1: \documentclass[twocolumn,aps,prl,amsmath,floatfix]{revtex4}

2: \usepackage{graphicx}

3: \begin{document}

4: \title{

5: Non-Fermi-liquid phases in the two-band Hubbard model:\\

6: Finite-temperature exact diagonalization study of Hund's rule coupling}

7: \author{A.~Liebsch and T.~A.~Costi} 

8: \affiliation{Institut f\"ur Festk\"orperforschung, 

9:              Forschungszentrum J\"ulich, 

10:              52425 J\"ulich, Germany}

11: \begin{abstract}

12: The two-band Hubbard model involving subbands of different widths is 

13: investigated via finite-temperature exact diagonalization (ED) and 

14: dynamical mean field theory (DMFT). In contrast to the quantum Monte Carlo 

15: (QMC) method which at low temperatures includes only Ising-like exchange 

16: interactions to avoid sign problems, ED permits a treatment of Hund's 

17: exchange and other onsite Coulomb interactions on the same footing.

18: The role of finite-size effects caused by the limited number of bath 

19: levels in this scheme is studied by analyzing the low-frequency behavior 

20: of the subband self-energies as a function of temperature, and by 

21: comparing with numerical renormalization group (NRG) results for an

22: effective one-band model. 

23: For half-filled, non-hybridizing bands, the metallic and insulating 

24: phases are separated by an intermediate mixed phase with an insulating  

25: narrow and a bad-metallic wide subband. The wide band in this phase 

26: exhibits different degrees of non-Fermi-liquid behavior, depending on 

27: the treatment of exchange interactions. Whereas for complete Hund's 

28: coupling, infinite lifetime is found at the Fermi level, in the 

29: absence of spin-flip and pair-exchange, this lifetime becomes finite. 

30: Excellent agreement is obtained both with new NRG and previous 

31: QMC/DMFT calculations. These results suggest that-finite temperature 

32: ED/DMFT might be a useful scheme for realistic multi-band materials.\\

33: \\

34: PACS. 71.20.Be  Transition metals and alloys - 71.27+a Strongly correlated

35: electron systems 

36: \end{abstract}

37: \maketitle

38: 

39: \section{1 \ Introduction}

40: 

41: Strongly correlated materials exhibit a wealth of fascinating 

42: physical phenomena associated with complex single-electron and 

43: many-electron interactions. Transition metal oxides, for example,

44: tend to have partially filled shells of highly correlated $d$ 

45: electrons, surrounded by complicated lattice geometries, with 

46: many atoms and electrons per unit cell. The theoretical description 

47: of the electronic properties of these materials is a challenging 

48: topic in condensed matter physics. Significant advances were 

49: achieved during recent years via the dynamical mean field theory 

50: (DMFT) \cite{rmp,reviews} which provides a treatment of 

51: single-electron and many-electron interactions on the same footing. 

52: 

53: For realistic materials, DMFT has been used extensively in 

54: combination with the quantum Monte Carlo (QMC) method \cite{qmc}. 

55: This approach has the advantage that it can be readily applied to 

56: systems consisting of two or more subbands. It has the drawback,

57: however, that, to avoid sign problems at low temperatures, it 

58: includes only Ising-like exchange interactions \cite{held}.

59: Improvements of the QMC method to include the full Hund's coupling 

60: are so far limited to $T=0$ or rather high temperatures 

61: \cite{arita,koga3,rubtsov}.  

62: In view of this limitation there is clearly a need to explore 

63: alternative methods that are applicable to multi-band materials 

64: and complete onsite exchange interactions.

65: 

66: The aim of the present work is two-fold: First, exact diagonalization 

67: (ED) \cite{ed,rmp} is proposed as a potentially highly useful impurity 

68: solver for finite-$T$ DMFT studies of realistic systems. The attractive 

69: feature of this approach is that, in contrast to finite-$T$ QMC, onsite 

70: Coulomb and exchange interactions are treated on the same basis. 

71: Second, we apply finite-$T$ ED/DMFT to a highly nontrivial system which 

72: has recently received considerable attention, namely, the two-band 

73: Hubbard model consisting of subbands of different widths

74: \cite{anisimov,epl,prl,koga1,prb70,koga2,koga3,prl05,inaba,knecht,

75: biermann05,ferrero,medici,arita,song}. The phase diagram of this 

76: system was recently evaluated in Refs.~\cite{prl05,inaba}. Here we focus 

77: on the electronic properties of the so-called orbital-selective phase

78: in which the narrow band is insulating while the wide band is still 

79: metallic. To examine the influence of finite-size effects in the ED

80: approach, and to investigate the electronic properties in the limit 

81: of low temperatures, we have also performed numerical renormalization 

82: group (NRG) DMFT calculations for a simplified two-band model which 

83: is particularly suited for the intermediate phase.

84: 

85: The main result of this paper is that the two-band ED/DMFT calculations 

86: provide a correct picture of the electronic properties of the Hubbard 

87: model involving nonequivalent subbands, including the unusual 

88: non-Fermi-liquid properties of the orbital-selective phase.  

89: This outcome is remarkable since, for computational reasons, the number 

90: of bath levels per impurity orbital is necessarily smaller than in 

91: analogous one-band models. Nevertheless, despite this limitation, 

92: the ED results are in good qualitative or, in some cases, quantitative 

93: agreement with the NRG results. In the Ising case they also agree very 

94: well with previous QMC/DMFT results \cite{prb70,biermann05}. 

95: 

96: The overall consistency with the NRG and QMC calculations suggests 

97: that finite-$T$ ED/DMFT might indeed be very useful in the future 

98: to analyze realistic materials, especially, if full diagonalization

99: is replaced by finite-$T$ Lanczos methods \cite{jaklic,capone}

100: in order to be able to deal with larger cluster sizes.

101: 

102: Hubbard models involving nonequivalent subbands are relevant for 

103: compounds such as Ca$_{2-x}$Sr$_x$RuO$_4$ and Na$_x$CoO$_2$, where, 

104: as a result of the nearly two-dimensional lattice structure, coexisting 

105: narrow and wide bands arise naturally. Thus, onsite Coulomb energies 

106: can be simultaneously large and small relative to the widths of 

107: important subbands. As a function of doping concentration, both 

108: materials give rise to a remarkably rich sequence of phases, including 

109: superconductivity and Mott insulating behavior \cite{maeno,nacoo}. 

110: Evidently, the competition between multiple kinetic energy 

111: scales and local Coulomb and exchange energies is an important 

112: feature of these strongly correlated systems.

113: 

114: A consistent treatment of Coulomb and exchange interactions in materials

115: of this kind is important since it has recently become clear that the 

116: Hund's coupling has a decisive influence on the nature of the Mott 

117: transition \cite{koga2,prl05,ferrero,medici,arita,pruschke}. 

118: In fact, the different treatments of exchange terms in earlier finite-$T$ 

119: QMC \cite{epl,prl,prb70} and zero-$T$ ED \cite{koga1} calculations have 

120: given rise to some confusion, with apparently contradictory results. 

121: As was clarified in Ref.~\cite{koga2} for $T=0$ and in Ref.~\cite{prl05} 

122: for $T>0$, however, the QMC and ED results are in agreement provided 

123: that exchange interactions are treated in the same manner. Thus, for 

124: full Hund's coupling the two-band Hubbard model exhibits successive 

125: first-order transitions. In striking contrast, in the absence of 

126: spin-flip and pair-exchange only the lower transition remains 

127: first-order \cite{prb70,prl05,knecht}. 

128: 

129: 

130: Moreover, as will be discussed in detail below, the nature of the 

131: intermediate phase depends in a subtle manner on the treatment of 

132: exchange interactions. In this regard the ED and NRG results yield 

133: the following picture: For full Hund's coupling, the wide band has 

134: infinite lifetime at $E_F$ but does not satisfy Fermi-liquid criteria 

135: at finite frequencies. This finding is consistent with recent results 

136: obtained by Biermann {\it et al.} \cite{biermann05} in $T=0$ two-band 

137: ED/DMFT calculations. For Ising exchange, instead, correlations 

138: are significantly enhanced and the lifetime becomes finite even at 

139: $E_F$, in agreement with QMC calculations \cite{prb70,biermann05}.   

140: 

141: {\it

142: Thus, the Mott transition in the Hubbard model involving different

143: subbands does not consist of equivalent sequential transitions. Instead,

144: when the narrow band becomes insulating, the wide band is forced into a

145: bad-metallic state whose deviations from Fermi-liquid behavior depend

146: in a qualitative manner on the treatment of exchange interactions.} 

147:          

148: In the past, finite-$T$ ED/DMFT methods have been applied mainly 

149: to single-band cases, where the incorporation of an appropriate 

150: cluster of bath levels is feasible. Throughout this paper we consider 

151: two impurity levels, each surrounded by either 2 or 3 bath levels, 

152: giving total cluster sizes $n_s=6$ or $n_s=8$ per spin. Below we 

153: demonstrate that even a cluster size of $n_s=6$ provides qualitatively 

154: correct subband self-energies. For instance, at $T=10$~meV the critical 

155: Coulomb energies of the orbital-selective Mott transitions for 

156: $n_s=6$ differ by only about {0.1\ldots0.2}~eV from those derived 

157: for $n_s=8$. This finding is interesting since it suggests that 

158: finite-$T$ ED/DMFT calculations for more realistic three-band models 

159: using a cluster size of $n_s=9$ (three impurity levels, each coupled

160: to two bath levels) should be useful. This would allow one to 

161: re-examine the Mott transition in systems that have been studied 

162: previously using QMC/DMFT for Ising exchange interactions.

163: 

164: A more accurate representation of low-frequency properties at low

165: temperatures requires three bath levels per impurity orbital: one 

166: near the Fermi level to provide adequate metallicity, and two for 

167: the upper and lower Hubbard bands, giving $n_s=8$ per spin. This 

168: extension leads to a significant reduction of finite-size effects.

169: Several comparisons of results for $n_s=6$ and $n_s=8$ are provided 

170: below to illustrate the range of applicability of the two-band 

171: ED/DMFT approach.    

172: 

173: The outline of the paper is as follows: In Section 2 the multiband

174: Hubbard model is specified and its numerical solution via the 

175: finite-$T$ exact diagonalization method is discussed. Section 3

176: describes the effective one-band model used in the NRG approach

177: to evaluate the electronic properties of the wide band in the 

178: intermediate phase when the narrow band is insulating. 

179: In Section 4 the ED/DMFT is applied to the purely metallic phase

180: just below the first Mott transition. 

181: Section 5 deals with the intermediate phase in the presence of full

182: Hund's coupling, where wide band exhibits infinite lifetime at

183: $E_F$, combined with non-Fermi-liquid behavior at finite frequencies.

184: In Section 6 this intermediate phase is considered in the absence of

185: spin-flip and pair-exchange terms, giving rise to finite lifetime 

186: even at $E_F$. In Section 7 the two-band ED approach is applied 

187: to the case studied previously within QMC/DMFT \cite{prb70}.

188: Section 8 presents analogous results for the case considered 

189: recently within QMC/DMFT by Biermann {\it et al.} \cite{biermann05}. 

190: Section 9 contains a brief discussion of iterated perturbation 

191: theory (IPT) DMFT results for the two-band model and Section 10 

192: provides a summary and outlook.             

193: 

194: \section{2 \ Multiband exact diagonalization method} %II

195: 

196: The two-band Hubbard model for non-equivalent subbands is represented

197: by the Hamiltonian:

198: \begin{eqnarray}

199:    H &=&  H_0 \ + \ H_1 \ + \ H_2 \             \nonumber\\ 

200:  H_0 &=& \sum_{lmi\sigma}  t_{lmi} 

201:              c_{li\sigma}^+ c_{mi\sigma}  \nonumber\\

202:  H_1 &=& \sum_{li} U n_{li\uparrow} n_{li\downarrow} \ + \

203:          \sum_{l\sigma\sigma'} (U'-J\delta_{\sigma\sigma'}) 

204:                  n_{l1\sigma} n_{l2\sigma'}                \nonumber

205: \end{eqnarray}

206: \begin{eqnarray}

207:  H_2 &=& -J'\sum_l [c_{l1\uparrow}^+ c_{l1\downarrow}

208:             c_{l2\downarrow}^+ c_{l2\uparrow}    +  {\rm H.c.} ] 

209:                                                             \nonumber\\

210:      &&  -J'  \sum_l [c_{l1\uparrow}^+ c_{l1\downarrow}^+                   

211:                   c_{l2\uparrow} c_{l2\downarrow}    + {\rm H.c.} ]\, ,

212: \end{eqnarray}

213: where $c_{li\sigma}^+$ and $c_{li\sigma}$ are creation and

214: annihilation operators for electrons at site $l$ in orbital $i=1,2$

215: with spin $\sigma$ and  $n_{li\sigma}=c_{li\sigma}^+ c_{li\sigma}$. 

216: H.c.~denotes Hermitian conjugate terms.

217: $H_0$ is the single-particle Hamiltonian which we represent by 

218: half-filled, non-hybridizing bands with density of states 

219: $N_i(\omega)= 2/(\pi D_i)\sqrt{1- (\omega/D_i)^2}$ and widths 

220: $W_1=2$~eV, $W_2=4$~eV, where  $W_i=2D_i$. $H_1$ represents the 

221: anisotropic, Ising-like onsite Coulomb and exchange interactions 

222: and $H_2$ additional spin-flip and pair-exchange contributions. For 

223: isotropic Hund's coupling one has $J'=J=(U-U')/2$. Below, we discuss 

224: results for $J'=J$ as well as $J'=0$. The latter case corresponds to 

225: the Ising-like exchange treatment in previous QMC/DMFT calculations 

226: \cite{prl,prb70}.

227: 

228: The ED/DMFT results are derived from a two-band generalization of the 

229: approach employed for single bands \cite{ed,rmp}. 

230: The effective Anderson impurity Hamiltonian includes impurity levels 

231: $\epsilon_{1,2}$ and bath levels $\epsilon_{k=3\ldots n_s}$ 

232: (see Fig.~1). For particle-hole symmetry $\epsilon_{1,2}=0$. 

233: Each impurity level interacts with its own bath, so that for $n_s=6$  

234: impurity level 1 couples to bath levels 3,4 with $\epsilon_4=-\epsilon_3$, 

235: while level 2 couples to bath levels 5,6 with $\epsilon_6=-\epsilon_5$. 

236: For $n_s=8$ impurity levels 1,2 couple in addition to the bath levels 

237: $\epsilon_7=0$ and $\epsilon_8=0$, respectively.

238: 

239: \begin{figure}[b!]%1  see Gnuz in  ED8/Jump  and   ED6/T=0.010

240:   \begin{center}

241:   \includegraphics[width=6.0cm,height=3cm,angle=-0]{fig1.ps}

242:   \end{center}

243:   \vskip-2mm

244: \caption{\label{ED}

245: Level diagram for ED scheme. Blue lines: impurity levels;

246: red solid (plus dashed) lines: bath levels for $n_s=6$ ($n_s=8$). 

247: Green lines: Coulomb and exchange interactions between impurity

248: levels; red dotted lines: hopping interactions between impurity

249: and bath levels.

250: }\end{figure}

251: 

252: Since at $T>0$ all states of the impurity Hamiltonian are used in the 

253: construction of the subband Green's functions, for $n_s=6$ the largest 

254: matrix to be diagonalized corresponds to the sector $n_\uparrow=

255: n_\downarrow=3$, with dimension $[n_s!/((n_s/2)!)^2]^2=400$. 

256: For $n_s=8$ this dimension increases to $4900$. 

257: We neglect hybridization between bands, so that self-energies and Green's 

258: functions are diagonal in orbital space. Denoting eigenvalues and 

259: eigenvectors of the impurity Hamiltonian by $E_\nu$ and $|\nu \rangle$, 

260: the 

261: subband Green's functions are evaluated from the expression 

262: \begin{equation}

263:  G_i(i\omega_n) = \frac{1}{Z} \sum_{\nu\mu} 

264:             \frac{\vert\langle \nu|c_{0i\sigma}^+ |\mu \rangle \vert^2}

265:                     {E_\nu - E_\mu - i\omega_n}

266:    [e^{-\beta E_\nu} + e^{-\beta E_\mu}].    

267: \end{equation}

268: %\begin{eqnarray}

269: % G_i(i\omega_n) &=& \frac{1}{Z} \sum_{\nu\mu} 

270: %            \frac{\vert\langle \nu|c_{0i\sigma}^+ |\mu \rangle \vert^2}

271: %                    {E_\nu - E_\mu - i\omega_n}

272: %   [e^{-\beta E_\nu} + e^{-\beta E_\mu}]\, .    

273: %                    {E_\nu - E_\mu - i\omega_n} \nonumber \\ 

274: %   &&\  \times\ [e^{-\beta E_\nu} + e^{-\beta E_\mu}]\, .    

275: %\end{eqnarray}

276: Here, $\beta=1/k_B T$, $\omega_n=(2n+1)\pi/\beta$ are Matsubara 

277: frequencies, and $Z=\sum_\nu {\rm exp}(-\beta E_\nu)$ is the partition 

278: function. $l=0$ denotes the impurity site. Since we consider only 

279: paramagnetic phases the spin index of Green's functions and 

280: self-energies is omitted for convenience. Because of the particle-hole

281: symmetry of the present system, Green's functions and self-energies

282: are purely imaginary quantities. 

283: 

284: \begin{figure}[t!]%1  see Gnuz in  ED8/Jump  and   ED6/T=0.010

285:   \begin{center}

286: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{ZU8.ps}

287: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{ZU6.ps}

288:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig2a.ps}

289:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig2b.ps}

290:   \end{center}

291:   \vskip-2mm

292: \caption{\label{ZU}

293: $Z_i(U)$ as a function of $U$ for $J=J'=U/4$ at $T=10$~meV, 

294: calculated within ED/DMFT for different cluster sizes:

295: (a) $n_s=8$; (b) $n_s=6$. 

296: Solid (red) curves: narrow band, dashed (blue) curves: wide band.  

297: }\end{figure}

298: 

299: To provide an overview of the Mott transitions in this model and 

300: to illustrate the sensitivity of the critical Coulomb energies 

301: to the cluster size used in the ED/DMFT, we show in Fig.~\ref{ZU} 

302: the quantities $Z_i(U)=1/[1- {\rm Im}\,\Sigma_i(i\omega_0)/\omega_0]$

303: for $n_s=6$ and $n_s=8$ at $T=10$~meV, assuming $J'=J=U/4$.

304: Here $\Sigma_i(i\omega_0)$ are the subband self-energies at the 

305: first Matsubara frequency. Thus, for isotropic Hund's 

306: exchange and $n_s=8$, $U_{c1}\approx2.1$~eV and $U_{c2}\approx3.1$~eV.

307: (In our notation the $U_{ci=1,2}$ refer here to the subband critical

308: Coulomb energies for increasing $U$ and should not be confused with 

309: the stability boundaries for increasing vs decreasing $U$ at the 

310: individual Mott transitions.) Near both critical Coulomb

311: energies $Z_i(U)$ exhibit hysteresis behavior characteristic of 

312: first-order transitions. For $n_s=6$, the common metallic phase 

313: is stable only up to about $U_{c1}\approx2$~eV.

314: Thus, inclusion of zero-energy bath levels supports the metallic 

315: character of the DMFT solution. A shift of about 0.2~eV is found 

316: for $U_{c2}$ where the wide band becomes insulating. These shifts 

317: are consistent with single-band ED/DMFT results\cite{prl05} upon 

318: increasing $n_s$ from 3 to 4; only minor additional shifts occur 

319: in this case between $n_s=4$ and the fully converged results for 

320: $n_s=6$.

321: 

322: According to the QMC/DMFT results discussed in Ref.~\cite{prb70}, 

323: the Mott transitions for $J=U/4$ and $J'=0$, i.e., for Ising-like 

324: exchange, are located at $U_{c1}\approx2.1$~eV and 

325: $U_{c2}\approx2.7$~eV. Moreover, apart from the shift of $U_{c2}$, 

326: this transition is no longer first-order. The two-band ED/DMFT 

327: results in Ref.~\cite{prl05} for $n_s=6$ confirmed this fundamental 

328: difference between the $J'=J$ and $J'=0$ treatments.  

329: 

330: As will be discussed in detail below, the wide band in the intermediate

331: phase, i.e., for $U_{c1}<U<U_{c2}$, does not satisfy Fermi-liquid 

332: criteria. Thus, the fact that $Z_2(U)$ as defined above is finite 

333: in this region does not imply existence of ordinary quasiparticles.

334: In the case of full Hund's coupling, the imaginary part of the 

335: self-energy of the wide band vanishes in the low-frequency limit, 

336: but does not increase linearly at small $i\omega_n$, implying 

337: non-quadratic variation at real frequencies. For Ising-like exchange,

338: Im\,$\Sigma_2(i\omega_n)$ remains finite for $\omega_n\rightarrow0$,

339: implying finite lifetime even for states at $E_F$. Plots like those 

340: in Fig.~1 are nevertheless useful since they permit a convenient

341: identification of phase transitions. Thus, $Z_i(U)\rightarrow0$ 

342: at $U_{ci}$ indicates that the narrow or wide subbands become 

343: insulating for $U>U_{ci}$, respectively. (Of course, due to the 

344: discrete representation along the imaginary frequency axis at finite 

345: $T$, $Z_i(U)$ does not fully vanish.) 

346: 

347: \begin{figure}[t!]%2

348:   \begin{center}

349: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{W1W2J.ps}

350: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{W1W2J0.ps}

351:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig3a.ps}

352:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig3b.ps}

353:   \end{center}

354:   \vskip-2mm

355: \caption{\label{W1W2}

356: Imaginary part of self-energy of wide band in intermediate phase

357: for decreasing widths $W_1$ of narrow band, with fixed $W_2=4$~eV, 

358: $U=2.4$~eV, $n_s=8$, $T=10$~meV. Alternating solid (red) and dashed

359: (blue) curves: $W_1=0.4,\,0.8,\,\,1.2,\,1.6,\,2.0$~eV (from bottom).

360: (a) $J=J'=U/4$; (b)  $J=U/4$, $J'=0$.

361: }\end{figure}

362: 

363: To check the convergence of the ED results with cluster size $n_s$

364: we compare them with NRG calculations for an effective one-band model

365: which is most suitable for the intermediate phase and which is

366: specified in the following section. Since the narrow band is insulating 

367: in this phase one of the key features of the effective model is the 

368: omission of one-electron hopping in the narrow band. Accordingly, the

369: upper and lower Hubbard peaks will be centered at about $\pm U/2$, but

370: the influence of their width is neglected. (In a local description, 

371: the width of the Hubbard bands is approximately given by $W$ \cite{rmp}.)

372: In Fig.~\ref{W1W2} we show that this approximation indeed has only a 

373: minor effect on the electronic properties of the metallic wide band. 

374: Keeping $W_2=4$~eV fixed and changing $W_1$ between $0.1\,W_2$ and 

375: $0.5\,W_2$, we notice that, both for isotropic and anisotropic exchange 

376: coupling, the self-energy of the wide band at $U=0.6\,W_2=2.4$~eV

377: is almost unaffected by the original single-particle width of the 

378: insulating narrow band. The detailed electronic properties of these 

379: phases will be discussed in Sections 5 and 6.    

380: 

381: \section{3 \ Numerical renormalization group approach for 

382:           effective one-band model}

383: 

384: In the limit of an insulating narrow band interacting with a metallic 

385: wide band, the two-band model can be simplified to an effective one-band 

386: model by eliminating the high energy states associated with the upper 

387: and lower Hubbard bands of the narrow band \cite{biermann05}.

388: Neglecting the one-electron hopping in the narrow band, i.e., assuming 

389: $W_1=0$, and fixing the occupations of these orbitals, the terms in 

390: Eq.~(1) involving interorbital Coulomb and pair-exchange interactions

391: disappear. The effective Hamiltonian then reduces to 

392: \begin{eqnarray}

393:   H  & = & \sum_{lm\sigma}t_{lm}c^{+}_{l\sigma}c_{m\sigma}

394:      + U\sum_l n_{l\uparrow} n_{l\downarrow}\nonumber\\ 

395:      &&\  - \sum_{l}[2 J S^{z}_{l}s^{z}_{l} +

396:       2J' (S^{+}_{l}s^{-}_{l}+S^{-}_{l}s^{+}_{l})]\, . \label{eq:fkl}

397: \end{eqnarray}

398: In this model, the low energy degrees of freedom of the narrow band 

399: are represented by local moments $\vec{S}_{l}$. These interact 

400: ferromagnetically with the local spin density $\vec{s}_{l}$ of the 

401: wide-band conduction electrons which are also subject to a local 

402: Coulomb repulsion $U$. The model corresponds to a double-exchange 

403: model with Coulomb interactions amongst the itinerant electrons or, 

404: equivalently, to the ferromagnetic Kondo lattice model with interactions 

405: in the band. The antiferromagnetic case, $J<0$, for $U=0$, has already 

406: been investigated in the context of heavy fermions  \cite{costi.02}.

407: We adapt these calculations to the case of interest here, namely,

408: ferromagnetic exchange, $J>0$, and $U>0$. The equivalence between the 

409: full two-band model and the effective model holds provided $U$ is large

410: enough so that a description of the low energy degrees of freedom of 

411: the narrow band in terms of local moments is possible and provided

412: that $J' \ll U/2$. The latter condition corresponds to eliminated

413: excited states of the full model being far from the first excited 

414: state of the effective model. Unless $J$ is chosen to be very small, 

415: we see that for realistic values of $J\approx U/4$,

416: the equivalence will hold well for Ising exchange coupling ($J'=0$) 

417: but less well for isotropic Hund's coupling ($J'=J$). In either case,

418: the effective model should be increasingly accurate in the limit 

419: $\omega, T \rightarrow 0$.

420: 

421: %\begin{figure}[t]

422: %\centering 

423: %\includegraphics[height=3.5cm,width=7.5cm]{low-energy-states.eps}

424: %\caption{\label{low-energy-states}

425: %Low lying two-electron states of the two-band and effective one-band

426: %models in the atomic limit. The $n=2,S_{z}=0$ states at $J+J'$ above the

427: %groundstate correspond to $\frac{1}{\sqrt{2}}(|\uparrow\rangle|\downarrow\rangle-

428: %|\downarrow\rangle|\uparrow\rangle)$ in both models, whereas the higher $n=2,S_{z}=0$

429: %state in the two-band model (at $J+U/2$ above the groundstate) corresponds to 

430: %$|0\rangle|\uparrow\downarrow\rangle$ (or $|\uparrow\downarrow\rangle|0\rangle$).

431: %For finite $W_{1}$, the latter would correspond to excitations to the upper and lower 

432: %``Hubbard bands'' of the narrow band, and are eliminated in the 

433: %effective model. The equivalence between the models holds when the 

434: %neglect of these states is a good approximation,

435: %i.e. for $J' \ll \frac{U}{2}$. 

436: %}

437: %\end{figure}

438: 

439: Within DMFT, we need to solve an effective quantum impurity model 

440: corresponding to a $S=1/2$ Kondo impurity coupled ferromagnetically 

441: with conduction electrons subject to a local Coulomb repulsion:

442: \begin{eqnarray}

443:   H  & = & \sum_{k,\sigma}\varepsilon_{k}c^+_{k\sigma}c_{k\sigma}

444:   + U n_{0\uparrow}n_{0\downarrow}\nonumber\\ 

445:   &&\   - 2J S_0^{z}s^{z}_{0} -

446:       2J' (S_0^{+}s^{-}_{0}+S_0^{-}s^{+}_{0})\, . \label{eq:quantum-imp}

447: \end{eqnarray}

448: We solve this model using the numerical renormalization 

449: group method \cite{wilson.74}, which allows

450: calculation of dynamical quantities on the real energy axis at both zero

451: and finite temperature \cite{costi.94,hofstetter.00}.

452: The impurity self-energy

453: $\Sigma(\omega)$ is calculated using the method

454: described in Ref.~\cite{bulla.98}. For comparison with ED, it is then 

455: evaluated on the imaginary axis $z=i\omega$ by analytic continuation

456: \begin{eqnarray}

457:  \Sigma(i\omega) = -\frac{1}{\pi}\int_{-\infty}^{+\infty} d\omega'\ 

458:     \frac{{\rm Im}\,\Sigma(\omega')}{i\omega-\omega'}\, .

459: \label{eq:matsubara-se}

460: \end{eqnarray}

461: The calculations use a logarithmic discretization of the conduction band 

462: $\varepsilon_{k_n}\rightarrow \pm D_2 \Lambda^{-\frac{n-1}{2}}$ with

463: $\Lambda=1.5$ and we retain of order 600 states per NRG iteration. 

464: Details of the calculation of spectra and other dynamical quantities 

465: can be found in Refs.~\cite{costi.94,bulla.01}.

466: 

467: The effective model specified in Eq.~(\ref{eq:fkl}) 

468: allows the non-Fermi-liquid physics of the intermediate phase of the 

469: two-band model to be understood qualitatively \cite{biermann05}. 

470: The case of Ising exchange ($J'=0$) describes interacting 

471: conduction electrons scattering from a disordered static configuration 

472: of local spins with $S_{l}^{z}=\pm 1/2$. The disorder potential

473: is proportional to the Ising exchange $J$. The conduction electrons

474: therefore have a finite lifetime at the Fermi level even at $T=0$, 

475: as is characteristic of a disordered metal. Switching on the

476: spin-flip part of the Hund's exchange gives the disordered spin 

477: configuration some dynamics at finite temperature, but as long as 

478: $J' < J$, we expect from the ferromagnetic Kondo model that the 

479: spin-flip part of the Hund's exchange will renormalize to zero 

480: at low temperature with the Ising part of the Hund's exchange remaining

481: finite. At $T=0$, the system will again be disordered with a disorder 

482: potential given by the finite Ising part of the Hund's exchange. 

483: This results in a finite lifetime for the conduction electrons at $T=0$. 

484: 

485: The situation changes for isotropic Hund's exchange ($J'=J$). 

486: In this case, the ferromagnetic exchange in the effective impurity 

487: model (\ref{eq:quantum-imp}) is known to be marginally irrelevant 

488: \cite{cragg.79,koller.05}. 

489: Thus, both Ising and spin-flip parts of the Hund's exchange renormalize

490: to zero and there will be no disorder scattering at $T=0$, giving rise

491: to an infinite lifetime for the conduction electrons at $T=0$. The vanishing

492: of the self-energy at $\omega=T=0$ also implies that the narrow-band

493: spectral function satisfies a pinning condition at $\omega = T =0$.

494: In Sections 5 and 6, these qualitative considerations will be seen 

495: to be in accord with the numerical results from ED for the full 

496: two-band model and NRG results for the effective one-band model.

497: 

498: \section{4 \ Metallic phase}

499: 

500: We begin our discussion of the multi-band ED results with the region 

501: in which both subbands are metallic, just below the Mott transition 

502: associated with the narrow band. To illustrate the applicability of 

503: the ED approach this region is very useful since it involves 

504: important redistribution of spectral weight between low and high 

505: frequencies that must be captured properly in order to describe

506: the Mott transition. 

507: 

508: 

509: \begin{figure}[t!]%3

510:   \begin{center}

511: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{M8.ps}

512: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{M6.ps}

513:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig4a.ps}

514:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig4b.ps}

515:   \end{center}

516:   \vskip-2mm

517: \caption{\label{M8M6}

518: Imaginary part of subband self-energies in metallic region 

519: for $J=J'=U/4$ at $T=2.5$~meV. 

520: Solid (red) curves: narrow band, dashed (blue) curves: wide band. 

521: From top: $U=1.8,\ 1.9,\,2.0,\,2.1$~eV.   (a) $n_s=8$; (b) $n_s=6$.

522: }\end{figure}

523: 

524: Fig.~\ref{M8M6} shows a comparison of the subband self-energies for 

525: $n_s=8$ and $n_s=6$ in the region close to the lower critical Coulomb 

526: energy for $T=2.5$~meV. For $n_s=8$, the $\Sigma_i(i\omega_n)$ vary 

527: roughly linearly at low frequencies, indicating that both bands are 

528: metallic. Of course, the narrow band is more strongly correlated than 

529: the wide band. The spacing between the lowest excited state and the 

530: ground state energy in these calculations is typically $1-2$~meV, 

531: i.e., evaluation of the self-energies in this low temperature range 

532: is indeed meaningful.

533: 

534: The results for $n_s=6$ are similar, except that the self-energy of the 

535: narrow band at $U=2.1$~eV is inversely proportional to $i\omega_n$,

536: indicating that the Mott transition in this case is located between

537: $U=2.0$~eV and $U=2.1$~eV. In addition, at the lowest Matsubara 

538: frequencies the self-energy of the wide band at 2.1~eV exhibits  

539: deviations from linear $i\omega_n$ variation. These deviations 

540: are related to finite-size effects and will be analyzed in more 

541: detail in the next section when we discuss the intermediate 

542: orbital-selective phase. Just below the transition for $n_s=6$, at  

543: $U=2$~eV, very small deviations from approximately linear $i\omega_n$ 

544: variation can also be seen in both $\Sigma_i(i\omega_n)$. 

545: 

546: It is remarkable that the ED results for $n_s=8$ capture the correlation 

547: effects in the common metallic phase of both subbands very well until 

548: close to $U_{c1}$ and down to rather low temperatures. This result 

549: is not at all obvious since so close to the Mott transition a large 

550: fraction of the spectral weight of the narrow band is transfered from 

551: the Fermi level to the Hubbard bands. Even the results for $n_s=6$ are  

552: qualitatively correct. The main effect of the cluster size $n_s=6$ is 

553: the underestimate of $U_{c1}$ by about $0.1$~eV. According to the 

554: single-band results \cite{prl05} inclusion of the zero energy bath levels 

555: in the $n_s=8$ cluster should provide the most important part of the 

556: shift towards the correct $U_{c1}$. In addition these extra bath levels 

557: yield an excellent representation of the metallic properties of both 

558: subbands right up to the critical Coulomb energy. 

559: 

560: Of course, slight finite-size effects should be manifest also in the 

561: low-temperature ED results for $n_s=8$. The precise form of  

562: $\Sigma_i(i\omega_n)$ at very low frequencies and temperatures,

563: in particular, the range of true Fermi-liquid behavior in the immediate 

564: vicinity of $U_{c1}$, can only be investigated by more accurate methods, 

565: such as a full two-band extension of the NRG approach which is applicable 

566: at zero and finite $T$ \cite{costi}. 

567: 

568: We emphasize that this region of the $T/U$ phase diagram for isotropic 

569: Hund's exchange is not yet accessible using QMC calculations. 

570: It would therefore be of great interest to extend the present results 

571: to $n_s=9$ and $n_s=12$ in order to explore the strongly correlated 

572: metallic phase of realistic three-band materials which have so far been 

573: investigated only for Ising-like exchange interactions. A more detailed 

574: comparison of the role of Hund vs. Ising exchange treatments in the 

575: metallic phase of the present two-band model will be given elsewhere

576: \cite{metallic}.     

577: 

578: \section{5 \ Intermediate phase: isotropic Hund's exchange}

579: 

580: As shown first by Koga {\it et al.}~\cite{koga1} within ED/DMFT 

581: calculations at $T=0$, the two-band Hubbard model with full Hund's exchange 

582: interaction exhibits successive, orbital-selective Mott transitions. 

583: In Ref.~\cite{prl05} we proved that analogous ED calculations at 

584: finite temperature are consistent with these findings, revealing 

585: sequential first-order transitions associated with the two subbands. 

586: In this section we discuss in more detail the low temperature properties 

587: of the intermediate phase between the fully metallic and insulating phases.

588: 

589: \begin{figure}[t!]%4

590:   \begin{center}

591: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{OSMTa.ps}

592: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{OSMTb.ps}

593:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig5a.ps}

594:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig5b.ps}

595:   \end{center}

596:   \vskip-2mm

597: \caption{\label{OSMT}

598: Imaginary part of subband 

599: self-energies in intermediate phase, for $U=2.4$~eV and $J=J'=U/4$.  

600: $T=2.5,\ 5,\ 10,\ 20,\ 30$~meV (from bottom). 

601: (a) $n_s=8$; (b) $n_s=6$; 

602: }\end{figure}

603: 

604: Fig.~\ref{OSMT} shows the subband self-energies for a Coulomb energy 

605: $U=2.4$~eV, i.e., between the lower and upper Mott transitions. 

606: The results 

607: for $n_s=8$ demonstrate that the narrow band is insulating, i.e., its

608: self-energy is inversely proportional to $i\omega_n$ at low frequencies. 

609: $\Sigma_1(i\omega_n)$ is also nearly independent of temperature in the 

610: range $T=2.5\ldots 30$~meV, suggesting that the Mott gap is considerably 

611: larger. The self-energy of the wide band, however, extrapolates to zero 

612: at low frequencies, indicating metallic behavior. These results 

613: suggest that the pinning condition $N_2(0)=A_2(0)$ is satisfied, i.e., 

614: the spectral density at $E_F$ of the interacting system coincides 

615: with that of the non-interacting system. Because of the rapid increase 

616: of Im\,$\Sigma_2(i\omega_n)$ at small $\omega_n$, the spectral 

617: distribution of the wide band in the intermediate phase consists of 

618: a very narrow peak at $E_F$ and large upper and lower Hubbard bands.

619: Nevertheless, as will be discussed further below, the comparison 

620: with the NRG results indicates that this band does not satisfy 

621: true Fermi-liquid behavior.

622: 

623: The results for $n_s=6$ shown in Fig.~\ref{OSMT}(b) agree qualitatively 

624: with those for $n_s=8$. The self-energies of the narrow band are nearly 

625: identical, with only a slightly larger dependence on temperature in 

626: the case $n_s=6$. The main difference is that $\Sigma_2(i\omega_n)$ 

627: at low $T$ and small $i\omega_n$ exhibits larger deviations as a result 

628: of finite-size effects. This is plausible since, as discussed above, 

629: the wide band is metallic, but highly correlated. The pronounced 

630: three-peak structure of its spectral function is therefore not 

631: represented accurately by only two non-zero bath levels per 

632: impurity orbital in the ED calculation for $n_s=6$. The extra bath

633: level with zero energy included for $n_s=8$ therefore plays a crucial 

634: role for a proper description of the metallic behavior of the wide

635: band for full Hund's coupling.   

636: 

637: \begin{figure}[t!]%5

638:   \begin{center}

639: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{GnuGJ.ps}

640:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig6.ps}

641:   \end{center}

642:   \vskip-2mm

643: \caption{\label{GreenJ}

644: Imaginary part of subband 

645: Green's functions in intermediate phase, for $U=2.4$~eV, $J=J'=U/4$,

646: $n_s=8$. $T=2.5,\ 5,\ 10,\ 20,\ 30$~meV. 

647: The dot indicates the pinning condition for the wide band.

648: }\end{figure}

649: 

650: Fig.~\ref{GreenJ} shows the subband Greens functions $G_i(i\omega_n)$ 

651: in the intermediate phase at various temperatures. Im\,$G_1$ is linear in 

652: $\omega_n$ at low frequencies, as expected for the insulating subband. 

653: Although the wide band exhibits clear signs of finite-size effects at 

654: small $\omega_n$ it is nevertheless seen to approximately satisfy the 

655: pinning condition Im\,$G_2(i\omega_n)\rightarrow -\pi N_2(0) = -1$ in 

656: the limit $\omega_n\rightarrow0$.   

657: 

658: For computational reasons the role of finite-size effects in the 

659: present finite-temperature ED approach based on full matrix 

660: diagonalization can at present not be checked by going beyond the 

661: cluster size $n_s=8$. To analyze the low-frequency properties of 

662: the wide band in the intermediate phase more closely, we have 

663: carried out NRG calculations for the effective one-band model  

664: discussed in Section 3. Although this model is more appropriate

665: for the anisotropic exchange treatment discussed in the next section,

666: we use it here (with some caution) since it is so far the only scheme 

667: capable of providing a guideline for the two-band Hubbard model 

668: at low finite temperatures.

669: 

670: \begin{figure}[t!]%6

671:   \begin{center}

672: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{NRG3.ps}

673: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{NRG4.ps}

674:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig7a.ps}

675:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig7b.ps}

676:   \end{center}

677:   \vskip-2mm

678: \caption{\label{NRG34}

679: Imaginary part of self-energy $\Sigma_2$ of wide band for $U=2.4$~eV 

680: and $J=J'=U/4$. (a) Solid (red) and dashed (blue) curves: ED results 

681: for $n_s=8$ at  $T=2.5,\,5,\,10,\,20,\,30$~meV (from bottom); 

682: green curves: NRG results for $T=3,\ 10,\ 15,\ 34$~meV (from top). 

683: (b) Same quantities on expanded low-frequency scale. Black curve (+):

684: self-energy $\Sigma_1$  of narrow band at $2.0$~eV. 

685: }\end{figure}

686: 

687: In Fig.~\ref{NRG34}(a) the ED results for Im\,$\Sigma_2(i\omega_n)$ are 

688: compared with the corresponding self-energy derived within the NRG. 

689: The overall frequency variation of $\Sigma_2$ is seen to be remarkably 

690: similar for both methods, implying similar spectral distributions.

691: The most noticeable difference is the larger dependence on temperature 

692: in the case of the ED results.

693: According to Fig.~\ref{OSMT}, the increase in cluster size from $n_s=6$ 

694: to $n_s=8$ diminishes the variation of $\Sigma_2$ with temperature and 

695: makes the minimum near $\omega_n\approx0.9$ less deep. This trend 

696: indicates that larger cluster sizes would bring the ED results

697: into better agreement with the NRG data. We point out, however, 

698: that part of the difference should be caused by the approximate 

699: nature of the NRG model in the case $J'=J$. 

700: As shown in the following section, for $J'=0$ the ED and NRG results 

701: for $\Sigma_2$ at $T=20\ldots30$~meV are in perfect agreement.  

702: 

703: The low-frequency region of $\Sigma_2(i\omega_n)$ is plotted in 

704: greater detail in Fig.~\ref{NRG34}(b). In the ED calculation, the 

705: spacing between the lowest excited level and the ground state energy 

706: is typically $10$~meV. Thus, temperatures below this range will give

707: rise to increasing finite-size effects, precluding any meaningful 

708: analysis at low frequencies. Despite this limitation, the ED data 

709: follow the frequency dependence of the NRG results rather well. 

710: 

711: In order to provide some insight into the variation of $\Sigma_2$ 

712: at finite $\omega_n$, we compare it in Fig.~\ref{NRG34}(b) also 

713: with the self-energy of the narrow band  $\Sigma_1$

714: at $U=2.0$~eV, i.e., 

715: just below the Mott transition at $U_{c1}\approx2.1$~eV. 

716: According to Fig.~1(a), we find: 

717: $Z_1(U=2.0~{\rm eV})\approx Z_2(U=2.4~{\rm eV})\approx 0.11$. 

718: Nevertheless, whereas $\Sigma_1$ at $U=2.0$~eV exhibits the typical 

719: behavior expected for a strongly correlated Fermi-liquid phase, 

720: $\Sigma_2$ evidently has much larger nonlinear corrections. A similar 

721: systematic difference is found between $\Sigma_2$ in the 

722: intermediate phase and the behavior the self-energy in a 

723: single-band model for $U$ just below the Mott transition.

724: 

725: The NRG results shown in Fig.~\ref{NRG34} indicate that   

726: $\Sigma_2(i\omega_n)\rightarrow0$ in the limit $\omega_n\rightarrow0$ 

727: and for low $T$. Thus, particles at $E_F$ have infinite lifetime. 

728: In addition, analogous NRG calculations at $T=0$ yield \cite{costi}  

729: %Im\,$\Sigma_2(\omega)\sim -1/({\rm ln}\omega)^2$, in contrast to 

730: non-Fermi-liquid behavior, in agreement with Ref.~\cite{biermann05}.

731: Thus, at small finite $i\omega_n$ Im\,$\Sigma_2$ does not vary 

732: linearly. The qualitative agreement between the finite-$T$ NRG and 

733: ED results seen in Fig.~\ref{NRG34} shows that the 

734: ED data are consistent with non-Fermi-liquid behavior.  

735: 

736: \begin{figure}[t!]%7

737:   \begin{center}

738: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{NRG1.ps}

739:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig8.ps}

740:   \end{center}

741:   \vskip-2mm

742: \caption{\label{NRG12}

743: Spectral distribution of wide band in intermediate phase,

744: calculated within NRG approach for $U=2.4$~eV, $J=J'=U/4$. 

745: Green curve: $T=3$~meV; red curve: $T=30$~meV;  

746: dashed blue curve: bare density of states.

747: }\end{figure}

748:     

749: Fig.~\ref{NRG12} shows the spectral function of the wide band in the

750: intermediate phase, calculated within the NRG. Spectra of this kind 

751: were also obtained by Arita and Held \cite{arita} for $U$ near $U_{c1}$ 

752: within QMC/DMFT calculations for the same two-band model for $T=0$ and

753: $J'=J=U/4$.

754: As mentioned above, the distribution exhibits a very narrow peak which 

755: satisfies the pinning condition at $E_F$ for temperatures up to about 

756: 10~meV. 

757: % This finding is consistent with the ED/DMFT results reported in 

758: % Ref.~\cite{prl05}, which revealed 

759: % $T_{c1}\approx 20$~meV and $T_{c2}\approx 15$~meV.    

760: Because of the sharp central peak of $A_2(\omega)$, the real and 

761: imaginary parts of the self-energy of the wide band $\Sigma_2(\omega)$ 

762: exhibit structure on a similar scale. Nevertheless, the variation 

763: of Im\,$\Sigma_2(i\omega_n)$ obtained in the NRG is very smooth. 

764: Thus, the weak shoulder seen in the ED results for $\Sigma_2$ near 

765: $\omega_n\approx0.06$ in Figs.~\ref{W1W2}(a), \ref{OSMT}(a) and 

766: \ref{NRG34}  must be attributed to finite-size effects.   

767: 

768: The ED and NRG results discussed above demonstrate that, in the 

769: presence of full Hund's exchange, the wide band above the main 

770: Mott transition at $U_{c1}$ remains metallic, without satisfying

771: Fermi-liquid criteria. In the following section it will 

772: be shown that the absence of spin-flip and pair-exchange terms

773: enhances this trend towards non-Fermi-liquid behavior, so that 

774: even particles at $E_F$ acquire a finite lifetime. Thus, in both

775: cases, the intermediate phase is bad-metallic.  

776: 

777: We point out that the differences between the self-energies for 

778: $n_s=8$ and $n_s=6$ appear mainly at rather low temperatures, below

779: about $T=20$~meV. Since QMC calculations also become difficult to 

780: converge in this range, three-band ED calculations with 

781: two bath levels per impurity orbital, i.e., cluster size $n_s=9$, 

782: should be competitive with analogous QMC calculations -- with the 

783: important additional benefit, that the ED approach can handle 

784: full Hund's coupling.          

785: 

786: \section{6 \ Intermediate phase: Ising exchange}

787: 

788: As shown in Ref.~\cite{prb70}, in the absence of spin-flip and 

789: pair-exchange contributions, the first-order Mott transition at 

790: $U_{c1}\approx 2.1$~eV affects the different subbands of the 

791: Hubbard model in fundamentally different ways: The narrow band 

792: undergoes a complete metal insulator transition, but the wide

793: band changes from a normal metal to a bad metal in the sense

794: that its self-energy exhibits progressive deviations from 

795: Fermi-liquid linear $\omega_n$ variation at low frequencies.

796: Thus, Im\,$\Sigma_2(i\omega_n)\rightarrow c(U)$ for     

797: $\omega_n\rightarrow 0$, where  $c(U)\approx 0\rightarrow-\infty$ 

798: for $U = U_{c1}\rightarrow U_{c2}$. Spectral functions obtained via 

799: the maximum entropy method showed that this breakdown of Fermi-liquid

800: behavior in the wide band in the intermediate phase leads to a narrow 

801: pseudogap near $E_F$ and to a violation of the pinning condition, 

802: i.e., $A_2(0)<N_2(0)$. The spectral function of the wide band then 

803: acquires a characteristic four-peak structure, with two maxima flanking 

804: the pseudogap, in addition to the Hubbard bands at higher energies.  

805: With increasing $U$, the pseudogap becomes deeper and wider, until 

806: it turns into a true insulating gap at $U_{c2}\approx 2.7$~eV.

807: As also shown in Ref.~\cite{prb70}, the upper bad-metal to insulator

808: transition is not a first-order transition, in contrast to the main

809: transition at $U_{c1}$.   

810: 

811: 

812: In this section we discuss the temperature dependence of the subband

813: self-energies in the intermediate phase and the finite-size effects 

814: associated with the limited number of bath levels included in our 

815: ED/DMFT approach. 

816: 

817: 

818: \begin{figure}[t!]%8

819:   \begin{center}

820: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{NRG3J0.ps}

821: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{NRG4J0.ps}

822: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{NRG5J0.ps}

823:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig9a.ps}

824:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig9b.ps}

825:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig9c.ps}

826:   \end{center}

827:   \vskip-2mm

828: \caption{\label{NRG345J0}

829: Imaginary part of subband self-energies $\Sigma_i(i\omega_n)$ 

830: in intermediate phase for $n_s=8$, $U=2.4$~eV, $J=U/4$, $J'=0$;

831: Solid (red) and dashed (blue) curves:  ED results for    

832: $T=2.5,\ 5,\ 10,\ 20,\ 30$~meV (from bottom). 

833: (a)  $\omega_n\Sigma_i(i\omega_n)$;

834: (b)  $\Sigma_2(i\omega_n)$; green curves: NRG results as in (c).

835: (c) Low-frequency behavior of self-energy of wide band: 

836: red and blue curves: ED results as in (b); green curves: NRG results 

837: for $T=2,\,4.4,\,10,\,20,\,30$~meV (from bottom). 

838: For clarity, successive curves in (c) are displaced vertically by 1.

839: }\end{figure}

840: 

841: Fig.~\ref{NRG345J0}

842: shows the subband self-energies for $n_s=8$ and $U=2.4$~eV,

843: assuming again $J=U/4$, but $J'=0$. The narrow band is fully insulating, 

844: so that $\omega_n\,{\rm Im}\,\Sigma_1(i\omega_n)\rightarrow {\rm const.}$ 

845: nearly independently of temperature, similarly to the case $J'=J$ plotted 

846: in Fig.~\ref{OSMT}. The self-energy of the wide band, however, differs 

847: qualitatively from the behavior found for isotropic Hund's coupling. 

848: Rather than vanishing in the limit $\omega_n\rightarrow0$, 

849: Im\,$\Sigma_2(i\omega_n)$ now approaches a sharp minimum, which gets 

850: more pronounced towards low temperatures. Clearly, since    

851: $\omega_n\,{\rm Im}\,\Sigma_2(i\omega_n)\rightarrow0$ at small 

852: $\omega_n$, this band is not yet insulating.  

853: Thus, instead of satisfying the pinning condition at $E_F$, the 

854: spectral function of the wide band exhibits a dip or pseudogap at

855: the Fermi level (see below).

856: 

857: Fig.~\ref{NRG345J0}(b) also shows the NRG results for 

858: $\Sigma_2(i\omega_n)$. 

859: They are seen to be in very good agreement with the ED data, 

860: except at low frequencies for $T\le10$~meV. 

861: The fact that for $T=20\ldots30$~meV there is now much better

862: coincidence between the ED and NRG results than in Fig.~\ref{NRG34} 

863: for $J'=J$ indicates that, as discussed in Section 3, the 

864: effective one-band model employed 

865: for the NRG is more appropriate in the case $J'=0$. The low-frequency 

866: behavior of Im\,$\Sigma_2(i\omega_n)$ is shown in more detail 

867: in Fig.~\ref{NRG345J0}(c). Both the ED and NRG results reveal that the 

868: sharp minimum of $\Sigma_2$ at small $\omega_n$ gets progressively 

869: deeper at low temperatures, suggesting that the pseudogap in the spectral

870: function becomes accordingly deeper. 

871: 

872: 

873: \begin{figure}[t!]%9

874:   \begin{center}

875: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{NFL6a.ps}

876: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{NFL6b.ps}

877:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig10a.ps}

878:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig10b.ps}

879:   \end{center}

880:   \vskip-2mm

881: \caption{\label{NFL6}

882: Imaginary part of subband self-energies $\Sigma_i(i\omega_n)$ 

883: in intermediate phase for $n_s=6$, $U=2.2$~eV and $J=U/4$, $J'=0$.   

884: $T=2.5,\ 5,\ 10,\ 20,\ 30$~meV (from bottom).  

885: }\end{figure}

886: 

887: Analogous results for $n_s=6$ are given in Fig.~\ref{NFL6}. Since 

888: according to Ref.~\cite{prl05} in this case the wide band becomes 

889: insulating near $U\approx2.4$~eV, we choose $U=2.2$~eV to illustrate 

890: the non-Fermi-liquid behavior of this band in the intermediate phase. 

891: As for $n_s=8$, the narrow band is insulating, i.e., 

892: $\omega_n\,{\rm Im}\,\Sigma_1(i\omega_n)\rightarrow{\rm const.}$ 

893: at low frequencies. The frequency variation of $\Sigma_2(i\omega_n)$ 

894: is seen to be more strongly affected by finite-size effects. 

895: Nevertheless, as in the case $n_s=8$, a sharp minimum is found 

896: for $\omega_n\rightarrow0$.

897: 

898: \begin{figure}[t!]%10

899:   \begin{center}

900: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{GnuGJ0.ps}

901:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig11.ps}

902:   \end{center}

903:   \vskip-2mm

904: \caption{\label{GreenJ0}

905: Imaginary part of subband 

906: Green's functions in intermediate phase, for $U=2.4$~eV, $J=U/4$, 

907: $J'=0$, $n_s=8$. $T=2.5,\ 5,\ 10,\ 20,\ 30$~meV (from top). 

908: }\end{figure}

909: 

910: Fig.~\ref{GreenJ0} shows the subband Green's functions $G_i(i\omega_n)$ 

911: for anisotropic Hund's exchange at various temperatures. 

912: As for the case $J'=J$ shown in Fig.~\ref{GreenJ}, Im\,$G_1$ is linear 

913: in $\omega_n$ at low frequencies, since the narrow band is insulating.

914: On the other hand, because of the more severe breakdown of Fermi-liquid 

915: behavior in the wide band for $J'=0$, with Im\,$\Sigma_2(i\omega_n)\ne 0$ 

916: in the low-frequency limit, the Green's function no longer satisfies the 

917: pinning condition. Thus, Im\,$G_2(i\omega_n)\rightarrow -c(T)$ with 

918: $c(T)<\pi N_2(0) = 1$, and $c(T)\rightarrow0$ for decreasing temperature.

919: This behavior implies that the spectral function of the wide band exhibits

920: a pseudogap at $E_F$ which becomes progressively deeper towards low $T$.  

921: 

922: This picture is fully confirmed by the NRG results shown in 

923: Fig.~\ref{NRG12J0}. In contrast to the narrow peak at $E_F$ in the

924: three-peak structure seen in Fig.~\ref{NRG12} for isotropic Hund's 

925: coupling, the spectra for $J'=0$ show a pseudogap which becomes deeper 

926: as $T$ decreases. As a result, the spectral distribution now exhibits a 

927: characteristic four-peak structure, with two maxima limiting the pseudogap 

928: and two shoulders associated with the Hubbard peaks. Spectra of this kind 

929: were first observed in the QMC/DMFT results at $T=31$~meV reported in 

930: Ref.~\cite{prb70} (see also next section). The comparison with the

931: spectra for $J'=J$ reveals nearly identical excitations for 

932: $\vert\omega\vert\gtrsim J$. Thus, as expected, the different treatments 

933: of exchange interactions affect primarily the low-frequency excitations 

934: in the metallic wide band.    

935: 

936: 

937: 

938: \begin{figure}[t!]%11

939:   \begin{center}

940: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{NRG1J0.ps}

941:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig12.ps}

942:   \end{center}

943:   \vskip-2mm

944: \caption{\label{NRG12J0}

945: Spectral distribution of wide band in intermediate phase, calculated 

946: within NRG approach for $U=2.4$~eV, $J=U/4$, $J'=0$.

947: Green curve: $T=3$~meV; red curve: $T=34$~meV;  

948: dashed blue curve: bare density of states. 

949: Black dotted curve: spectrum for $J'=J$, $T=3$~meV from 

950: Fig.~\ref{NRG12}. 

951: }\end{figure}

952: 

953: 

954: 

955: \section{7 \ Comparison with previous QMC/DMFT results}

956: 

957: In this and the following sections we compare the ED and NRG results 

958: for $J'=0$ with available QMC/DMFT data in order to illustrate 

959: the consistency between these impurity treatments and to explore 

960: further the role of finite-size effects in the ED approach.

961: 

962: \begin{figure}[t!]%12

963:   \begin{center}

964: %  \includegraphics[width=5.5cm,height=8cm,angle=-90]{QMC.ED.ps}

965: %  \includegraphics[width=5.5cm,height=8cm,angle=-90]{QMCnew.ps}

966: %  \includegraphics[width=5.5cm,height=8cm,angle=-90]{QMC.NRG.ps}

967:   \includegraphics[width=5.9cm,height=8cm,angle=-90]{fig13a.ps}

968:   \includegraphics[width=5.9cm,height=8cm,angle=-90]{fig13b.ps}

969:   \includegraphics[width=5.9cm,height=8cm,angle=-90]{fig13c.ps}

970:   \end{center}

971:   \vskip-2mm

972: \caption{\label{QMCprb}

973: Comparison of subband self-energies $\Sigma_i(i\omega_n)$ in intermediate 

974: non-Fermi-liquid phase for $U=2.1,\ 2.4,\ 2.7$~eV, $J=U/4$, $J'=0$,

975: $T=31$~meV, calculated via three different impurity solvers:

976: (a) ED results for $n_s=8$: red solid curves: narrow band; 

977: blue dashed curves: wide band; 

978: (b) QMC results from Fig.~10 of Ref.~\cite{prb70};

979: red solid curves: narrow band; blue dashed curves: wide band;  

980: (c) comparison of ED (x) and NRG self-energies of wide band.

981: }\end{figure}

982: 

983: As shown in Ref.~\cite{prb70}, in the absence of spin-flip and

984: pair-exchange terms, purely metallic and insulating phases exist

985: for $U<U_{c1}\approx2.1$~eV and  $U>U_{c2}\approx2.7$~eV, 

986: respectively. Fig.~\ref{QMCprb} shows a comparison of subband 

987: self-energies at three representative Coulomb energies within 

988: the intermediate `bad-metal' non-Fermi-liquid region. The ED 

989: results for $n_s=8$ are seen to be in excellent agreement with 

990: the NRG self-energies, confirming the validity of the effective 

991: one-band model in the intermediate phase for $J'=0$. Moreover, 

992: both schemes agree very well with the QMC/DMFT self-energies 

993: reported in Ref.~\cite{prb70} for $T=31$~meV. Minor differences 

994: between the QMC and ED/NRG results are found only in the steepest 

995: parts of $\Sigma_i$ close to the first Matsubara frequency. 

996: (In contrast to the QMC calculations which are carried out at 

997: discrete Matsubara frequencies, the NRG self-energy is available 

998: continuously as a function of frequency. The ED self-energies 

999: could in principle also be obtained at arbitrary $i\omega$ but 

1000: were calculated here at $i\omega_n$. This explains the slightly 

1001: different form of the curves plotted in panel (c). At 

1002: $i\omega_n$ the ED and NRG data nearly coincide.)  

1003: 

1004: \begin{figure}[t!]%13

1005:   \begin{center}

1006: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{QMCa.ps}

1007: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{QMCb.ps}

1008:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig14a.ps}

1009:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig14b.ps}

1010:   \end{center}

1011:   \vskip-2mm

1012: \caption{\label{QMCab}

1013: Subband self-energies $\Sigma_i(i\omega_n)$ in intermediate 

1014: non-Fermi-liquid phase for $T=31$~meV, $U=2.1,\ 2.4,\ 2.7$~eV,

1015: $J=U/4$, $J'=0$. Solid (red) curves: $n_s=8$; dashed (blue) curves:   

1016: $n_s=6$. 

1017: (a) narrow subband: $\omega_n\Sigma_1(i\omega_n)$; 

1018: (b) wide   subband: $\Sigma_2(i\omega_n)$.  

1019: }\end{figure}

1020: 

1021: The overall 

1022: trend obtained previously within the QMC/DMFT is fully confirmed 

1023: by the new ED and NRG calculations: The narrow band is insulating 

1024: throughout this range of Coulomb energies, whereas the wide band 

1025: changes gradually from metallic to insulating via progressive 

1026: non-Fermi-liquid behavior. Thus, for small $\omega_n$,   

1027: Im\,$\Sigma_2(i\omega_n)\rightarrow c(U)$, 

1028: where $c(U)\rightarrow-\infty$ at $U\approx2.7$~eV.   

1029: 

1030: Evidently, despite their intrinsic numerical uncertainties,

1031: the three complementary impurity solvers: ED, NRG and QMC provide 

1032: perfectly consistent descriptions of the electronic properties of 

1033: the orbital-selective phase for Ising-like exchange. 

1034: 

1035: Since the temperature $T=31$~meV in these results is relatively 

1036: high, finite-size effects tend to be less pronounced than in the cases 

1037: discussed in the previous sections at lower $T$. This is illustrated 

1038: in Fig.~\ref{QMCab} where we compare the above ED results for $n_s=8$ 

1039: with those for $n_s=6$. The self-energies of the narrow band are 

1040: nearly identical and satisfy 

1041: $\omega_n\Sigma_1(i\omega_n)\rightarrow {\rm const.}$ at low 

1042: frequencies. This is plausible since the additional zero-energy 

1043: bath level for $n_s=8$ carries very little weight in the insulating 

1044: state. Because of breakdown of Fermi-liquid behavior, the self-energy 

1045: of the wide band approaches a finite value in the limit    

1046: $\omega_n\rightarrow0$. The differences between the results for 

1047: $n_s=6$ and $n_s=8$ are very small as long as this band is either 

1048: nearly metallic, like at $U=2.1$~eV, or nearly insulating, like at 

1049: $U=2.7$~eV. Slightly larger differences are found only in the middle 

1050: of the bad-metallic region near $U=2.4$~eV. As discussed in the 

1051: preceding section, the spectral function then has a more complicated

1052: four-peak structure as a result of the narrow pseudogap at $E_F$,

1053: which cannot be adequately represented using only two bath levels. 

1054: This region could possibly be even more accurately described by using 

1055: four bath levels for the wide band and two levels for the insulating 

1056: narrow band, maintaining the total $n_s=8$.      

1057: 

1058:  \begin{figure}[t!]%14

1059:   \begin{center}

1060: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{GnuQMCNRG.ps}

1061:   \includegraphics[width=6.1cm,height=8cm,angle=-90]{fig15.ps}

1062:   \end{center}

1063:   \vskip-2mm

1064: \caption{\label{QMCNRG}

1065: Spectral distributions of subbands in intermediate phase, calculated 

1066: within QMC/DMFT for $U=2.4$~eV, $J=U/4$, $J'=0$, $T=31$~meV.

1067: Red curve: metallic wide band; magenta curve: insulating narrow band;

1068: blue and black curves: bare densities of states. From Fig.~11(b) 

1069: of Ref.~\cite{prb70}.

1070: }\end{figure}

1071:          

1072: We close this section by showing in Fig.~\ref{QMCNRG} the spectral 

1073: distributions of both subbands, as calculated within the QMC/DMFT and 

1074: the maximum entropy method \cite{prb70}. The spectrum of the bad-metallic 

1075: wide band can be compared with the corresponding NRG spectra plotted in 

1076: Fig.~\ref{NRG12J0}. Both distributions exhibit the marked four-peak

1077: structure induced by the pseudogap and the Hubbard bands. The 

1078: low-frequency region is seen to be in excellent agreement. Both methods

1079: coincide in that, at $T\approx30$~meV, the interacting density of 

1080: states at $E_F$ of the wide band is only about one third of the 

1081: noninteracting one. The slightly larger differences at higher energies,

1082: in particular, the position and relative weight of the Hubbard bands, 

1083: may be caused by the choice of fitting parameters in the maximum entropy 

1084: procedure, and by the less accurate nature of the NRG approach at high 

1085: energies. 

1086: %Since the ED and QMC self-energies shown in Fig.~\ref{QMCprb}

1087: %are in excellent agreement, the corresponding ED results for 

1088: %$G_i(i\omega_n)$ could also be used, via $G_i(\tau)$ and the maximum

1089: %entropy method, to derive spectral distributions similar to those 

1090: %shown in Fig.~\ref{QMCNRG}.   

1091: 

1092: 

1093: \section{8 \ Comparison with QMC/DMFT results for $\bf W_2=10W_1$}

1094: 

1095: The breakdown of Fermi-liquid behavior in the orbital-selective phase

1096: with Ising-like exchange was recently also studied by Biermann 

1097: {\it et al.}~\cite{biermann05} for the same two-band Hubbard model 

1098: as above, except for $W_1=0.2$~eV and $W_2=2$~eV. 

1099: The subbands were also assumed to be non-hybridizing and half-filled. 

1100: QMC/DMFT calculations were carried out for $T=1/120\ldots1/40$~eV, 

1101: $U=0.8$~eV, $J=0$ and $J=U/4$, with $J'=0$, i.e., in the absence of 

1102: spin-flip and pair exchange terms. For $J=0$ as well as $J=U/4$, the 

1103: self-energy of the narrow band was found to diverge at low frequencies, 

1104: demonstrating that this band is insulating since $W_1\ll U$. The wide 

1105: band is in the intermediate orbital-selective region between the purely 

1106: metallic and insulating phases since $U\ll W_2$. The self-energy of this 

1107: band showed a striking variation with the magnitude of $J$: 

1108: For $J=0$, $\Sigma_2(i\omega_n)\sim i\omega_n$ at low frequencies, as 

1109: expected for Fermi-liquid behavior. For  $J=U/4$, however, 

1110: Im\,$\Sigma_2(i\omega_n)\rightarrow{\rm const.}\approx -0.09$ nearly 

1111: independently of temperature for $T=1/120\ldots1/40$~eV, indicating 

1112: breakdown of Fermi-liquid properties. Accordingly, the pinning condition 

1113: $N_2(0)=A_2(0)$ was found to be satisfied for $J=0$, but not for $J=U/4$. 

1114: These results are fully consistent with the trend discussed in 

1115: Ref.~\cite{prb70} for $W_1=2$~eV and $W_2=4$~eV.   

1116: 

1117: To check the accuracy of our finite-$T$ ED approach we have applied 

1118: it to the case investigated in Ref.~\cite{biermann05}. To provide a 

1119: picture of the Mott transitions in this two-band system we show first 

1120: in Fig.~\ref{Paris.zu} the variation of $Z_i(U)$ for three different 

1121: treatments of Hund's exchange. In all cases the narrow band 

1122: becomes insulating at about $U_{c1}=0.3\ldots0.4$~eV. 

1123: However, the range and nature of the orbital-selective phase of the 

1124: wide band, $U_{c1} < U < U_{c2}$, depend sensitively on the magnitudes 

1125: of $J/U$ and $J'/U$. For $J=J'=0$ the upper transition occurs for 

1126: $U_{c2}$ slightly larger than $W_2$, with a pronounced hysteresis 

1127: loop indicative of first-order behavior. A very weak hysteresis 

1128: is found also for $J=J'=U/4$, with $U_{c2}\approx 1.5$~eV, similar 

1129: to the one in Fig.~\ref{ZU} near $U\approx3$~eV. Finally, for $J=U/4$ and 

1130: $J'=0$, $U_{c2}\approx 1.2$~eV without evidence of first-order behavior.

1131: (The latter result is consistent with $U_{c2}\approx 2.4$~eV in Fig.~2

1132: of Ref.~\cite{prl05}.)  Although $Z_2(U)\approx 0.2\ldots0.8$ at 

1133: $U=0.8$~eV, i.e., for $U_{c1}\ll U \ll U_{c2}$, the analysis of the 

1134: self-energy reveals fundamentally different electronic properties 

1135: of the wide band in this region.     

1136: 

1137: \begin{figure}[t!]%15  see Gnuz in  ED8/Jump  and   ED6/T=0.010

1138:   \begin{center}

1139: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{Pariszu.ps}

1140:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig16.ps}

1141:   \end{center}

1142:   \vskip-2mm

1143: \caption{\label{Paris.zu}

1144: $Z_i(U)$ as a function of $U$ for $W_1=0.2$~eV, $W_2=2.0$~eV, 

1145: $T=1/120$~eV, calculated within ED/DMFT, $n_s=6$, for 

1146: $J=J'=0$,  $J=J'=U/4$, and $J=U/4$, $J'=0$.    

1147: Red and blue curves: wide band, black curves: narrow band.

1148: The $Z_1$ for  $J=J'=U/4$ and $J=U/4$, $J'=0$ are indistinguishable. 

1149: The dots mark the three cases shown in Fig.~\ref{Paris}.

1150: }\end{figure}

1151:    

1152: This is illustrated in Fig.~\ref{Paris} which shows 

1153: Im\,$\Sigma_2(i\omega_n)$ for $n_s=8$ and $n_s=6$. The ED results for 

1154: $n_s=8$ are in excellent agreement with the QMC data \cite{biermann05}

1155: for $J=J'=0$ as well as $J=U/4$, $J'=0$, with the exception of small 

1156: finite-size effects at the lowest Matsubara frequencies and lowest 

1157: temperature. (Since $W_2=2$~eV, $T=1/120\ldots1/40$~eV in these 

1158: calculations corresponds to about $T=16\ldots50$~meV in the case 

1159: $W_2=4$~eV considered in the preceding sections.)  Also, the 

1160: variation with $T$ is in our results slightly more pronounced 

1161: than in the case of the QMC/DMFT.

1162: \begin{figure}[t!]%16

1163:   \begin{center}

1164: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{Parisa.ps}

1165: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{Parisb.ps}

1166: %  \includegraphics[width=5.0cm,height=8cm,angle=-90]{Parisc.ps}

1167:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig17a.ps}

1168:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig17b.ps}

1169:   \includegraphics[width=5.0cm,height=8cm,angle=-90]{fig17c.ps}

1170:   \end{center}

1171:   \vskip-2mm

1172: \caption{\label{Paris}

1173: (a) Imaginary part of self-energy of wide band 

1174: for $J=0$ and $J=U/4$ in intermediate 

1175: phase for $U=0.8$~eV, $J'=0$, calculated within ED/DMFT with $n_s=8$. 

1176: $W_1=0.2$~eV, $W_2=2.0$~eV.  Red solid, blue dashed, and black dotted 

1177: curves: $T=1/120$~eV, $T=1/80$~eV, and $T=1/40$~eV, respectively.

1178: The dots denote the results for $T=1/120$~eV obtained in 

1179: Ref.~\cite{biermann05}. 

1180: (b) Same as (a) except for $n_s=6$.  

1181: (c) Same as (a) except for $J'=J=U/4$.

1182: }\end{figure}

1183: 

1184: For $n_s=6$ the ED results exhibit larger finite-size effects, as shown 

1185: in Fig.~\ref{Paris}(b). The variation with temperature is in this case 

1186: also larger than for $n_s=8$. Evidently, the subtle features of spectral 

1187: functions in the intermediate phase are not so well represented by 

1188: including only two bath levels. Nevertheless, the qualitative 

1189: difference between Fermi-liquid behavior for $J=0$ and the clear 

1190: deviation from this behavior for $J=U/4$ are very well reproduced 

1191: by these $n_s=6$ ED/DMFT calculations.   

1192: 

1193: For completeness we show in Fig.~\ref{Paris}(c) the $n_s=8$ ED/DMFT 

1194: results for isotropic Hund's coupling, i.e., $J'=J=U/4$. This case 

1195: is not yet accessible within QMC/DMFT because of sign problems at low 

1196: temperatures. 

1197: The inclusion of spin-flip and pair-exchange terms is seen to restore

1198: the limiting behavior $\Sigma_2(i\omega_n)\rightarrow 0$ for $\omega_n

1199: \rightarrow 0$. Nevertheless, compared to the case $J'=J=0$ shown in (a),

1200: $\Sigma_2$ now increases much more rapidly at small finite frequencies, 

1201: similarly to the data shown in Figs.~\ref{OSMT} and \ref{NRG34}.  

1202: (As argued in Section 5, the shoulder near $\omega_n\approx0.06$ is  

1203: caused by finite-size effects.)

1204: Thus, instead of quasiparticles with weight $Z_2\approx0.8$ for $J=J'=0$,

1205: finite Hund's exchange with $J'=J=U/4$ supports a state of infinite 

1206: lifetime at $E_F$, but does not satisfy Fermi-liquid criteria at 

1207: $\omega>0$. This picture is consistent with the $T=0$ ED/DMFT results

1208: by Biermann {\it et al.}~\cite{biermann05}.

1209: The spectral function of the wide band therefore should exhibit 

1210: a narrow peak at $E_F$, similar to the one shown in Fig.~\ref{NRG12} 

1211: for $U=2.4$~eV. 

1212: 

1213: 

1214: \section{9 \ Comments on IPT/DMFT results}

1215: 

1216: In Ref.~\cite{prb70} we reported DMFT calculations for the two-band 

1217: Hubbard model within QMC and iterated perturbation theory \cite{ipt}. 

1218: Both approaches were shown to yield a consistent picture of the

1219: Mott transition, in the sense that both exhibit a single first-order

1220: transition at which the narrow band becomes insulating and the 

1221: wide band begins to show progressive bad-metallic behavior.

1222: This band becomes fully insulating at a higher Coulomb energy in a 

1223: non-first-order manner. As shown via ED/DMFT at $T=0$ in Ref.~\cite{koga2} 

1224: and at $T>0$ in Ref.~\cite{prl05}, the nature of this upper 

1225: transition depends in a critical way on the treatment of onsite 

1226: exchange interactions. In the absence of spin-flip and pair-exchange

1227: terms it is continuous, consistent with the QMC results, while for

1228: full Hund's coupling it becomes first-order.

1229: 

1230: The question then arises whether the IPT approach also supports this

1231: picture. In fact, as pointed out in Ref.~\cite{prl05}, it is surprising 

1232: that the IPT gives only one common first-order transition even though 

1233: spin-flip and pair-exchange terms are included. It can easily be shown,

1234: however, that to second-order in the Coulomb interaction the subband 

1235: self-energies with and without spin-flip and pair-exchange are the same, 

1236: except for a slightly different relation between $U$ and $J$. Thus, 

1237: $J'=J=0.25\,U$ gives the same result as $J'=0$ and $J\approx0.22\,U$. 

1238: Corrections to the simple second-order diagram via renormalization of subband

1239: energies to yield the correct atomic limit \cite{kajueter,lichtenstein}

1240: are also insensitive to the choice of $J'$ for the present two-band model. 

1241: 

1242: To resolve this puzzle we have repeated the IPT calculations reported 

1243: in Ref.~\cite{prb70} at even lower temperatures and found indeed a tiny 

1244: hysteresis loop also at the upper Mott transition, with a critical 

1245: temperature of approximately $T_{c2}\approx5$~meV, i.e., significantly 

1246: lower than the critical temperature of the lower Mott transition, 

1247: $T_{c1}\approx50$~meV. These findings suggest that present formulations 

1248: of IPT are too simple to deal with the full complexity of Hund's  

1249: exchange. Diagrams beyond second-order are required to distinguish

1250: more clearly between $J'=J$ and $J'=0$ treatments.      

1251: 

1252: 

1253: \section{10 \ Summary and outlook} 

1254: 

1255: Finite-temperature ED/DMFT studies are carried out in order to 

1256: explore the usefulness of this approach for multi-band systems.

1257: The important feature here is that onsite Coulomb and 

1258: exchange interactions are fully included. Since in the past ED

1259: has been used mainly for single-band systems, a particular aim 

1260: of this work is to illustrate the dependence of self-energies 

1261: on the cluster size and to test the range of applicability of 

1262: this method. As an example, we focus on the Hubbard model for

1263: two bands of different widths and investigate the metal insulator 

1264: transition as a function of Hund's exchange.   

1265: 

1266: The surprising and potentially very useful result of this work is

1267: that even a cluster size of $n_s=6$, with only two bath levels per 

1268: impurity orbital, provides a qualitatively correct picture in all 

1269: of the important phases of the $T/U$ phase diagram. Moreover, 

1270: finite-size effects are substantially reduced for $n_s=8$, i.e., using

1271: one extra bath level per impurity orbital. Thus, in the phase just 

1272: below the main first-order Mott transition, both subbands exhibit 

1273: clear metallic properties, albeit with strongly reduced 

1274: quasi-particle weights. The intermediate region between the purely

1275: metallic and insulating phases depends in a subtle manner on the 

1276: exchange interactions included in the ED calculation.  

1277: For full Hund's coupling coexisting metallic and insulating

1278: subbands are found, where the wide band exhibits infinite lifetime

1279: at $E_F$, but non-Fermi-liquid behavior at finite frequencies.

1280: For Ising-like exchange, this breakdown of Fermi-liquid behavior

1281: is enhanced, giving finite lifetime even at $E_F$. 

1282: 

1283: To explore the influence of finite-size effects in the two-band 

1284: ED/DMFT approach NRG calculations were performed for an  effective 

1285: one-band model suitable to describe the wide band in the intermediate 

1286: phase. Both for full Hund's coupling and Ising-like exchange,

1287: very good agreement with the ED results is found for 

1288: temperatures in the range $T=20\ldots30$~meV. At lower $T$, larger

1289: differences appear on a quantitative level at low frequencies.

1290: Nevertheless, the important qualitative differences between various 

1291: phases, especially the characteristic low-frequency variation of 

1292: the self-energy in the two types of non-Fermi-liquid regions for 

1293: isotropic and anisotropic exchange coupling, are fully reproduced 

1294: by the ED approach, both for $n_s=6$ and $n_s=8$.     

1295: 

1296: To test the accuracy of the two-band ED approach we also have applied

1297: it to models studied earlier within QMC/DMFT, neglecting spin-flip

1298: and pair-exchange. Nearly quantitative agreement is obtained for

1299: a cluster size $n_s=8$, but even the ED results for $n_s=6$ are

1300: found to be qualitatively reliable. 

1301: 

1302: The present ED approach utilizes full diagonalization of the 

1303: impurity Hamiltonian. Since at low temperatures only a limited 

1304: range of excited states is relevant for the local Green's functions 

1305: and self-energies it should be very useful to generalize finite-$T$ 

1306: Lanczos one-band methods \cite{jaklic,capone} in order to bridge the 

1307: gap between the present work and the true $T=0$ limit, and to apply 

1308: the finite-temperature ED/DMFT approach to realistic two-band and 

1309: three-band materials.

1310: 

1311: \bigskip

1312: {\bf Acknowledgements}: \ 

1313: One of us (A. L.) likes to thank A. I. Lichtenstein for parts of 

1314: the multi-band ED code. Some of the ED and NRG DMFT calculations 

1315: were carried out on the IBM supercomputer (JUMP) of the 

1316: Forschungszentrum J\"ulich.

1317: 

1318: \bigskip

1319: Email: \ a.liebsch@fz-juelich.de; \ t.costi@fz-juelich.de 

1320: 

1321: \begin{thebibliography}{99}

1322: 

1323: %\bibitem{dmft1}

1324: %   W. Metzner and D. Vollhardt, 

1325: %      Phys. Rev. Lett. {\bf 62}, 324 (1989);

1326: %   A. Georges and G. Kotliar, 

1327: %      Phys. Rev. B {\bf 45}, 6479 (1992);

1328: %   M. Jarrel,                 

1329: %      Phys. Rev. Lett. {\bf 69}, 168 (1992).

1330: 

1331: \bibitem{rmp}

1332:    For a review of early papers, see:

1333:    A. Georges, G. Kotliar, W. Krauth and M. J. Rozenberg, 

1334:    Rev. Mod. Phys. {\bf 68}, 13 (1996).  

1335: 

1336: \bibitem{reviews}

1337:    For recent reviews, see: K. Held, cond-mat/0511293;

1338:    A. Georges {\it et al.}, cond-mat/0403123;  

1339:    G. Kotliar and S. Savrasov, cond-mat/0208241.  

1340: 

1341: \bibitem{qmc} 

1342:    J. E. Hirsch and R. M. Fye, 

1343:       Phys. Rev. Lett. {\bf 56}, 2521 (1986).

1344: 

1345: \bibitem{held}  

1346:    For a discussion of this problem, see: 

1347:    K. Held and D. Vollhardt, 

1348:       Eur. Phys. J. B {\bf 5}, 473 (1998). 

1349: 

1350: \bibitem{arita}

1351:    R. Arita and K. Held,

1352:       Phys. Rev. B {\bf 96}, 201102  (2005); see also:

1353:    S. Sakai, R. Arita and H. Aoki,

1354:       Phys. Rev. B {\bf 70}, 172504  (2004).

1355: 

1356: \bibitem{koga3}

1357:    A. Koga, N. Kawakami, T. M. Rice, and  M. Sigrist,

1358:       Phys. Rev. B  {\bf 72}, 045128 (2005).

1359: 

1360: \bibitem{rubtsov}

1361:    See also: A. N. Rubtsov, V. V. Savkin, and A. I. Lichtenstein,  

1362:       Phys. Rev. B  {\bf 72}, 035122 (2005).

1363: 

1364: 

1365: \bibitem{ed}  

1366:    M. Caffarel and W. Krauth, 

1367:       Phys. Rev. Lett. {\bf 72}, 1545 (1994).

1368: 

1369: \bibitem{anisimov}

1370:    V. I. Anisimov, I. A. Nekrasov, D. E. Kondakov, T. M. Rice 

1371:       and M. Sigrist,

1372:    Euro. Phys. J. B {\bf 25}, 191 (2002). 

1373: 

1374: \bibitem{epl} 

1375:    A. Liebsch, 

1376:       Europhys. Lett. {\bf 63}, 97 (2003). 

1377: 

1378: \bibitem{prl}  

1379:    A. Liebsch, 

1380:       Phys. Rev. Lett. {\bf 91}, 226401 (2003).

1381: 

1382: \bibitem{koga1}

1383:    A. Koga, N. Kawakami, T. M. Rice, and  M. Sigrist,

1384:       Phys. Rev. Lett. {\bf 92}, 216402 (2004).

1385: 

1386: \bibitem{prb70}  

1387:    A. Liebsch, 

1388:       Phys. Rev. B  {\bf 70}, 165103 (2004).

1389: 

1390: \bibitem{koga2}

1391:    A. Koga, N. Kawakami, T. M. Rice, and  M. Sigrist,

1392:       Physica B     {\bf 359-361}, 1366   (2005).

1393: 

1394: \bibitem{song}

1395:    Yun Song and L.-J. Zou, 

1396:       Phys. Rev. B     {\bf 72}, 085114 (2005).

1397: 

1398: \bibitem{prl05}  

1399:    A. Liebsch, 

1400:       Phys. Rev. Lett. {\bf 95}, 116402 (2005).

1401: 

1402: \bibitem{inaba}

1403:    K. Inaba, A. Koga, S. Suga and N. Kawakami,

1404:       J. Phys. Soc. Jpn. {\bf 74}, 2393 (2005).

1405: 

1406: 

1407: \bibitem{medici}

1408:    L. de' Medici, A. Georges and S. Biermann,

1409:       Phys. Rev. B     {\bf 72}, 205124 (2005).

1410: 

1411: \bibitem{ferrero}

1412:    M. Ferrero, F. Becca, M. Fabrizio, and M. Capone,

1413:       Phys. Rev. B     {\bf 72}, 205126 (2006).

1414: 

1415: \bibitem{knecht}

1416:    C. Knecht, N. Bl\"umer, and P.G.J. van Dongen,

1417:       Phys. Rev. B     {\bf 72}, 081103(R) (2005).

1418: 

1419: %\bibitem{comment}

1420: %  A. Liebsch, cond-mat/0506138.  

1421: 

1422: \bibitem{biermann05}

1423:    S. Biermann, L. de' Medici, and A. Georges,

1424:       Phys. Rev. Lett. {\bf 95}, 206401 (2005).

1425: 

1426: \bibitem{jaklic} 

1427:      J. Jakli\v{c} and P. Prelov\v{s}ek, 

1428:         Adv. Phys. {\bf 49}, 1 (2000).

1429: 

1430: \bibitem{capone}

1431:    M. Capone, L. de' Medici, and A. Georges,

1432:       cond-mat/ 0512484.

1433: 

1434: \bibitem{maeno}

1435:    T. Maeno, T. M. Rice, and M. Sigrist, 

1436:       Phys. Today {\bf 54}, (1) 42 (2001).

1437: 

1438: \bibitem{nacoo}

1439:    K. Takada {\it et al.}, Nature {\bf 422}, 53 (2003); 

1440:    R. E. Schaak {\it et al.}, Nature {\bf 424}, 527 (2003);

1441:    M. L. Foo {\it et al.}, Phys. Rev. Lett. {\bf 92}, 247001 (2004). 

1442: 

1443: \bibitem{pruschke}

1444:    Th. Pruschke and R. Bulla, 

1445:      Eur. Phys. J. B {\bf 44}, 217 (2005).

1446: 

1447: \bibitem{costi.02} 

1448:      T. A. Costi and N. Manini, 

1449:       J. Low Temp. Phys. {\bf 126}, 835 (2002)

1450: 

1451: \bibitem{wilson.74} 

1452:       K. G. Wilson, 

1453:       Rev. Mod. Phys. {\bf 47}, 773 (1975).

1454: 

1455: \bibitem{costi.94} 

1456:        T. A. Costi, A. C. Hewson and V. Zlatic, 

1457:        J. Phys.: Condens. Matter {\bf 6}, 2519 (1994).

1458: 

1459: \bibitem{hofstetter.00} 

1460:       W. Hofstetter, 

1461:       Phys. Rev. Lett. {\bf 85}, 1508 (2000).

1462: 

1463: \bibitem{cragg.79} 

1464:     D. M. Cragg and P. Lloyd, J. Phys. C: Solid State 

1465:        Phys. {\bf 12}, L215 (1979).

1466: 

1467: \bibitem{koller.05} 

1468:      W. Koller,  A. C. Hewson and D. Meyer, 

1469:         Phys. Rev. B{\bf 72}, 045117 (2005).

1470: 

1471: \bibitem{bulla.98} 

1472:       R. Bulla, T. Pruschke and A. C. Hewson, 

1473:       J. Phys.: Condens. Matter {\bf 10 }, 8365 (1998).

1474: 

1475: \bibitem{bulla.01}

1476:       R. Bulla, T. A. Costi and D. Vollhardt, 

1477:         Phys. Rev. B{\bf 64}, 045103 (2001).

1478: 

1479: \bibitem{costi} 

1480:        T. A. Costi, to be published.

1481: 

1482: \bibitem{metallic}

1483:    A. Liebsch, to be published.

1484: 

1485: 

1486: \bibitem{kajueter}

1487:    H. Kajueter and G. Kotliar, 

1488:       Phys. Rev. Lett. {\bf 77}, 131  (1997).

1489: 

1490: \bibitem{ipt}

1491:    A. Georges and G. Kotliar, 

1492:       Phys. Rev. B {\bf 45}, 6479 (1992).

1493: 

1494: \bibitem{lichtenstein}

1495:    A. I. Lichtenstein and M. I. Katsnelson,

1496:       Phys. Rev. B     {\bf 57}, 6884 (1998).

1497: 

1498: \end{thebibliography}

1499: \end{document}

1500: 

1501: