X-ray absorption spectroscopy of Na_ {x} CoO_ {2} layered cobaltates

Abstract

Measurements of polarization and temperature dependent soft x-ray absorption have been performed on NaxCoO2 single crystals with x=0.4 and x=0.6. They show a deviation of the local trigonal symmetry of the CoO6 octahedra, which is temperature independent in a temperature range between 25 K and 372 K. This deviation was found to be different for Co3+ and Co4+ sites. With the help of a cluster calculation we are able to interpret the Co L23-edge absorption spectrum and find a doping dependent energy splitting between the t2g and the eg levels (10 Dq) in NaxCoO2.

Full text

Page 1

arXiv:cond-mat/0606518v1 [cond-mat.str-el] 20 Jun 2006
X-ray absorption spectroscopy on layered cobaltates NaxCoO2
T. Kroll1,∗M. Knupfer1, J. Geck1, C. Hess1, T. Schwieger1, G.
Krabbes1, C. Sekar1, D.R. Batchelor2, H. Berger3, and B. B¨ uchner1
1IFW Dresden, P.O. Box 270016, D-01171 Dresden, Germany
2Universit¨ at W¨ urzburg, Am Hubland, D-97074 W¨ urzburg, Germany and
3Institute of Physics of Complex Matter, EPFL, CH-1015 Lausanne, Switzerland
Measurements of polarization and temperature dependent soft x-ray absorption have been per-
formed on NaxCoO2 single crystals with x=0.4 and x=0.6. They show a deviation of the local
trigonal symmetry of the CoO6 octahedra, which is temperature independent in a temperature
range between 25 K and 372 K. This deviation was found to be different for Co3+and Co4+sites.
With the help of a cluster calculation we are able to interpret the Co L23–edge absorption spectrum
and find a doping dependent energy splitting between the t2g and the eg levels (10Dq) in NaxCoO2.
PACS numbers:
I.INTRODUCTION
The discovery of an unexpectedly large thermopower
accompanied by low resistivity and low thermal conduc-
tivity in NaxCoO2raised significant research interest in
these materials [1] and lead to a number of experimental
and theoretical investigations [2, 3, 4, 5, 6]. This interest
has strongly been reinforced by the discovery of supercon-
ductivity in the hydrated compound Na0.35CoO2·1.3H2O
in 2003 [7], and thus NaxCoO2 experiences an again
increasing attention [8, 9, 10, 11, 12, 13, 14, 15, 16,
17, 18, 19, 20, 21, 22, 23, 24].
Na cobaltates to the high temperature superconductors
(HTSC) - both are transition metal oxides and adopt
a layered crystal structure with quasi two dimensional
(Cu,Co)O2layers - is an important aspect of the research
activities. In contrast to the HTSC cuprates however,
the CoO2layers consist of edge sharing CoO6octahedra
which are distorted in a way that the resulting symme-
try is trigonal. The trigonal coordination of the Co–sites
results in geometric frustration which favors unconven-
tional electronic ground states. The geometrically frus-
trated CoO2–sublattice also exists in the non–hydrated
parent compound NaxCoO2, which has been investigated
in this work. The intercalation of water into the parent
compound is expected to have little effect on the Fermi
surface beside the increase in two–dimensionality due to
the effect of expansion [25, 26].
Upon lowering the symmetry from cubic to trigonal,
the t2gstates split into states with eπ
which can be represented as
The similarity of the
gand a1gsymmetry,
eπ
g±= ∓1
√3[|xy? + exp±i2π/3|yz? + exp±i4π/3|xz?] (1)
and
a1g=
1
√3[|xy? + |yz? + |xz?].
(2)
∗Electronic address: t.kroll@ifw-dresden.de
eg
egπ
egσ
t2g
a1g
FIG. 1: Illustration of the splitting of the energy levels for
cubic and trigonal symmetry. Here due to the trigonal distor-
tion the eπ
gstates have been assumed to be lowest in energy.
The cubic egstates remain degenerate and will be named
eσ
gin the following in order to avoid confusion. As has
been predicted by calculations [15] and shown experimen-
tally in Ref. 16, the eπ
g-states are lower in energy and are
therefore filled, while the a1g-states are partially filled
as a function of x. Therefore, the a1g–states govern the
relevant low energy excitations.
From magnetization, specific heat, and conductivity
measurements for various doping levels, three phase tran-
sitions as a function of temperature have been observed:
a low temperature bulk antiferromagnetic transition at
TN≈ 20 K for 0.75 ≤ x ≤ 0.9 [5, 19, 20] and two high
temperature transitions at about 280 K in a Na0.82CoO2
sample [17] and 323–340 K in samples with x ≈ 0.75
[27, 28]. The origin of the 280 K transition has been
discussed in terms of charge ordering and the formation
of magnetopolarons due to a charge induced Co 3d spin
state transition from a low spin state to an intermedi-
ate spin state [17]. Such a spin state transition occurs
for instance in the related compounds La1−ySryCoO3
[30, 31, 32, 33]. In some samples, a third transition at
around 340 K is observed, and it was suggested that this
is due to a structural transition involving Na ordering
[27, 28, 29], which is supported by Huang et al. who
found a structural transition concomitant with a shift of
a large fraction of the Na ions from a high–symmetry
position to a lower–symmetry position [28].
In this article, we present measurements of the near
edge x–ray absorption fine structure (NEXAFS) of

Page 2

2
NaxCoO2 (x = 0.4 and x = 0.6) which have been car-
ried out in order to investigate the electronic properties
of this interesting class of materials.
II. EXPERIMENTAL
The used single crystals were grown by two different
methods. First, single crystals of Na0.75CoO2were grown
using the travelling solvent floating–zone method, crys-
tals with a lower Na concentration were produced using
de–intercalation with Bromium from the highly doped
samples. The initial characterization of the samples has
been carried out using energy dispersive x-ray analysis
(EDX), x–ray diffraction and chemical analysis. Details
of the crystal growth, de–intercalation, and characteriza-
tion of the resulting samples will be presented elsewhere
[34]. Second, high–quality single crystals were grown by
the sodium chloride flux method as thoroughly described
in Ref. 35. Both methods lead to the same spectra.
The NEXAFS measurements of the absorption coef-
ficient were performed at the UE52-PGM beamline at
the synchrotron facility BESSY II, Berlin, analyzing the
drain current. The energy resolution was set to 0.09 eV
and 0.16 eV for photon energies of 530 eV and 780 eV,
respectively. We performed measurements on different
non–hydrated single crystals with a sodium content of
x = 0.4 and x = 0.6 at various temperatures and differ-
ent polarizations of the incident synchrotron light at the
oxygen K- and cobalt L2,3–edge. The two doping levels
that have been investigated belong to two different and
interesting regions in the phase diagram [24], x=0.4 lies
in the region of a paramagnetic metal whereas and x=0.6
is a Curie-Weiss metal. All the crystals were of the same
size of about 3x3 mm2area and 1 mm thickness. Crys-
tals were freshly cleaved in–situ under ultra–high vacuum
conditions (about 2·10−10mbar) at 25 K, which resulted
in shiny sample surfaces.
Special attention has been paid to the reproducibility
of the experimental data and the effects of surface con-
taminations. The freshly cleaved surfaces turned out to
be very sensitive to adsorbates. We observed irreversible
changes in the O K–spectra when the sample surfaces
were exposed to pressures above 2 · 10−9mbar which we
attribute to adsorbed, oxygen containing molecules at
the surface. No such changes have been found for the
Co absorption edges. A comparison to the temperature
dependent behavior of the Co L2,3–edge proves that the
origin of this irreversible effect is due to surface contam-
ination and not due to a change or a transition of the
whole sample, since the Co L2,3–edge remains unchanged
with temperature.
According to the dipole selection rules the O K and Co
L2,3 excitations as probed by these experiments, corre-
spond to core electron transitions into unoccupied oxy-
gen 2p and cobalt 3d electronic states. Upon variation
of the incident light polarization, different O 2p and Co
3d orbitals can be probed [36]. For polarization depen-
dent measurements, the samples were oriented such that
the direction of incident photons and the sample surface
normal (i.e. the c–axis) enclose an angle of α = 70◦. The
used undulator allows a rotation of the beam polarization
by 90◦using the vertical and horizontal mode. This pro-
cedure avoids experimental artifacts related to the differ-
ences in the optical path and the probed area. All NEX-
AFS results referred to E parallel to the c–axis (E||c) are
corrected using the formula I||c=
where I⊥cand I are measured NEXAFS intensities with
E ⊥ c and E in the plane defined by the c–axis and the
incident photon beam, respectively. In order to compare
the data, the NEXAFS spectra are normalized at higher
energies where the absorption is doping independent and
isotropic and the spectra for different measurements and
settings should show the same intensities. The spectra
are normalized at 600 eV for the O K–edge and at 810
eV for the Co L2,3–edge.
1
sin2(α)(I − I⊥ccos2α)
III. RESULTS AND DISCUSSION
Using NEXAFS, the unoccupied energy levels close to
the Fermi level can be studied by excitations from core
electrons into unoccupied states. We have measured exci-
tations from O 1s core levels into unoccupied O 2p states
that are hybridized with states of primary Co and Na
character, as well as excitations from Co 2p into Co 3d
states. If the influence of the core hole is neglected, as
is reasonable for the O 1s core hole excitations, a direct
interpretation of the NEXAFS results in terms of the par-
tial unoccupied density of states is possible, analogous to
a one electron addition process [37, 38]. If the core hole
cannot be neglected, as is the case for Co 3d excitations,
the interpretation of the data requires consideration of
multiplet splitting, hybridization and crystal field effects.
A. Co L-edge
In Fig. 2 experimental spectra of the Co L2,3–edges
are shown, which display three main features at each
edge: one strong central peak (peak b and b’) with a
shoulder towards higher energies (peak c and c’) and
a peak/shoulder towards lower energies (peak a and a’)
(Fig. 2). This result can easily be compared to the sim-
ilar compound LiCoO2 which nominally contains only
Co3+ions with S=0 [39]. In the NEXAFS spectra of the
Co L2,3–edge of LiCoO2 one finds only one main peak
[40] different from the spectrum observed for the mixed
valence system NaxCoO2. Especially the low energy fea-
ture (peak a and a’) is absent in LiCoO2, consequently
we assign this peak to be caused by excitations into unoc-
cupied 3d states of nominal Co4+ions which are missing
in LiCoO2. This interpretation is furthermore supported
by the doping dependence of the L–edge as the low en-
ergy peak appears stronger for lower sodium doping (i.e.
higher Co4+concentration) and weaker for higher sodium

Page 3

3
?
?
?
?
?
?
?
?
??
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
??
?
?
?
?
?
?
?
?
?
?
?
?
?
?
??
?
?
?
?
?
?
?
?
FIG. 2:
NEXAFS spectra of the Co L2,3–edge for NaxCoO2 with
x=0.4 (top) and x=0.6 (middle) at 25 K. ∆Ex represents the
difference in energy between the first and the main peak. The
insets show an enlargement of the region around the first peak
(ii) and around the top of the main peak (iii). Bottom: No
significant temperature dependence is observable between 25
and 372 K for x=0.6. All intensities are normalized at 810 eV
where no stoichiometric, polarization, or temperature depen-
dence is observable.
Top and middle figure: Polarization dependence
doping (i.e. lower Co4+concentration), as well as by cal-
culations as described below.
The local electronic structure around a Co atom in
NaxCoO2has been modelled in a cluster calculation us-
ing many–body wave functions. Within this approach, a
CoO6cluster containing the Co 3d and the O 2p valence
electrons has been solved exactly including all interac-
tions between 3d electrons in Ohsymmetry with a ground
state for Co4+as given in equation 2 [41]. For simplicity
we will use in the following the expression Co4+, referring
to a (distorted) CoO6octahedra with a formal Co4+cen-
tral ion containing five holes and, analogously, Co3+for
an octahedra containing four holes with a formal Co3+
central ion. It has been found from the calculations that
the first peak at lower energies in the NEXAFS Co L–
edge (peak a and a’ in Fig. 2) originates from excitations
into Co4+final states with an A1gsymmetry, while the
main peak (b and b’) and the shoulder (c and c’) are due
to excitations into final states of Co3+with Egsymme-
try and Co4+with T1gand T2gsymmetry, respectively.
Note that in order to avoid confusion we labeled the fi-
nal states by using capital letters (e.g. A1g) while the
ground is labeled with lower case letter (e.g. a1g). The
ground state of the system has been found to be strongly
covalent with a moderate positive charge transfer energy
∆CT= E(dn+1L) − E(dn) for Co3+and a negative ∆CT
for Co4+[41].
In the experimental spectrum of the Co L–edge one ad-
ditionally observes that the energy difference between the
first peak (A1g) and the largest peak (Eg) in Fig. 2 dif-
fers between the two different stoichiometries x=0.4 and
x=0.6 being ∆Ex=0.4 = −2.0 eV and ∆Ex=0.6 = −1.8
eV. Theoretically, this behavior can be best explained by
a change of the energy splitting between the t2g states
and the eg states (10Dq). An explanation for a lower
10Dq for higher sodium intercalation can be found from
neutron powder diffraction by Huang et al. An increasing
Co–O distance with increasing sodium content is found
[28] which guides a lower influence of the crystal field and
charge transfer, i.e. a lower 10Dq.
The a1g orbital of the ground state in trigonal sym-
metry points along the (1 1 1) direction of the distorted
CoO6octahedra, i.e. parallel to the crystal c–axis, while
the the two eπ
gorbitals point perpendicular to it. From
polarization dependent measurements with the E vector
of photons parallel and 70◦to the crystal c axis one ob-
serves the a1g peak of Co4+to be stronger for E||c as
compared to E ⊥ c. This behavior is expected for a local
trigonal distortion, where the t2gground states split into
states with a1gand eπ
remain untouched (c.f. Fig. 1) and therefore should not
show a strong polarization dependence. Different from
that, the intensity of the Co3+central peak is signifi-
cantly larger for E ⊥ c compared to E||c (Fig. 2). This
effect points to an additional distortion that splits the eσ
levels. Such a splitting might be caused by two mecha-
nisms. From spectral ellipsometry on a Na0.82CoO2sam-
ple, Bernhard et al. find a transition at 280 K which they
gsymmetry, whereas the eσ
gstates
g

Page 4

4
?
?
?
?
?
?
??
?
?
?
FIG. 3: Stoichiometric dependence of NaxCoO2 at the Oxy-
gen K-edge. A, B, and C represent the positions of the three
main peaks. The intensities are normalized at 600 eV.
explain by the formation of magnetopolarons using the
idea of a spin–state transition [17] similar to the related
compound La1−ySryCoO3[30, 31, 32, 33]. In NaxCoO2,
this mechanism would be driven by a displacement of
the neighboring oxygen ligands towards the central Co4+
ion which may favor an intermediate–spin (IS) state with
S=1 over a low–spin (LS) state with S=0 of the Co3+ions
[17]. This displacement would result in a splitting of the
eσ
glevels. Another possibility for an additional distortion
could arise from the effect of sodium ordering at special
doping levels [28, 29, 42, 43], it is assumed that this re-
sults in orthorombic symmetry so that a perturbation of
the trigonal distortion of the octahedra could occur.
Both possible contributions are related to a corre-
sponding ordering temperature, the formation of magne-
topolarons is observed at 280 K, whereas the sodium or-
dering appears at temperatures below 350 K for x=0.75.
However, no significant temperature dependence for
Na0.6CoO2 has been found between 25 K and 370 K
in the present NEXAFS studies. This implies that the
spin–state of Co as well as the electronic structure in
this energy range as seen by NEXAFS is not affected by
temperature neither by a spin state transition nor by a
structural transition involving Na ordering (Fig. 2 bot-
tom). A temperature independent distortion would be
expected if the non–trigonal distortion is purely struc-
tural.
B.O K-edge
In Figure 3, we present the results for the O K absorp-
tion edge of Na0.4CoO2 and Na0.6CoO2 measured with
light polarized parallel to the crystallographic c axis. For
both stoichiometries we observed three pronounced fea-
tures A, B and C above the absorption threshold, show-
?
?
?
?
?
?
??
?
?
?
?
?
?
?
?
?
??
?
?
?
FIG. 4: NEXAFS polarization dependence of the oxygen K-
edge of NaxCoO2 for x=0.6 (top) and x=0.4 (bottom). Po-
larization E||c is indicated by filled symbols, E||c is indicated
by open symbols. The intensities are normalized at 600 eV
where the spectrum is isotropic.
ing a significant doping dependence. Features A and B
increase for smaller x, while feature C decreases, i.e. fea-
tures A and B increase with an increasing hole doping. In
the related compound LiCoO2the situation is very simi-
lar to Na1CoO2, since both systems are assumed to have
a low–spin state meaning that all six t2gstates are occu-
pied by electrons while the four egstates are empty. The
resulting absorption spectrum shows only one peak due
to O 2p orbitals hybridized with the Co 3d orbitals with
egsymmetry [40]. In NaxCoO2, excitations into unoccu-
pied Co3+states should therefore only be responsible for
a single peak in the whole O K-absorption edge spectra of
NaxCoO2. We therefore attribute the first two features
A and B in Fig. 3 to doping induced states related to the
formation of Co4+sites and feature C to the formation
of of Co3+sites, surrounded by oxygen octahedra. It has
been shown previously that the pre–edge peaks in the O
K NEXAFS spectra of the late transition–metal (TM)

Page 5

5
oxides are shifted by about 1 eV to lower energies when
the TM valence increases by 1 [44]; therefore, spectral
features from two different valence states can be well re-
solved. The energetic downshift can be explained by a
decrease of the energy between the partially filled Co 3d
states; the valence electrons will be screened by 1 eV more
with every added valence electron, resulting in a situa-
tion that for late transition metals a higher valency corre-
sponds to a lower excitation energy. Upon hole addition
(decreasing x), the increase of features A and B, and the
decrease of feature C indicate that both are related to the
doping process and it is natural to ascribe them to excita-
tions into unoccupied O 2p states hybridized with Co3+
eσ
gstates (feature C) and, about 1 eV lower in energy, to
excitations into unoccupied O 2p states hybridized with
Co4+eσ
gstates (feature B) and those hybridized with
Co4+a1g states (feature A). Additionally, these results
show that the holes in NaxCoO2have a significant oxygen
character, which is in good agreement with other cobalt
based compounds [32, 40, 45].
Next we turn to the polarization dependence of the
O K absorption edges as shown in Fig. 4. From po-
larization dependent absorption measurements, informa-
tion about the orientation of the corresponding orbitals
are obtained. Our data signal that the doping induced
absorption feature A is strong for E||c, and substantially
weaker for E ⊥ c. Consequently, from the orientation of
the a1g–orbital and the attribution of the three peaks as
described above, the holes doped into the CoO2 layers
of NaxCoO2 have a predominant a1g character similar
to the result found by Wu et al. using x–ray absorp-
tion spectroscopy [16]. Band structure calculations find,
that although the a1gand eπ
some extend, the centers of these states are energetically
separated with those closer to EFhaving dominant a1g
character [4], in good agreement with our results. In ad-
dition, while feature B is not or only slightly polarization
dependent, the intensity of peak C (Co3+) is significantly
stronger for E ⊥ c than for E||c. The same result has
already been observed at the Co L–edge, but because
of the large central peak no quantitative analysis of the
polarization dependence of the Co4+shoulder could be
made. At the oxygen K–edge these two peaks are well
separated and a difference in the response due to differ-
ent polarizations is observable. From this we conclude
that the trigonal symmetry is better realised in octahe-
dra containing a Co4+central ion than in octahedra con-
taining a Co3+central ion. Such a situation has already
been suggested by Bernhard and coworkersas an origin of
magnetopolarons [17]. As already mentioned in section
IIIA, the underlying idea for such an effect is a lowering
of the local symmetry around the Co3+. Thus, the t2g
triplet and the egdoublet split and become polarization
dependent. The effect may be stronger at temperatures
below 20 K, which is also the critical temperature for a
magnetic transition to a bulk antiferromagnetic ordered
state [20, 24], at temperatures lower than 20 K the mo-
bility of the magnetopolarons is assumed to be lower due
gstates overlap, and mix to
?
?
?
?
?
?
??
?
?
?
?
?
?
?
??
?
?
FIG. 5: Stoichiometric and polarization dependence of the
NEXAFS spectra of NaxCoO2 at the oxygen K–edge. The
energy range right above the oxygen threshold is shown where
the the hybridization between O and Na can be monitored.
All curves are normalized at 600 eV.
to an increased self–trapping energy in the antiferromag-
netic state [17].
Somewhat higher in energy at E ≈ 535 eV, one finds
the excitations into unoccupied O levels which are hy-
bridized with Na orbitals [16]. As expected, the resulting
peaks increase in intensity with increasing sodium inter-
calation, but in addition they are strongly polarization
dependent. As is shown in Fig. 5 the intensity for?E||c is
stronger than for?E ⊥ c. This leads to a finite Na–O hy-
bridization along c, a consequence of this hybrid could be
a 3 dimensional magnetism as has been proposed by Jo-
hannes et al. [46]. From our data it becomes clear, that
the inter–planar binding is more likely to have a covalent
rather than an ionic character, so that a 3D magnetism
is reasonable.
IV.CONCLUSION AND SUMMARY
The study of single crystals with a stoichiometry
Na0.4CoO2and Na0.6CoO2reveal polarization dependen-
cies that cannot be explained by a simple trigonal dis-
tortion of the CoO6 octahedra. Taking the results of
both, the Co L2,3–edge and the O K–edge into account,
it follows that an additional distortion is present which
is stronger for octahedra with a formal Co3+central
ion than for octahedra with a formal Co4+central ion.
A possible explanation for such a phenomena has been
given by Bernhard et al. who find magnetopolarons in
Na0.82CoO2due to a lowering of the local symmetry [17].
Furthermore, we find doping dependent relative peak po-
sitions at the Co L2,3–edge which can be explained by a
doping dependent splitting of the t2g–eg levels (10Dq)
as has been found from cluster calculations and can be