Journal of Electron Spectroscopy and Related Phenomena 110111 (2000) 213233

www.elsevier.nl / locate / elspec

Resonant inelastic X-ray scattering as a probe of optical scale

excitations in strongly electron-correlated systems:
quasi-localized view
Sergei M. Butorin
Department of Physics, Uppsala University, Box 530, S-751 21 Uppsala, Sweden

Abstract

An application of resonant inelastic X-ray scattering technique for studying optical scale excitations in electron-correlated
materials is discussed. Examples are given including data obtained for 3d transition metal, lanthanide, and actinide systems.
In some cases, the data are compared with the results of crystal-field multiplet and Anderson impurity model calculations.
Advantages of this technique are pointed out, such as an ability to probe an extended multiplet structure of the ground state
configuration, which is not fully accessible by other spectroscopies, an extreme sensitivity of spectral profiles to the chemical
state of the element in question and to the crystal-field strength, and a great potential in probing the ground state character
(for example, ground state J-mixing in rare-earths) due to the techniques elemental selectivity and strict selection rules.
Issues are addressed, such as a possible deviation from the linear dispersion of inelastic scattering structures, corresponding
to charge-transfer excitations, with varying excitation energies and an estimation of values for model parameters, involved in
the description of charge-transfer processes. 2000 Elsevier Science B.V. All rights reserved.

Keywords: Resonant inelastic X-ray scattering; 3d transition metal, lanthanide and actinide compounds; Elementary and charge-transfer
excitations; Anderson impurity model calculations

1. Technique and models particle picture and an atomic-like approach to

characterize the electronic structure of these com-
To successfully describe various physical prop- pounds is more appropriate.
erties of a system in question it is necessary to obtain In this case, a state of the system without a core
knowledge about the ground state and low-energy hole is described in terms of intra-atomic neutral
excited states of this system. For 3d transition excitations (a multiplet structure of the ground state
element, lanthanide, and actinide compounds with a electronic configuration due to exchange, crystal
partly filled d or f shell, strong correlation effects, field, spinorbit interactions, etc.) and / or inter-
when the dispersional part of d or f bandwidth is atomic charge-transfer excitations. The latter are the
smaller than the on-site Coulomb interaction U result of electron hopping from delocalized states to
between localized electrons, break down a single- a localized state and are treated by short-range

0368-2048 / 00 / $ see front matter 2000 Elsevier Science B.V. All rights reserved.
PII: S0368-2048( 00 )00166-3
214 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233

models, such as an Anderson impurity model [1], By now, it has been proven that at some core-
using a set of parameters. The models are repre- thresholds of electron-correlated systems, X-ray
sented by the Hamiltonian emission spectroscopy with monochromatic photon
H5 O n 1O n
k, a , s
k a k as
m, s
m ms
excitation can be considered as an analog of these
techniques so that the excitationradiative de-excita-

1 O (V c c 1 H.c.)
tion channel can be treated as resonant inelastic
ka m ms k as X-ray scattering (RIXS) process. Final states probed
k, a ,m, s
via such a channel are related to eigenvalues of the
1U O n n . m s m9 s 9 (1) ground state Hamiltonian. The core-hole lifetime is
m,m9, s, s 9
not a limit on the resolution in this spectroscopy (see
Important physical quantities included in this e.g. Ref. [7]). It is important to distinguish between
Hamiltonian are the delocalized- and localized-state the many-body description of RIXS and a single-
energies k a and m , hopping matrix element Vk a m , particle approach which is usually applied to wide-
and U. Here k, a, s, and m denote a wave vector, an band materials [8]. The differences between two
index of the energy level in the valence band, a spin formalisms are schematically illustrated in Fig. 1.
index, and an azimutal quantum number, respective- According to the many-body picture, an energy of
ly. For the description of core spectroscopies, a a photon, scattered on a certain low-energy excita-
further term is added to the Hamiltonian to account tion, should change by the same amount as a change
for coupling between localized electron and a core in an excitation energy of the primary beam (see a
hole. The values of model parameters are optimized decay route of core excitation B versus that of A) so
by fitting both high-energy spectroscopic and low- that inelastic scattering structures have constant
energy transport data and then employed to describe energy losses and follow the elastic peak on the
the character of the ground state, different ground- emitted-photon energy scale. In the single-particle
state properties, the nature and size of the band gap view, energy positions of specific inelastic-scattering
in insulators [2], etc. structures with respect to the elastic peak which are
Since the interpretation of transport measurements defined by the momentum conservation rule may
in these regards is often hampered by the presence of vary only within the energy range covered by the
defects and by the importance of electronlattice occupied part of the valence band. In Fig. 1a, this is
interactions, high-energy spectroscopies which di- reflected in the situation when, for core excitation B
rectly probe the electronic degrees of freedom are
often used for preliminary estimations of model
parameters. For these estimations, it is important to
take into account significant configurational depen-
dence of model parameters which is predicted by
first-principles calculations [36]. In particular, re-
moving / adding of a valent d or f electron is expected
to result in a decrease / increase in the value of the
hybridization strength V which in turn may lead to
renormalization effects for U. These effects are more
pronounced for core-level spectroscopies. In the
presence of a core hole V is strongly reduced since
the waverfunctions become more localized. The
renormalization of model parameters in the final state
can produce significant uncertainty in estimated
values of these parameters in the ground state. In this
situation, X-ray scattering techniques become very
attractive because the scattering process is charge- Fig. 1. Schematic representation of the radiative de-excitation
neutral. process for two different core excitations A and B.
 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233 215

with the higher energy, the radiative decay results in (EELS) spectroscopies, there are some advantages in
a transition with the lower energy than those for A, using this method:
respectively. In spite of simplifications made here,
the outlined differences can be used to test the 1. the technique is not surface-sensitive, helping to
validity of one or another model for a system in avoid the confusion with a formation of additional
question. states because of surface defects;
As an example, we use data from Ref. [9] (see Fig. 2. its element-specificity enables one to study very
2) which were obtained at the U 3d 5 / 2 edge of UO 3 . dilute compounds since metal states can be prob-
The inelastic scattering structure with the energy loss ed separately from ligand states;
of about 5 eV is observed to follow the elastic peak 3. the cross-section for inelastic X-ray scattering is
up to 20 eV above the 3d 5 / 2 threshold while the strongly enhanced on the resonance in contrast to
width of the occupied part of the valence band is weak dipole-forbidden transitions in optical ab-
only | 4 eV. This indicates the importance of elec- sorption spectra;
tron correlation effects in UO 3 . 4. the dipole nature of radiative transitions makes it
Although, the information provided by the RIXS easier to calculate RIXS intensities compared to
technique is similar to that obtained from optical dd ( f f ) intensities in optical spectroscopy or in
absorption or low-energy electron-energy-loss EELS.

In calculations of resonant X-ray scattering as a

second order optical process, only a resonant term of
the modified KramersHeisenberg equation is usual-
ly used, where the spectral intensity is given by

OUO U
2
k f uD (q19 ) uilkiuD q(1 ) ugl
Iqq9 (V,v ) 5 ]]]]]]
f i Eg 1 V 2 Ei 2 iGi / 2
3 d (Eg 1 V 2 Ef 2 v ). (2)

Here, ugl, uil, and u f l are the ground, intermediate,

and final states with energies Eg , Ei , and Ef , respec-
tively, while V and v represent energies of incident
and scattered photons, respectively. D is the dipole
operator, G stands for the intermediate state lifetime
and q and q9 are polarizations of the light with
respect to the quantization axis.
For the q-polarized incident photons, spectra
detected in different directions with respect to the
quantization axis can be described [10] as

I iso 5 Iq0 1 Iq1 1 Iq21 ,

1
I 908 5 Iq0 1 ](Iq1 1 Iq21 ),
2
I 08 5 Iq1 1 Iq 21 (3)

Fig. 2. U M5 X-ray fluorescence spectra recorded at the U 3d 5 / 2 For the case of studying the multiplet structure of the
threshold of UO 3 [9]. Excitation energies used in these measure- ground state configuration, the ability of resonant
ments are indicated by arrows on the absorption spectrum. X-ray emission spectroscopy to probe low-energy
216 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233

excitations was first discussed by Tanaka and Kotani For instance, dd excitations of MnO with energies
[11] in the description of resonant Cu 3d 2p of around 5 eV and higher can hardly be observed in
spectra of La 2 CuO 4 and CuO. The difference in dd optical and EELS spectra [18] for that reason.
excitation profiles for 3d 9 multiplets between these Whereas, RIXS does not have this disadvantage as
two oxides was predicted. However, no experimental can be seen from the analysis of the data in Figs. 3
data were available with the energy resolution being and 4.
sufficiently high to support conclusions made by When the energy of incident photons is set at the
authors. The first experiment which unambiguously Mn 2p threshold, an excited electron itself screens
confirmed the ability of this resonant technique to the core hole. Due to dipole selection rules, there are
probe elementary excitations was performed on MnO different radiative transitions for the de-excitation
[12] (two years later, high-resolution RIXS data at process: back to the ground state or to low-lying
the Cu 3p resonance of cuprates were published [7] excited states so that the multiplet structure of the
which are in good agreement with theoretical predic- ground state configuration can be probed. Mn L2,3
tions). Prior to this, probing of f f excitations in (3d,4s 2p transitions) X-ray fluorescence spectra
rare-earths was discussed in Ref. [13]. The efficiency of single-crystal MnO (100), displayed in Fig. 3, can
of the technique in studies of charge-transfer excita- be ordered by assigning the peaks to one of three
tions for valence electrons in correlated systems was
first demonstrated by Butorin et al. [9] for both soft
and intermediate-energy X-ray regions.
The present paper is to large extent centered on
the description of RIXS as a tool for studying
elementary excitations. Various aspects of probing
the charge-transfer excitations by this spectroscopy
and data interpretation within the Anderson impurity
model framework are discussed extensively in the
contribution by Kotani [14]. A reader is also referred
to publications by the present author et al. [9,1517]
where the latter issue is addressed.

2. 3d Transition metal compounds

The study of dd excitations is particularly im-

portant for elements from the middle of the 3d row
because a multiplet structure of the ground state
configuration is fairly rich. In the final state of the
nonresonant X-ray emission and photoemission pro-
cesses, a system has one electron less compared to
the ground state, as a consequence the multiplet
structure observed in the X-ray emission and valence
photoemission spectra is different from that for the
ground state configuration. Instead, optical absorp-
tion and EELS are usually used to study low-energy
dd excitations. These spectroscopies are however
Fig. 3. Mn L2,3 X-ray fluorescence spectra of single-crystal MnO
not element selective so that intra-atomic dd transi-
(100) recorded at different excitation energies near Mn 2p
tions often appear as weak structures on the slope of thresholds [12]. Excitation energies are indicated by ticks on the
the main absorption edge. Rich multiplets can be absorption spectrum shown in the inset. The excitation for the
partly hidden under intense inter-band transitions. uppermost spectrum was 716 eV.
 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233 217

be understood as a consequence of the spin ordering

of the excited states. The lowest 2p 5 3d 6 intermediate
states would have the highest possible spin (also
sextets). If these states decay, they are likely to end
up again as a sextet, the 6 S ground state, contributing
to the recombination line. But at higher energies in
the L3 absorption multiplet one finds quartets and
doublets (spin is not conserved because of the large
2p spinorbit interaction). These higher-lying quar-
tets are less likely to recombine into the sextet
ground state, which explains the relatively weak
recombination line. This general ordering of the spin
states in the intermediate state is not only true within
the L3 region, but also over the whole spectrum, as
explained by Thole and Van der Laan [19]. Within
the L2 region, one can see a clearly lower intensity of
the elastic peak on the high-energy slope (Fig. 3h)
than on the first maximum (Fig. 3f). Another quali-
tative explanation for the trends in the intensity of
the elastic peak is that if the absorption is relatively
weak, the optical matrix element from the excited
state back to the ground state is also small, and the
core hole is more likely to decay to an excited final
state.
Likewise in the elastic peak, inelastic scattering
structures reveal the strong dependence on varying
excitation energies. To make it easier to identify
excitations in the final state, the Mn L2,3 X-ray
fluorescence data are displayed on the energy-loss
scale in Fig. 4 along with results of model atomic-
multiplet calculations for the Mn 21 ion from Ref.
[12]. The calculations were performed using Eq. (2)
where interference terms were neglected. In other
Fig. 4. Resonant Mn L2,3 X-ray fluorescence data (dots) of single-
words, intensities were summed incoherently. The
crystal MnO plotted as energy loss spectra together with results of atomic multiplet structures and matrix elements of
model calculations (solid lines) for the Mn 21 ion [12]. Letters dipole transitions between 2p 5 3d 6 and 3d 5 configura-
correspond to the same excitation energies as in Fig. 3. tions were calculated in intermediate coupling in the
spherical O3 group using Cowans programs [20].
categories: the recombination (elastic) peak, the The radial parts of the 3d3d and 2p3d Coulomb
resonating loss structures due to dd excitations and and exchange multipole interactions (so-called Slater
charge transfer transitions (i.e. RIXS structures), and integrals) were scaled down to 80% of their Hartree
the normal La , b X-ray emission lines which appear at Fock values to account for intra-atomic configuration
constant energies of emitted photons. The electronic interaction and hybridization effects.
recombination peak is at the same energy as the One can see in Fig. 4 that these calculations of
excitation energy, except for the possibility of 3d 5 2p 5 3d 6 3d 5 transitions are very successful
phonon losses. in reproducing the spectra excited on the L3 absorp-
The relative intensity of the recombination peak tion multiplet (spectra a,b,c,d). Although the crystal
decreases with increasing excitation energy. This can field interaction was neglected, the atomic approach
218 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233

is adequate for the interpretation of experimental spectra of this compound are entirely dominated by
data due to an extra stabilization of the 3d 5 high-spin the X-ray scattering contribution with the La normal
configuration by the Hunds rule coupling. Such an emission intensity being significant only when the
intra-atomic exchange stabilization results in a large L3 L2 M4,5 CosterKronig decay channel is open.
energy separation between the ground high-spin 6 S Experimental data, obtained with the polarization
and first excited low-spin 4 G states [21] compared to vector E in of incident photons being parallel to the c
the crystal field splitting. The first distinctly-resolved axis of the FeCO 3 crystal, are displayed in Fig. 5.
inelastic scattering structure in resonant Mn L2,3 The corresponding calculated spectra for the Fe 21
X-ray fluorescence spectra of MnO is observed at an
energy loss of about 3 eV. According to the atomic-
multiplet calculations, this structure corresponds to
the transitions to 4 G, 4 P, and 4 D-derived states. When
the crystal-field interaction is taken into account (see
e.g. Ref. [22]), the | 3-eV structure can be described
to have contributions of 4 T 1g , 4 T 2g , 4 A 1g , and 4 Eg
symmetries. Further increase in the excitation energy
results in a development of a shoulder at a loss
energy of about 5 eV (Fig. 4, spectra c,d) which is
mainly composed by transitions to states with 4 A 2g ,
4
T 1g , and 4 T 2g character. The spectral weight in this
energy-loss region becomes strongly enhanced for
excitation energies set to the L3 absorption multiplet.
In addition, non-zero intensity can be observed for
dd excitations within | 10 eV below the recombi-
nation peak despite some overlap of inelastic scatter-
ing spectra with the La emission line at the fixed
energy of 638 eV (increasing loss energy). The latter
line appears as a result of excitations into the L3
continuum and CosterKronig decay from the L2
hole states. As a whole, the RIXS data obtained for
MnO indicate that the RIXS technique indeed offers
an opportunity to study dd excitations in the
extended energy range which are often not accessible
with optical spectroscopy and EELS.
The natural extension of the RIXS spectroscopy to
ease the symmetry identification for elementary
excitations is polarization-dependent measurements.
While the reader is referred to papers by Duda et al.
[23,24] and Hague et al. [25], which are in the same
issue of the journal, for more extensive description
of linear and circular magnetic dichroism studies,
here, Fe L2,3 X-ray fluorescence data of single-
crystal FeCO 3 from Ref. [24] are used as an exam-
ple. We show the success of crystal-field multiplet Fig. 5. Fe L2,3 X-ray fluorescence spectra of single-crystal FeCO 3
recorded at different excitation energies across Fe 2p thresholds
calculations in reproducing structures in resonant
with the polarization vector E in of incident photons parallel to the
spectra due to the dipole nature of the spectroscopic c axis of the crystal (adopted from Ref. [23]). The spectra are
process. We provide the evidence of that, for the normalized to the same height. Excitation energies are indicated
excitation close to Fe 2p thresholds, the experimental by arrows on the absorption spectrum shown in the top panel.
 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233 219

system are shown in Fig. 6. The calculations were set to very small values, for simplicity. The 10Dq
performed using Eq. (2) within the framework of parameter was equal to 1.1 eV. The polarization of
crystal-field multiplet theory. Slater integrals and incident photons was taken to be along the trigonal
matrix elements were obtained using Cowans [20] axis with the 908 angle between directions of the
and Butlers [26] codes, respectively, which were incoming and outgoing radiation in the horizontal
modified by Thole [27]. A 20% reduction was plane. The lifetime G of the intermediate state was
applied to HartreeFock values of Slater integrals. set to 0.2 and 0.4 eV for L3 and L2 , respectively.
Regarding a small trigonal distortion of FeCO 3 , the We find that calculated RIXS spectra are very
calculations were done in the basis of C3v symmetry, sensitive to the value of the 10Dq parameter. Distinct
although crystal-field parameters Ds and Dt were splittings observed in experimental spectra and re-
produced in calculations with 10Dq 5 1.1 eV become
obscure in calculated profiles at 10Dq 5 1.0 eV. The
most sensitive spectrum is the one for the excitation
d. Its highest structure on the low photon energy side
(at | 706.1 eV in Fig. 6) shows the extreme depen-
dence on small variations of 10Dq, thus providing a
good fingerprint of the crystal-field strength. Indeed,
the 10Dq value derived in present calculations is
consistent with estimations from other publications
(see e.g. Ref. [28]).
A very good agreement between calculated reson-
ant X-ray scattering spectra of Fe 21 and the resonant
part of experimental Fe L2,3 X-ray fluorescence data
of FeCO 3 indicates a significantly low contribution
to the spectra from charge-transfer excitations in the
latter system, which were not taken into account in
the calculations, as well as from normal emission. In
fact, a sizeable contribution of normal emission in
the L3 region is observed only for excitation energies
set to the L2 edge as a result of the CosterKronig
decay of the L2 hole. The ionic character of Fe
chemical bonds in FeCO 3 enables crystal-field theory
to describe RIXS data in detail and as a consequence
to provide knowledge about the ground state and
low-lying excited states.
For highly covalent compounds, it is however
necessary to take into account charge-transfer excita-
tions and configurational mixing in the ground and
intermediate states of the spectroscopic process in
analysis of experimental data. The configuration
interaction modifies (sometimes significantly) the
spacing between energy levels resulting from crystal
field, spinorbit, exchange interactions, etc. The
character of states, expressed as a linear combination
of wave functions, may change significantly as well.
Fig. 6. Results of crystal-field multiplet calculations of spectra Charge-transfer effects can produce intense struc-
displayed in Fig. 5. Spectral profiles are calculated for zero tures (charge-transfer satellites) in RIXS spectra, the
temperature. energy-loss of which is related to physical quantities
220 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233

being crucial for understanding of ground state

properties. In particular, the charge-transfer energy
D 5 d 2 p , required for a transfer of an electron
from a ligand site to a metal site, and hybridization
strength V between metal 3d states and ligand states
are employed in the description of coupling between
the ground state configuration 3d n and excited
3d n11L ] (L
] denotes
]]] a hole in the valence band), thus
yielding the D 2 1 4 V eff
2
separation between bond-
ing and anti-bonding states. U for 3d electrons is
also included in the description of coupling between
3d n11L n12 2
L configuration to define the D 1
] and 3d ]
U separation between gravity centers of these con-
figurations in the limit of V 0. Charge-transfer
satellites in RIXS spectra, corresponding to transi-
tions to states of predominantly 3d n11L n12 2
] or 3d L ]
character, can be strongly enhanced by setting the
excitation energy to their 2p 5 3d n12] L or 2p 5 3d n13L 2
]
counterparts in metal 2p absorption spectra. Due to
this resonant behavior, energy positions of satellites
can be determined more accurately, thus resulting in
more accurate estimations of related physical quan-
tities.
In Figs. 7 and 8, resonant Co and Ni L2,3 X-ray
fluorescence spectra of CoO and NiO, respectively,
are shown on the photon energy scale, covering the
corresponding absorption edges. Besides the struc-
tures clearly dispersing on this scale with varying
excitation energies, there are lines which barely Fig. 7. Co L2,3 X-ray fluorescence spectra of single-crystal CoO
show any energy shifts. While the dispersing struc- (100) recorded at different excitation energies near Co 2p
thresholds [29]. Excitation energies are indicated by arrows on the
tures can be tentatively attributed to dd excitations absorption spectrum shown on top.
the assignment of other lines to normal emission is
not that obvious for different spectra. In particular,
for CoO, setting the excitation energy to the high- elements were obtained by applying a chain of
energy tail of the L3 absorption line gives rise to an programs which, in addition to Cowans and Butlers
appearance of an intense line at about 777 eV in codes, comprises a charge-transfer program written
spectrum e in Fig. 7. The energy position of this line by Thole and Ogasawara. The 3d n and 3d n11] L
is somewhat different from that of La in spectrum f, configurations for the ground and final states of the
thus indicating that it cannot entirely consist of the spectroscopic process and the 2p 5 3d n11 and
normal emission contribution. The effect is more 2p 5 3d n12]
L configurations for the intermediate state
pronounced for NiO. Low photon energy lines in were included in these calculations (n is equal to 7
and 8 for CoO and NiO, respectively). The 3d n12L 2
spectra d and e in Fig. 8 reveal a distinct low energy ]
shift relative to La recorded at an excitation energy configuration was not taken into account in the
set far above 2p3 / 2 and 2p1 / 2 thresholds. The results description of the ground state because its contribu-
of Anderson impurity model calculations (see Figs. 9 tion was estimated to be only | 1% in these oxides
and 10) suggest that these lines mainly originate [30,31]. Spectra, displayed in Figs. 9 and 10, were
from O 2p metal 3d charge-transfer excitations. calculated for the incident light linearly-polarized
Calculations were done using formula (2). Matrix along the z-axis and 908 detection geometry with the
 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233 221

Fig. 8. Ni L2,3 X-ray fluorescence spectra of single-crystal NiO

(100) recorded at different excitation energies near Ni 2p thres-
Fig. 9. Experimental X-ray scattering spectra of single-crystal
holds [29]. Excitation energies are indicated by arrows on the
CoO (100) at Co 2p thresholds (dots) together with the results of
absorption spectrum shown on top. The excitation for spectrum h
Anderson impurity model calculations for the Co 21 system in
was 916.5 eV.
octahedral symmetry. Letters correspond to the same energies of
incident photons as in Fig. 7.

parameter values summarized in Table 1. The inter-

mediate state lifetime G was set to 0.5 and 0.7 eV at resonance of transitions to the states of essentially
L3 and L2 , respectively, for CoO and to 0.6 and 0.8 3d 8L
] character. A similar situation occurs for NiO.
eV for NiO. The hybridization strength V was taken The intense lines in spectra d and e at the energy loss
to be weaker in the intermediate state than that in the of | 5.7 and | 8.2 eV, respectively, were found to
ground and final states due to its significant configu- have mainly the 3d 9L ] origin (see Fig. 9) and
rational dependence, as a priori expected [36]. represent the enhanced inelastic scattering structure
One can see in Fig. 9 that the line in spectrum e of due to charge-transfer excitations. This structure
CoO discussed above (here it is at | 8 eV on the does not follow an increasing excitation energy and
energy loss scale) is reproduced by calculations as a is observed in the spectra rather at the constant
222 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233

Table 1
Values for parameters used in Anderson impurity model calcula-
tions of resonant X-ray scattering spectra at metal 2p thresholds in
CoO and NiO a
Parameters CoO NiO
k 0.8 0.8
D 4.0 3.5
Ve g , ground state 2.2 2.2
Ve g , intermediate state 1.8 1.8
W 4.0 5.0
N 8 10
Q 2U 0.0 1.0
10Dq 0.5 0.5
Exchange field 0.3 0.15
a
k is a scaling coefficient for Slater integrals, D is defined as an
energy difference between gravity centers of 3d n 11L and 3d n
]
configurations, Ve g represents hopping for e g orbitals (Vt 2g taken as
half of the Ve g value), W is the width of the O 2p band which
shape is approximated by a circle, N is the number of levels in the
valence band, Q is a core-hole potential. The inter-atomic-ex-
change field is applied along the z-axis. All the values, except for
those for k and N, are in units of eV.

trum (in this case, intermediate states of mainly

2p 5 3d n12]
L character), the corresponding charge-
transfer satellite spectrum for inelastic scattering can
be decoupled for V and v in the limit of G 0. No
correlation between incident and emitted photon
energies can be observed for any changes of V
within a range of DV , W so that the charge-transfer
structure behaves as a normal-emission-like line.
Present calculations also reveal the sensitivity of
RIXS profiles to the non-zero exchange field giving
rise to the spectral weight transfer [33,34]. Inter-
atomic exchange interactions which correspond to
very strong effective magnetic fields and act only on
Fig. 10. Experimental X-ray scattering spectra of single-crystal the valence electron spins were taken into account
NiO (100) at Ni 2p thresholds (dots) together with the results of
Anderson impurity model calculations for the Ni 21 system in using a mean-field theory treatment. In principle, the
octahedral symmetry. Letters correspond to the same energies of exchange field strength can be different in the ground
incident photons as in Fig. 8. and intermediate states of the spectroscopic process.
For example, for a charge-transfer gap material like
photon energy than at the constant energy loss. Such NiO, the (super)exchange parameter, which is de-
behavior is not predicted by the simplistic considera- fined as J 2 V 4e g /D 3eff [35], is expected to be larger
tions of the scattering process, described in Section in the presence of a core hole due to the smaller
1. Nevertheless it is not surprising. A possibility of effective D. In the calculations, the exchange fields
the non-linear dispersion of inelastic scattering struc- in the ground and intermediate states were set to be
tures, corresponding to charge-transfer excitations, the same, for simplicity. The value of the exchange
was discussed earlier by Tanaka et al. [32]. It was field used in spectral simulations for CoO seems to
shown that for the excitation energy set to the region be too high since the Neel temperature in this oxide
of charge-transfer satellites in the absorption spec- is lower than that in NiO. However, it is very
 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233 223

difficult to reproduce experimental spectra of CoO

without turning on the exchange field even if the
Boltzmann distribution in the population of low-
lying excited states at room temperature is taken into
account in the calculations. One possible reason for
applying an excessive exchange field may be an
underestimation of spinspin correlation effects on
the spectral shape [36] in the molecular-field approx-
imation. Another reason could be a possible exist-
ence of Co vacancies in the bulk of the single-crystal
so that induced electronic holes, having oxygen
character, would be antiferromagnetically coupled
with Co 21 . The exchange interaction between these
holes and nearest-neighbor Co ions is expected to be
much stronger than the CoCo (super)exchange
interaction [37], thus producing larger spectral ef-
fects and requiring larger effective exchange fields
for the spectral simulations.
As a whole, results of Anderson impurity model
calculations are in good agreement with experimental
data except for spectrum d of CoO. Although, all the
structures are reproduced, their calculated relative
intensities are not entirely correct. In particular, the
structure at the energy loss of | 5.6 eV, which has a
charge-transfer origin, is clearly lacking an intensity
in the calculated spectrum. This suggests that the
difference Q 2 U and / or bandwidth W, as parameters
to large extent defining the charge-transfer satellite
weight in the absorption spectrum at the corre-
sponding excitation energy, may be somewhat larger
than those used in calculations. Fig. 11. Co L2,3 X-ray fluorescence spectra of LaCoO 3 recorded
Despite the relative success of Anderson impurity at different excitation energies near Co 2p thresholds [29].
model calculations of RIXS in CoO and NiO, it is Excitation energies are indicated by arrows on the absorption
spectrum shown on top.
not really clear to what extent this approach may be
valid for even more covalent Co and Ni systems,
where 3d states have higher degree of delocalization. La 0.1 Sr 0.9 CoO 3 and NdNiO 3 , the interpretation of
In fact, for highly covalent compounds, such as these spectra as mostly consisting of a normal
La 12x Sr x CoO 3 and NdNiO 3 , only small changes in emission contribution, which would be a projection
the shape of Co and Ni L2,3 X-ray fluorescence of the partial density of states, can be challenged by
spectra with varying excitation energies across the their Anderson impurity model description as spectra
corresponding 2p absorption edges are observed (see of compounds with negative D. In the latter case, the
Figs. 1113). Although relatively high resolution ground and low-lying excited states will mainly have
was used to measure these data (total resolution was the 3d n11L ] character so that main spectral structures
about 0.7 eV for both Co and Ni compounds), the may behave in a similar manner as charge-transfer
spectra do not show many structures and look quite satellites in the RIXS spectra of CoO and NiO, i.e.
broad. The energy position of the main peak has reveal no shifts on the emitted photon energy scale
almost no dependence on the excitation energy tuned for some range of incident photon energies. Detailed
above or at the 2p threshold. However, at least for calculations of spectral shapes using both band
224 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233

Fig. 12. Co L2,3 X-ray fluorescence spectra of La 0.1 Sr 0.9 CoO 3

recorded at different excitation energies near Co 2p thresholds
[29]. Excitation energies are indicated by arrows on the absorption Fig. 13. Ni L3 X-ray fluorescence spectra of NdNiO 3 recorded at
spectrum shown on top. different excitation energies near the Ni 2p5 / 2 threshold [29].
Excitation energies are indicated by arrows on the absorption
spectrum shown in the top panel.
theory and quasi-atomic approach are needed to
clarify the role of partial 3d localization and dd
correlation in these systems. vide information about J-mixing in the ground state
through studies of intra-atomic f f excitations.
The symmetry in a solid is not spherical, the
angular momentum of f electrons is not conserved
3. Lanthanide compounds and therefore J is not a good quantum number.
However, the crystal field is an order of magnitude
Since the nature of the ground state defines smaller than the spinorbit splitting and can be
various physical properties of a system in question, treated as a perturbation. In the case of weak
knowing the ground state character is a key issue, hybridization effects, the interlevel coupling and
especially in materials science. It turns out that for consequently J-mixing in the ground state of the
rare-earth compounds the RIXS technique can pro- system are often disregarded in the interpretation of
 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233 225

experimental data by applying a pure atomic approx- obtained at the Dy 4d threshold of DyF 3 at room
imation (mainly for high-energy spectroscopies) or temperature. Measurements at the 4d threshold of
by using a first order crystal-field theory where the rare-earths provide naturally higher resolution than
crystal field interaction is assumed to act only within those at the 3d threshold, thus allowing one to study
the separate J manifolds. This is partly due to elementary excitations in greater detail (see e.g. Ref.
complications in extracting information about the [41]). Experimental spectra of DyF 3 are displayed in
ground state J-mixing directly from the data. For Figs. 14 and 15 on both photon-energy and energy-
example, the estimation of the J-mixing degree in loss scales. Two distinct groups of pronounced
high-order crystal-field theory by adjusting the crys- inelastic-scattering peaks are observed in these spec-
tal-field parameters from the fit of optical absorption tra. The first group is distinguished by small energy
or low-energy electron-energy-loss spectra [38,39] losses on the tail of the elastic line, whereas the
may result in a large uncertainty originating from second is characterized by energy losses more than
difficulties calculating the intensities of dipole-for-
bidden transitions. In turn, the possible influence of
the weak metalligand hybridization is difficult to
analyze quantitatively in the absence of so-called
charge-transfer satellites in high-energy spectro-
scopic data.
In this situation, the use of alternate spectroscopic
means to obtain ground-state J-mixing information is
essential. Recently, Finazzi et al. [40] have shown
that this mixing can be studied by taking advantage
of dichroic properties of rare-earth 3d X-ray absorp-
tion. However, the method is limited to magnetically
ordered systems. Here, we discuss a potential of
resonant X-ray scattering spectroscopy in studying of
the ground-state J-mixing when applied to com-
pounds without distinct long-range magnetic order
and significant metalligand hybridization.
Similar to optical absorption and EELS with
respect to probing the optical scale excitations, RIXS
at the same time provides an additional level of the
transition selectivity due to the element specificity
and dipole selection rules. In contrast to systems
with the strong metalligand hybridization where the
charge-transfer process leads to an appearance of
intense lines in RIXS spectra as a result of inter-ionic
excitations, J-mixing in systems with weak hybridi-
zation effects is expected to manifest itself in an
intensity gain of some intra-ionic ( f f ) transitions
which are disallowed for the pure Hund-rule ground
state. In other words, transitions with DJ other than
0, 61, and 62 are probed in the resonant excitation
deexcitation process. Although J is not a good
quantum number in the J-mixing case, we use this
Fig. 14. The total electron yield spectrum at the Dy 4d edge and
terminology for simplicity. resonant X-ray scattering spectra of DyF 3 normalized to the
A discussion about the RIXS potential to probe the incident photon flux [42]. The letters correspond to the excitation
ground-state J-mixing is based on analysis of data energies indicated in the absorption spectrum.
226 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233

ably well reproduced by atomic multiplet calcula-

tions for the Dy 31 ion [44]. The results of calcula-
tions show that the dominant elastic peak in all of the
RIXS spectra of DyF 3 is to large extent a conse-
quence of strong interference effects in the inter-
mediate state of the coherent second-order optical
process. The states constituting the main 4d absorp-
tion edge have a lifetime broadening of about 2 eV
largely because of the 4d 2 4f 4f CosterKronig
decay. However, a close inspection of experimental
RIXS spectra shows that there are some spectral
structures which are not revealed in calculations
within the pure atomic approximation. Thus, the
feature with the energy loss of about 1.17 eV is
observed in spectra h, k, and l, presented in detail by
Fig. 15. While atomic multiplet theory predicts non-
zero intensities for resonant inelastic X-ray scattering
transitions to the 6 H13 / 2 , 6 H11 / 2 , and 6 F11 / 2 sextets of
the 4f 9 configuration (the Hund rules ground state is
6
H15 / 2 ), the energy of the extra-feature in experimen-
tal spectra h, k, and l (Fig. 15) is close to those of
6
F9 / 2 and 6 H7 / 2 manifolds of Dy 31 in LaF 3 [45].
This is an indication of J-mixing and the presence of
J 5 13 / 2 and J 5 11 / 2 components in the ground
state of DyF 3 . Indications of other extra-structures
missing from atomic multiplet calculations can be
Fig. 15. Enlarged inelastic X-ray scattering part of the resonant
Dy 4f 4d spectra from DyF 3 together with optical absorption seen in the energy loss range between 1.2 and 2.0 eV,
transitions of Dy 31 in LaF 3 at 4.2 K [45]. For optical transitions, as in spectra l and m.
the most intense lines in each group of Stark levels belonging to To simulate the effect of J-mixing, we calculated
each J are all normalized to the same value. The eight clearly transition intensities within the pure atomic approxi-
separated manifolds with energies under 2 eV are assigned to
6 mation by choosing 6 H13 / 2 and 6 H11 / 2 as initial
H13 / 2 , 6 H11 / 2 , ( 6 H9 / 2 , 6 F11 / 2 ), ( 6 H7 / 2 , 6 F9 / 2 ), 6 H5 / 2 , 6 F7 / 2 , 6 F5 / 2 , and
6
F3 / 2 (in order of the increasing transition energy), respectively. states for the scattering process. The excitation
energy was taken to be the same as that for ex-
perimental spectrum l (since the intensity of extra-
2.5 eV. When the excitation energy approaches the structures is higher for excitation energies at the
main broad maximum of the Dy 4d absorption edge, main absorption edge) and the core-hole lifetime
the first group still possesses significant intensity broadening was set to 2.0 eV. According to these
while the structures of the second group become calculations, the spectral weight at an energy loss of
relatively faint. Regarding the energy scale on which about 1.17 eV (transitions to 6 F9 / 2 and 6 H7 / 2 ) should
the spectral variations occur, the observed transitions constitute 56% and 217% of that at 1.0 eV (transi-
can be attributed to intra-ionic f f excitations. The tions to 6 F11 / 2 ) for 6 H13 / 2 and 6 H11 / 2 as initial states,
energy gap between two groups of inelastic X-ray respectively. Although, accounting for weak but
scattering structures reflects the separation between finite Dy 4f F 2p hybridization may give rise to
sextets and quartets of trivalent Dy [21,43] which some changes in calculated relative intensities in
can be reached due to the excitationdeexcitation both cases, it is clear that an admixture of the
process. J 5 11 / 2 component in the ground state results in a
Indeed, the overall spectral shapes and their stronger effect than that for the J 5 13 / 2 component.
behavior with varying excitation energies are reason- Since the degree of J-mixing in DyF 3 is relatively
 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233 227

low, the J 5 11 / 2 contribution to the ground states is structures is rather a combined effect of the crystal-
expected to be comparable with the J 5 13 / 2 contri- field interaction and Dy 4f F 2p hybridization.
bution in order to explain the noticeable weight of However, the calculations which would take into
forbidden transitions in the inelastic scattering account both the crystal field and F 2p Dy 4f
spectra in Fig. 15. This is not unusual. For example, charge transfer excitations are complicated by a huge
the J 5 11 / 2 component has been found to be number of multiplets and require large computational
comparable to the J 5 13 / 2 component in the ground resources. At present, they are out of the scope of the
state of Dy 31 -doped yttrium scandium gallium gar- paper.
net [39] as a result of the crystal-field interaction. The existence of J-mixing in the intermediate state
To estimate the effect of this interaction on the raises a question about how strongly the inelastic-
shape of RIXS spectra, we also performed model scattering intensity at energy losses between 1.0 eV
crystal-field multiplet calculations for the Dy 31 ion and 2.0 eV is related to J-mixing in the ground state
in the crystal field of Oh symmetry with the strength of DyF 3 . To estimate this, crystal-field multiplet
of 35 meV. Fig. 16 shows the 1.02.2 eV energy-loss calculations with the crystal-field interaction
region of the spectra calculated using a pure atomic switched off in the intermediate state were per-
approximation and crystal field multiplet theory. It is formed. Thus, any J-mixing in the core-excited state
clear that switching on the crystal field gives rise to was disallowed. A comparison of the results of
additional transitions. The calculated intensities are calculations with and without crystal-field interaction
too low to fully account for the observed spectral in the intermediate state (Fig. 16) shows no signifi-
weight in experimental data at the corresponding cant changes in the inelastic-scattering intensities of
energy loss. This suggests that inter-atomic coupling forbidden structures on switching off J-mixing in
is also important for the description of the inelastic- the core-excited state. One of the main reasons for
scattering profile in the energy loss range between that is a large core-hole lifetime broadening. As a
1.0 and 2.0 eV and that the appearance of additional whole, the calculations indicate that the spectral
weight in the energy loss region between 1.0 and 2.0
eV is largely determined by J-mixing in the ground
state of DyF 3 .

4. Actinide compounds

The usual perspective on the actinides is that their

5f electrons are more localized than 3d electrons of
transition elements but less localized than 4f elec-
trons of lanthanides. Charge-transfer effects are
expected to be significant in actinide compounds as a
result of metal 5f 2 ligand 2p hybridization. Indeed,
the analysis of our data obtained at U 3d 5 / 2 threshold
shows that the ligand 2p U 5f charge-transfer
plays an important role in uranium compounds, such
as UO 2 , UO 2 (NO 3 ) 2 ? 6H 2 O, and even in UF 4 .
Fig. 16. A 1.02.2 eV energy-loss region of inelastic X-ray The most extensive studies of the electronic
scattering spectra of the Dy 31 system calculated using a pure structure of these aforementioned compounds have
atomic approach (intermediate coupling) and crystal field multi- been carried out for UO 2 [46,47]). Molecular-orbital
plet theory (Oh symmetry). In the calculations, Slater integrals calculations by several research groups [4851] gave
F k (5f,5f ), F k (5d,5f ), and G k (5d,5f ) were scaled down to 80%,
75%, and 66%, respectively, from the HartreeFock values. An values for the 5f occupancy, which range from 2.3 to
excitation energy was set to the main absorption maximum 2.9 electrons, while this occupancy was estimated at
(spectrum l in Fig. 14) and G was taken to be 2.0 eV. about 2.3 electrons from the analysis of X-ray
228 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233

absorption and photoemission data within an Ander- about 3570 and 3586 eV for UO 2 were earlier
son impurity model [5254]. These results indicate assigned to multiple-scattering resonances) [5961].
significant degree of covalency for UO chemical While for UO 2 (NO 3 ) 2 ? 6H 2 O, the structure observed
bonds in UO 2 . For UF 4 , a 5f contribution of | 0.3 at about 4 eV above the main absorption maximum
electrons to the bonding orbitals was also predicted was suggested to represent a charge-transfer satellite
from relativistic DiracSlater local-density calcula- [59] (other structures at about 10 eV and 32 eV above
tions [55]. For compounds containing U 61 , the the main absorption maximum were attributed to
degree of covalency for metal 2 ligand bonds is multiple scattering resonances), for UO 2 and UF 4 ,
expected to be even higher than that for U 41 where charge-transfer effects are less pronounced, an
systems. For example, molecular-orbital calculations identification of possible charge-transfer satellites is
yielded the 5f occupancy of | 2.6 electrons for the hampered due to the substantial smearing out of the
uranyl ion UO 212 [56,57]. Although, the values for spectral structures. In this situation, the virtually
the 5f occupancy obtained from molecular-orbital unlimited resolution (defined by the response func-
calculations seem to be overestimated [58] one can tion of the instrument) of the RIXS technique and its
not rule out the importance of the U 5f ligand 2p ability to enhance transitions to charge-transfer ex-
hybridization even in a compound with ionic bonds cited states are especially useful.
such as UF 4 . The U 5f 3d X-ray fluorescence spectra of UO 2
As discussed above, one of the consequences of (7p 3d transition probability is much lower) de-
high covalency and hybridization in the ground state tected in the horizontal plane at 908 angle between
is an appearance of charge-transfer satellites in the directions of incident and scattered photons and for
high-energy spectroscopic data. For actinides, the 3d different excitation energies across the U M5 absorp-
core-hole lifetime broadening is quite large, thus tion edge are displayed in Fig. 18. One can identify
reducing the efficiency of the X-ray absorption contributions from scattering and normal fluores-
technique. As a result, the U 3d 5 / 2 X-ray absorption cence in these spectra. The scattering part follows
spectra of UF 4 , UO 2 , and UO 2 (NO 3 ) 2 ? 6H 2 O, dis- varying excitation energies while the normal fluores-
played in Fig. 17, do not exhibit many sharp cence part appears at constant emitted-photon ener-
features. In particular, spectra of UF 4 and UO 2 gies. For excitation energies set near the U 3d 5 / 2
appear as a single line with some asymmetry on the threshold, the spectra consist of the recombination
high-energy side (weak and broad structures in the line and low-energy structure, extending over 10 eV
continuum at about 3576 and 3590 eV for UF 4 and at (structures present at about 19 eV below the recombi-
nation peak correspond to U 6p3 / 2 3d 5 / 2 transi-
tions).
The shape of resonant spectra can not be attributed
solely to the 5f 2 3d 9 5f 3 5f 2 excitationde-exci-
tation process. The 3d 9 5f 3 5f 2 multiplet spread is
about 4 eV while the separation between centroids of
the recombination line and low-energy structure is
approximately 6.5 eV. Due to significant U 5f 2 O 2p
hybridization, the ground state of UO 2 can be
described as a mixture of primarily 5f 2 and 5f 3] L
configurations. Then, the intermediate state of the
spectroscopic process is mainly a mixture of 3d 9 5f 3
and 3d 9 5f 4L
] configurations so that there is a radia-
tive decay to 5f 2 and 5f 3L] states, i.e. transitions back
to the ground state and to low-lying excited states.
Final states of this second order optical process can
Fig. 17. Total electron yield spectra of UF 4 , UO 2 , and be divided into three categories: bonding (the re-
UO 2 (NO 3 ) 2 ? 6H 2 O near the U M5 absorption edge. combination line), nonbonding, and antibonding (the
 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233 229

into account the configuration-dependent hybridiza-

tion [36], the 8-eV enhancement in the scattering
spectrum can be tentatively associated with transi-
tions to anti-bonding states between 5f 2 and 5f 3] L
configurations. Setting V to 1.3 eV (due to higher
localization of 5f wavefunctions, bare V in this oxide
is expected to be lower than the one derived for
UO 2 (NO 3 ) 2 ? 6H 2 O, see below) we then estimate D
( 5 f 2 p ) in UO 2 to be about 6 eV. This value is
larger than 4.5 eV derived for the on-site f f
Coulomb interaction Uff from photoemission and
bremsstrahlung isochromat data [54,63], thus sug-
gesting that UO 2 is a MottHubbard-type insulator
[2]. This conclusion is consistent with those made
from the analysis of both resonant and non-resonant
4f photoemission spectra [5254].
The low-energy structure that corresponds to
transitions involving ligand 2p U 5f charge-trans-
fer excitations is also present in the RIXS spectra of
UF 4 (Fig. 19) obtained at the excitation energy set to
the U M5 absorption maximum and to 5 eV above it.
The situation for UF 4 contrasts to that for CeF 3
where charge-transfer effects are not observed in the
RIXS data recorded at Ce 3d thresholds. The shape
of the UF 4 spectra in Fig. 19 thus concurs with the
common expectation that 5f states in actinides are
more extended than 4f states in lanthanides. The
relative intensity of the charge-transfer structure in
the RIXS spectra of UF 4 is lower than that in UO 2
whereas the energy separation between centroids of
Fig. 18. The total electron yield spectrum of UO 2 near the U M5
this structure and the recombination line, which is
absorption edge and resonant U 5f 3d X-ray fluorescence
spectra. The arrows on the total electron yield spectrum indicate mainly determined by values of D and V in the
the excitation energies used to collect the fluorescence spectra. ground state, increases up to | 9 eV. Since V in UF 4
is rather close to that in UO 2 , an increase of this
energy separation observed for uranium fluoride is
low-energy structure) states between 5f 2 and 5f 3L mainly due to an increase of D.
]
configurations. The whole low-energy structure Relative intensities of structures in scattering
grows slightly and then decreases with increasing spectra depend not only on values of model parame-
excitation energies, showing a hint of some enhance- ters in the ground state of the system but also on the
ment at about 8 eV below the recombination line channel interference and relationship between on-site
for 3558.1-eV incident photons. f f Coulomb interaction Uff and the attractive core-
The energy separation between the slightly re- hole potential Ucf in the intermediate state of the
sonating structure and the recombination line is spectroscopic process. For the spectral intensity,
close to that between the 8.5-eV satellite and 1-eV given by formula (2), the interference effects lead to
main line in resonant valence band photoemission an enhancement or suppression of transitions to a
spectra of UO 2 [62]) which have been attributed to certain final state due to a summation over i inside
antibonding and bonding states between 5f 1 and the modulus. Since Ucf is usually somewhat larger
5f 2L than Uff , the effective charge-transfer energy Deff 5
] configurations, respectively [52,53]. Taking
230 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233

changes in the shape of resonant X-ray scattering

spectra with varying excitation energies (Fig. 20).
For the excitation energy set to the U 3d 5 / 2 absorp-
tion satellite at 3557.5 eV, one can observe an
enhancement in the inelastic scattering weight at an
energy loss of about 9.5 eV (spectrum c). Similar
enhancement was detected in the RIXS spectra of
UO 3 (Fig. 2) at corresponding excitation energies
although, in the present case, this resonance is more

Fig. 19. Resonant X-ray scattering spectra of UF 4 recorded at

excitation energies set to the main U M5 absorption maximum and
to 5 eV above it together with those of CeF 3 obtained at both Ce
3d 5 / 2 and 3d 3 / 2 thresholds [13].

D 2 Ucf 1 Uff decreases in the intermediate state,

thus giving rise to stronger configuration mixing. On
the other hand, V is shown to be reduced in the
presence of a core hole [36]. This leads to a
weakening of the U 5f 2 ligand 2p hybridization. In
addition, the relative intensities of scattering struc-
tures may be affected by self-absorption. Therefore,
the difference in relative intensities of the recombi-
nation line and the charge-transfer structure between
UO 2 and UF 4 may not necessarily be determined by
the difference in the 5f 3L ] admixture in the ground
state of these compounds.
In UO 2 (NO 3 ) 2 ? 6H 2 O, a degree of covalent
character of the chemical bonds is high because U is
in the highest oxidation state 61. The admixture of
Fig. 20. Resonant X-ray scattering spectra of UO 2 (NO 3 ) 2 ? 6H 2 O
the 5f 2L 2
] configuration is expected to be significant recorded at various energies of incident photons near the U 3d 5 / 2
so that the ground state of the system can be threshold. The arrows on the total electron yield spectrum of the U
described as a mixture of 5f 0 , 5f 1L, 2 2
] and 5f L ] M5 absorption edge, shown in the top panel, indicate the excitation
configurations. As a consequence, there are large energies used for scattering measurements.
 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233 231

pronounced. The resonance indicates the charge- distance. The values of D may also be different for
transfer character of the absorption satellite at about inequivalent UO bonds.
4 eV above the U 3d 5 / 2 maximum. Referring to the The determination of the energies of transitions to
discussion for UO 3 [9], the elastic peak and struc- bonding, nonbonding and antibonding states between
tures with energy losses of | 5.1 eV and | 9.5 eV in 5f 0 , 5f 1L, 2 2
] and 5f L ] configurations from resonances
the scattering spectra of UO 2 (NO 3 ) 2 ? 6H 2 O can be in scattering spectra puts additional constrains on
associated with transitions to bonding, nonbonding, values of V, D, and Uff in the ground state of
0 1
and antibonding states between 5f and 5f L ] con- UO 2 (NO 3 ) 2 ? 6H 2 O. Neglecting the inequivalence of
figurations, respectively. UO bonds, model parameters can be estimated by
For UO 3 [9], a resonance of transitions to non- diagonalizing a simplified Hamiltonian so that its
bonding states between 5f 1L 2 2
] and 5f ] L configura- eigenvalues coincide with energies of corresponding
tions was also observed in scattering spectra at an states. This gives 1.4, 3.5, and 4 eV for V, D, and Uff ,
energy loss of about 14.5 eV when the excitation respectively. The derived average values suggest that
energy was set to the U 3d 5 / 2 absorption satellite UO 2 (NO 3 ) 2 ? 6H 2 O is in the intermediate regime of
| 10 eV above the main maximum (see also Fig. 2). the ZaanenSawatzkyAllen diagram [2].
This in turn supported the assignment of the latter RIXS measurements at the actinide 5d threshold
satellite to the one originating from the O 2p U 5f provide an opportunity to study in detail elementary
charge-transfer. For UO 2 (NO 3 ) 2 ? 6H 2 O, spectrum f excitations in actinide compounds due to the natu-
recorded at similar excitation energy (3564.2 eV rally higher resolution of such experiments in com-
versus 3563.9 eV for UO 3 ) exhibits a broad line with parison with those at the actinide 3d and 4d thres-
energy losses of around 15.4 eV. However, the origin holds. An example of probing the f f excitations in
of this line is not clear because of uncertainty in the actinide systems is illustrated in Fig. 21 where the
energy of normal fluorescence transitions and their RIXS spectra of solid UF 4 , recorded for different
relative contribution to spectrum f (unfortunately, the incident photon energies in the pre-5d-threshold
high-energy excited spectra, where normal fluores- region, are displayed. The assignment of sharp
cence dominates, were not recorded). The broad line inelastic scattering structures to the f f transitions is
can belong to normal fluorescence, or it can corre- supported by atomic multiplet calculations for the
spond to a resonance of charge-transfer excited states U 41 ion. The spectra were calculated using Eq. (2),
as a result of coupling between 5f 1L 2 2
] and 5f L ] where the varying lifetime of core-excited states due
configurations. The possibility of some contribution to the auto-ionization via the 5d5f 5f super Coster
of transitions to the latter states is suggested by the Kronig decay was taken into account. The auto-
shape of other spectra of UO 2 (NO 3 ) 2 ? 6H 2 O re- ionization into the continuum of g symmetry was
corded at lower excitation energies. For example, only considered since it is the most dominant path.
spectra c and e (Fig. 20) contain structures with Matrix elements were obtained from Cowans pro-
similar energy losses to those of the broad line in grams so that Slater integrals F k (5f,5f ), F k (5d,5f ),
spectrum f. G k (5d,5f ), and R k (5de g,5f ) were scaled down to
Some differences in the behavior of RIXS spectra 75%, 75%, 66%, and 80%, respectively, from the
between UO 2 (NO 3 ) 2 ? 6H 2 O and UO 3 which both HartreeFock values. The density of states of the
contain U 61 are due to somewhat different environ- continuum was assumed to be constant and the
ment for U in these compounds. In UO 2 (NO 3 ) 2 ? kinetic energy of the continuum electron was set to
6H 2 O, the U ion is surrounded by eight O ions the value which made the average energies of 5d 9 5f 3
[64,65] which create two short and six long UO and 5d 10 5f 1 e g equal.
bonds of 1.76 and 2.48 A, respectively. In UO 3 , U The calculations reproduce all of the spectral
has six nearest O neighbors [66] with two of them structures very well especially an enhancement of the
located at the 1.79-A distance and others at 2.30 A. peak at about 1.2 eV with increasing excitation
This strong inequivalence of O sites implies a large energies. The growth of the peak is due to enhanced
variation in the value of V for the same compound transitions into the 1 G4 state. Changes in absolute
since V is expected to scale with the cationanion intensities of inelastic scattering structures corre-
232 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233

peak contains some contribution of diffuse scattering

which may vary with varying excitation energies.
RIXS profiles, corresponding to the f f excita-
tions, are found to be very sensitive to the chemical
state of U in different systems [67]. For example, it
is a matter of the presence or absence of these
excitations when going from U 41 to U 61 com-
pounds. Even for the same oxidation state, the
corresponding RIXS structures are observed to be
broadened in compounds with the increasing degree
of covalency in chemical bonding. Therefore, RIXS
measurements near the U 5d threshold provide good
fingerprints for the chemical state of U in different
systems in contrast to X-ray absorption spectra
which show only small differences at the U 5d edge
[60,67].

Acknowledgements

Much of the experimental work reviewed here has

been carried out in collaboration with Prof. Joseph
Nordgren, Dr. Akane Agui, Dr. Laurent Duda, Dr.
Jinghua Guo, Dr. Pieter Kuiper, Dr. Yanjun Ma, Dr.
Martin Magnuson, Dr. Derrick Mancini, Dr. Ken
Miyano, Dr. David Shuh, and Conny Sathe. The
author is also grateful to Dr. Frank de Groot for
making available TT-MULTIPLETS programs and for his
help in using those. This work was supported by the
Swedish Natural Science Research Council (NFR)

and the Goran Gustavsson Foundation for Research
Fig. 21. Resonant X-ray scattering spectra of UF 4 [67] recorded in Natural Sciences and Medicine.
at different excitation energies close to the U 5d threshold (lines
with markers) together with the results of atomic multiplet
calculations (sticks with thin lines) for the U 41 ion. Excitation References
energies are indicated by arrows on the total electron yield
spectrum at the U 5d absorption edge shown in the top panel.
[1] P.W. Anderson, Phys. Rev. B 124 (1961) 41.
[2] J. Zaanen, G.A. Sawatzky, J.W. Allen, Phys. Rev. Lett. 55
(1985) 418.
[3] O. Gunnarsson, O. Jepsen, Phys. Rev. B 38 (1988) 3568.
sponding to the f f transitions are reproduced on
[4] O. Gunnarsson, O.K. Andersen, O. Jepsen, J. Zaanen, Phys.
going from spectrum a to spectrum b. For spectra c Rev. B 39 (1989) 1708.
and d, such changes in calculated intensities are
[5] O. Gunnarsson, K. Schonhammer, Phys. Rev. B 40 (1989)
about three times higher as compared to those in the 4160.
experiment. The discrepancy may originate from the [6] O. Gunnarsson, N.E. Christensen, Phys. Rev. B 42 (1990)
2363.
normalization procedure for the experimental spectra

[7] P. Kuiper, J.-H. Guo, C. Sathe, L.-C. Duda, J. Nodgren,
to account for variations in the incident photon flux. J.J.M. Pothuizen et al., Phys. Rev. Lett. 80 (1998) 5204.
The intensity of the elastic peak was used as a [8] Y. Ma, N. Wassdahl, P. Skytt, J. Guo, J. Nordgren, P.D.
reference in this procedure. However, the elastic Johnson et al., Phys. Rev. Lett. 69 (1992) 2598.
 S.M. Butorin / Journal of Electron Spectroscopy and Related Phenomena 110 111 (2000) 213 233 233

[9] S.M. Butorin, D.C. Mancini, J.-H. Guo, N. Wassdahl, J. [40] M. Finazzi, F.M.F. de Groot, A.-M. Dias, B. Kierren, F.
Nordgren, M. Nakazawa et al., Phys. Rev. Lett. 77 (1996) Bertran, Ph. Sainctavit et al., Phys. Rev. Lett. 75 (1995)
574. 4654.
[10] B.T. Thole, G. vand der Laan, M. Fabrizio, Phys. Rev. B 50 [41] A. Moewes, T. Eskildsen, D.L. Ederer, J. Wang, J. McGuire,
(1994) 11466. T.A. Callcott, Phys. Rev. B 57 (1998) R8059.
[11] S. Tanaka, A. Kotani, J. Phys. Soc. Jpn. 62 (1993) 464. [42] S.M. Butorin, J.-H. Guo, D. Shuh, J. Nordgren, in: ALS
[12] S.M. Butorin, J.-H. Guo, M. Magnuson, P. Kuiper, J. Compendium of User Abstracts and Technical Reports 1997,
Nordgren, Phys. Rev. B. 54 (1996) 4405. LBNL, University of California, Berkeley, 1998, p. 143.
[13] S.M. Butorin, D.-C. Mancini, J.-H. Guo, N. Wassdahl, J. [43] W.T. Carnall, P.R. Fields, K. Rajnak, J. Chem. Phys. 49
Nordgren, J. Alloys Comp. 225 (1995) 230. (1968) 4424.
[14] A. Kotani, (this issue). [44] A. Kotani, in: Proceedings of the Second International
[15] S.M. Butorin, J.-H. Guo, M. Magnuson, J. Nordgren, Phys. Conference On Synchrotron Radiation in Materials Science,
Rev. B 55 (1997) 4242. Kobe, 1998, Technical Report of ISSP, Vol. Ser. A No. 3456,
[16] S.M. Butorin, L.-C. Duda, J.-H. Guo, N. Wassdahl, J. 1999, in press.
Nordgren, M. Nakazawa, A. Kotani, J. Phys. Condens. [45] J.L. Fry, H.H. Caspers, H.E. Rast, S.A. Miller, J. Chem.
Matter 9 (1997) 8155. Phys. 48 (1968) 2342.
[17] S.M. Butorin, M. Magnuson, K. Ivanov, D.K. Shuh, T. [46] J.W. Allen, R.Z. Bachrach (Eds.), Synchrotron Radiation
Takahashi, S. Kunii et al., J. Electr. Spectrosc. 101103 Research: Advances in Surface and Interface Studies, Vol. 1,
(1999) 783. Plenum Press, New York, 1992, p. 253.
[18] B. Fromme, U. Brunokowski, E. Kisker, Phys. Rev. B 58 [47] J.R. Naegele, in: A. Goldmann (Ed.), Electronic Structure of
(1998) 9783.
Solids: Photoemission and Related Data, Landolt-Bornstein
[19] B.T. Thole, G. van der Laan, Phys. Rev. B 38 (1988) 3158. New Series, Solid State Phys, Vol. 23B, Springer-Verlag,
[20] R.D. Cowan, The Theory of Atomic Structure and Spectra, 1994, p. 183.
University of California Press, Berkeley, 1981. [48] V.A. Gubanov, A. Rosen, D.E. Ellis, Solid State Commun.
[21] D. van der Marel, G.A. Sawatzky, Phys. Rev. B 37 (1988) 22 (1977) 219.
10674. [49] V. Heera, G. Seifert, P. Ziesche, Phys. Stat. Sol. (b) 118
[22] J. van Elp, R.H. Potze, H. Eskes, R. Berger, G.A. Sawatzky, (1983) K107.
Phys. Rev. B 44 (1991) 1530. [50] D.E. Ellis, G.L. Goodman, Int. J. Quant. Chem. 25 (1984)
[23] L.-C. Duda, (this issue). 185.
[24] L.-C. Duda, J. Nordgren, G. Drager, S. Bocharov, T. [51] G.L. Goodman, J. Alloys Comp. 181 (1992) 33.
Kirchner, (this issue).
[52] O. Gunnarsson, D.D. Sarma, F.U. Hillebrecht, K. Schonham-
[25] C. F. Hague, J.-M. Mariot, L. Journel, J.-J. Gallet, A. mer, J. Appl. Phys. 63 (1988) 3676.
Rogalev, G. Krill et al., (this issue). [53] O. Gunnarsson, T.C. Li, Phys. Rev. B 36 (1987) 9488.
[26] P.H. Butler, Point Group Symmetry Applications: Methods [54] A. Kotani, H. Ogasawara, Physica C 186188 (1993) 16.
and Tables, Plenum Press, New York, 1981. [55] K. Pierloot, A. Reinders, G.L. Goodman, D. Devoghel, C.
[27] B.T. Thole, G. van der Laan, P.H. Butler, Chem. Phys. Lett.
Gorller-Walrand, L.G. Vanquickenborne, J. Chem. Phys. 94
149 (1988) 295. (1991) 2928.
[28] Z. Yi-Yang, Y. Chun-Hao, Phys. Rev. B 47 (1993) 5451. [56] P.F. Walsh, D.E. Ellis, J. Chem. Phys. 65 (1976) 2387.
[29] S. M. Butorin, M. Magnuson, C. Sathe, A. Agui, T. [57] J.H. Wood, M. Boring, S.B. Woodruff, J. Chem. Phys. 74

Kaambre, J.-H. Guo, et al., unpublished results. (1981) 5225, and references therein.
[30] K. Okada, A. Kotani, J. Phys. Soc. Jpn. 61 (1992) 449. [58] L.E. Cox, J. Electr. Spectrosc. 26 (1982) 167.
[31] F.M.F. de Groot, J. Electr. Spectrosc. 92 (1998) 207. [59] J. Petiau, G. Calas, D. Petitmaire, A. Bianconi, M. Benfatto,
[32] S. Tanaka, Y. Kayanuma, A. Kotani, J. Phys. Soc. Jpn. 59 A. Marcelli, Phys. Rev. B 34 (1986) 7350.
(1990) 1488. [60] G. Kalkowski, G. Kaindl, W.D. Brewer, W. Krone, Phys.
[33] D. Alders, J. Vodel, C. Levelut, S.D. Peacor, T. Hibma, M. Rev. B 35 (1987) 2667.
Sacchi et al., Europhys. Lett. 32 (1995) 259. [61] J. Guo, D.E. Ellis, E. Alp, L. Soderholm, G.K. Shenoy, Phys.
[34] F.M.F. de Groot, P. Kuiper, G.A. Sawatzky, Phys. Rev. B 57 Rev. B 39 (1989) 6125.
(1998) 14584. [62] L.E. Cox, W.P. Ellis, R.D. Cowan, J.W. Allen, S.-J. Oh, I.
[35] J. Zaanen, G.A. Sawatzky, Can. J. Phys. 65 (1987) 1262. Lindau et al., Phys. Rev. B 35 (1987) 5761.
[36] M.A. van Veenendaal, D. Alders, G.A. Sawatzky, Phys. Rev. [63] Y. Baer, J. Shoenes, Solid State Commun. 33 (1980) 885.
B 51 (1995) 13966. [64] D. Hall, A.D. Rae, T.N. Waters, Acta Crystallogr. 19 (1965)
[37] P. Kuiper, G. Kruizinga, J. Chijsen, G.A. Sawatzky, H. 389.
Verweij, Phys. Rev. Lett. 62 (1989) 221.
[65] J.C. Taylor, M.H. Muller, Acta Crystallogr. 19 (1965) 536.
[38] J.D. Axe, G.H. Dieke, J. Chem. Phys. 37 (1962) 2364. [66] S. Siegel, H.R. Hoekstra, Inorg. Nucl. Chem. Lett. 7 (1971)
[39] M.D. Seltzer, A.O. Wright, C.A. Morrison, D.E. Wortman, 455.
J.B. Gruber, E.D. Filer, J. Phys. Chem. Solids 57 (1996) [67] D.K. Shuh, K.E. Ivanov, S.M. Butorin, J.-H. Guo, M.
1175. Magnuson, J. Nordgren, et al., unpublished.