoparticle is a stable multiply bonded system
representing a new structural principle in gold chemistry. Based on the idea of the aurophilicity and relativistically strengthened WAu and AuAu bonds its existence and properties was
predicted by Pyykko and Runeberg [452] in 2002. The existence of this species was experimentally confirmed in the same year [453]. Later on, Pyykko and coworkers [454] studied the
finitetemperature dynamics of this species.
At this occasion we wish to cite the title of the paper Gold goes nano from small clusters
to lowdimensional assemblies in which Schwerdtfeger [351] emphasised that bulk gold is a
noble metal but the atomic and nanocrystalline gold compounds can be very reactive and reach
applications in different areas. It was not until 1987 that the interest in gold nanostructures was
really awakened through the report of Haruta et al. [455] on CO oxidation on supported gold
catalysts. Since then nanocrystalline gold compounds have been the subject of intense research
and have found far reaching applications in areas as catalysis, sensors, molecular electronics or as
bioconjugate probes for amplification tags in gene analysis, antibody or antigen detection, DNA
sequencing, or gene mapping [351]. Catalytic properties of gold, its ability to create metalmetal
374
bonds, aurophilicity and other phenomena linked with relativistic effects in chemical bonds as
pioneered in groups of Pyykko, Schwarz, Schmidbaur, Schwerdtfeger,Schroder and other were
discussed in Section 5.5, see, e.g., Refs. [12, 13, 20, 350353]. The role of the gold nanofilms in
biomedical applications was analysed by Tokonami et al. [456].
Extremely interesting is discussion of the relation between molecular properties and the
solidstate symmetry. It may help in understanding of experimental observations and some
trends within a group of analogous compounds and their solid state structures, but in any case
it is not straightforward. In two papers dealing with the Cu, Ag, and Au halides Schwerdtfeger
et al. [457, 458] compared the structures calculated with and without relativistic effects. Along
with the bonding characteristics of molecules involved in solidstate structures they also calculated sublimation energies for different most stable crystal modifications and their low and
high temperature stabilities and their transitions. Some crystals, like the fluorides CuF, are experimentally unknown since they easily disproportionate at normal conditions. In general, the
coinage halides crystallize in the zinc blende, rock salt or in the chainlike structures. At the
nonrelativistic correlated level the cubic structures are favoured. Relativity stabilizes chainlike
structures particularly for the AuX crystals. This is linked with the relativistic increase of the covalency in AuX (X is F, Cl, Br, or I) molecules and disfavours in this way ionic structure types
for which the rock salt crystals are more typical. Different from the AuX compounds, the CuX
and AgX halides can be regarded as mainly ionic. How relativity reduces the polarity of MX
molecules is demonstrated in Fig. 5.8. Considerable reduction of the dipole moment is observed
particularly for AuF. The silver halides, which have highest dipole moments, all crystallize at
room temperatures and low pressures in the simple rock salt structure. The most interesting situation is found for the gold halides [458], for which the unusual chainlike structures with short
AuAu distances are favoured compared to the cubic structures. The main message is [457] that
relativistic effects drastically influence the solidstate symmetry.
It appears that similar analysis as was presented for the group 11 monohalides is also applicable to the group 12 solidstate oxides, ZnO, CdO, and HgO [459]. The solidstate HgO at
low pressure is analogous with the AuF structure. It is represented by the chainlike montroydite
and cinnabar structures in contrast with ZnO and CdO structures which have typically hexagonal
wurtzite or rock salt structures. This marked difference is caused by relativistic effects. There
is some analogy between the bonding situation in AuF and HgO. As it is presented in Tab. 5.19,
the bond strength of both molecules is significantly relativistically reduced, and, also, both molecules become less polar, i.e. their dipole moments diminish due to relativistic effects, see Fig. 5.8
and Refs. [300, 459]. Besides of changes in the solidstate structure, relativistic effects also reduce the cohesive energy of HgO. It appears that the unusual yellow to red colour of HgO is a
relativistic effect as well [459].
Exceptional position of Hg containing crystals in the group 12 dihalides is also attributed
to relativistic effects by Donald et al. [414]. Their analysis is based on DFT calculations of
monomers and dimers of MX2 molecules (M = Zn, Cd, Hg and X = F, Cl, Br, I). The zinc and
cadmium dihalide dimers have the usual D2h symmetry geometry, whereas the mercury dihalide
dimers are looselybound units with the C2h symmetry. Donald et al. found an obvious connection between the structures and characteristics of monomers, their dimers, and the crystals
they form. The crystal structures of the group 12 MX2 dihalides exhibit a wide range of structural types, from threedimensional extended solids to molecular crystals. Since the ionization
potential of Hg is high due to the relativistic stabilization of its valence 6s2 electrons, somewhat
375
surprising is the structure of HgF2 , which becomes an ionic solid. The crystals of HgCl2 and
HgBr2 are molecular solids which is connected with smaller electronegativity of Cl and Br so
that extended structures cannot be formed.
Calculations [460] of the interatomic interaction energies for the Hg dimer and small clusters
of the mercury atoms allowed to develop a manybody expansion of the correlation energy which
is added to the solid state HF energy [461]. The Hg dimer is more strongly bound than the Zn
and Cd dimers due to relativistic effects. The cluster studies were extended to the solid state and
optimized lattice parameters and cohesive energies for bulk mercury. For smaller clusters the
highly accurate CCSD(T) method was still applicable. DFT methods applicable to larger clusters
lead to diverse results for, e.g., the cohesive energies for bulk mercury. Accurate interaction
potentials including at least threebody effects are critically needed for reliable description of
the bulk properties by Monte Carlo simulations [462]. Accurate interaction potentials, based on
the fitting of CCSD(T) data, lead to excellent agreement of the bulk melting temperature of argon,
84.7 K, within less than 1 K of the experimental value. Such accuracy is not yet achievable for
metallic systems or molecular solids, but preliminary results results are at least very promising.
A cluster modelling of the interaction of the eka-Hg (element 112) and ekaPb atom (element
114) with the stable Au(111) surface using accurate smallcore relativistic pseudopotentials and
twocomponent noncollinear DFT was published by Zaitsevskii et al. [416, 463].
376
Conclusions
Real life of molecules is a relativistic quantum mechanical life. This holds for all atoms and
molecules throughout the Periodic Table. In light elements are relativistic effects visible in fine
structure effects and are growing dramatically for molecules containing elements with increasing
atomic number. We have demonstrated that relativistic effects alter many atomic and molecular
properties not only quantitatively but also qualitatively. Most important aspect of this paper is
understanding of trends within the Periodic Table. For specific groups of elements in the Table we analyse trends of ionization potentials, electron affinities, geometries, electric properties,
magnetic properties, interaction and reaction energies. When going to heavier atoms relativistic effects lead frequently to characteristic V shaped behaviour of properties with respect to
the atomic number. These relativistic changes are related to the electronic structure of atoms
and molecules. We also discuss briefly the link between relativistic changes in properties of
individual atoms or molecules and their behaviour in the condensed phase.
We are witnessing vigorous progress in the development of new relativistic theoretical approaches and methods, which allow treating of gradually larger manyelectron molecules at the
correlated relativistic level with a reasonable accuracy. Both wave function and density functional theories participate on the extension of applicability to a wide scale of problems which
need and can be treated by relativistic methods of quantum chemistry.
Acknowledgments
This research was supported by the Slovak Grant Agency, grant VEGA1/0520/10 and grants of
the Slovak Research and Development Agency APVV (Contracts No. LPP011007 and LPP
015509). We are grateful to Professor Maria Barysz, Dr. Lukas Bucinsky, Assoc. Professor
Pavel Neogrady, and Professor Jan Pisu t for careful reading the manuscript, for their valuable
suggestions and their kind help in correcting inaccuracies in the text. We also acknowledge the
effort of Professor Vladimr Buzek for his enthusiastic pushing for finishing this paper.
Conclusions
377
Appendix: Atomic units

Introducing atomic units, e = m0 = h
= 40 = 1, considerably simplifies all formulae of
quantum mechanics. In Tab. 0.1 we present SI values of atomic units.
Tab. 0.1. The conversion factors of atomic units to SI and other units based on the 2006 CODATA adjustment of the value of constants [45].
a.u. of
Symbol
Numerical value
Unit
charge: elementary charge

mass: electron mass
action: reduced Planck constant
length: Bohr radius (bohr)
e
me
h
a0
energy: Hartree energy (hartree)
Eh
time
force
velocity
momentum
current
charge density
electric potential
electric field
electric field gradient
electric dipole moment
/Eh
h
Eh /a0
a0 Eh /h
h/a0
eEh /h
e/a30
Eh /e
Eh /ea0
Eh /ea20
ea0
electric quadrupole moment

electric polarizability
1st hyperpolarizability
2nd hyperpolarizability
magnetic flux density
magnetic dipole moment
magnetizability
permitivity
ea20
e2 a20 /Eh
e3 a30 /Eh2
e4 a40 /Eh3
h/ea20
he/me
e2 a20 /me
e2 /a0 Eh
1.602 176 487(40)1019

9.109 382 15(45)1031
1.054 571 628(53)1034
0.529 177 208 59(36)1010
0.529 177 208 59(36)1010
4.359 743 94(22)1018
27.211 383 86(68)
2.194 746 313 705(15)105
2.625 499 617103
2.418 884 326 505(16)1017
8.238 722 06(41)108
2.187 691 2541(15)106
1.992 851 565(99)1024
6.623 617 63(17)103
1.081 202 300(27)1012
27.211 383 86(68)
5.142 206 32(13)1011
9.717 361 66(24)1021
8.478 352 81(21)1030
2.541 746 23
4.486 551 07(11)1040
1.648 777 2536(34)1041
3.206 361 533(81)1053
6.235 380 95(31)1065
2.350 517 382(59)105
1.854 801 830(46)1023
7.891 036 433(27)1029
1.112 650 056...1010
C
kg
Js
m
A
J
eV
cm1
kJ mol1
s
N
m s1
kg m s1
A
C m3
V
V m1
V m2
Cm
Debye
C m2
C2 m2 J1
C3 m3 J2
C4 m4 J3
T
J T1
J T2
F m1
378
References
[1] S. L. Glashow, B. Bova: Interactions : a journey through the mind of a particle physicist and the
matter of this world: Warner Books, New York, NY (1988)
[2] P. A. M. Dirac: Proc. Roy. Soc. Lond. A 123 (1929) 714
[3] W. Heitler, F. London: Z. Phys. 44 (1927) 455
[5] W. Koos, L. Wolniewicz: Acta Phys. Polon. 20 (1961) 129
[6] B. Swirles: Proc. Roy. Soc. Lond. A 152 (1935) 625
[7] I. P. Grant: Proc. Roy. Soc. Lond. A 262 (1961) 555
[8] J. P. Desclaux: At. Data Nucl. Data Tabes 12 (1973) 311
[9] P. Pyykko, J. P. Desclaux: Accounts Chem. Res. 12 (1979) 276
[10] P. Pyykko: Chem. Rev. 88 (1988) 563
[11] P. Pyykko: Chem. Rev. 97 (1997) 597
[12] P. Pyykko: Angew. Chem.-Int. Edit. 41 (2002) 3573
[13] P. Pyykko: Angew. Chem.-Int. Edit. 43 (2004) 4412
[14] W. Kutzelnigg: Phys. Rev. A 54 (1996) 1183
[15] J. Almlof, O. Gropen: In: Reviews in computational chemistry, eds. K. B. Lipkowitz, D. B. Boyd,
vol. 8, pp. 203244. Willey-VCH, New York (2001)
[16] P. Schwerdtfeger: In: Atoms, Molecules and Clusters in Electric fields: Theoretical Approaches to
the calculation of electric polarizability, ed. G. Maroulis, pp. 132. Imperial College Press, London
(2006)
[17] P. Schwerdtfeger, M. Seth: In: Encyclopedia of Computational Chemistry, ed. P. v. R. Schleyer, pp.
24802499. Wiley, Chichester (1998)
[18] B. A. Hess: In: Encyclopedia of Computational Chemistry, ed. P. v. R. Schleyer, pp. 24992508.
Wiley, Chichester (1998)
[19] W. H. E. Schwarz, A. Rutkowski, S. G. Wang: Int. J. Quantum Chem. 57 (1996) 641
[20] H. Schwarz: Angew. Chem.-Int. Edit. 42 (2003) 4442
[21] M. Urban, V. Kello: Adv. Quantum Chem. 50 (2005) 249
[22] M. Urban, V. Kello: Comput. Lett. 2 (2006) 259
[23] P. Schwerdtfeger (ed.): Relativistic electronic structure theory. Part 1. Fundamentals.: Theoretical
and computational chemistry v. 11. Elsevier, Amsterdam (2002)
[24] P. Schwerdtfeger (ed.): Relativistic electronic structure theory. Part 2. Applications.: Theoretical
and computational chemistry v. 14. Elsevier, Amsterdam (2004)
[25] K. G. Dyall, K. Fgri: Introduction to relativistic quantum chemistry: Oxford University Press,
New York (2007)
[26] M. Reiher, A. Wolf: Relativistic quantum chemistry : the fundamental theory of molecular science:
Wiley-VCH, Weinheim (2009)
[28] I. P. Grant, H. M. Quiney: Int. J. Quantum Chem. 80 (2000) 283
[29] K. Pachucki, J. Komasa: Phys. Rev. Lett. 92 (2004) 213001
[30] D. Mendeleev: Zhurnal Russkoe Fiziko-Khimicheskoe Obshchestvo 1 (1869) 66
[31] D. Mendelejeff: Zeitschrift fur Chemie 12 (1869) 405
References
379
[32] D. R. Lide (ed.): CRC handbook of chemistry and physics, CD-ROM Version 2005: CRC Press,
Boca Raton (2005)
[33] P. Neogrady, V. Kello, M. Urban, A. J. Sadlej: Int. J. Quantum Chem. 63 (1997) 557
[34] J. Gazo, J. Kohout, M. Serator, T. Sramko,

M. Zikmund: Vseobecna a anorganicka chemia: Alfa,
Bratislava (1974)
[35] M. Urban, J. Noga, S. J. Cole, R. J. Bartlett: J. Chem. Phys. 83 (1985) 4041
[36] K. Raghavachari, G. W. Trucks, J. A. Pople, M. HeadGordon: Chem. Phys. Lett. 157 (1989) 479
[37] M. Urban, I. Cernu

sa k, J. Noga, V. Kello: In: Methods in Computational Chemistry, ed. S. Wilson,
vol. 1, pp. 117250. Plenum Press, New York (1987)
[38] J. Paldus, X. Z. Li: Adv. Quantum Chem. 110 (1999) 1
[39] T. Helgaker, P. Jrgensen, J. Olsen: Molecular electronic-structure theory: Wiley, Chichester ; New
York (2000)
[40] I. Shavitt, R. J. Bartlett: Many-body methods in chemistry and physics : MBPT and coupled-cluster
theory: Cambridge molecular science. Cambridge University Press, Cambridge ; New York (2009)
[41] J. P. Desclaux, Y. K. Kim: J. Phys. B-At. Mol. Opt. Phys. 8 (1975) 1177
[42] N. Bohr: Philos. Mag. 26 (1913) 1
[43] A. Einstein: Ann. Phys.-Berlin 17 (1905) 891
[44] J. C. Maxwell: Philos. Mag. 21 (1861) 161
[45] P. J. Mohr, B. N. Taylor, D. B. Newell: Rev. Mod. Phys. 80 (2008) 633
[46] J. F. Harrison: Spinorbit interaction (2006): http://www.cem.msu.edu/cem987/spin orbit.pdf
[47] E. Schrodinger: Phys. Rev. 28 (1926) 1049
[48] W. Gerlach, O. Stern: Z. Phys. 9 (1922) 353
[49] G. E. Uhlenbeck, S. Goudsmit: Nature 117 (1926) 264
[50] J. Pisu t, R. Zajac: O atomoch a kvantovan: FMFI UK, Bratislava (2009)
[51] L. Piela: Ideas of quantum chemistry: Elsevier, Amsterdam ; Boston (2007)
[52] G. E. Uhlenbeck, S. Goudsmit: Naturw. 13 (1925) 953
[53] W. Pauli: Z. Phys. 43 (1927) 601: See footnote 2 on page 607
[54] M. Born, R. Oppenheimer: Ann. Phys.-Berlin 84 (1927) 0457
[55] J. C. Slater: Phys. Rev. 34 (1929) 1293
[56] V. Fock: Z. Phys. 61 (1930) 126
[57] D. R. Hartree, W. Hartree: Proc. Roy. Soc. Lond. A 150 (1935) 9
[58] D. R. Hartree: Nature 138 (1936) 1080
[59] C. C. J. Roothaan: Rev. Mod. Phys. 32 (1960) 179
[60] J. A. Pople, R. K. Nesbet: J. Chem. Phys. 22 (1954) 571
[61] S. F. Boys: Proc. R. Soc. London Ser. A 200 (1950) 542
arsky, M. Urban: Ab initio calculations : methods and applications in chemistry: Lecture notes
[62] P. C
in chemistry 16. Springer-Verlag, Berlin ; New York (1980)
[63] C. C. J. Roothaan: Rev. Mod. Phys. 23 (1951) 69
Malmqvist, P. Neogrady, T. B. Pedersen,
[64] F. Aquilante, L. De Vico, N. Ferre, G. Ghigo, P.-A.
M. Pitona k, M. Reiher, B. O. Roos, L. Serrano-Andres, M. Urban, V. Veryazov, R. Lindh: J. Comput.
Chem. 31 (2010) 224
[65] M. Urban, P. Neogrady, I. Hubac: In: Recent Advances in Computational Chemistry, ed. R. J.
Bartlett, vol. 3, pp. 275306. World Scientific, Singapore (1997)
380
[66]
[67]
[68]
[69]
[70]
[71]
[72]
[73]
[74]
[75]
[76]
[77]
[78]
[79]
[80]
[81]
[82]
[83]
[84]
[85]
[86]
[87]
[88]
[89]
[90]
[91]
[92]
[93]
[94]
[95]
[96]
[97]
[98]
[99]
[100]
J. D. Watts, J. Gauss, R. J. Bartlett: J. Chem. Phys. 98 (1993) 8718

P.-O. Lowdin: Phys. Rev. 97 (1955) 1509
C. D. Sherrill, H. F. Schaefer: Adv. Quantum Chem. 34 (1999) 143
B. Roos: Chem. Phys. Lett. 15 (1972) 153
M. Barbatti, H. Lischka: J. Phys. Chem. A 111 (2007) 2852
T. D. Crawford, H. F. Schaefer: In: Reviews in Computational Chemistry, eds. K. B. Lipkowitz,
D. B. Boyd, pp. 33136. Wiley-VCH, Inc., Hoboken (2007)
M. Gell-Mann, K. A. Brueckner: Phys. Rev. 106 (1957) 364
H. A. Bethe: Phys. Rev. 103 (1956) 1353
F. Coester: Nucl. Phys. 7 (1958) 421
F. Coester, H. Kummel: Nucl. Phys. 17 (1960) 477
z ek: J. Chem. Phys. 45 (1966) 4256
J. C
z ek: Adv. Chem. Phys. 14 (1969) 35
J. C
z ek: Phys. Rev. A 5 (1972) 50
J. Paldus, I. Shavitt, J. C
R. J. Bartlett: J. Phys. Chem. 93 (1989) 1697
z ek: Adv. Quantum Chem. 9 (1975) 105
J. Paldus, J. C
J. Paldus: In: Methods in Computational Molecular Physics, eds. S. Wilson, G. H. F. Diercksen, vol.
293 of NATO Advanced Science Institutes Series, Series B, Physics, pp. 99194. Plenum Press, New
York (1992)
S. A. Kucharski, R. J. Bartlett: Theor. Chim. Acta 80 (1991) 387
I. Lindgren, S. Salomonson, D. Hedendahl: In: Recent progress in coupled cluster methods: Theory
arsky, J. Paldus, J. Pittner, pp. 357374. Springer, Berlin (2010)
and applications, eds. P. C
E. Eliav, U. Kaldor: In: Recent progress in coupled cluster methods: Theory and applications, eds.
arsky, J. Paldus, J. Pittner, pp. 113144. Springer, Berlin (2010)
P. C
arsky, J. Paldus, J. Pittner (eds.): Recent progress in coupled cluster methods: Theory and
P. C
applications: Springer, Berlin (2010)
G. D. Purvis, R. J. Bartlett: J. Chem. Phys. 76 (1982) 1910
J. Noga, R. J. Bartlett: J. Chem. Phys. 86 (1987) 7041
R. J. Bartlett, M. Musial: Rev. Mod. Phys. 79 (2007) 291
M. Kallay, P. R. Surjan: J. Chem. Phys. 115 (2001) 2945
S. A. Kucharski, R. J. Bartlett: J. Chem. Phys. 108 (1998) 5243
M. Musial, S. A. Kucharski, R. J. Bartlett: J. Chem. Phys. 116 (2002) 4382
X. Z. Li, J. Paldus: J. Chem. Phys. 101 (1994) 8812
P. Neogrady, M. Urban: Int. J. Quantum Chem. 55 (1995) 187
M. Douglas, N. M. Kroll: Ann. Phys. (N.Y.) 82 (1974) 89
B. A. Hess: Phys. Rev. A 32 (1985) 756
J. Noga, W. Kutzelnigg: J. Chem. Phys. 101 (1994) 7738
X. Z. Li, J. Paldus: J. Chem. Phys. 119 (2003) 5320
J. Pittner, P. Piecuch: Mol. Phys. 107 (2009) 1209
arsky, I. Hubac: J. Phys. Chem. A 108 (2004) 3125
O. Demel, J. Pittner, P. C
Malmqvist, L. Serrano-Andres, K. Pierloot,
B. O. Roos, K. Andersson, M. P. Fulscher, P.-A.
M. Merchan: Adv. Chem. Phys. 93 (1996) 219
References
381
Malmqvist, B. O. Roos, A. J. Sadlej, K. Wolinski: J. Phys. Chem. 94 (1990)

[101] K. Andersson, P.-A.
5483
Malmqvist: Phys. Chem. Chem. Phys. 6 (2004) 2919
[102] B. O. Roos, P.-A.
[103] P.-A. Malmqvist, B. O. Roos, B. Schimmelpfennig: Chem. Phys. Lett. 357 (2002) 230
[104] Z. Azizi, B. O. Roos, V. Veryazov: Phys. Chem. Chem. Phys. 8 (2006) 2727
Malmqvist, V. Veryazov, P. O. Widmark: J. Phys. Chem. A 108 (2004)
[105] B. O. Roos, R. Lindh, P.-A.
2851
Malmqvist, V. Veryazov, P. O. Widmark: J. Phys. Chem. A 109 (2005)
[106] B. O. Roos, R. Lindh, P.-A.
6575
[107] H. A. Lorentz: Proc. Acad. Science Amsterdam 1 (1899) 427
[108] H. Poincare: Comptes rendus de l Academie des Sciences de Paris 140 (1905) 1504
[109] A. Einstein: Ann. Phys.-Berlin 18 (1905) 639
[110] W. Gordon: Z. Phys. 40 (1926) 117
[111] O. Klein: Z. Phys. 41 (1927) 407
[112] W. Pauli, V. Weisskopf: Helv. Phys. Acta 7 (1934) 709
[113] P. A. M. Dirac: In: History of twentieth century physics, ed. C. Weiner, pp. 109146. Academic
Press, New York (1977)
[114] C. D. Anderson: Phys. Rev. 41 (1932) 405
[115] C. G. Darwin: Proc. Roy. Soc. Lond. A 118 (1928) 654
[116] H. A. Bethe, E. E. Salpeter: Quantum mechanics of one-and two-electron atoms: Plenum Press,
New York (1977)
[117] R. E. Moss: Advanced molecular quantum mechanics; an introduction to relativistic quantum mechanics and the quantum theory of radiation: Chapman and Hall, London (1973)
[118] M. E. Rose: Relativistic electron theory: Wiley, New York (1961)
[119] A. Messiah: Quantum mechanics: North-Holland, Amsterdam (1961)
[120] H. E. White: Phys. Rev. 38 (1931) 514
[121] A. Szabo: J. Chem. Educ. 46 (1969) 678
[122] C. W. Sherwin: Introduction to quantum mechanics: Holt, Rinehart and Winston, New York (1959)
[123] I. P. Grant, H. M. Quiney: Adv. At. Mol. Phys. 23 (1988) 37
[124] Y. Ishikawa, H. M. Quiney: Int. J. Quantum Chem. Suppl. 21 (1987) 523
[125] O. Visser, P. J. C. Aerts, D. Hegarty, W. C. Nieuwpoort: Chem. Phys. Lett. 134 (1987) 34
[126] L. Visscher, K. G. Dyall:
At. Data Nucl. Data Tables 67 (1997) 207:
see also
http://www.chem.vu.nl/visscher/FiniteNuclei/FiniteNuclei.htm
[127] M. Gell-Mann: Nuovo Cimento Suppl. 4 (1956) 848
[128] T. Saue, L. Visscher: In: Theoretical chemistry and physics of heavy and superheavy elements,
eds. U. Kaldor, S. Wilson, vol. 11 of Progress in theoretical chemistry and physics, p. 211. Kluwer,
Dordrecht (2003)
[129] P. A. M. Dirac: In: Structure et Propertietes des Noyaux Atomiques. Septieme Conseil du Physique
Solvay, Bruxelles, 2229 October 1933, eds. J. Cockcroft, J. Chadwick, F. Joliot, J. Joliot, N. Bohr,
G. Gamov, P. A. M. Dirac, W. Heisenberg, pp. 203230. Gautier-Villars, Paris (1934)
[130] G. Breit: Physical Review 34 (1929) 553
[131] J. A. Gaunt: Proc. Roy. Soc. Lond. A 122 (1929) 513
[132] P. M. Kozlowski, L. Adamowicz: Chem. Rev. 93 (1993) 2007
382
[133] G. E. Brown, D. G. Ravenhall: Proc. Roy. Soc. Lond. A 208 (1951) 552
[134] M. H. Mittleman: Phys. Rev. A 24 (1981) 1167
[135] H. M. Quiney: Private communication; the information is provided in a letter from Lady Jeffreys
(born Swirles) to H.M.Q., 3 March 1997 ()
[136] G. Malli, J. Oreg: J. Chem. Phys. 63 (1975) 830
[137] R. E. Stanton, S. Havriliak: J. Chem. Phys. 81 (1984) 1910
[138] P. J. C. Aerts, W. C. Nieuwpoort: Chem. Phys. Lett. 113 (1985) 165
[139] K. G. Dyall, I. P. Grant, S. Wilson: J. Phys. B-At. Mol. Opt. Phys. 17 (1984) 493
[140] L. Visscher, P. J. C. Aerts, O. Visser: In: The effects of relativity in atoms, molecules, and the solid
state, eds. S. Wilson, I. P. Grant, B. L. Gyorffy, p. 197. Plenum Press, New York (1991)
[141] J. Thyssen: Develompment and applications of methods for correlated relativistic calculations of
molecular properties: Phd. thesis, University of Southern Denmark, Odense (2001)
[142] H. J. Aa. Jensen, K. G. Dyall, T. Saue, K. Fgri: J. Chem. Phys. 104 (1996) 4083
[143] J. Thyssen, T. Fleig, H. J. Aa. Jensen: J. Chem. Phys. 129 (2008) 034109
[144] T. Fleig, C. M. Marian, J. Olsen: Theor. Chem. Acc. 97 (1997) 125
[145] H. A. Kramers: Proc. Acad. Sci. Amsterdam 33 (1930) 959
[146] P. Hafner: J. Phys. B-At. Mol. Opt. Phys. 13 (1980) 3297
[147] H. A. Bethe: Annal. Phys. 3 (1929) 133208
[148] S. L. Altmann: Rotations, quaternions, and double groups: Clarendon Press, Oxford (1986)
[149] C. J. Bradley, A. P. Cracknell: The mathematical theory of symmetry in solids; representation theory
for point groups and space groups: Clarendon Press, Oxford, (1972)
[150] P. J. C. Aerts: Chem. Phys. Lett. 104 (1984) 28
[151] A. Rosen: Int. J. Quantum Chem. 13 (1978) 509
[152] J. Meyer: Int. J. Quantum Chem. 33 (1988) 445
[153] J. Meyer, W. D. Sepp, B. Fricke, A. Rosen: Comput. Phys. Commun. 54 (1989) 55
[154] L. Visscher: Chem. Phys. Lett. 253 (1996) 20
[155] T. Saue, H. J. Aa. Jensen: J. Chem. Phys. 111 (1999) 6211
[156] K. Rykhlinskaya, S. Fritzsche: Comput. Phys. Commun. 171 (2005) 119
[157] T. Yanai, R. J. Harrison, T. Nakajima, Y. Ishikawa, K. Hirao: Int. J. Quantum Chem. 107 (2007)
1382
[158] D. L. Peng, J. Y. Ma, W. J. Liu: Int. J. Quantum Chem. 109 (2009) 2149
[159] B. A. Hess, C. M. Marian: In: Computational molecular spectroscopy, eds. P. Jensen, P. R. Bunker,
p. 169. Wiley, Chichester ; New York (2000)
[160] K. Balasubramanian: Relativistic effects in chemistry. Part A: Theory and techniques: Wiley Interscience, New York (1997)
[161] K. G. Dyall: Theor. Chem. Acc. 115 (2006) 441
[162] K. G. Dyall: Theor. Chem. Acc. 117 (2007) 483
[163] K. G. Dyall: J. Phys. Chem. A 113 (2009) 12638
[164] K. G. Dyall, A. S. P. Gomes: Theor. Chem. Acc. 125 (2010) 97
[165] A. Gomes, K. Dyall, L. Visscher: Theor. Chem. Acc. (2010): DOI 10.1007/s00214-009-0725-7
[166] L. Visscher: Theor. Chem. Acc. 98 (1997) 68
[167] M. Abe, T. Yanai, T. Nakajima, K. Hirao: Chem. Phys. Lett. 388 (2004) 68
[168] C. Mller, M. Plesset: Phys. Rev. 46 (1934) 618
References
383
[169] J. K. Laerdahl, T. Saue, K. Fgri: Theor. Chem. Acc. 97 (1997) 177

[170] DIRAC: a relativistic ab initio electronic structure program, release dirac08 (2008): written by
L. Visscher, H. J. Aa. Jensen, and T. Saue, with new contributions from R. Bast, S. Dubillard,
K. G. Dyall, U. Ekstrom, E. Eliav, T. Fleig, A. S. P. Gomes, T. U. Helgaker, J. Henriksson, M. Ilias,
Ch. R. Jacob, S. Knecht, P. Norman, J. Olsen, M. Pernpointner, K. Ruud, P. Saek, and J. Sikkema,
see http://dirac.chem.sdu.dk
[171] M. Abe, T. Nakajima, K. Hirao: J. Chem. Phys. 125 (2006) 234110
[172] L. Visscher, T. J. Lee, K. G. Dyall: J. Chem. Phys. 105 (1996) 8769
[173] M. Pernpointner, L. Visscher: J. Comput. Chem. 24 (2003) 754
[174] L. Visscher, K. G. Dyall, T. J. Lee: Int. J. Quantum Chem. Suppl. 29 (1995) 411
[175] A. Landau, E. Eliav, Y. Ishikawa, U. Kaldor: J. Chem. Phys. 121 (2004) 6634
[176] L. Visscher, E. Eliav, U. Kaldor: J. Chem. Phys. 115 (2001) 9720
[177] A. Landau, E. Eliav, U. Kaldor: Chem. Phys. Lett. 313 (1999) 399
[178] A. Landau, E. Eliav, U. Kaldor: Int. J. Mod. Phys. B 17 (2003) 5335
[179] E. Eliav, M. J. Vilkas, Y. Ishikawa, U. Kaldor: J. Chem. Phys. 122 (2005) 224113
[180] T. Fleig, J. Olsen, C. M. Marian: J. Chem. Phys. 114 (2001) 4775
[181] T. Fleig, J. Olsen, L. Visscher: J. Chem. Phys. 119 (2003) 2963
[182] T. Fleig, H. J. Aa. Jensen, J. Olsen, L. Visscher: J. Chem. Phys. 124 (2006) 104106
[183] S. Knecht, H. J. Aa. Jensen, T. Fleig: J. Chem. Phys. 132 (2010) 014108
[184] T. Fleig: Phys. Rev. A 77 (2008) 062503
[185] Y. S. Lee: In: Relativistic electronic structure theory. Part 2. Applications., ed. P. Schwerdtfeger,
Theoretical and computational chemistry v. 14, pp. 352416. Elsevier, Amsterdam (2004)
[186] K. A. Peterson, D. Figgen, E. Goll, H. Stoll, M. Dolg: J. Chem. Phys. 119 (2003) 11113
[187] D. Figgen, K. A. Peterson, M. Dolg, H. Stoll: J. Chem. Phys. 130 (2009) 164108
[188] H. Stoll: Chem. Phys. Lett. 429 (2006) 289
[189] D. Figgen, A. Wedig, H. Stoll, M. Dolg, E. Eliav, U. Kaldor: J. Chem. Phys. 128 (2008) 024106
[190] L. L. Foldy, S. A. Wouthuysen: Phys. Rev. 78 (1950) 29
[191] B. A. Hess, C. M. Marian, S. D. Peyerimhoff: In: Modern electronic structure theory, Part I, ed. D. R.
Yarkony, Advanced series in physical chemistry, Vol. 2, pp. 152278. World Scientific, Singapore
(1995)
[192] B. A. Hess: Phys. Rev. A 33 (1986) 3742
[193] T. Nakajima, K. Hirao: J. Chem. Phys. 113 (2000) 7786
[194] T. Nakajima, K. Hirao: J. Chem. Phys. 119 (2003) 4105
[195] C. Park, J. E. Almlof: Chem. Phys. Lett. 231 (1994) 269
[196] C. van Wullen: J. Chem. Phys. 120 (2004) 7307
[197] M. Reiher, A. Wolf: J. Chem. Phys. 121 (2004) 2037
[198] M. Reiher, A. Wolf: J. Chem. Phys. 121 (2004) 10945
[199] J. L. Heully, I. Lindgren, E. Lindroth, S. Lundqvist, A. M. Martensson-Pendrill: J. Phys. B-At. Mol.
Opt. Phys. 19 (1986) 2799
[200] W. Kutzelnigg: J. Chem. Phys. 110 (1999) 8283
[201] W. Kutzelnigg, W. J. Liu: J. Chem. Phys. 123 (2005) 241102
[202] E. van Lenthe, E. J. Baerends, J. G. Snijders: J. Chem. Phys. 105 (1996) 2373
[203] K. G. Dyall: J. Chem. Phys. 106 (1997) 9618
384
[204] K. G. Dyall: In: Theoretical chemistry and physics of heavy and superheavy elements, eds.
U. Kaldor, S. Wilson, p. 325. Kluwer, Dordrecht (2003)
[205] C. Chang, M. Pelissier, P. Durand: Phys. Scr. 34 (1986) 394
[206] E. van Lenthe, E. J. Baerends, J. G. Snijders: J. Chem. Phys. 101 (1994) 9783
[207] E. van Lenthe, J. G. Snijders, E. J. Baerends: J. Chem. Phys. 105 (1996) 6505
[208] K. G. Dyall, E. van Lenthe: J. Chem. Phys. 111 (1999) 1366
[209] R. van Leeuwen, E. van Lenthe, E. J. Baerends, J. G. Snijders: J. Chem. Phys. 101 (1994) 1272
[210] Y. Nakatsuka, T. Nakajima, M. Nakata, K. Hirao: J. Chem. Phys. 132 (2010) 054102
[211] M. Barysz, A. J. Sadlej, J. G. Snijders: Int. J. Quantum Chem. 65 (1997) 225
[212] M. Barysz, A. J. Sadlej: Theochem-J. Mol. Struct. 573 (2001) 181
[213] M. Barysz: J. Chem. Phys. 114 (2001) 9315
[214] M. Barysz, A. J. Sadlej: J. Chem. Phys. 116 (2002) 2696
[215] M. Barysz: In: Theoretical Chemistry and Physics of Heavy and Super-Heavy Elements., eds.
U. Kaldor, S. Wilson, vol. 11 of Progress in Theoretical Chemistry and Physics, pp. 349397. Kluwer
Academic Publishers, Dordrecht (2003)
[216] M. Barysz: In: Relativistic methods for chemists, eds. M. Barysz, Y. Ishikawa, vol. 10 of Challenges
and advances in computational chemistry and physics, pp. 165190. Springer, Dordrecht (2010)
[217] R. D. Cowan, D. C. Griffin: J. Opt. Soc. Am. 66 (1976) 1010
[218] J. H. Wood, A. M. Boring: Phys. Rev. B 18 (1978) 2701
[219] S. Komorovsky, M. Repisky, O. L. Malkina, V. G. Malkin, I. Malkin, M. Kaupp: J. Chem. Phys.
124 (2006) 084108
[220] V. Kello, A. J. Sadlej: Int. J. Quantum Chem. 68 (1998) 159
[221] M. Barysz: Pol. J. Chem. 74 (2000) 1329
[222] D. Kedziera, M. Barysz, A. J. Sadlej: Struct. Chem. 15 (2004) 369
[223] J. Seino, W. Uesugi, M. Hada: J. Chem. Phys. 132 (2010) 164108
[224] M. Ilias, T. Saue: J. Chem. Phys. 126 (2007) 064102
[225] L. Visscher, T. Saue: J. Chem. Phys. 113 (2000) 3996
[226] K. G. Dyall: J. Chem. Phys. 100 (1994) 2118
[227] H. J. Aa. Jensen, M. Ilias in preparation
[228] W. J. Liu, D. L. Peng: J. Chem. Phys. 125 (2006) 044102
[229] D. Kedziera, M. Barysz: Chem. Phys. Lett. 446 (2007) 176
[230] M. Barysz, L. Mentel, J. Leszczynski: J. Chem. Phys. 130 (2009) 164114
[231] C. E. Moore: Ionization Potentials and Ionization Limits Derived from the Analyses of Optical
Spectra: Natl. Stand. Ref. Data Ser. U.S. Natl. Bur. Stand. (1970)
[232] T. Saue, H. J. Aa. Jensen: J. Chem. Phys. 118 (2003) 522
[233] C. M. Marian: In: Reviews in computational chemistry, eds. K. B. Lipkowitz, D. B. Boyd, vol. 17,
pp. 99204. Willey-VCH, New York (2001)
[234] B. A. Hess, C. M. Marian, U. Wahlgren, O. Gropen: Chem. Phys. Lett. 251 (1996) 365
[235] M. Ilias, V. Kello, L. Visscher, B. Schimmelpfennig: J. Chem. Phys. 115 (2001) 9667
[236] L. Visscher, O. Visser, P. J. C. Aerts, H. Merenga, W. C. Nieuwpoort: Comput. Phys. Commun. 81
(1994) 120
[237] H. N. Russel, F. A. Saunders: Astrophys. J. 61 (1925) 38
[238] W. E. Lamb, R. C. Retherford: Phys. Rev. 72 (1947) 241
References
[239]
[240]
[241]
[242]
[243]
[244]
[245]
[246]
[247]
[248]
[249]
[250]
[251]
[252]
[253]
[254]
[255]
[256]
[257]
[258]
[259]
[260]
[261]
[262]
[263]
[264]
[265]
[266]
[267]
[268]
[269]
[270]
[271]
[272]
[273]
[274]
[275]
[276]
385
H. A. Bethe: Phys. Rev. 72 (1947) 339

J. P. Briand, P. Chevallier, P. Indelicato, K. P. Ziock, D. D. Dietrich: Phys. Rev. Lett. 65 (1990) 2761
I. Goidenko, L. Labzowsky, E. Eliav, U. Kaldor, P. Pyykko: Phys. Rev. A 67 (2003) 020102
K. G. Dyall, C. W. Bauschlicher, D. W. Schwenke, P. Pyykko: Chem. Phys. Lett. 348 (2001) 497
M. J. Vilkas, Y. Ishikawa, K. Hirao: Chem. Phys. Lett. 321 (2000) 243
T. H. Dunning, K. A. Peterson, D. E. Woon: In: Encyclopedia of Computational Chemistry, ed.
P. v. R. Schleyer, pp. 88115. Wiley, Chichester (1998)
A. J. Sadlej: Collect. Czech. Chem. Commun. 53 (1988) 1995
A. J. Sadlej: Theor. Chim. Acta 79 (1991) 123
A. J. Sadlej, M. Urban: Theochem-J. Mol. Struct. 234 (1991) 147
A. J. Sadlej, M. Urban, O. Gropen: Phys. Rev. A 44 (1991) 5547
V. Kello, A. J. Sadlej: Theor. Chim. Acta 83 (1992) 351
V. Kello, A. J. Sadlej, K. Fgri: Phys. Rev. A 47 (1993) 1715
P. Neogrady, V. Kello, M. Urban, A. J. Sadlej: Theor. Chim. Acta 93 (1996) 101
M. Urban, A. J. Sadlej: Mol. Phys. 92 (1997) 587
I. Cernu
sa k, V. Kello, A. J. Sadlej: Collect. Czech. Chem. Commun. 68 (2003) 211
V. Kello, A. J. Sadlej, B. A. Hess: J. Chem. Phys. 105 (1996) 1995
I. Miadokova, V. Kello, A. J. Sadlej: Theor. Chem. Acc. 96 (1997) 166
V. Kello, A. J. Sadlej, K. Fgri: J. Chem. Phys. 108 (1998) 2056
I. Miadokova, V. Kello, A. J. Sadlej: Mol. Phys. 96 (1999) 179
Z. Benkova, A. J. Sadlej: Mol. Phys. 102 (2004) 687
K. Fgri, K. G. Dyall: In: Relativistic electronic structure theory, Part 1. Fundamentals, ed.
P. Schwerdtfeger, Theoretical and computational chemistry. Vol. 11, pp. 259290. Elsevier, Amsterdam (2002)
N. B. Balabanov, K. A. Peterson: J. Chem. Phys. 123 (2005) 064107
K. A. Peterson, C. Puzzarini: Theor. Chem. Acc. 114 (2005) 283
P. Hohenberg, W. Kohn: Phys. Rev. 136 (1964) B864
W. Kohn, L. J. Sham: Phys. Rev. 140 (1965) A1133
A. K. Rajagopal, J. Callaway: Phys. Rev. B 7 (1973) 1912
A. K. Rajagopal: J. Phys. C 11 (1978) L943
E. Engel, S. Keller, A. F. Bonetti, H. Muller, R. M. Dreizler: Phys. Rev. A 52 (1995) 2750
E. K. U. Gross, R. M. Dreizler (eds.): Density functional theory: Plenum Press, New York (1995)
C. J. Cramer: Essentials of computational chemistry: theories and models: Wiley, Chichester, 2nd
edn. (2004)
T. Saue, T. Helgaker: J. Comput. Chem. 23 (2002) 814
O. Fossgaard, O. Gropen, M. C. Valero, T. Saue: J. Chem. Phys. 118 (2003) 10418
P. Salek, T. Helgaker, T. Saue: Chem. Phys. 311 (2005) 187
N. Gaston, P. Schwerdtfeger, T. Saue, J. Greif: J. Chem. Phys. 124 (2006) 044304
386
[277] I. S. Lim, M. Pernpointner, M. Seth, J. K. Laerdahl, P. Schwerdtfeger, P. Neogrady, M. Urban: Phys.

Rev. A 60 (1999) 2822
[278] I. S. Lim, P. Schwerdtfeger, B. Metz, H. Stoll: J. Chem. Phys. 122 (2005) 104103
[279] E. Eliav, U. Kaldor, Y. Ishikawa: Phys. Rev. A 50 (1994) 1121
[280] W. C. Martin, A. Musgrove, S. Kotochigova, J. E. Sansonetti: Ground Levels and Ionization Energies for the Neutral Atoms: National Institute of Standards and Technology, Gaithersburg (2003):
http://physics.nist.gov/IonEnergy
[281] I. S. Lim, P. Schwerdtfeger: Phys. Rev. A 70 (2004) 062501
[282] F. E. Jorge, A. C. Neto, G. G. Camiletti, S. F. Machado: J. Chem. Phys. 130 (2009) 064108
[283] M. Ilias, P. Neogrady: Chem. Phys. Lett. 309 (1999) 441
I. Cernu
[284] P. Neogrady, M. Medved,

sa k, M. Urban: Mol. Phys. 100 (2002) 541
[285] A. Borschevsky, V. Pershina, E. Eliav, U. Kaldor: Chem. Phys. Lett. 480 (2009) 49
[286] M. Ilias, H. J. Aa. Jensen, V. Kello, B. O. Roos, M. Urban: Chem. Phys. Lett. 408 (2005) 210
[287] M. L. Polak, M. K. Gilles, R. F. Gunion, W. C. Lineberger: Chem. Phys. Lett. 210 (1993) 55
[288] L. Demovic, V. Kello, M. Urban to be published
[289] Y. Ralchenko, A. E. Kramida, J. Reader: NIST Atomic Spectra Database (version 3.1.5): National
Institute of Standards and Technology, Gaithersburg, MD (2008): http://physics.nist.gov/asd3
[290] F. Wang, T. Ziegler, E. van Lenthe, S. van Gisbergen, E. J. Baerends: J. Chem. Phys. 122 (2005)
204103
[291] S. Yamamoto, H. Tatewaki: J. Chem. Phys. 132 (2010) 054303
[292] M. Urban, J. D. Watts, R. J. Bartlett: Int. J. Quantum Chem. 52 (1994) 211
[293] R. Pou-Amerigo, M. Merchan, I. Nebotgil, P. O. Widmark, B. O. Roos: Theor. Chim. Acta 92 (1995)
149
[294] L. Demovic, I. Cernu

sa k, G. Theodorakopoulos, I. D. Petsalakis, M. Urban: Chem. Phys. Lett. 447
(2007) 215
[295] V. Kello, A. Antusek, M. Urban: J. Comput. Methods Sci. Eng. 4 (2004) 753
[296] K. Huber, G. Herzberg: Molecular Spectra and Molecular Structure IV. Constants of Diatomic
Molecules: Van Nostrand Reinhold Co., New York (1979)
Malmqvist: Adv. Quantum. Chem. 47 (2004) 37
[297] B. O. Roos, P.-A.
[298] R. R. Reddy, Y. N. Ahammed, K. R. Gopal, P. A. Azeem, B. S. Devi, T. V. R. Rao, S. H. Behere:
Opt. Mater. 22 (2003) 7
[299] T. Fleig, A. J. Sadlej: Phys. Rev. A 65 (2002) 032506
[300] M. Ilias, P. Furdk, M. Urban: J. Phys. Chem. A 102 (1998) 5263
[301] A. Gijsbertsen, W. Siu, M. F. Kling, P. Johnsson, P. Jansen, S. Stolte, M. J. J. Vrakking: Phys. Rev.
Lett. 99 (2007) 213003
[302] C. Thierfelder, P. Schwerdtfeger, A. Koers, A. Borschevsky, B. Fricke: Phys. Rev. A 80 (2009)
022501
[303] M. Schadel: Angew. Chem.-Int. Edit. 45 (2006) 368
[304] P. Pyykko: Mol. Phys. 99 (2001) 1617
[305] P. Pyykko: Mol. Phys. 106 (2008) 1965
[306] E. Arimondo, M. Inguscio, P. Violino: Rev. Mod. Phys. 49 (1977) 31
[307] W. Gordy, R. L. Cook: Microwave molecular spectra: Wiley, New York (1984)
[308] D. A. Shirley: Rev. Mod. Phys. 36 (1964) 339
References
[309]
[310]
[311]
[312]
[313]
[314]
[315]
[316]
[317]
[318]
[319]
[320]
[321]
[322]
[323]
[324]
[325]
[326]
[327]
[328]
[329]
[330]
[331]
[332]
[333]
[334]
[335]
[336]
[337]
[338]
[339]
[340]
[341]
[342]
[343]
[344]
[345]
[346]
387
P. Pyykko: Z. Naturforsch. 47a (1992) 189

M. Urban, A. J. Sadlej: Chem. Phys. Lett. 173 (1990) 157
V. Kello, A. J. Sadlej: Mol. Phys. 89 (1996) 127
V. Kello, A. J. Sadlej: Chem. Phys. Lett. 292 (1998) 403
V. Kello, A. J. Sadlej, P. Pyykko, D. Sundholm, M. Tokman: Chem. Phys. Lett. 304 (1999) 414
V. Kello, A. J. Sadlej: Mol. Phys. 96 (1999) 275
V. Kello, A. J. Sadlej: Phys. Rev. A 60 (1999) 3575
V. Kello, P. Pyykko, A. J. Sadlej, P. Schwerdtfeger, J. Thyssen: Chem. Phys. Lett. 318 (2000) 222
V. Kello, A. J. Sadlej: J. Chem. Phys. 112 (2000) 522
V. Kello, A. J. Sadlej, P. Pyykko: Chem. Phys. Lett. 329 (2000) 112
V. Kello, P. Pyykko, A. J. Sadlej: Chem. Phys. Lett. 346 (2001) 155
J. Bieron, P. Pyykko, D. Sundholm, V. Kello, A. J. Sadlej: Phys. Rev. A 64 (2001) 052507
V. Kello, A. J. Sadlej: J. Chem. Phys. 120 (2004) 9424
P. Schwerdtfeger, R. Bast, M. C. L. Gerry, C. R. Jacob, M. Jansen, V. Kello, A. V. Mudring, A. J.
Sadlej, T. Saue, T. Sohnel, F. E. Wagner: J. Chem. Phys. 122 (2005) 124317
L. Demovic, V. Kello, A. J. Sadlej, S. A. Cooke: J. Chem. Phys. 124 (2006) 184308
V. Kello, A. J. Sadlej: Collect. Czech. Chem. Commun. 72 (2007) 64
A. D. Buckingham: J. Chem. Phys. 36 (1962) 3096
A. Wolf, M. Reiher: J. Chem. Phys. 124 (2006) 064102
A. Wolf, M. Reiher: J. Chem. Phys. 124 (2006) 064103
R. Mastalerz, G. Barone, R. Lindh, M. Reiher: J. Chem. Phys. 127 (2007) 074105
M. Pernpointner, M. Seth, P. Schwerdtfeger: J. Chem. Phys. 108 (1998) 6722
M. Barysz, A. J. Sadlej: Theor. Chem. Acc. 97 (1997) 260
L. Visscher, T. Enevoldsen, T. Saue, J. Oddershede: J. Chem. Phys. 109 (1998) 9677
J. N. P. van Stralen, L. Visscher: Mol. Phys. 101 (2003) 2115
F. J. Lovas, E. Tiemann: J. Phys. Chem. Ref. Data 3 (1974) 609
A. F. Oluwole, S. G. Schmelling, H. A. Shugart: Phys. Rev. C 2 (1970) 228
D. Sundholm, P. Pyykko, L. Laaksonen, A. J. Sadlej: Chem. Phys. Lett. 112 (1984) 1
K. D. Hensel, C. Styger, W. Jager, A. J. Merer, M. C. L. Gerry: J. Chem. Phys. 99 (1993) 3320
N. J. Martin, P. G. H. Sandars, G. K. Woodgate: Proc. Roy. Soc. A 305 (1968) 139
J. S. M. Harvey, L. Evans, H. Lew: Can. J. Phys. 50 (1972) 17191727
R. Weber, B. Jeckelmann, J. Kern, U. Kiebele, B. Aas, W. Beer, I. Beltrami, K. Bos, G. de Chambrier,
P. F. A. Goudsmit, H. J. Leisi, W. Ruckstuhl, G. Strassner, A. Vacchi: Nucl. Phys. A 377 (1982) 361
S. A. Beaton, M. C. L. Gerry: J. Chem. Phys. 110 (1999) 10715
S. Buttgenbach: Hyperfine structure in 4d- and 5d-shell atoms: Springer tracts in modern physics
96. Springer-Verlag, Berlin ; New York (1982): p. 83
L. Young, C. A. Kurtz, D. R. Beck, D. Datta: Phys. Rev. A 48 (1993) 173
S. Bouazza, J. Dembczynski, E. Stachowska, G. Szawiola, J. Ruczkowski: Eur. J. Phys. D 4 (1998)
39
K. Murakawa: Phys. Rev. 100 (1955) 1369
A. Svane: Phys. Rev. B 68 (2003) 064422
S. A. Cooke, M. C. L. Gerry: Phys. Chem. Chem. Phys. 6 (2004) 4579
388
[347]
[348]
[349]
[350]
[351]
[352]
[353]
[354]
[355]
[356]
[357]
[358]
[359]
[360]
[361]
[362]
[363]
[364]
[365]
[366]
[367]
[368]
[369]
[370]
[371]
[372]
[373]
[374]
[375]
[376]
[377]
[378]
[379]
[380]
[381]
[382]
[383]
S. A. Cooke, M. C. L. Gerry: J. Mol. Spectrosc. 234 (2005) 195

R. L. A. Haiduke, A. B. F. da Silva, L. Visscher: J. Chem. Phys. 125 (2006) 064301
S. Huzinaga, M. Klobukowski: Theochem-J. Mol. Struct. 167 (1988) 1
J. Roithova, D. Schroder: Coord. Chem. Rev. 253 (2009) 666
P. Schwerdtfeger: Angew. Chem.-Int. Edit. 42 (2003) 1892
P. Pyykko, P. Zaleski-Ejgierd: J. Chem. Phys. 128 (2008) 124309
H. Schmidbaur: Gold Bull. 33 (2000) 3
E. Kraka, M. Filatov, D. Cremer: Croat. Chem. Acta 82 (2009) 233
T. Okabayashi, Y. Nakaoka, E. Yamazaki, M. Tanimoto: Chem. Phys. Lett. 366 (2002) 406
P. Schwerdtfeger, J. S. McFeaters, M. J. Liddell, J. Hrusa k, H. Schwarz: J. Chem. Phys. 103 (1995)
245
P. Schwerdtfeger, M. Dolg, W. H. E. Schwarz, G. A. Bowmaker, P. D. W. Boyd: J. Chem. Phys. 91
(1989) 1762
B. Metz, H. Stoll, M. Dolg: J. Chem. Phys. 113 (2000) 2563
W. Liu, C. van Wullen, Y. K. Han, Y. J. Choi, Y. S. Lee: Adv. Quantum Chem. 39 (2001) 325
M. Barysz, P. Pyykko: Chem. Phys. Lett. 285 (1998) 398
C. Puzzarini, K. A. Peterson: J. Chem. Phys. 122 (2005) 084323
F. A. Cotton, C. B. Harris: Inorg. Chem. 4 (1965) 330
S. Ritter: Chem. Eng. News 83 (2005) 9
B. O. Roos: Collect. Czech. Chem. Commun. 68 (2003) 265
B. O. Roos, L. Gagliardi: Inorg. Chem. 45 (2006) 803
T. Nguyen, A. D. Sutton, M. Brynda, J. C. Fettinger, G. J. Long, P. P. Power: Science 310 (2005)
844
M. Brynda, L. Gagliardi, P. O. Widmark, P. P. Power, B. O. Roos: Angew. Chem.-Int. Edit. 45 (2006)
3804
L. Gagliardi, B. Roos: In: Lecture Series on Computer and Computational Sciences, ed. G. Maroulis,
vol. 6, pp. 622. Brill Academic Publishers, Leiden (2006)
L. Gagliardi, B. O. Roos: Nature 433 (2005) 848
Malmqvist, L. Gagliardi: J. Am. Chem. Soc. 128 (2006) 17000
B. O. Roos, P.-A.
E. S. Kryachko, F. Remacle: J. Chem. Phys. 127 (2007) 194305
E. S. Kryachko: J. Mol. Struct. 880 (2008) 23
P. Pyykko, N. Runeberg: Chem. Asian J. 1 (2006) 623
J. Li, Y. X. Qiu, S. G. Wang: J. Phys. Chem. A 113 (2009) 1646
I. Stich:
Acta Phys. Slovaca 57 (2007) 1
M. Konopka, R. Rousseau, I. Stich,

D. Marx: J. Am. Chem. Soc. 126 (2004) 12103
M. Konopka, R. Turansky, M. Dubecky, D. Marx, I. Stich:

J. Phys. Chem. C 113 (2009) 8878
H. Gronbeck, H. Hakkinen: J. Phys. Chem. B 111 (2007) 3325
V. De Renzi: Surf. Sci. 603 (2009) 1518
Y. Akinaga, T. Nakajima, K. Hirao: J. Chem. Phys. 114 (2001) 8555
A. Casini, C. Hartinger, C. Gabbiani, E. Mini, P. J. Dyson, B. K. Keppler, L. Messori: J. Inorg.
Biochem. 102 (2008) 564
J. V. Burda, M. Zeizinger, J. Leszczynski: J. Comput. Chem. 26 (2005) 907
H. Chojnacki, J. Kuduk-Jaworska, I. Jaroszewicz, J. J. Janski: Pol. J. Chem. 83 (2009) 1013
References
[384]
[385]
[386]
[387]
[388]
[389]
[390]
[391]
[392]
[393]
[394]
[395]
[396]
[397]
[398]
[399]
[400]
[401]
[402]
[403]
[404]
[405]
[406]
[407]
[408]
[409]
[410]
[411]
[412]
[413]
[414]
[415]
[416]
[417]
[418]
[419]
389
N. P. Labello, A. M. Ferreira, H. A. Kurtz: J. Phys. Chem. A 110 (2006) 13507

A. Antusek, M. Urban, A. J. Sadlej: J. Chem. Phys. 119 (2003) 7247
J. Granatier, M. Urban, A. J. Sadlej: J. Phys. Chem. A 111 (2007) 13238
J. Granatier, M. Urban, A. J. Sadlej: Chem. Phys. Lett. 484 (2010) 154
S. Li, K. A. Peterson, D. A. Dixon: J. Chem. Phys. 128 (2008) 154301
P. Archirel, V. Dubois, P. Maitre: Chem. Phys. Lett. 323 (2000) 7
J. Miyawaki, K. Sugawara, H. Takeo, C. Dedonder-Lardeux, S. Martrenchard-Barra, C. Jouvet,
D. Solgadi: Chem. Phys. Lett. 302 (1999) 354
V. Dubois, P. Archirel, P. Maitre: Chem. Phys. Lett. 323 (2000) 1
N. A. Lambropoulos, J. R. Reimers, N. S. Hush: J. Chem. Phys. 116 (2002) 10277
S. F. Boys, F. Bernardi: Mol. Phys. 19 (1970) 553
J. Granatier, M. Urban, A. J. Sadlej (2010): to be published
H. M. Lee, S. K. Min, E. C. Lee, J. H. Min, S. Odde, K. S. Kim: J. Chem. Phys. 122 (2005) 064314
H. M. Lee, M. Diefenbach, S. B. Suh, P. Tarakeshwar, K. S. Kim: J. Chem. Phys. 123 (2005) 074328
C. Gourlaouen, J. P. Piquemal, T. Saue, O. Parisel: J. Comput. Chem. 27 (2006) 142
J. Hrusa k, D. Schroder, H. Schwarz: Chem. Phys. Lett. 225 (1994) 416
S. G. Lias: In: NIST Chemistry WebBook, NIST Standard Reference Database Number 69, eds.
P. J. Linstrom, W. G. Mallard. National Institute of Standards and Technology, Gaithersburg ():
http://webbook.nist.gov
M. Barysz, J. Leszczynski, A. Bilewicz: Phys. Chem. Chem. Phys. 6 (2004) 4553
M. Barysz, D. Kedziera, J. Leszczynski, A. Bilewicz: Int. J. Quantum Chem. 106 (2006) 2422
H. Fang, S. G. Wang: Theochem-J. Mol. Struct. 773 (2006) 15
O. D. Haberlen, N. Rosch: J. Phys. Chem. 97 (1993) 4970
T. M. Bernhardt, J. Hagen, S. M. Lang, D. M. Popolan, L. D. Socaciu-Siebert, L. Woste: J. Phys.
Chem. A 113 (2009) 2724
P. Schwerdtfeger, M. Lein, R. P. Krawczyk, C. R. Jacob: J. Chem. Phys. 128 (2008) 124302
S. G. Wang, W. H. E. Schwarz: J. Am. Chem. Soc. 126 (2004) 1266
B. Ruscic, J. Berkowitz: J. Chem. Phys. 97 (1992) 1818
G. D. Bent: J. Chem. Phys. 89 (1988) 7298
R. S. Mulliken, W. B. Person: Molecular complexes; a lecture and reprint volume: WileyInterscience, New York, (1969)
I. S. Lim, P. Botschwina, R. Oswald, V. Barone, H. Stoll, P. Schwerdtfeger: J. Chem. Phys. 127
(2007) 104313
G. Frenking: J. Organomet. Chem. 635 (2001) 9
J. Kundu, O. Neumann, B. G. Janesko, D. Zhang, S. Lal, A. Barhoumi, G. E. Scuseria, N. J. Halas:
J. Phys. Chem. C 113 (2009) 14390
M. Hargittai: Accounts Chem. Res. 42 (2009) 453
K. J. Donald, M. Hargittai, R. Hoffmann: Chem.-Eur. J. 15 (2009) 158
E. S. Kryachko, F. Remacle: Mol. Phys. 106 (2008) 521
A. Zaitsevskii, C. van Wullen, A. V. Titov: J. Chem. Phys. 132 (2010) 081102
T. Helgaker, M. Jaszunski, K. Ruud: Chem. Rev. 99 (1999) 293
J. Vaara: Phys. Chem. Chem. Phys. 9 (2007) 5399
M. Ilias, T. Saue, T. Enevoldsen, H. J. Aa. Jensen: J. Chem. Phys. 131 (2009) 124119
390
[420] T. Saue: In: Relativistic electronic structure theory, Part 1. Fundamentals, ed. P. Schwerdtfeger,
Theoretical and computational chemistry. Vol. 11, pp. 332400. Elsevier, Amsterdam (2002)
[421] G. A. Aucar, T. Saue, L. Visscher, H. J. Aa. Jensen: J. Chem. Phys. 110 (1999) 6208
[422] F. London: J. Phys. Radium 8 (1937) 397
[423] T. Helgaker, P. Jrgensen: J. Chem. Phys. 95 (1991) 2595
[424] T. Saue, K. Fgri, T. Helgaker, O. Gropen: Mol. Phys. 91 (1997) 937
[425] S. Komorovsky, M. Repisky, O. L. Malkina, V. G. Malkin, I. M. Ondk, M. Kaupp: J. Chem. Phys.
128 (2008) 104101
[426] L. Cheng, Y. Xiao, W. Liu: J. Chem. Phys. 131 (2009) 244113
[427] S. Komorovsky, M. Repisky, O. L. Malkina, V. G. Malkin: J. Chem. Phys. 132 (2010) 154101
[428] M. Repisky, S. Komorovsky, E. Malkin, O. L. Malkina, V. G. Malkin: Chem. Phys. Lett. 488 (2010)
94
[429] P. Manninen, P. Lantto, J. Vaara, K. Ruud: J. Chem. Phys. 119 (2003) 2623
[430] M. Jaszunski, K. Ruud: Mol. Phys. 104 (2006) 2139
[431] J. Gauss, J. F. Stanton: Adv. Chem. Phys. 123 (2002) 355
[432] A. Antusek, M. Jaszunski: Mol. Phys. 104 (2006) 1463
[433] A. Antusek, D. Kedziera, K. Jackowski, M. Jaszunski, W. Makulski: Chem. Phys. 352 (2008) 320
[434] A. Antusek, M. Pecul, J. Sadlej: Chem. Phys. Lett. 427 (2006) 281
[435] V. G. Malkin, O. L. Malkina, D. R. Salahub: Chem. Phys. Lett. 261 (1996) 335
[436] M. Buhl, M. Kaupp, O. L. Malkina, V. G. Malkin: J. Comput. Chem. 20 (1999) 91
[437] M. Kaupp, M. Buhl, V. G. Malkin: Calculation of NMR and EPR parameters : theory and applications: Wiley-VCH, Weinheim (2004)
[438] S. K. Wolff, T. Ziegler, E. van Lenthe, E. J. Baerends: J. Chem. Phys. 110 (1999) 7689
[439] S. K. Wolff, T. Ziegler: J. Chem. Phys. 109 (1998) 895
[440] R. Bouten, E. J. Baerends, E. van Lenthe, L. Visscher, G. Schreckenbach, T. Ziegler: J. Phys. Chem.
A 104 (2000) 5600
[441] V. Pershina, J. Anton, B. Fricke: J. Chem. Phys. 127 (2007) 134310
[442] V. Pershina, T. Bastug: Chem. Phys. 311 (2005) 139
[443] V. Pershina, J. Anton, T. Bastug: Eur. Phys. J. D 45 (2007) 87
[444] V. Pershina, A. Borschevsky, E. Eliav, U. Kaldor: J. Chem. Phys. 129 (2008) 144106
[445] I. Cernu
sa k, M. Aranyosiova, O. Vollarova, D. Velic, O. Kirdajova, J. Benko: Int. J. Quantum Chem.
109 (2009) 2365
[446] M. Ilcin, V. Lukes, V. Laurinc, S. Biskupic: Collect. Czech. Chem. Commun. 72 (2007) 363
[447] L. Bucinsky, S. Biskupic, M. Ilcin, V. Lukes, V. Laurinc: J. Comput. Chem. 30 (2009) 65
[448] M. Ilcin, V. Lukes, L. Bucinsky, V. Laurinc, S. Biskupic: Int. J. Quantum Chem. 108 (2008) 2150
[449] C. Vogt, R. D. Hoffmann, U. C. Rodewald, G. Eickerling, M. Presnitz, V. Eyert, W. Scherer,
R. Pottgen: Inorg. Chem. 48 (2009) 6436
[450] J. A. Li, G. Schreckenbach, T. Ziegler: Inorg. Chem. 34 (1995) 3245
[451] G. Eickerling, R. Mastalerz, V. Herz, W. Scherer, H. J. Himmel, M. Reiher: J. Chem. Theory Comput.
3 (2007) 2182
[452] P. Pyykko, N. Runeberg: Angew. Chem.-Int. Edit. 41 (2002) 2174
[453] X. Li, B. Kiran, J. Li, H. J. Zhai, L. S. Wang: Angew. Chem.-Int. Edit. 41 (2002) 4786
[454] K. Manninen, P. Pyykko, H. Hakkinen: Phys. Chem. Chem. Phys. 7 (2005) 2208
References
391
[455] M. Haruta, T. Kobayashi, H. Sano, N. Yamada: Chem. Lett. (1987) 405

[456] S. Tokonami, H. Shiigi, T. Nagaoka: In: Nanostructured thin films and surfaces, ed. C. S. S. R.
Kumar, pp. 175202. Wiley-VCH, Weinheim (2010)
[457] T. Sohnel, H. Hermann, P. Schwerdtfeger: Angew. Chem.-Int. Edit. 40 (2001) 4382
[458] T. Sohnel, H. Hermann, P. Schwerdtfeger: J. Phys. Chem. B 109 (2005) 526
[459] S. Biering, A. Hermann, J. Furthmuller, P. Schwerdtfeger: J. Phys. Chem. A 113 (2009) 12427
[460] N. Gaston, P. Schwerdtfeger: Phys. Rev. B 74 (2006) 024105
[461] B. Paulus, K. Rosciszewski, N. Gaston, P. Schwerdtfeger, H. Stoll: Phys. Rev. B 70 (2004)
[462] E. Pahl, F. Calvo, P. Schwerdtfeger: Int. J. Quantum Chem. 109 (2009) 1812
[463] A. Zaitsevskii, C. van Wullen, E. A. Rykova, A. V. Titov: Phys. Chem. Chem. Phys. 12 (2010) 4152
Miroslav Ilias born in Bojnice, Slovakia (1975), graduated in Physical Chemistry at the Comenius University in Bratislava (1998). At the same institution he received his PhD. degree in
Chemical Physics (2002). Afterwards he spent more than 4 years at various institutions abroad
(Denmark, France, Israel) as a post-doctoral fellow, working on the development and applications of methods of relativistic quantum chemistry with professors L. Visscher, T. Saue, H. J. A.
Jensen, U. Kaldor. Up to 2009 he worked at the University of Zilina,

presently he is a faculty
member at the Matej Bel University in Banska Bystrica, Slovakia.
Vladimr Kello was born in Bratislava, Czechoslovakia (1947). He studied chemistry at the
Comenius University Bratislava, where he graduated in 1972 and in 1977 he received PhD. from
the same institution working with Miroslav Urban on a topic ab initio calculations of equilibrium
constants of chemical reactions. He has been on the faculty of the Comenius University since
1978. He was a postdoctoral fellow at Max-Planck Institute of Astrophysics in Garching, Germany with Geerd H. F. Diercksen in the period 1981-1982 and later was a frequent visitor and
spent many months at this institute. A close scientific collaboration with Andrzej J. Sadlej began
during the Garching postdoctoral stay and lasted for 30 years. In the frame of this collaboration
he many times visited University of Lund, Sweden and later the Copernicus University Torun,
Poland. He became Associated Professor in 1991, received the D.Sc. degree in 1995, and was
promoted to Full Professor at the Comenius University in 1998. His research centers on calculations of atomic and molecular properties by means of highly electron correlated methods with
inclusion of relativistic effects, with particular interest in electric dipole moments, polarizabilities
and field gradients.
Miroslav Urban (1942) is a Professor at the Slovak University of Technology in Bratislava,

Faculty of Materials Science and Technology in Trnava, Slovakia. Since 1964 he worked at
the Comenius University, Faculty of Natural Sciences, up to January 2009 as a Full Professor
of Theoretical and Computational Chemistry. He received his master degree in Physical chemistry at the Comenius University (1964), Phd. Degree at the Charles University in Prague (1972;
supervisors J. Koutecky, J. Paldus and R. Polak). He worked repeatedly as a visiting scientist/visiting Professor at the Quantum Theory Project, University of Florida, Gainesville, USA
(with Professor R. J. Bartlett), Max-Planck Institute for Astrophysics, Garching, Germany (with
Professor G. H. F. Diercksen), Department of Theoretical Chemistry, University of Lund, Sweden (with Professor A. J. Sadlej), Center for Computational Quantum Chemistry, University of
Georgia, Athens, USA (with Professor H. F. Schaefer). He published 108 scientific papers, 3
books, 3 chapters in monographs. His scientific interests are focused to the development of the
Many-Body Perturbation Theory and Coupled Cluster methods, developments of efficient computer codes, applications to accurate calculations of, e.g., electric and optical properties, electron
affinities, reaction and interaction energies, relativistic and electron correlation effects in molecular properties. He is a member of the scientific board of the World Association of Theoretical
and Computational Chemists, a member of the International Academy of Quantum Molecular
Science, Slovak Academic Society, and Learned Society of the Slovak Academy of Sciences.

Relativistic Effects in Atomic and Molecular Properties

Uploaded by

Copyright:

Available Formats

Relativistic Effects in Atomic and Molecular Properties

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Relativistic Effects in Atomic and Molecular Properties

Uploaded by

Copyright:

Available Formats

acta physica slovaca vol. 60 No.

RELATIVISTIC EFFECTS IN ATOMIC AND MOLECULAR PROPERTIES

Received 17 May 2010, accepted 1 June 2010

Relativistic effects, Electron correlation, Change of picture, Ionization

Relativistic effects in atomic and molecular properties

Manifestation of relativistic effects on atomic properties. Basic notes.

Basic notes on nonrelativistic electronic structure methods for manyelectron atoms

Relativistic theory of manyelectron atoms and molecules

4.7.2 RASSISO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324

Relativistic effects in atoms and molecules, their properties and interactions

Relativistic effects in atomic and molecular properties

Relativistic effects in atomic and molecular properties

Manifestation of relativistic effects on atomic properties. Basic notes.

Historically, a breakdown in understanding of the chemical and physical behaviour of elements

Fig. 2.1. Mendeleev Periodic Table as published in 1869, Ref. [31].

Manifestation of relativistic effects on atomic properties. Basic notes.

Relativistic effects in atomic and molecular properties

Shell HartreeFock oneelectron calculation (ROHF), which treats interelectronic interactions

Manifestation of relativistic effects on atomic properties. Basic notes.

Relativistic effects in atomic and molecular properties

Manifestation of relativistic effects on atomic properties. Basic notes.

Zdependence of the relativistic effects

Relativistic effects in atomic and molecular properties

ical model is more suitable for describing studied chemical system.

Manifestation of relativistic effects on atomic properties. Basic notes.

Semiclassical estimate of spinorbit effects

Relativistic effects in atomic and molecular properties

where the electric field, E, for a central potential V (r) is given by

The intrinsic magnetic moment of the electron is (in a.u.)

where s is the spin moment, see Section 3.1.2. The term

which allows to estimate SOsplittings by using quantum numbers j, l and s, related to j 2 , l2

Manifestation of relativistic effects on atomic properties. Basic notes.

Relativistic effects in atomic and molecular properties

Basic notes on nonrelativistic electronic structure methods for manyelectron atoms

Schrodinger equation and the hydrogenlike atoms

The Schrodinger equation, published in 1926, is a fundamental equation of the nonrelativistic

n (r, t) = ( 1 + V (r))n (r, t),

Basic notes on nonrelativistic electronic structure methods

The zcomponent l of the angular momentum is quantized as

Relativistic effects in atomic and molecular properties

The Pauli matrices,

Basic notes on nonrelativistic electronic structure methods

Relativistic effects in atomic and molecular properties

Basic notes on nonrelativistic electronic structure methods

The independentelectron model: The HartreeFock method

is the oneelectron operator, see Eq. 3.11, Ji and K

Relativistic effects in atomic and molecular properties

Basic notes on nonrelativistic electronic structure methods

define a reference wave function.

Relativistic effects in atomic and molecular properties

Electron correlation in the nonrelativistic theory of many electron systems

Basic notes on nonrelativistic electronic structure methods

where ai are monoexcited configurations, ab

Relativistic effects in atomic and molecular properties

Basic notes on nonrelativistic electronic structure methods

Coupled Cluster theory

Relativistic effects in atomic and molecular properties

Basic notes on nonrelativistic electronic structure methods

Next, we multiply Eq. 3.35 from left by exp(T) and obtain:

Relativistic effects in atomic and molecular properties