1: %--------%---------%---------%---------%---------%---------%---------%---------%
2: % \documentclass[preprint,eqsecnum,aps,prb,epsf]{revtex4}
3: % \documentclass[preprint,eqsecnum,aps,prb]{revtex4}
4: \documentclass[twocolumn,eqsecnum,aps,prb,floatfix]{revtex4}
5: \usepackage[dvips]{graphicx}
6: % \documentstyle[preprint,eqsecnum,aps]{revtex}
7: % \documentclass[eqsecnum,aps,prb,epsf]{revtex4}
8: %
9: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
10: %
11: \providecommand{\mB}{\mu_{\rm B}}
12: \providecommand{\veck}{\mathbf{k}}
13: \providecommand{\ve}{\varepsilon}
14: \providecommand{\vecQ}{\mathbf{Q}}
15: \providecommand{\vecG}{\mathbf{G}}
16: \providecommand{\vecq}{\mathbf{q}}
17: \providecommand{\vecr}{\mathbf{r}}
18: \providecommand{\murakami}
19: {Murakami98a,Murakami98b,murakami99a,murakami99b,murakami99c,murakami00b}
20: \providecommand{\nmechanism}{Elfimov,Benfatto,Taka1,Taka2,Taka3,Benedetti00}
21: %
22: %--------%---------%---------%---------%---------%---------%---------%---------%
23: %
24: %\draft
25: %\preprint{}
26: %
27: %--------%---------%---------%---------%---------%---------%---------%---------%
28: %
29: \begin{document}
30: \title{Magnetic Resonant X-Ray Scattering in KCuF$_3$}
31: \author{Manabu Takahashi}
32: \affiliation{Faculty of Engineering, Gunma University, Kiryu, Gunma 376-8515, Japan}
33: \author{Manabu Usuda}
34: \author{Jun-ichi Igarashi}
35: \affiliation{Synchrotron Radiation Research Center, Japan Atomic Energy Research Institute,\\ Mikazuki, Sayo, Hyogo 679-5148, Japan}
36: \date{\today}
37: %
38: %--------%---------%---------%---------%---------%---------%---------%---------%
39: %
40: \begin{abstract}
41: We study the magnetic resonant x-ray scattering (RXS) spectra around the $K$
42: edge of Cu in KCuF$_3$ on the basis of an {\em ab initio} calculation. We use
43: the full-potential linearlized augmented plane wave method in the LDA$+U$
44: scheme, and introduce the lattice distortion as inputs of the calculation.
45: We obtain finite intensity on magnetic superlattice spots, about three
46: orders of magnitude smaller than on orbital superlattice spots, by taking
47: account of the spin-orbit interaction (SOI). No intensity appears without
48: the SOI, indicating that the intensity arises not from the spin
49: polarization but from the orbital polarization in $4p$ states. The present
50: calculation reproduces well the experimental spectra as functions of photon
51: energy and of azimuthal angle. We also calculate the RXS intensity on
52: orbital superlattice spots. It is found that the intensity increases with
53: increasing Jahn-Teller distortion. The spectra remain nearly the same in
54: the nonmagnetic state given by the simple LDA, in which the orbital
55: polarization in the $3d$ states is much smaller. This strongly suggests
56: that the intensity on orbital spots is mainly controlled by the lattice
57: distortion, not by the $3d$ orbital order itself.
58: \end{abstract}
59:
60: \pacs{78.70.Ck, 71.28.+d, 71.20.-b}
61:
62: \maketitle
63:
64: %
65: %--------%---------%---------%---------%---------%---------%---------%---------%
66: %\narrowtext
67:
68: \section{introduction}
69:
70: The resonant x-ray scattering (RXS) has attracted much attention, since the
71: orbital order can be directly probed by the RXS signals on the orbital
72: superlattice spots. Several RXS experiments have been carried out on
73: typical perovskite compounds, \cite{\murakami,Nakao00} in which an
74: orbitally ordered state may be stabilized below a critical
75: temperature.\cite{Kanamori,Khomskii72} For the $K$ edge in transition-metal
76: compounds, RXS intensities arise from the modulation of $4p$ states in the
77: intermediate states of the resonant process. Since $4p$ states are not the
78: states of orbital ordering, this causes complications on the interpretation
79: of RXS intensities.
80:
81: A mechanism was proposed that the modulation of $4p$ states comes from
82: orbitally polarized $3d$ states on the Mn$^{3+}$ ion through the
83: anisotropic terms of the $3d$--$4p$ Coulomb interaction in
84: LaMnO$_3$,\cite{Ishihara,Ishihara2} and thereby it was argued that the RXS
85: intensities on the superlattice spots are a direct reflection of orbital
86: order. On the other hand, subsequent studies based on band structure
87: calculations\cite{Elfimov,Benfatto,Taka1} have revealed that $4p$ states of
88: Mn$^{3+}$ ion so extend in space that they are considerably modified
89: by neighboring electronic states through the Jahn-Teller distortion (JTD).
90: It has been concluded that the effect of the JTD on the RXS spectra is
91: much larger than that of the $3d$-$4p$ Coulomb interaction.
92:
93: RXS experiments were carried out also on YVO$_3$[\onlinecite{murakami00b}] and
94: YTiO$_3$[\onlinecite{Nakao00}]. Since the JTD is considerably smaller in these
95: materials than in LaMnO$_3$, one may think the effect of the Coulomb
96: interaction important. This is not the case, though; {\em ab initio}
97: calculations indicated that the effect of lattice distortion on RXS
98: intensities is much larger than that of the $3d$-$4p$ Coulomb interaction
99: even in these materials. \cite{Taka2,Taka3} The calculation was carried out
100: within the muffin-tin (MT) approximation on the lattice parameters
101: determined from the experiment. Since the MT approximation averages the
102: potential coming from orbitally polarized $3d$ states, it works to
103: eliminate the effect of the anisotropic terms of the $3d$-$4p$ Coulomb
104: interaction on the $4p$ states. We found that the spectra consist of
105: several peaks, to which the GdFeO$_3$-type distortion and the JTD
106: contribute differently, in good agreement with the experiment.
107:
108: Recently, an RXS experiment has been carried out on KCuF$_3$, in which the
109: intensities on the magnetic and orbital superlattice spots have been
110: reported.\cite{Paolasini,Caciuffo} The magnetic RXS has been studied for
111: typical materials such as CoO[\onlinecite{Neubeck1}] and NiO[\onlinecite{Neubeck2}].
112: However, the analyses of the experimental data are limited to model
113: calculations.\cite{Iga1,Iga2} As regards the intensities on the orbital
114: spots, they increase with going down through the Ne\'el temperature in the
115: experiment, and are interpreted as a consequence of the increase of the $3d$
116: orbital order parameter in the magnetic phase.\cite{Paolasini} This
117: interpretation seems obscure unless it is clarified how the $3d$ orbital
118: order controls the RXS spectra. The purpose of this paper is to
119: elucidate the mechanism of the RXS spectra on the orbital and magnetic
120: superlattice spots through {\em ab initio} calculations. We use the
121: full-potential linearlized augmented plane wave (FLAPW) method in the
122: LDA$+U$ scheme, where the local $3d$--$3d$ Coulomb interaction is introduced on Cu
123: sites.\cite{Anisimov}
124:
125: We obtain the RXS intensity on magnetic spots by taking account of the
126: spin-orbit interaction (SOI). The calculation has to be accurate, since
127: the intensities are about three orders of magnitudes smaller than those on
128: orbital spots. This seems the first case to have evaluated the magnetic
129: RXS spectra in the {\em ab initio} level. We reproduce well the
130: experimental spectra as functions of photon energy and of azimuthal angle.
131: No intensity appears without the SOI, indicating that the intensity arises
132: from the orbital polarization, not from the spin polarization, in the $4p$
133: states. On the other hand, the spectra change little with turning off the
134: SOI from the $3d$ states. Since the $3d$ orbital moment disappears in this
135: condition, the $3d$ orbital moment plays no role to induce the orbital
136: polarization in $4p$ states. This shows a contrast with the magnetic RXS
137: in NiO and CoO, where the $3d$ orbital moment plays an important role
138: through the mixing of the $4p$ states with the $3d$ states of neighboring
139: Cu sites and the intra-atomic $4p$-$3d$ Coulomb interaction.
140:
141: We also calculate the RXS spectra on the orbital superlattice spots. A
142: similar calculation has already been carried out.\cite{Caciuffo} We come to
143: the same conclusion that the RXS intensity is mainly controlled by lattice
144: distortion. We add several aspects of making clear the mechanism by
145: calculating the spectra with changing magnitude of the JTD and in various
146: magnetic phases. We explicitly show that the magnetic order itself has
147: little influence on the intensity and that the intensity increases with
148: increasing JTD strength. In this respect, the increase of the intensity
149: with passing through the Ne\'el temperature has to accompany a considerable
150: increase of the JTD strength in the magnetic phase. We also predict a new
151: azimuthal angle dependence different from the previously assumed one but
152: consistent with the experiment.\cite{Caciuffo}
153:
154: In Sec.~II, we study electronic structures in various phases with the FLAPW
155: method in the LDA$+U$ scheme. In Sec~III, the absorption spectra and the
156: RXS spectra are presented in comparison with the experiment. Section IV is
157: devoted to concluding remarks.
158:
159: \section{electronic structures}
160: %--------+---------+---------+---------+---------+---------+---------+---------+
161: \begin{figure}
162: \includegraphics[width=7cm,keepaspectratio]{cryst.eps}
163: \caption{Sketch of the ordering pattern of KCuF$_3$.
164: Orbitals $\psi_{x^2-z^2}$ and $\psi_{y^2-z^2}$ are
165: antiferro-orbitally ordered with the ordering vector
166: $\vecQ_{\rm orb}=(\pi/a,\pi/b,\pi/c)$. Arrows represent the direction of
167: magnetic moments with the ordering vector $\vecQ_{\rm mag}=(0,0,\pi/c)$.
168: Small solid spheres represent F atoms.
169: K atoms are omitted in the figure.
170: }
171: \label{fig:crys}
172: \end{figure}
173: %--------+---------+---------+---------+---------+---------+---------+---------+
174: KCuF$_3$ has a typical perovskite structure, and undergoes a structural
175: phase transition at $T\approx 800$ K, below which the JTD and the
176: associated antiferro-orbital (AFO) long-range order emerges with the
177: ordering vector $\vecQ_{\rm orb}=(\pi/a,\pi/b,\pi/c)$. Figure \ref{fig:crys} shows the
178: schematic view of the ordering pattern. It is tetragonal with $a=b=4.141$\AA,
179: $c=3.924$\AA\ at room temperature; three different bond lengths are
180: given by $\ell=2.253$\AA, $s=1.888$\AA, $m=1.962$\AA.\cite{Buttner,Hutings}
181: Since the lattice parameters are not known at
182: low temperatures, we use these values in the following calculation. There
183: are two inequivalent Cu sites in a unit cell, which will be called as A and
184: B sites. With further decreasing temperatures, an antiferromagnetic (AFM)
185: long-range order develops below 38 K with the ordering vector $\vecQ_{\rm
186: mag}=(0,0,\pi/c)$. Therefore four inequivalent Cu sites, A$_+$, A$_-$,
187: B$_+$, B$_-$, exist. The local magnetic moment is pointing to either the
188: $(110)$ or $(1\bar1 0)$ direction with the value 0.48 $\mB$. The recent
189: magnetic scattering experiment gives the ratio of the orbital angular
190: momentum ($L$) to the spin angular momentum ($S$) as
191: $L/S=0.29$.\cite{Caciuffo}
192:
193: We carry out the band structure calculation in the LDA$+U$ scheme by setting
194: the parameters of the intraatomic Coulomb interaction between $3d$ orbitals
195: such that $U=3.0$ eV and $J=1.0$ eV. We take account of the SOI to deal
196: with the orbital moment, which is important for the RXS spectra on the
197: magnetic spots. First assuming the AFM long-range order, we obtain an
198: orbital long-range order with the ordering vector $\vecQ_{\rm orb}$; the
199: {\em hole} numbers inside the Cu MT sphere in the $3d$ states $\varphi_{x^2-z^2}$
200: ($\varphi_{y^2-z^2}$) and $\varphi_{3y^2-r^2}$ ($\varphi_{3x^2-r^2}$) are
201: given by $n_{x^2-z^2}(n_{y^2-z^2})=0.65$ $n_{3y^2-r^2}(n_{3x^2-r^2})=0.08$
202: in A$_\pm$ (B$_\pm$) sites, respectively. Note that the difference in the
203: values between A$_+$ and A$_-$ sites and B$_+$ and B$_-$ sites is
204: negligibly small. The local magnetic moment is obtained as $0.79\mB$
205: with $S=0.34$, $L=0.11$, $L/S=0.33$, which is somewhat larger than the
206: experimental one. In order to see the effect of the magnetic order on
207: orbital polarization, we carry out the same LDA$+U$ calculation in the
208: presence of the ferromagnetic (FM) long-range order. We obtain the same
209: type of orbital order, in which $n_{x^2-z^2}(n_{y^2-z^2})=0.66$
210: $n_{3y^2-r^2}(n_{3x^2-r^2})=0.10$ in A (B) sites, respectively. Since
211: the hole numbers are similar to that of the antiferromagnetic state, we
212: conclude that the magnetic order has little influence on the orbital order
213: parameter. On the other hand, the orbital polarization is rather sensitive
214: to the intraatomic Coulomb interaction. The simple LDA ($U=0$) gives the
215: orbital polarization much smaller than the values mentioned above, that is,
216: $n_{x^2-z^2}(n_{y^2-z^2})=0.26$ and $n_{3y^2-r^2}(n_{3x^2-r^2})=0.11$ per
217: spin in the A(B) sites in the nonmagnetic (NM) phase.
218:
219: %--------+---------+---------+---------+---------+---------+---------+---------+
220: \begin{figure}
221: \includegraphics[width=7cm,keepaspectratio]{KCF3cudos.eps}
222: \caption{Density of states projected on the $s$, $p$ and $d$
223: symmetric states inside the Cu muffin-tin sphere,
224: calculated in the AFM phase of the LDA$+U$ scheme.
225: The origin of energy is the bottom of the conduction band.}
226: \label{fig:doscalc}
227: \end{figure}
228: %--------+---------+---------+---------+---------+---------+---------+---------+
229: Figure \ref{fig:doscalc} shows the calculated density of states (DOS)
230: projected on the $d$ and $p$ symmetries of a Cu site, which is defined
231: inside the MT sphere, in the AF phase given by the LDA$+U$ method. The
232: origin of the energy is the bottom of the conduction band. The DOS projected
233: onto the $d$ symmetric states is almost concentrated in the region of
234: energy less than 2 eV. It has a finite gap $\sim 0.82$ eV. The left part of
235: DOS is shown in a 20 times magnified scale. The most part of probability
236: distributes in the interstitial region or on neighboring MT spheres, and
237: thereby the magnitude of the DOS projected onto the $p$ symmetry becomes
238: much smaller than the $3d$ DOS. We refer to the $p$ symmetric states as
239: $4p$ states, because the wave functions are quite close to the atomic $4p$
240: wave functions within the MT sphere. Of course the ``4p" states lose the
241: atomic character outside the MT sphere, forming an energy band with its
242: width being as large as $30$ eV. The $4p$ DOS shows almost no exchange
243: splitting in spite of the magnetic ground state, indicating that the $3d$
244: states have little influence on $4p$ states.
245:
246: \section{Absorption Coefficient and RXS spectra}
247:
248: We define the $1s$--$4p$ dipole transition density matrix
249: $\tau^{(j)}_{mm'}(\ve)$ assigned to each Cu site as
250: \begin{eqnarray}
251: \tau^{(j)}_{mm'}(\ve) &=& \sum_{n,\veck}\int r^2 dr\, r'^2 dr' [R_{j1s}^*(r ) r {\cal P}^{(j)}_m\phi_{n,\veck}(r )]^* \nonumber \\
252: &\times& [R_{j1s}^*(r') r' {\cal P}^{(j)}_{m'}\phi_{n,\veck}(r')] \delta(\ve-\ve_{n,\veck}),
253: \label{transition}
254: \end{eqnarray}
255: where $j$ stands for sublattices A$_\pm$ or B$_\pm$. Suffixes $m$, $m'$
256: represent the $x$, $y$, and $z$ axes,
257: which are chosen to be parallel to the $a$, $b$, and $c$ axes,
258: respectively (see Fig. \ref{fig:crys}).
259: The $\phi_{n,\veck}$ represents the wave function with the band index
260: $n$, wave-vector $\veck$ and energy $\ve_{n,\veck}$ which is
261: larger than the Fermi energy. Operator ${\cal P}^{(j)}_m$ projects the wave
262: function $\phi_{n,\veck}$ on the $m$ ($m\in x,y,z$) component of the
263: $p$ symmetric part on the Cu $j$ site. The $R_{j1s}$ represents the Cu
264: $1s$ wave function on the Cu $j$ site. We evaluate this density matrix using the FLAPW
265: wave function.
266:
267: \subsection{Absorption Coefficient}
268:
269: We first discuss the absorption
270: coefficient $A(\omega)$ around the $K$ edge.
271: Neglecting the core-hole potential working on the $4p$ states
272: in the final state of the absorption process,
273: we have the expression for $A(\omega)$,
274: \begin{eqnarray}
275: &&A(\omega)\propto \sum_{n,\veck}\left|\int r^2 dr R_{1s}^*(r ) r
276: {\cal P}_m\phi_{n,\veck}(r )\right|^2 \delta(\omega-\ve_{n,\veck}-\ve_{1s})
277: \nonumber \\
278: &&= \tau^{(j)}_{xx}(\omega-\ve_{1s})+\tau^{(j)}_{yy}(\omega-\ve_{1s})
279: +\tau^{(j)}_{zz}(\omega-\ve_{1s}) ,
280: \label{eq.abs}
281: \end{eqnarray}
282: where $\ve_{1s}$ is the energy of the $1s$ state.
283: The dependence on sublattices can be neglected.
284: %Since the factor before the $\delta$-function in Eq.~(\ref{eq.abs})
285: %is nearly independent of $n$ and $\veck$,
286: $A(\omega)$ is nearly proportional to the $4p$ DOS.
287: Figure \ref{fig:absorp} shows the calculated result
288: in comparison with the experiment,
289: where $\ve_{1s}$ is adjusted such that
290: the position of the main peak coincides with the experimental one.
291: It is difficult to determine the absolute value of $\ve_{1s}$
292: within the present status of the {\em ab initio} calculation,
293: since several screening processes are neglected.
294: The spectral shape is found in good agreement with the experimental curve.
295: If the core-hole potential would be taken into account,
296: the intensities in low frequency region may be further enhanced.
297: %--------+---------+---------+---------+---------+---------+---------+---------+
298: \begin{figure}
299: \includegraphics[width=7cm,keepaspectratio]{KCF3abs.eps}
300: \caption{
301: Absorption coefficient $A(\omega)$ as a function of photon energy
302: around the $K$ edge of Cu, in comparison with the experimental data of
303: fluorescence (Ref. \onlinecite{Caciuffo}).
304: The core-hole energy is adjusted such that the calculated peak position
305: coincides with the experimental one.
306: }
307: \label{fig:absorp}
308: \end{figure}
309: %--------+---------+---------+---------+---------+---------+---------+---------+
310:
311: \subsection{RXS Spectra}
312: %--------+---------+---------+---------+---------+---------+---------+---------+
313: \begin{figure}
314: \includegraphics[width=7cm,keepaspectratio]{geom.eps}
315: \caption{RXS geometry. Incident photon with wave vector $\vecq_i$
316: and polarization $\sigma$ or $\pi$ is scattered into the state with
317: wave vector $\vecq_f$ and polarization $\sigma'$ or $\pi'$ at Bragg angle
318: $\theta$. The sample crystal is rotated by azimuthal angle $\psi$ around
319: the scattering vector $\vecG$.
320: }
321: \label{fig:geometry}
322: \end{figure}
323: %--------+---------+---------+---------+---------+---------+---------+---------+
324:
325: The conventional RXS geometry is shown in Fig.~\ref{fig:geometry};
326: photon with frequency $\omega$, momentum $\vecq_i$,
327: polarization $\mu$ ($=\sigma$ or $\pi$) is scattered into state with
328: momentum $\vecq_f$ and polarization $\mu'$ ($=\sigma'$ or $\pi'$).
329: The scattering vector $\vecG$ is defined by $\vecq_f-\vecq_i$.
330: In such a situation, we have the RXS intensity $I(\vecG,\omega)$
331: \begin{equation}
332: I(\vecG,\omega)\propto \left| \sum_{mm'}E_m^{\rm out}
333: M_{mm'}(\vecG,\omega)E_{m'}^{\rm in}\right|^2,
334: \end{equation}
335: with
336: \begin{eqnarray}
337: && M_{mm'}(\vecG,\omega) = \nonumber \\
338: &&\frac{1}{\sqrt{N}}\sum_{j,\Lambda}
339: \frac{\langle g|x_m(j)|\Lambda\rangle
340: \langle\Lambda|x_m'(j)|g\rangle}
341: {\hbar\omega-(E_{\Lambda}-E_g)+i\Gamma}\exp(-i\vecG\cdot\vecr_j). \nonumber \\
342: \end{eqnarray}
343: Here $E^{\mbox{\scriptsize in(out)}}_x$ is the $x$ component
344: of polarization vector of the incident (scattered) photon,
345: and $j$ runs over Cu lattice sites.
346: The $|g\rangle$ represents the ground state with energy $E_g$.
347: The $|\Lambda\rangle$ represents the intermediate state with energy $E_\Lambda$;
348: it consists of the excited electron on the $4p$ states and a hole
349: on the $1s$ state. $\Gamma$ describes the broadening due to the $1s$
350: core-hole lifetime. The RXS spectra are rather insensitive with
351: varying $\Gamma$ values; we set ${\Gamma}=1$ eV.
352: The dipole operators $x_\alpha(j)$ at site $j$ are defined as
353: $x_1(j)=x$, $x_2(j)=y$, $x_3(j)=z$ in the coordinate frame fixed to
354: the crystal axes with the origin located at the center of site $j$.
355: Hereafter $\vecG$ is expressed in unit of $(2\pi/2a,2\pi/2b,2\pi/2c)$.
356: Just like the absorption coefficient, we neglect the core-hole potential
357: in the intermediate state. Then the amplitude $M_{mm'}(\vecG,\omega)$
358: is expressed in terms of the density matrices:
359: \begin{eqnarray}
360: && M_{mm'}(\vecG,\omega) = \nonumber \\
361: && \frac{1}{\sqrt{N}}\sum_j \exp(-i\vecG\cdot\vecr_j)
362: \int \frac{\tau^{(j)}_{mm'}(\ve){\rm d}\ve}
363: {\hbar\omega-\ve+\ve_{1s}+i\Gamma}.
364: \label{eq.amp}
365: \end{eqnarray}
366:
367: \subsubsection{on Magnetic Superlattice Spots}
368: %--------+---------+---------+---------+---------+---------+---------+---------+
369: \begin{figure}
370: \includegraphics[width=7cm,keepaspectratio]{KCF3rxs005_888.eps}
371: \caption{Calculated RXS spectra on a magnetic superlattice spot $(001)$
372: in comparison with the experiment (Ref. \onlinecite{Caciuffo}).
373: The core-hole energy is set the same as in the case of absorption coefficient
374: (Fig. \ref{fig:absorp}).
375: }
376: \label{fig:RXSmag}
377: \end{figure}
378: %--------+---------+---------+---------+---------+---------+---------+---------+
379: For the magnetic superlattice spots $\vecG=(00m)$
380: with $m$ being odd integers, Eq.(\ref{eq.amp}) contains the following
381: combination of density matrices:
382: \begin{eqnarray}
383: && \tau^{({\rm A}_+)}(\ve)- \tau^{({\rm A}_-)}(\ve)
384: + \tau^{({\rm B}_+)}(\ve)-\tau^{({\rm B}_-)}(\ve) \nonumber \\
385: &&=\left( \begin{array}{rrr} 0 & 0 &\eta(\ve) \\
386: 0 & 0 &\eta(\ve) \\
387: -\eta(\ve) &-\eta(\ve) & 0
388: \end{array} \right).
389: \end{eqnarray}
390: Only the off-diagonal elements survives with an antisymmetric form.
391: \cite{Iga1,Iga2}
392: They would vanish without the SOI, indicating that they arise from
393: the {\em orbital polarization}\footnote{
394: The orbital polarization in the $4p$ states here means the polarization
395: related to imaginary wavefunctions, which gives rise to orbital moment.
396: On the other hand, the $4p$ orbital polarization for the case of orbital
397: superlattice spots means the polarization with respect to real wavefunctions
398: such as $x$, $y$, $z$ types.}
399: in the $4p $ states.
400: From this expression, Eq.(\ref{eq.amp}) becomes
401: \begin{equation}
402: I(\vecG,\omega) \propto \left|
403: (C_{yz}+C_{zx})\int d\ve\frac{\eta(\ve)}
404: {\omega-\ve+\ve_{1s}+{\rm i}{\Gamma}}\right|^2,
405: \label{eq.RXSmag}
406: \end{equation}
407: with $C_{ij}= E^{\rm out}_i E^{\rm in}_j - E^{\rm out}_j E^{\rm in}_i$.
408: Since the polarization dependent part $C_{yz}+C_{zx}$ is factored out,
409: the photon energy dependence is independent of polarization.
410:
411: The actual calculation in the LDA$+U$ scheme is rather heavy, since the
412: number of atoms in the unit cell becomes as large as 20 in the AFM phase.
413: Highly accurate calculations are required, since the intensity is about
414: three orders of magnitude smaller than that for the orbital superlattice
415: spots, as shown below. It is not easy to sum up many k-points in
416: evaluating $\eta(\ve)$, though. Fortunately, summing $8\times 8\times 8$
417: k-points in the first Brillouin zone seems sufficient in comparison with
418: the result for summing $6\times 6\times 6$ k-points. Figure
419: \ref{fig:RXSmag} shows the calculated spectra for $\vecG=(001)$ as a
420: function of photon energy in comparison with the experiment.
421: \cite{Caciuffo} The intensity in the the $\sigma\to\sigma'$ channel does
422: not appear in agreement with the experiment. Its spectral shape consists
423: of several peaks. The highest and next highest peaks correspond well to
424: peaks a and b, although their energies are slightly shifted. Note that the
425: core-hole energy has already been adjusted such that the calculated
426: absorption peak coincides with the experimental one. We have checked that
427: the spectra change little with turning off the SOI from the $3d$ states.
428: This suggests that the $3d$ orbital moment has little influence on inducing
429: the $4p$ orbital polarization.
430:
431: From Eq.~(\ref{eq.RXSmag}), we derive explicitly
432: the azimuthal angle dependence,
433: \begin{eqnarray}
434: I(\vecG,\omega) &\propto& 0 \quad
435: {\rm for}\ \sigma\to\sigma' \nonumber \\
436: &\propto& |\cos\theta\cos\psi|^2 \quad
437: {\rm for}\ \sigma\to\pi',
438: \end{eqnarray}
439: where $\sin\theta=0.088$ for the $(001)$ spot and $\sin\theta=0.439$
440: for the $(005)$ spot.
441: Here the azimuthal angle $\psi=0$ is defined such that the $(\bar110)$ axis is contained
442: in the scattering plane. Figure \ref{fig:azim} shows the peak intensity
443: on $(005)$ as a function of $\psi$, in comparison with the experiment.
444: The dependence agrees with the experiment.
445: %--------+---------+---------+---------+---------+---------+---------+---------+
446: \begin{figure}
447: \includegraphics[width=7cm,keepaspectratio]{KCF3azm.eps}
448: \caption{Azimuthal angle dependence of the peak intensity
449: on a magnetic superlattice spot $(005)$ and on an orbital superlattice
450: spot $(331)$, in comparison with the experiment (Ref. \onlinecite{Caciuffo}).
451: }
452: \label{fig:azim}
453: \end{figure}
454: %--------+---------+---------+---------+---------+---------+---------+---------+
455:
456: \subsubsection{on Orbital Superlattice Spots}
457:
458: For the orbital superlattice spots $\vecG=(h\ell m)$
459: with $h$, $l$, and $m$ being odd integers, Eq.~(\ref{eq.amp}) contains
460: the following combination of density matrices:
461: \begin{eqnarray}
462: && \tau^{({\rm A}_+)}(\ve)+ \tau^{({\rm A}_-)}(\ve)
463: -\tau^{({\rm B}_+)}(\ve)-\tau^{({\rm B}_-)}(\ve) \nonumber \\
464: &&\approx\left( \begin{array}{rrr} \alpha(\ve)-\beta(\ve) &0 &\gamma(\ve) \\
465: 0 &\beta(\ve)-\alpha(\ve) &\delta(\ve) \\
466: -\gamma(\ve) & -\delta(\ve) & 0
467: \end{array} \right).
468: \end{eqnarray}
469: The off-diagonal elements are about two orders of magnitude smaller than
470: the diagonal terms.
471: Neglecting such terms, we obtain
472: \begin{equation}
473: I(\vecG,\omega) \propto \left|C_{x^2-y^2}
474: \int d\ve\frac{\alpha(\ve)-\beta(\ve)}
475: {\omega-\ve+\ve_{1s}+ i{\Gamma}}\right|^2,
476: \label{eq.RXSorb}
477: \end{equation}
478: with
479: \begin{equation}
480: C_{x^2-y^2}=E^{\mbox{\scriptsize out}}_x E^{\mbox{\scriptsize in}}_x
481: -E^{\mbox{\scriptsize out}}_y E^{\mbox{\scriptsize in}}_y.
482: \label{eq.azim}
483: \end{equation}
484: Just like Eq.~(\ref{eq.RXSmag}), the polarization dependent part $C_{x^2-y^2}$
485: is factored out, so that the photon energy dependence becomes independent
486: of polarization.
487:
488: %--------+---------+---------+---------+---------+---------+---------+---------+
489: \begin{figure}
490: \includegraphics[width=7cm,keepaspectratio]{KCF3rxs331.eps}
491: \caption{Calculated RXS spectra on an orbital superlattice spot $(331)$
492: in comparison with the experiment.
493: The thick solid and broken lines show the spectra on the AFM and FM phases
494: in the LDA$+U$ scheme, respectively. The thin solid line shows the spectra
495: on the NM phase in the LDA scheme. The dotted line shows the experimental
496: data (Ref.[\onlinecite{Caciuffo}]).
497: The core-hole energy is set the same as in the case of absorption coefficient.
498: }
499: \label{fig:RXSorb}
500: \end{figure}
501: %--------+---------+---------+---------+---------+---------+---------+---------+
502: Figure \ref{fig:RXSorb} shows the calculated spectrum for $\vecG=(331)$,
503: as a function of photon energy, in comparison with the experiment. The
504: calculation is carried out on both the AFM and FM phases in the LDA$+U$
505: scheme, and on the NM phase in the simple LDA scheme. All three cases of
506: calculation give nearly the same result, indicating that the $4p$ states
507: are little influenced by the details of the $3d$ states, in agreement of
508: the previous calculation of the same kind. \cite{Caciuffo} In particular,
509: the fact that the last case gives the similar result in spite of
510: considerably smaller orbital polarizations gives a strong support to the
511: mechanism that the $3d$ orbital polarization has little relation to the RXS
512: intensity. This is consistent with previous studies on other
513: transition-metal compounds, \cite{Elfimov,Benfatto,Taka1} where the RXS
514: intensity are mainly controlled by lattice distortion. The effect of the
515: SOI is negligibly small for the these cases. The present calculation
516: reproduce well the spectra consisting of three peaks, although peaks c and
517: e are much smaller than the experimental ones.
518:
519: The azimuthal angle dependence for $\vecG=(331)$ is explicitly evaluated
520: from Eq.(\ref{eq.azim}). It is given by
521: \begin{eqnarray}
522: I(\vecG,\omega) &\propto& |\cos\beta\sin 2\psi|^2 \quad
523: {\rm for}\ \sigma\to\sigma' \nonumber \\
524: &\propto& |\sin\theta\cos\beta\cos 2\psi
525: + \cos\theta\sin\beta\sin\psi|^2 \nonumber \\
526: &&\quad {\rm for}\ \sigma\to\pi',
527: \end{eqnarray}
528: with $\sin\theta=0.364$, $\tan\beta=3\sqrt{2}c/a=4.02$.
529: The azimuthal angle $\psi=0$ is again defined such that the $(\bar110)$ axis
530: is contained in the scattering plane.
531: The intensity has a period of $2\pi$ not $\pi$ with respect to $\psi$.
532: This rather complex behavior seems reasonable, since the scattering vector
533: $\vecG$ is not in a direction of high symmetry.
534: This is obviously different from the form previously assumed in the analysis
535: of the experimental data.\cite{Caciuffo}
536: Figure \ref{fig:azim} shows the peak intensity
537: as a function of $\psi$, in comparison with the experiment.
538: The present form is consistent with the experiment.
539:
540: So far, all the spectra are calculated with the lattice parameters at room
541: temperature mentioned before. If the RXS intensity arises from lattice
542: distortion, it increases with increasing JTD strength. To confirm this, we
543: calculate the intensity with changing the JTD strength. Figure
544: \ref{fig:RXSjtd} shows the intensity of the peak at $\hbar\omega=8995$ eV
545: as a function of $\ell-s$ (the values of $m$ and $\ell+s$ are kept to be
546: $1.962$\AA\ and $4.141$\AA, respectively). The intensity monotonically
547: increases with increasing values of $\ell-s$, consistent with the above
548: observation. It has been observed in the experiment that the intensity
549: increases by factor 2 with temperature going through the magnetic phase
550: transition temperature, and this behavior has been interpreted as a
551: consequence of a strong coupling between orbital and spin degrees of
552: freedom. \cite{Paolasini} The result that the spectra are mainly
553: controlled by lattice distortion indicates that the lattice distortion
554: becomes large in the magnetic phase. The direct determination of lattice
555: parameters have not been carried out yet.
556: %--------+---------+---------+---------+---------+---------+---------+---------+
557: \begin{figure}
558: \includegraphics[width=7cm,keepaspectratio]{KCF3rxs_i-q2.eps}
559: \caption{Peak intensity of RXS spectra on $(331)$ with varying JTD strength.
560: The values of $m$ and $\ell+s$ are fixed at $1.962$\AA\ and $4.141$\AA,
561: respectively. The circle indicated by an arrow is the result for the actual
562: value of $\ell-s$ at room temperature.
563: }
564: \label{fig:RXSjtd}
565: \end{figure}
566: %--------+---------+---------+---------+---------+---------+---------+---------+
567:
568: \section{Concluding Remarks}
569:
570: We have calculated the RXS spectra around the $K$ edge of Cu in KCuF$_3$ on
571: the basis of the {\em ab initio} calculation, that is, the FLAPW method in the
572: LDA$+U$ scheme. The lattice distortion is introduced as inputs of
573: calculation. We have obtained finite intensities on magnetic superlattice
574: spots by taking account of the SOI. The intensities are about three orders
575: of magnitude smaller than those on orbital superlattice spots, consistent
576: with the experiment. This is the first result to have evaluated the
577: magnetic RXS spectra in the {\em ab initio} level. Since the intensity
578: disappears without the SOI, the present result indicates that the spectra
579: arises from the orbital polarization, not from the spin polarization, in
580: the $4p$ states. On the other hand, the spectra change little with turning
581: off the SOI from the $3d$ states. This indicates that the $3d$ orbital
582: moment has no role to induce the $4p$ orbital polarization. This situation
583: is different from our previous finding in the study of the magnetic RXS in
584: CoO and NiO that the $4p$ orbital polarization is induced by the $3d$
585: orbital moment through the mixing of the $4p$ states with the $3d$ states
586: of {\em neighboring} Cu atoms and the intra-atomic $4p$-$3d$ Coulomb
587: interaction.\cite{Iga1,Iga2} This difference may come from the fact that
588: the $3d$ orbital moment is about 1/3 of that in NiO and 1/10 of that in
589: CoO. Furthermore, the mixing effect may become smaller than in NiO and CoO,
590: since neighboring Cu sites are always intervened by F atoms. Closely
591: related is the phenomenon of the magnetic circular dichroism (MCD) in the
592: $K$-edge absorption in ferromagnetic metals Fe, Co, Ni.\cite{Schutz} In this
593: case, the $4p$ orbital polarization was found to be induced by the $3d$
594: orbital moment through the $4p$ mixing with the $3d$ states at {\em
595: neighboring} sites.\cite{Iga3,Iga4}
596:
597: We have also calculated the RXS intensities on orbital superlattice spots.
598: The spectra are independent of whether the system is in the AFM phase or
599: in the FM phase, and of the orbital polarization in the $3d$ states. We
600: have explicitly shown that the spectra are mainly controlled by the lattice
601: distortion through the calculation with varying JTD strength. The present
602: result is consistent with the previous studies of the RXS on
603: transition-metal oxides,\cite{Elfimov,Benfatto,Taka1,Taka2,Taka3} but shows
604: contrast with the RXS around the $L_{\rm III}$ edge in the quadrupole
605: ordering phase of the rare-earth compound CeB$_6$, where the spectra seems
606: directly controlled by the quadrupole order. \cite{Nakao01,Nagao,Nagao2}
607:
608: \begin{acknowledgments}
609: We have used the FLAPW code developed by N. Hamada. We
610: thank him for allowing to use his code. This work was partially supported
611: by a Grant-in-Aid for Scientific Research from the Ministry of Education,
612: Culture, Sports, Science, and Technology, Japan.
613: \end{acknowledgments}
614:
615: \bibliography{p3}
616:
617: \newpage
618: %
619: %--------%---------%---------%---------%---------%---------%---------%---------%--------%---------%---------%---------%---
620: %
621: \end{document}
622: \newpage
623: %--------%---------%---------%---------%---------%---------%---------%---------%--------%---------%---------%---------%---
624: %%
625: %%
626: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
627: % %
628: % F I G U R E S %
629: % %
630: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
631: %
632: %--------+---------+---------+---------+---------+---------+---------+---------+
633: \begin{figure}
634: \includegraphics[width=7cm,keepaspectratio]{cryst.eps}
635: \caption{Sketch of the ordering pattern of KCuF$_3$.
636: Orbitals $\psi_{x^2-z^2}$ and $\psi_{y^2-z^2}$ are
637: antiferro-orbitally ordered with the ordering vector
638: $\vecQ_{\rm orb}=(\pi/a,\pi/b,\pi/c)$. Arrows represent the direction of
639: magnetic moments with the ordering vector $\vecQ_{\rm mag}=(0,0,\pi/c)$.
640: Small solid spheres represent F atoms.
641: K atoms are omitted in the figure.
642: }
643: \label{fig:crys}
644: \end{figure}
645: %--------+---------+---------+---------+---------+---------+---------+---------+
646: \begin{figure}
647: \includegraphics[width=7cm,keepaspectratio]{KCF3cudos.eps}
648: \caption{Density of states projected on the $s$, $p$ and $d$
649: symmetric states inside the Cu muffin-tin sphere,
650: calculated in the AFM phase of the LDA$+U$ scheme.
651: The origin of energy is the bottom of the conduction band.}
652: \label{fig:doscalc}
653: \end{figure}
654: %--------+---------+---------+---------+---------+---------+---------+---------+
655: \begin{figure}
656: \includegraphics[width=7cm,keepaspectratio]{KCF3abs.eps}
657: \caption{
658: Absorption coefficient $A(\omega)$ as a function of photon energy
659: around the $K$ edge of Cu, in comparison with the experimental data of
660: fluorescence (Ref. \onlinecite{Caciuffo}).
661: The core-hole energy is adjusted such that the calculated peak position
662: coincides with the experimental one.
663: }
664: \label{fig:absorp}
665: \end{figure}
666: %--------+---------+---------+---------+---------+---------+---------+---------+
667: \begin{figure}
668: \includegraphics[width=7cm,keepaspectratio]{geom.eps}
669: \caption{RXS geometry. Incident photon with wave vector $\vecq_i$
670: and polarization $\sigma$ or $\pi$ is scattered into the state with
671: wave vector $\vecq_f$ and polarization $\sigma'$ or $\pi'$ at Bragg angle
672: $\theta$. The sample crystal is rotated by azimuthal angle $\psi$ around
673: the scattering vector $\vecG$.
674: }
675: \label{fig:geometry}
676: \end{figure}
677: %--------+---------+---------+---------+---------+---------+---------+---------+
678: \begin{figure}
679: \includegraphics[width=7cm,keepaspectratio]{KCF3rxs005_888.eps}
680: \caption{Calculated RXS spectra on a magnetic superlattice spot $(001)$
681: in comparison with the experiment (Ref. \onlinecite{Caciuffo}).
682: The core-hole energy is set the same as in the case of absorption coefficient
683: (Fig. \ref{fig:absorp}).
684: }
685: \label{fig:RXSmag}
686: \end{figure}
687: %--------+---------+---------+---------+---------+---------+---------+---------+
688: \begin{figure}
689: \includegraphics[width=7cm,keepaspectratio]{KCF3azm.eps}
690: \caption{Azimuthal angle dependence of the peak intensity
691: on a magnetic superlattice spot $(005)$ and on an orbital superlattice
692: spot $(331)$, in comparison with the experiment (Ref. \onlinecite{Caciuffo}).
693: }
694: \label{fig:azim}
695: \end{figure}
696: %--------+---------+---------+---------+---------+---------+---------+---------+
697: \begin{figure}
698: \includegraphics[width=7cm,keepaspectratio]{KCF3rxs331.eps}
699: \caption{Calculated RXS spectra on an orbital superlattice spot $(331)$
700: in comparison with the experiment.
701: The thick solid and broken lines show the spectra on the AFM and FM phases
702: in the LDA$+U$ scheme, respectively. The thin solid line shows the spectra
703: on the NM phase in the LDA scheme. The dotted line shows the experimental
704: data (Ref.[15]).
705: The core-hole energy is set the same as in the case of absorption coefficient.
706: }
707: \label{fig:RXSorb}
708: \end{figure}
709: %--------+---------+---------+---------+---------+---------+---------+---------+
710: \begin{figure}
711: \includegraphics[width=7cm,keepaspectratio]{KCF3rxs_i-q2.eps}
712: \caption{Peak intensity of RXS spectra on $(331)$ with varying JTD strength.
713: The values of $m$ and $\ell+s$ are fixed at $1.962$\AA\ and $4.141$\AA,
714: respectively. The circle indicated by an arrow is the result for the actual
715: value of $\ell-s$ at room temperature.
716: }
717: \label{fig:RXSjtd}
718: \end{figure}
719: %--------+---------+---------+---------+---------+---------+---------+---------+
720: \end{document}
721: