1: \documentclass[twocolumn,aps,prb,amsmath,floatfix]{revtex4}
2: \usepackage{graphicx}
3: \begin{document}
4: \title{Multisite versus multiorbital Coulomb correlations studied
5: within finite-temperature exact diagonalization dynamical mean-field theory}
6: \author{A.~Liebsch$^1$, H. Ishida$^2$, and J. Merino$^3$}
7: \affiliation{$^1$Institut f\"ur Festk\"orperforschung,
8: Forschungszentrum J\"ulich,
9: 52425 J\"ulich, Germany \\
10: $^2$College of Humanities and Sciences, Nihon University,~Tokyo 156,
11: Japan\\
12: $^3$Departamento de F\'\i sica Te\'orica de la Materia Condensada,
13: Universidad Aut\'onoma de Madrid, Madrid 28049, Spain}
14: \begin{abstract}
15: The influence of short-range Coulomb correlations on
16: the Mott transition in the single-band Hubbard model at half-filling
17: is studied within cellular dynamical mean field theory for square and
18: triangular lattices. Finite-temperature exact diagonalization is used to
19: investigate correlations within two-, three-, and four-site clusters.
20: Transforming the non-local self-energy from a site basis to a molecular
21: orbital basis, we focus on the inter-orbital charge transfer between
22: these cluster molecular orbitals in the vicinity of the Mott transition.
23: In all cases studied, the charge transfer is found to be small,
24: indicating weak Coulomb induced orbital polarization despite sizable
25: level splitting between orbitals. These results demonstrate that
26: all cluster molecular orbitals take part in the Mott transition and that
27: the insulating gap opens simultaneously across the entire Fermi surface.
28: Thus, at half-filling we do not find orbital-selective Mott transitions,
29: nor a combination of band filling and Mott transition in different orbitals.
30: Nevertheless, the approach towards the transition differs greatly between
31: cluster orbitals, giving rise to a pronounced momentum variation along the
32: Fermi surface, in agreement with previous works.
33: The near absence of Coulomb induced orbital polarization in
34: these clusters differs qualitatively from single-site multi-orbital studies
35: of several transition metal oxides, where the Mott phase exhibits nearly
36: complete orbital polarization as a result of a correlation driven
37: enhancement of the crystal field splitting.
38: The strong single-particle coupling among cluster orbitals in the
39: single-band case is identified as the source of this difference.
40: \\
41: \mbox{\hskip1cm} \\
42: PACS. 71.20.Be Transition metals and alloys - 71.27+a Strongly correlated
43: electron systems
44: \end{abstract}
45: \maketitle
46:
47: \section{Introduction}
48:
49: Considerable progress has recently been achieved in the understanding of the
50: Mott transition in a variety of transition metal oxides.\cite{kotliar06}
51: Whereas density functional theory in the local density approximation (LDA)
52: predicts many of these materials to be metallic, the explicit treatment of
53: local Coulomb interactions via dynamical mean field theory (DMFT)\cite{dmft}
54: correctly yields insulating behavior for realistic values of the on-site
55: Coulomb energy $U$. In the metallic phase, the non-cubic structure of some
56: of these systems gives rise to non-equivalent, partially filled subbands
57: that are split by a crystal field and exhibit orbital dependent electron
58: occupancies. The hallmark of the Mott transition of these oxides is that
59: orbital polarization can be greatly increased by Coulomb correlations
60: and that the insulating phase is nearly completely orbitally polarized.
61: For instance, in the case of LaTiO$_3$, the $e'_g$ bands are pushed above
62: the Fermi level and the remaining singly occupied $a_g$ subband
63: is split into lower and upper Hubbard bands.\cite{pavarini,prb08}
64: In the case of V$_2$O$_3$, Coulomb correlations push the
65: $a_g$ band above the Fermi level, and the doubly degenerate
66: $e'_g$ subbands exhibit a Mott gap.\cite{keller,poteryaev}
67: Also, in the insulating phase of Ca$_2$RuO$_4$,
68: the $d_{xy}$ like band is completely filled and the $d_{xz,yz}$ like
69: subbands are split into Hubbard bands.\cite{anisimov,prl07}
70: The common feature of the Mott transition in these materials is that
71: the effective band degeneracy is reduced from three to two or one,
72: so that the critical Coulomb energy is lower than it would be if the
73: $t_{2g}$ bands were fully degenerate.
74: On the other hand, other materials can exhibit a quite different behavior.
75: For instance, orbital polarization in BaVS$_3$ was shown to decrease with
76: increasing local Coulomb interaction.\cite{lechermann}
77: Also, the Mott transition in LaVO$_3$ and YVO$_3$ occurs before orbital
78: polarization is complete.\cite{ray}
79: Moreover, in a hypothetical tetragonal structure of LaTiO$_3$,
80: relevant for heterostructures, the Mott phase is reached when
81: $n_{xz,yz}$ approaches $1/4$ and $n_{xy}$ vanishes.\cite{tetra}
82: Finally, the possibility of so-called orbital selective Mott transitions
83: in multi-band systems has been discussed extensively in the literature.
84: \cite{anisimov,koga,prb04,song,inaba,costi}
85: These different trends underline the remarkably rich physics of
86: Mott transitions in multi-orbital materials.
87:
88: The aim of this work is to investigate the relationship between Coulomb
89: correlations in single-site multi-orbital systems as described above to
90: those occurring within a single band when inter-site Coulomb correlations
91: are taken into account. The influence of short-range correlations on
92: the nature of the Mott transition is currently of great interest and
93: has been studied by many groups.
94: \cite{tohyama,preuss,moreo,hettler,senechal,lich,huscroft,moukouri,%
95: kotliar01,imai,maier2,onoda,kyung,potthoff,senechal1,parcollet,capone,%
96: senechal2,civelli,maier,kyung2,capone2,kyung3,merino,zhang,ohashi,saheb,%
97: park,koch,ferrero,gull,lee,balzer,senechal3,Ferrero}
98: Here we examine the role of correlation
99: driven orbital polarization in the vicinity of the Mott transition.
100: For example, it is well known that in a minimal two-site cluster model,
101: \cite{ferrero} which permits explicit treatment of short-range Coulomb
102: correlations in an isotropic square lattice, the Green's function
103: and self-energy become diagonal if one transforms the site basis to a
104: diagonal bonding - antibonding molecular orbital basis.
105: In a four-site cluster model, diagonality is obtained by
106: transforming sites to cluster molecular orbitals characterized by
107: $\Gamma=(0,0)$, $X=(\pi,0),(0,\pi)$, and $M=(\pi,\pi)$.\cite{park}
108: The molecular orbital components of the self-energy provide
109: qualitative information on the importance of correlations in
110: the corresponding sections of the Brillouin Zone.
111: In the case of an isotropic triangular lattice, Green's function and
112: self-energy can be diagonalized by an analogous transformation to
113: molecular orbitals appropriate for a three-site cluster.\cite{koch}
114: The question then arises whether these cluster molecular orbitals in
115: the single-band case obey a similar scenario as the multi-orbital
116: systems mentioned above.
117:
118: Since an approximate momentum variation of the lattice self-energy in
119: these models can be derived from a linear superposition of the respective
120: molecular orbital components of the cluster self-energy, the effect of
121: correlation enhanced orbital polarization is of direct relevance for the
122: question of whether the Mott gap opens uniformly across the Fermi
123: surface, or whether it opens first in certain regions of the Brillouin
124: Zone (e.g. near the so-called hot spots) and only at larger $U$ in the
125: remaining regions (the so-called cold spots). The latter picture would
126: be analogous to the orbital selective Mott transition which can occur
127: within single-site DMFT treatments of certain multi-band systems.
128: \cite{anisimov,koga,prb04,song,inaba,costi} Another possibility,
129: analogous to multi-orbital materials such as LaTiO$_3$, V$_2$O$_3$,
130: and Ca$_2$RuO$_4$, is that a subset of cluster orbitals could exhibit
131: a genuine Mott transition, while the remaining ones are pushed above
132: or below the Fermi level at about the same critical $U$.
133:
134: To account for inter-site correlations we use DMFT combined with finite
135: temperature exact diagonalization (ED).\cite{ed}
136: It was recently shown\cite{perroni} that this method can be generalized
137: to multi-band materials by computing only those excited states of the
138: impurity Hamiltonian that are within a narrow range above the ground state,
139: where the Boltzmann factor provides the convergence criterion. Exploiting
140: the sparseness of the Hamiltonian, these states can be computed very
141: efficiently by using the Arnoldi algorithm.\cite{arnoldi} Higher excited
142: states enter via Green's functions which are evaluated using the Lanczos
143: method. This approach has proved to be highly useful for the study of
144: strong correlations in several transition metal oxides.
145: \cite{prb08,perroni,naco,tetra,v2o3-al} An important feature of ED/DMFT is
146: that low temperatures and large Coulomb energies can be reached.
147: The adaptation of single-site multi-orbital ED to multi-site single-band
148: systems is discussed in detail below. In particular, we introduce a
149: mixed site - molecular orbital basis which permits a more flexible
150: and more accurate projection of the lattice Green's function onto the
151: cluster than in a pure site representation. Previous multi-site ED/DMFT
152: studies focussed on $T=0$.\cite{kyung,civelli,kyung2,capone2,kyung3,merino,zhang}
153: The extension to finite $T$ discussed here
154: is especially useful for the evaluation of the $T/U$ phase diagram.
155:
156: The main result of this work is that in all cluster models studied
157: here for half-filled square and triangular lattices, there is little
158: enhancement of orbital polarization in the vicinity of the Mott
159: transition. Thus, despite sizable level splitting
160: between these cluster orbitals, they all exhibit Mott gaps at the
161: same critical Coulomb energy. As a consequence, the Mott gap in these
162: models opens uniformly across the Fermi surface. For the square lattice
163: we show explicitly that the Mott gap at the cold spot $M/2=(\pi/2,\pi/2)$
164: of the Brillouin Zone is driven by Coulomb correlations at the hot spot
165: $X=(\pi,0)$. Therefore, there is
166: no orbital selective Mott transition. Moreover, there is no evidence
167: for the combination of partial band filling and Mott transition in
168: remaining subbands that is characteristic of single-site DMFT
169: treatments of the multi-orbital materials LaTiO$_3$, V$_2$O$_3$,
170: and Ca$_2$RuO$_4$, as mentioned above.
171:
172: The outline of this paper is as follows. Section II discusses the
173: theoretical aspects of our cluster ED/DMFT implementation of finite
174: temperature exact diagonalization. Section III provides the results
175: for the two-site and four-site clusters of the square lattice, and
176: the three-site cluster of the isotropic triangular lattice.
177: In Section IV we briefly discuss analogies and differences between
178: these multi-site correlation effects and those investigated previously
179: in single-site DMFT treatments of multi-orbital materials.
180: The conclusions are presented in Section V.
181:
182: \section{Multi-Site ED/DMFT}
183:
184: Let us consider the single-band Hubbard model
185: \begin{equation}
186: H = -t \sum_{\langle ij\rangle\sigma} ( c^+_{i\sigma} c_{j\sigma} + H.c.)
187: + U \sum_i n_{i\uparrow} n_{i\downarrow}
188: \end{equation}
189: where the sum in the first term extends over nearest neighbor sites.
190: The hopping integral $t$ will be set equal to unity throughout this paper.
191: Thus, the band widths of the square and triangular lattices are $W=8$ and
192: $W=9$, respectively.
193: Within cellular dynamical mean field theory (CDMFT)\cite{kotliar01,maier}
194: the interacting lattice Green's function in the cluster site basis is given by
195: \begin{equation}
196: G_{ij}(i\omega_n) = \sum_{\vec k} \left( i\omega_n + \mu - t(\vec k) -
197: \Sigma(i\omega_n)\right)^{-1}_{ij} \label{G}
198: \end{equation}
199: where the $\vec k$ sum extends over the reduced Brillouin Zone,
200: $\omega_n=(2n+1)\pi k_BT$ are Matsubara frequencies and $\mu$ is the chemical
201: potential. The lattice constant is unity.
202: $t(\vec k)$ denotes the hopping matrix for the superlattice and
203: $\Sigma(i\omega_n)$ represents the cluster self-energy matrix.
204: To make contact to other recent works,\cite{parcollet,zhang,park,gull,ferrero}
205: we consider here the paramagnetic metal insulator transition.
206:
207: In the site basis, the Green's functions for two-site, three-site and
208: four-site clusters have the structure
209: \begin{eqnarray}
210: G^{(2)} &=& \left( \begin{array}{ll}
211: a & b \\
212: b & a \\
213: \end{array} \right)\label{G2}\\
214: G^{(3)} &=& \left( \begin{array}{lll}
215: a & b & b \\
216: b & a & b \\
217: b & b & a \\
218: \end{array} \right)\label{G3} \\
219: G^{(4)} &=& \left( \begin{array}{llll}
220: a & b & b & c \\
221: b & a & c & b \\
222: b & c & a & b \\
223: c & b & b & a \\
224: \end{array} \right)\label{G4}
225: \end{eqnarray}
226: with $a=G_{11}$, $b=G_{12}$ and $c=G_{14}$. Site labels in the square lattice
227: refer to $1\equiv(0,0),\ 2\equiv(1,0),\ 3\equiv(0,1),\ 4\equiv(1,1)$ and in the
228: triangular lattice to $1\equiv(0,0),\ 2\equiv(1,0),\ 3\equiv(1/2,\sqrt3/2)$.
229: The superscript denotes the cluster size $n_c$ in the square lattice
230: ($n_c=2$ or $n_c=4$) or triangular lattice ($n_c=3$), respectively.
231: In the site bases, the corresponding self-energy matrices
232: $\Sigma^{(n_c)}(i\omega_n)$ have the same symmetry properties
233: as the Green's functions.
234:
235: A key aspect of DMFT is that, to avoid double-counting of Coulomb
236: interactions in the quantum impurity calculation, it is necessary to
237: remove the self-energy from the cluster in which correlations are
238: treated exactly. This removal yields the Green's function
239: \begin{equation}
240: G_0(i\omega_n) = [G(i\omega_n)^{-1} + \Sigma(i\omega_n)]^{-1} .
241: \label{G0}
242: \end{equation}
243: These matrices also exhibit the symmetry properties specified above.
244:
245: For the purpose of the ED calculations it is convenient to transform the
246: site bases into molecular orbital bases in which the Green's functions and
247: self-energies become diagonal. For the two-site cluster, molecular orbitals
248: are given by the bonding - anti-bonding combinations
249: $\phi_{1,2}=(|1\rangle\pm|2\rangle)/\sqrt2$.
250: For the four-site cluster, they are formed by the plaquettes
251: $\phi_1=(|1\rangle+|2\rangle+|3\rangle+|4\rangle)/2$,
252: $\phi_2=(|1\rangle+|2\rangle-|3\rangle-|4\rangle)/2$,
253: $\phi_3=(|1\rangle-|2\rangle+|3\rangle-|4\rangle)/2$,
254: $\phi_4=(|1\rangle-|2\rangle-|3\rangle+|4\rangle)/2$.
255: Finally, for the three-site cluster of the triangular lattice they
256: can be written as:
257: $\phi_1=( |1\rangle+|2\rangle+|3\rangle)/\sqrt3$,
258: $\phi_2=(-2|1\rangle+|2\rangle+|3\rangle)/\sqrt6$,
259: $\phi_3=( |2\rangle-|3\rangle)/\sqrt2$.
260: In theses cluster molecular orbital bases, the above
261: Green's functions take the form
262: \begin{eqnarray}
263: G^{(2)} &=& \left( \begin{array}{cc}
264: a+b & 0 \\
265: 0 & a-b \\
266: \end{array} \right)\label{g2}\\
267: G^{(3)} &=& \left( \begin{array}{ccc}
268: a+2b & 0 & 0 \\
269: 0 & a-b & 0 \\
270: 0 & 0 & a-b \\
271: \end{array} \right) \label{g3} \\
272: G^{(4)} &=& \left( \begin{array}{cccc}
273: a+2b+c & 0 & 0 & 0 \\
274: 0 & a-c & 0 & 0 \\
275: 0 & 0 & a-c & b \\
276: 0 & 0 & 0 & a-2b+c \label{g4} \\
277: \end{array} \right)
278: \end{eqnarray}
279: The self-energies $\Sigma(i\omega_n)$ and Green's functions
280: $G_0(i\omega_n)$ can be diagonalized in the same fashion.
281: We denote these elements as $G_m(i\omega_n)$, $\Sigma_m(i\omega_n)$
282: and $G_{0,m}(i\omega_n)$.
283:
284: \begin{figure} % [b!] %1
285: \begin{center}
286: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{DCA.dos.nc=2.ps}
287: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{DCA.dos.nc=4.ps}
288: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{DCA.dos.nc=3.ps}
289: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig1a.ps}
290: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig1b.ps}
291: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig1c.ps}
292: \end{center}
293: \caption{(Color online)
294: Total density of states $\rho(\omega)$ and molecular orbital components
295: $\rho_m(\omega)$ for two-site and four-site clusters of square lattice
296: (top panels), and of three-site cluster of triangular lattice (bottom panel).
297: For clarity, the molecular orbital components are divided by $n_c$.
298: }\end{figure}
299:
300: In the site basis, the local density of states in the non-interacting
301: limit is given by
302: \begin{eqnarray}
303: \rho_{ii}(\omega) &=& -\frac{1}{\pi}\,{\rm Im}\, G_{ii}(\omega)
304: \end{eqnarray}
305: with $\Sigma=0$. Since we consider isotropic clusters, all sites
306: are equivalent, so that $\rho_{ii}(\omega)$ coincides with the density
307: of states $\rho(\omega)$. In the molecular orbital basis, the density
308: of states components $\rho_m(\omega)$ have different shapes and different
309: centroids, analogous to the crystal field split density of states
310: components of many transition metal oxides.
311: Figure~1 shows these densities for the two-site and four-site clusters
312: of the square lattice and the three-site cluster of the triangular
313: lattice, as described above. According to Eqs.~(\ref{g2}-\ref{g4}),
314: the cluster molecular orbital densities of states are given by
315: \begin{eqnarray}
316: \rho^{(2)}_{1 } &=& \rho^{(2)}_{11} + \rho^{(2)}_{12} \label{rho2} \\
317: \rho^{(2)}_{2 } &=& \rho^{(2)}_{11} - \rho^{(2)}_{12} \nonumber \\
318: \rho^{(3)}_{1 } &=& \rho^{(3)}_{11}+2 \rho^{(3)}_{12} \label{rho3} \\
319: \rho^{(3)}_{2} = \rho^{(3)}_{3} &=& \rho^{(3)}_{11} - \rho^{(3)}_{12} \nonumber \\
320: \rho^{(4)}_{1} &=& \rho^{(4)}_{11} +2\rho^{(4)}_{12}+\rho^{(4)}_{14}\label{rho4}\\
321: \rho^{(4)}_{2} =\rho^{(4)}_{3} &=& \rho^{(4)}_{11} -\rho^{(4)}_{14} \nonumber \\
322: \rho^{(4)}_{4} &=& \rho^{(4)}_{11}- 2\rho^{(4)}_{12}+\rho^{(4)}_{14}\nonumber
323: \end{eqnarray}
324: where $\rho^{(n_c)}_{ij}$ are the site components for cluster $n_c$.
325:
326: From these cluster molecular orbital densities of states an approximate
327: momentum variation across the Brillouin Zone can be constructed (see below).
328: For instance, in the case of the square lattice with $n_c=4$, densities
329: associated with the high-symmetry points of the original lattice are given by
330: $\rho_\Gamma(\omega)=\rho_1(\omega)$,
331: $\rho_X(\omega)=\rho_{2}(\omega)=\rho_{3}(\omega)$ and
332: $\rho_M(\omega)=\rho_4(\omega)$.
333: At $M/2=(\pi/2,\pi/2)$, the density of states corresponds to the local
334: density $\rho(\omega)=\rho_{11}(\omega)=(\rho_\Gamma+\rho_M+2\rho_X)/4$.
335: Note, however, that all molecular
336: orbital densities extend across the entire band width. Thus, they are not
337: identical with those sections of the local density of states that originate
338: in momentum regions surrounding the high-symmetry points, as would be the
339: case in the dynamical cluster approximation (DCA).\cite{hettler,maier}
340:
341:
342: % The centroids of these distributions are
343: % $\epsilon_1=-\epsilon_2=-0.975$ for the two-site cluster,
344: % $\epsilon_1=-2.73,\ \epsilon_{2,3}=0.23$ in the three-site case,
345: % (an on-site energy $\epsilon_0=-0.83$ was added to ensure that the Fermi
346: % level is zero), and
347: % $\epsilon_1=-\epsilon_4=-1.96,\ \epsilon_{2,3}=0$ for the four-site cluster.
348: % The effective crystal field splittings between cluster orbitals, i.e.
349: % the differences between their centroids, are then
350: % $\Delta^{(2)}_{21 }=1.95$,
351: % $\Delta^{(3)}_{21=31}=2.96$, and
352: % $\Delta^{(4)}_{43=21}=1.96$.
353: % Relative to the total band width these splittings are much larger than
354: % in the multi-orbital materials mentioned in the Introduction. It will
355: % therefore be of interest to inquire whether Coulomb interactions also
356: % give rise to a similar kind of enhanced orbital polarization.
357:
358: We now project the Green's function $G_0(i\omega_n)$ defined in Eq.~(\ref{G0})
359: onto a cluster consisting of $n_c$ impurity levels and $n_b$ bath levels.
360: The total number of levels is $n_s=n_c+n_b$.
361: In the site basis we have
362: \begin{eqnarray}
363: G_0(i\omega_n) &\approx& G^{cl}_0(i\omega_n) \nonumber\\
364: &=& \left( i\omega_n + \mu - h - \Gamma(i\omega_n) \right)^{-1}
365: \end{eqnarray}
366: where $h$ is the non-interacting impurity cluster Hamiltonian and $\Gamma(i\omega_n)$
367: the hybridization matrix describing the coupling between impurity cluster and bath.
368: Thus,
369: \begin{eqnarray}
370: h^{(2)} &=& \left( \begin{array}{ll}
371: \epsilon_0 & t \\
372: t & \epsilon_0 \\
373: \end{array} \right)\\
374: h^{(3)} &=& \left( \begin{array}{lll}
375: \epsilon_0 & t & t \\
376: t & \epsilon_0 & t \\
377: t & t & \epsilon_0 \\
378: \end{array} \right) \label{h3} \\
379: h^{(4)} &=& \left( \begin{array}{llll}
380: \epsilon_0 & t & t & 0 \\
381: t & \epsilon_0 & 0 & t \\
382: t & 0 & \epsilon_0 & t \\
383: 0 & t & t & \epsilon_0 \\
384: \end{array} \right)
385: \end{eqnarray}
386: For the square lattice we choose $\epsilon_0=0$ and for the triangular lattice
387: $\epsilon_0=-0.83$, so that the Fermi level coincides with $\omega=0$.
388:
389: Instead of expressing the non-diagonal hybridization matrix $\Gamma(i\omega_n)$
390: in a site basis, it is convenient to go over to the molecular
391: orbital basis in which $G_0(i\omega_n)$ is diagonal. Assuming that
392: each component $G_{0,m}(i\omega_n)$ couples only with its own bath,
393: we have
394: \begin{eqnarray}
395: G_{0,m}(i\omega_n) &\approx& G^{cl}_{0,m}(i\omega_n) \nonumber\\
396: &=& \left( i\omega_n + \mu -\epsilon_m -
397: \sum_k \frac{\vert V_{mk}\vert^2}{i\omega_n - \epsilon_k}\right)^{-1}
398: \label{G0m}
399: \end{eqnarray}
400: where $\epsilon_m$ represents an impurity level, $\epsilon_k$ the bath
401: levels, and $V_{mk}$ the hybridization matrix elements. The incorporation
402: of the impurity level $\epsilon_m$ ensures a much better fit of
403: $G_{0,m}(i\omega_n)$ than by projecting only onto bath orbitals.
404:
405: For instance, for $n_c=3$ and $n_s=12$ (i.e., three bath levels per impurity
406: orbital), each component
407: $G_{0,m}(i\omega_n)$ is fitted using seven parameters: one impurity level
408: $\epsilon_m$, three bath levels $\epsilon_k$ and three hopping integrals
409: $V_{mk}$. Since according to Eq.~(\ref{g3}) there are two independent functions,
410: we use a total of 14 fit parameters to represent these two $G_{0,m}$ components.
411: This procedure allows for a considerably more flexible projection
412: of the Green's function matrix $G_0(i\omega_n)$ onto the bath. In a site basis
413: for an isotropic triangular lattice (taking again three bath levels per site)
414: one would have instead only 6 fit parameters if each site couples to
415: its own bath. Effectively, therefore, the molecular orbital basis accounts
416: for several additional cross hybridization terms as well as internal cluster
417: couplings (see below). Moreover, it is much more
418: reliable to fit the two independent molecular orbital components
419: $G_{0,m}(i\omega_n)$ than a non-diagonal site matrix $G_{0,ij}(i\omega_n)$
420: with an equivalent number of parameters. Analogous considerations hold
421: for the two-site and four-site clusters of the square lattice.
422: For example, for $n_c=4$, $n_s=12$ there are two independent functions
423: $G_{0,1}$ and $G_{0,2}$ ($G_{0,4}$ is related to $G_{0,1}$), giving a total
424: of 10 fit parameters, compared to only two parameters in a simple site
425: picture with fourforld and particle hole symmetry.
426:
427: \begin{figure}[t] %2
428: \begin{center}
429: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{GnuImG0.nc=3.ps}
430: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{GnuReG0.nc=3.ps}
431: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig2a.ps}
432: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig2b.ps}
433: \end{center}
434: \caption{(Color online)
435: Projection of lattice Green's function components $G_{0,m}(i\omega_n)$
436: onto bath for $n_c=3$, $U=9$, $T=0.02$. Upper panel: Im\,$G_{0,m}$,
437: lower panel: Re\,$G_{0,m}$.
438: Red curves: lattice Green's functions,
439: blue curves: approximate expression, right-hand side of Eq.~(\ref{G0m}).
440: }\end{figure}
441:
442: Figure 2 illustrates the typical quality of the projection of the lattice
443: components of $G_{0,m}(i\omega_n)$ onto the bath for $n_c=3$ and $n_s=9$.
444: Thus, although only two bath levels per orbital are included (i.e., using
445: five parameters per orbital), the fit of both real and imaginary parts is
446: excellent. For these cluster sizes and low temperatures, iterations take
447: only a few minutes. Fits of similar quality are achieved for multi-orbital
448: materials.\cite{prb08,perroni,naco,tetra,v2o3-al} To achieve even better
449: agreement at low frequencies, it is preferable to minimize not the
450: bare difference $G_{0,m}(i\omega_n) - G^{cl}_{0,m}(i\omega_n)$ but to divide
451: these functions first by $\omega_n$.
452:
453: We now discuss the evaluation of the finite temperature interacting cluster
454: Green's function. If this step is carried out in the diagonal molecular
455: orbital basis, the Coulomb interaction must be expressed as a matrix containing
456: many inter-orbital components. For $n_c=4$ it can be easily shown that
457: $U_{m_1m_2m_3m_4}=U/4$ for 64 of the possible 256 configurations.
458: All other matrix elements vanish. This step can be circumvented by working
459: in a mixed basis consisting of cluster sites $i$ and bath orbitals $k$.
460: We illustrate this procedure here for the triangular lattice with $n_c=3$.
461: Let us denote the transformation between sites and orbitals as $T^{(n_c)}$,
462: where
463: \begin{eqnarray}
464: T^{(3)}_{im} &=& \left( \begin{array}{rrr}
465: 1/\sqrt3 & -2/\sqrt6 & 0 \\
466: 1/\sqrt3 & 1/\sqrt6 & 1/\sqrt2 \\
467: 1/\sqrt3 & 1/\sqrt6 &-1/\sqrt2 \\
468: \end{array} \right) .
469: \end{eqnarray}
470: In this mixed basis, the effective site block of the cluster Hamiltonian
471: becomes
472: \begin{eqnarray}
473: h^{(3)} &=& \left( \begin{array}{rrr}
474: \epsilon' & t' & t' \\
475: t' & \epsilon' & t' \\
476: t' & t' & \epsilon' \\
477: \end{array} \right)
478: \end{eqnarray}
479: with $\epsilon'=\epsilon_0 + (\epsilon_1+2\epsilon_2)/3$ and
480: $t'=t + (\epsilon_1-\epsilon_2)/3$, where $\epsilon_0$ and $t$ are the
481: elements of the original cluster Hamiltonian defined in Eq.~(\ref{h3}).
482: The new terms involving the molecular orbital cluster levels
483: $\epsilon_m$ arise from the projection specified in Eq.~(\ref{G0m}).
484: In the mixed basis, the hybridization matrix elements $V_{mk}$
485: between cluster and bath orbitals introduced in Eq.~(\ref{G0m}) are
486: transformed to new hybridization matrix elements between cluster sites
487: $i$ and bath orbitals $k$. They are given by
488: \begin{equation}
489: V'_{ik} = (T^{(3)} V)_{ik} = \sum_m T^{(3)}_{im} V_{mk}\ .
490: \end{equation}
491:
492:
493: Using the elements $\epsilon'$, $t'$ and $V'_{ik}$ together with the
494: on-site Coulomb energy $U$, the non-diagonal interacting cluster Green's
495: function at finite temperature is derived from the expression\cite{perroni,luca}
496: \begin{eqnarray}
497: G^{cl}_{ij}(i\omega_n) &=& \frac{1}{Z} \sum_{\nu\mu}\,e^{-\beta E_\nu}\,
498: \Big(\frac{\langle\nu\vert c_{i\sigma} \vert\mu\rangle
499: \langle\mu\vert c_{j\sigma}^+\vert\nu\rangle}
500: {E_\nu - E_\mu + i\omega_n} \nonumber\\
501: &&\hskip9mm + \ \ \frac{\langle\nu\vert c_{i\sigma}^+\vert\mu\rangle
502: \langle\mu\vert c_{j\sigma} \vert\nu\rangle}
503: {E_\mu - E_\nu + i\omega_n} \Big)
504: \label{Gcl}
505: \end{eqnarray}
506: where $E_\nu$ and $|\nu \rangle$ denote the eigenvalues and eigenvectors of the
507: impurity Hamiltonian, $\beta=1/k_B T$ and $Z=\sum_\nu {\rm exp}(-\beta E_\nu)$
508: is the partition function. At low temperatures only a relatively small number of
509: excited states in few spin sectors contributes to $G^{cl}_{ij}$. They can be
510: efficiently evaluated using the Arnoldi algorithm.\cite{arnoldi} The excited
511: state Green's functions are computed using the Lanczos procedure. Further
512: details can be found in Ref.\cite{perroni}. The non-diagonal elements
513: of $G^{cl}_{ij}$ are derived by first evaluating the diagonal components
514: $G^{cl}_{ii}$ and then using the relation\cite{senechal3}
515: \begin{equation}
516: G^{cl}_{(i+j)(i+j)} = G^{cl}_{ii}+G^{cl}_{ij}+G^{cl}_{ji}+ G^{cl}_{jj}.
517: \end{equation}
518: Since $G^{cl}_{ij}=G^{cl}_{ji}$, this yields:
519: \begin{equation}
520: G^{cl}_{ij} = \frac{1}{2}(G^{cl}_{(i+j)(i+j)} - G^{cl}_{ii} - G^{cl}_{jj}).
521: \end{equation}
522:
523: For the two-site cluster,
524: we have used 3 or 4 bath levels for each impurity orbital ($n_s=8$ or 10),
525: for the three-site cluster 2 or 3 bath levels per impurity orbital
526: ($n_s=9$ or 12), and for the four-site cluster 2 bath levels per impurity
527: orbital ($n_s=12$). $G^{cl}_{ij}(i\omega_n)$ obeys the same symmetry properties
528: as the lattice Green's functions given in Eqs.~(\ref{G2}--\ref{G4}).
529: It therefore can be diagonalized as indicated in Eqs.~(\ref{g2}--\ref{g4}).
530: We denote these diagonal elements as $G^{cl}_{m}(i\omega_n)$. For $n_c=4$,
531: we have checked that the evaluation of $G^{cl}_{m}(i\omega_n)$ in the
532: non-diagonal site - orbital basis and in the diagonal molecular
533: orbital basis yield identical results.
534:
535: The key assumption in DMFT is now that the resulting impurity cluster self-energy
536: is a physically reasonable representation of the lattice self-energy. Thus,
537: using a relation analogous to Eq.~(\ref{G0}), we find
538: \begin{eqnarray}
539: \Sigma^{cl}_{m}(i\omega_n) &=& 1/G^{cl}_{0,m}(i\omega_n)-1/G^{cl}_{m}(i\omega_n)
540: \nonumber\\
541: &\approx& \Sigma_{m}(i\omega_n) .
542: \label{S}
543: \end{eqnarray}
544: After transforming $\Sigma_{m}(i\omega_n)$ back to the non-diagonal site basis,
545: it is used as input in the lattice Green's function Eq.~(\ref{G}) in the
546: next iteration step.
547:
548: To summarize the procedure discussed above, the multi-site ED/DMFT
549: calculation consists of the following steps:\\
550: (a) evaluate the lattice Green's function $G_{ij}(i\omega_n)$, Eq.~(\ref{G}),
551: in the non-diagonal site basis, using as input the self-energy obtained
552: in a previous iteration step. The entire iteration
553: procedure is started at small $U$ with $\Sigma=0$.\\
554: (b) transform $G_{ij}$ and $\Sigma_{ij}$ to the diagonal molecular
555: orbital basis and compute the components $G_{0,m}$. \\
556: (c) project the $G_{0,m}(i\omega_n)$ onto independent baths to determine
557: $\epsilon_m$, $\epsilon_k$ and $V_{mk}$ as indicated in Eq.~(\ref{G0m}).\\
558: (d) from the fit parameters $\epsilon_m$ and $V_{mk}$ determine the
559: Hamiltonian matrix elements $\epsilon'$, $t'$ and $V'_{ik}$
560: in the mixed site - orbital basis. \\
561: (e) evaluate the non-diagonal cluster Green's function
562: $G^{cl}_{ij}(i\omega_n)$ using the Arnoldi and Lanczos methods. \\
563: (f) transform this Green's function to the diagonal orbital basis and
564: compute the cluster self-energy components $\Sigma^{cl}_{m}(i\omega_n)$
565: defined in Eq.~(\ref{S}).\\
566:
567:
568: We emphasize that ED/DMFT involves, at each iteration, two projections:
569: (1) The lattice Green's function $G_0$ is projected onto the cluster
570: Green's function $G_0^{cl}$, as indicated in Eq.~(\ref{G0m}).
571: By definition, $G_0$ has a continuous spectrum at real frequencies,
572: while $G_0^{cl}$ is discrete.
573: (2) The cluster self-energy $\Sigma^{cl}$, which evidently has a
574: discrete spectrum at real $\omega$, is used as an approximation of the
575: lattice self-energy $\Sigma$, which by definition is continuous along
576: the real frequency axis.
577: Thus, both projections, $G_0\approx G_0^{cl}$ and $\Sigma^{cl}\approx\Sigma$,
578: rely on the well-known fact that continuous and discrete spectra at real
579: $\omega$ can yield nearly identical distributions at Matsubara frequencies.
580: Since the cluster size determines the number of discrete spectral features
581: of $G_0^{cl}$ and $\Sigma^{cl}$, there exists evidently an infinite number
582: of discrete spectra which may in principle be used to represent the
583: continuous spectra of the lattice quantities $G_0$ and $\Sigma$.
584:
585: \section{Results and Discussion}
586:
587: \begin{figure}[t] %3
588: \begin{center}
589: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{DCA.occ.nc=2.ps}
590: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{DCA.occ.nc=4.ps}
591: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{DCA.occ.nc=3.ps}
592: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig3a.ps}
593: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig3b.ps}
594: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig3c.ps}
595: \end{center}
596: \caption{(Color online)
597: Occupancies of cluster molecular orbitals (per spin) for two- and four-site
598: clusters of square lattice (top panels), and of three-site cluster of
599: triangular lattice (bottom panel) as functions of Coulomb energy $U$.
600: The temperature is $T=0.02$. The sum of these molecular occupancies is
601: $n_c/2$ and their average is $0.5$.
602: The Mott transition for increasing $U$ occurs at
603: $U_{c2}\approx 5.5$ for the square lattice ($n_c=2$ and $n_c=4$), and near
604: $U_{c2}\approx 9.5$ for the triangular lattice ($n_c=3$).
605: }\end{figure}
606:
607: Figure~3 shows the occupancies of the cluster molecular orbitals for three
608: cluster sizes as functions of increasing $U$. For the square lattice
609: the Mott transition occurs near $U_{c2}\approx 5.5$,
610: while for the triangular lattice $U_{c2}\approx 9.5$ (see below).
611: Evidently all orbital occupancies vary smoothly across the transition.
612: There is no indication of orbital selective Mott transitions, nor for
613: complete filling or emptying of any orbitals at large Coulomb energies.
614: This behavior differs qualitatively from the one found
615: in materials such as LaTiO$_3$, V$_2$O$_3$, and Ca$_2$RuO$_4$, where
616: the insulating phase exhibits nearly complete orbital polarization as
617: a result of a Coulomb driven enhancement of the crystal field splitting
618: between $t_{2g}$ orbitals\cite{pavarini,prb08,keller,poteryaev,anisimov,prl07}
619: (see following section).
620:
621: \begin{figure}[t] %4
622: \begin{center}
623: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{A.nc=4.U=56.ps}
624: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{A.nc=4.U=5.ps}
625: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{A.nc=4.U=6.ps}
626: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig4a.ps}
627: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig4b.ps}
628: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig4c.ps}
629: \end{center}
630: \caption{(Color online)
631: Upper panel: total spectral distribution below and above the Mott transition
632: ($U_c\approx 5.5$) for the square lattice with $n_c=4$ at temperature
633: $T=0.02$: $U=5$ (red curve) and $U=6$ (blue curve).
634: Lower two panels: molecular orbital contributions at $U=5$ and $U=6$;
635: blue curves: $A_1(\omega)$ (associated with $\Gamma$),
636: green curves: $A_4(\omega)$ ($M$),
637: red curves: $A_{2,3}(\omega)$ ($X$) (broadening $\delta=0.1$).
638: }\end{figure}
639:
640: %\begin{figure}[t] %5
641: %\begin{center}
642: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{A.nc=4.U=56r.ps}
643: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{A.nc=4.U=5r.ps}
644: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{A.nc=4.U=6r.ps}
645: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig5ax.ps}
646: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig5bx.ps}
647: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig5c.ps}
648: %\end{center}
649: %\caption{(Color online)
650: %Same as Figure~4, except that the spectra are derived via extrapolation
651: %of $G_m(i\omega_n)$ to the real frequency axis (broadening $\delta=0.1$).
652: %}\end{figure}
653:
654: The fact that all cluster orbitals remain partially occupied across the
655: transition implies that the gap opens simultaneously in all orbitals.
656: This can be seen most clearly in the spectral distributions, as shown in
657: Figure~4 for the square lattice with $n_c=4$. Since we are here concerned
658: with the transition from metallic to insulating behavior, we show the
659: spectra obtained from the interacting cluster Green's function,
660: $A_{ij}(\omega)=-(1/\pi)\,{\rm Im}\, G^{cl}_{ij}(\omega+i\delta)$,
661: Eq.~(\ref{Gcl}), with $\delta=0.1$. These spectra can be evaluated without
662: requiring analytic continuation from Matsubara to real frequencies.
663: Using the transformations indicated in Eqs.~(\ref{g4}) and (\ref{rho4}),
664: the cluster molecular orbital densities are given by
665: $A_1 =A_{11}+2A_{12}+A_{14}$,
666: $A_2=A_3=A_{11} -A_{14}$ and
667: $A_4 =A_{11}-2A_{12}+A_{14}$.
668: The total density $A(\omega)=\sum_m A_m(\omega)/4$ coincides with the on-site
669: distribution $A_{11}(\omega)$, which also represents the density corresponding
670: to $M/2=(\pi/2,\pi/2)$ (see below). To our knowledge, this orbital decomposition
671: has not been addressed before.
672: Clearly, all orbitals contribute to the spectral weight at $E_F$ in the metallic
673: phase, as well as to the lower and upper Hubbard bands in the insulating phase.
674: The spectral distributions shown in the upper panel are consistent with those
675: by Kyung et al.\cite{kyung2} and Zhang and Imada\cite{zhang} within ED/DMFT at
676: $T=0$. The spectra reveal a characteristic four-peak structure, consisting of
677: low-frequency peaks limiting the pseudogap due to short-range correlations and
678: high-frequency peaks associated with the Hubbard bands.\cite{preuss,moreo,kyung2}
679: The small peak at $E_F$ in the metallic phase at $U=5$ appears only at finite $T$.
680: It vanishes at $T=0$. At such low frequencies, however, ED finite-size effects
681: cannot be ruled out.
682:
683: %Figure~5 shows the same spectra as in Figure~4, except that the routine
684: %{\it ratint}\cite{ratint} was used to extrapolate the cluster molecular orbital
685: %components $G_m(i\omega_n)$ to real frequencies. Since this extrapolation is
686: %ill-posed, there is appreciable uncertainty at high frequencies.
687: %Nevertheless, the low $\omega$ variation of the total spectral distribution
688: %shown in the upper panel is consistent with analogous results obtained
689: %via the continuous time quantum Monte Carlo method and the maximum entropy
690: %scheme by Park et al.\cite{park} ($T=0.01$) and Gull et al.\cite{gull} ($T\ge0.1$).
691: %As in the cluster spectra shown in Fig.~4, the insulating phase exhibits a
692: %four-peak structure, and all orbitals contribute to the upper and lower
693: %Hubbard bands. Thus, there is no orbital selective band filling or emptying
694: %at the Mott transition.
695:
696: \begin{figure}[t] %6
697: \begin{center}
698: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Gnur9.0.ps}
699: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Gnur9.6.ps}
700: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig5a.ps}
701: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig5b.ps}
702: \end{center}
703: \caption{(Color online)
704: Spectral distributions of cluster molecular orbitals below and above
705: the Mott transition ($U_c\approx 9.5$) for the triangular lattice
706: with $n_c=3$ at temperature $T=0.02$.
707: Upper panel: $U=9.0$; lower panel: $U=9.6$.
708: Blue curves: $A_1(\omega)$, red curves: $A_{2,3}(\omega)$.
709: }\end{figure}
710:
711: Similar results are obtained for the square lattice in the two-site cluster
712: model. Results for the triangular lattice with $n_c=3$ are shown in Figure~5.
713: According to Eqs.~(\ref{g3}) and (\ref{rho3}) the cluster molecular orbital
714: densities are given by $A_{1}=A_{11}+2A_{12}$ and $A_2=A_3=A_{11}-A_{12}$.
715: As for the square lattice, the Mott gap opens simultaneously in all orbitals
716: at about the same critical $U$, and all orbitals contribute to the lower and
717: upper Hubbard bands.
718: While the metallic phase of the unfrustrated square lattice close to the
719: transition exhibits a pseudogap due to short-range antiferromagnetic
720: correlations,\cite{kyung2,zhang,park,gull} this phenomenon is absent in the
721: triangular lattice as a result of geometrical frustration.\cite{kyung3}
722:
723: \begin{figure}[t] %7
724: \begin{center}
725: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{A.beta=50.ps}
726: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Gnudocc.ps}
727: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig6a.ps}
728: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig6b.ps}
729: \end{center}
730: \caption{(Color online)
731: Hysteresis behavior of spectral weights $A_m(0)$ of cluster molecular orbitals
732: at $E_F=0$ (upper panel) and average double occupancy (lower panel) as functions
733: of Coulomb energy for triangular lattice with $n_c=3$ ($T=0.02$).
734: Red curves: increasing $U$, blue curves: decreasing $U$.
735: }\end{figure}
736:
737: To illustrate the first-order nature of the metal insulator transition we show
738: in the upper panel of Figure~6 the spectral weight of the $n_c=3$ cluster orbitals
739: at $E_F=0$ as a function of $U$. The lower panel shows the average double occupancy
740: $d_{\rm occ}=\sum_m \langle n_{m\uparrow}n_{m\downarrow}\rangle/3$.
741: Both quantities exhibit hysteresis for increasing and decreasing $U$, indicating
742: the coexistence of metallic and insulating solutions.
743: %To our knowledge, the finite temperature hysteresis for the triangular lattice
744: %with $n_c=3$ has not been identified before.
745: The complete $T / U$ phase diagram will be published elsewhere.
746: The phase diagram for the isotropic square lattice with $n_c=4$ was recently
747: mapped out in detail by Park et al.\cite{park}
748:
749: The cluster molecular orbital components of the self-energy and Green's function
750: may be used to derive an approximate expression for the momentum variation of the
751: lattice self-energy and Green's function in the original Brillouin Zone:\cite{parcollet}
752: \begin{equation}
753: \Sigma(\vec k,i\omega_n) \approx \frac{1}{n_c}
754: \sum_{ij} e^{i\vec k\cdot(\vec R_i-\vec R_j)} \Sigma_{ij}(i\omega_n)
755: \label{Slattice}
756: \end{equation}
757: where $\vec R_i$ are the cluster site positions and $\Sigma_{ij}$ are the site
758: components of the self-energy. An analogous expression holds for the lattice Green's
759: function. For the clusters discussed above this superposition implies that the Mott
760: gap opens uniformly along the Fermi surface since all orbitals undergo a common
761: transition.
762:
763: Writing the site elements in terms of the orbital components, one has for $n_c=4$:
764: $\Sigma(\Gamma,i\omega_n)=\Sigma_1(i\omega_n)$,
765: $\Sigma(X ,i\omega_n)=\Sigma_2(i\omega_n)=\Sigma_3(i\omega_n)$ and
766: $\Sigma(M ,i\omega_n)=\Sigma_4(i\omega_n)$.
767: In agreement with results of previous authors
768: \cite{parcollet,civelli,zhang,park,gull,ferrero,balzer}
769: we find the behavior of $\Sigma(\Gamma,i\omega_n)$ and $\Sigma(M,i\omega_n)$
770: near the Mott transition to differ qualitatively from that of $\Sigma(X,i\omega_n)$
771: (not shown here): Whereas Im\,$\Sigma(X,i\omega_n)$ exhibits $\sim\omega_n$
772: variation at low frequencies in the metallic phase and $\sim 1/\omega_n$ variation
773: in the insulating phase (the real part vanishes because of particle hole symmetry,
774: see Fig.~1), Im\,$\Sigma(\Gamma,i\omega_n)$ and Im\,$\Sigma(M,i\omega_n)$ remain
775: $\sim\omega_n$ in both phases, but their real parts increase rapidly across the
776: transition.
777:
778: Although this behavior might suggest a Mott transition for the $X$ cluster orbitals
779: combined with a band-filling or band-emptying mechanism for the $\Gamma$ and $M$
780: orbitals, the orbital occupancies (Fig.~3) and spectral distributions (Fig.~4)
781: demonstrate this not to be the case. Moreover, for the crucial question of whether
782: or not the Mott gap opens simultaneously across the Fermi surface, it is important
783: to compare the self-energy at $X$ with its behavior at $M/2$, where the electron
784: band also crosses $E_F$. According to Eq.~(\ref{Slattice}),
785: $\Sigma(M/2,i\omega_n)$ coincides with the diagonal on-site element of the cluster
786: self-energy, which is identical with the local lattice self-energy. Thus:
787: \begin{eqnarray}
788: \Sigma(M/2,i\omega_n) &=& \frac{1}{4}[\Sigma(\Gamma,i\omega_n)+\Sigma(M,i\omega_n)
789: \nonumber\\
790: && \ \ \ + \ 2\, \Sigma(X,i\omega_n)].
791: \end{eqnarray}
792: The real parts of the first terms on the {\it rhs} cancel since the corresponding
793: density of states components are mirrors of each other (see Fig.~1). Thus,
794: $\Sigma(M/2,i\omega_n)$
795: and $\Sigma(X,i\omega_n)$ are purely imaginary because of particle hole symmetry.
796: The above relation demonstrates that Coulomb correlations at the cold spot $M/2$
797: are essentially driven by those at the hot spot $X$. In fact, the magnitude of
798: the self-energy at $M/2$ is a factor of 2 smaller than the singular term at $X$,
799: with weak additional, non-singular contributions associated with $\Gamma$ and $M$.
800:
801: Along the Fermi surface between $X$ and $M/2$ (i.e. for $k_x+k_y=\pi$)
802: the self-energy is given by
803: \begin{equation}
804: \Sigma(\vec k,i\omega_n) = {\rm cos}^2(k_x)\Sigma(X,i\omega_n)
805: +{\rm sin}^2(k_x)\Sigma(M/2,i\omega_n). \\
806: \label{XtoM/2}
807: \end{equation}
808: This function is imaginary, i.e., there are no band shifts due to a finite real
809: part of the self-energy. Thus, within the four-site cluster DMFT, the opening
810: of the Mott gap on the entire Fermi surface is determined solely by the singularity
811: of Im\,$\Sigma(X,i\omega_n)$.
812:
813: The above analysis leads to a surprisingly simple picture for the momentum
814: variation of correlations along the Fermi surface. It consists of two sinusoidal
815: contributions: The singular $X$ term oscillates with amplitude $1$ at $X$ and
816: $1/2$ at $M/2$, and the non-singular term due to the $\Gamma,\ M$ orbitals
817: oscillates with amplitude $1$ at $M/2$ and zero at $X$.
818:
819: \begin{figure}[t] %8
820: \begin{center}
821: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Gnus.ps}
822: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Gnug.ps}
823: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig7a.ps}
824: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig7b.ps}
825: \end{center}
826: \caption{(Color online)
827: Lattice self-energy, Eq.~(\ref{Slattice}), (upper panel) and Green's function
828: (lower panel) at $X=(\pi,0)$ (red curves) and $M/2=(\pi/2,\pi/2)$ (blue curves)
829: for metallic phase ($U=5$) and insulation phase ($U=6$) of square lattice
830: ($n_c=4$) at $T=0.02$. Both $\Sigma(\vec k,i\omega_n)$ and $G(\vec k,i\omega_n)$
831: are purely imaginary between $X$ and $M/2$. The sinusoidal variation of the
832: self-energy between these points is given by Eq.~(\ref{XtoM/2}).
833: }\end{figure}
834:
835: \begin{figure}[t] %9
836: \begin{center}
837: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Gnug5.0.ps}
838: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Gnug6.0.ps}
839: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig8a.ps}
840: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig8b.ps}
841: \end{center}
842: \caption{(Color online)
843: Lattice Green's function Eq.~(2) in $n_c=4$ cluster molecular orbital basis
844: for $U=5$ (upper panel) and $U=6$ (lower panel) at $T=0.02$.
845: Red curves: diagonal version of Eq.~(\ref{G});
846: blue curves: approximate expression, Eq.~(\ref{Grho}).
847: }\end{figure}
848:
849: This is illustrated in Figure 7 which shows the lattice self-energy and
850: Green's function at $X$ and $M/2$ below and above the Mott transition.
851: In the metallic phase, the self-energy at $X$ is seen to be
852: larger than the one at $M/2$. Thus, quasi-particle lifetimes decrease between
853: $M/2$ and $X$. Nevertheless, the singular behavior at $M/2$ is governed
854: by the one at $X$. Accordingly, the Green's function at $X$ and $M/2$ displays
855: a change from metallic to insulating behavior at the same Coulomb energy.
856:
857: To understand the strong coupling between different sections of the
858: Brillouin Zone it is important to recall that, in the diagonal cluster
859: molecular orbital basis, the single-particle part of the lattice Hamiltonian
860: appearing in Eq.~(\ref{G}) is not diagonal. Thus, the orbital component of the
861: lattice Green's function
862: $G_m(i\omega_n)$ is influenced not only by the corresponding self-energy
863: $\Sigma_m(i\omega_n)$, but by the other orbital elements as well. This point
864: becomes clear if we compare $G_m(i\omega_n)$ with the approximation
865: \begin{equation}
866: G_m(i\omega_n) \approx \int d\omega\frac{\rho_m(\omega)}
867: { i\omega_n + \mu - \omega - \Sigma_m(i\omega_n)}.
868: \label{Grho}
869: \end{equation}
870: Figure 8 shows that in the metallic region at $U=5$ there is little difference
871: with regard to the actual $G_m(i\omega_n)$. Also, at $U=6$ the key components
872: responsible for the metal insulator transition, namely $G_{2,3}(i\omega_n)$
873: corresponding to $X=(\pi,0)$, are well represented by this approximation.
874: The $\Gamma,M$ components, $G_{1,4}(i\omega_n)$, however, do not reveal
875: insulating behavior since at small $\omega_n$ the imaginary parts do not
876: extrapolate to zero. This demonstrates that the Mott gaps seen in
877: $A_{1,4}(\omega)$ in Figs.~4 and 5 are not caused by the rapidly varying
878: real parts of $\Sigma_{1,4}(i\omega_n)$. Instead, the gaps at $\Gamma$ and
879: $M$ are driven by the singular behavior of Im\,$\Sigma_{2,3}(i\omega_n)$
880: which contributes to $A_{1,4}(\omega)$ via the non-diagonal elements of
881: $t(\vec k)$.\cite{kmixing}
882:
883: If the approximate components obtained via Eq.~(\ref{Grho}) were used
884: to generate the spectral distributions $A_m(\omega)$, it is clear that
885: only $A_{2,3}(\omega)$ corresponding to $X$ would exhibit the Mott transition,
886: while $A_{1,4}(\omega)$ would retain considerable metallicity. This implies
887: that the physics at the cold spot $M/2$ is incorrectly represented
888: via Eq.~(\ref{Grho}), suggesting that the gap at $M/2$ does not open at
889: the same $U$ as at $X$. Instead, as argued above, the metal insulator
890: transition at $M/2$ is caused by the same self-energy terms as at $X$,
891: i.e., the Mott gap opens uniformly.
892:
893: As pointed out earlier, in the molecular orbital basis the Coulomb matrix
894: has a large number of non-zero elements. Thus, in this basis there is not
895: only single-particle hybridization arising from $t(\vec k)$, but also
896: strong inter-orbital Coulomb repulsion. Nevertheless, the above analysis
897: reveals that, although these Coulomb interaction terms are properly taken
898: into account, the spectral distributions at $\Gamma$, $M$ and $M/2$ do not
899: exhibit a Mott gap unless the non-diagonal terms of $t(\vec k)$ in the
900: orbital basis are included. It is therefore the single-particle part of
901: the Hamiltonian that provides the correct connection between the self-energy
902: components and thereby generates the true momentum variation of the spectral
903: distribution of the single band.
904:
905: Because of the strong coupling between orbitals, the notion that some of
906: these orbitals undergo a Mott transition while others do not, does not appear
907: appropriate. As shown consistently by all orbital-resolved spectra in the
908: present multi-site ED/DMFT study, there is a single Mott transition common
909: to all cluster molecular orbitals, implying a simultaneous opening of the
910: Mott gap across the entire Fermi-surface. Nevertheless, in agreement with
911: previous authors, we find pronounced momentum variation of quasi-particle
912: properties close to the Mott transition.
913:
914: In view of the approximate nature of the momentum variation of the lattice
915: self-energy and Green's function derived within the CDMFT it would be very
916: interesting to compare the above results with analogous ones obtained within
917: the DCA,\cite{hettler,maier} in particular, since the cluster molecular orbital
918: components of the density of states, as stated above, differ appreciably between
919: these two cluster DMFT schemes.\cite{haule} This comparison will be addressed
920: in a future publication.
921:
922: The scenario discussed in this section differs strikingly from the one found
923: for several multi-orbital materials which have been studied previously by
924: various groups. For the sake of comparison we review some of these systems
925: in the following section.
926:
927: \section{Comparison with multi-orbital systems}
928:
929: \begin{figure}[t]
930: \begin{center}
931: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{LaTiOn1.ps}
932: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{V2O3nvsU.ps}
933: %\includegraphics[width=4.5cm,height=6.5cm,angle=-90]{CROnvsU.ps}
934: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig9a.ps}
935: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig9b.ps}
936: \includegraphics[width=4.5cm,height=6.5cm,angle=-90]{Fig9c.ps}
937: \end{center}
938: \caption{(Color online)
939: Subband occupancies obtained within finite temperature multi-band ED/DMFT
940: as functions of $U$.
941: Upper panels: LaTiO$_3$ ($3d^1$ configuration) for full Hund exchange with
942: $J=0.65$~eV\cite{prb08};
943: V$_2$O$_3$ ($3d^2$ configuration) for full Hund exchange
944: (blue dots) and Ising-like exchange (red solid dots) with $J=0.7$~eV\cite{v2o3-al};
945: red empty dots: quantum Monte Carlo DMFT results.\cite{keller}
946: Lower panel: Ca$_2$RuO$_4$ ($4d^4$ configuration) for full Hund exchange
947: with $J=U/4$ and $\Delta=0.4$~eV $t_{2g}$ crystal field splitting.\cite{prl07}
948: }\end{figure}
949:
950: During the recent years single-site multi-orbital DMFT has been used
951: extensively to investigate the metal insulator transition of a variety
952: of materials.\cite{kotliar06} Here, we briefly discuss some of these
953: systems which are regarded as typical Mott insulators, and which all
954: exhibit characteristic changes of the electronic structure as the metallic
955: phase is replaced by the insulator at large Coulomb energies.
956:
957: Figure~9 shows the correlation driven enhancement of orbital polarization
958: for LaTiO$_3$, V$_2$O$_3$ and Ca$_2$RuO$_4$.
959: Because of the orthorhombic structure of LaTiO$_3$, LDA calculations
960: reveal that the $a_g$ subbands of the $t_{2g}$ sector are slightly more
961: occupied than the two $e'_g$ components.\cite{pavarini} Local Coulomb
962: interactions enhance this $t_{2g}$ crystal field splitting,
963: so that at the Mott transition close to $U=5$~eV the $e'_g$ bands become
964: nearly empty and the $a_g$ band half-filled.\cite{pavarini,prb08}
965: In the case of V$_2$O$_3$, the corundum lattice structure ensures that
966: the doubly degenerate $e'_g$ bands have slightly larger binding energy
967: than the $a_g$ bands.\cite{keller} With increasing Coulomb interaction
968: this crystal field splitting is strongly enhanced, until in the range
969: $U\approx 5\ldots6$~eV, the $a_g$ bands are pushed above the Fermi level
970: and the $e'_g$ bands become half-filled.\cite{keller,poteryaev,v2o3-al}
971: %
972: %For Ising-like exchange, nearly perfect
973: %agreement is found between the orbital polarization obtained within
974: %ED/DMFT\cite{v2o3-al} and QMC/DMFT.\cite{keller} (Similar QMC results
975: %were found by Poteryaev et al.\cite{poteryaev}) For full Hund exchange,
976: %the physical picture is the same as for Ising-like exchange, except for
977: %a slightly larger critical Coulomb energy.
978: %
979: Finally, in the case of Ca$_x$Sr$_{2-x}$RuO$_4$, Sr substitution via the
980: smaller Ca ions gives rise to an enlarged crystal field splitting between
981: $d_{xy}$ and $d_{xz,yz}$ like subbands.\cite{fang} Coulomb correlations
982: increase this splitting, until in the Mott phase the $d_{xy}$ bands are
983: fully occupied and the half-filled $d_{xz,yz}$ bands are split into lower
984: and upper Hubbard bands.\cite{prl07}
985:
986: \begin{figure}[t] %11
987: \begin{center}
988: %\includegraphics[width=7.7cm,height=7.5cm,angle=-90]{mott1.ps}
989: \includegraphics[width=5.7cm,height=7.5cm,angle=-90]{Fig10.ps}
990: \end{center}
991: \vskip-17mm
992: \caption{(Color online)
993: Schematic illustration of correlation driven enhancement of orbital
994: polarization.
995: Upper row: crystal field split $t_{2g}$ LDA densities of states for
996: occupancies $n=1$, 2, and 4, corresponding to LaTiO$_3$, V$_2$O$_3$
997: and Ca$_2$RuO$_4$, respectively. Blue curves: singly-degenerate $a_g$
998: band, red curves: doubly-degenerate $e'_g$ bands. In the case of
999: Ca$_2$RuO$_4$, $a$ refers to $d_{xy}$, $e$ to $d_{xz,yz}$.
1000: The vertical bars denote the Fermi level.
1001: Lower row: orbitally polarized Mott phase.
1002: $n=1$: empty $e'_g$ bands,
1003: lower and upper Hubbard peaks of half-filled $a_g$ band;
1004: $n=2$: empty $a_g$ band,
1005: lower and upper Hubbard peaks of half-filled $e'_g$ bands;
1006: $n=4$: filled $a_g$ band,
1007: lower and upper Hubbard peaks of half-filled $e'_g$ bands.
1008: }\end{figure}
1009:
1010: Schematically, the uncorrelated densities of states of these transition
1011: metal oxides and the spectra derived within single-site multi-orbital DMFT for
1012: realistic Coulomb energies are shown in Figure~10. Note that in the metallic
1013: phase, both orbital symmetries contribute to the spectral weight at the
1014: Fermi level. In the Mott phase, the gap involes transitions between states
1015: of opposite symmetry character.
1016:
1017: Despite the different subband occupancies of these materials, they exhibit
1018: a similar correlation driven enhancement of orbital polarization.\cite{others}
1019: By pushing some subbands above or below the Fermi level, the effective degeneracy
1020: is reduced from three to two or one. The Mott transition therefore occurs at
1021: lower critical Coulomb energy than in a cubic environment with equivalent subbands.
1022:
1023: According to Figure~1, the cluster molecular orbital densities of the
1024: single band Hubbard model in a multi-site picture also exhibit a
1025: substantial splitting relative to the total band width. Nevertheless,
1026: Coulomb correlations in these cases only lead to moderate charge transfer
1027: between these cluster orbitals, as demonstrated by the results given in
1028: Fig.~2. The main physical reason for this qualitative difference with
1029: respect to the multi-orbital materials is the strong single-particle
1030: hybridization among orbitals so that any tendency towards orbital selective
1031: Mott transitions is suppressed. For the same reason, partial band filling
1032: or emptying, with a Mott transition in the remaining subset of bands, as
1033: found in several multi-orbital materials, is also absent.
1034:
1035: A certain amount of inter-orbital hybridization exists also in multi-orbital
1036: systems since the single-electron Hamiltonian in the orbital basis is not
1037: diagonal throughout the Brillouin Zone. This residual coupling, however,
1038: is much weaker than in the single-band multi-site system, so that Coulomb
1039: correlations can indeed lead to nearly complete orbital polarization. This is
1040: supported by the DMFT results for LaTiO$_3$ and V$_2$O$_3$ which have been
1041: studied both by evaluating the lattice Green's function via Eq.~(\ref{G}) (see
1042: Refs.\cite{pavarini,poteryaev}) and via the approximate version, Eq.~(\ref{Grho})
1043: (see Refs.\cite{prb08,keller,v2o3-al}). In these systems both formulations give
1044: very similar results, in particular, both confirm the scenario of strong orbital
1045: polarization.
1046:
1047: \bigskip
1048:
1049: \section{Conclusion}
1050:
1051: Cellular DMFT combined with
1052: finite temperature exact diagonalization has been used to investigate the
1053: influence of short range correlations on the Mott transition in the single-band
1054: Hubbard model. Both square and triangular lattices at half-filling were studied.
1055: A mixed basis consisting of cluster sites and bath molecular orbitals was shown
1056: to provide an efficient method for the evaluation of the cluster self-energies
1057: and Green's functions. Since in the cluster molecular orbital representation
1058: these quantities become diagonal, an intriguing analogy exists between Coulomb
1059: correlations in these multi-site single-band systems and several multi-orbital
1060: materials which were studied previously within single-site DMFT.
1061:
1062: In remarkable
1063: contrast to LaTiO$_3$, V$_2$O$_3$, and Ca$_2$RuO$_4$, which exhibit pronounced
1064: orbital polarization at the Mott transition, the single-band systems
1065: show very little correlation driven enhancement of orbital polarization.
1066: Thus, all cluster molecular orbitals take part in the metal insulator transition.
1067: Moreover, the transition occurs at the same critical $U$ for all cluster orbitals.
1068: Since an approximate momentum variation of the lattice self-energy and Green's
1069: function can be constructed from a superposition of these molecular orbital
1070: components, this finding yields the important result that the Mott gap opens
1071: simultaneously across the entire Fermi surface. Thus, for both square and
1072: triangular lattices at half filling, there is no orbital selective Mott
1073: transition, where certain sections of the Brillouin Zone would open a gap
1074: at lower Coulomb energy than other parts. Moreover, there is no evidence for
1075: the combination of subband-filling and Mott transition in other subbands,
1076: that is characteristic of the multi-orbital materials mentioned above.
1077: It would be of great interest to investigate whether these findings also
1078: hold at a finer momentum resolution which would require cluster sizes larger
1079: than $n_c=4$.
1080:
1081: \bigskip
1082:
1083: {\bf Acknowledgements}\ \
1084: We like to thank A. Georges, N. Kawakami, G. Kotliar, R.H. McKenzie, M. Potthoff,
1085: B.J. Powell and H. Tsunetsugu for useful discussions. The computational work was
1086: carried out on the J\"ulich JUMP computer.
1087:
1088: \bigskip
1089:
1090:
1091: \begin{thebibliography}{99}
1092:
1093: \bibitem{kotliar06}
1094: For recent reviews, see:
1095: K. Held, Adv. in Physics, {\bf 56}, 829 (2007);
1096: G. Kotliar, S.Y. Savrasov, K. Haule, V.S. Oudovenko, O. Parcollet,
1097: and C.A. Marianetti,
1098: Rev. Mod. Phys. {\bf 78}, 865 (2006).
1099:
1100: \bibitem{dmft}
1101: A. Georges, G. Kotliar, W. Krauth and M.J. Rozenberg,
1102: Rev. Mod. Phys. {\bf 68}, 13 (1996).
1103:
1104: \bibitem{pavarini}
1105: E. Pavarini, S. Biermann, A. Poteryaev, A.I. Lichtenstein, A. Georges,
1106: and O. K. Andersen,
1107: Phys. Rev. Lett. {\bf 92}, 176403 (2004).
1108:
1109: \bibitem{prb08}
1110: A. Liebsch,
1111: Phys. Rev. B {\bf 77}, 115115 (2008).
1112:
1113: \bibitem{keller}
1114: G. Keller, K. Held, V. Eyert, D. Vollhardt, and V.I. Anisimov,
1115: Phys. Rev. B {\bf 70}, 205116 (2004).
1116:
1117: \bibitem{poteryaev}
1118: A.I. Poteryaev, J.M. Tomczak, S. Biermann, A. Georges, A.I. Lichtenstein,
1119: A.N. Rubtsov, T. Saha-Dasgupta, and O.K. Andersen,
1120: Phys. Rev. B {\bf 76}, 085127 (2007).
1121:
1122: \bibitem{anisimov}
1123: V.I. Anisimov, I.A. Nekrasov, D.E. Kondakov, T.M. Rice and M. Sigrist,
1124: Eur. Phys. J. B {\bf 25}, 191 (2002).
1125:
1126: \bibitem{prl07}
1127: A. Liebsch and H. Ishida,
1128: Phys. Rev. Lett. {\bf 98}, 216403 (2007).
1129:
1130: \bibitem{lechermann}
1131: F. Lechermann, S. Biermann, and A. Georges,
1132: Phys. Rev. Lett. {\bf 94}, 166402 (2005).
1133:
1134: \bibitem{ray}
1135: M. De Raychaudhury, E. Pavarini, and O.K. Andersen,
1136: Phys. Rev. Lett. {\bf 99}, 126402 (2007).
1137:
1138: \bibitem{tetra}
1139: H. Ishida and A. Liebsch,
1140: Phys. Rev. B {\bf 77}, 115350 (2008).
1141:
1142: \bibitem{koga}
1143: A. Koga, N. Kawakami, T.M. Rice, and M. Sigrist,
1144: Phys. Rev. Lett. {\bf 92}, 216402 (2004);
1145: Physica B {\bf 359-361}, 1366 (2005).
1146:
1147: \bibitem{prb04}
1148: A. Liebsch,
1149: Phys. Rev. B {\bf 70}, 165103 (2004);
1150: Phys. Rev. Lett. {\bf 95}, 116402 (2005).
1151:
1152: \bibitem{song}
1153: Yun Song and L.-J. Zou,
1154: Phys. Rev. B {\bf 72}, 085114 (2005).
1155:
1156: \bibitem{inaba}
1157: K. Inaba, A. Koga, S. Suga and N. Kawakami,
1158: J. Phys. Soc. Jpn. {\bf 74}, 2393 (2005).
1159:
1160: \bibitem{costi}
1161: T.A. Costi and A. Liebsch,
1162: Phys. Rev. Lett. {\bf 99}, 236404 (2007).
1163:
1164: \bibitem{tohyama}
1165: T. Tohyama and S. Maekawa,
1166: Phys. Rev. B {\bf 49}, 3596 (1994).
1167:
1168: \bibitem{preuss}
1169: R. Preuss, W. Hanke and W. von der Linden,
1170: Phys. Rev. Lett. {\bf 75}, 1344 (1995).
1171:
1172: \bibitem{moreo}
1173: A. Moreo, S. Haas, A.W. Sandvik, and E. Dagotto,
1174: Phys. Rev. B {\bf 51}, 12045 (1995).
1175:
1176: \bibitem{hettler}
1177: M.H. Hettler, A.N. Tahvildar-Zadeh, M. Jarrell, T. Pruschke, and
1178: H.R. Krishnamurthy, Phys. Rev. B {\bf58}, R7475 (1998).
1179:
1180: \bibitem{senechal}
1181: D. Senechal, D. Perez, and M. Pioro-Ladriere,
1182: Phys. Rev. Lett. {\bf 84}, 522 (2000).
1183:
1184: \bibitem{lich}
1185: A.I. Lichtenstein and M.I. Katsnelson,
1186: Phys. Rev. B {\bf 62}, R9283 (2000).
1187:
1188: \bibitem{huscroft}
1189: C. Huscroft, M. Jarrell, Th. Maier, S. Moukouri, and A.N. Tahvildarzadeh,
1190: Phys. Rev. Lett. {\bf 86}, 139 (2001).
1191:
1192: \bibitem{moukouri}
1193: S. Moukouri and M. Jarrell,
1194: Phys. Rev. Lett. {\bf 87}, 167010 (2001).
1195:
1196: \bibitem{kotliar01}
1197: G. Kotliar, S.Y. Savrasov, G. Palsson, and G. Biroli,
1198: Phys. Rev. Lett. {\bf 87}, 186401 (2001).
1199:
1200: \bibitem{imai}
1201: Y. Imai and N. Kawakami,
1202: Phys. Rev. B {\bf 65}, 233103 (2002).
1203:
1204: \bibitem{maier2}
1205: Th.A. Maier, Th. Pruschke, and M. Jarrell,
1206: Phys. Rev. B {\bf 66}, 226402 (2002).
1207:
1208: \bibitem{onoda}
1209: S. Onoda and M. Imada,
1210: Phys. Rev. B {\bf 67}, 161102 (2003).
1211:
1212: \bibitem{kyung}
1213: B. Kyung, J.S. Landry, D. Poulin, and A.-M.S. Tremblay,
1214: Phys. Rev. Lett. {\bf 90}, 099702 (2003).
1215:
1216: \bibitem{potthoff}
1217: M. Potthoff, M. Aichhorn, and C. Dahnken,
1218: Phys. Rev. Lett. {\bf 91}, 206402 (2003).
1219:
1220: \bibitem{senechal1}
1221: D. Senechal and A.-M.S. Tremblay,
1222: Phys. Rev. Lett. {\bf 92}, 126401 (2004).
1223:
1224: \bibitem{parcollet}
1225: O. Parcollet, G. Biroli, and G. Kotliar,
1226: Phys. Rev. Lett. {\bf 92}, 226402 (2004).
1227:
1228: \bibitem{capone}
1229: M. Capone, M. Civelli, S.S. Kancharla, C. Castellani, and G. Kotliar,
1230: Phys. Rev. B {\bf 69}, 195105 (2004).
1231:
1232: \bibitem{senechal2}
1233: D. Senechal, P.-L. Lavertu, M.-A. Marois, and A.-M. S. Tremblay,
1234: Phys. Rev. Lett. {\bf 94}, 156404 (2005).
1235:
1236: \bibitem{civelli}
1237: M. Civelli, M. Capone, S.S. Kancharla, O. Parcollet, and G. Kotliar,
1238: Phys. Rev. Lett. {\bf 95}, 106402 (2005).
1239:
1240: \bibitem{maier}
1241: T. Maier, M. Jarrell, T. Pruschke, and M.H. Hettler,
1242: Rev. Mod. Phys. {\bf 77}, 1027 (2005).
1243:
1244: \bibitem{kyung2}
1245: B. Kyung, S.S. Kancharla, D. Senechal, A.-M.S. Tremblay,
1246: M. Civelli, and G. Kotliar,
1247: Phys. Rev. B {\bf 73}, 165114 (2006).
1248:
1249: \bibitem{capone2}
1250: M. Capone and G. Kotliar,
1251: Phys. Rev. B {\bf 74}, 054513 (2006).
1252:
1253: \bibitem{kyung3}
1254: B. Kyung,
1255: Phys. Rev. B {\bf 75}, 033102 (2007).
1256:
1257: \bibitem{merino}
1258: J. Merino,
1259: Phys. Rev. Lett. {\bf 99}, 036404 (2007).
1260:
1261: \bibitem{zhang}
1262: Y.Z. Zhang and M. Imada,
1263: Phys. Rev. B {\bf 76}, 045108 (2008).
1264:
1265: \bibitem{ohashi}
1266: T. Ohashi, T. Momoi, H. Tsunetsugu, and N. Kawakami
1267: Phys. Rev. Lett. {\bf 100}, 076402 (2008).
1268:
1269: \bibitem{saheb}
1270: P. Sahebsara and D. Senechal,
1271: Phys. Rev. Lett. {\bf 100}, 136402 (2008).
1272:
1273: \bibitem{park}
1274: H. Park, K. Haule, and G. Kotliar,
1275: arXiv:0803.1324.
1276:
1277: \bibitem{koch}
1278: E. Koch, G. Sangiovanni, and O. Gunnarsson,
1279: arXiv:0804.3320.
1280:
1281: \bibitem{gull}
1282: E. Gull, Ph. Werner, M. Troyer, and A.J. Millis,
1283: arXiv:0805.3778.
1284:
1285: \bibitem{ferrero}
1286: M. Ferrero, P.S. Cornaglia, L. De Leo, O. Parcollet, G. Kotliar, and A. Georges,
1287: arXiv:0806.4383.
1288:
1289: \bibitem{lee}
1290: H. Lee, G. Li, and H. Monien,
1291: arXiv:0807.1683.
1292:
1293: \bibitem{balzer}
1294: M. Balzer and M. Potthoff,
1295: arXiv:0808.2364.
1296:
1297: \bibitem{senechal3}
1298: An excellent introduction to quantum cluster models can be found in:
1299: D. Senechal, arXiv:0806.2609.
1300:
1301: \bibitem{Ferrero}
1302: See also:
1303: M. Ferrero, L. De Leo, Ph. Lecheminant, and M. Fabrizio,
1304: J. Phys. Condend. Mat. {\bf 19}, 433201 (2007).
1305:
1306: \bibitem{ed}
1307: M. Caffarel and W. Krauth,
1308: Phys. Rev. Lett. {\bf 72}, 1545 (1994).
1309:
1310: \bibitem{perroni}
1311: C.A. Perroni, H. Ishida, and A. Liebsch,
1312: Phys. Rev. B {\bf 75}, 045125 (2007).
1313: See also: A. Liebsch and T.A. Costi,
1314: Eur. Phys. J. B {\bf 51}, 523 (2006).
1315:
1316: \bibitem{arnoldi}
1317: R.B. Lehoucq, D.C. Sorensen, and C. Yang,
1318: ARPACK {\it Users' Guide} (SIAM, Philadelphia, 1997).
1319:
1320: \bibitem{naco}
1321: A. Liebsch and H. Ishida,
1322: Eur. Phys. J. B {\bf 61}, 405 (2008).
1323:
1324: \bibitem{v2o3-al}
1325: A. Liebsch, ED/DMFT for V$_2$O$_3$,
1326: unpublished (2007).
1327:
1328: %\bibitem{ratint}
1329: % {\it Numerical Recipes in Fortran 77},
1330: % Cambridge University Press, p. 106 (1986-1992).
1331: % See also: J. Stoer and R. Burlisch, {\it Introduction to
1332: % Numerical Analysis} (New York, Springer, 1980).
1333:
1334: %\bibitem{haule}
1335: % K. Kaule and G. Kotliar,
1336: % Phys. Rev. B {\bf 76}, 104509 (2007).
1337:
1338: \bibitem{luca}
1339: See also:
1340: M. Capone, L. de' Medici, and A. Georges,
1341: Phys. Rev. B {\bf 76}, 245116 (2007).
1342:
1343: \bibitem{kmixing}
1344: The fact that differences between Eqs.~(\ref{G}) and (\ref{Grho})
1345: can be important was also emphasized by Poteryaev et al.\cite{poteryaev}
1346:
1347: \bibitem{haule}
1348: For the square lattice with $n_c=4$, see, e.g.,
1349: K. Kaule and G. Kotliar,
1350: Phys. Rev. B {\bf 76}, 104509 (2007).
1351:
1352: \bibitem{fang}
1353: Z. Fang, N. Nagaosa, and K. Terakura,
1354: Phys. Rev. B {\bf 69}, 045116 (2004).
1355:
1356: \bibitem{others}
1357: Note, however, other trends such as those discussed in Refs.
1358: \cite{lechermann,ray,tetra}.
1359:
1360:
1361: \end{thebibliography}
1362: \end{document}
1363: