1: \documentclass[twocolumn,aps,prl,amsmath,floatfix]{revtex4}
2: \usepackage{graphicx}
3: \begin{document}
4:
5: \title{Novel Mott Transitions in the Nonisotropic Two-Band Hubbard Model}
6: \author{A. Liebsch}
7: \affiliation{Institut f\"ur Festk\"orperforschung,
8: Forschungszentrum J\"ulich,
9: 52425 J\"ulich, Germany}
10: \begin{abstract}
11: The Mott transition in a two-band Hubbard model involving subbands of
12: different widths is studied as a function of temperature using dynamical
13: mean field theory combined with exact diagonalization. The phase diagram
14: is shown to exhibit two successive first-order transitions if the full
15: Hund's rule coupling is included. In the absence of spin-flip and
16: pair-exchange terms the lower transition remains first-order while the
17: upper becomes continuous.
18: \\ \\
19: PACS numbers: 71.20.Be, 71.27.+a.
20: \end{abstract}
21: \maketitle
22:
23: The nature of the metal insulator transition in materials involving
24: subbands of different widths has been intensively debated during the
25: recent years \cite{anisimov,epl,prl,koga1,prb,koga2,koga3,ferrero,%
26: medici,arita,song,knecht}.
27: This issue is relevant for the understanding of the effect of strong
28: local Coulomb interactions in systems such as Ca$_{2-x}$Sr$_x$RuO$_4$.
29: In this layer perovskite the partially filled Ru $t_{2g}$ bands for $x=2$
30: consist of wide $d_{xy}$ and narrow $d_{xz,yz}$ bands \cite{oguchi}.
31: As pointed out in \cite{prl2000}, the onsite Coulomb energy lies between
32: the subband widths: $W_{xz,yz}<U<W_{xy}$. The usual criterion with the
33: parameter $U/W$ as a measure of the importance of correlations must then
34: be generalized. The pure Sr compound is superconducting below 1.5 K
35: \cite{maeno}.
36: Iso-electronic replacement of Sr by Ca leads to an effective band
37: narrowing due to octahedral distortions \cite{fang}
38: and a metal insulator transition \cite{nakatsuji}.
39: As a consequence of non-cubic crystal fields many other transition metal
40: oxides also involve non-equivalent partially occupied subbands.
41:
42: A key question in these materials is whether the wide and narrow subbands
43: exhibit separate Mott transitions or whether single-particle hybridization
44: and inter-orbital Coulomb interactions ensure a single transition for all
45: bands simultaneously. This issue was studied initially by Anisimov et
46: al.~\cite{anisimov} and \mbox{Liebsch} \cite{epl} for Ca$_{2-x}$Sr$_x$RuO$_4$
47: using simplified band structure models which did not yet include the full
48: complexity due to Ca-induced octahedral distortions. Correlations were
49: treated in the dynamical mean field theory \cite{dmft} (DMFT) combined
50: with the non-crossing approximation (NCA) and Quantum Monte Carlo (QMC)
51: method, respectively. Since the NCA calculations (at $T=0$) neglected
52: interorbital Coulomb interactions the results showed separate,
53: `orbital-selective' Mott transitions for the narrow and wide subbands
54: \cite{anisimov}. In contrast, the QMC calculations (at $T=0.125$~eV)
55: included interorbital Coulomb interactions and suggested a common transition
56: for all $t_{2g}$ bands \cite{epl}.
57:
58: Recent theoretical studies of the Mott transition in a paramagnetic
59: two-band model system have led to apparent contradictions
60: \cite{prl,koga1,prb,koga2,koga3,ferrero,medici,arita}.
61: Including the full Hund's rule coupling Koga {\it et al.}~\cite{koga1}
62: found orbital-selective metal insulator transitions at $T=0$.
63: Neglecting spin-flip and pair-exchange terms Liebsch \cite{prb} obtained
64: at $T>0$ a single first-order transition, followed by a
65: bad-metallic/non-Fermi-liquid (NFL) phase.
66: As we argue below, the various results and conclusions are
67: consistent provided that the choice of crucial parameters such as
68: temperature and Hund's rule coupling is properly taken into account.
69:
70: In the present work we combine finite temperature DMFT with exact
71: diagonalization \cite{ed} (ED) to determine the $T/U$ phase diagram
72: of a two-band system consisting of non-hybridizing, half-filled subbands
73: with semi-circular density of states of width $W_1=2$~eV and $W_2=4$~eV.
