1: \documentclass [prb,superscriptaddress,floatfix,showpacs] {revtex4}
2: %\documentclass [superscriptaddress,floatfix,showpacs,draft] {revtex4}
3:
4: \usepackage {amsmath,amssymb,epsfig,color}
5: %\documentstyle[prb,aps,psfig,graphicx,epsf,multicol,tabularx]{revtex}
6:
7: \def\red{\color{red}}
8:
9: \begin{document}
10:
11: \title
12: {Crystal-field splitting for low symmetry systems in ab initio calculations}
13:
14: \author{S.V.~Streltsov}
15: \affiliation{Ural State Technical University, Mira St. 19, 620002 Ekaterinburg, Russia}
16: \affiliation{Institute of Metal Physics, S.Kovalevskoy St. 18, 620219 Ekaterinburg GSP-170, Russia}
17: \email{streltsov@optics.imp.uran.ru}
18:
19: \author{A.S.~Mylnikova}
20: \affiliation{Ural State Technical University, Mira St. 19, 620002 Ekaterinburg, Russia}
21: \affiliation{Institute of Metal Physics, S.Kovalevskoy St. 18, 620219 Ekaterinburg GSP-170, Russia}
22: \author{A.O.~Shorikov}
23: \affiliation{Institute of Metal Physics, S.Kovalevskoy St. 18, 620219 Ekaterinburg GSP-170, Russia}
24: \author{Z.V.~Pchelkina}
25: \affiliation{Institute of Metal Physics, S.Kovalevskoy St. 18, 620219 Ekaterinburg GSP-170, Russia}
26: \author{D.I.~Khomskii}
27: \affiliation{II. Physikalisches Institut, Universit$\ddot a$t zu K$\ddot o$ln,
28: Z$\ddot u$lpicher Stra$\ss$e 77, D-50937 K$\ddot o$ln, Germany}
29: \affiliation{Groningen University, Nijenborgh 4, 9747 AG Groningen, The
30: Netherlands}
31: \author{V.I.~Anisimov}
32: \affiliation{Institute of Metal Physics, S.Kovalevskoy St. 18, 620219 Ekaterinburg GSP-170, Russia}
33:
34:
35:
36: \date{\today}
37:
38: \begin{abstract}
39: In the framework of the LDA+U approximation we propose the direct way
40: of calculation of crystal-field excitation energy and
41: apply it to La and Y titanates. The method developed can be
42: useful for comparison with the results of spectroscopic measurements
43: because it takes into account fast relaxations of electronic system.
44: For titanates these relaxation processes reduce the value of
45: crystal-field splitting by $\sim30\%$ as compared with
46: the difference of LDA one electron energies. However, the
47: crystal-field excitation energy in these systems is still large enough
48: to make an orbital liquid formation rather unlikely
49: and experimentally observed isotropic magnetism remains
50: unexplained.
51: \end{abstract}
52:
53: \pacs{71.15.-m, 71.20.-b, 71.30.+h}
54: \maketitle
55:
56:
57: \section{Introduction}
58: \label{intro}
59: The magnitude of crystal-field splitting (CFS) is known to be very
60: important characteristic of transition metal compounds, necessary
61: for understanding different physical phenomena such as magnetism
62: and metal-insulator transitions.\cite{Goodenough,Ballhausen}
63: It is
64: often used for detailed fitting of spectroscopic data and in
65: various model calculations. An important question is: how can one
66: calculate CFS in a reliable way?
67:
68: One can obtain character of CFS using the group theory analysis.
69: It provides qualitative description of ionic levels structure in
70: the presence
71: of crystal-field of given symmetry and can even give some ratios between
72: splittings using Wigner-Eckart theorem.\cite{Elliott}
73: However, for quantitative estimates the detailed structure of ionic
74: potentials has to be taken into account.
75:
76: Generally, there are different contributions to the total value of
77: CFS to be considered. A point-charge
78: contribution can be calculated ``by hand'' using for
79: instance direct Madelung potential summation (see e.g.~\cite{Cwik-03}).
80: However, this method has some drawbacks and neglects the
81: overlap of the electron clouds of neighboring ions, which can lead to
82: several important consequences.\cite{Goodenough}
83:
84: Another important contribution comes from the covalency. It can also be
85: estimated using some model calculations (for instance taking Slater-Koster
86: parameters), but without precise knowledge of the band structure one
87: does not actually know the parameters of the models accurately enough.
88: In order to take into consideration all these contributions
89: to the total value of CFS in a reliable way one should use {\it
90: ab initio} methods based on density functional theory (DFT)
91: calculations. Nevertheless, even in this case there are several ways to
92: define and calculate CFS. Each of them has its own virtues, but
93: also difficulties.
94:
95: The first possibility is a direct calculation of the center of
96: gravity of corresponding bands. It works well when
97: splitting is large enough. Thus it is most useful and rather
98: accurate for $t_{2g}-e_g$ splitting (typically in 3d-oxides CFS
99: between $t_{2g}$ and $e_g$ states is $\sim$2-2.5~eV). For example the behavior of
100: spin-state transition temperature in cobaltites, which depends on
101: this splitting has been recently estimated quite accurately.\cite{Nekrasov-03}
102: The problems might arise in computation of smaller values such as the
103: splitting inside $t_{2g}$ or $e_g$ shells.
104: The general problem is
105: that CFS is a well-defined characteristic only for isolated ions
106: which have energy levels but not bands. The bigger the corresponding
107: band width (in comparison with CFS) the more questionable it becomes to treat
108: the difference of centers of gravity as an estimate of CFS
109: value.
110:
111:
112: The second way to calculate CFS is the downfolding or projection
113: procedure which could give a few orbital on-site Hamiltonian in the
114: minimal basis set of functions $\psi_1$, $\psi_2$,
115: ..., $\psi_N$.\cite{Andersen-00,Anisimov-04}
116: In Sec.~IV the method of obtaining such few orbital Hamiltonian
117: from a full orbital one is briefly described.
118:
119: Having this Hamiltonian one
120: can diagonalize it to obtain its eigenvalues $\epsilon_1$, $\epsilon_2$,
121: ..., $\epsilon_N$ and eigenvectors
122: $\Psi_1$, $\Psi_2$, ..., $\Psi_N$ which are in general linear
123: combinations of the initial wave functions:
124: \begin{equation}
125: \label{Psi}
126: |\Psi_j> = \sum^N_{i} a_{ij} |\psi_i>.
127: \end{equation}
128: The eigenvalues can be considered as the energies of the orbitals.
129: The differences between these eigenvalues define in this case CFS.
130:
131: In spite of great generality of this method, it has some
132: disadvantages. There is an ambiguity in the definition of unitary
133: transformation matrix $U^{({\bf k})}_{ji}$ for Wannier function
134: construction (see Sec. IV)
135: and in the choice of the basis set wave functions
136: $\psi_1$, $\psi_2$, ..., $\psi_N$ in any downfolding or projection
137: procedure. Using different basis sets few orbital non-interacting
138: Hamiltonian can be constructed in different ways. All of
139: them are equally good if the resulting Hamiltonian gives the same
140: bands as the LDA full orbital one. It is essentially important
141: for low symmetry systems, where it is not clear what
142: kind of linear
143: combinations of $d-$wave functions and in which local coordinate
144: system (LCS) should be taken as a basis set for few orbital Hamiltonian
145: construction.
146:
147: In addition there is one more serious and general problem.
148: LDA calculations are widely known to give good description
149: of the {\it ground state} characteristics. However, it often
150: fails to describe the excited states. LDA eigenvalues
151: should not be treated as the real excitation energies.
152: This general shortcoming concerns all methods of
153: calculating CFS which use LDA eigenvalues.
154:
155:
156: On the other hand, care should be taken in what we actually mean
157: by CFS. In direct study of CF excitations e.g. by optics
158: one should take into account possible relaxation of both the
159: electronic (fast process) and lattice (relatively slow relaxation)
160: subsystems. Fast electronic relaxation definitely has to be taken
161: into consideration, whereas the lattice
162: relaxation can be often treated separately (direct optical
163: absorptions usually occur at a frozen lattice, according to the
164: Frank-Condon principle).
165:
166: In this paper we propose {\it the direct method} of
167: calculation of CF excitation energy in the framework of the LDA+U
168: approximation which allows for such relaxation of
169: electron system and enables to avoid ambiguity in the choice of basis set
170: (as in projection or downfolding procedure) as well. CF excitations are
171: calculated as difference between the ground state energy and the
172: total energy of an excited state in which the electron is
173: ``artificially'' put to one of the higher-lying $d-$levels.
174: These states - the ground state and the
175: excited state with the electron constrained in the higher level - are
176: both treated in the self-consisted LDA+U scheme. This allows us to obtain the values of
177: CF excitation energy which, in particular, take into account fast relaxation of the electronic system.
178:
179:
180: Below we develop this general method and test it on the example of
181: Y and La titanates.
182: We compare our results with the previously obtained ones.\cite{Pavarini-04,Solovyev-03}
183: The physics of these compounds is discussed on the basis
184: of calculated values of CF excitation energy, as well as the other results of
185: {\it ab initio} LDA and LDA+U calculations.
186:
187:
188: \section{PHYSICS OF TITANATES AND CRYSTAL-FIELD SPLITTING}
189:
190: Unusual and very rich physics of titanates attracts much
191: attention since the possibility of existence of an orbital liquid
192: has been proposed for LaTiO$_3$.\cite{Keimer-00,Khaliullin-00} In
193: the ground state both LaTiO$_3$ and YTiO$_3$ crystallize in
194: perovskite structure. The lattice distortions due to different
195: ionic radii of Y and La ions seems to lead to the cardinal changes
196: in electronic and magnetic properties of these compounds.
197:
198: At low temperatures LaTiO$_3$ was found to be G-type
199: antiferromagnet with N$\acute e$el temperature for stoichiometric
200: samples T$_N$=146~K~\cite{Cwik-03} while YTiO$_3$ is an isotropic
201: ferromagnet with relatively low T$_C\sim$30~K.\cite{Hester-97}
202: Local magnetic moment on Ti$^{3+}$ ion in YTiO$_3$ was found to be
203: 0.84~$\mu_B$.\cite{Hester-97} The ordered moment in LaTiO$_3$
204: amounts to 0.46 - 0.58~$\mu_B$~\cite{Cwik-03,Keimer-00,Meijer-99}
205: and strongly differs from 1~$\mu_B$ expected for $d^1$
206: configuration with quenched orbital moment. Another interesting
207: feature is nearly isotropic magnon spectra with small spin gap
208: observed both in LaTiO$_3$ and YTiO$_3$ despite of the intrinsic
209: orthorhombic distortion.\cite{Keimer-00,Ulrich-02}
210:
211: In order to describe these puzzling properties the exciting idea of the
212: orbital liquid has been proposed.\cite{Khaliullin-00} According to this
213: theory the large degeneracy of $t_{2g}$ shell can lead to the
214: orbitally-disordered ground state. The crucial point in this case is
215: the value of energy required for electron excitation from one $t_{2g}$
216: orbital to another: orbital liquid can exist only if this energy is zero or
217: very small.
218:
219: However, as the real symmetry of LaTiO$_3$ and YTiO$_3$ is not cubic
220: but orthorhombic, one should expect certain splitting of
221: $t_{2g}$-levels. First calculations of the CFS
222: in LaTiO$_3$ was performed by R.
223: Schmitz and E. M\"uller-Hartmann.\cite{Cwik-03,Schmitz-04}
224: Using realistic crystal structure\cite{Cwik-03} they carried out
225: model calculations taking into account point-charge and covalency contributions
226: and obtained that the lowest singlet
227: is separated from two higher-lying almost degenerate levels by
228: about 0.24 eV. Similar calculations were also performed by
229: Mochizuki and Imada.\cite{Imada-03} They stressed the importance of
230: the GdFeO$_3$-type distortion and found the value of CFS to be
231: $0.77/\epsilon_{TiLa}$~eV, where $\epsilon_{TiLa}$ is an effective
232: dielectric constant. This constant is hard to evaluate in model
233: calculations in a reliable way because of local screening effects
234: in a solid, which could be explicitly taken into account only in
235: {\it ab initio} band structure calculations.
236:
237:
238: The {\it ab initio} calculations were done by Pavarini
239: {\it et al.}\cite{Pavarini-04} and Solovyev.\cite{Solovyev-03}
240: They used the diagonalization of few orbital on-site effective LDA
241: Hamiltonian in real space to obtain CFS. Solovyev has performed
242: an exact procedure of full-orbital Hamiltonian transformation
243: to the small energy-dependent Hamiltonian. After that the energy\vspace{-0.3cm}
244: \begin{center}
245: \begin{figure*}
246: \centering
247: \includegraphics[clip=false,width=0.42\textwidth]{bands-wien.ps}
248: \includegraphics[clip=false,width=0.35\textwidth]{bnds-LMTO.ps}
249: \caption{\label{bands} (color online) Band structure for LaTiO$_3$ obtained within the LDA
250: approximation in the framework of two methods: the full potential
251: linearized augmented plane waves -- FP-LAPW (left) and linear
252: muffin-tin orbitals -- LMTO (right). The energy region is chosen
253: to illustrate the behavior of $t_{2g}$ bands. The Fermi level
254: corresponds to zero energy.}
255: \end{figure*}
256: \end{center}\noindent was fixed in the center of the $t_{2g}$
257: band.\cite{Solovyev-03} Pavarini {\it et al.}\cite{Pavarini-04}
258: have used formalism of Nth-order muffin-tin
259: orbitals (NMTOs) to define small Hamiltonian.\cite{Andersen-00}
260: Despite the similarity of the methods, the
261: results are qualitatively different. The CFS between lowest and
262: next $t_{2g}$ level obtained in Ref.~\onlinecite{Solovyev-03} is rather small,
263: $\sim$30-50~meV for both compounds, which would not prevent
264: an orbital liquid formation.
265: However, the results of calculations presented in Ref.~\onlinecite{Pavarini-04}
266: are incompatible with this scenario: obtained values of CFS
267: are $\sim$ 150-200~meV for both compounds, similar to the results of
268: model calculations.\cite{Cwik-03,Imada-03}
269:
270:
271: \section{LDA: BAND STRUCTURE}
272: Discrepancy between the results of previous {\it ab initio} CFS
273: calculations~\cite{Pavarini-04,Solovyev-03} is probably caused by
274: (i) choice of MT spheres radii in LMTO method~\cite{Andersen-75} or
275: (ii) the details of different projection procedures. To
276: resolve first problem we have carried out the conventional LDA
277: calculations using LMTO method and verified its band structure by
278: performing the full-potential calculations in the framework of
279: linearized augmented plane waves (FP-LAPW) method realized on
280: Wien2k program code.\cite{Wien2k} We chose FP-LAPW because it is
281: known to be the most accurate method for band structure
282: calculations.
283:
284: Crystallographic data for LaTiO$_3$ (T=8~K) and YTiO$_3$ (T=293~K)
285: were taken from Ref.~\onlinecite{Cwik-03}.
286: The radii of the muffin-tin (MT) spheres for LMTO calculations
287: were chosen to be R$_{\rm La }$=3.28~a.u., R$_{\rm Ti}$=2.52~a.u., and for
288: both oxygens R$_{\rm O}$=1.92~a.u. Remaining unit cell volume was
289: filled by empty spheres (atomic spheres with zero nuclear charge)
290: with different MT radii. Ionic radius of Y is known to be smaller than
291: La one.\cite{Shannon} According to this fact
292: the MT radius of Y was also taken to be smaller: R$_{\rm Y }$=3.01~a.u.
293:
294:
295: The Ti($4s$,$4p$,$3d$), O($3s$,$2p$,$3d$) and
296: Y($5s$,$5p$,$4d$,4$f$), La($6s$,$6p$,$5d$) states were included
297: in the basis set in our calculations. Almost empty La-$4f$
298: states were treated as pseudo-core, since
299: La-$4f$ states are localized and do not strongly contribute to the
300: relevant electronic states. Due to rather
301: large values of the on-site Coulomb interaction $U$ (of the order
302: of 10~eV) La-4f states will be located
303: approximately at 5-10~eV above the Fermi level.\cite{Chainani-92}
304:
305: The Brillouin-zone (BZ) integration in the course of the
306: self-consistent iterations was performed over a mesh of 27 {\bf
307: k}-points in the irreducible part of the BZ. Density of states (DOS)
308: were calculated by the tetrahedron method with 512 {\bf k}-points in
309: the whole BZ.
310:
311: The band structure obtained within the LDA approximation in the
312: framework of LMTO (with the MT radii chosen above) and FP-LAPW
313: methods for LaTiO$_3$ is presented in Fig.~\ref{bands}. Twelve
314: Ti$-t_{2g}$ (four Ti per unit cell) bands are placed in the
315: vicinity of the Fermi level and have band width
316: $\sim 1.95$ eV both in FP-LAPW and LMTO methods. The energy gap
317: between the top of the $t_{2g}$ and the bottom of $e_g$ bands is
318: estimated to be 0.25~eV in FP-LAPW and 0.18~eV in LMTO. Comparing this figure
319: and Fig.~2 of Ref.~\onlinecite{Solovyev-03} one can see
320: that the present LMTO bands agree better in all
321: high-symmetry points of BZ with FP-LAPW ones than those presented
322: in Ref.~\onlinecite{Solovyev-03}. So being firmly convinced of the
323: correctness of LMTO band structure one could carry out the CFS
324: calculations.
325:
326:
327: \section{LDA: WANNIER FUNCTION PROJECTION FOR CRYSTAL-FIELD SPLITTING CALCULATION}
328:
329: One of the ways to calculate CFS is to use the minimal basis set Hamiltonian for
330: the orbitals of interest in real space. To construct such Hamiltonian we used
331: Wannier functions projection procedure. In this section we present
332: brief description of the method. For more details, see
333: Ref.~\onlinecite{Anisimov-04}.
334:
335: For an LDA Hamiltonian
336: $\hat H$ one has a Hilbert space of eigenfunctions (Bloch states
337: $|\psi_{i\bf k}\rangle$) with the basis set $|\phi_\mu\rangle$
338: defined by the particular method (for LMTO these are
339: LMT-orbitals,\cite{Andersen-75} for LAPW - Augmented Plane
340: Waves,\cite{Mattheiss86} etc.). In this basis set the Hamiltonian
341: operator is defined as:
342: \begin{eqnarray}
343: \label{Ham_gen}
344: \widehat H & = & \sum_{\mu\nu} |\phi_\mu\rangle
345: H_{\mu\nu} \langle\phi_\nu|,
346: \end{eqnarray}
347: where greek indices are used for full-orbital matrices.
348:
349: If we consider a certain subset of the Hamiltonian eigenfunctions,
350: for example Bloch states of partially filled bands $|\psi_{n\bf
351: k}\rangle$, then we can define a corresponding subspace in the
352: total Hilbert space. The Hamiltonian matrix is diagonal in the
353: Bloch states basis, however, physically more appealing is a basis which
354: would have a form of site-centered atomic orbitals. That is a set
355: of Wannier functions $|W_{n}^{\bf T}\rangle$ defined as a Fourier
356: transformation of certain linear combination of Bloch functions
357: belonging to this subspace (see Eq. (\ref{WF_psi_def}) bellow). They
358: are labeled in real space according to the band $n$ and the
359: lattice vector of the unit cell {\bf T} which they belong to. The
360: Hamiltonian operator $\hat H^{WF}$ defined in this basis set is
361: \begin{eqnarray}
362: \label{Ham_genWF} \widehat H^{WF} & = & \sum_{nn'{\bf
363: T}}|W_{n}^{\bf 0}\rangle H_{nn'}({\bf T}) \langle W_{n'}^{\bf T}|.
364: \end{eqnarray}
365:
366: The total Hilbert space can be divided into a direct sum of above
367: introduced subspace (of partially filled Bloch states)
368: and the subspace formed by all other states orthogonal to it.
369: Those two subspaces are decoupled since they are the eigenfunctions
370: corresponding to different eigenvalues.
371: The Hamiltonian matrix in Wannier function basis is block-diagonal, so
372: that the matrix elements between different subspaces in Hilbert
373: space are zero. The block $H_{nn'}$ in (\ref{Ham_genWF})
374: corresponding to the partially filled bands can be considered as a
375: projection of the full Hamiltonian operator (\ref{Ham_gen}) to the
376: subspace defined by its Wannier functions.
377:
378:
379: Localized Wannier functions $|W_{i}^{\bf T}\rangle$ were defined
380: in Ref.~\onlinecite{wannier} as Fourier transforms of the Bloch functions
381: $|\psi_{i\bf k}\rangle$
382: \begin{eqnarray}
383: \label{WF_psi_def}
384: |W_i^{\bf T}\rangle & = & \frac{1}{\sqrt{N}} \sum_{\bf k}
385: e^{-i{\bf kT}}|\psi_{i{\bf k}}\rangle,
386: \end{eqnarray}
387: where $N$ is the number of discrete $\bf k$ points in the
388: first BZ.
389:
390: Wannier functions are not uniquely defined because for a certain
391: set of bands any orthogonal linear combination of Bloch functions
392: $|\psi_{i\bf k}\rangle$ can be used in (\ref{WF_psi_def}). In
393: general it means that the freedom of choice of Wannier functions
394: corresponds to freedom of choice of a {\bf k}-depended unitary transformation
395: matrix $U^{({\bf k})}_{ji}$:\cite{vanderbildt}
396: \begin{eqnarray}
397: \label{psi_def}
398: |\widetilde\psi_{i\bf k}\rangle & = & \sum_j
399: U^{({\bf k})}_{ji} |\psi_{j\bf k}\rangle.
400: \end{eqnarray}
401:
402: The most serious drawback of Wannier representation is that there is no rigorous
403: way to define $U^{({\bf k})}_{ji}$. As one can see from
404: (\ref{WF_psi_def}) and (\ref{psi_def}), Wannier functions can vary
405: significantly in shape and range because variations in
406: $|\psi_{i\bf k}\rangle$ or $U^{({\bf k})}_{ji}$ change relative phases and
407: amplitudes of Bloch functions at different $\bf k$ and bands $i$.
408:
409: In order to avoid this disadvantage some additional restrictions on the
410: properties of Wannier functions have been proposed.\cite{Koster-53,Kohn-73,vanderbildt}
411: Among others Marzari and
412: Vanderbildt~\cite{vanderbildt} suggested the condition of maximum
413: localization for Wannier functions. That gave the variational
414: procedure to calculate $U^{({\bf k})}_{ji}$. To get a good initial
415: point the authors suggested to choose a set of trial localized
416: orbitals $|\phi_n\rangle$ and projecting them onto the Bloch
417: functions $|\psi_{i\bf k}\rangle$. It was found that this starting
418: guess is usually quite good.\cite{vanderbildt} This fact led
419: later to the simplified calculating scheme proposed in
420: \cite{pickett} where variational procedure was abandoned and the
421: result of aforementioned projection was considered as a final
422: step.
423:
424: In order to start projection procedure one needs to determine
425: the set of trial orbitals $|\phi_n\rangle$ and the bands which will be
426: used for the Wannier functions construction. The latter can be
427: defined either by the bands numbers (from $N_1$ to $N_2$) or by the
428: energy interval ($E_1, E_2$).
429:
430: Non-orthogonalized Wannier functions in real
431: $|\widetilde{W}_{n}^{\bf T}\rangle$ and reciprocal space
432: $|\widetilde{W}_{n\bf k}\rangle$ are then the projection of the
433: set of trial site-centered atomic-like orbitals $|\phi_n\rangle$
434: on the Bloch functions $|\psi_{i\bf k}\rangle$
435: of the chosen bands
436: \begin{eqnarray}
437: \label{WF_psi} |\widetilde{W}_{n}^{\bf T}\rangle & = &
438: \frac{1}{\sqrt{N}} \sum_{\bf
439: k} e^{-i{\bf kT}} |\widetilde{W}_{n\bf k}\rangle, \\ \nonumber
440: |\widetilde{W}_{n\bf k}\rangle & \equiv & \sum_{i=N_1}^{N_2}
441: |\psi_{i\bf k}\rangle\langle\psi_{i\bf k}|\phi_n\rangle =
442: \sum_{i(E_1\le \varepsilon_{i}({\bf k})\le E_2)} |\psi_{i\bf
443: k}\rangle\langle\psi_{i\bf k}|\phi_n\rangle.
444: \end{eqnarray}
445:
446: Note that the Wannier functions in reciprocal space
447: $|\widetilde{W}_{n\bf k}\rangle$ do not coincide with the Bloch
448: functions $|\psi_{n\bf k}\rangle$ for multi-band case due to the
449: summation over band index $i$ in (\ref{WF_psi}). One can consider
450: them as Bloch sums of Wannier functions analogous to the basis
451: functions Bloch sums $\phi_j^{\bf k}({\bf r})$, see (\ref{psik})
452: below.
453:
454: \begin{table*}
455: \centering \caption{The magnitudes of CFS in $t_{2g}$ shell (in
456: meV) obtained in the LDA approach by different authors are
457: presented in first 3 columns. First value is an energy difference
458: between the lowest energy level and the middle one; the second is
459: the difference between the middle and the highest energy
460: levels in $t_{2g}$ shell. The results of the constrained LDA+U
461: calculations, which take into account fast electron relaxations
462: are shown in the last column. } \vspace{0.2cm} \label{SplitTable}
463: \begin{tabular}{ccccc}
464: \hline
465: \hline
466: & Present results & E. Pavarini {\it et al.}~\footnote{Reference \cite{Pavarini-04}} & I.V.
467: Solovyev~\footnote{Reference \cite{Solovyev-03}} & Present results\\
468: & (WF projection) & & & (LDA+U)\\
469: \hline
470: $LaTiO_3$ & 230; 40 & 140; 60 & 54; 39 & 160; -\\
471: $YTiO_3$ & 180; 80 & 200; 130 & 27; 154 & 150; -\\
472: \hline
473: \end{tabular}
474: \end{table*}
475:
476: The coefficients $\langle\psi_{i\bf k}|\phi_n\rangle$ in
477: (\ref{WF_psi}) define (after orthonormalization) the unitary
478: transformation matrix $U^{({\bf k})}_{ji}$ in (\ref{psi_def}).
479: However, projection procedure defined in (\ref{WF_psi}) can be
480: considered as a more general formalism than unitary transformation
481: (\ref{psi_def}).
482:
483: The Bloch functions in LMTO basis take the form
484: \begin{eqnarray}
485: \label{psi} |\psi_{i\bf k}\rangle=
486: \sum_{\mu} c^{{\bf k}}_{\mu i}|\phi_{\mu}^{\bf k}\rangle,
487: \end{eqnarray}
488: where $\mu$ is the combined index of the $qlm$ ($q$ - atomic number
489: in the unit cell, $lm$ are orbital and magnetic quantum numbers),
490: and $\phi_{\mu}^{\bf k}({\bf r})$ are Bloch sums of the basis orbitals
491: $\phi_{\mu}({\bf r- T})$
492: \begin{eqnarray}
493: \label{psik} \phi_{\mu}^{\bf
494: k}({\bf r}) & = & \frac{1}{\sqrt{N}} \sum_T e^{ikT} \phi_{\mu}({\bf r}-{\bf T}),
495: \end{eqnarray}
496:
497: \noindent and the coefficients have the property
498: $c^{\bf k}_{\mu i} = \langle\phi_{\mu}|\psi_{i\bf k}\rangle.$
499:
500: If $n$ in $|\phi_n\rangle$ corresponds to the particular $qlm$
501: combination (in other words $|\phi_n\rangle$ is an {\it orthogonal} LMTO
502: basis set orbital), then $\langle\psi_{i\bf k}|\phi_n\rangle =
503: c_{ni}^{{\bf k}*}$ and hence
504: \begin{eqnarray}
505: \label{WF}
506: |\widetilde{W}_{n\bf
507: k}\rangle & = & \sum_{i=N_1}^{N_2} |\psi_{i\bf k}\rangle
508: c_{ni}^{{\bf k}*} \\ \nonumber & = &
509: \sum_{i=N_1}^{N_2} \sum_{\mu} c_{\mu i}^{\bf k} c_{ni}^{{\bf k}*}
510: |\phi_{\mu}^{\bf k}\rangle =
511: \sum_{\mu} \tilde{b}^{\bf k}_{\mu n}
512: |\phi_{\mu}^{\bf k}\rangle,
513: \end{eqnarray}
514: \begin{eqnarray}
515: \nonumber
516: \tilde{b}^{\bf k}_{\mu n} =\sum_{i=N_1}^{N_2} c_{\mu i}^{\bf k} c_{ni}^{{\bf k}*}.
517: \end{eqnarray}
518:
519: In order to orthonormalize the WF (\ref{WF}) an overlap matrix
520: $O_{nn'}({\bf k})$ and its inverse square root $S_{nn'}({\bf k})$
521: can be defined as
522: \begin{eqnarray}
523: \label{O-S} O_{nn'}({\bf k})&\equiv& \langle\widetilde{W}_{n\bf
524: k}|\widetilde{W}_{n'\bf k}\rangle = \sum_{i=N_1}^{N_2} c_{ni}^{\bf
525: k} c_{n'i}^{{\bf k}*}, \\ \nonumber S_{nn'}({\bf k})
526: &\equiv& O^{-1/2}_{nn'}({\bf k}),
527: \end{eqnarray}
528: The orthogonality of Bloch states $\langle\psi_{n\bf
529: k}|\psi_{n'\bf k}\rangle=\delta_{nn'}$ was used.
530:
531: Orthonormalized Wannier functions in $k$-space $|W_{n\bf
532: k}\rangle$ can be obtained as
533: \begin{eqnarray}
534: \label{WF_orth} |W_{n\bf k}\rangle & = &\sum_{n'} S_{nn'}({\bf k})
535: |\widetilde{W}_{n'\bf k}\rangle =
536: \sum_{i=N_1}^{N_2} |\psi_{i\bf k}\rangle \bar{c}_{ni}^{{\bf
537: k}*}
538: \\ \nonumber &=&\sum_{\mu} b^{\bf k}_{\mu n} |\phi_{\mu}^{\bf k}\rangle,
539: \end{eqnarray}
540:
541: where
542: \begin{eqnarray}
543: \label{other}
544: \bar{c}_{ni}^{{\bf k}*}&\equiv& \langle\psi_{i {\bf k}}|W_{n {\bf k}}\rangle=
545: \sum_{n'} S_{nn'}({\bf k})
546: c_{n'i}^{{\bf k}*},
547: \\ b^{\bf k}_{\mu n} &\equiv& \langle\phi_{\mu}^{\bf k}|W_{n {\bf k}}\rangle=
548: \sum_{i=N_1}^{N_2} \bar{c}_{\mu i}^{\bf k} \bar{c}_{ni}^{{\bf k}*}.
549: \end{eqnarray}
550:
551:
552:
553: Thus, matrix elements of the few orbital Hamiltonian $\widehat H^{WF}$ in the basis
554: of WF in real space where both orbitals are in the same unit cell
555: are given by the following expression
556: \begin{eqnarray}
557: \label{E_WF} H^{WF}_{nm}(0) & = & \langle W_{n}^{\bf
558: 0}|\frac{1}{N}\biggl(\sum_{\bf k}\sum_{i=N_1}^{N_2}|\psi_{i\bf k}\rangle
559: \epsilon_{i}({\bf k})\langle\psi_{i\bf k}|\biggr)|W_{m}^{\bf
560: 0}\rangle = \nonumber \\
561: & = & \frac{1}{N} \sum_{\bf k}\sum_{i=N_1}^{N_2} \bar{c}_{ni}({\bf k})
562: \bar{c}_{mi}^{*}({\bf k})\epsilon_{i}({\bf k}).
563: \end{eqnarray}
564: Here $\epsilon_{i}({\bf k})$ is an eigenvalue for a particular band.
565:
566:
567: For both LaTiO$_3$ and YTiO$_3$ we are interested in CFS in the
568: $t_{2g}$ shell. For that 3$\times$3 Hamiltonians were
569: derived for $t_{2g}$ bands placed in the vicinity of the Fermi
570: level in the energy range (-0.8; 1.25)~eV for LaTiO$_3$ and
571: (-0.65; 1.55)~eV for YTiO$_3$. Then we diagonalize them and
572: calculate the difference between corresponding eigenvalues in
573: order to determine the CFS.
574:
575:
576: The obtained values of CFS together with results of
577: Ref.~\onlinecite{Pavarini-04,Solovyev-03} are presented in Tab.~\ref{SplitTable}.
578: The CFS calculated by Pavarini {\it et al.}\cite{Pavarini-04} has the same
579: character and similar values as ours.
580: According to these results, in LaTiO$_3$ the lowest energy level is
581: widely separated from two other, which are almost degenerate.
582: Similar result was obtained by Cwik {\it et al.} using a full
583: Madelung-sum point charge model, where the first splitting was found
584: to be 240~meV.\cite{Cwik-03}
585:
586:
587: The situation is a little bit different in YTiO$_3$. The
588: magnitude of the first splitting (between lowest and middle
589: energy levels) has the same order as in La titanate,
590: while the second one (between middle and highest
591: energy levels) increases. It is interesting to note that in
592: calculations of Solovyev the same effect of the second
593: splitting enhancement going from LaTiO$_3$ to YTiO$_3$ is
594: observed.\cite{Solovyev-03} Nevertheless, the character of CFS
595: obtained by Solovyev~\cite{Solovyev-03} is quite
596: different from the present one and from that calculated in
597: Ref.~\onlinecite{Pavarini-04}.
598:
599:
600: \section{LDA+U: BAND STRUCTURE}
601:
602: The application of LDA to solids provides important information about details
603: of band structure. However, for transition metal compounds it
604: usually leads to metallic type of the electronic structure due to
605: the presence of the partially filled $d-$bands at the Fermi level,
606: often in contrast with an experiment. This is also the
607: case in the LDA approach presented above (Sec. IV). The LDA+U
608: approximation has been developed to include in the calculation scheme
609: orbital-dependent Hubbard-like correction $U$ which acts differently
610: on the occupied and unoccupied $d-$orbitals giving as a result
611: correct description for localized states.\cite{Anisimov-91}
612:
613: In order to analyze of the results of the LDA+U
614: calculations in terms of $t_{2g}$ and $e_{g}$ orbitals
615: one has to chose in the local coordinate system (LCS). Usually
616: it is possible to define the LCS where axes are pointed directly
617: along $Ti-O$ bonds, but for strongly distorted compounds
618: (like LaTiO$_3$ and YTiO$_3$) such coordinate system in general might be
619: not rectangular. We use the rectangular local coordinate system where the
620: axes are directed as much as possible to the nearest oxygens.
621:
622: The interorbital on-site Coulomb interaction parameter $U$ and
623: intra-atomic exchange coupling $J$ for $t_{2g}$ shell were estimated
624: to be 3.3~eV and 0.8~eV, respectively, using constrained superscell
625: calculations~\cite{Pickett-98} in the framework of LMTO method and
626: taking into account screening by $e_g$ electrons.\cite{Solovyev-96}
627: These values are in good agreement with previous
628: findings.\cite{Solovyev-96,Mizokawa-96}
629: The experimentally observed magnetic structures: YTiO$_3$ -
630: ferromagnet, LaTiO$_3$ - G-type antiferromagnet are used
631: for the conventional and constrained LDA+U calculations.
632:
633: The top panel of Fig.~\ref{LDAU-DOS} shows LDA+U
634: electronic structure of LaTiO$_3$ in details. There are three
635: distinguishable sets of bands: completely filled O-$2p$ bands,
636: partially filled Ti-$3d$ bands and empty La-$5d$ bands. The bands in
637: the energy range from -7.2~eV to -2.7~eV originate mainly from O-2$p$
638: states. The gap $\sim$2.3~eV appears between O-$2p$ and the narrow
639: peak of Ti-$3d$($t_{2g}$) states. The band gap in LaTiO$_3$ is found to
640: be 0.57~eV. It divides Ti-$3d(t_{2g})$ band into two parts in such a way
641: that both the top of the valence and the bottom of the conduction band are
642: predominantly formed by Ti-$3d(t_{2g})$ states (see Fig.~\ref{LaTiO3-d-DOS}).
643: The bands lying higher than $\sim$2~eV have
644: basically La-$3d$, O-$2p$ and Ti-$3d(e_g)$ contributions.
645: The calculated local magnetic moment on Ti$^{3+}$ ion amounts to 0.78$\mu_B$
646: being by $\sim$0.2$\mu_B$ larger than the experimentally observed value.
647: \cite{Cwik-03}
648: \vspace{-0.2cm}
649: \begin{center}
650: \begin{figure}
651: \centering
652: \includegraphics[clip=false,width=0.42\textwidth]{LaTiO3-ldau.eps}
653: \caption{\label{LDAU-DOS}(color online). Total and partial DOS of LaTiO$_3$ (top panel) and
654: YTiO$_3$ (bottom panel) calculated within the LDA+U approximation for
655: AFM-G and FM magnetic structures, respectively. The Fermi level corresponds to zero energy.}
656: \end{figure}
657: \end{center}
658:
659: The electronic structure obtained in the LDA+U approximation
660: for YTiO$_3$ is shown in the bottom panel of Fig.~\ref{LDAU-DOS}.
661: Generally it has similar character.
662: However, the value of the band gap is larger than in LaTiO$_3$
663: ($\sim$0.78~eV). The value of the local magnetic moment is found
664: to be 0.89$\mu_B$ in good agreement with experiment.\cite{Hester-97}
665:
666:
667: There are two opposite factors influencing the magnitude of the band gap.
668: On one hand the larger lattice distortions in YTiO$_3$ due to smaller
669: ionic radii of Y in comparison with La make Ti($3d$)-O($2p$)
670: hybridization weaker and as a result the band gap larger. On the contrary,
671: ferromagnetism observed in YTiO$_3$ leads to an increase of the band
672: width resulting in a band gap reduction.
673:
674:
675: \begin{table*}
676: \centering
677: \caption{Total energy difference between three magnetic
678: solutions per Ti ion and magnitudes of the band gaps and local magnetic
679: moments on Ti ions obtained in LDA+U calculation. Zero energy is the
680: lowest total energy for compound under consideration.}
681: \vspace{0.2cm}
682: \label{EnergyTable}
683: \begin{tabular}{l|ccc|ccc}
684: \hline\hline
685: \multicolumn {1}{c}{}& \multicolumn {3}{c}{$LaTiO_3$} & \multicolumn
686: {3}{c}
687: {$YTiO_3$} \\
688: \multicolumn {1}{c}{}&
689: \multicolumn {1}{c}{Band gap}&
690: \multicolumn {1}{c}{Mag. moment}&
691: \multicolumn {1}{c}{Total energy}
692: &
693: \multicolumn {1}{c}{Band gap}&
694: \multicolumn {1}{c}{Mag. moment}&
695: \multicolumn {1}{c}{Total energy}\\
696: \hline
697: %FM
698: \multicolumn {1}{c}{FM}&
699: \multicolumn {1}{c}{0.45~eV}&
700: \multicolumn {1}{c}{0.88~$\mu_B$}&
701: \multicolumn {1}{c}{125~K}
702: &
703: \multicolumn {1}{c}{0.78~eV}&
704: \multicolumn {1}{c}{0.89~$\mu_B$}&
705: \multicolumn {1}{c}{0}
706:
707: \\
708: %AFM-A
709: \multicolumn {1}{c}{AFM-A}&
710: \multicolumn {1}{c}{0.54~eV}&
711: \multicolumn {1}{c}{0.83~$\mu_B$}&
712: \multicolumn {1}{c}{0}
713: &
714: \multicolumn {1}{c}{0.89~eV}&
715: \multicolumn {1}{c}{0.87~$\mu_B$}&
716: \multicolumn {1}{c}{60~K} \\
717: %AFM-G
718: \multicolumn {1}{c}{AFM-G}&
719: \multicolumn {1}{c}{0.57~eV}&
720: \multicolumn {1}{c}{0.78~$\mu_B$}&
721: \multicolumn {1}{c}{40~K}
722: &
723: \multicolumn {1}{c}{1.04~eV}&
724: \multicolumn {1}{c}{0.81~$\mu_B$}&
725: \multicolumn {1}{c}{230~K} \\
726: \hline
727: \end{tabular}
728: \end{table*}
729:
730:
731: In order to clarify the role of magnetism we performed additional
732: band structure calculations for magnetic structures different
733: from experimental one.
734:
735: The FM, AFM-G and AFM-A structures
736: were considered. The values of the band gap and local magnetic moments on
737: Ti$^{3+}$ ions for all calculated configurations are presented in Tab.
738: \ref{EnergyTable}. One can see that the band gap
739: reduction due to ferromagnetism amounts to 21$\%
740: $ (0.12~eV) in the case of LaTiO$_3$ and 25$\%
741: $ (0.26~eV) in YTiO$_3$. The modifications of an
742: electronic structure with the change of magnetic structure
743: from AFM-G to FM in the vicinity of Fermi level for LaTiO$_3$ are presented
744: in the inset of Fig.\ref{LaTiO3-d-DOS}.
745:
746:
747: Continuing investigation of the interplay between lattice, electronic and
748: magnetic degrees of freedom one can isolate the influence of ``pure lattice
749: distortions'' on the electronic structure of La and Y titanates comparing
750: the results of calculations performed in identical magnetic structures.
751: The analysis shows significant changes in the value of the band gap. Only
752: due to the lattice distortions it changes by 0.32-0.47~eV depending on the
753: magnetic structure under consideration.
754:
755:
756: The main reason for such a
757: strong influence of local geometry on the electronic structure of these
758: compounds is probably connected with the stronger covalency of $d^1$
759: configuration and hence higher sensibility to the distortions in
760: comparison with other configurations of $d-$shell (with the exclusion of $d^9$).
761: In contrast to the band gap, magnetic
762: moment is only reduced by 10$\%$ with the change of magnetic configuration.
763:
764: In addition, it should be mentioned that our calculations
765: unlike the experiment give for LaTiO$_3$ the lowest total energy
766: for A-type AFM, although experimentally observed G-type lies quite close.
767: This result is supported by the calculation of the exchange interaction
768: parameters~\cite{Solovyev-03} and connected with the orbital pattern of the compound
769: discussed in details in Sec.~\ref{OS}.
770:
771: \section{CALCULATION OF CRYSTAL-FIELD EXCITATION ENERGY IN
772: LDA+U}
773:
774: The direct study of CFS, for instance by optical measurements, implies the
775: excitation of an electron from one energy level to another.
776: Simultaneously with this excitation the external ``bath'' formed by
777: the rest of the electrons can relax giving the system a chance
778: to lower the energy. The simple CFS computations using the centers of
779: gravity of corresponding bands, projection or downfolding procedure
780: do not take into account such processes and as a result overestimate the
781: value of CFS. In this section we propose
782: the direct calculation of the total energy difference between ground
783: and first excited states ({\it CF excitation energy calculation})
784: in $t_{2g}$ shell using
785: the constrain procedure adopted for the LDA+U approximation.
786: %\vspace{-0.4cm}
787: \begin{center}
788: \begin{figure}[b]
789: \centering
790: \includegraphics[clip=false,angle=270,width=0.4\textwidth]{Ti-d-LaTiO3.ps}
791: \caption{\label{LaTiO3-d-DOS}(color online). LaTiO$_3$. Ti-$3d$ partial
792: DOS calculated within LDA+U approach.
793: The inset shows Ti-$3d$ partial DOS in AFM-G (solid line) and FM
794: (dashed line) configurations. Parts of plots with positive (negative)
795: ordinates denote majority (minority) spin DOS.
796: The Fermi level corresponds to zero energy.}
797: \end{figure}
798: \end{center}
799:
800: The occupation matrix in LDA+U can be defined in the usual way:
801: \begin{equation}
802: \label{eq:Occ}
803: n_{mm'}^\sigma =-\frac 1\pi\int^{E_F}ImG_{inlm,inlm'}^\sigma(E)dE,
804: \end{equation}
805: where $\sigma$ is the spin, $i$ denotes the site, $n,l,m$ are principle,
806: orbital and magnetic
807: quantum numbers respectively, $G_{inlm,inlm'}^\sigma(E)=
808: \langle inlm\sigma|(E-\widehat H^{LDA+U})^{-1}| inlm'\sigma \rangle$ are
809: the elements of the Green function matrix and $\widehat H^{LDA+U}$ is
810: a single particle LDA+U Hamiltonian (for its definition see
811: Ref.~\onlinecite{Anisimov-91}). The occupation matrix for an isolated ion
812: is diagonal. However, for ions in solids it can have more complex
813: structure in low symmetry systems and one needs to diagonalize it.
814: The eigenvectors of the \eqref{eq:Occ} can be used for transformation
815: of occupation matrix to diagonal form. Using this procedure one can
816: obtain the information on the orbitals where the electrons are localized.
817: \begin{center}
818: \begin{figure*}
819: \centering
820: \includegraphics[clip=true,width=0.49\textwidth]{LaTiO3-plan.eps}
821: \includegraphics[clip=true,width=0.49\textwidth]{YTiO3-plan.eps}
822: \caption{\label{ab-plane}(color online) Orbital ordering in LaTiO$_3$ (left) and YTiO$_3$ (right) in
823: $ab$-plane obtained in the framework of the LDA+U calculations for AFM-G
824: and FM structures respectively.}
825: \end{figure*}
826: \end{center}
827:
828: \begin{table}[b]
829: \centering
830: \caption{Decomposition of the lowest in energy orbital
831: into $xy$, $xz$ and $yz$ orbitals. The results are
832: presented in local coordinate system where the orthogonal
833: axes point as much as possible to neighboring oxygens.}
834: \vspace{0.2cm}
835: \begin{tabular}{lccc}
836: \hline
837: \hline
838: & $LaTiO_3$ & $YTiO_3$ \\
839: \hline
840: LDA & 0.63$xy$+0.66$yz$+0.40$xz$&
841: 0.76$xy$+0.64$yz$-0.06$xz$\\
842: LDA+U & 0.62$xy$+0.72$yz$+0.32$xz$&
843: 0.75$xy$+0.66$yz$-0.04$xz$\\
844: \hline
845: \label{orb-table}
846: \end{tabular}
847: \end{table}
848:
849: In fact, in the case of spin-polarized calculations (like LDA+U) there
850: are two occupation matrices for every transition metal ion on each
851: site: spin majority and spin minority occupation matrices. According to
852: the Hund's rule, at first the spin majority states start to be occupied.
853: Nevertheless, due to the large spatial extension and sizable overlap between
854: $e_g$ and oxygen $p$ orbitals (see Fig.~\ref{LaTiO3-d-DOS})
855: there are always some non-negligible occupation numbers for
856: $e_g$ states in a real band structure calculation for both
857: spins. In order to avoid the influence of these effects
858: (they can lead to the ``symmetrization'' of $d$-orbitals) we use for the
859: analysis not the occupation matrix for majority
860: spin (Ti is $d^1$ system in the compounds under consideration), but
861: the difference between matrices for two spins.
862: Here we suppose the oxygen influence on $d-$states for both
863: spins to be equal.
864:
865:
866: The diagonalization of the difference between occupation matrices for
867: majority and minority spins for LaTiO$_3$ gives following
868: eigenvalues: 0, 0, 0.01, 0.02, 0.76. The eigenvector corresponding to the
869: largest eigenvalue defines the occupied orbital as a linear combination of
870: all $d-$orbitals:
871: \begin{equation}
872: |\Psi_{GS}\rangle = \sum_{i} a_i |\psi_i \rangle.
873: \end{equation}
874: Thus, in case of LaTiO$_3$ in the LCS:
875: \begin{eqnarray}
876: \label{LaTiO3-occ-orb}
877: |\Psi^{LaTiO_3}_{GS}\rangle = 0.62xy+0.72yz+0.32xz+\\ \nonumber
878: 0.02(3y^2-r^2)+0.02(z^2-x^2).
879: \end{eqnarray}
880: Similar decomposition for YTiO$_3$ is presented in the Table III.
881:
882:
883: The excitation energy (from the ground to the first excited state)
884: in this case is the total energy difference between states where
885: this orbital is occupied and where it
886: is empty and another one is occupied. According to this scheme we
887: performed the calculation where the system is constrained by
888: external potential $\widehat V_{constr}$ to change the occupied orbital:
889: \begin{equation}
890: \widehat V_{constr} = |\Phi_{GS} \rangle \delta V \langle \Phi_{GS}|.
891: \end{equation}
892: In other words $\widehat V_{constr}$ just pushes up the orbital
893: where the electron has been localized in LDA+U.
894: It is not important how big is the correction
895: $\delta V$. One needs just be sure that it is large enough
896: to force the electron to hop to another orbital.
897: The total energy of excited state does not depend on
898: the value of $\delta V$, because the correction is applied
899: to the orbital which has to be empty in the excited state.
900: The total energy difference between
901: the ground and first excited states is an estimate of the CF excitation energy.
902:
903: The results of calculations of CF excitation energies for LaTiO$_3$ and
904: YTiO$_3$ are presented in the fourth
905: column of Tab.~\ref{SplitTable}.
906: They have the same order of magnitude
907: as those obtained using the WF projection in the present
908: work and downfolding performed by Pavarini {\it et al.}\cite{Pavarini-04}
909: As has been mentioned above, for proper estimation of CF excitation
910: energy the relaxation of electron system should be taken into
911: consideration. Such relaxation decreases the energy costs
912: of the electron excitation. According to the present
913: results the gain in energy can amount to $\sim$70~meV ($\sim$30\%).
914: However, it is obviously insufficient to reduce the value of CFS
915: significantly.
916:
917: Thus, the results of the CFS calculations both using WF projection and
918: constrained LDA+U which takes into account the electron system relaxation
919: indicate that the splittings in LaTiO$_3$ and YTiO$_3$ are relatively large.
920: They are definitely bigger than the spin-orbital coupling in titanates
921: which is expected to be $\Lambda_{SO}\sim20$~meV~\cite{Abragam}.
922: The latter result is supported by the recent XAS measurements, where
923: shown that the orbital momentum in LaTiO$_3$ is essentially
924: quenched.\cite{Haverkort-04}
925:
926: But the most important conclusion is that CF excitation energy
927: obtained in the present work and value of CFS calculated
928: in Ref.~\onlinecite{Pavarini-04} as well as extracted from
929: Ti L$_{2,3}$ XAS~\cite{Haverkort-04} and optical
930: measurements~\cite{Gruninger-05} are all the order of 200 meV,
931: that makes orbital liquid scenario for LaTiO$_3$ rather unlikely.
932:
933:
934: \section{ORBITAL STRUCTURE}
935: \label{OS}
936: Orbital degrees of freedom are known to play an important role
937: and should be correctly taken into account in theories describing
938: magnetic interactions in strongly correlated materials.\cite{Kugel-82}
939:
940: As one can see from Tab. \ref{orb-table} the composition of the
941: occupied orbitals in Y and La titanates is quite different. Contribution
942: of $xy$ and $yz$ components to the resulting orbital in LaTiO$_3$ are
943: nearly the same and by $\sim$~40--60 $\%
944: $ bigger than $xz$ one. There is quite different
945: situation in YTiO$_3$, where $xz$ contribution is almost zero.
946:
947: This fact can be explained by means of the analysis of the crystal
948: structure distortions,
949: which are quite different in these compounds. Distortions in $ab-$plane
950: in LaTiO$_3$ have in general large trigonal $D_{3d}$ component. It
951: gives rise to square-to-rectangle transformation in this
952: plane~\cite{Cwik-03} and
953: results in localization of the electron on the orbital of almost
954: $a_{1g}=(xy+yz+zx)/\sqrt3$ character.
955: At the same time the predominant distortions in YTiO$_3$ make rhombus from
956: the initial square, revealing the sizable contribution of local tetragonal
957: distortions.
958:
959:
960: However, even in case of LaTiO$_3$ the occupied orbital
961: deviates from a$_{1g}$:
962: \begin{eqnarray}
963: \label{LDAandLDAUorbitals}
964: |\langle \Psi_{LDA+U}|a_{1g} \rangle|^2 = 91 \%
965: \\
966: |\langle \Psi_{LDA}|a_{1g} \rangle|^2 = 96 \%
967: \end{eqnarray}
968:
969:
970: The variation of $Ti-O$ distances in YTiO$_3$~\cite{MacLean-79} leads
971: to antiferroorbital ordering (Fig.~4,5)
972: inducing according to Goodenough-Kanamori rules the ferromagnetic structure in
973: agreement with experiment.\cite{Hester-97,Akimitsu-01}
974:
975: The basal plane in LaTiO$_3$ undergoes the elongation in the direction
976: of orthorhombic $a-$axis. It leads to such kind of orbital ordering then the
977: orbitals of Ti ions placed in the $ab-$plane point almost along the same
978: direction, but not to the oxygens (Fig.~\ref{ab-plane}, left panel).
979: Thus, this in-plane ``ferroorbital ordering''
980: implies large hopping integrals not only between two occupied
981: (AFM interactions), but also between occupied and unoccupied (FM
982: interactions) $d-$orbitals on different sites.
983: The latter agrees with direct calculation of the exchange interaction
984: parameters in Hartree-Fock approximation~\cite{Solovyev-03}
985: and have to be taken into account in model calculations.
986:
987:
988: The presence of a mirror plane in GdFeO$_3$-type structure
989: perpendicular to $c-$axis~\cite{Imada-03} results in
990: such kind of orbital ordering in $c-$direction that the lobes of
991: the orbitals point to each other (see Fig.~\ref{c-axis}, left panel).
992: This is in strong contrast to the
993: orbital ordering in $ab-$plane, where orbitals are
994: directed away from the oxygens.
995:
996: The complicated orbital structure of LaTiO$_3$ does not favor the
997: isotropic magnetic interactions in LaTiO$_3$. But this is not a unique
998: situation. It is generally accepted that YTiO$_3$ shows Jahn-Teller
999: distortions and corresponding orbital ordering (Ref.~\onlinecite{Sawada-97}
1000: and references therein). Experimentally however
1001: it also has isotropic magnetic properties, which are hard to explain
1002: by orbital ordering. This would require an unrealistic set of
1003: parameters.\cite{Ulrich-02}
1004:
1005: Thus the isotropic character of exchange in both LaTiO$_3$ and YTiO$_3$
1006: remains an open problem. Whether the orbital
1007: fluctuations~\cite{Keimer-00,Ulrich-02} with
1008: CF splitting of $\sim$200~meV can resolve this problem, is not clear to
1009: us. Another option could be that the actual crystal (and consequently
1010: orbital) structure of LaTiO3 is somewhat different from the
1011: accepted one - see the recent results in Ref.~\onlinecite{Arao-02},
1012: where the indications of monoclinic distortions have been found.
1013: This question requires further investigations.
1014:
1015:
1016: \vspace{-0.3cm}
1017: \begin{center}
1018: \begin{figure}[h!]
1019: \centering
1020: \includegraphics[clip=true,width=0.2\textwidth]{La-c-axe-2.ps}
1021: \includegraphics[clip=true,width=0.25\textwidth]{Y-c-axe.ps}
1022: \caption{\label{c-axis}(color online) Orbital ordering in LaTiO$_3$ (left) and
1023: YTiO$_3$ (right) in
1024: $c$-direction obtained in the framework of the LDA+U calculations for AFM-G
1025: and FM structures, respectively.}
1026: \end{figure}
1027: \end{center}
1028:
1029:
1030: \section{CONCLUSIONS}
1031:
1032: Following the logic of the present paper we would
1033: like to stress two general points.
1034:
1035: (I) In the framework of LDA+U approximation we propose
1036: the {\it direct} method of calculation of crystal-field excitation energy.
1037: It can be useful for comparison with
1038: the results of spectroscopic measurements because it takes into account
1039: fast relaxations of electronic system. Any electron excitations in
1040: spectroscopy are accompanied by such relaxations and it should be
1041: definitely taken into consideration in the calculation of these
1042: excitations.
1043:
1044: Moreover, this scheme is more reliable then others because it uses
1045: not eigenvalues of LDA but its total energies. In addition, the method
1046: does not have an ambiguity in the definition of
1047: the basis set as any projection or downfolding procedure does, because
1048: all states included in the Hamiltonian are used.
1049: The latter is especially important for low symmetry systems
1050: where it is not clear what kind of linear
1051: combination of $d-$wave function and in which coordinate
1052: system should be taken for construction of a few orbital
1053: tight-binding Hamiltonian.
1054:
1055: (II) We probe this method on Y and La titanates, for which
1056: the value of crystal-field excitation energy is indeed
1057: essential. Using Wannier function projection of LDA full-orbital Hamiltonian
1058: it is found that CFS between the lowest and the next in energy $t_{2g}$
1059: level is 230~meV and 180~meV for La and Y titanates, in
1060: agreement with previous estimate.\cite{Pavarini-04}
1061:
1062: According to the results of present calculations,
1063: fast electronic relaxations reduce CFS by $\sim$30\%
1064: (it amounts to 70~meV in case of LaTiO$_3$) as compared with
1065: the difference of LDA one electron energies or CFS obtained
1066: by Madelung point charge summation.\cite{Cwik-03}
1067: However, the value of crystal-field excitation energy is still
1068: large enough to make orbital liquid formation in titanates rather
1069: unlikely. At the same time the intricate orbital ordering
1070: in contrast to the expectations from the model calculations
1071: does not favor isotropic magnetic interaction in LaTiO$_3$.
1072:
1073:
1074:
1075: \section{ACKNOWLEDGMENTS}
1076: We would like to thank I.V. Solovyev, D.E. Kondakov, M.A. Korotin,
1077: A.I. Poteryaev, S. Okatov and A.I. Lichtenstein for very useful
1078: discussion of calculation details. We also acknowledge fruitful
1079: communications with G. Khaliullin, M. Haverkort, M. Gr\"uninger,
1080: M. Cwik, E. M\"uller-Hartmann, M. Braden and especially L.H. Tjeng.
1081: This work is supported by Russian Foundation for
1082: Basic Research under the grants RFFI-04-02-16096 and RFFI-03-0239024,
1083: Netherlands Organization for Scientific Research through NWO 047.016.005
1084: and by the Deutsche Forschungsgemeinschaft through SFB 608.
1085:
1086:
1087: \begin{thebibliography}{21}
1088: \bibitem{Goodenough} J.B. Goodenough, {\it Magnetism and the Chemical
1089: Bond}, (Interscience publishers, New York-London. 1963).
1090: \bibitem{Ballhausen} C.J. Ballhausen, {\it Introduction to Ligand Field
1091: Theory}, (McGraw-Hill, New York, 1962).
1092: \bibitem{Elliott} J.P. Elliott and P.G. Dawber,
1093: {\it Symmetry in Physics}, (Oxford University Press, New York, 1989).
1094: \bibitem{Cwik-03} M. Cwik, T. Lorenz, J. Baier,
1095: R. M\"uller, G. Andr$\acute e$, F. Bour$\acute e$e,
1096: F. Lichtenberg, A. Freimuth, R. Schmitz, E. M\"uller-Hartmann,
1097: and M. Braden, Phys. Rev. B {\bf 68}, 060401(R) (2003).
1098: \bibitem{Nekrasov-03}I.A. Nekrasov, S.V. Streltsov, M.A. Korotin, and
1099: V.I. Anisimov, Phys. Rev. B {\bf 68}, 235113 (2003).
1100: \bibitem{Andersen-00} O.K. Andersen and T. Saha-Dasgupta,
1101: Phys. Rev. B {\bf 62}, R16219 (2000).
1102: \bibitem{Anisimov-04} V.I. Anisimov {\it et. al.,} cond-mat/0407359.
1103: \bibitem{Pavarini-04} E. Pavarini, S. Biermann, A. Poteryaev, A.I. Lichtenstein,
1104: A. Georges, and O. K. Andersen, Phys. Rev. Lett. {\bf 92}, 176403 (2004).
1105: \bibitem{Solovyev-03} I.V. Solovyev, Phys. Rev. B {\bf69}, 134403 (2004).
1106: \bibitem{Keimer-00} B. Keimer, D. Casa, A. Ivanov, J.W. Lynn,
1107: M.v. Zimmermann, J.P. Hill, D. Gibbs, Y. Taguchi, and Y. Tokura,
1108: Phys. Rev. Lett. {\bf 85}, 3946 (2000).
1109: \bibitem{Khaliullin-00} G. Khaliullin and S. Maekawa, Phys. Rev. Lett. {\bf 85}, 3950 (2000).
1110: \bibitem{Hester-97} J.R. Hester, K. Tomimoto, H. Noma, F.P. Okamure,
1111: and J. Akimitsu, Acta Cryst. {\bf B53}, 739 (1997).
1112: \bibitem{Meijer-99} G.I. Meijer, W. Henggeler, J. Brown, O.-S. Becker,
1113: J.G. Bednorz, C. Rossel, and P. Wachter, Phys. Rev. B {\bf 59}, 11832 (1999).
1114: \bibitem{Ulrich-02} C. Ulrich, G. Khaliullin, S. Okamoto,
1115: M. Reehuis, A. Ivanov, H. He, Y. Taguchi, Y. Tokura, and B. Keimer,
1116: Phys. Rev. Lett. {\bf 89}, 167202 (2002).
1117: \bibitem{Schmitz-04} R. Schmitz, O. Entin-Wohlman, A. Aharony, A. B. Harris,
1118: and E. M\"uller-Hartmann, cond-mat/0407524.
1119: \bibitem{Imada-03} M. Mochizuki and M. Imada, J. Phys. Soc. Jpn. {\bf 73}, 1833 (2004).
1120: \bibitem{Andersen-75} O.K. Andersen, Phys. Rev. B {\bf 12}, 3060 (1976).
1121: \bibitem{Wien2k} P. Blaha {\it et al.}, WIEN2k,
1122: An Augmented Plane Wave + Local Orbitals Program for Calculating Crystal
1123: Properties, Karlheinz Schwarz, Techn. Universit$\ddot a$t
1124: Wien, Austria, 2001.
1125: \bibitem{Shannon} R.D. Shannon, Acta Cryst., {\bf A32}, 751 (1976).
1126: \bibitem{Chainani-92} A. Chainani, M. Mathew, and D.D. Sarma, Phys. Rev. B {\bf 46}, 9976 (1992).
1127: \bibitem{Mattheiss86}L.F. Mattheiss and D.R.~Hamann, Phys. Rev. B {\bf 33}, 823 (1986).
1128: \bibitem{wannier} G.H. Wannier, Phys. Rev. {\bf 52}, 191 (1937).
1129: \bibitem{vanderbildt} N. Marzari and D. Vanderbilt, Phys. Rev. B {\bf 56}, 12847 (1997).
1130: \bibitem{Koster-53} G.F. Koster, Phys. Rev. {\bf 89}, 67 (1953).
1131: \bibitem{Kohn-73} W. Kohn, Phys. Rev. B {\bf 7}, 4388 (1973).
1132: \bibitem{pickett} Wei Ku, H. Rosner, W.E. Pickett, R.T. Scalettar, Phys. Rev. Lett. {\bf 89}, 167204 (2002).
1133: \bibitem{Anisimov-91} V.I. Anisimov, J. Zaanen, and O.K. Andersen, Phys.
1134: Rev. B {\bf 44}, 943 (1991).
1135: \bibitem{Pickett-98} W.E. Pickett, S.C. Erwin, and E.C. Ethridge, Phys. Rev. B {\bf 58}, 1201 (1998).
1136: \bibitem{Solovyev-96} I. Solovyev, N. Hamada, and K. Terakura,
1137: Phys. Rev. B {\bf 53}, 7158 (1996).
1138: \bibitem{Mizokawa-96} T. Mizokawa and A. Fujimori, Phys. Rev. B {\bf 54}, 5368 (1996).
1139: \bibitem{Abragam} A. Abragam and B. Bleaney, {\it Electron Paramagnetic
1140: Resonance of Transition Ions} (Oxford University Press, New York, 1970).
1141: \bibitem{Haverkort-04} M.W. Haverkort {\it et al.}, Phys. Rev. Lett.
1142: {\bf 94}, 056401 (2005).
1143: \bibitem{Gruninger-05} M. Gr\"uninger {\it et al.}, to be published.
1144: \bibitem{Kugel-82} K.I. Kugel and D.I. Khomskii, Usp. Fiz. Nauk {\bf 136}, 621 (1982).
1145: \bibitem{MacLean-79} D.A. MacLean {\it et al.}, J. Solid State Chem. {\bf 30}, 35 (1979).
1146: \bibitem{Akimitsu-01} J. Akimitsu {\it et al.}, J. Phys. Soc. Jpn. {\bf 70}, 3475 (2001).
1147: \bibitem{Sawada-97} H. Sawada, N. Hamada, and K. Terakura, Physica B,
1148: {\bf 237-238}, 46 (1997).
1149: \bibitem{Arao-02} M. Arao, Y. Inouem, and Y. Koyama,
1150: J. Phys. Chem. Solids {\bf 63}, 995 (2002).
1151: \end{thebibliography}
1152:
1153: \end{document}
1154: