Wohlfeld 2013

PHYSICAL REVIEW B 88, 195138 (2013)
Microscopic origin of spin-orbital separation in Sr2 CuO3
Krzysztof Wohlfeld,1,2 Satoshi Nishimoto,2 Maurits W. Haverkort,3,4,5 and Jeroen van den Brink2,6
1
Stanford Institute for Materials and Energy Sciences, SLAC National Laboratory and Stanford University, 2575 Sand Hill Road,
Menlo Park, California 94025, USA
2
Institute for Theoretical Solid State Physics, IFW Dresden, D-01069 Dresden, Germany
3
Max-Planck-Institut für Festkörperforschung, D-70569 Stuttgart, Germany
4
Department of Physics and Astronomy, University of British Columbia, Vancouver, Canada V6T-1Z1
5
Max Planck Institute for Chemical Physics of Solids, D-01187 Dresden, Germany
6
Department of Physics, TU Dresden, D-01062 Dresden, Germany
(Received 23 July 2013; published 20 November 2013)
A recently performed resonant inelastic x-ray scattering experiment (RIXS) at the copper L3 edge in the
quasi-one-dimensional Mott insulator Sr2 CuO3 has revealed a significant dispersion of a single-orbital excitation
(orbiton). This large and unexpected orbiton dispersion has been explained using the concept of spin-orbital
fractionalization in which the orbiton, which is intrinsically coupled to the spinon in this material, liberates
itself from the spinon due to the strictly one-dimensional nature of its motion. Here, we investigate this
mechanism in detail by (i) deriving the microscopic spin-orbital superexchange model from the charge-transfer
model for the CuO3 chains in Sr2 CuO3 , (ii) mapping the orbiton motion in the obtained spin-orbital model
into a problem of a single hole moving in an effective half-filled antiferromagnetic chain t-J model, and
(iii) solving the latter model using the exact diagonalization and obtaining the orbiton spectral function. Finally,
the RIXS cross section is calculated based on the obtained orbiton spectral function and compared with the RIXS
experiment.
DOI: 10.1103/PhysRevB.88.195138 PACS number(s): 75.25.Dk, 75.30.Ds, 71.10.Fd, 78.70.Ck
I. INTRODUCTION there were just three experimental indications of the existence

of mobile orbitons: (i) indirectly in the form of Davydov
Long and difficult “search”’for orbitons. A relatively well-
splittings23 in Cr2 O3 , (ii) more recently and also indirectly
understood problem in strongly correlated electron systems
in a pump-probe experiment in the doped manganite,24 and
concerns the propagation of collective magnetic (spin) excita-
(iii) in the RIXS spectrum on titanates, where a very small
tions in Mott insulators such as, e.g., three-dimensional (3D)
(w.r.t. the experimental resolution) dispersion was found.25
LaMnO3 , two-dimensional (2D) La2 CuO4 , ladder SrC2 O3 , and
From the theoretical side, the situation is also complex.
one-dimensional (1D) Sr2 CuO3 .1 The theoretically calculated
To understand the orbiton dispersion, one has to take into
dispersion of such magnetic excitations (magnons in 2D or account the interaction between orbitons and (i) the lattice
3D, triplons in the ladder, or spinons in 1D) agrees very well (phonons) and (ii) the spin degrees of freedom. Although
with the one measured using the inelastic neutron scattering2–4 the former has been investigated in several studies26–28 and
or the resonant inelastic x-ray scattering (RIXS).5–8 The for long “blamed” for causing a confinement of the orbiton
origin of this fact is the relative simplicity of the spin-spin motion,26–28 it turned out not to be of great importance in
interactions, which are usually modeled using the Heisenberg- the here discussed case of orbitons in Sr2 CuO3 .22 Therefore,
type spin Hamiltonians.1,2 The excitation spectrum of such while still far from being understood, the interaction with
Hamiltonians can then be obtained using, e.g., the linear the lattice will not be discussed in what follows. At the
spin-wave approximation in 2D/3D (Ref. 9) or the Bethe- same time, however, the spin-orbital interaction,29 which
ansatz–based approaches in 1D.9,10 stems from the inherent entanglement of the spin and orbital
This situation is very different when one considers prop- degrees of freedom30–33 in the Kugel-Khomskii superexchange
agation of the collective orbital interactions: the orbitons11 (and/or direct exchange)11,34 models, which describe the
(coined as such in Ref. 12). On the experimental side, this propagation of spin or orbital excitations,11 has a profound
originates from the lack of experimental probe to measure impact on the orbiton motion.13,17,35–39 Moreover, as already
orbiton dispersion.13–15 Even if neutrons do couple to the discussed in Refs. 40–48, and in direct relevance to the here
orbital excitations16 and can in principle detect orbital waves,17 discussed problem in Refs. 22 and 49 (see also below), in
this can not be easily realized experimentally.18 This is due to order to correctly describe the collective orbital excitations,
the usually low transfers of energy in the neutron scattering this interaction should not be treated on a mean-field level.
experiments w.r.t. the energies needed to trigger the orbital This latter feature of the spin-orbital interaction severely
excitations (for an exception, see Ref. 19). Inelastic light complicates matter and is one of the main motivations for
scattering in the form of (optical) Raman scattering can not the study presented in this paper.
transfer much momentum to orbital excitations, leading to Recent experimental and theoretical findings. This brief
controversial interpretations of the observed features.17,20,21 overview of the problems with finding mobile orbitons makes it
Only recently it has been proposed13,14 and then experimen- clear that the recent experimental finding of the mobile orbiton
tally and theoretically established22 that RIXS may be used in Sr2 CuO3 (Ref. 22) and its short theoretical description in
to probe the orbitons’ motion. Therefore, until last year, Refs. 22 and 49, signifies a breakthrough in the study of orbital
1098-0121/2013/88(19)/195138(21) 195138-1 ©2013 American Physical Society

WOHLFELD, NISHIMOTO, HAVERKORT, AND VAN DEN BRINK PHYSICAL REVIEW B 88, 195138 (2013)
excitations. We therefore briefly summarize these findings in Khomskii” spin-orbital model which describes the spin and
the following. orbital dynamics in Sr2 CuO3 , and thus defines the Hamiltonian
The RIXS measurements performed at copper L3 edge in that is used to calculate the orbiton spectral function. In
Sr2 CuO3 (Ref. 22) revealed two dispersive orbital excitations Sec. IV, we calculate the orbiton spectral function using the
(due to large crystal-field splitting also called dd excitations). newly developed concept of spin-orbital separation.49 Next,
First, the dxz orbital excitation, which (in the here used hole in Sec. V, we establish the relation between the RIXS cross
language) corresponds to a transfer of a hole from the ground section and the orbiton spectral function calculated in Sec. IV
state dx 2 −y 2 orbital to the excited dxz excitation, showed and compare the obtained RIXS spectra with those obtained in
a sinelike dispersion. This dispersion was of the order of the experiment.22 Finally, in Sec. VI we discuss the possible
200 meV, had a dominant π period component, and a large other scenarios which might explain the experimental results
incoherent spectrum which led to peculiar “oval”-like features presented in Ref. 22 and end with the concluding remarks.
in the RIXS spectrum (see Fig. 1 in Ref. 22). Second, also The paper is supplemented by three Appendices in which
the dxy orbital excitation had a small dispersion with visible (i) we discuss some details of the calculations performed in
π period component. Finally, the other two orbital excitations Sec. III A (Appendix A), (ii) we compare the results of Secs. IV
(the dyz and the d3z2 −r 2 orbital excitations) did not show any and V to those obtained using the linear orbital wave theory
significant dispersion. (Appendix B), and (iii) we compare the results of Sec. V
While these experimental results are the first unambiguous with those obtained assuming all orbital excitations to be
observation of an orbiton (cf. discussion above), they turned dispersionless (Appendix C).
out to constitute a challenge from a theoretical perspective.
It was shown22 that the above-mentioned particular features
of dispersion could only be explained if the concept of II. CHARGE-TRANSFER MODEL
spin-orbital separation was invoked and applied.49 In short, this
A. Hamiltonian
concept suggests that (i) the orbiton in Sr2 CuO3 is so strongly
coupled to the spin excitations (spinons in this 1D case) that As noted in Sec. I, the purpose of this study is to describe
its coherent motion can only be explained if this coupling the propagation of orbital excitation in the quasi-1D cuprate
was explicitly taken into account, (ii) during its motion the Sr2 CuO3 .50–52 Therefore, as our starting point, we take the
orbiton can nevertheless “liberate” from the spinon. This following multiband charge-transfer Hamiltonian (which is an
scenario can explain the reason why this orbiton dispersion extended version of the charge-transfer model discussed in
was not observed before: Since (on one hand) the spin-orbital Ref. 53):
separation phenomenon is rather unique to 1D and (on the other
H = H0 + H1 + H2 + H3 , (1)
hand) the experimental searches were constrained to mostly 2D
or 3D compounds, the orbiton was finally only observed when with H0 the tight-binding Hamiltonian written in a basis
the attention was turned into a purely 1D system. consisting of five 3d orbitals per Cu site and three 2p
Aim and plan of the paper. In this paper, we show how orbitals per O site. The many-body interactions are included
to apply the spin-orbital separation concept developed and in H1 , H2 , and H3 (onsite Coulomb interaction on Cu atoms,
discussed in Refs. 22 and 49 to the problem of the orbiton onsite Coulomb interaction on O atoms, and nearest-neighbor
motion in Sr2 CuO3 . We start from (Sec. II) the proper Coulomb interaction between electrons on Cu and O sites). In
charge-transfer model for Sr2 CuO3 supplemented by the terms second quantized form and using the hopping parameters as
which describe the dynamics of the excited orbitals. From indicated in Fig. 1, we obtain for the tight-binding Hamiltonian
this model, we derive in Sec. III the corresponding “Kugel- H0
† †
† †
H0 = −tσ (fiaσ fixσ − fi+1,aσ fixσ + H.c.) − tσ o (fiaσ fiyo+σ − fiaσ fiyo−σ + H.c.)
i,σ i,σ
† † † †
† †
− tπ (ficσ fiyσ − fi+1,cσ fiyσ + fibσ fizσ − fi+1,bσ fizσ + H.c.) − tπo (ficσ fixo+σ − ficσ fixo−σ + H.c.)
i,σ i,σ

+ x (nix − nia ) + y (niy − nic ) + z (niz − nib ) + yo (niyo+ − nia )
i i i i

+ xo (niyo− − nia ) + xo (nixo+ − nib ) + xo (nixo− − nib ) + εa nia + εb nib + εc nic ; (2)
i i i i i i
for the onsite Coulomb repulsion on copper sites term

αβ
αβ
αβ † †
H1 = U − JH niασ niβ σ̄ + U − 2JH niασ niβσ + U niα↑ niα↓ − JH fiασ fiασ̄ fiβ σ̄ fiβσ
i,σ,α<β i,σ,α<β i,α i,σ,α<β
αβ † †
+ JH fiα↑ fiα↓ fiβ↓ fiβ↑ ; (3)
i,α<β
195138-2
MICROSCOPIC ORIGIN OF SPIN-ORBITAL SEPARATION . . . PHYSICAL REVIEW B 88, 195138 (2013)
for the onsite Coulomb repulsion on oxygen sites the Hamiltonian

H2 = Up − JHμν niμσ niν σ̄ + Up − 2JHμν niμσ niνσ + Up niμ↑ niμ↓
i,σ,μ<ν i,σ,μ<ν i,μ
μν † †
μν † †
− JH fiμσ fiμσ̄ fiν σ̄ fiνσ + JH fiμ↑ fiμ↓ fiν↓ fiν↑ ; (4)
i,σ,μ<ν i,μ<ν
for the nearest-neighbor Coulomb repulsion

H3 = Vdp niα (niμ + niμ+ + niμ− + ni+1μ ). (5)
i,μα
Here, one needs to consider the following: (v) the structure of the dominant hopping elements follows
(i) the CuO3 chain is oriented along the x axis (Cu-Cu the Slater-Koster scheme [and was verified by our local
distance is set to 1) [cf. Fig. 1(a)]; for simplicity hole notation density approximation (LDA) calculations] and is depicted
is used; in Figs. 1(b)–1(d); the rather large hopping between oxygens
(ii) the charge-transfer model unit cell [cf. Fig. 1(a)] tpp (cf. Ref. 53) is neglected; although this may give rise
includes (a) one copper atom with three 3d orbitals: 3dx 2 −y 2 ≡ to a significantly smaller hole occupation on copper sites
a, 3dzx ≡ b, and 3dxy ≡ c, (b) one oxygen atom within (and consequently may reduce the intensity of the dispersive
the Cu-O-Cu-O- . . . chain with three 2p orbitals: 2px ≡ x, dd excitations), in the approach presented below it will not
2py ≡ y, 2pz ≡ z, (c) two equivalent oxygen atoms outside the contribute to the superexchange processes;
Cu-O-Cu-O- . . . chain with two 2p orbitals: above this chain – (vi) the charge-transfer energy μ is measured for the
2px ≡ xo+, 2py ≡ yo+ and below this chain – 2px ≡ xo−, particular 2p orbital from the relevant 3d orbital (in the here
2py ≡ yo−; used hole notation, see also above), i.e., from that 3d orbital
(iii) the copper orbital indices are α,β ∈ {a,b,c}, the chain which hybridizes with this particular 2p orbital;
oxygen orbital indices are μ,ν ∈ {x,y,z}, and the spin index (vii) due to crystal field there are distinct onsite energies εα
σ ∈ {↑,↓} (σ̄ = −σ ); for each 3d orbital;
(iv) fiκσ annihilates a hole at site i in orbital κ with spin (viii) the structure of the Coulomb interaction follows
σ while density operators are niκ = niκ↑ + niκ↓ with niκσ = Refs. 54 and 55 and up to two-orbital interaction terms exactly
† reproduces the correct onsite Coulomb interaction; note that
fiκσ fiκσ ;
the Coulomb interaction on oxygens above and below the chain
is not considered because we are interested merely in the Mott
(a) insulating case with one hole per copper site, and in the analysis
O O O O
that follows this particular Coulomb interaction plays only a
minor role.
Cu O Cu O Cu O Cu O We note at this point that the above model does not contain
the dyz ≡ d and d3z2 −r 2 ≡ e orbitals. This is because there
y will not be any sizable dispersion due to the very small
O O O O superexchange processes for the dd excitations involving these
x orbitals. Note further that (i) the hopping from the dyz orbital
Cu: a b c orbitals O: x y z orbitals to the neighboring oxygen along the chain direction x is
negligible, and (ii) the hopping from the d3z2 −r 2 to the px
orbital on the neighboring oxygen is particularly small in
(b) (c) (d) xo+
yo+ this compound (much smaller than tπ according to our LDA
tσ o tπ o calculations). We will therefore include these orbitals only
when calculating the RIXS cross section in Sec. V.
tσ tσ tπ tπ y tπ tπ y
x a x z z c
b
tσ o tπ o
B. Parameters
yo− xo−
In the model Hamiltonian (1), a large number of parameters
FIG. 1. The relevant atoms and orbitals that are taken into account appear, which need to be fixed in order to obtain quantitative
in the model, Eq. (1): (a) orientation of the CuO3 chain with one results that can be compared to experiment (cf. left column of
Cu site and three O sites in the unit cell (dotted line), (b) nearest- Table I).
neighbor O orbitals which hybridize with the Cu a orbital following In principle, we used the basic set of the parameters that
Eq. (2), (c) nearest-neighbor O orbitals which hybridize with the Cu b was proposed in Ref. 53. The only exception is the intersite
orbital following Eq. (2) (note that there is no hybridization between Coulomb repulsion Vdp which is set to a somewhat smaller
O orbitals that lie above/below the Cu-O-Cu-O- . . . chain), and (d) value of 1 eV than the exceptionally large one suggested in
nearest-neighbor O orbitals which hybridize with the Cu c orbital Ref. 53. Note, however, that the smaller value is still generally
following Eq. (2). accepted for the cuprates.56 Besides, this value leads to the
195138-3
TABLE I. Assumed values of the parameters used in the paper (see text for further details). All parameters except R, Rnm , and rnm (which
are dimensionless) are given in eV. Tilde before the value of the parameter denotes the fact that this precise value is not used in the analysis.
Model parameters
Charge-transfer model Spin-orbital model
(Sec. II) (Sec. III) Spectral function/RIXS parameters
tσ 1.5 J1 0.088 Local excitation energies

tπ 0.83 J2b 0.021 (Secs. IV, V, and Appendix C)
tσ o 1.8 J2c 0.010 Eb 2.15
b
tπ o 1.0 J12 0.043 Ec 1.41
c
x 3.0 J12 0.030 Ed 2.06
y 2.8 R 1.7 Ee 2.44
z 2.2 R1b 2.5 EAF 0.33
xo 2.5 R1c 2.3 Effective t-J models (Sec. IV)
yo 3.5 R2b 2.0 tb 0.084
U 8.8 R2c 1.9 tc 0.051
Up 4.4 r1b 1.7 J 0.24
JHb ≡ JHab 1.2 r1c 1.3 Linear orbital wave approximation
JHc ≡ JHac 0.69 r2b 1.2 (Appendix B)
p
JH ≡ JHμν 0.83 r2c 1.1 B 2.48
Vdp 1.0 ε̄a 0.0 C 1.74
εa 0.0 ε̄b ∼2.2 Jb −0.019
εb ∼0.5 ε̄c ∼2.0 Jc −0.014
εc ∼0.5
spin superexchange parameter that is equal to 0.24 eV (see III. DERIVATION OF THE SPIN-ORBITAL MODEL
below), which is the experimentally observed value.22 Finally,
Since the Coulomb repulsion U and the charge-transfer
we used the following values for parameters not considered in
energies μ present in model Eq. (1) are far larger than the
Ref. 53: (i) The Hund’s exchange JHc and JHb are calculated
p hoppings tn (tn U and tn μ where n = σ,π,σ o,π o)
using Slater integrals from Ref. 57, while JH is taken from
(cf. Table I), the ground state of H is a Mott insulator. This is
Ref. 56. (ii) εb and εc are estimated to be 0.5 eV from the
because, in the zeroth-order approximation in the perturbation
LDA calculations. (iii) While following Ref. 53, x = 3.0 eV
theory in hopping tn and in the regime of one hole per copper
and y0 = 3.5 eV, the values of the other charge-transfer
site, there is one hole localized in the a orbital at each copper
parameters are not given in this reference and have to be
site i. Similarly, when a single orbital excitation is made, then
obtained in another way. It seems reasonable to assume first
in the zeroth order the hole will be localized on a single copper
that values of xo , y , and z are roughly of the order of x .
site in a particular b or c orbital (because the charge-transfer
But, since the charge-transfer parameters are defined as equal
energy μ is always positive).
to the difference in energy between the particular hybridizing
In the second- and fourth-order perturbation theory in tn
c or b orbital and the particular 2p orbital, they have to
(the terms obtained from tn and tn3 perturbation vanish),58 the
be lower than x . Quantum chemical calculations suggest,
hole can delocalize which leads to a particular low-energy
however, that the actual values of y and z might still be
Hamiltonian: the spin-orbital Hamiltonian. This Hamiltonian
different: it occurs that the values of the onsite energies of
has the following generic structure:
the b and c orbitals are not identical and that the b orbital
has a higher energy than the c orbital by ca. 0.7 eV (cf.
H̄ = H̄0 + H̄a + H̄b + H̄c . (6)
Ref. 22). Altogether, this suggests the following values for
these two parameters: y = 2.8 eV and z = 2.2 eV. We
will show later that these values give the orbiton dispersion It consists of two kinds of terms: (i) H̄0 , which is a result of the
in reasonably good agreement with the RIXS experiment.22 second-order perturbation theory in tn , and (ii) H̄a + H̄b + H̄c
(iv) tπ and tπo are assumed to be roughly of the order of 55% terms which follow from the fourth-order perturbation theory
of tσ and tσ o (respectively).56 in tn and can be called “superexchange” terms. Note that
Note that Table I contains also a few other parameters which the latter terms can be classified in two classes: (i) H̄a , the
are later introduced in this paper. While they mostly follow so-called “standard” or “spin” superexchange terms, which
from the charge-transfer model parameters mentioned above, contribute when all holes are in the a orbitals (i.e., no orbital
we will comment on their origin once they become relevant in excitations are present), and (ii) H̄b + H̄c , the spin-orbital
the following sections. superexchange (Kugel-Khomskii–type) terms with one orbital
excitation present on one site of the bond (in the b or c orbital)
195138-4
(a) (b) (c)

tσ x tσ tσ tσ
a a
Cu O Cu Cu O Cu
Cu O Cu
a a
Cu O Cu tσ tσ tσ tσ
a a
Cu O Cu Cu O Cu
Cu O Cu
FIG. 2. (Color online) Schematic view of superexchange interactions when no orbital excitations are present, i.e., between spins (arrows)
of the holes in the a orbitals: the “initial” state [panel (a)] can be brought into the so-called “virtual” state of the superexchange process [circle
with two spins (arrows) on the right side of panel (b)] by the virtual hopping [left side of panel (b)] which can “decay” [right side of panel
(b)] via the virtual hopping into the “final” state of the superexchange [panel (c)]. Panel (b) shows two kinds of “virtual” states with doubly
occupied ions: on the oxygen (copper) on the lower (upper) side of the panel with the energy cost ∝ U (∝Up + 2), respectively. Panel (c)
shows two possible low-energy final configurations without double occupancies: without spin flip (i.e., identical to the initial state) and with
spin flip. Note that panel (b) shows relevant copper and oxygen orbitals in a schematic way, i.e., depicted by horizontal bars, while panels (a)
and (c) explicitly show the relevant copper orbitals (oxygen orbitals are not shown on these panels).
and no orbital excitation present on the other site of the bond. above the operators denotes the fact that we prohibit double
In the following sections, we discuss these terms step by step. occupancies in this low-energy Hamiltonian due to the large
onsite Hubbard U and Up .
A. Renormalization of onsite energies: H̄0 It is convenient to define at this point the orbital pseudospin
operators σ = 12 where
In the second-order perturbation theory in tn , the hole can
delocalize to the four neighboring oxygen sites surrounding † †
σiz = 12 (ñib − ñia ), σi+ = f˜bi fãi , σi− = fãi f˜bi , (8)
the copper sites forming bonding and antibonding states.
Although there are many important consequences of such
and τ = 1
where
tn2 processes, let us now just explore one of them which 2
actually turns out to be very important: the renormalization † †
of the onsite energies of the orbitals. In Appendix A, we τiz = 12 (ñic − ñia ), τi+ = f˜ci fãi , τi− = fãi f˜ci . (9)
discuss another, perhaps less important, consequence of these
Here, the tilde above the operators denotes the fact that double
processes: the renormalization of the hopping within the chain
occupancies are forbidden in this low-energy Hamiltonian due
due to hybridization with oxygen orbitals above and below the
to large onsite Coulomb repulsion U and Up . Setting ε̄a = 0,
chain (these renormalization factors are called λa and λc ).
we can rewrite Eq. (7) as follows:
When the hole delocalizes into the bonding and antibonding
states formed by the a, b, or c orbitals with the four surrounding 1 1
oxygen sites, the effective onsite energies of the orbital levels H̄0 = ε̄b + σi + ε̄c
z
+ τi .
z
(10)
2 2
are strongly renormalized with respect to the energy levels of i i
the pure a, b, or c orbitals. Although the proper calculation of
this phenomenon can be done analytically by diagonalizing a
five-level problem defined separately for each of the copper B. Spin superexchange: H̄a
α orbitals, we do not perform it here. Instead, we take the Let us first study the superexchange interactions when only
values obtained from the ligand field theory program59 based one type of orbital is occupied along the superexchange bond.
on the multiplet ligand field theory using Wannier orbitals on a In this case, it is very straightforward to show that the model (1)
CuO4 cluster. It occurs that, for realistic values of parameters can be easily reduced to the low-energy Heisenberg model for
of model (1) (see Table I), the antibonding states are well spins S = 12 using the perturbation theory to fourth order in tσ
separated from the bonding states and we can safely neglect (Ref. 58) (cf. Fig. 2):
the latter ones in the low-energy limit that is of interest here.
This leads to the following term in our spin-orbital model: 1
H̄a = J1 (1 + R) Pi,i+1 Si · Si+1 − , (11)
4
H̄0 = ε̄a ñia + ε̄b ñib + ε̄c ñic , (7) i
i i i
where Pi,i+1 denotes the fact that there are no orbital
where the values of the parameters ε̄α are shown in Table I. excitations present along the bond i,i + 1 and is defined
Here, we use the operators niα from Eq. (1), although a as
rigorous treatment would require the use of the operators
1 1 1 1
actually creating the particular bonding states centered around Pi,i+1 = + τiz
+ τi+1
z
+ σiz
+ σi+1 .
z
a copper α orbital at site i. We discuss in Appendix A why such 2 2 2 2
simplification is to a large extent justified. Besides, the tilde (12)
195138-5
Note that here the superexchange interactions involve not only in tn we obtain (cf. Figs. 3 and 4)
the spin degree of freedom S = 12 , b
3 b J1 + J2
† † H̄b = Si · Si+1 + R1 J12 + r1
b b
Siz = 12 (ñi↑ − ñi↓ ), Si+ = f˜i↑ f˜↓i , Si− = f˜i↓ f˜↑i , (13) i
4 2

but also the orbital degree of freedom pseudospin operators 1 R + r1 b + −
b b
defined in the previous section. Again, the tilde above × σiz σi+1 z
− + 1 J12 (σi σi+1 + σi− σi+1
+
)
4 2
1 b
the operators denotes the fact that double occupancies are
forbidden in this low-energy Hamiltonian due to large onsite b J1 + J2
+ − Si · Si+1 R2 J12 + r2
b b
Coulomb repulsion U and Up . i
4 2
The superexchange constant contains contributions due to
charge excitations on copper sites (∼J1 ) and on the oxygen 1 R2b + r2b b + − − +
× σi σi+1 −
z z
− J12 (σi σi+1 + σi σi+1 ) ,
sites located in-between the copper sites (∼J1 R), where 4 2
2 (16)
2t¯σ2 1
J1 = , (14) where the superexchange constant J1 is the one given by
x + Vdp U
Eq. (14) while
with t¯σ = λa tσ (see Appendix A for origin of the factor λa ) 2
and 2tπ2 1
J2b = (17)
2U z + Vdp U
R= . (15)
2x + Up and
Two remarks are in order here. First, when no orbital (2tπ t¯σ )2 1
excitations are present, the Hamiltonian (11) is equal to the
b
J12 = . (18)
(z + Vdp )(x + Vdp ) U
well-known spin-only Heisenberg model. This is in agreement
with the “common wisdom” stating that the orbital degrees of The complex structure of Hamiltonian (16) is a con-
freedom can be easily integrated out in systems with only sequence of the fact that the proper derivation of such a
one orbital occupied in the ground state. Second, in the low-energy model has to include the superexchange processes
above derivation we neglected intermediate states with 1A1 with four distinct intermediate states:
or 1E symmetry. In principle, superexchange processes which (i) The high-spin state 3 T1 on copper sites (middle bottom
involve these intermediate states should also be taken into panel of Fig. 3) which involves d 2 = b1 a 1 orbital configuration
account. However, due to the crystal-field splitting this would and with energy in terms of Racah parameters54 A − 5B ≡
mean that the final states of the superexchange process would U − 3JHb . This leads to r1b = 1−3J1b /U .
H
contain high-energy orbital excitations. Consequently, these (ii) The-low spin state 1 T1 on copper sites (middle bottom
processes are suppressed. panel of Fig. 4) which involves d 2 = b1 a 1 orbital configuration
and with energy in terms of Racah parameters54 A + B +
C. Spin-orbital superexchange for b orbital: H̄b 2C ≡ U − JHb . This leads to r2b = 1−J1b /U .
H
If along a bond there is one hole in the b orbital (due to, (iii) The high-spin state 3 T1 on oxygen sites (middle top
e.g., an orbital excitation created in RIXS) and another one in panel of Fig. 3; note that due to the equivalence between the
2
the a orbital, then using the perturbation theory to fourth order t2g and p2 configurations,54 we can label the multiplet states on
(a) (b) tπ tσ (c)

z tπ or
tσ x tσ
or
tπ
b a b
a
Cu O Cu Cu O Cu Cu O Cu
a b tσ tσ
or or
tσ tπ tπ
Cu O Cu tσ b a
Cu O Cu Cu O Cu
Cu O Cu
FIG. 3. (Color online) Schematic view of superexchange interactions with one orbital excitation (here: into orbital b) on one site and no
orbital excitations on the other site (hole in orbital a) in the case that both spins (arrows) on the neighboring orbitals are parallel: similarly as
in Fig. 2 the initial state [panel (a)] can be brought into the so-called virtual state of the superexchange process [circle with two spins (arrows)
on the right side of panel (b)] by the virtual hopping [left side of panel (b)] which can decay [right side of panel (b)] via the virtual hopping
into the final state of the superexchange [panel (c)]. Panel (b) shows two kinds of virtual states with doubly occupied ions, on the oxygen and
copper sites (cf. Fig. 2). Panel (c) shows two possible low-energy final configurations without double occupancies: without orbital flip (i.e.,
identical to the initial state) and with orbital flip.
195138-6
(c)
a b
Cu O Cu
tπ tσ
(a) (b) z tπ or or
tσ x tπ
tσ
b a b
a
Cu O Cu Cu O Cu Cu O Cu
a b tσ tσ
or or
tσ tπ
Cu O Cu tσ tπ
b a
Cu O Cu Cu O Cu
Cu O Cu
b a
Cu O Cu
FIG. 4. (Color online) Schematic view of superexchange interactions with one orbital excitation (here: into orbital b) on one site and no
orbital excitations on the other site (hole in orbital a) in the case that spins (arrows) on the neighboring orbitals are antiparallel: similarly as
in Fig. 2 the initial state [panel (a)] can be brought into the so-called virtual state of the superexchange process [circle with two spins (arrows)
on the right side of panel (b)] by the virtual hopping [left side of panel (b)] which can decay [right side of panel (b)] via the virtual hopping
into the final state of the superexchange [panel (c)]. Panel (b) shows two kinds of virtual states with doubly occupied ions, on the oxygen and
copper sites (cf. Fig. 2). Panel (c) shows four possible low-energy final configurations without double occupancies: with/without orbital flip
and with/without spin flip.
the p shell by those known from the t2g sector) which involves where the superexchange constants are J1 [cf. Eq. (14)] and
p2 = x 1 z1 orbital configuration and with energy in terms of 2
p
Racah parameters54 Ao − 5Bo ≡ Up − 3JH (where o denotes 2t¯π2 1
J2 =
c
, (20)
the fact that the Racah parameters are for oxygen sites). This y + Vdp U
leads to R1b = + +U2U(1−3J p /U ) .
x z p H p with t¯π = λc tπ (see Appendix A for origin of the factor λc )
(iv) The low-spin state 1T2 on oxygen sites (middle top panel and
of Fig. 4) which involves p2 = x 1 z1 orbital configuration and
with energy in terms of Racah parameters54 Ao + Bo + 2Co ≡ (2t¯π t¯σ )2 1
p
c
J12 = . (21)
Up − JH . This leads to R2b = + +U2U(1−J p /U ) . (y + Vdp )(x + Vdp ) U
x z p H p
Similarly to the b orbital case discussed above, the complex

D. Spin-orbital superexchange for c orbital: H̄c
structure of Hamiltonian (19) is a consequence of the fact that
If along a bond there is one hole in the c orbital (due to, e.g., the proper derivation of such a low-energy model has to include
an orbital excitation created in RIXS) and another one in the the superexchange processes with four distinct intermediate
a orbital, then using the perturbation theory to fourth order in states:
tn we obtain (cf. Figs 3 and 4 showing an analogous situation (i) The high-spin state 3 T1 on copper sites which involves
in the case of the orbital superexchange between the b and a d = c1 a 1 orbital configuration and with energy in terms of
2
orbitals) Racah parameters54 A + 4B ≡ U − 3JHc . This leads to r1c =
1
3 J1 + J2c 1−3JHc /U
.
H̄c = Si · Si+1 + R1c J12
c
+ r1c (ii) The low-spin state 1 T1 on copper sites which involves
4 2
i
d 2 = c1 a 1 orbital configuration and with energy in terms of
1 R1c + r1c c + − − + Racah parameters54 A + 4B + 2C ≡ U − JHc . This leads to
× τi τi+1 −
z z
+ J12 (τi τi+1 + τi τi+1 )
4 2 r2c = 1−J1c /U .
1 H
J1 + J2c (iii) The high-spin state 3 T1 on oxygen sites which involves

+ − Si · Si+1 R2c J12
c
+ r2c p2 = x 1 y 1 orbital configuration and with energy in terms
4 2 p

i
of Racah parameters54 Ao − 5Bo ≡ Up − 3JH . This leads to
1 R c + r2c c + − R1 = + +U (1−3J p /U ) .
c 2U
× τiz τi+1z
− − 2 J12 (τi τi+1 + τi− τi+1
+
) , x y p H p
4 2 (iv) The low-spin state 1T2 on oxygen sites which involves
(19) p = x 1 y 1 orbital configuration and with energy in terms of
2
195138-7
p
Racah parameters54 Ao + Bo + 2Co ≡ Up − JH . This leads to Hamiltonian (6). Thus, we express the above formulas for
R2c = + +U2U(1−J p /U ) . orbiton spectral functions in terms of the orbital psuedospinon
x y p p
H
operators acting in the restricted Hilbert space of the spin-
orbital Hamiltonian without double occupancies60 :
E. Remarks on the derivation and parameters 1 1 †
Ab (k,ω) = lim Im 0̄|σk σ |0̄ , (24)
First, we would like to remark that the physics of superex- π η→0 ω + E0̄ − H̄ − iη k
change interactions is very similar in both “orbital exchange” 1 1 †
cases discussed above [cf. Eqs. (16) and (19)]. Thus, the main Ac (k,ω) = lim Im 0̄|τk τ |0̄ , (25)
π η→0 ω + E0̄ − H̄ − iη k
(quantitative) difference between these two cases originates in
slightly renormalized model parameters. Second, we should where |0̄ is the ground state of the spin-orbital Hamiltonian
comment on the superexchange paths which, due to their H̄ with energy E0̄ . It is now easy to verify that, for the realistic
small relative contribution to the low-energy Hamiltonian, regime of parameters defined in Table I, the ground state is
are neglected in the above derivation: (i) There is a finite insulating, ferro-orbital (FO), i.e., only orbital a is occupied,
1 1
probability that, e.g., the c1 a 1 (i.e., t2g eg ) configuration in the and antiferromagnetic (AF) (due to its 1D nature and lack of
3 2
intermediate state T1 decays into a t2g configuration. However, long-range order called “quantum” AF in what follows). This
such process can be neglected since this means that in the is because the energy cost of populating b or c orbital states ε̄b
final state of the superexchange process we would then be and ε̄c is much larger than hopping tn (cf. Table I). Thus, it is
left with a transition to a higher-energy sector: starting from only the spin Heisenberg Hamiltonian H̄a which dictates what
the initial state with one hole in a t2g configuration we would is the spin ground state (which is always AF for any positive
end with two holes in the t2g configuration in the final state J1 and R, cf. Table I).
of the superexchange process and this would cost the energy In the following sections, we calculate the orbiton spectral
∼ ε̄b or ∼ ε̄c ; the latter energies are typically much larger functions (24) and (25): first by mapping them onto the spectral
than the scales of the superexchange interactions. (ii) We also functions of the effective t-J model problems and then by
neglect intermediate states of the kind 1A1 or 1E or 3A2 since solving these simplified problems numerically. While this
they all require transitions to the higher-energy sector ∼ ε̄b or method is not entirely exact, it gives a far better approximation
∼ε̄c . (iii) Finally, from the above structure one can see that of the actual spectral function than the commonly used
it is impossible to have a “mixing” between the t2g orbital linear orbital wave approximation (cf. Appendix B 1 and
excitations, i.e., to have transitions between the states with, Appendix B 3).
e.g., ai1 ci+1
1
configuration and, e.g., bi1 ai+1
1
configuration; this
is due to (a) the flavor-conserving hoppings between the t2g A. Mapping onto the effective t- J models
orbitals, and (b) no onsite hopping [cf. Eq. (1)] allowing for
a transition from the c1 a 1 to b1 a 1 state (cf. Table A26 from Mapping for the b orbiton case. To address the issues
Ref. 54). mentioned above, we rewrite Eq. (24) in the following way:
Finally, let us note that all of the parameters of the spin-
1 1
orbital model directly follow from the charge-transfer model Ab (k,ω) = lim Im 0̄| eikj σj − Sjz
parameters. Their values are shown in Table I. π η→0 j
2

1 ikj † 1
× e σj − Sj |0̄
z
IV. ORBITON SPECTRAL FUNCTIONS

ω + E0̄ − H̄ − iη j 2

Our main purpose is to calculate the orbiton dispersion, 1 1
+ lim Im 0̄| eikj σj + Sjz
which follows from the two orbiton spectral functions π η→0 j
2
1
† 1 † 1
Ab (k,ω) ≡ lim Im 0| eikj fj aσ fj bσ × eikj σj + Sjz |0̄ .
π η→0 jσ ω + E0̄ − H̄ − iη j 2
1 †
× eikj fj bσ fj aσ |0 (22) (26)
ω + E0 − H − iη j σ
Here, we used an approximation that the spectral function
and for an orbiton is only nonzero when the spin of the hole in
1 † the excited orbital is conserved [i.e., we assumed that the
Ac (k,ω) ≡ lim Im 0| eikj fj aσ fj cσ spectral functions for orbiton, which contains the “cross terms”
π η→0 jσ of the kind ∝( 21 + Sjz )( 21 − Sjz ), can be neglected]. In fact,
1 † this approximation amounts to neglecting the process which
× eikj fj cσ fj aσ |0 , (23)
ω + E0 − H − iη j σ describes orbiton propagation with an additional spin flip (see
Appendix B 3 for justification that such process has relatively
where |0 is the ground state of the charge-transfer Hamiltonian small amplitude and can be neglected). Furthermore, due to the
H [Eq. (1)] with energy E0 . SU(2) spin invariance of both the ground state |0̄ and of the
In what follows, we will concentrate on the low-energy ver- Hamiltonian H̄, the two contributions to the spectral function
sion of the charge-transfer Hamiltonian, i.e., the spin-orbital [as written on the right-hand side of Eq. (26)] are equal, i.e.,
195138-8
we can write with

2 1 0
HJW = ε̄b
†
βi βi , (32)
Ab (k,ω) = lim Im 0̄| e σj − Sjz
ikj
π η→0 j
2

i

1 †
1 † 1
a
HJW = J1 (1 + R) (1 − niβ ) (α αi+1 + H.c.)
× eikj σj − Sjz |0̄ . i
2 i
ω + E0̄ − H̄ − iη j 2
1 1
(27) − niα − ni+1α + niα ni+1α (1 − ni+1β ) (33)
2 2
(with the pseudospinon operators originating in the projection
Next, we introduce fermions (to be called spinons) α
operators Pi,i+1 ) and
through the Jordan-Wigner (JW) transformation for spins and
fermions β (to be called pseudospinons) through the JW 1 b † †
transformation for pseudospins (cf. Ref. 49). We define
b
HJW = − R1b + r1b + R2b + r2b J12 (αi βi βi+1 αi+1 + H.c.)
4 i
⎛ ⎞
1 b
† − R1 + r1b J12 b †
(βi βi+1 + H.c.)
Sj+ = exp ⎝−iπ Qn ⎠ αj , 2 i
n=1,...,j −1 b
⎛ ⎞ b J1 + J2 †
− R1 J12 + r1
b b
βi βi , (34)
2
Sj− = αj exp ⎝iπ Qn ⎠ , (28) i
n=1,...,j −1 where in addition we assumed that only one pseudospinon
1 in the bulk is present (which corresponds to the FO ground
Sjz = nj α − , state with one orbital excitation) and we skipped the terms
2 p
niα ni+1β + ni+1α niβ , as for realistic values of JHb and JH (cf.
†
where Qn = αn αn and αn are fermions. Besides, we define the Table I) they are of the order of 10%–20% of the value of the
orbital fermionic operators β as hopping tb [see Eq. (37) and Table I].
⎛ ⎞ Next, we perform a transformation that connects the above-
†
derived Hamiltonian with the effective t-J model. Thus,
σj+ = exp ⎝−iπ Q̄n ⎠ βj , we introduce the auxiliary fermions p̃iσ acting in a Hilbert
n=1,...,j −1 space without double occupancies [we have checked that the
⎛ ⎞
operators p̃iσ fulfill the appropriate commutation rules (cf.

σj− = βj exp ⎝iπ Q̄n ⎠ , (29) Ref. 61)]:
⎛ ⎞
n=1,...,j −1
† †

1 p̃j ↑ = βj , p̃j ↓ = βj αj exp ⎝iπ Qn ⎠ , (35)
σjz = njβ − , n=1,...,j −1
2
† †
where Q̄n = βn βn and βn are fermions. where again Qn = αn αn . Besides, we introduce back spin
It turns out that when calculating the orbiton spectral operators S following Eq. (28). Thus, we obtain
function (27) with spins and pseudospins expressed in terms ab
Ht-J ≡ Ht-J
0
+ Ht-J
a
+ Ht-J b
of the JW fermions following the above transformation, a †
pseudospinon and spinon are not present on the same site = = −tb (p̃j σ p̃j +1σ + H.c.)
(cf. Ref. 49). In other words, we have a constraint j,σ
† † 1

∀i (βi βi + αi αi ) 1, (30) +J Sj Sj +1 − ñj ñj +1 − Ebh ñj , (36)
j
4 j
since otherwise the right-hand side of the spectral function
† † where the parameters are defined as
in Eq. (27) is zero because σj ( 12 − Sjz ) = βj (1 − nj α ). The
physical understanding of this phenomenon is as follows: 1 b
suppose one promotes a hole with spin down to the b orbital at tb ≡ 3R1b + R2b + 3r1b + r2b J12 , (37)
8
site i, which means that we have no pseudospinon and spinon
at this site. Now, this pseudospinon can move only via such J ≡ J1 (1 + R), (38)
processes which do not flip the spin of the hole in the b orbital,
J1 + J2b
i.e., we prohibit creating spinon and pseudospinon at the same Ebh ≡ ε̄b − R1b J12
b
+ r1b , (39)
site (cf. Appendix B 3). 2
Altogether, this means that while rewriting the low-energy and we furthermore neglected the difference between tb↓ ≡
Hamiltonians (10), (11), and (16) in terms of fermions α and β, 1
4
(R1b + R2b + r1b + r2b )J12
b
and tb↑ ≡ 12 (R1b + r1b )J12
b
hopping
we can skip all the terms which contain the pseudospinon and element. Note, however, that (i) this difference is of ca. 10% for
spinon at the same site. We arrive at the following Hamiltonian: realistic parameters from Table I and therefore can be neglected
to simplify the calculations, (ii) keeping this difference while at
ab
HJW ≡ HJW
0
+ HJW
a
+ HJW
b
(31) the same time neglecting the possibility of orbiton propagation
195138-9
with an additional spin flip (the so-called B1 process in

Appendix B 3) violates the SU(2) spin symmetry of the original
a b a a
Hamiltonian, and finally (iii) we have verified that including
this difference not only does not lead to qualitatively different Cu O Cu O Cu O Cu O
RIXS cross section but also the quantitative changes are

negligible.
a a b a
As the last step, we express also the b orbiton spectral
function (27) in the t-J model language. Using the same Cu O Cu O Cu O Cu O
transformations as for the Hamiltonian above, we obtain

2 † 1 a a a b
Ab (k,ω) = lim Im |p̃k↑ p̃k↑ | ,
π η→0 ω + E − Ht-J ab
− iη Cu O Cu O Cu O Cu O
(40)
here | is the ground state of Ht-J
ab
at half-filling with energy FIG. 5. (Color online) Schematic view of the propagation of the
E (i.e., is a 1D quantum AF). Let us note that the t-J model orbiton in the spin-orbital separation scenario: the orbiton hops to
spectral function does not depend on the spin σ of the fermion the neighboring site (top panel) and initially excites a single spinon
p̃kσ which is consistent with the fact that the choice of spin σ (middle panel) but then separates from the spinon and can freely
in Eq. (35) was arbitrary. travel in the 1D AF (bottom panel).
Mapping for the c orbiton case. Following the same steps
as for the b orbiton spectral function we obtain the effective
t-J Hamiltonian
ac
Ht-J ≡ Ht-J
0
+ Ht-J
a
+ Ht-Jc B. Spin-orbital separation and numerical results
† Altogether, we see that we managed to map the spin-orbital
= −tc (p̃j σ p̃j +1σ + H.c.)
problem with an FO and AF ground state and one excitation
j,σ
in the b or c orbital onto an effective t-J model with an
1 AF ground state and one empty site (“hole”) without a spin
+J Sj Sj +1 − ñj ñj +1 − Ech ñj , (41)
j
4 j
(cf. Ref. 49 and also Ref. 62 which also shows a mapping
of a spinlike problem onto an effective t-J model). As the
where the parameters are defined as latter problem is well known,63 even before calculating the
1 c spectral function, we can draw an interesting conclusion: The
tc ≡ 3R1c + R2c + 3r1c + r2c J12 , (42)
8 t-J model spectral function at half-filling, when calculated in
1D, describes a phenomenon called spin-charge separation.
J ≡ J1 (1 + R), (43)
This means that the “hole” in the 1D AF separates into an
c independent holon, which carries charge quantum number,
c J1 + J2
Ec ≡ ε̄c − R1 J12 + r1
h c b
. (44) and spinon which carries spin quantum number. Thus, also the
2
here discussed spin-orbital problem shows such a separation
The spectral function is then defined as phenomenon, to be called spin-orbital separation. In fact, this
2 † 1 can also be understood by looking at the cartoon picture in
Ac (k,ω) = lim Im |p̃k↑ p̃k↑ | . Fig. 5: (i) the orbiton moves in such a way that the spin of
π η→0 ω + E − Ht-J
ac
− iη
the hole in this excited orbital is conserved, (ii) this motion
(45) introduces a single defect in the AF ground state (spinon), and
Parameters after the mapping. As shown above, the (iii) the created spinon and the “pure” orbiton (holon in the
parameters tb , tc , and J in the effective t-J model are expressed t-J model language) can move independently and completely
in terms of the spin-orbital model parameters from the middle separate.49
column of Table I. Thus, they can be easily calculated and their Nevertheless, i.e., despite the fact that the t-J model
precise values are given in the right column of Table I. On the spectral function is well known, we calculate the spectral
other hand, while the energies of the onsite orbital excitations functions (40) and (45) using the Lanczos exact diagonal-
Ebh and Ech also follow from these parameters, in order to ization on a 28-site chain separately for each orbiton case.
stay in line with Ref. 22, we directly estimate them following The spectral function for the b orbiton [Ab (k,ω)] and c
the ab initio quantum chemistry calculations on three CuO3 orbiton case [Ac (k,ω)] is shown in Fig. 6. The spectrum
plaquettes in Sr2 CuO3 (cf. Ref. 22). Note that these ab initio for each orbiton case consists of a lower-lying orbiton
calculations are performed for ferromagnetic chain and hence branch with dispersion ∝tb (or ∝tc ), period π , and mixed
they are well suited to our needs since we have Ebh Eb spinon-orbiton
√ excitation bounded from above by the edge
(Ech Ec ) where Eb (Ec ) is defined as the onsite cost of a ∝ J 2 + 4t 2 + 4tJ cos k with t ≡ tb or t ≡ tc depending on
b (c) orbital excitation in a ferromagnetic environment. Let the orbiton under consideration (cf. Ref. 49). Note that this
us note that both methods lead to rather similar results, i.e., spectrum is quantitatively (but not qualitatively) different than
estimating Ebh and Ech directly from the spin-orbital model the “usual” spin-charge separation. The latter is “normally”
parameters given in Table I would lead to similar values as the calculated for the case J < t (whereas in “our” spin-orbital
reported here ab initio values. case J > t in the effective t-J model).49
195138-10
y
k in k out
ein eout
θ Ψ=130
0
kx x
FIG. 7. (Color online) Geometry of the RIXS experiment per-

formed on Sr2 CuO3 and presented in Ref. 22 shown here for the
= 130◦ scattering angle. The momentum transfer kx is denoted as
k in the main text.
Here, (i) orbital d = 3dyz , orbital e = 3d3z2 −r 2 , and the other

†
orbitals are defined as in Eq. (1), (ii) operators fj ασ and fj ασ
FIG. 6. (Color online) Spectral function Ab (k,ω) + Ac (k,ω) as a
function of momentum k and energy transfer ω in the spin-orbital
are defined as in Eq. (1) with α ∈ {a,b,c,d,e}, (iii) Bσ,σ , Cσ,σ ,
scenario and calculated using Lanczos exact diagonalization on a 28- Dσ,σ , and Eσ,σ are complex numbers which define the so-
site chain. Results for broadening η = 0.05 eV which gives FWHM = called RIXS matrix elements. The latter ones can be easily
0.1 eV, i.e., the experimental resolution of RIXS in Sr2 CuO3 (Ref. 22). calculated in the fast collision approximation (see immediately
following).
V. RIXS CROSS SECTION
A. RIXS matrix elements
In this section, we calculate the RIXS spectra of the
orbital excitations in Sr2 CuO3 . As it is well established13,14,20 To calculate the above-defined RIXS matrix elements in
that RIXS is an excellent probe of orbital excitations, the the fast collision approximation, and to be able to compare
calculations are rather straightforward provided the orbiton the obtained results with the experimental ones reported in
spectral function is known. Ref. 22, we assume that (i) the incoming energy of the photon
Following Refs. 6 and 15, using the dipole approximation is tuned to the copper L3 edge, i.e., ωin 930 eV, and thus
and the so-called fast collision approximation,64,65 the RIXS the wave vector of the incoming photon is kin 0.471/Å,
cross section for orbital excitations at the Cu2+ L edge in the (ii) the wave vector at the edge of the Brillouin zone along the
1D copper oxygen chain reads as x direction in Sr2 CuO3 is 0.8051/Å as the lattice constant is52
3.91 Å, (iii) in the ionic picture the ground-state configuration
1 † at the copper site is 3d 9 , (iv) the relatively small spin-orbit
I (k,ω; e) = lim Im 0|TS (k,e)
π η→0 coupling in the 3d shell can be neglected, (v) the incoming
1 polarization vector ein is parallel to the scattering plane and
× TS (k,e)|0 , (46) the outgoing polarization vector is not measured (cf. Fig. 7),
ω + E0 − H − iη
(vi) the scattering plane is the xy plane, i.e., the one in which
where ω ≡ ωout − ωin is the photon energy loss, k ≡ kin − kout the copper oxygen chain lies (which runs along the x direction,
is the photon momentum loss [with k ≡ kx = k · x̂ being the see Fig. 1) (cf. Fig. 7), and (vii) the angle between the outgoing
momentum loss along the x direction of the copper oxygen and the incoming photon momentum is either ψ = 90◦ or 130◦
chain in the studied case of Sr2 CuO3 (cf. Fig. 1)], and |0 (cf. Fig. 7). The latter defines the two scattering geometries
is the ground state of Hamiltonian H with energy E0 [see used in the RIXS experiment reported in Ref. 22.
Eq. (1) and cf. Eqs. (22) and (23)]. Finally, TS (k,e) is the Next, we calculate the RIXS matrix elements in three
RIXS scattering operator, which depends on the incoming steps: First, following inter alia Ref. 15, we express the
† matrix elements in terms of the different components of
and outgoing photon polarization e = ein eout in the RIXS
experiment. It reads as 6,15,65 the incoming and outgoing polarization vectors and the spin
operator. These expressions can be easily obtained from Fig. 1
T S = T b + Tc + Td + Te of Ref. 15 and hence we do not write them here. Second,
† † we express the incoming and outgoing polarization vectors
= eikj [Bσ,σ (e)fj bσ fj aσ + Cσ,σ (e)fj cσ fj aσ
in terms of the angle θ measured between the momentum of
j,σ,σ
the incoming photon and the x chain direction (cf. Fig. 7)
† †
+ Dσ,σ (e)fj dσ fj aσ + Eσ,σ (e)fj eσ fj aσ ]. (47) (note the convention that if k < 0, then θ → 0): (i) the
195138-11
vector of the incoming polarization of the photon in terms Finally, the “interference” terms in the above equation cancel
of the angle θ is ein = [sin θ, cos θ,0], (ii) the vector of the due to the identity relations between the RIXS matrix elements
outgoing polarization of the photon in terms of the angle θ is ∗ ∗
B↑,↑ B↓,↑ = −B↓,↓ B↑,↓ , (50)
eout = [− cos θ, sin θ,0] for π polarization and eout = [0,0,1]
for σ polarization in the ψ = 90◦ geometry, and (iii) the which leads to
vector of the outgoing polarization of the photon in terms Ib (k,ω) = (|B↑,↑ |2 + |B↓,↑ |2 )Ab (k,ω), (51)
of the angle θ is eout = [− cos(θ − 40◦ ), sin(θ − 40◦ ),0] for
π polarization and eout = [0,0,1] for σ polarization in the where we used that (see Sec. IV A)
ψ = 130◦ geometry. Third, we express the angle θ in terms 1 † 1
of the transferred momentum k along the x direction: (i) for Ab (k,ω) = lim Im |p̃k↑ p̃k↑ |
π η→0 ω + E − Ht-J
ab
− iη
ψ = 90◦ geometry the transferred √ momentum as a function
of angle θ ∈ (0◦ ,90◦ ) is k 0.58 2π sin(θ − 45◦ ) where the 1 † 1
= lim Im |p̃k↓ p̃k↓ |
distance between the copper sites along the chain is assumed π η→0 ω + E − Ht-J
ab
− iη
to be equal to unity, and (ii) for ψ = 130◦ the angle changes (52)
as θ ∈ (0◦ ,130◦ ) and k 1.07π sin(θ − 65◦ ).
Altogether, this shows how to calculate the RIXS matrix and Ab (k,ω) was calculated in Sec. IV. Employing the
elements for orbital excitations and that the latter effectively same transformation for the c orbiton, we obtain Ic (k,ω) =
becomes a function of transferred momentum k and energy (|C↑,↑ |2 + |C↓,↑ |2 )Ab (k,ω) with Ac (k,ω) also calculated in
ω. Therefore, in what follows, we simplify notation and write Sec. IV.
I (k,ω; e) → I (k,ω). Second, to complete the RIXS calculations we also have to
add the spectra for the dispersionless excitations to the d and
e orbitals. As this task is straightforward (cf. Appendix C), we
B. Numerical results
obtain for the total RIXS cross section
We first express the RIXS cross section for the b and c
I (k,ω) = (|B↑,↑ |2 + |B↓,↑ |2 )Ab (k,ω)
orbital excitations in terms of the previously calculated (see
Sec. IV) b and c orbiton spectral functions. In order to do so, + (|C↑,↑ |2 + |C↓,↑ |2 )Ac (k,ω)
we use the fact that according to the analysis in Sec. IV A, the + (|D↑,↑ |2 + |D↑,↓ |2 )δ(ω − Ed − EAF )
spin of the hole in the excited orbital does not change during
the orbiton propagation process. Thus, e.g., for only the b part + (|E↑,↑ |2 + |E↑,↓ |2 )δ(ω − Ee − EAF ). (53)
of the RIXS cross section we can write Note that the values of the onsite orbital energies of the
dispersionless orbital excitations Ed and Ee are obtained
1 †
Ib (k,ω) = lim Im 0| Bσ∗1 ,↑ eikj fj aσ1 fj c↑ from the quantum chemistry ab initio calculations for a
π η→0 σ ,j
ferromagnetic chain consisting of three CuO3 plaquettes (cf.
1
Table I). Since these values are given for a ferromagnetic chain,
1 † we have to add the energy cost of a single spin flip (EAF ), which
× B↑,σ2 eikj fj c↑ fj aσ2 |0
ω + E0 − H − iη σ ,j is also calculated using the same ab initio method (cf. Table I
2
for its precise value).

1 †
+ lim Im 0| Bσ∗1 ,↓ eikj fj aσ1 fj c↓ Comparison with the experiment. The RIXS cross section
π η→0 σ calculated according to Eq. (53) is shown in Fig. 8. Comparing
1 ,j
1 †
this theoretical spectrum against the experimental one shown
× B↓,σ2 eikj fj c↓ fj aσ2 |0 . in Fig. 4(a) in Ref. 22 (for the case of the scattering angle
ω + E0 − H − iη σ ,j
2 = 130◦ ; a similar agreement is obtained for the unpublished
(48) RIXS experimental results66 for the scattering angle = 90◦ ),
we note the following similarities between the two:
(i) The c orbiton spectrum: both the dispersion and the
Next, we can employ the transformations used in Secs. III
intensities agree qualitatively and quantitatively; in particular,
and IV A to map the above problem first onto a spin-orbital
the theoretical spectrum has the largest intensity at k = 0 mo-
model and then onto an effective t-J model problem. We obtain
mentum which is solely a result of the dispersion originating
then
in the spin-orbital separation scenario (the RIXS local matrix
elements for the c orbiton are momentum independent in the
1 ∗ † ∗ †
Ib (k,ω) = lim Im |(B↑,↑ p̃k↑ + B↑,↓ p̃k↓ ) RIXS geometry of Fig. 7).
π η→0 (ii) The b orbiton dispersion: the dispersion has the same
1 particular cosinelike shape with a period π and minima at
× (B↑,↑ p̃k↑ + B↑,↓ p̃k↓ )|
ω + E − Ht-J
ab
− iη ±π/2; the spectral weights agree qualitatively and quantita-
1 tively.
∗ † ∗ †
+ lim Im |(B↓,↓ p̃k↑ + B↓,↑ p̃k↓ ) (iii) “Shadow” (“oval”-like) bands above the b orbiton: the
π η→0
width and shape of the shadow band are very similar both in
1 the experiment and in theory; spectral weights agree relatively
× (B↓,↓ p̃k↑ + B↓,↑ p̃k↓ )| .
ω + E − Ht-J
ab
− iη well (e.g., larger spectral weights for the negative than for the
(49) positive momentum transfer).
195138-12
FIG. 8. (Color online) RIXS cross section for ψ = 90◦ (ψ = 130◦ ) scattering geometry as calculated in the spin-orbital separation scenario
and convoluted with the results from the local model (Fig. 13) on the top (bottom) panel. Left (right) panels show line (color map) spectra.
Results for broadening η = 0.05 eV (cf. caption of Fig. 6).
The characteristic spin-orbital separation spectrum is much the more itinerant character of the system and larger dispersion
better visible for the b orbiton than for the c orbiton. The relation for the orbiton.
reason for this is twofold. First, the overall sensitivity of RIXS Comparison with other theoretical calculations. There are
to the b orbiton excitations is much larger than to the c orbiton actually two other simple approximations which might naively
due to the chosen geometry of the RIXS experiment. Thus, be employed to calculate the RIXS spectra and which could
all features related to the b orbiton are better visible than be compared against the experiment: (i) the “local model”
those related to the c orbiton. Second, even despite this, the approximation which assumes that all orbital excitations are
spin-orbital separation can be better observed for the b orbiton local, i.e., also both the b and the c orbiton spectral functions
case than for the c orbiton (cf. Fig. 6). This is because the do not have any momentum dependence (cf. Appendix C),
effective hopping element tb for the b orbiton is larger than and (ii) the one which assumes that the spectral functions
the hopping tc for the c orbiton (cf. Table I), which is due to of the b and c orbitons are calculated using the linear
(i) the renormalization ∝λc of the copper oxygen hopping tπ orbital wave approximation (cf. Appendix B 2). However, as
for the hopping from c orbiton due to the formation of the shown in detail in the above-mentioned Appendices, the RIXS
bonding and antibonding states with the neighboring oxygens spectra calculated using these approximations do not fit the
(see Appendix A), and (ii) the larger effective charge-transfer experimental ones.
gap for the c orbital than for the b orbital (see Table I; note that
this effective charge-transfer gap is defined as the difference VI. DISCUSSION AND CONCLUSIONS
in energy between a particular 3d orbital and the hybridizing
2p orbital and thus is larger for lower-lying 3d orbitals). We first list a few alternative scenarios that might lead to
The main discrepancy between the experiment and theory the dispersive orbital excitations in Sr2 CuO3 and argue why
is related to the somewhat smaller dispersion in the theoretical they do not lead to a plausible explanation of the experimental
calculations than in the experiment. While there might be results reported in Ref. 22. At the end of the section, we present
several reasons explaining this fact, let us point to two plausible our conclusions.
ones. First, the neglected spin-orbit coupling in the 3d orbitals
would mix the b and d (i.e., xz and yz) orbital excitations A. Alternative scenarios leading to dispersive features in the
and would lead to a finite dispersion in the d orbital channel. RIXS spectrum of Sr2 CuO3
This would mean that the present dispersionless d orbital Spin-charge separation observed directly. The spin-charge
excitation would no longer “cover” parts of the dispersive b separation where a hole created in a 1D AF decays into a
orbiton. Thus, effectively this would lead to a large dispersive holon and a spinon can be observed with ARPES: e.g., in
feature around the b and d excitation energies. Second, the SrCuO2 (Ref. 63) or in Sr2 CuO3 (Ref. 67). However, not
relatively high covalency of the Sr2 CuO3 compound, which is only that holon dispersion is much larger than the dispersion
not taken into account in the present derivation, might lead to under consideration in this paper [it is of the order of 1.1 eV
195138-13
(Ref. 67)], but also, what is more important, one can not chain in Sr2 CuO3 .22 We explained that these dispersive
directly probe the spin-charge separation in RIXS since in features can indeed be attributed to the dispersive orbital
RIXS the total charge is conserved.68 excitations (orbitons). The unexpectedly strong dispersion of
Holon, antiholon, and two spinons, i.e., spin-charge separa- these excitations is not only a result of the relatively strong
tion indirectly. Nevertheless, it occurs that there is a possibility superexchange interactions in the system, but is also due to
to observe spin-charge separation with RIXS or electron the fractionalization of the spin and orbital degrees of freedom
energy loss spectroscopy (EELS) in an indirect way. If one which, as shown in this paper and in Refs. 22 and 49, is possible
transfers a hole from the copper site i either to the neighboring in this quasi-1D strongly correlated system.
copper site i + 1 to form a doubly occupied site or to the Finally, one may wonder whether the spin-orbital separation
neighboring oxygen plaquette surrounding the central copper phenomenon can also be observed in other transition-metal
site i + 1 to form a Zhang-Rice singlet, then one ends up with oxides. While we suggest that this should be possible in most
one hole in the spin background on site i and another hole in other quasi-1D systems which are again mostly cuprates, it
the spin background on site i + 1. Next, both of these objects is impossible to observe this phenomenon in 2D and 3D
can become mobile and experience the spin-charge separation: systems, such as La2 CuO4 , LaMnO3 , or LaVO3 , with long-
the first one can move by decaying to a holon and a spinon, range magnetic order (cf. Ref. 73). However, many theoretical
while the second one can move by decaying to an antiholon studies have discussed the nature of the orbital excitations in
and a spinon. This rather complicated scenario was invoked to these systems, which leaves a large field to be still explored
explain the K-edge RIXS spectra in various quasi-1D cuprates experimentally (cf. Refs. 11, 13, 17, 35, 37–42, and 45).
(cf. Refs. 68 and 69) and the EELS spectrum in Sr2 CuO3 .70 Furthermore, the mapping of the spin-orbital model into the
However, the common feature of all these experiments is effective simpler t-J model presented here is actually valid
that there is a large dispersion (of the order of 1 eV) which also in higher dimensions.49 However, the lack of experimental
has a periodicity of 2π and a minimum at k = 0. Thus, clearly results, which can verify various theories concerning these
it is not the spectrum that is observed in the L-edge RIXS orbital excitations, means that it remains a challenge both for
experiment in Ref. 22. The reason for this is that RIXS at L theory and for experiment to explore the nature of the orbital
edge is much more sensitive to the onsite excitations on the excitations in higher dimensions.
copper site than to the intersite charge-transfer excitations on
the neighboring oxygen or copper sites.71
Orbital excitations propagating via the O(2p) orbitals. In ACKNOWLEDGMENTS
that case, the propagation would entirely happen via the O(2p)
First and foremost, we acknowledge very stimulating dis-
orbitals on a kind of a zigzag chain along the CuO4 plaquettes.
cussions and common work on this subject with our theoretical
This, however, can not lead to a momentum dependence in the
collaborators M. Daghofer, L. Hozoi, and G. Khaliullin as
observed spectrum.
well as with our experimental collaborators J. Schlappa, T.
Similar experiments. One should also compare the here
Schmit, and H. Ronnow. We also thank S.-L. Drechsler,
reported theoretical results to the experimental ones which
A. M. Oleś, G. A. Sawatzky, M. van Veenendaal, and V.
were discussed in Ref. 72. There, a somewhat similar disper-
Yushhankhai for very valuable comments. K.W. acknowledges
sive feature, as the one discussed here, was discovered in the
support from the Alexander von Humboldt Foundation, the
RIXS spectra at the 1s → 3d edge. Although this dispersion
Polish National Science Center (NCN) under Project No.
was attributed to an orbital excitation, it remained unclear
2012/04/A/ST3/00331, and from the U. S. Department of
to the authors of that paper how to correctly interpret this
Energy, Materials Sciences and Engineering Division, under
phenomenon. An obvious suggestion is that the spectrum
Contract No. DE-AC02-76SF00515. This research benefited
observed in Ref. 72 might be of similar origin as the one
from the RIXS collaboration supported by the Computational
discussed here: the dispersion also has a π periodicity and
Materials Science Network program of the Division of Mate-
is of the order of 0.2 eV. However, there are two problems
rials Science and Engineering, U. S. Department of Energy,
with this scenario: (i) there is just one dispersive peak but no
Grant No. DE-SC0007091.
other dispersive modes and there is no shoulder peak, (ii) the
dispersion is shifted by π/2 in the momentum space. This
first problem can perhaps be “solved”: RIXS at the 1s → 3d
APPENDIX A: REDUCTION OF THE EFFECTIVE
edge involves quadrupolar transitions, the RIXS signal is rather
HOPPING IN THE LINEAR CHAIN DUE TO ∝tn2
weak, and thus it is possible that one can not observe all details
PERTURBATIVE PROCESSES
of the spectra. However, the second one remains a challenge for
theory. One suggestion might be that the effective “dispersion” Let us first state that, due to the tn2 processes, the number of
that one sees in the spectrum in Ref. 72 is the top part of the holes residing in the orbitals within the Cu-O-Cu-O- . . . chain
“shadow” bands that we reported here for the b orbiton; this for a particular CuO4 cluster depends on the particular α orbital
would require that the “true” (lower) b orbiton band is covered forming the bonding state (cf. Fig. 1). More precisely, for the
by some other excitations in that experiment. bonding states formed around the b orbital and occupied by one
hole, the whole charge is concentrated in the orbitals within
the Cu-O-Cu-O- . . . chain, while for the bonding states formed
B. Conclusions by the a or c orbitals, it is not the case. This is because in
We have considered in detail the origin of the dispersive the latter case the bonding state is formed by orbitals situated
features observed in the RIXS spectra of the quasi-1D CuO3 above and below the Cu-O-Cu-O- . . . chain.
195138-14
Looking in detail at this problem, we concentrate first Sec. III A, and (ii) the use of the factors λa and λc which
at the case with the hole being initially doped into the a renormalize the number of holes present within the chain
orbital. The part of the charge-transfer Hamiltonian (1) which when a hole is doped into the a or c orbital, respectively
is responsible for the effect mentioned above is (see above). We have verified that the renormalization of other
† † parameters has a much smaller effect. In particular, (i) the
−tσ o (fiaσ fiyo+σ − fiaσ fiyo−σ + H.c.) charge-transfer energies in the bonding states should change
i,σ similarly for all orbitals with respect to their values in the

+ yo (niyo+ + niyo− ), (A1) charge-transfer model defined in the 2p and 3d orbital bases,
i
(ii) the matrix elements of the Coulomb interaction in the
bonding/antibonding basis are similar to the ones calculated
with tσ o being the main “actor” here, i.e., it is this hopping in the 2p and 3d orbital bases.
element which makes the hole escape from the Cu-O-Cu-O-
. . . chain (see Fig. 1). We can now easily diagonalize this
three-level problem and calculate the number of holes left in APPENDIX B: LINEAR ORBITAL WAVE APPROXIMATION
the a orbital by evaluating the following quantum mechanical The “standard” way to obtain the orbiton dispersion in the
amplitude: spin-orbital model is to use the linear orbital wave (LOW)
yo − ea approximation (cf. Ref. 43) for the orbital pseudospin degrees
λa ≡ a|ψa = , (A2) of freedom and to integrate out the spin degrees of freedom
2tσ2o + (yo − ea )2
in a mean-field way.49 This in general may be justified here

due to the presence of the long-range orbital order. In order to
where ea = (yo − 2yo + 8tσ o )/2, while |ψa is the bond-
test this scenario, in this section of the Appendix, we perform
ing state coming from the above diagonalization procedure and the LOW approximation and calculate the orbiton spectral
which is well separated from the antibonding and nonbonding function (Sec. B 1) together with the RIXS cross section (see
states (so that we can skip the latter two when studying the Sec. B 2). We also discuss why the LOW approximation fails
low-energy regime). in properly describing the experimental results reported in
A similar analysis as above but for the c orbital leads to Ref. 22 (cf. Sec. B 3).
xo − ec
λc ≡ c|ψc = , (A3)
2
2tπo + (xo − ec )2 1. Spectral function in linear orbital wave approximation
LOW for b orbiton. Following, e.g., Refs. 43 and 49 we
where ec = (xo − 2xo + 8tπo )/2. Here again |ψc is the
first introduce the following bosonic creation (annihilation)
bonding state but this time centered around the c orbital. †
Finally, since the b orbital does not hybridize with the operators βj (βj ) for the orbital pseudospin operator:
oxygens lying above or below the Cu-O-Cu-O- . . . chain, the † †
corresponding λb would be equal to unity and could be skipped σjz = βj βj − 12 , σj+ = βj , σj− = βj , (B1)
in what follows. where we already skipped the three-orbiton terms in the above
These renormalized values of the number of holes within expressions since we will keep only quadratic terms in the
the chain directly lead to renormalized values of the hopping bosonic degrees of freedom in the effective Hamiltonian below.
elements from the ψa and ψc orbitals with respect to the Besides, we decouple the orbital operators from the spins and
hopping elements from the pure a and c orbitals. A rigorous assume for the spins their appropriate mean-field values. The
calculation would now require that together with using the latter is a standard procedure when calculating the spin-wave
parameters λa and λc as renormalizing the hopping, we dispersion in the spin and orbitally ordered systems.55,75
should also use the basis spanned by the ψa and ψc orbitals. Applying these transformations to the Hamiltonian H̄, we
However, we avoid this in our calculations. We justify this obtain the following LOW Hamiltonian:
“approximation” as follows.
In general, to properly account for all the effects arising
ab
HLOW ≡ HLOW
0
+ HLOW
a
+ HLOW
b

from the ∝tn2 perturbative processes, the rigorous treatment =
†
(B + 2Jb cos k)βk βk
would require using the so-called cell perturbation theory:74 k
that is, to rewrite the full charge-transfer Hamiltonian using 1

the bonding/antibonding states and then to calculate the su- + J1 (1 + R) Si · Si+1 − , (B2)
perexchange interactions in this basis. This, however, requires i
4
very tedious calculations as even for the much simpler case of
with the constants B and Jb defined as
Ref. 74 the problem is nontrivial and complex.
b
b J1 + J2
Therefore, we follow the more standard route, i.e., we
calculate the superexchange interactions in Secs. III B–III E B ≡ ε̄b − A R1 J12 + r1
b b
2
using the orbital basis that was already used to write the
charge-transfer model (1). The only two remnants of the J1 + J2b
− B R2b J12
b
+ r2b + 2J1 (1 + R)B (B3)
cell perturbation theory, or in other words of the fact that 2
the superexchange should be modified due to the formation and
of bonding and antibonding states, are (i) the use of the b
renormalized parameters ε̄α instead of εα as discussed in Jb ≡ 12 J12
c
A R1 + r1b − B R2b + r2b , (B4)
195138-15
with Since | is a vacuum for boson operators β| = 0, we easily

obtain
A ≡ |Si · Si+1 + 34 | (B5) 1 1
Ab (k,ω) = lim Im . (B9)
π η→0 ω − B − 2Jb cos k − iη
and
The orbiton spectral function consists of a single quasiparticle
B ≡ | 14 − Si · Si+1 | . (B6) peak with a sinelike dispersion, with period 2π and bandwidth
4|Jb | 0.08 eV (cf. Fig. 10). This result can be intuitively
Here, | is the AF and FO ground state (cf. Sec. IV) of HLOW b understood by looking at the cartoon picture of the orbiton
with energy E . The values of the spin-spin correlations have propagation in the LOW approximation (cf. Fig. 9).
to be calculated for the spin ground state of the Hamiltonian LOW for c orbiton. Following the same steps as above, we
b
HLOW , which is a quantum AF (see Sec. IV). This can easily obtain
be obtained from the well-known exact Bethe-ansatz-based ac
HLOW ≡ HLOW
0
+ HLOW
a
+ HLOW
c
solution for a 1D quantum AF: A = 0.31 and B = 0.69. (Let †
us note that these numbers are significantly different from those = (C + 2Jc cos k)βk βk
that are known for the not-realized-here “classical” case, i.e., k
for the ordered Néel AF; in that case, A = 0.5, B = 0.5, and 1

Jb 0.011 eV.) Using the spin-orbital model parameters from + J1 (1 + R) Si · Si+1 − (B10)
4
Table I, we finally obtain Jb −0.019 eV as also reported in i
Table I. with the constants C and Jc defined as

Although one could directly use Eq. (B3) to calculate the
onsite cost of an orbital excitation B, this value can also be J1 + J2c
C ≡ ε̄c − A R1c J12c
+ r1c
calculated by using the ab initio quantum chemistry calculation 2
c
for a ferromagnetic chain with four CuO3 plaquettes. The J1 + J
latter gives the value of a single orbital excitation in the − B R2c J12c
+ r2c 2
+ 2J1 (1 + R)B (B11)
2
ferromagnetic chain Eb = 2.15 eV (cf. discussion in Sec. IV
and Table I) and leads to and
c
Jc ≡ 12 J12
c
A R1 + r1c − B R2c + r2c , (B12)
B Eb + EAF (B7)
which gives Jc −0.014 eV (cf. Table I). Again, C 1.74 eV
and, since EAF = 0.33 eV, B 2.48 eV (cf. Table I). To be can be estimated using the ab initio calculated value (cf.
in line with Ref. 22, we use the latter value as the cost of the discussion above for the b orbiton) of Ec 1.41 eV for a
local b orbital excitations in the AF chain.
Having defined the Hamiltonian and its parameters, we are
now ready to compute the orbiton spectral function Ab (k,ω)
[Eq. (24)], which, when expressed in the new bosonic operators
[Eq. (B1)], reads as
1 1 †
Ab (k,ω) = lim Im |βk β | .
π η→0 ω + E − HLOW
ab
− iη k
(B8)
a b a a
Cu O Cu O Cu O Cu O
a a b a
Cu O Cu O Cu O Cu O
a a a b
Cu O Cu O Cu O Cu O
FIG. 9. (Color online) Schematic view of the propagation of the FIG. 10. (Color online) Spectral function Ab (k,ω) + Ac (k,ω) as
orbiton in the LOW approximation: note that the orbiton moves in a function of momentum k and energy transfer ω in the LOW
such a way that it does not disturb the AF correlations which are due approximation and quantum AF case. Results for broadening η =
to the mean-field decoupling of spin and orbital degrees of freedom. 0.05 eV (cf. caption of Fig. 6).
195138-16
local c orbital excitation in the ferromagnetic chain and the Note that in the above expression the spin-dependent part [the
relation (see above) three last terms of the right-hand side of Eq. (B15)] in Eq. (B16)
is skipped. This is because it does not lead to any dispersive
C Ec + EAF . (B13) excitations since there are no terms in the LOW Hamiltonian
Finally, we obtain the following spectral function Ac (k,ω) : which could move a spin excitation together with an orbital
excitation (in the LOW approximation). (This is somewhat
1 1 similar to the problem of a hole doped into the orbitally ordered
Ac (k,ω) = lim Im , (B14)
π η→0 ω − C − 2Jc cos k − iη state in a 1D chain which can “visit” the neighboring sites but
which is also shown in Fig. 10 and which qualitatively the spectrum of which is k independent.76 ) However, these
resembles the above-calculated b-orbiton dispersion. three neglected terms will contribute to the total RIXS cross
Spin-orbital waves. As a side remark, let us note that the section as dispersionless excitations.
joint spin-orbital wave defined as in Ref. 39 can not be present When, apart from the above-discussed b orbiton case, we
in the considered spin-orbital model. The reason is that this also include the contribution from the c orbiton (which is
would require such terms as, e.g., (Si+ Sj− + Si− Sj+ )(τi+ τj+ + analogous to the b orbiton case) and from the dispersionless d
and e orbitons, we obtain
τi− τj− ) to be present in the spin-orbital Hamiltonian (6), which
is not the case here. I (k,ω) = 14 |B↑,↑ + B↓,↓ |2 Ab (k,ω)

+ 14 |B↑,↑ − B↓,↓ |2 + |B↓,↑ |2 δ(ω − Eb − EAF )
2. RIXS in linear orbital wave scenario
+ 14 |C↑,↑ + C↓,↓ |2 Ac (k,ω)
In order to calculate the RIXS cross section in the LOW
approximation, we first express the RIXS operator (47) in + 14 |C↑,↑ − C↓,↓ |2 + |C↓,↑ |2 δ(ω − Ec − EAF )
terms of the orbital pseudospins. For the b orbiton case (i.e.,
+ [|D↑,↑ (k)|2 + |D↑,↓ (k)|2 ]δ(ω − Ed − EAF )
Tb operator), following Eqs. (8) and (13), we obtain
+ [|E↑,↑ (k)|2 + |E↑,↓ (k)|2 ]δ(ω − Ee − EAF ).
1 ikj 1
Tb = √ e (B↑,↑ + B↓,↓ ) σj+ + (B↑,↑ − B↓,↓ )Sjz σj+ (B17)
N j 2
Here, the spectral functions Ab (k,ω) and Ac (k,ω) are calcu-
+ + − +
+ B↑,↓ Sj σj + B↓,↑ Sj σj . (B15) lated in Sec. B 1 and the RIXS matrix elements follow from
Sec. V A.
Next, following Eq. (B1), we express the first term of Comparison with the experiment. The RIXS cross section
the above-written RIXS operator in terms of the Holstein- calculated using Eq. (B17) is shown in Fig. 11. A small
Primakoff bosons βk+ : dispersion of the orbiton excitations, known already from
the spectral functions in Sec. B 1, is relatively well visible
Tb(1) = 12 (B↑,↑ + B↓,↓ )βk+ . (B16) in the RIXS cross section. However, there is a qualitative
FIG. 11. (Color online) RIXS cross section for 90◦ (130◦ ) scattering geometry as calculated in the LOW approximation and convoluted with
the results from the local model (Fig. 13) on the top (bottom) panel. Left (right) panels show line (color map) spectra. Results for broadening
η = 0.05 eV (cf. caption of Fig. 6).
195138-17
A PROCESS and
b
tB = tB1 + tB2 = 12 J12
c
R1 + r1b |Siz Si+1
z
+ 34 |
b
b a a b − R2 + r2b | 41 − Siz Si+1
z
| , (B20)
Cu O Cu Cu O Cu
with

tB1 = 12 J12
b
μ R1b + r1b ↑↓|Siz Si+1
z
+ 34 |↑↓
B1 PROCESS b
− R2 + r2b ↑↓| 14 − Siz Si+1
z
|↑↓ , (B21)
and
b a a b b
Cu O Cu Cu O Cu tB2 = 12 J12
b
R1 + r1b ν ↓↓|Siz Si+1
z
+ 34 |↓↓ , (B22)
where |↓↓ denotes a ferromagnetic state, |↓↑ denotes a Néel
AF state, and μ = | |↓↑ |2 ∼ 0.8 and ν = | |↓↓ |2 ∼ 0.2.
B2 PROCESS
Substituting parameters from Table I and spin correlations for
the quantum AF, Néel AF, and ferromagnetic state we obtain
a a b that tA ∼ −0.046 eV, tB1 ∼ 0.009 eV, and tB2 ∼ 0.018 eV (one
b
can check that altogether they indeed give Jb ∼ −0.019 eV as
Cu O Cu Cu O Cu
calculated in the previous section, cf. Table I). Thus, we see
that (i) the A process has a surprisingly large contribution and
FIG. 12. (Color online) Schematic view of the three possible an opposite sign to the other processes, so, unlike in the LOW
superexchange processes which lead to the orbiton propagation (here result presented above, we should treat it separately as we make
shown for the b orbiton, but the c orbiton case is analogous). In the a huge error when we add all of these processes together; (ii)
LOW approximation, all of these processes, the amplitudes of which
the B1 process is not only much smaller than the A process,
have different signs, are summed up. When treated separately, as in
but also it is twice smaller than the B2 process.77
the spin-orbital separation approach, it occurs that for the orbiton
This means that it is reasonable to try to define such an
propagation in a quantum AF only processes A and B2 matter.
approximation, when calculating the orbiton propagation in the
disagreement between these theoretical calculations and the spin-orbital model (6), that, unlike the LOW approximation,
experimental results [cf. Fig. 4(a) in Ref. 22 for the case of the will not average over these three processes. At the same time,
RIXS with scattering angle = 130◦ ]; a similar disagreement such approximation could neglect the B1 process due to its
is obtained for the unpublished RIXS experimental results66 relatively small amplitude. In order to verify what kind of
for the scattering angle = 90◦ . The main differences are approximation can be used, let us try to intuitively understand
as follows: (i) the theoretical dispersion has its minimum at the difference between these three processes (cf. Fig. 12).
k = 0 according to the calculations, while this is not the case While process A denotes an orbiton hopping accompanied by
in the experiment, (ii) the theoretical results do not predict the a spin flip, the B processes describe orbiton hoppings without
onset of a continuum above the b excitation, (iii) the obtained any change in the spin background: the B1 for the case when
dispersion of the quasiparticle peaks is much smaller than in the spins on the bond are antiparallel, while the B2 when the
the experiment, and (iv) there is a disagreement between the spins are parallel. However, one can also look at this problem in
calculated and measured RIXS intensities. a different way: for the A and B2 process the spin of the hole in
orbital b is conserved during the spin-orbital exchange. Hence,
when process B1 is neglected, one can safely assume that the
3. Why linear orbital wave approximation fails spin of the hole in the excited orbital does not change during
General considerations. Let us show why the LOW theory, orbiton propagation, and this is the essence of the mapping to
employed above in calculating the orbiton spectral function, the t-J model discussed in the main text of the paper.
can not correctly reproduce the orbiton propagation in the here
discussed spin-orbital model. APPENDIX C: RIXS FOR DISPERSIONLESS ORBITAL
In order to do that, we take a closer look at all possible EXCITATIONS
channels of the orbiton propagation (see Fig. 12) and calculate
their relative contribution to the orbiton propagation in the In order to verify that the RIXS cross section in the so-called
LOW approximation. (In what follows, we concentrate on the b local model, i.e. with all orbital excitations dispersionless,
orbiton case, but similar arguments apply to the c orbiton case.) indeed does not follow the experimental RIXS cross section,22
Thus, we split the effective orbiton superexchange process we study the RIXS response for the following local Hamilto-
∝Jb , as calculated in the previous section [Eq. (B12)] into nian:

three different contributions: H = (Eb + EAF ) (nib − nia ) + (Ec + EAF ) (nic − nia )
Jb = tA + tB1 + tB2 , (B18) i

i
where + (Ed + EAF ) (nid − nia )

b
i
tA = 12 J12
b
R1 + r1b + R2b + r2b | 21 (Si+ Si+1
−
+ H.c.)|
+ (Ee + EAF ) (nie − nia ). (C1)
(B19) i
195138-18
FIG. 13. (Color online) RIXS cross section for 90◦ (130◦ ) scattering geometry as calculated in the local model on the top (bottom) panel.
Left (right) panels show line (color map) spectra. Results for broadening η = 0.05 eV (cf. caption of Fig. 6).
Here Eb , Ec , Ed , and Ee are the costs of the local orbital excitations as calculated using the ab initio quantum chemistry cluster
calculations for the ferromagnetic CuO3 chain in Sr2 CuO3 (see Table I), while EAF = 0.24 eV is the estimated correction to these
values due to the quantum AF ground state (see Table I). Substituting Eq. (C1) into (46) and using Eq. (47), we easily obtain
I (k,ω) = [|B↑,↑ (k)|2 + |B↑,↓ (k)|2 ]δ(ω − Eb − EAF ) + [|C↑,↑ (k)|2 + |C↑,↓ (k)|2 ]δ(ω − Ec − EAF )
+ [|D↑,↑ (k)|2 + |D↑,↓ (k)|2 ]δ(ω − Ed − EAF ) + [|E↑,↑ (k)|2 + |E↑,↓ (k)|2 ]δ(ω − Ee − EAF ), (C2)
which is shown in Fig. 13 for the two discussed scattering geometries. It can be easily verified that the cross section calculated in
this way does not agree with the RIXS experimental cross section, as shown in Fig. 4(a) in Ref. 22 (for the case of the scattering
angle = 130◦ ; a similar disagreement is obtained for the unpublished RIXS experimental results66 for the scattering angle
= 90◦ ).
1 8
M. Imada, A. Fujimori, and Y. Tokura, Rev. Mod. Phys. 70, 1039 L. Braicovich, J. van den Brink, V. Bisogni, M. M. Sala, L. J. P.
(1998). Ament, N. B. Brookes, G. M. De Luca, M. Salluzzo, T. Schmitt,
2
R. Coldea, S. M. Hayden, G. Aeppli, T. G. Perring, C. D. Frost, V. N. Strocov, and G. Ghiringhelli, Phys. Rev. Lett. 104, 077002
T. E. Mason, S.-W. Cheong, and Z. Fisk, Phys. Rev. Lett. 86, 5377 (2010).
9
(2001). A. Auerbach, Interacting Electrons and Quantum Magnetism
3
S. Notbohm, P. Ribeiro, B. Lake, D. A. Tennant, K. P. Schmidt, G. (Springer, New York, 1994).
10
S. Uhrig, C. Hess, R. Klingeler, G. Behr, B. Büchner, M. Reehuis, T. Giamarchi, Quantum Physics in One Dimension (Oxford
R. I. Bewley, C. D. Frost, P. Manuel, and R. S. Eccleston, Phys. University Press, Oxford, 2004).
11
Rev. Lett. 98, 027403 (2007). K. I. Kugel and D. I. Khomskii, Zh. Eksp. Teor. Fiz. 64, 1429 (1973)
4
I. A. Zaliznyak, H. Woo, T. G. Perring, C. L. Broholm, C. D. Frost, [Sov. Phys. JETP 37, 725 (1973)].
12
and H. Takagi, Phys. Rev. Lett. 93, 087202 (2004). H. F. Pen, J. van den Brink, D. I. Khomskii, and G. A. Sawatzky,
5
L. J. P. Ament, G. Ghiringhelli, M. M. Sala, L. Braicovich, and Phys. Rev. Lett. 78, 1323 (1997).
13
J. van den Brink, Phys. Rev. Lett. 103, 117003 (2009). S. Ishihara and S. Maekawa, Phys. Rev. B 62, 2338
6
M. W. Haverkort, Phys. Rev. Lett. 105, 167404 (2010). (2000).
7 14
L. Braicovich, L. J. P. Ament, V. Bisogni, F. Forte, C. Aruta, F. Forte, L. J. P. Ament, and J. van den Brink, Phys. Rev. Lett. 101,
G. Balestrino, N. B. Brookes, G. M. De Luca, P. G. Medaglia, 106406 (2008).
15
F. Miletto Granozio, M. Radovic, M. Salluzzo, J. van den Brink, P. Marra, K. Wohlfeld, and J. van den Brink, Phys. Rev. Lett. 109,
and G. Ghiringhelli, Phys. Rev. Lett. 102, 167401 (2009). 117401 (2012).
195138-19
16 45
Note that due to the typically quenched angular momentum of the J. Kim, D. Casa, M. H. Upton, T. Gog, Y.-J. Kim, J. F. Mitchell,
orbital ordered ground state, neutrons can not detect the orbital M. van Veenendaal, M. Daghofer, J. van den Brink, G. Khaliullin,
order (cf. Ref. 15). and B. J. Kim, Phys. Rev. Lett. 108, 177003 (2012).
17 46
S. Ishihara, Phys. Rev. B 69, 075118 (2004). W.-L. You, A. M. Oleś, and P. Horsch, Phys. Rev. B 86, 094412
18
S.-I. Shamoto, M. Tsubota, G. Iga, B. Faak, and T. Kajitani, (2012).
47
J. Neutron Res. 13, 175 (2005). B. Kumar, Phys. Rev. B 87, 195105 (2013).
19 48
Y.-J. Kim, A. P. Sorini, C. Stock, T. G. Perring, J. van den Brink, D. Vieira, arXiv:1212.3241.
49
and T. P. Devereaux, Phys. Rev. B 84, 085132 (2011). K. Wohlfeld, M. Daghofer, S. Nishimoto, G. Khaliullin, and J. van
20
E. Saitoh, S. Okamoto, K. T. Takahashi, K. Tobe, K. Yamamoto, T. den Brink, Phys. Rev. Lett. 107, 147201 (2011).
50
Kimura, S. Ishihara, S. Maekawa, and Y. Tokura, Nature (London) T. Ami, M. K. Crawford, R. L. Harlow, Z. R. Wang, D. C. Johnston,
410, 180 (2001). Q. Huang, and R. W. Erwin, Phys. Rev. B 51, 5994 (1995).
21 51
M. Grüninger, R. Rückamp, M. Windt, P. Reutler, C. Zobel, N. Motoyama, H. Eisaki, and S. Uchida, Phys. Rev. Lett. 76, 3212
T. Lorenz, A. Freimuth, and A. Revcolevschi, Nature (London) (1996).
52
418, 39 (2002). K. M. Kojima, Y. Fudamoto, M. Larkin, G. M. Luke, J. Merrin,
22
J. Schlappa, K. Wohlfeld, K. J. Zhou, M. Mourigal, M. W. B. Nachumi, Y. J. Uemura, N. Motoyama, H. Eisaki, S. Uchida,
Haverkort, V. N. Strocov, L. Hozoi, C. Monney, S. Nishimoto, S. K. Yamada, Y. Endoh, S. Hosoya, B. J. Sternlieb, and G. Shirane,
Singh, A. Revcolevschi, J. Caux, L. Patthey, H. M. Ronnow, J. v. d. Phys. Rev. Lett. 78, 1787 (1997).
53
Brink, and T. Schmitt, Nature (London) 485, 82 (2012). R. Neudert, S.-L. Drechsler, J. Málek, H. Rosner, M. Kielwein,
23
R. M. Macfarlane and J. W. Allen, Phys. Rev. B 4, 3054 (1971). Z. Hu, M. Knupfer, M. S. Golden, J. Fink, N. Nücker, M. Merz,
24
D. Polli, M. Rini, S. Wall, R. W. Schoenlein, Y. Tomioka, Y. Tokura, S. Schuppler, N. Motoyama, H. Eisaki, S. Uchida, M. Domke, and
G. Cerullo, and A. Cavalleri, Nat. Mater. 6, 643 (2007). G. Kaindl, Phys. Rev. B 62, 10752 (2000).
25 54
C. Ulrich, L. J. P. Ament, G. Ghiringhelli, L. Braicovich, M. M. Sala, J. S. Griffith, The Theory of Transition Metal Ions (Cambridge
N. Pezzotta, T. Schmitt, G. Khaliullin, J. van den Brink, H. Roth, University Press, Cambridge, 1964).
55
T. Lorenz, and B. Keimer, Phys. Rev. Lett. 103, 107205 (2009). A. M. Oleś, G. Khaliullin, P. Horsch, and L. F. Feiner, Phys. Rev.
26
P. B. Allen and V. Perebeinos, Phys. Rev. Lett. 83, 4828 (1999). B 72, 214431 (2005).
27 56
J. van den Brink, Phys. Rev. Lett. 87, 217202 (2001). J. B. Grant and A. K. McMahan, Phys. Rev. B 46, 8440 (1992).
28 57
K. P. Schmidt, M. Grüninger, and G. S. Uhrig, Phys. Rev. B 76, M. W. Haverkort, Ph.D. thesis, University of Cologne, 2005.
58
075108 (2007). J. Zaanen and A. M. Oleś, Phys. Rev. B 37, 9423 (1988).
29 59
Note that this intersite spin-orbital interaction has nothing to do M. W. Haverkort, M. Zwierzycki, and O. K. Andersen, Phys. Rev.
with the relativistic onsite spin-orbit interaction, which in the here B 85, 165113 (2012).
60
discussed case of 3d transition-metal oxides is vanishingly small Let us note that these two different formulas for the orbiton spectral
with respect to the spin-orbital interaction (Refs. 55 and 57). functions are not fully equivalent: Eqs. (24) and (25) do not contain
30
A. M. Oleś, P. Horsch, L. F. Feiner, and G. Khaliullin, Phys. Rev. the orbital excitations which are above the Hubbard gap. However,
Lett. 96, 147205 (2006). we can safely neglect the coherent propagation of the orbital
31
A. M. Oleś, J. Phys.: Condens. Matter 24, 313201 (2012). excitations in the Hilbert space with double occupancies, since
32
W. Brzezicki, J. Dziarmaga, and A. M. Oleś, Phys. Rev. Lett. 109, (i) for the coherent propagation of the orbital excitations the relevant
237201 (2012). gap is of the order of the Hubbard U and not of the order of the
33
W. Brzezicki, J. Dziarmaga, and A. M. Oleś, Phys. Rev. B 87, charge-transfer gap , and (ii) ε̄b and ε̄c are much smaller than the
064407 (2013). Hubbard U .
34 61
K. I. Kugel and D. I. Khomskii, Usp. Fiz. Nauk 136, 621 (1982) G. Martinez and P. Horsch, Phys. Rev. B 44, 317 (1991).
62
[Sov. Phys.-Usp. 25, 231 (1982)]. P. Bouillot, C. Kollath, A. M. Läuchli, M. Zvonarev, B. Thielemann,
35
L. F. Feiner, A. M. Oleś, and J. Zaanen, Phys. Rev. Lett. 78, 2799 C. Rüegg, E. Orignac, R. Citro, M. Klanjšek, C. Berthier,
(1997). M. Horvatić, and T. Giamarchi, Phys. Rev. B 83, 054407 (2011).
36 63
L. F. Feiner, A. M. Oleś, and J. Zaanen, J. Phys.: Condens. Matter C. Kim, A. Y. Matsuura, Z.-X. Shen, N. Motoyama, H. Eisaki,
10, L555 (1998). S. Uchida, T. Tohyama, and S. Maekawa, Phys. Rev. Lett. 77, 4054
37
S. Ishihara, J. Inoue, and S. Maekawa, Phys. Rev. B 55, 8280 (1997). (1996).
38 64
J. van den Brink, W. Stekelenburg, D. I. Khomskii, G. A. Sawatzky, J. Luo, G. T. Trammell, and J. P. Hannon, Phys. Rev. Lett. 71, 287
and K. I. Kugel, Phys. Rev. B 58, 10276 (1998). (1993).
39 65
A. M. Oleś, L. F. Feiner, and J. Zaanen, Phys. Rev. B 61, 6257 F. M. F. de Groot, P. Kuiper, and G. A. Sawatzky, Phys. Rev. B 57,
(2000). 14584 (1998).
40 66
G. Khaliullin and V. Oudovenko, Phys. Rev. B 56, R14243 (1997). J. Schlappa and T. Schmitt (private communication).
41 67
G. Khaliullin and S. Maekawa, Phys. Rev. Lett. 85, 3950 H. Fujisawa, T. Yokoya, T. Takahashi, S. Miyasaka, M. Kibune, and
(2000). H. Takagi, Phys. Rev. B 59, 7358 (1999).
42 68
K. Kikoin, O. Entin-Wohlman, V. Fleurov, and A. Aharony, Phys. Y.-J. Kim, J. P. Hill, H. Benthien, F. H. L. Essler, E. Jeckelmann,
Rev. B 67, 214418 (2003). H. S. Choi, T. W. Noh, N. Motoyama, K. M. Kojima, S. Uchida,
43
K. Wohlfeld, A. M. Oleś, and P. Horsch, Phys. Rev. B 79, 224433 D. Casa, and T. Gog, Phys. Rev. Lett. 92, 137402 (2004).
69
(2009). M. Z. Hasan, P. A. Montano, E. D. Isaacs, Z.-X. Shen, H. Eisaki,
44
A. Herzog, P. Horsch, A. M. Oleś, and J. Sirker, Phys. Rev. B 83, S. K. Sinha, Z. Islam, N. Motoyama, and S. Uchida, Phys. Rev.
245130 (2011). Lett. 88, 177403 (2002).
195138-20
70 75
R. Neudert, M. Knupfer, M. S. Golden, J. Fink, W. Stephan, F. Moussa, M. Hennion, J. Rodriguez-Carvajal, H. Moudden,
K. Penc, N. Motoyama, H. Eisaki, and S. Uchida, Phys. Rev. Lett. L. Pinsard, and A. Revcolevschi, Phys. Rev. B 54, 15149
81, 657 (1998). (1996).
71 76
J. Geck and V. Bisogni (private communication). K. Wohlfeld, M. Daghofer, A. M. Oleś, and P. Horsch, Phys. Rev.
72
J. W. Seo, K. Yang, D. W. Lee, Y. S. Roh, J. H. Kim, H. Eisaki, B 78, 214423 (2008).
77
H. Ishii, I. Jarrige, Y. Q. Cai, D. L. Feng, and C. Kim, Phys. Rev. B If we were to make a similar analysis as above but for the
73, 161104 (2006). spin sector of the ground state | being a Néel AF instead of
73
K. Wohlfeld, M. Daghofer, G. Khaliullin, and J. van den Brink, the quantum AF, we would notice that A and B2 processes are
J. Phys.: Conf. Ser. 391, 012168 (2012). not present in that case and it is only the B1 process which
74
J. H. Jefferson, H. Eskes, and L. F. Feiner, Phys. Rev. B 45, 7959 contributes. Thus, we see that in that case the LOW approach works
(1992). fine.
195138-21

Wohlfeld 2013

Uploaded by

Copyright:

Available Formats

Wohlfeld 2013

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Wohlfeld 2013

Uploaded by

Copyright:

Available Formats

PHYSICAL REVIEW B 88, 195138 (2013)

Microscopic origin of spin-orbital separation in Sr2 CuO3

DOI: 10.1103/PhysRevB.88.195138 PACS number(s): 75.25.Dk, 75.30.Ds, 71.10.Fd, 78.70.Ck

I. INTRODUCTION there were just three experimental indications of the existence

1098-0121/2013/88(19)/195138(21) 195138-1 ©2013 American Physical Society

for the onsite Coulomb repulsion on copper sites term

for the onsite Coulomb repulsion on oxygen sites the Hamiltonian

for the nearest-neighbor Coulomb repulsion

tσ 1.5 J1 0.088 Local excitation energies

(a) (b) (c)

(a) (b) tπ tσ (c)

Similarly to the b orbital case discussed above, the complex

J1 + J2c (iii) The high-spin state 3 T1 on oxygen sites which involves

IV. ORBITON SPECTRAL FUNCTIONS

we can write with

with an additional spin flip (the so-called B1 process in

RIXS cross section but also the quantitative changes are

transformations as for the Hamiltonian above, we obtain

FIG. 7. (Color online) Geometry of the RIXS experiment per-

Here, (i) orbital d = 3dyz , orbital e = 3d3z2 −r 2 , and the other

for its precise value).

with Since | is a vacuum for boson operators β| = 0, we easily

Table I. with the constants C and Jc defined as

where + (Ed + EAF ) (nid − nia )

You might also like

with Since | is a vacuum for boson operators β| = 0, we easily