74: The subbands interact
75: via intra- and interorbital Coulomb matrix elements $U$ and $U'=U-2J$,
76: where $J$ is the Hund's rule exchange integral. In contrast to the
77: multi-band QMC approach, which includes only Ising-like exchange terms to
78: avoid sign problems at low temperatures \cite{held}, ED permits also the
79: consideration of spin-flip and pair-exchange interactions. To be specific
80: we take $J=U/4$ which is approximately satisfied in several transition
81: metal oxides.
82:
83: The main result of this work is that the $T/U$ phase diagram in the
84: presence of the full Hund's rule exchange exhibits
85: {\it two successive first-order phase transitions}, with separate
86: hysteresis loops and coexistence regions. The intermediate region
87: corresponds to the $T>0$ analog of the orbital-selective Mott
88: (OSM) phase obtained at \mbox{$T=0$} in \cite{koga1}. On the other hand,
89: if spin-flip and pair-exchange terms are omitted, we find a
90: {\it single first-order transition succeeded by a non-Fermi-liquid
91: phase}, in agreement with previous QMC results \cite{prb}.
92: Both trends are consistent with those obtained by several groups
93: \cite{koga1,koga2,ferrero,medici,arita} for the same two-band model
94: at $T=0$.
95:
96: The ED/DMFT results are derived from a two-band generalization of the
97: approach employed for single-bands \cite{dmft,ed,gracias}. Since at $T>0$
98: all states of the impurity Hamiltonian are used in the construction
99: of the subband Green's functions $G_i(i\omega_n)$, two bath
100: levels per impurity level are taken into account ($n_s=6$ per spin).
101: To check the accuracy of this approximation we have evaluated the $T/U$
102: phase diagram of a single band for $n_s=3,\dots,6$. As shown in Fig.~1,
103: the stability boundaries $U_{c1}(T)$ and $U_{c2}(T)$ for $n_s=3$ are
104: slightly too low. Nevertheless, the overall shape of the phase diagram
105: agrees qualitatively with the converged results for $n_s=6$ \cite{tong}.
106: Thus, we are confident that in the two-band case $n_s=6$ also yields a
107: reasonable picture of the phase diagram. Preliminary results for $n_s=8$
108: will also be presented.
109:
110: \begin{figure}[t!]%1
111: \begin{center}
112: \includegraphics[width=4.5cm,height=8cm,angle=-90]{Fig1.ps}
113: \end{center}
114: \vskip-2mm
115: \caption{
116: Phase diagram for one-band Hubbard model, calculated within ED/DMFT
117: for $n_s=3,4,6$. Symbols (x): $T=0$ transitions obtained in \cite{bulla}.
118: }\end{figure}
119:
120: To analyze the metal/insulator transition we study the quasiparticle weights
121: $Z_i = 1/[1-d{\rm Re}\,\Sigma_i(\omega)/d\omega\vert_{\omega=0}]$,
122: which in the metallic range can be represented as
123: $Z_i \approx 1/[1- {\rm Im}\,\Sigma_i(i\omega_0)/ \omega_0]$, where
124: $\Sigma_i(i\omega_0)$ is the subband self-energy at the first Matsubara
125: frequency. Fig.~2(a) shows $Z_i(U)$ for $J'=J=U/4$, where $J'$
126: denotes spin-flip and pair-exchange terms \cite{fit}. Two critical regions
127: can be identified, each with hysteresis loops characteristic of first-order
128: phase transitions. The coexistence areas are $U^<_{c1}(T) < U < U^<_{c2}(T)$
129: near 2.0~eV and $U^>_{c1}(T) < U < U^>_{c2}(T)$ near 3.0~eV, where
130: $U^<_{cn}(T)$ and $U^>_{cn}(T)$ are the stability boundaries obtained for
131: increasing ($n=2$) and decreasing ($n=1$) $U$.
132: Let us denote the true critical energies of these transitions
133: as $U^<_c(T)$ and $U^>_c(T)$. Below $U^<_c(T)$ both bands are metallic
134: while above $U^>_c(T)$ both are insulating. At the lower transition
135: both bands undergo first-order transitions -- but in fundamentally
136: different ways: $Z_1(U)$ becomes very small while $Z_2(U)$ drops to a
137: finite value. The narrow band therefore undergoes a `complete'
138: metal/insulator transition, whereas the wide band exhibits an `incomplete'
139: transition to a new, considerably more correlated phase. This band becomes
140: fully insulating near 3.0~eV, where it exhibits a weak second hysteresis loop.
141: To our knowledge this is the first time that {\it sequential first-order
142: transitions} are identified in the $T/U$ phase diagram of a Hubbard model
143: involving non-equivalent bands \cite{kawakami}.
144: If such a material could be encountered experimentally, the conductivity
145: as a function of pressure would show two consecutive jumps.
146:
147: Fig.~2(b) shows $Z_i(U)$ for $J'=0$, $J=U/4$, i.e., in the absence of
148: spin-flip and pair-exchange terms. The results are similar to those in
149: Fig.~2(a), with the important exception that the wide band above the lower
150: transition is even more correlated and the upper transition is now continuous
151: at smaller $U$ \cite{continuous}.
152: Remarkably, the lower transition remains first-order for both subbands.
153: In this case the conductivity shows a jump at $U^<_c(T)$ but a change of slope
154: at $U^>_c(T)$. The results in Fig.~2(b) confirm the picture obtained previously
155: within the QMC for $T>0$, $J'=0$ which showed the existence of a single
156: first-order transition followed by a mixed insulating/bad-metallic phase
157: \cite{prb,ipt,bluemer1}. They also demonstrate that the two-band ED/DMFT
158: for $n_s=6$ is qualitatively accurate \cite{hyst}.
159:
160: \begin{figure}[t!]%2
161: \begin{center}
162: \includegraphics[width=8cm,height=8cm,angle=-90]{Fig2a.ps}
163: \includegraphics[width=8cm,height=8cm,angle=-90]{Fig2b.ps}
164: \end{center}
165: \vskip-2mm
166: \caption{
167: $Z_i(U)$ of nonisotropic two-band Hubbard model, calculated within ED/DMFT.
168: (a) $J'=J=U/4$, (b) $J'=0$, $J=U/4$. Solid (dashed) curves: narrow (wide)
169: band. Results for different temperatures are displaced vertically by $0.3$.
170: }\end{figure}
171:
172: The phase diagrams deduced from the ED results for $T\ge2.5$~meV are shown
173: in Fig.~3. For $J'=J$ as well as $J'=0$ the transition at $U^<_c(T)$ is
174: first-order for both subbands. The subsequent transition of the wide band at
175: $U^>_c(T)$ is first-order for $J'=J$ but continuous for $J'=0$.
176: At $U^<_c(T)$ the metal/insulator transition is complete only for the narrow
177: band. The wide band first undergoes a transition to a more strongly correlated
178: phase and becomes truly insulating at the second transition at $U^>_c(T)$.
179: The overall shape of the phase diagram for $J'=J$ agrees with the one
180: recently obtained by Inaba {\it et al.} \cite{kawakami}.
181:
182: \begin{figure}[t!]%3
183: \begin{center}
184: \includegraphics[width=5cm,height=8cm,angle=-90]{Fig3a.ps}
185: \includegraphics[width=5cm,height=8cm,angle=-90]{Fig3b.ps}
186: \end{center}
187: \vskip-2mm
188: \caption{
189: Phase diagram for nonisotropic two-band Hubbard model, calculated within
190: ED/DMFT. (a) $J'=J=U/4$, (b) $J'=0$, $J=U/4$. Solid dots in (a), (b):
191: stability boundaries of both subbands near lower first-order transition.
192: Open dots in (a): stability boundaries of wide band near upper first-order
193: transition. Dashed line in (b): approximate location of continuous transition
194: of wide band. Symbols (x): $T=0$ transitions obtained in \cite{koga1,koga2,arita};
195: (+): transitions at $T=31$~meV obtained in \cite{prb,knecht}.
196: Lines are guides to the eye.
197: }\end{figure}
198:
199: Fig.~3 also shows the critical Coulomb energies obtained at $T=0$
200: \cite{koga1,koga2,arita} and $T=31$~meV \cite{prb,knecht}. As illustrated
201: in Fig.~1 for the one-band case, the ED scheme with small $n_s$ underestimates
202: the critical Coulomb energies by about 0.1 to 0.2~eV. Preliminary results
203: for $n_s=8$ indicate similar shifts in the two-band case. Taking these
204: displacements into account the ED results shown in Fig.~3 are in excellent
205: correspondence with those obtained at $T=0$ and $T=31$~meV.
206: Evidently the conflicting conclusions reached in
207: \cite{anisimov,epl,prl,koga1,prb,koga2,ferrero,medici,arita} concerning
208: the nature of the Mott transition in multi-band systems were caused by
209: different behaviors obtained for $T=0$ vs. $T>0$ and $J'=J$ vs. $J'=0$.
210: Accounting for these different parameter choices, the DMFT treatments
211: are consistent.
212:
213: Quasiparticle spectra derived within the QMC/DMFT for $J'=0$ \cite{prb}
214: showed that in the intermediate phase the self-energy of the wide band
215: at small $\omega_n$ deviates significantly from metallic $\sim\omega_n$
216: behavior \cite{biermann}. Accordingly, the quasi-particle spectra show
217: a pseudogap which for larger $U$ gets more pronounced \cite{bad}, indicating
218: progressive non-Fermi-liquid properties. A true gap opens at $U^>_c(T)$.
219: Thus, $Z_2(U)>0$
220: in the region $U^<_c(T)<U<U^>_c(T)$ does not imply existence of quasiparticles.
221: As indicated in Fig.~3(b) the intermediate `orbital-selective Mott' phase in
222: the absence of spin-flip and pair-exchange is in fact a mixed insulating/NFL
223: phase. The same trend is found using the $T>0$ ED/DMFT \cite{future}.
224:
225: \begin{figure}[t!]%4
226: \begin{center}
227: \includegraphics[width=5cm,height=4cm,angle=-90]{Fig4a.ps}\hskip-3mm
228: \includegraphics[width=5cm,height=4cm,angle=-90]{Fig4b.ps}
229: \end{center}
230: \vskip-2mm
231: \caption{
232: Self-energy of wide band for $J'=J$ in intermediate phase at $U=2.4$~eV.
233: (a) $n_s=6$; (b) $n_s=8$. Solid dots: $T=10$~meV; empty dots: $T=5$~meV;
234: (+): $T=2.5$~meV.
235: }\end{figure}
236:
237: For $J'=J$, the low-frequency analysis of $\Sigma_2(i\omega_n)$ is more
238: intricate. As shown in Fig.~4(a) for $n_s=6$, $\Sigma_2(i\omega_n)$
239: reveals deviations from metallic $\sim\omega_n$ variation, giving rise to
240: small pseudogaps in the quasiparticle spectra. This behavior is incompatible
241: with Im\,$\Sigma_2(\omega)\sim\omega^2$ and suggests that, as for $J'=0$, in
242: the intermediate phase the wide band at $T>0$ violates Fermi-liquid theory.
243: % \cite{0505393}.
244: The extension of the present ED approach to $n_s=8$ indicates, however, that
245: the additional bath levels are important for the low-frequency variation of
246: $\Sigma_2(i\omega_n)$. As can be seen in Fig.~4(b), the deviations are absent
247: and the $\sim\omega_n$ variation is consistent with Fermi-liquid behavior.
248: In fact, the shoulder near $\omega_0=0.06$ suggests that Fermi-liquid
249: properties persist up to about $T=\omega_0/\pi\approx 20$~meV. Thus,
250: the OSM phase in Fig.~3(a) is the $T>0$ analog of the orbital-selective
251: Mott phase identified first by Koga {\it el al.}~\cite{koga1} at $T=0$.
252: A more complete discussion of the results for $n_s=8$ will be given elsewhere
253: \cite{future}. Because of finite size limitations of the present ED/DMFT
254: scheme associated with the small number of bath levels, a precise
255: determination of low-temperature properties is not possible. Nevertheless,
256: an approximate extrapolation of $\Sigma_2(i\omega_n)$ indicates that the
257: $T\rightarrow0$ limit for $J'=J$ satisfies Fermi-liquid criteria, in
258: agreement with previous $T=0$ studies \cite{koga1,koga2,ferrero,medici,arita}.
259:
260: In view of the importance of spin-flip and pair-exchange terms for
261: the Mott transition in multi-band materials \cite{pruschke} it is
262: desirable to investigate the $T/U$ phase diagram by methods
263: which permit adequate treatment of the complete Hund's rule matrix,
264: for instance, a two-band extension of $T>0$ numerical renormalization
265: group studies \cite{costi}; see also \cite{kawakami}.
266: Because of sign problems, recent QMC extensions including spin-flip
267: and pair-exchange terms are limited to $T\ge 1/6$~eV$\ \gg T_c$
268: \cite{koga3} or $T=0$ \cite{arita}. Also, other recent works
269: \cite{ferrero,medici} employing a variety of quantum impurity
270: methods deal so far mainly with $T=0$ and do not yet allow the
271: identification of the multi-band Mott transition at general $T,U$
272: values.
273:
274: In summary, the $T/U$ phase diagram of the Hubbard model involving
275: half-filled, non-equivalent subbands is shown to be remarkably rich.
276: The competing kinetic energy scales, coupled via Coulomb and exchange
277: energies, give rise to {\it sequential first-order phase transitions}.
278: The lower transition separates a purely metallic phase from a mixed
279: phase with insulating narrow and strongly-correlated wide subbands.
280: The wide band becomes insulating at the second first-order transition.
281: Omission of spin-flip and pair-exchange terms enhances the correlations
282: in the wide band in the intermediate phase so that it no
283: longer satisfies Fermi-liquid criteria, and modifies the upper phase
284: transition from first-order to continuous.
285:
286: For the analysis of experimental data of materials such as
287: Ca$_{2-x}$Sr$_x$RuO$_4$ it is necessary to account also for hybridization
288: between orbitals. Preliminary studies of this effect within two-band
289: models for $T\gg T_c$ \cite{koga3} and $T=0$ \cite{medici,song} suggest
290: significant changes. Moreover, spatial fluctuations \cite{kotliar} and
291: deviations from half-filling might play a decisive role close to the
292: Mott transition. More work is needed to investigate whether both
293: first-order transitions persist in the presence of these effects, or
294: whether the weak first-order behavior of the upper transition disappears
295: and only the dominant lower transition survives as the common first-order
296: Mott transition for all bands.
297:
298: I like to thank Th. Costi, K. Held, N. Kawakami, and E. Koch for valuable
299: discussions.
300:
301: Email: a.liebsch@fz-juelich.de
302: \begin{thebibliography}{99}
303: \bibitem{anisimov}
304: V. I. Anisimov, I. A. Nekrasov, D. E. Kondakov, T. M. Rice
305: and M. Sigrist,
306: Euro. Phys. J. B {\bf 25}, 191 (2002).
307:
308: \bibitem{epl}
309: A. Liebsch,
310: Europhys. Lett. {\bf 63}, 97 (2003).
311:
312: \bibitem{prl}
313: A. Liebsch,
314: Phys. Rev. Lett. {\bf 91}, 226401 (2003).
315:
316: \bibitem{koga1}
317: A. Koga, N. Kawakami, T. M. Rice, and M. Sigrist,
318: Phys. Rev. Lett. {\bf 92}, 216402 (2004).
319:
320: \bibitem{prb}
321: A. Liebsch,
322: Phys. Rev. B {\bf 70}, 165103 (2004).
323:
324: \bibitem{koga2}
325: A. Koga, N. Kawakami, T. M. Rice, and M. Sigrist,
326: cond-mat/0406457.
327:
328: \bibitem{koga3}
329: A. Koga, N. Kawakami, T. M. Rice, and M. Sigrist,
330: cond-mat/0503651.
331:
332: \bibitem{ferrero}
333: M. Ferrero, F. Becca, M. Fabrizio, and M. Capone,
334: cond-mat/0503759
335:
336: \bibitem{medici}
337: L. de' Medici, A. Georges and S. Biermann,
338: cond-mat/0503764.
339:
340: \bibitem{arita}
341: R. Arita and K. Held,
342: cond-mat/0504040.
343:
344: \bibitem{song}
345: Yun Song and L.-J. Zou,
346: cond-mat/0504420.
347:
348: \bibitem{knecht}
349: C. Knecht, N. Bl\"umer, and P.G.J. van Dongen,
350: cond-mat/0505106.
351:
352: \bibitem{oguchi}
353: T. Oguchi,
354: Phys. Rev. B {\bf 51}, 1385 (1995);
355: I. I. Mazin and D. Singh,
356: Phys. Rev. Lett. {\bf 79}, 733 (1997).
357:
358: \bibitem{prl2000}
359: A. Liebsch and A. I. Lichtenstein,
360: Phys. Rev. Lett. {\bf 84}, 1591 (2000).
361:
362: \bibitem{maeno}
363: Y. Maeno, T. M. Rice and M. Sigrist,
364: Physics Today {\bf 54}(1), 42 (2001).
365:
366: \bibitem{fang}
367: A. Fang and K. Terakura,
368: Phys. Rev. B {\bf 64}, R020509 (2001);
369: A. Fang, N. Nagaosa, and K. Terakura,
370: Phys. Rev. B {\bf 69}, 045116 (2004).
371:
372: \bibitem{nakatsuji}
373: S. Nakatsuji and Y. Maeno,
374: Phys. Rev. Lett. {\bf 84}, 2666 (2000).
375:
376: \bibitem{dmft}
377: For a review, see:
378: A. Georges, G. Kotliar, W. Krauth and M. J. Rozenberg,
379: Rev. Mod. Phys. {\bf 68}, 13 (1996).
380:
381: \bibitem{ed}
382: M. Caffarel and W. Krauth,
383: Phys. Rev. Lett. {\bf 72}, 1545 (1994).
384:
385: \bibitem{held}
386: For a discussion of this problem, see:
387: K. Held and D. Vollhardt,
388: Eur. Phys. J. B {\bf 5}, 473 (1998).
389:
390: \bibitem{gracias}
391: I like to thank the authors of \cite{dmft} for making the single-band ED
392: code available, and A.I. Lichtenstein for parts of the two-band ED code.
393:
394: \bibitem{bulla}
395: See: R. Bulla, T. A. Costi, and D. Vollhardt,
396: Phys. Rev. B {\bf 64}, 045103 (2001) and references herein.
397:
398: \bibitem{tong}
399: N.H. Tong, S.Q. Shen, and F.C. Pu,
400: Phys. Rev. B {\bf 64}, 235109 (2001).
401:
402: \bibitem{fit}
403: More accurate results at low $\omega_n$ and $T$ are obtained by minimizing
404: differences between impurity and Anderson model Green's functions via
405: $\sum_n \vert\Delta G_i(i\omega_n)\vert^2$ rather than
406: $\sum_n \vert\Delta G_i(i\omega_n)\vert$. I thank E. Koch for this suggestion.
407:
408: \bibitem{kawakami}
409: After completion of this work we learned of similar results
410: by K. Inaba {\it et al.}, cond-mat/0506151.
411:
412: \bibitem{continuous}
413: The term `continuous' indicates here that there is no jump; it may
414: also signify crossover behavior.
415:
416: \bibitem{ipt}
417: Surprisingly, two-band DMFT using iterated perturbation theory for
418: $J'=J$ reveals the lower transition as first-order but the upper as
419: continuous \cite{prb}.
420:
421: \bibitem{bluemer1}
422: Recent QMC/DMFT results \cite{knecht} for $J=U/4,\,J'=0$ are in
423: agreement with those of \cite{prb}. Compare $Z_i(U)$ and $N_i(\omega)$
424: in Figs.~9(b) and 11 \cite{prb} with Figs.~1(b)(inset) and 3,4
425: \cite{knecht}.
426: Thus, the claims in \cite{knecht} that ``the second transition was
427: not seen in earlier QMC and IPT studies'' and ``our high-precision
428: data correct earlier QMC results'' are unfounded.
429: See also: A. Liebsch, cond-mat/0506138.
430:
431: \bibitem{hyst}
432: The QMC hysteresis region is slightly wider than in ED.
433: This could be due to slow convergence of QMC at $T=0.02$~eV
434: and too small $n_s$ in ED.
435:
436: \bibitem{biermann}
437: Similar QMC results were obtained for $J'=0$, $T>0$, $W_1=0.1W_2$ by
438: Biermann {\it et al.}, cond-mat/0505737.
439:
440: \bibitem{bad}
441: See also Fig.~3 of \cite{knecht}.
442:
443: %\bibitem{0505393}
444: % A.Liebsch, cond-mat/0505393.
445:
446: \bibitem{future}
447: A. Liebsch, to be published.
448:
449: \bibitem{pruschke}
450: See also: Th. Pruschke and R. Bulla, cond-mat/0411186.
451:
452: \bibitem{costi}
453: Th. Costi, to be published.
454:
455: \bibitem{kotliar}
456: O. Parcollet, G. Biroli, and G. Kotliar,
457: Phys. Rev. Lett. {\bf 92}, 226402 (2004).
458: \end{thebibliography}
459: \end{document}
460:
461:
462: