1: %\documentclass[prb,floatfix,twocolumn,showpacs,amsmath,amssymb]{revtex4}

2: %\documentclass[rmp,aps,preprint,nofootinbib,nofloatfix,amsmath,amssymb]{revtex4}

3: %\documentclass[rmp,aps,twocolumn,nofootinbib,floatfix,amsmath,amssymb]{revtex4}

4: %endfloats,

5: \documentclass[12pt,iopams]{iopart}

6: \usepackage{graphicx,color}% Include figure files

7: \usepackage{graphics,lscape,epsfig}

8: \usepackage{dcolumn}% Align table columns on decimal point

9: \usepackage{bm}% bold math

10: \usepackage{epsfig,color}

11: \usepackage{natbib}

12: \usepackage{amssymb}

13: \newcommand{\pdag}{{\phantom{\dagger}}}

14: \newcommand{\be}{\begin{equation}}

15: \newcommand{\ee}{\end{equation}}

16: \newcommand{\ba}{\begin{eqnarray*}}

17: \newcommand{\ea}{\end{eqnarray*}}

18: \newcommand{\bea}{\begin{eqnarray}}

19: \newcommand{\eea}{\end{eqnarray}}

20: \newcommand{\bk}{{\mathbf{k}}}

21: \newcommand{\bkp}{{\mathbf{k'}}}

22: \newcommand{\Ima}{{\Im m}}

23: \newcommand{\Rea}{{\Re e}}

24: \newcommand{\newblock}{\hskip .11em plus .33em minus .07em}

25: 

26: 

27: \begin{document}

28: 

29: \topical{Strong correlations in a nutshell}

30: \author{Michel Ferrero$^1$, Lorenzo De Leo$^{1,2}$, Philippe Lecheminant$^3$ and Michele Fabrizio$^{4,5}$}

31: \address{$^1$ Centre de Physique Th\'eorique, Ecole Polytechnique, 91128 Palaiseau Cedex, France}

32: %\author{Lorenzo De Leo}

33: \address{$^2$ Center for Material Theory, Serin Physics Laboratory, Rutgers University, 136

34: Frelinghuysen Road, Piscataway, NJ 08854-8019, USA}

35: %\author{Philippe Lecheminant}

36: \address{$^3$ Laboratoire de Physique Th\'eorique et Mod\'elisation, Universit\'e de Cergy-Pontoise, CNRS UMR 8089, 

37: 2 Avenue Adolphe Chauvin, 95302, Cergy-Pontoise, France}

38: %\author{Michele Fabrizio}

39: \address{$^4$ International

40: School for Advanced Studies (SISSA), and INFM-Democritos, National Simulation Center, I-34014 Trieste, Italy}

41: \address{$^5$ The Abdus Salam International Center for Theoretical Physics

42: (ICTP), P.O.Box 586, I-34014 Trieste, Italy}

43: 

44: \date{\today}

45: \begin{abstract}

46: We present the phase diagram of clusters made of two, three and four coupled Anderson impurities.

47: All three clusters share qualitatively similar phase diagrams

48: that include Kondo screened and unscreened regimes separated by almost critical crossover regions reflecting

49: the proximity to barely avoided critical points. This suggests the emergence of universal paradigms 

50: that apply to clusters of arbitrary size. We discuss how these crossover regions of the impurity models

51: might affect the approach to the Mott transition within a cluster extension of dynamical mean-field theory.

52: \end{abstract}

53: \pacs{71.30.+h, 71.10.Fd, 71.27.+a}

54: \maketitle

55: 

56: \tableofcontents

57: 

58: %%%%%%%%%%%%%%%  POINT 1. ADDED TABLE

59: \begin{table}[htb]

60: \caption{Acronyms and main notations used in the text. }

61: %\label{acronymous}}

62: \begin{indented}

63: \item[] \begin{tabular}{@{}ll}

64: \br

65: $W$ & Non-interacting electron bandwidth \\

66: $U$ & Hubbard on-site repulsion \\

67: $T_F^*$ & Quasi-particle effective Fermi temperature\\

68: $\Gamma$ & Impurity hybridization width \\

69: $T_K$ & Kondo temperature \\

70: $\rho(\epsilon)$ & Impurity density of states\\

71: $\Sigma(i\omega)$ & Impurity self-energy in Matsubara frequencies\\

72: MIT & Mott metal-to-insulator transition\\

73: DMFT &  Dynamical mean-field theory\\

74: NRG & Wilson's numerical renormalization group \\

75: CFT & Conformal field theory\\

76: DOS & Single-particle density of states \\

77: %BC &  Boundary condition \\

78: %$Z_2$  & Ising conformal field theory \\

79: %TIM & Tri-critical Ising model in two dimensions\\

80: \br

81: \end{tabular}

82: \end{indented}

83: \end{table}

84: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

85: 

86: 

87: 

88: \section{Introduction}

89: \label{Introduction}

90: 

91: The Mott metal-to-insulator transition~\cite{Mott1949,Mott} emerges out of the competition

92: between the opposite tendencies of the electrons to delocalize throughout the lattice

93: in order to maximize the band-energy gain and their mutual Coulomb

94: repulsion which, on the contrary, tends

95: to suppress valence fluctuations by localizing the carriers. If

96: the band-energy gain, which can be identified with the ``bare''

97: bandwidth $W$, is small enough with respect to the short-range

98: Coulomb repulsion, commonly parametrized by an on-site Hubbard

99: $U$, and the average electron-density per site is integer, charge gets 

100: localized and the system is a Mott insulator; otherwise it remains metallic.

101: 

102: Despite its intuitive nature, the Mott phenomenon is extremely difficult to study

103: because it is inherently non-perturbative and because it escapes any simple

104: single-particle description. Those can only deal with band-insulators, characterized by an energy gap separating totally filled

105: from unfilled bands. The simplest example of a Mott insulator is provided by the single-band Hubbard model at half-filling,

106: which always has an insulating phase at sufficiently large repulsion.

107: Yet, in order to make this phase appear for instance in Hartree-Fock

108: theory, one is obliged to assume an antiferromagnetic order parameter that

109: doubles the unit cell so as to fulfill the necessary requirement for

110: a band-insulator -- an even number of electrons per unit cell.

111: With this trick, the Mott transition is effectively turned into a metal to band-insulator 

112: transition driven by magnetism. 

113: In reality, local moments form and eventually order as a consequence of charge localization by the Mott phenomenon.

114: This distinction might look pedantic since the ground state is anyway both insulating and magnetic,

115: but in fact it is not, as we are going to argue by the qualitative behavior of the

116: entropy in the Mott insulator and in the contiguous metal.

117: 

118: \begin{figure}

119: \centerline{\includegraphics[width=12cm]{fabrizio_fig01.eps}}

120: \caption{\label{entropyMott} Qualitative behavior of the entropy in a Mott insulator}

121: \end{figure}

122: 

123: In Fig.~\ref{entropyMott}, we sketch the typical temperature dependence of the entropy 

124: deep inside a Mott insulating phase, $U\gg W$.

125: %%%%%%%%%% POINT 3. MODIFIED TEXT 

126: At high temperature, $T\lesssim U$, valence fluctuations are suppressed and the local charge gets locked to some 

127: fixed value $n_0$. Yet, all local electronic configurations with $n_0$ electrons are thermally occupied with equal probability, 

128: leading to a constant entropy regime that could be identified as the \underline{ideal} Mott insulator, where    

129: charge degrees of freedom are frozen while all other degrees of freedom, in particular 

130: spin, are completely free.  However, at

131: some lower temperature, other energy scales come into play whose

132: role is to lock these additional degrees of freedom, namely to favor one or several among 

133: all the local electronic configurations with $n_0$ electrons. These energy scales

134: may include for instance the on-site Coulomb exchange, responsible for the Hund's

135: rules, the inter-site direct- or super-exchange, the crystal field, the coupling

136: to the lattice, etc. We will collectively denote these energy scales by $J$,

137: which may be identified as the temperature below which the entropy of the residual degrees of

138: freedom of the ideal Mott insulator starts to be suppressed.

139: Consequently, at low temperature, a \underline{realistic}

140: insulating phase is established, which is commonly

141: accompanied by a symmetry breaking phase transition at $T=T_c\leq J$, for instance a magnetic

142: ordering, a collective Jahn-Teller distortion, etc.  

143: Below $T_c$, the entropy decreases to zero as $T\to 0$, generically faster than linearly.

144: For instance, in the half-filled single-band Hubbard model, the \underline{ideal} Mott insulator corresponds  

145: to a regime in which each site is singly occupied but its spin can be with equal probability either up or down, leading to an entropy 

146: $\ln 2$ per site. However, below a temperature of the order of the inter-site spin-exchange, the $\ln 2$ entropy decreases  

147: until the system crosses a magnetic phase transition, below which its entropy vanishes according to the dimensionality of the 

148: system and to the dispersion relation of the spin waves. 

149: %%%%%%%%%%%%%%%%%%%%%%%%%% END MODIFIED TEXT

150: 

151: Recently, an amount of research activity has focused on the

152: possibility that different symmetry broken phases may compete in

153: the insulator, leading to exotic low temperature phenomena~\cite{misguich-2005-}.

154: Here, we will completely discard this event and concentrate on a different competition which

155: emerges in the metallic phase adjacent the Mott insulator.

156: 

157: \begin{figure}

158: \centerline{\includegraphics[width=12cm]{fabrizio_fig02.eps}}

159: \caption{Qualitative behavior of the entropy of a strongly-correlated metal.}

160: \label{entropymetal}

161: \end{figure}

162: 

163: In Fig.~\ref{entropymetal}, we draw how the entropy versus temperature might look like 

164: for a strongly-correlated Fermi-liquid-like metal, assuming

165: that no symmetry breaking intervenes down to zero temperature.

166: As before, we expect that the charge entropy is, this time only partially,

167: reduced at some high temperature of order $U\simeq W$. The rest of it,

168: as well as the entropy of the other degrees of freedom, are instead suppressed by the

169: formation of the degenerate quasi-particle gas. This occurs below a temperature $T_F^*$, that can

170: be identified as the effective quasi-particle Fermi temperature. Since quasi-particles carry

171: the same quantum numbers as the electrons, the entropy quenching involves all degrees of freedom

172: at once, including the charge. Below $T_F^*$ the entropy vanishes linearly,

173: $S(T) \simeq \gamma_*\, T$, with a specific heat coefficient $\gamma_*$ usually larger than its

174: non-interacting value $\gamma \sim 1/W$.

175: 

176: Let us suppose that the Mott transition (MIT) were continuous and try to guess how that might happen from

177: the point of view of the entropy.

178: Obviously, since quasi-particles disappear in the Mott insulator, $T_F^*$ has to vanish at the MIT. Therefore,

179: sufficiently close to the MIT, the quasi-particle Fermi temperature $T_F^*$ must become smaller than $J$.

180: When this happens we should expect, by continuity with the insulating side, that part of the spin entropy

181: gets suppressed already at temperatures of order $J$, above the onset of Fermi-degeneracy. This amounts to 

182: some kind of pseudo-gap opening above $T_F^*$, which is at odds with the conventional Landau-Fermi-liquid theory. 

183: One way out, apart from a first-order MIT, is that something new occurs

184: when $T_F^* \simeq J$. Indeed, the presence of $J$ provides the metallic phase

185: with an alternative mechanism to freeze spin

186: degrees of freedom independently of the charge ones, a mechanism that becomes competitive

187: with the onset of a degenerate quasi-particle gas when $T_F^* \simeq J$. This competition is likely

188: to lead to an instability of the Landau-Fermi-liquid towards a low-temperature symmetry-broken phase 

189: (this happens in the insulating side), but may also signal a real break-down of Fermi-liquid theory.

190: 

191: 

192: Notice that, unlike the competition between different symmetry broken Mott-insulating

193: phases, which requires a fine tuning of the Hamiltonian parameters

194: that may only accidentally occur in real materials, this other type 

195: of competition -- whose effects have not been discussed in the literature before to the extent we believe they deserve -- 

196: should be encountered whenever it is possible to

197: move gradually from a Mott insulator into a metallic phase, for

198: instance by doping or by pressure. We also know several examples

199: where this competition is argued to be at the origin of interesting

200: phenomena. For instance, in heavy fermion materials the Kondo

201: effect, favoring the formation of a coherent band of heavy

202: quasi-particles, competes with the RKKY interaction (for a comprehensive review 

203: see e.g. Ref.~\cite{Hewson}). Here this

204: competition is supposedly the key to understand the anomalies

205: which appear at the transition between the heavy fermion

206: paramagnet and the magnetically ordered phase~\cite{millis-1993,coleman-2001-13,si-2004-}.

207: 

208: 

209: \subsection{Competing screening mechanisms in Anderson impurity models}

210: \label{Intro:A}

211: 

212: The heavy-fermion example is a particularly pertinent one to 

213: introduce the subject of this Topical Review. Indeed, the competition

214: between Kondo effect and RKKY coupling has interesting

215: consequences not only in the periodic Anderson model but already

216: at the level of Anderson impurity models. 

217: 

218: For instance, the phase

219: diagram of two spin-1/2 impurities coupled to a conduction bath

220: and mutually by a direct antiferromagnetic exchange has two limiting

221: regimes: one where each impurity is independently Kondo screened

222: by the conduction electrons; and another where the 

223: exchange locks the impurity spins into a singlet

224: state, now transparent to the conduction electrons. Under

225: particular circumstances -- the two involved scattering channels must be independently coupled each to one impurity --

226: these two regimes are separated by a

227: quantum critical point, at which non-Fermi liquid behavior

228: emerges~\cite{Jones87,Jones88,Jones89,Affleck92PRL,Affleck95}.

229: 

230: The phase diagram grows richer when one consider three

231: antiferromagnetically coupled spin-1/2

232: impurities~\cite{paul-1996-,ingersent-2005-95}. Here, besides a

233: Kondo screened regime, there are other phases where the direct

234: exchange prevails, but is unable to fully quench all impurity

235: degrees of freedom. This leads to stable non-Fermi liquid phases

236: analogous to overscreened multi-channel Kondo

237: models~\cite{Affleck:1990by,Affleck:1990iv}. These impurity-cluster models are interesting not only as simple attempts

238: towards an understanding of the fully periodic Anderson model, but also

239: because compact cluster of impurities are

240: achievable experimentally by adsorbing atoms on metallic surfaces.

241: Trimers of Cr atoms have already been realized on gold

242: surfaces~\cite{Jamneala2001mc}, which has actually motivated the

243: most recent theoretical activity on impurity

244: trimers~\cite{Kudasov2002u,savkin-2005-94,lazarovits-2005-95,ingersent-2005-95}.

245: In this context, the major task is to identify those phases which

246: are stable towards perturbations generically present on

247: metallic surfaces. Therefore the quantum critical points

248: that separate stable phases are of minor interest from an experimental point of view, as they require such 

249: a fine tuning that is extremely unlikely to occur in physical systems.  

250: 

251: 

252: 

253: 

254: \subsection{Impurity models and Dynamical Mean Field Theory}

255: 

256: Unstable critical points arise when the competition between Kondo screening and

257: RKKY coupling is maximum. This is nothing but the impurity counterpart of the situation

258: in which $T_F^* \simeq J$ we previously met in connection with the Mott transition.

259: This weak analogy turns into an actual equivalence within the so-called

260: dynamical mean-field theory (DMFT)~\cite{George96}, 

261: the quantum analogue of classical mean-field theory which, like the latter,

262: is exact for infinite coordination lattices. In this limit, the

263: single-particle self-energy becomes fully local but maintains a non-trivial

264: time-dependence, obtained within DMFT by solving an auxiliary single-impurity Anderson model

265: that is designed so as to have an impurity self-energy that coincides with the local self-energy of

266: the lattice model. This requirement translates into an impurity

267: model identified by the same local interaction as the

268: lattice model and by a coupling to a conduction bath that must 

269: be self-consistently determined. The single-site formulation of DMFT has provided

270: a lot of useful information about the Mott transition {\sl per se}, disentangled from magnetism

271: or whatever symmetry breaking occurs in the insulating state. However, 

272: even though single-site DMFT can account in a Hartree-Fock manner for simple magnetic phases on bipartite lattice,  

273: it is inadequate to study our anticipated competition. For instance, it misses precursor effects in the 

274: paramagnetic phase close to the magnetic phase transition, caused by inter-site

275: processes which disappear in infinite coordination lattices.

276: For this reason, several extensions of DMFT have been recently proposed to include

277: short-range spatial correlations~\cite{Kotliar2001spb,Maier2005,senechal-2000-84,potthoff-2003-91,Lichtenstein2000k}.

278: In these novel versions, the lattice model is mapped onto \underline{a cluster} of

279: Anderson impurities, subject to self-consistency conditions. 

280: 

281: The physics of the Anderson impurity model turned out to be a precious guideline

282: to interpret single-site DMFT results~\cite{George96}. Similarly, we expect

283: that the preliminary knowledge of the general properties of impurity clusters is useful, perhaps even 

284: necessary, in connection with any cluster-extension of DMFT. However, apart from few

285: exceptions~\cite{Jones87,paul-1996-}, little is known about impurity clusters.

286: In addition, since models of impurity clusters involve many intra- and inter-impurity energy scales, it is not

287: {\sl a priori} evident that there should be  a common interpreting framework

288: like the Kondo physics in the single-impurity case.

289: 

290: This is actually the purpose of this Topical Review. Specifically, we are going to present the

291: phase diagram of the simplest among impurity clusters, namely dimers, trimers and tetramers, that could be used to

292: implement a cluster DMFT calculation on strongly correlated models. 

293: Besides our main objective to identify the common features among different clusters,

294: which should presumably play the most significant role in a DMFT approach, we will also try to argue how much of the

295: impurity cluster physics might survive the DMFT self-consistency, hence the possible consequences on the phase diagram

296: of the lattice models.

297: Needless to say, the interest in impurity clusters

298: goes beyond its possible relevance to strongly correlated models near a Mott transition.

299: As we previously mentioned, these clusters may be experimentally realized

300: on metallic surfaces or, eventually, by arranging quantum dots in

301: proper geometries. Moreover,

302: these models represent a theoretical challenge by themselves, which

303: requires the full machinery of Wilson's numerical renormalization

304: group (NRG)~\cite{Wilson75,Krishnamurthy80i,Krishnamurthy80ii} and conformal

305: field theory (CFT)~\cite{DiFrancesco} for a detailed comprehension.

306: 

307: \bigskip

308: 

309: Before entering into the details of our calculations, it is worth briefly

310: presenting the physical idea that guided this work.

311: First of all, let us recall some basic facts of the

312: single-site DMFT mapping onto impurity models. Within this mapping, the quasi-particle

313: effective Fermi temperature, $T_F^*$, translates into the Kondo temperature, $T_K$, of the impurity model.

314: The self-consistency condition causes $T_K$ to vanish at a finite

315: value of $U$, which signals, in the lattice counterpart, the

316: onset of the Mott transition. This also implies that the metallic

317: phase just prior to the Mott transition translates into an

318: Anderson impurity model deep inside the Kondo regime, with a very

319: narrow Kondo resonance and pre-formed Hubbard side-bands~\cite{George96}.

320: The same behavior should occur even when dealing with a cluster of impurities,

321: which should therefore translate into a cluster of Kondo impurities.

322: The novelty stems from the other energy scales which we collectively denoted as

323: $J$, and that take care of quenching in the Mott insulator the degrees of

324: freedom other than the charge. Indeed, near the Mott transition, $J$ translates into

325: additional processes, like for instance a direct exchange between the impurity-spins, which tend to remove,

326: completely or partially, the degeneracy of the cluster. Consequently, $J$

327: competes with the Kondo effect, an agent that takes more advantage the

328: more degenerate the impurity-cluster ground state.

329: \begin{figure}

330: \centerline{\includegraphics[width=12cm]{fabrizio_fig03.eps}}

331: \caption{\label{Fig1} Behavior versus $U$ and doping $\delta$ 

332: of the quasiparticle Fermi temperature, $T_F^*$, which translates within DMFT into the

333: Kondo temperature $T_K$ of the effective impurity model. $J$ is an effective intra-cluster energy scale.}

334: \end{figure}

335: 

336: We notice that this competition is always active in impurity clusters, 

337: while it is commonly absent in single-impurity models except in 

338: multi-orbital cases~\cite{Capone02,Capone04} whose physics is in fact close to clusters. We believe that this

339: additional ingredient is precisely the common denominator of all impurity-cluster models, which endows them

340: with the capability of providing a more faithful description of a realistic

341: Mott transition within DMFT.

342: 

343: Indeed, in the presence of the intra-cluster coupling $J$, the approach to the Mott transition

344: changes as qualitatively shown in Fig.~\ref{Fig1}, with a Kondo

345: temperature smoothly decreasing from its initial value $W$ as

346: $U/W$ increases and becoming of order $J$ just before the transition. 

347: Analogously, see also Fig.~\ref{Fig1}, starting 

348: from the Mott insulator and doping it, $T_K$ will smoothly increase

349: from its value $T_K=0$ at zero doping, until it will again cross a

350: value of order $J$. In other words, any impurity cluster

351: should experience, within DMFT, two different regimes. The first, when $T_K\gg J$, in which full Kondo

352: screening takes place and the impurity density of states

353: displays the usual Kondo-resonance. In the lattice model, this regime translates into a conventional correlated

354: metal. The second, when $T_K\ll J$, particularly close to the Mott transition, in which

355: no or only partial Kondo screening occurs. Here the impurity density of states is pseudo-gaped at the

356: chemical potential. As we will show, these two regimes of the impurity cluster are generically separated by

357: an almost critical crossover-region that reflects the proximity to a true quantum critical point.

358: How do the unscreened phase and the almost critical crossover-region of the impurity cluster translate

359: in the lattice model? The answer to this question would be simple if a true impurity critical point

360: existed, as discussed in Refs.~\cite{DeLeo03f} and \cite{DeLeo04f}. Indeed, near a

361: critical point, the impurity model displays strongly enhanced

362: local susceptibilities, equivalently enhanced local irreducible four-leg vertices,

363: in several instability channels. Within DMFT, the irreducible four-leg

364: vertices, which enter the Bethe-Salpeter equations, coincide with the

365: local ones~\cite{George96}. Therefore, it is reasonable to argue that,

366: after full DMFT self-consistency is carried out, these local

367: instabilities may turn into symmetry-breaking bulk-instabilities that correspond to the same instability channels

368: of the impurity critical point. However, establishing which one of these symmetry breakings really occurs

369: requires full DMFT calculations, as it depends on other details, like for instance the nesting of the

370: Fermi surface. These speculations have been tested with success by a DMFT analysis of

371: a two-orbital Hubbard model~\cite{Capone04}. Although these criticalities

372: are, rigorously speaking, avoided in impurity-cluster models pertinent to DMFT, still we believe that these

373: models approach a critical point so closely that the physics does not change qualitatively.

374: 

375: In the following, we will describe in succession the case of dimer, trimer and tetramer clusters, and, in spite of their 

376: obvious differences, we will identify the universal aspects of the competition discussed above. 

377: 

378: 

379: \section{The impurity dimer}

380: \label{The impurity dimer}

381: The simplest impurity cluster that is relevant to DMFT is a dimer described by the Hamiltonian

382: \bea

383: \mathcal{H} &=& \sum_{a=1}^2\,\sum_{\bk\sigma} \, \epsilon_\bk \, c^\dagger_{a\,\bk\sigma} \, c^\pdag_{a\,\bk\sigma} \,

384:   + \, \sum_{\bk\sigma} \, t_{\perp\,\bk}\,\left( c^\dagger_{1\,\bk\sigma} \, c^\pdag_{2\,\bk\sigma} + H.c. \right) \nonumber \\

385:   && - \sum_{a=1}^2\,\sum_{\bk\sigma} \,\left(V_\bk\,c^\dagger_{a\,\bk\sigma} d^\pdag_{a\,\sigma} + H.c.\right)\label{AIM-dimer}\\

386:   && -t_\perp\,\sum_\sigma \Big(d^\dagger_{1\,\sigma} d^\pdag_{2\,\sigma}

387:   + H.c.\Big)+ \frac{U}{2}\sum_{a=1}^2 \, (n_a - 1)^2,\nonumber

388: \eea

389: where $c^\dagger_{a\,\bk\sigma}$ creates a conduction electron in channel $a=1,2$ with

390: momentum $\bk$, energy $\epsilon_\bk$, measured with respect to the chemical potential, and spin $\sigma$,

391: while $d^\dagger_{a\,\sigma}$ is the creation

392: operator at the impurity site $a=1,2$ with spin $\sigma$, $n_a = \sum_\sigma\,

393: d^\dagger_{a\,\sigma}  d^\pdag_{a\,\sigma}$ being the occupation number.

394: This model describes two Anderson impurities, each hybridized with its own conduction bath

395: and in turn coupled to the other impurity by a single-particle hopping $t_\perp$.

396: Since within cluster DMFT the self-consistent baths 

397: must mimic the effects of the rest of the lattice on the two sites of the dimer,

398: also the two baths are coupled by a hybridization $t_{\perp\,\bk}$.

399: %%%%%%%%%%%%%%%%% POINT 4. ADDED SENTENCE

400: The role of the inter-bath hybridization is to generate a frequency-dependent contribution to the

401: inter-impurity hopping that, together with $t_\perp$, produce off-diagonal elements $a \not = b$ to

402: the impurity Green's function

403: \[

404: \mathcal{G}_{ab}(\tau) = -\langle T_\tau\Big(d^\pdag_{a\,\sigma}(\tau)\,d^\dagger_{b\,\sigma}\Big)\rangle.

405: \]

406: In fact, any coupling among the baths transfers into a coupling among the impurities and vice-versa, 

407: apart from the frequency dependence that can be anyway neglected in the asymptotic low-frequency regime.

408: For this reason, in what follows we will indifferently refer to inter-bath or to inter-site depending upon the context.  

409: %%%%%%%%%%%%%%%%% END ADDED SENTENCE

410: 

411: 

412: Close to the Mott transition

413: the effective impurity model resides well inside the Kondo regime, where $U\gg V_\bk,t_\perp$.

414: Here, the model can be mapped via a Schrieffer-Wolff transformation onto two spin-1/2 impurities that, up to order $1/U$,

415: are coupled to the two conduction baths by a Kondo exchange

416: \be

417: J_K = \frac{8}{U}\,\sum_\bk\, \left|V_\bk\right|^2 \, \delta\left(\epsilon_\bk\right)

418: \label{JK}

419: \ee

420: and together by an antiferromagnetic

421: $J = 4t_\perp^2/U$. This means that the spectral distribution of the

422: inter-impurity hybridization

423: \[

424: \sum_\sigma\, \langle \Big(d^\dagger_{1\,\sigma} d^\pdag_{2\,\sigma}

425: + H.c.\Big)\rangle = -4\,\int_{-\infty}^0 \frac{d\omega}{\pi}\, \Ima\, \mathcal{G}_{12}(\omega) ,

426: \]

427: is transferred to high energy, and what remains at low energy is mainly the

428: exchange $J$. Within the DMFT self-consistency scheme, we should then expect that also the

429: direct hybridization among the baths, $t_{\perp\,\bk}$, behaves

430: similarly, which suggests that one could start the analysis with the large $U$-limit of the Hamiltonian

431: \bea

432: \mathcal{H} &=& \sum_{a=1}^2\Bigg[\sum_{\bk\sigma}

433: \,\epsilon_\bk\, c^\dagger_{a\,\bk\sigma} c^\pdag_{a\,\bk\sigma} -

434: \left(V_\bk\,c^\dagger_{a\,\bk\sigma} d^\pdag_{a\,\sigma} + H.c.\right)\Bigg]\nonumber\\

435: &+& \frac{U}{2}\sum_{a=1}^2 \, (n_a - 1)^2 + J\, \mathbf{S}_1\cdot\mathbf{S}_2 \nonumber \\

436: && \equiv \sum_{a=1}^2\,\mathcal{H}_{a}^K + J\, \mathbf{S}_1\cdot\mathbf{S}_2,

437: \label{Ham-dimer}

438: \eea

439: where $\mathbf{S}_a$ is the spin-density operator of impurity $a=1,2$, plus a weak inter-bath hybridization

440: to be considered as a perturbation. This does not at all imply that the latter is irrelevant.

441: Rather, as we are going to discuss, this hybridization turns out to be a relevant perturbation. It only means that

442: this perturbation becomes influential at energy scales much smaller than those at which the main effects

443: caused by the competition between $J$ and $J_K$ start to appear, as we shall discuss later. For the time being, let us

444: consider the Hamiltonian (\ref{Ham-dimer}). This model was originally

445: studied with NRG by Jones and Varma~\cite{Jones87,Jones88,Jones89}. They found that the phase diagram

446: includes two stable phases. When the Kondo temperature, $T_K$, is much larger than $J$, each impurity is

447: Kondo screened by its conduction bath. On the contrary, when $J\gg T_K$, the two impurities lock into a singlet and

448: no Kondo screening is required anymore. These two stable phases were found to be separated by a critical point with

449: non-Fermi liquid properties~\cite{Jones89}. We notice that, since $T_K$ is a decreasing function of $U$, the

450: phase diagram at fixed $J/U\ll 1$ as a function of $U/\Gamma$, where $\Gamma=\Gamma(0)$ is the hybridization width defined through

451: \be

452: \Gamma(\epsilon) = \pi\,\sum_{\bk}\, V_\bk^2 \,\delta\left(\epsilon - \epsilon_\bk\right),

453: \label{Gamma(e)}

454: \ee

455: also includes a critical point at some $(U/\Gamma)_*$, see Fig.~\ref{dimer-phd}.

456: %%%%%% POINT 9. ADDED SENTENCE 

457: More specifically, since the Kondo temperature in units of half the conduction bandwidth behaves, at $J=0$ and for large $U$, 

458: as~\cite{Hewson}

459: \be

460: T_K \sim \sqrt{\frac{8\Gamma}{\pi U}}\;\mathrm{e}^{-\pi U/8\Gamma},

461: \label{estimate-TK}

462: \ee

463: one expects the critical point to occur approximately around  

464: \be

465: \frac{U}{\Gamma} \sim \frac{8}{\pi}\,\ln\frac{1}{J}.

466: \label{estimate-U/Gamma}

467: \ee 

468: %%%%%%%%% END ADDED SENTENCE

469: \begin{figure}

470: \centerline{\includegraphics[width=12cm]{fabrizio_fig04.eps}}

471: \caption{\label{dimer-phd} Phase diagram of the dimer model (\ref{Ham-dimer}) as function of $U/\Gamma$ at

472: fixed $J/U\ll 1$. }

473: \end{figure}

474: In other words, perturbation theory breaks down at a finite value of the interaction within

475: model (\ref{Ham-dimer}), which is {\sl per se} an interesting situation uncommon in interacting Fermi systems.

476: 

477: The detailed properties of the critical point were later unraveled in Refs.~\cite{Affleck92PRL} and

478: \cite{Affleck95} by means of conformal field theory (CFT). 

479: The use of CFT to study impurity models relies on the fact that only a finite number of conduction electron  

480: scattering channels are hybridized with the impurity. This implies that, when $\Gamma(\epsilon)$, Eq.~(\ref{Gamma(e)}), 

481: is smooth around the chemical potential, $\epsilon=0$, on a scale larger than the Kondo temperature, 

482: the asymptotic low temperature/frequency behavior is similar to a  

483: conventional one-dimensional semi-infinite chain of non-interacting electrons, the impurity sitting at the edge. 

484: It is known that non-interacting electrons in one dimension can be mapped through bosonization~\cite{Sasha} 

485: onto a 1+1 critical field theory -- the criticality corresponding to the fermionic spectrum being gapless -- 

486: that is not only scale but also conformal invariant.   

487: This allows to fully identify and classify all critical properties: thermodynamic quantities, correlation functions and  

488: finite-size energy spectra~\cite{Zamo,Zamo1,Cardy1,DiFrancesco}.

489: In impurity models, the effective one-dimensional chain turns out to be semi-infinite, implying that 

490: one has actually to deal with a boundary CFT~\cite{Cardy}. Since a single impurity can not induce any gap in the 

491: bulk spectrum, the conformal invariance of the free electrons remains intact; the only effect of the impurity is to 

492: change the boundary conditions (BCs) among the conformally invariant ones. 

493: A crucial step to determine the allowed BCs is the so-called {\sl conformal

494: embedding}~\cite{DiFrancesco}, which amounts to identifying the

495: conformal field theories corresponding to the symmetry groups 

496: under which the Hamiltonian of the conduction electrons plus the impurity stays

497: invariant. Note that the number of gapless degrees of freedom must be conserved, which corresponds, 

498: within CFT, to the fact that the 

499: CFTs in the absence and presence of the impurity have the same total {\sl central charge}~\cite{DiFrancesco}.

500: A conformal embedding can be justified rigorously 

501: by identifying the partition function of free electrons with that obtained by combining the partition functions of the 

502: CFTs that emerge out of the embedding. Some simple applications of this very powerful method are given in the Appendix.  

503: In the most favorable cases, the proper BCs correspond to conformally invariant BCs only within one of

504: the different CFTs of the embedding. The next useful information

505: is that the conformally invariant BCs within each sector can be

506: obtained by the so-called {\sl fusion

507: hypothesis}~\cite{Cardy,affleck-1995-26}, according to which, starting

508: from the spectrum of a known BC, one can obtain all the others upon

509: {\sl fusion} with the proper scaling fields, called {\sl primary

510: fields}, of the CFT, see the Appendix. {\sl Fusion} is just the technical word to denote

511: the process in which the impurity with its quantum numbers 

512: dissolves into the conduction electron Fermi sea.

513: 

514: Let us consider for instance model (\ref{Ham-dimer}), and assume, for simplicity, that the two baths are

515: particle-hole invariant. As a consequence, the baths, in the absence of the impurities,

516: can be described by a CFT~\cite{affleck-1995-26} which includes

517: independent spin $SU(2)_1$ and charge isospin $SU(2)_1$ for each

518: bath, see Appendix, namely an overall

519: \[

520: \big(SU(2)_1^{(1)}\times SU(2)_1^{(2)}\big)_{charge}\,\times \,

521: \big(SU(2)_1^{(1)}\times SU(2)_1^{(2)}\big)_{spin}.

522: \]

523: The subscript in $SU(2)_k^{(a)}$ can be regarded here as the

524: number of copies of spin- or isospin-1/2 electrons participating to the

525: $SU(2)$ algebra~\cite{Sasha}, while the superscript refers to the bath.

526: Since the charge isospin generators

527: commute with the Hamiltonian even when the coupling to the impurities is switched on,

528: the charge sector can still be represented by two independent

529: isospin $SU(2)_1$ CFTs. On the other hand, only the total spin

530: $SU(2)$ transformations leave the Hamiltonian invariant, which

531: translates into an $SU(2)_2$ (two copies of electrons) CFT. As a

532: result the proper embedding in the spin sector

533: is~\cite{Affleck92PRL}, see also the Appendix, 

534: \[

535: \Bigg(SU(2)_1^{(1)}\times SU(2)_1^{(2)}\Bigg)_{spin} \rightarrow SU(2)_2 \times Z_2,

536: \]

537: where $Z_2$ denotes an Ising CFT which reflects the symmetry under

538: permutation of the two baths. The Ising CFT has three primary fields, the identity $I$, with dimension 0, 

539: $\epsilon$ (the thermal operator), with dimension 1/2, 

540: and $\sigma$ (the Ising order parameter), with dimension 1/16. 

541: Each state and operator can then be identified by the 

542: quantum numbers $(I_1,I_2,S,\mathrm{Ising})$ where $I_1$ and $I_2$ refer to the charge isospin of each channel, 

543: $S$ to the total spin and $\mathrm{Ising}$ to the $Z_2$ sector. 

544: 

545: Affleck and Ludwig~\cite{Affleck92PRL,Affleck95} realized that the different fixed

546: points found by NRG, see Fig.~\ref{dimer-phd}, correspond to the three different BCs of the Ising CFT,

547: namely two fixed BCs, the stable phases -- where one Ising-spin

548: orientation is prohibited at the boundary -- and one free BC, the

549: unstable fixed point -- where both orientations are allowed.

550: Starting from the unscreened phase, the Kondo screened phase is

551: obtained by fusion with the Ising primary field $\epsilon$, while the unstable critical point is obtained by fusion with the

552: Ising order parameter $\sigma$. Furthermore, the critical point is identified

553: by a finite residual entropy $S(T=0)= \frac{1}{2}\;\ln 2$, showing that part of the impurity degrees of freedom

554: remains unscreened.

555: 

556: 

557: CFT also determines, by the so-called

558: double-fusion~\cite{affleck-1995-26}, the 

559: dimensions of the relevant operators, see the Appendix. It turns out that there are

560: three equally relevant (i.e. with dimension less than one) 

561: symmetry breaking perturbations which can destabilize the unstable

562: fixed point, identified by two asterisks in Table~\ref{k=2-fixed-content} of the Appendix. 

563: They all have the same dimension $1/2$ as the

564: invariant operator, single asterisk in Table~\ref{k=2-fixed-content} of the Appendix, which moves away from the fixed point and

565: corresponds to a deviation of $J$ from its fixed point value $J_*$

566: at fixed $T_K$, or, vice versa, a deviation of $T_K$ at fixed $J$.

567: The first symmetry breaking operator is an opposite spin magnetization for the

568: two baths, namely a local operator of the form

569: \be

570: h_{AF}\,\Big(\mathbf{S}_1 - \mathbf{S}_2\Big), 

571: \label{dimer-AF}

572: \ee

573: and corresponds in Table~\ref{k=2-fixed-content} of the Appendix to the operator with quantum numbers 

574: $(I_1,I_2,S,\mathrm{Ising})=(0,0,1,0)$. 

575: The second is a BCS term in the inter-bath Cooper

576: singlet channel:

577: \be

578: h_{SC}\,\left( d^\dagger_{1\uparrow}d^\dagger_{2\downarrow} + d^\dagger_{2\uparrow}d^\dagger_{1\downarrow}\right)

579: + H.c.\,.

580: \label{dimer-SC}

581: \ee

582: The last perturbation is a direct

583: hybridization between the two baths,

584: \be

585: \sum_\sigma\, h_\perp\,d^\dagger_{1\sigma}d^\pdag_{2\sigma}\, + \, H.c.\, ,

586: \label{dimer-tperp}

587: \ee

588: which breaks the $O(2)$ channel symmetry. Both (\ref{dimer-SC}) and (\ref{dimer-tperp}) correspond 

589: in Table~\ref{k=2-fixed-content} of the Appendix to the spin-singlet operator with quantum numbers 

590: $(I_1,I_2,S,\mathrm{Ising})=(1/2,1/2,0,0)$. On the contrary, both a chemical potential shift that

591: moves away from particle hole symmetry, or a perturbation that splits the two conduction channels, do not

592: destabilize the critical point. Indeed, if the position of the impurity levels is modified, so that the

593: average number of electrons on each impurity moves away from $\langle n_a\rangle = 1$, $a=1,2$, the critical

594: point is still encountered, although it will shift to larger $U/\Gamma$ at fixed $J/U\ll 1$~\cite{DeLeo04f}, see

595: Fig.~\ref{dimerpdmu}.

596: \begin{figure}

597: \centerline{\includegraphics[width=8cm]{fabrizio_fig05.eps}}

598: \caption{\label{dimerpdmu} Phase diagram of (\ref{Ham-dimer}) as function of $U/\Gamma$ and

599: $\delta = |\langle n\rangle -1|$, $\langle n\rangle$ being the average occupancy of each impurity, at fixed $J/U\ll 1$.}

600: \end{figure}

601: 

602: 

603: 

604: 

605: 

606: \subsection{Dynamical behavior of the impurity dimer}

607: 

608: The instability channel (\ref{dimer-tperp}) is very important since, as we mentioned, the inter-bath as well as the inter-impurity

609: hybridization are always present. Therefore the relevant issue becomes whether this

610: hybridization completely washes out the critical behavior of the underneath

611: critical point, or whether a critical region still remains. In order to answer this question, it is

612: convenient to analyze the impurity spectral function, which is also the key ingredient of the

613: DMFT self-consistency procedure.

614: 

615: We start by noticing that, in spite of the fact that both Kondo screened and unscreened

616: phases of the Hamiltonian (\ref{Ham-dimer}) are Fermi-liquid-like in Nozi\`eres' sense~\cite{NozieresJLTP} 

617: %%%%%% POINT 5. ADDED SENTENCE

618: (namely they correspond asymptotically to well defined limits of free-electrons scattering off a structure-less impurity potential, 

619: infinite in the Kondo screened phase and zero in the unscreened one),  

620: %%%%%% END ADDED SENTENCE

621: the dynamical properties of the impurities are completely different. Indeed,

622: the conduction-electron scattering $S$-matrix at the chemical potential should satisfy

623: \be

624: \mathcal{S} = 1 - 2\,\frac{\rho(0)}{\rho_0},

625: \label{rho.vs.S}

626: \ee

627: where $\rho(0)$ is the actual density of states (DOS) of the impurity at the chemical potential,

628: while $\rho_0=1/(\pi\Gamma)$ is its non interacting $U=J=0$ value. Since both stable phases are Fermi-liquid like,

629: %%%% POINT 5. ADDED SENTENCE

630: it follows that the $S$ matrix is unitary, hence can be written as $\mathcal{S} = \mathrm{e}^{2i\delta}$,

631: %%%% END ADDED SENTENCE 

632: where $\delta$ is the phase-shift at the chemical potential. The Kondo screened phase is identified by a phase-shift $\delta=\pi/2$,

633: that implies $\mathcal{S}=-1$ hence $\rho(0)=\rho_0$; the DOS at the chemical potential is unaffected

634: by the interaction. On the contrary, in the unscreened phase $\delta=0$, thus $\mathcal{S}=1$ and

635: $\rho(0)=0$. In other words, while in the Kondo screened phase the DOS is peaked at the chemical potential 

636: -- the conventional Kondo resonance behavior -- it vanishes in the unscreened one. Furthermore, according to

637: CFT, right at the critical point $\mathcal{S}=0$, namely $\rho(0)=\rho_0/2$. These results are actually

638: reproduced by NRG, see e.g. Ref.~\cite{DeLeo04f}. In Fig.~\ref{DOSexE}, we draw our NRG results

639: for the impurity DOS of the dimer model (\ref{Ham-dimer}) for $U=8$, $J=0.00125$, in units of half the

640: conduction bandwidth, and for various values of $\Gamma$ across the critical point $\Gamma_*$, which lies between

641: 0.42 and 0.44. The upper inset shows that, on large scales, the DOSs in the screened and unscreened phases

642: are practically indistinguishable. The differences emerge at very low energies. Apart from the value of the

643: DOS at the chemical potential, which, as mentioned, can be anticipated by general scattering theory arguments,

644: other useful information can be extracted from the whole low-energy behavior. As was realized in Ref.~\cite{DeLeo04f},

645: the DOS is controlled by two energy scales. In the Kondo screened phase,

646: there is a narrow Kondo peak on top of a broad resonance. The Kondo peak shrinks as the critical point

647: is approached, while the width of the large resonance remains practically constant.

648: Indeed, right at the critical point, only the latter survives. On the contrary, inside the unscreened phase,

649: the Kondo peak turns into a narrow pseudo-gap within the broad resonance, leading to 

650: a low-energy DOS $\rho(\epsilon) \sim \epsilon^2$.

651: \begin{figure}

652: \centerline{\includegraphics[width=12cm]{fabrizio_fig06.eps}}

653: \caption{\label{DOSexE} Main panel: low energy behavior of the impurity DOS of the dimer model

654: (\ref{Ham-dimer}) with $U=8$, $J=0.00125$ and, from top to bottom, $\Gamma=0.44,~0.42,~0.4,~0.35,~0.3$, in

655: units of half the conduction bandwidth. 

656: %%%%% POINT 9. ADDED SENTENCE 

657: We observe that the rough estimate of the order of magnitude of 

658: the critical $U/\Gamma \sim -(8/\pi)\ln J \simeq 17.3$, see Eq.~(\ref{estimate-U/Gamma}), agrees  

659: with the actual numerical value $(U/\Gamma)_* \sim 18.2$.  

660: %%%%%% END ADDED SENTENCE

661: Upper inset: the DOS behavior in the whole energy range

662: with the same $U$ and $J$ and with $\Gamma=0.6$, top curve, and $\Gamma=0.3$. 

663: %%%%% MODIFIED TEXT FOR CLARITY

664: The Hubbard bands are clearly visible, while the low energy parts are hardly distinguishable. 

665: The discretization

666: parameter~\cite{Krishnamurthy80i,Krishnamurthy80ii} that we used is $\Lambda = 2$.

667: %%%%% END MODIFIED TEXT

668: }

669: \end{figure}

670: This behavior has been found to be well reproduced by the following model-DOS at low energy~\cite{DeLeo04f}:

671: \be

672: \rho_\pm(\epsilon) = \frac{\rho_0}{2}\,\Bigg(

673: \frac{T_+^2}{\epsilon^2+T_+^2}\pm \frac{T_-^2}{\epsilon^2+T_-^2}\Bigg),

674: \label{model-DOS}

675: \ee

676: where the + sign refers to the screened phase and the - one to the unscreened. $T_+$ is the width of the

677: broad resonance, $T_-$ the one of the narrow peak in the screened phase and the amplitude of the 

678: pseudo-gap in the unscreened regime. As the critical point is approached on both sides,

679: $T_-\sim |\Gamma -\Gamma_*|^2\to 0$~\cite{DeLeo04f}, in accordance with the CFT prediction that the relevant operator has

680: dimension 1/2. Away from particle-hole symmetry, $\langle n_a\rangle \not = 1$, the two stable phases are still identified

681: by a relative $\pi/2$-shift of $\delta$, although the unscreened phase value, $\delta_0$, is different from zero. In this case, the 

682: following model-DOS was found to reproduce well the NRG data~\cite{DeLeo04f}:

683: \be

684: \rho_\pm(\epsilon) = \frac{\rho_0}{2}\,\Bigg[

685: \frac{\displaystyle T_+^2 + \mu_{\pm}^2}{\displaystyle \left(\epsilon +\mu_{\pm}\right)^2 + T_+^2}

686: \pm \cos 2\delta_0\, \frac{\displaystyle T_-^2}{\epsilon^2 + T_-^2}\Bigg],

687: \label{model-DOS-no-ph}

688: \ee

689: with $\mu_{\pm} = \pm T_+\,\sin 2\delta_0$. This formula shows that, in the unscreened phase, the pseudo-gap

690: remains pinned at the chemical potential, even if, since the broad resonance shifts, the pseudo-gap fills in

691: of an amount proportional to the ``doping'', i.e. $|\langle n_a\rangle - 1|$.

692: 

693: 

694: 

695: By the model DOS (\ref{model-DOS}), one can extract a model self-energy, $\Sigma(i\omega)$

696: in Matsubara frequencies, which, for low $\omega>0$, behaves as

697: \be

698: \Sigma_+(i\omega) \simeq - i\omega\,\Bigg(

699: \frac{\displaystyle \Gamma\left(T_+ + T_-\right)}{\displaystyle 2T_+ T_-} - 1\Bigg)

700: + i\omega^2\, \frac{\displaystyle \Gamma\,\left(T_+-T_-\right)^2}{\displaystyle 4T_+^2 T_-^2},

701: \label{Sigma_+}

702: \ee

703: in the Kondo screened phase, as

704: \be

705: \Sigma_-(i\omega) \simeq -i \frac{1}{\omega}\,\frac{2\Gamma T_+ T_-}{T_+ - T_-}

706: - i\Gamma\,\frac{T_+ + 3T_-}{T_+ - T_-} -i\omega\,\frac{2\Gamma-T_+ + T_-}{T_+ - T_-},

707: \label{Sigma_-}

708: \ee

709: in the  unscreened one, and finally as

710: \be

711: \Sigma_*(i\omega) \simeq -i \Gamma -i\omega\,\frac{2\Gamma - T_+}{T_+},

712: \label{Sigma_*}

713: \ee

714: exactly at the critical point, $T_-=0$, or in the range $T_-\ll \omega \ll T_+$. 

715: The model self-energy reproduces well the actual NRG results, 

716: shown in Fig.~\ref{SigmaexE}. 

717: %%%%%%%%%%% POINT 6. MODIFIED TEXT

718: Only in the screened phase, panels (a) and (b) in Fig.~\ref{SigmaexE}, the self-energy has the standard perturbative

719: behavior,  $\Sigma(i\omega) \sim \Big(1- 1/Z\Big)\,i\omega$, with $Z$ the quasiparticle residue,

720: which breaks down at the critical point, where $\Sigma(i\omega)$ goes to a constant value for $\omega\to 0$, 

721: and even more in the unscreened regime where, as shown on different frequency ranges in panels (a), (c) and (d) of 

722: Fig.~\ref{SigmaexE}, $\Sigma(i\omega)$ diverges as $\omega\to 0$.

723: %%%%%%%%%%% END MODIFIED TEXT

724: \begin{figure}

725: \centerline{\includegraphics[width=12cm]{fabrizio_fig07.eps}}

726: \caption{\label{SigmaexE} Imaginary part of $\Sigma(i\omega)$ versus $\omega$ for $U=8$

727: and $J=0.00125$: (a) whole frequency-range behavior for $\Gamma=0.5,~0.48,~0.44,~0.35,~0.3$, from

728: top to bottom; (b) low frequency behavior for the Kondo screened values $\Gamma=0.5,~0.48,~0.44$;

729: (c) and (d) low frequency behavior for the unscreened values $\Gamma=0.35,~0.3$. These results

730: were obtained with $\Lambda = 2$.}

731: \end{figure}

732: 

733: Let us now consider the original dimer model (\ref{AIM-dimer}). In order not to deal with too many

734: Hamiltonian parameters, we consider the case in which the inter-bath hybridization is zero, $t_{\perp\, \bk}=0$,

735: and take $U=8$, as before, and $t_\perp=0.05$, such that $J=4t_\perp^2/U=0.00125$, the same value used previously.

736: In this model $t_\perp$ has a double role: on one side it generates a spin-exchange able to drive the model across

737: the critical point, but at the same time it also breaks the relevant $O(2)$ channel symmetry thus making 

738: the critical point inaccessible. Indeed, as shown by the behavior of the impurity DOS in Fig.~\ref{DOStperp},

739: a crossover now joins the Kondo screened phase and the unscreened one. There are still quite distinct

740: DOSs deep inside the screened and unscreened phases, the former characterized by a Kondo resonance, the latter by a

741: pseudo-gap. However, the transition from the two limiting behaviors is now just a crossover, although quite sharp.

742: \begin{figure}

743: \centerline{\includegraphics[width=12cm]{fabrizio_fig08.eps}}

744: \caption{\label{DOStperp} Main panel: low energy behavior of the impurity DOS of the model

745: (\ref{AIM-dimer}) with $U=8$, $t_\perp=0.05$ and, from top to bottom, $\Gamma=0.5,~0.47,~0.45,~0.4,~0.3$, in

746: units of half the conduction bandwidth. Upper inset: the DOS behavior in the whole energy range

747: with the same $U$ and $J$ and with $\Gamma=0.5$, top curve, and $\Gamma=0.2$. These results

748: were obtained with $\Lambda = 2$.}

749: \end{figure}

750: 

751: 

752: More interesting are the changes that intervene in the impurity self-energy with respect to the model (\ref{Ham-dimer}).

753: In presence of $t_\perp$, the self-energy, besides diagonal elements, $\Sigma_{11}(i\omega) = \Sigma_{22}(i\omega)$, which

754: are imaginary, also acquires off-diagonal components, $\Sigma_{12}(i\omega) = \Sigma_{21}(i\omega)^*$, which

755: turn out to be purely real. In Fig.~\ref{Sigma}, we plot $\mathcal{I}m\, \Sigma_{11}(i\omega)$

756: and $\mathcal{R}e\, \Sigma_{12}(i\omega)$ versus $\omega$ for the same values of $\Gamma$'s as in

757: Fig.~\ref{DOStperp}, on two different energy ranges. We notice that the gross features of $\Sigma_{11}(i\omega)$

758: remain intact, namely, as the model moves from the screened regime towards the unscreened one, the diagonal self-energy

759: increases quite fast in absolute value. However, a linearly-vanishing ``Fermi-liquid'' behavior is eventually recovered

760: at very low energies, as shown in the left-bottom panel of Fig.~\ref{Sigma}. In fact, the model in the presence of $t_\perp$

761: does not need to develop a singular self-energy anymore to open up a pseudo-gap at the chemical potential

762: as in model (\ref{Ham-dimer}). It is the off-diagonal self-energy, $\Sigma_{12}$, that accomplishes the job in this

763: case. Indeed, as shown in the right panels of Fig.~\ref{Sigma}, $\Sigma_{12}$ becomes so large at low energy to open up

764: an appreciable hybridization gap, not explainable by the tiny value of $t_\perp$ as compared to the hybridization width

765: $\Gamma$. This result could be justified simply by stating that the strong repulsion $U$ enhances the \underline{effective} 

766: strength of $t_\perp$. However, the previous results on the dimer-model (\ref{Ham-dimer}) and the strong energy-dependence 

767: of the self-energy, see Fig.~\ref{Sigma}, suggest that this anomalous behavior reflects rather the properties of the avoided 

768: critical point which exists in the presence of $J$ at $t_\perp=0$. In other words, we believe that our results 

769: testify that a sizable critical region still exits and largely explains the physical behavior.

770: 

771: \begin{figure}

772: \centerline{\includegraphics[width=12cm]{fabrizio_fig09.eps}}

773: \caption{\label{Sigma} $y$-axis: Imaginary part of $\Sigma_{11}(i\omega)$, left panels, and real part of

774: $\Sigma_{12}(i\omega)$, right panels, versus $\omega$ ($x$-axis) for $U=8$, $t_\perp=0.05$ and, from top to bottom,

775: $\Gamma=0.5,~0.47,~0.45,~0.4,~0.3$ and $\Lambda = 2$. In the top figures the whole frequency range

776: is showed, while in the bottom ones only the very-low frequency behavior.}

777: \end{figure}

778: 

779: 

780: Obviously, as in any other case

781: of avoided criticality, the width of the critical region depends on the actual value of $t_\perp$ with respect to the other

782: parameters $U$ and $\Gamma$. Since both $t_\perp$ and $\Gamma$ are self-consistently

783: determined within DMFT as function of $U$ and the bare

784: bandwidth, we cannot establish with certainty what might happen in a DMFT simulation

785: of a Hubbard model using a dimer as a representative

786: cluster. However, because the exchange $J$ derives from high-energy processes and survives even inside the Mott insulator, while the

787: coherent hopping dies out at the Mott transition, we believe that a sizable critical region should exist in the effective

788: impurity-cluster model and plays an influential role in determining the bulk properties after the DMFT

789: self-consistency.

790: 

791: 

792: \section{The impurity trimer}

793: \label{The impurity trimer}

794: 

795: \begin{figure}

796: \centerline{\includegraphics[width=10cm]{fabrizio_fig10.eps}}

797: \caption{\label{Fig-trimer} The impurity trimer with the Hamiltonian

798: (\ref{Ham-trimer})}

799: \end{figure}

800: 

801: An important lesson of the impurity dimer was that, in order to identify in all details the properties of the

802: critical region, it is more convenient to study a model in which the impurities are coupled together by an

803: antiferromagnetic exchange rather than by a hopping as would be the case in reality. In the end, we will

804: discover that, just like in the dimer example, the hopping is a relevant perturbation, and yet a critical

805: region survives.

806: Therefore, let us consider the next simple cluster, which is

807: the impurity trimer drawn in Fig.~\ref{Fig-trimer} with the Hamiltonian

808: \be

809: \mathcal{H} = \sum_{a=1}^3\,\mathcal{H}_a^K\, +

810: J\,\left(\mathbf{S}_1+\mathbf{S}_3\right)\cdot \mathbf{S}_2

811: + J'\,\mathbf{S}_1\cdot \mathbf{S}_3,

812: \label{Ham-trimer}

813: \ee

814: where $\mathcal{H}_a^K$ has been defined in (\ref{Ham-dimer}). This model describes

815: three spin-1/2 impurities, coupled together by antiferromagnetic $J$ and $J'$ and each

816: of them hybridized to a conduction bath. As before, we assume that the baths are degenerate

817: and particle-hole invariant. This model, although simplified by the absence of any inter-impurity

818: hopping, is much more complicated than the dimer model (\ref{Ham-dimer}). Therefore, in order to

819: unravel its phase diagram, we need to combine the NRG analysis, which provides the low-energy

820: spectra of the various fixed points, with CFT, which allows to identify each fixed point with

821: a particular boundary CFT, whose properties can be determined exactly. For this reason, we cannot

822: start our analysis before introducing some CFT preliminaries.

823: 

824: \subsection{CFT preliminaries for the trimer}

825: \label{CFT preliminaries for the trimer}

826: 

827: As in the dimer example, also in the trimer model

828: (\ref{Ham-trimer}) at particle-hole symmetry the charge degrees of freedom can be described by

829: three independent isospin $SU(2)_1^{(a)}$ CFTs, $a=1,2,3$. For

830: the spin degrees of freedom, the expression of 

831: the inter-impurity exchange suggests naturally that we must first 

832: couple the spin sectors of baths 1 and 3 into an overall

833: $SU(2)_2$ via the embedding

834: \be

835: SU(2)_1^{(1)}\times SU(2)_1^{(3)}

836: \rightarrow SU(2)_2^{(1-3)} \times Z_2,

837: \ee

838: and finally couple the $SU(2)_2$ to the bath 2 into an $SU(2)_3$, according to

839: \be

840: SU(2)_2^{(1-3)} \times SU(2)_1^{(2)} \rightarrow SU(2)_3 \times

841: \left({\rm TIM}\right),

842: \ee

843: where TIM stands for the tricritical Ising

844: model CFT with central charge $c=7/10$. 

845: It describes for instance the tricritical point of the 

846: two-dimensional Blume-Capel model which involves an Ising spin variable

847: and a vacancy variable indicating if the site is empty

848: or occupied~\cite{Blume,Capel}.

849: The above conformal embedding can be rigorously justified by the {\sl character

850: decomposition}~\cite{DiFrancesco}, although we do not give here the details of this lengthy  

851: and involved construction.  

852: 

853: The primary fields of an $SU(2)_k$ as well as of the Ising CFTs, together with their fusion, i.e. multiplicative, rules, 

854: are discussed in the Appendix. For what concerns the TIM, it contains 

855: six primary fields, the identity $I$, with dimension 0,

856: the thermal energy operator $\epsilon$, with dimension 1/10,

857: the energy density of annealed vacancies $t$, with dimension 3/5, $\epsilon^{''}$,

858: with dimension 3/2, the magnetization $\sigma$, with dimension 3/80, 

859: and the subleading magnetization operator $\sigma^{'}$, with

860: dimension 7/16~\cite{DiFrancesco}. Their fusion rules can be found for instance in

861: Ref.~\cite{DiFrancesco}, pg.~224.

862: 

863: 

864: As we previously mentioned, the possible conformally invariant

865: boundary conditions can be classified by means of the fusion

866: hypothesis~\cite{Affleck:1990by,Affleck:1990iv,Affleck95}. Namely,

867: starting from the spectrum of a simple BC, one can obtain the spectra of other allowed BCs upon

868: fusion with primary fields of the CFTs. By comparing the low-energy spectra determined in this way with those obtained by NRG,

869: one can identify and characterize all fixed points of the model.

870: 

871: 

872: 

873: \subsection{Fixed points in the trimer phase diagram}

874: \label{Fixed points in the trimer phase diagram}

875: 

876: %%%%%%% POINT 2. REVISED FIGURE 11

877: \begin{figure}

878: \centerline{\includegraphics[width=12cm]{fabrizio_fig11.eps}}

879: \caption{Phase diagram of the trimer model (\ref{Ham-trimer}). All different phases are discussed in the text, and their 

880: properties briefly summarized in Table~\ref{Table trimer}. 

881: \label{trimer-phd}}

882: \end{figure}

883: %%%%% END REVISED FIGURE 11

884: 

885: In Fig.~\ref{trimer-phd}, we draw the phase diagram of the trimer

886: as obtained by NRG~\cite{Wilson75,Krishnamurthy80i,Krishnamurthy80ii}.

887: In order to have a classification

888: scheme which works equally well for Fermi-liquid and

889: non-Fermi-liquid phases, the fixed points are identified through

890: the zero-frequency values of the scattering $S$-matrices of the baths,

891: $(S_1,S_2,S_3)$, which can be obtained by

892: CFT~\cite{Affleck:1991tk,Affleck93} through the modular

893: $S$-matrix~\cite{DiFrancesco}. Note that, through Eq.~(\ref{rho.vs.S}),

894: the values of the scattering $S$-matrices give direct access to the values of the DOS at the chemical potential

895: of each impurity. We just recall that $S_a=-1$ means that the impurity-$a$ DOS has a Kondo resonance,

896: $S_a=1$ that it has a pseudo-gap, $\rho_a(\epsilon)\sim \epsilon^2$, while any intermediate value implies a

897: non-Fermi liquid behavior. 

898: %%%%%% POINT 2. TABLE TRIMER

899: The physical properties of the different phases are furthermore summarized in Table~\ref{Table trimer}. 

900: 

901: 

902: \begin{table}[htb]

903: \caption{Summary of the main physical properties of the different phases in Fig.~\ref{trimer-phd}, 

904: including the behavior of the zero-frequency DOSs for the three impurities, $\rho_i(0)$ with $i=1,2,3$, 

905: with respect to the non-interacting value $\rho_0$, and the dimension of the relevant 

906: symmetry-breaking single-particle operators. ``NOT'' means that the operator is not relevant, i.e. has dimension 

907: not smaller than one. $\phi$ is the golden ratio.

908: \label{Table trimer}}

909: \begin{indented}

910: \item[] \begin{tabular}{@{}llllll}

911: \br

912: ~~~ & $(0,1,0)$ & $\phi^{-2}\,(-1,1,-1)$ & $(-1,-1,-1)$ & $(0,-1,0)$ & $(1,-1,1)$ \\

913: \mr

914: $\rho_1(0)/\rho_0=\rho_3(0)/\rho_0$ & $1/2$ & $\left(1+\phi^{-2}\right)/2$ & $1$ & $1/2$ & 0 \\

915: %

916: $\rho_2(0)/\rho_0$ & 0 & $\left(1-\phi^{-2}\right)/2$ & $1$ & $1$ & 1 \\ 

917: %

918: $\mathbf{S}_1 + \mathbf{S}_3 - 2\mathbf{S}_2$ & 1/2 & 2/5 & NOT & NOT & NOT \\

919: %

920: $\mathbf{S}_1 - \mathbf{S}_3$ & NOT & NOT & NOT & 1/2 & NOT \\

921: %

922: $d^\dagger_1 d^\pdag_3$$^a$ & 1/2 & 3/5 & NOT & 1/2 & NOT \\

923: %

924: $d^\dagger_1 d^\dagger_3$$^a$ & 1/2 & 3/5 & NOT & 1/2 & NOT \\

925: %

926: $\left(d^\dagger_1+d^\dagger_3\right)d^\pdag_2$$^a$ & NOT$^{b}$ & 3/5 & NOT & NOT$^{b}$ & NOT \\

927: %

928: $\left(d^\dagger_1+d^\dagger_3\right)d^\dagger_2$$^a$ & NOT$^{b}$ & 3/5 & NOT & NOT$^{b}$ & NOT \\

929: \br

930: \end{tabular}

931: \item[]$^a$ These particle-hole and particle-particle operators are spin-singlets.

932: \item[]$^b$ These operators are not relevant in the sense that their dimension is not smaller than one. However, they 

933: do generate in perturbation theory one of the truly relevant perturbations, with dimension smaller than one, so that 

934: in reality they are relevant too.  

935: \end{indented}

936: \end{table}

937: %%%%%%%%% END TABLE TRIMER

938: 

939: 

940: 

941: 

942: Let us now present briefly the features of each fixed point. 

943: \\[1.5ex]

944: \noindent \textsl{\bf 1.\quad $\mathbf{(S_1,S_2,S_3)=(-1,-1,-1)}$}

945: \\[1.5ex]

946: This fixed point, that describes a conventional perfectly Kondo-screened

947: phase, will be used as the ancestor BC which, upon fusion with primary fields,

948: will provide all other BCs.

949: It is quite obvious that this phase exists and extends in a whole region around the

950: origin $J=J'=0$ in Fig.~\ref{trimer-phd}. Indeed, when $J=J'=0$,

951: each impurity is independently Kondo screened by its own

952: conduction bath and this perfect screening cannot be affected by

953: finite $J$ and $J'$ much smaller than the Kondo temperature. It is

954: far less obvious that this fixed point remains stable for

955: large $J\simeq J'$. When $J'=J\gg T_K$, the impurities lock into

956: two degenerate S=1/2 configurations. In the first, sites 1 and 3

957: are coupled into a triplet which in turn is coupled with site 2

958: into an overall spin-1/2 configuration. Since this is even by

959: interchanging 1 with 3, we denote it as $|e\rangle$. The other

960: configuration, which we denote as $|o\rangle$ as it is odd under

961: $1\leftrightarrow 3$, corresponds to sites 1 and 3 coupled into a

962: singlet, leaving behind the free spin-1/2 moment of site 2. The

963: Kondo exchange projected onto this subspace reads

964: \bea

965: &&

966: \frac{J_K}{3}\; |e\rangle\langle e |\; \mathbf{S}\cdot

967: \big(2\mathbf{J}_1(0) - \mathbf{J}_2(0) + 2\mathbf{J}_3(0)\big) \nonumber \\

968: && + J_K\;

969: |o\rangle\langle o |\; \mathbf{S}\cdot \mathbf{J}_2(0)

970: \nonumber \\

971: && - \frac{J_K}{\sqrt{3}}\,\Big( |e\rangle\langle o | + |o\rangle\langle e |\Big)\,

972: \mathbf{S}\cdot \big(\mathbf{J}_1(0) - \mathbf{J}_3(0)\big),

973: \label{J'= J>>0}

974: \eea

975: where $\mathbf{S}$ describes the effective S=1/2 of the trimer, while $\mathbf{J}_a(0)$

976: is the spin density of bath $a=1,2,3$ at the impurity site, assumed to be the origin.

977: All the above screening channels flow to strong coupling within a simple one-loop calculation.

978: Since it can be readily shown that the impurity can be perfectly screened, both in the spin

979: and in the even-odd channels, one has to conclude that the whole line $J=J'$ at finite $J_K$

980: corresponds to the Kondo screened fixed point $(-1,-1,-1)$, as in Fig.~\ref{trimer-phd}.

981: A small deviation from $J=J'$ is an irrelevant perturbation that splits the degeneracy between

982: $|e\rangle$ and $|o\rangle$. Only a finite deviation eventually destabilizes this fixed point, the faster

983: the smaller $J_K$.

984: \\[1.5ex]

985: \noindent \textsl{\bf 2. \quad $\mathbf{(S_1,S_2,S_3)=(0,1,0)}$}

986: %\label{(0,1,0)}

987: \\[1.5ex]

988: This fixed point occurs for $J\gg T_K,J'$, see

989: Fig.~\ref{trimer-phd}. The NRG spectrum is compatible with that 

990: obtained by fusing the $(-1,-1,-1)$ fixed point with the field $\sigma^{'}$

991: of the TIM. It is not difficult to realize that this fixed point

992: is equivalent to the non-Fermi liquid phase of the S=1/2

993: two-channel Kondo model~\cite{Affleck:1990by,Affleck:1990iv}.

994: Indeed if $J'=0$ and $J$ is large, the trimer locks into the

995: S=1/2 configuration which we previously denoted as $|e\rangle$, to

996: indicate the even parity upon $1\leftrightarrow 3$. The Kondo

997: exchange projected onto this configuration is, see Eq.~(\ref{J'=

998: J>>0}),

999: \be

1000: \mathbf{S}\cdot \sum_{a=1}^3 \, J_K^{(a)}\,

1001: \mathbf{J}_a(0) = \frac{J_K}{3}\, \mathbf{S}\cdot \big(2\mathbf{J}_1(0)

1002: - \mathbf{J}_2(0) + 2\mathbf{J}_3(0)\big).

1003: \label{J'=0 J>>0}

1004: \ee

1005: Hence, while baths 1 and 3 are still antiferromagnetically coupled, the

1006: coupling with bath 2 turns effectively ferromagnetic. The ordinary

1007: one-loop renormalization group calculation would predict that the

1008: Kondo exchanges $J_K^{(1)}=J_K^{(3)}>0$ flow towards strong

1009: coupling, while $J_K^{(2)}<0$ flows towards zero. This suggests

1010: that a model with $J_K^{(1)}=J_K^{(3)} \gg -J_K^{(2)} >0$ should

1011: behave asymptotically as (\ref{J'=0 J>>0}). If $J_K^{(2)}=0$ this

1012: is just the two-channel spin-1/2 impurity

1013: model~\cite{Affleck:1990by,Affleck:1990iv}, which is non

1014: Fermi-liquid with $S$-matrices

1015: $S_1=S_3=0$~\cite{Affleck:1991tk,Affleck93}. It is easy to show 

1016: that the small ferromagnetic $J_K^{(2)}$ transforms into an

1017: antiferromagnetic exchange of the form $\mathbf{J}_2(0)\cdot\left(\mathbf{J}_1(0)+\mathbf{J}_3(0)\right)$, 

1018: which is irrelevant. Consequently, we expect that

1019: this phase should remain non-Fermi liquid and identified by the

1020: $S$-matrices $(S_1,S_2,S_3)=(0,1,0)$, as indeed confirmed by CFT.

1021: In addition, through the modular $S$-matrix, one can show that the

1022: zero-temperature entropy $S(0) = 1/2 \, \ln 2$ is finite and

1023: coincides with that of the S=1/2 two channel Kondo model.

1024: Since $\sigma^{'}\times \sigma^{'} = I + \epsilon^{''}$, with the

1025: latter having dimension $3/2>1$, this fixed point is stable to

1026: symmetry-preserving perturbations.

1027: Yet, there are several

1028: symmetry-breaking relevant perturbations of dimension 1/2. One of

1029: them corresponds to the staggered magnetization

1030: \be

1031: \boldsymbol{J}_1- 2\boldsymbol{J}_2 +

1032: \boldsymbol{J}_3.

1033: \label{trimer-Ms}

1034: \ee

1035: %%%%%%%% MODIFIED TEXT TO MAKE STATEMENT CLEARER 

1036: All the other

1037: relevant operators break the degeneracy between bath 1 and 3,

1038: as for instance the spin-singlet operator

1039: \be

1040: \boldsymbol{J}_2 \cdot \big(\boldsymbol{J}_1 -

1041: \boldsymbol{J}_3\big),

1042: \label{trimer-s2(s1-s3)}

1043: \ee

1044: and the direct hopping or singlet-pairing between baths 1 and 3, both known to be relevant perturbations at the overscreened

1045: non-Fermi-liquid fixed point~\cite{Affleck92PRB}.  

1046: Note that, although the hopping/pairing operators between

1047: baths 1 and 2 as well as 3 and 2 have dimension one,   

1048: they do induce indirectly a symmetry-breaking coupling among baths 1 and 3 of dimension 1/2, 

1049: hence they are effectively relevant, specifically marginally relevant.

1050: This phase with $S$-matrices $(S_1,S_2,S_3)=(0,1,0)$ 

1051: extends at finite $J'$ just because $J'$ does not generate

1052: any symmetry-breaking relevant perturbation. 

1053: %%%%%%%% END MODIFIED TEXT

1054: 

1055: 

1056: The approach to the fixed point is controlled by two leading

1057: irrelevant operators of dimension 3/2: $\epsilon^{''}$ and the

1058: scalar product of the staggered magnetization (\ref{trimer-Ms})

1059: with the first spin descendant. Similarly to the overscreened

1060: two-channel Kondo model~\cite{Affleck:1990by,Affleck:1990iv}, these

1061: operators produce logarithmic singularities in the impurity

1062: contribution to the specific heat coefficient and to the magnetic

1063: susceptibility, $C_{imp}/T \sim \chi_{imp} \sim \ln(1/T)$.

1064: \\[1.5ex]

1065: \noindent \textsl{\bf 3.\quad $\mathbf{(S_1,S_2,S_3)= \phi^{-2}\, (-1,1,-1)}$}

1066: \\[1.5ex]

1067: Since the Kondo screened phase, $(-1,-1,-1)$ and the non-Fermi

1068: liquid one, $(0,1,0)$, are essentially different, it is clear

1069: that an unstable critical line separates the two, see

1070: Fig.~\ref{trimer-phd}. We find that the NRG spectrum can be

1071: reproduced by fusing the $(-1,-1,-1)$ fixed point with $\epsilon$ of the TIM.

1072: The $S$-matrices are $\phi^{-2}\,(-1,1,-1)$ and the residual

1073: entropy is $S(0) = \ln \phi $, where $\phi = (1+\sqrt{5})/2$ is

1074: the golden ratio. Since $\epsilon\times \epsilon = I + t$, the

1075: operator which moves away from the critical line has dimension

1076: 3/5. The most

1077: relevant symmetry breaking operator is still the staggered

1078: magnetization (\ref{trimer-Ms}), which has now dimension 2/5. Once

1079: more, the approach to this fixed point is controlled by the scalar

1080: product of the staggered magnetization with the first Kac-Moody

1081: descendant of the $SU(2)_3$ CFT, which has dimension $1+2/5$.

1082: Analogously to the multichannel

1083: Kondo~\cite{Affleck:1990by,Affleck:1990iv}, this operator produces

1084: impurity contributions to the specific-heat coefficient and

1085: magnetic susceptibility that diverge like $T^{-1/5}$.

1086: 

1087: The spin-singlet operator (\ref{trimer-s2(s1-s3)}) is also

1088: relevant, although with a larger dimension 3/5. In addition, there

1089: is a new class of dimension-3/5 operators which correspond to

1090: coupling into a spin-singlet two particles, or one hole and one

1091: particle, belonging to bath 2 and either bath 1 or 3.

1092: Finally, this critical line is stable towards moving away from particle-hole symmetry, as it was the case

1093: in the dimer.

1094: \\[1.5ex]

1095: \noindent \textsl {\bf 4.\quad $\mathbf{(S_1,S_2,S_3)=(1,-1,1)}$ and $\mathbf{(S_1,S_2,S_3)=(0,-1,0)}$}

1096: \\[1.5ex]

1097: These two fixed points occur when $J'>J$ is larger or comparable

1098: with the Kondo temperature. They have a very simple explanation.

1099: Indeed, when $J=0$, site 2 is only coupled to bath 2 with a Kondo

1100: exchange, leading to a full screening, i.e. $S_2=-1$. Sites 1 and

1101: 3 plus their own baths realize a two-impurity Kondo

1102: model which, as discussed before, has two stable regimes. One is

1103: Kondo screened, $S_1=S_3=-1$, for $J'\ll T_K$,

1104: $(S_1,S_2,S_3)=(-1,-1,-1)$ in Fig.~\ref{trimer-phd}, and the other

1105: unscreened for $J'\gg T_K$, $(S_1,S_2,S_3)=(1,-1,1)$ in

1106: Fig.~\ref{trimer-phd}. These two regimes are stable

1107: towards switching on a small $J\ll J'$. When $J=0$, we also know

1108: that an unstable fixed point at $J'=J'_*\sim T_K$ separates these

1109: two stable phases, which is identified by $S_1=S_3=0$, hence the

1110: label (0,-1,0) in Fig.~\ref{trimer-phd}. Since site 2 is tightly

1111: bound into a singlet state with bath 2, a finite but small $J\ll

1112: T_K$ will simply generate a ferromagnetic exchange of order

1113: $-J^2/T_K$ by virtually exciting the singlet state. The net effect

1114: is that the unstable fixed point at $J=0$ is just the endpoint of

1115: another critical line which, for $J\ll T_K$, moves to larger

1116: values of $J'$. From the CFT viewpoint, the $(1,-1,1)$ and

1117: $(0,-1,0)$ fixed points can be obtained by fusing with $\epsilon^{''}$ of the TIM

1118: or $\sigma_I$ of the Ising CFT, respectively. The properties of

1119: the unstable $(0,-1,0)$ critical line are the same as those of the

1120: dimer critical point. In particular there is a relevant operator

1121: in the singlet Cooper channel that now involves pairing among

1122: baths 1 and 3, as well as an equally relevant operator which

1123: corresponds to an opposite magnetization of bath 1 and 3, i.e.

1124: $\mathbf{J}_1 - \mathbf{J}_3$.

1125: 

1126: 

1127: \subsection{Dynamical properties of the trimer}

1128: 

1129: In the $J$-$J'$ parameter-space of the trimer model (\ref{Ham-trimer}), the

1130: $J'=0$ line is actually pertinent to a Hubbard model on a bipartite lattice and with nearest neighbor hopping

1131: simulated within cluster-DMFT. In this case, the phase diagram is qualitatively similar to that of the

1132: dimer model, see Fig.~\ref{dimerpdmu}, apart from the fact that the unscreened phase corresponds now to the

1133: $(S_1,S_2,S_3)=(0,1,0)$ non-Fermi liquid phase, while the critical line represents the

1134: $(S_1,S_2,S_3)= \phi^{-2}\, (-1,1,-1)$ unstable fixed point.

1135: 

1136: Both the critical line and the non-Fermi liquid phase are unstable to a nearest-neighbor hopping. 

1137: If, instead of three impurities coupled by $J$, we consider three impurities coupled by a hopping,

1138: the transition from the Kondo screened phase to the non-Fermi liquid one transforms into a crossover

1139: between Fermi-liquid regimes, exactly like in the dimer. In Fig.~\ref{Sigmatrimer}, we draw the self-energies

1140: in Matsubara frequencies of the 3 impurities coupled by the spin-exchange $J$ (panels (a) and (b)),

1141: as well as of the impurities coupled by a hopping $t_\perp$ along the bonds 1-2 and 2-3 (panels (c) to (f)). The Hamiltonian parameters

1142: are $U=8$, $J=0.00125$ and $t_\perp=0.05$, like in Figs.~\ref{SigmaexE} and \ref{Sigma}, while $\Gamma= 0.4,~0.44,~0.5$.

1143: \begin{figure}

1144: \centerline{\includegraphics[width=12cm]{fabrizio_fig12.eps}}

1145: \caption{Non-vanishing self-energies of the trimer. The three curves correspond to $\Gamma=0.5$ (green), 0.44 (red) and 0.4 (black). Panels (a) $\Ima \Sigma_{11}(i\omega) = \Ima \Sigma_{33}(i\omega)$

1146: and (b) $\Ima \Sigma_{22}(i\omega)$, for the case of 3 impurities coupled by the spin exchange $J$.

1147: Panels (c) $\Ima \Sigma_{11}(i\omega) = \Ima \Sigma_{33}(i\omega)$, (d) $\Ima \Sigma_{22}(i\omega)$,

1148: (e) $\Rea \Sigma_{12}(i\omega) = \Rea \Sigma_{32}(i\omega)$ and (f) $\Ima \Sigma_{13}(i\omega)$, for the case

1149: of impurities coupled by a single-particle hopping $t_\perp$, leading to the same value of $J$ as before.

1150: We used a discretization parameter $\Lambda = 3.0$.

1151: \label{Sigmatrimer}}

1152: \end{figure}

1153: In the absence of $t_\perp$, the self-energies, which are diagonal and imaginary, behave as $\Ima \Sigma_{aa}(i\omega) \sim -\omega$

1154: in the Kondo screened phase, $\Gamma=0.5$, and tend to a constant at the fixed point, $\Gamma\simeq 0.44$. In the unscreened

1155: phase, $\Gamma=0.4$, $\Ima \Sigma_{11}(i\omega) = \Ima \Sigma_{33}(i\omega) \to \mathrm{const.}$

1156: while $\Ima \Sigma_{22}(i\omega) \sim -1/\omega$, in accordance with the values of the scattering $S$-matrices.

1157: Similarly to the dimer, when the impurities are instead coupled

1158: by a single-particle hopping, a Fermi-liquid behavior $\Ima \Sigma_{aa}(i\omega) \sim -\omega$ is eventually recovered at very low

1159: energy, see panels (c) and (d), although the DOS at site 2 may still display a pseudo-gaped behavior due to the

1160: large value of $\Rea \Sigma_{12}(i\omega) = \Rea \Sigma_{32}(i\omega)$.

1161: 

1162: In conclusion, the dynamical behavior with $J$ or in the presence of $t_\perp$ is qualitatively

1163: similar to the 2-impurity cluster.

1164: 

1165: 

1166: 

1167: \section{The impurity tetramer}

1168: 

1169: Let us move finally to the last type of cluster investigated, the tetramer drawn in Fig.~\ref{Fig-tetramer}

1170: with the Hamiltonian

1171: \bea

1172: \mathcal{H} &=& \sum_{a=1}^4\,\mathcal{H}_a^K\, +

1173: J\,\left(\mathbf{S}_1+\mathbf{S}_3\right)\cdot \left(\mathbf{S}_2 + \mathbf{S}_4\right)\nonumber \\

1174: && + J'\,\left(\mathbf{S}_1\cdot \mathbf{S}_3 + \mathbf{S}_2\cdot \mathbf{S}_4\right).

1175: \label{Ham-tetramer}

1176: \eea

1177: \begin{figure}

1178: \centerline{\includegraphics[width=10cm]{fabrizio_fig13.eps}}

1179: \caption{\label{Fig-tetramer} The 4-impurity cluster}

1180: \end{figure}

1181: This model now describes four

1182: spin-1/2 impurities, coupled together by nearest $J$, and next-nearest

1183: neighbor $J'$, antiferromagnetic exchanges. In addition, each spin is Kondo-coupled

1184: to a conduction bath by $J_K>0$. The four baths are once more assumed to be degenerate

1185: and particle-hole invariant. As before, the impurities are for convenience only coupled through a spin-exchange

1186: and not by hopping terms, which we will take into account as perturbations.

1187: 

1188: 

1189: 

1190: \subsection{CFT preliminaries for the tetramer}

1191: \label{CFT preliminaries for the tetramer}

1192: 

1193: Given our choice of the model (\ref{Ham-tetramer}), the charge degrees of

1194: freedom can be still represented by four independent $SU(2)_1$ CFTs, one for each bath.

1195: Concerning the spin degrees of freedom, the way in which the impurities are exchange-coupled 

1196: naturally leads to the following conformal embedding scheme

1197: \bea

1198: && \left(SU(2)_1^{(1)} \times SU(2)_1^{(3)}\right)\times

1199: \left(SU(2)_1^{(2)} \times SU(2)_1^{(4)}\right) \nonumber \\

1200: && \rightarrow

1201: \left(SU(2)_2^{(1-3)} \times Z_2^{(1)}\right)\times

1202: \left(SU(2)_2^{(2-4)} \times Z_2^{(2)}\right)\nonumber\\

1203: && \rightarrow SU(2)_4 \times Z_2^{(1)}\times Z_2^{(2)}

1204: \times \Bigg[c=1 \left(\mbox{CFT}\right)_{p'=6}\Bigg]

1205: \label{embedding plaquette}

1206: \eea

1207: where $c=1$ CFT stands for the 

1208: $Z_2$ orbifold of a free bosonic CFT (the bosonic field $\phi$ and $-\phi$ must be identified) with compactification radius

1209: $R = \sqrt{2p'}$ and $p'=6$~\cite{DiFrancesco,affleck-2001-594}.

1210: The two step process represented by~\ref{embedding plaquette} corresponds to the

1211: coupling of the $SU(2)_1$ spin sectors of the impurities on the diagonals into $SU(2)_2$,

1212: followed by the coupling of these two new sectors into an $SU(2)_4$. The resulting cosets

1213: are the two Ising sectors and the $c=1$ CFT.

1214: The embedding

1215: can again be rigorously proven through the character decomposition, but the proof is very technical 

1216: so we prefer not to give any detail. 

1217: It is also convenient to represent the two Ising sectors as a

1218: single $c=1$ free bosonic CFT, now with compactification radius

1219: $R = \sqrt{4}$, i.e. $p'=2$.

1220: 

1221: The $Z_2$ orbifold of a $c=1$ CFT with compactification radius $R=\sqrt{2p'}$ includes, besides the

1222: identity, the following primary fields~\citep[pg.~785]{DiFrancesco}:

1223: \begin{itemize}

1224: \item[(i)] $p'-1$ fields, $\phi_{h}$, with dimensions $h=\lambda^2/4p'$, with $\lambda=1,\dots,p'-1$;

1225: \item[(ii)] a doubly-degenerate field, $\phi^{(a)}_{p'/4}$, $a=1,2$, with dimension $p'/4$;

1226: \item[(iii)] the twist operators $\sigma^{(a)}$ and $\tau^{(a)}$, $a=1,2$, with dimensions

1227: 1/16 and 9/16, respectively;

1228: \item[(iv)] the dimension-1 operator $\theta$.

1229: \end{itemize}

1230: The fusion rules among the primary fields can be found for

1231: instance in Ref.~\cite[pg.~786]{DiFrancesco}.

1232: 

1233: The embedding

1234: \be

1235: SU(2)_2 \times SU(2)_2 \rightarrow SU(2)_4 \times \Bigg[c=1 (\mathrm{CFT})_{p'=6}\Bigg],

1236: \label{embed-Georges}

1237: \ee

1238: has already been discussed in Ref.~\cite{Georges&Sengupta} in the context of the 2-impurity 2-channel Kondo model.

1239: We will see in what follows that some of the anomalies found by those authors do also appear in our model.

1240: 

1241: 

1242: 

1243: \subsection{Fixed points in the tetramer phase diagram}

1244: 

1245: %%%%% POINT 2. REVISED FIGURE 14

1246: \begin{figure}

1247: \centerline{\includegraphics[width=12cm]{fabrizio_fig14.eps}}

1248: \caption{\label{tetramer-phd} Phase diagram of the

1249: impurity tetramer described by the Hamiltonian (\ref{Ham-tetramer}). All phases are discussed in the text, and their 

1250: properties summarized in Table~\ref{Table tetramer}. Since the calculation is numerically heavy, we 

1251: had to use a large $\Lambda=10$, which is sufficient to characterize the low energy spectra, hence the various phases, but 

1252: not adequate to provide accurate estimates of the critical points (for instance, when $J=0$, 

1253: we find $(J'/T_k)_*\simeq 0.24$ instead of the two-impurity value $\simeq 2$~\cite{Jones89}). 

1254: However, even though the absolute values of the critical $J/T_K$ and $J'/T_K$ are underestimated, 

1255: we still believe that the overall shape of the phase diagram should be representative. For this reason, 

1256: we have decided to plot the NRG data rescaled in such a way that, when $J=0$, 

1257: the critical point has the value found in the two-impurity model.~\cite{Jones89}}

1258: \end{figure}

1259: %%%%% END REVISED FIGURE 14

1260: 

1261: In Fig.~\ref{tetramer-phd}, we draw the phase diagram of

1262: (\ref{Ham-tetramer}) as obtained by NRG. As before, each fixed point

1263: is identified by the $S$-matrices $(S_1,S_2,S_3,S_4)^{(n)}$, where

1264: the superscript $(n)$ is introduced to distinguish between

1265: different fixed points with the same $S$-matrices. 

1266: %%%%%%% POINT 2. TABLE TETRAMER

1267: In Table~\ref{Table tetramer} the main physical properties of the different phases are summarized.

1268: 

1269: \begin{table}[htb]

1270: \caption{Summary of the main physical properties of the different phases in Fig.~\ref{tetramer-phd}.  

1271: $\rho(0)$ is the zero-frequency DOS of any of the four impurities,  $\rho_0$ being 

1272: its non-interacting value. As in Table~\ref{Table trimer}, several single-particle 

1273: operators are considered, specifying whether they are relevant, in which case their dimension is 

1274: indicated, or not. 

1275: $d^\dagger_i d^\pdag_{i+1}$ and $d^\dagger_i d^\dagger_{i+1}$ denotes hopping and singlet-pairing, 

1276: respectively, along the sides of the plaquette, $i - i+1$ meaning 

1277: $1-2$, $2-3$, $3-4$ or $4-1$.  $d^\dagger_i d^\pdag_{i+2}$ and $d^\dagger_i d^\dagger_{i+2}$  

1278: have the same meaning but along the diagonals, i.e. $i - i+2$ stems for $1-3$ and $2-4$. $x$ is a 

1279: parameter (not to be confused with the coordinate $x$ that is used in the first row to identify the different phases, according to 

1280: Fig.~\ref{tetramer-phd}) that changes continuously along the critical lines I and II, reaching at the first order point 

1281: along the diagonal $J=J'$ the value $x=1$ for line I and $x=2$ for line II.  \\[1.5ex]

1282: \label{Table tetramer}}

1283: %\begin{indented}

1284: %\item[] 

1285: \begin{tabular}{@{}llllll}

1286: \br

1287: ~~~~ &  $(x^2-y^2)$ & line II & screened & line I & $(xy)$ \\

1288: \mr

1289: $\rho(0)/\rho_0$ & 0 & 1/2 & 1 & 1/2 & 0 \\

1290: $\mathbf{S}_1-\mathbf{S}_2+\mathbf{S}_3-\mathbf{S}_4$ & NOT & $1/3 + x^2/6$ & NOT & NOT & NOT \\

1291: $\mathbf{S}_1 - \mathbf{S}_3$ & NOT & NOT & NOT & $1/2 + x^2/2$ & NOT \\

1292: $\mathbf{S}_2 - \mathbf{S}_4$ & NOT & NOT & NOT & $1/2 + x^2/2$ & NOT \\

1293: $d^\dagger_i d^\pdag_{i+1}$ & NOT & $5/8 + (1-2x)^2/24$ & NOT & NOT & NOT \\

1294: $d^\dagger_i d^\dagger_{i+1}$ & NOT & $5/8 + (1-2x)^2/24$ & NOT & NOT & NOT \\

1295: $d^\dagger_i d^\pdag_{i+2}$ & NOT & NOT & NOT & $1/2 + x^2/2$ & NOT \\

1296: $d^\dagger_i d^\dagger_{i+2}$ & NOT & NOT & NOT & $1/2 + x^2/2$ & NOT \\

1297: \br

1298: \end{tabular}

1299: %\end{indented}

1300: \end{table}

1301: %%%%%%%%%%% END TABLE TETRAMER

1302: \noindent \textsl{\bf 1.\quad $\mathbf{(S_1,S_2,S_3,S_4) = (-1,-1,-1,-1)}$}

1303: \\[1.5ex]

1304: This fixed point corresponds to a perfectly Kondo screened phase.

1305: It occurs when $T_K$ is large compared with both $J$ and

1306: $J'$. Once again, we will use the Kondo screened $(-1,-1,-1,-1)$ fixed

1307: point as the ancestor BC to generate all the others through

1308: fusion.

1309: \\[1.5ex]

1310: \noindent \textsl{\bf 2.\quad $\mathbf{(S_1,S_2,S_3,S_4) = (1,1,1,1)^{(1)}}$}

1311: \\[1.5ex]

1312: If $J'=0$ and $J\gg T_K$, the tetramer locks into a non-degenerate

1313: singlet state which is obtained by coupling sites 1 and 3 into a

1314: triplet, as well as sites 2 and 4, and coupling the two triplets

1315: into an overall singlet. This configuration transforms like the function $x^2-y^2$ under the $C_4$ point-group

1316: of the plaquette, hence the label in Fig.~\ref{tetramer-phd}. The singlet decouples from the

1317: conduction electrons which do not feel the presence of the

1318: impurities anymore, resulting in a phase-shift $\delta=0$ in every channel. This phase is Fermi-liquid-like 

1319: and remains stable even in the presence of a finite $J'$, provided the lowest

1320: excitation gap from the ground state of the isolated tetramer is

1321: much larger than $T_K$. Within CFT, there are several possible

1322: fusions which lead the $(-1,-1,-1,-1)$ fixed point to this new

1323: one. One is for instance the fusion with the primary

1324: field $\theta$ of the $c=1$ CFT with $p'=6$.

1325: \\[1.5ex]

1326: \noindent \textsl{\bf 3. \quad $\mathbf{(S_1,S_2,S_3,S_4) = (1,1,1,1)^{(2)}}$}

1327: \\[1.5ex]

1328: If $J=0$, sites 1 and 3 are decoupled from sites 2 and 4, hence the tetramer

1329: reduces to two independent Kondo dimers. If $J'\gg T_K$ each pair of impurities,

1330: 1 and 3 or 2 and 4, is strongly bound into a singlet which decouples from the

1331: conduction electrons. This impurity configuration transforms like $xy$ under the $C_4$ point group,

1332: which explains the label in Fig.~\ref{tetramer-phd}.

1333: This fixed point is obviously stable to a small $J$ being turned on, hence,

1334: in analogy with the dimer, it should be obtainable by the $(-1,-1,-1,-1)$ fixed point upon

1335: fusion with $\epsilon_{I}^{(1)}\,\epsilon_{I}^{(2)}$, where $\epsilon_I^{(a)}$, $a=1,2$,

1336: are the energy operator of the two Ising CFTs. This is also the dimension-1

1337: primary field $\theta$ of the $c=1$ CFT with $p'=2$.

1338: The NRG spectrum agrees with this prediction

1339: not only for small $J$, but for the whole region $J'>J$ with $J'\gg T_K$, see

1340: Fig.~\ref{tetramer-phd}.

1341: \\[1.5ex]

1342: \noindent \textsl{\bf 4. \quad First order line}

1343: \\[1.5ex]

1344: If $J=J'\gg T_K$, the tetramer locks into a doubly degenerate spin-singlet, the states

1345: with symmetry $x^2-y^2$ and $xy$ previously mentioned.

1346: The Kondo exchange provides a coupling between these two

1347: configurations only at second order, namely via a quartic conduction-electron

1348: operator, which is therefore irrelevant. Hence the tetramer decouples asymptotically

1349: from the conduction baths, and its degeneracy remains untouched. This is confirmed by

1350: the NRG calculation, which shows the same Fermi liquid spectrum as in the absence of

1351: the impurity-cluster apart from each state being doubly degenerate. This phase is

1352: the analogous of a first order line, hence its name in

1353: Fig.~\ref{Fig-tetramer}, with a relevant operator of dimension 0 that

1354: describes the splitting of the double degeneracy of the tetramer.

1355: \\[1.5ex]

1356: \noindent \textsl{\bf 5. \quad $\mathbf{(S_1,S_2,S_3,S_4) = (0,0,0,0)^{(1)}}$ and $\mathbf{(0,0,0,0)^{(2)}}$}

1357: \\[1.5ex]

1358: The Kondo screened phase at small $J$ and $J'$ is essentially different from the

1359: two unscreened phases at large $J>J'$ and large $J'>J$, respectively.

1360: Hence there are two critical lines that start from the $J'=0$ axis as well as

1361: from the $J=0$ one, see Fig.~\ref{tetramer-phd}, and finally merge with the first order line

1362: at large $J=J'$. This might happen either through a multicritical point or by a gradual evolution

1363: of each line into a first order critical point. The latter scenario is actually realized, since,

1364: unlike in the trimer model, the NRG low-energy spectra along both critical lines

1365: varies continuously, signaling the existence of marginal perturbations.

1366: For the same reason, a precise identification of these critical lines with appropriate boundary CFTs is

1367: not a simple task.

1368: 

1369: Let us start from the simplest case at $J=0$, which corresponds to two independent dimers, sites 1 plus 3

1370: and sites 2 plus 4. The fixed point which separates the Kondo screened phase from the unscreened one is

1371: obviously the superposition of the fixed points of each dimer, discussed previously.

1372: It is obtained by the Kondo screened fixed point upon

1373: fusion with the product $\sigma_{I}^{(1)}\,\sigma_{I}^{(2)}$ of the two Ising CFTs~\cite{Affleck92PRL},

1374: and is identified by zero scattering matrices, hence $(0,0,0,0)^{(2)}$ in Fig.~\ref{tetramer-phd}, as well as by a 

1375: residual $\ln 2$ entropy.

1376: We already showed that the fixed point of a dimer can be destabilized only by the symmetry-breaking operators

1377: (\ref{dimer-AF})-(\ref{dimer-tperp}), which are not generated by a small $J$.

1378: Therefore, a finite $J\ll J'$ does not spoil the unstable fixed point $(0,0,0,0)^{(2)}$, but only

1379: moves its position to larger $J'$, as it generates a weak ferromagnetic exchange along each dimer.

1380: We notice that, upon double fusion,

1381: \ba

1382: && \Bigg(\sigma_{I}^{(1)}\,\sigma_{I}^{(2)}\Bigg) \times \Bigg(\sigma_{I}^{(1)}\,\sigma_{I}^{(2)}\Bigg) =

1383: I + \epsilon_{I}^{(1)} + \epsilon_{I}^{(2)} + \epsilon_{I}^{(1)}\,\epsilon_{I}^{(2)} \\

1384: && \equiv I + \phi_{1/2} + \theta,

1385: \ea

1386: where the last expression on the right-hand side is written in terms of the corresponding fields

1387: of the $p'=2$, $c=1$ CFT. In agreement with NRG, the operator content includes the marginal

1388: operator $\theta$, besides the dimension-1/2 relevant operator that moves away from the fixed point.

1389: Since for two independent dimers we do know that there is no such marginal operator at the unstable fixed point,

1390: we must conclude that $\theta$ acquires a finite coupling-constant only for $J \neq 0$. This situation, which is quite

1391: exceptional in impurity models, resembles that found in Ref.~\cite{Georges&Sengupta} in the 2-impurity 2-channel Kondo model.

1392: An important discovery of Ref.~\cite{Georges&Sengupta} was that this marginal operator not only

1393: influences the spectrum but also the operator content.

1394: Specifically, Georges and Sengupta recognized, by abelian bosonization of the model, that the fixed point Hamiltonian

1395: in the presence of the marginal operator is similar to an X-ray edge problem in bosonization

1396: language~\cite{Schotte}. Therefore any operator which

1397: involves creation or annihilation of the corresponding ``core-hole'' acquires an additional dimension.

1398: It is not difficult to realize that the same happens in our case. Indeed the dimension-1/2 relevant operator

1399: can be mapped within bosonization~\cite{Sasha} into the operator

1400: \be

1401: \epsilon_{I}^{(1)} + \epsilon_{I}^{(2)} \rightarrow d^\dagger\,\Psi(0) + \Psi(0)^\dagger\,d,

1402: \label{X-ray-1}

1403: \ee

1404: which represents the creation (annihilation) of a core-electron, $d^\dagger$($d$), and the contemporary

1405: annihilation (creation) of a conduction electron at the core-hole site, $\Psi(0)$ $\left(\Psi(0)^\dagger\right)$.

1406: Analogously, the marginal operator transforms like

1407: \[

1408: \epsilon_{I}^{(1)}\,\epsilon_{I}^{(2)} \rightarrow \left(1-d^\dagger\,d\right)\;\Psi(0)^\dagger\,\Psi(0),

1409: \]

1410: which corresponds to the interaction between the core-hole and the conduction electrons.

1411: In the presence of this term, the dimension of the relevant operator (\ref{X-ray-1}) changes according to

1412: \[

1413: \frac{1}{2} \rightarrow \frac{(2-2x)^2}{8},

1414: \]

1415: where $x$ parametrizes the critical line, and is actually related to the phase-shift induced by the core-hole

1416: in the equivalent X-ray edge problem. Since the end-point at $J=J'$ is expected to be a first order one,

1417: we conclude that $x$ moves from 0, $J=0$, to 1, $J=J'$, along the critical line. The dimensions of all the other

1418: dimension-1/2 symmetry-breaking operators changes instead as

1419: \[

1420: \frac{1}{2} \rightarrow \frac{1}{2} + \frac{(2x)^2}{8},

1421: \]

1422: so that they all become marginal at $J=J'$. Notice that, since the twist operators are not affected by the

1423: marginal perturbation, the $S$-matrices do not change along the line.

1424: 

1425: Concerning the other critical line, $(S_1,S_2,S_3,S_4) = (0,0,0,0)^{(1)}$ in Fig.~\ref{tetramer-phd},

1426: we find that, when $J'=0$, the NRG spectrum is reproduced with a good approximation by fusing the 

1427: $(-1,-1,-1,-1)$ BC with the primary field $\phi_{1/6}$ of the

1428: $p'=6$, $c=1$ CFT. Once again, since $\phi_{1/6}\times \phi_{1/6} = I + \theta + \phi_{2/3}$, the operator content

1429: includes, besides the relevant operator of dimension 2/3 that moves away from criticality,

1430: the marginal operator $\theta$ of the $p'=6$, $c=1$ CFT, which explains 

1431: the continuous evolution of the NRG spectrum from $J'=0$ to $J'=J$. The role of this marginal operator should be

1432: similar to its analogous on the other critical line. Therefore, following the previous analysis and

1433: in accordance with Ref.~\cite{Georges&Sengupta}, we expect the critical line to be parametrized by a

1434: ``phase-shift'' $x$ that modifies not only the spectrum but also the operator content. In particular, the

1435: dimension of the operator $\phi_{2/3}$ that moves away from the critical line changes according to

1436: \[

1437: \frac{2}{3} \rightarrow \frac{(4-2x)^2}{24}.

1438: \]

1439: Since this operator eventually acquires vanishing dimension at $J'\to J$, we conclude that $x\to 2$ at the

1440: end-point. The precise determination of $x$ along the line is however difficult to extract from our NRG spectra.

1441: 

1442: At $x=0$, the most relevant symmetry breaking operator

1443: would correspond to the staggered magnetization

1444: \be

1445: \boldsymbol{J}_1-\boldsymbol{J}_2+\boldsymbol{J}_3- \boldsymbol{J}_4,

1446: \label{tetramer-Ms}

1447: \ee

1448: with dimension 1/3, which, at finite $x$, changes into

1449: \[

1450: \frac{1}{3} + \frac{(2x)^2}{24},

1451: \]

1452: hence becomes marginal at $J=J'$. This is physically sound, since at $J=J'$ there is maximum spin frustration.

1453: Besides the staggered magnetization, there are other less relevant

1454: symmetry-breaking operators of dimension 2/3 at $x=0$. One of them is the singlet four-fermion operator

1455: \[

1456: \big(\boldsymbol{J}_1 - \boldsymbol{J}_3\big)\cdot

1457: \big(\boldsymbol{J}_1 - \boldsymbol{J}_2 + \boldsymbol{J}_3

1458: - \boldsymbol{J}_4\big) + (1,3)\leftrightarrow (2,4),

1459: \]

1460: whose dimension at $x\not =0$ is $1/2 + (2+2x)^2/24$, thus becoming soon irrelevant.

1461: 

1462: The other symmetry-breaking operators of dimension 2/3 correspond actually to all possible

1463: mean-field decoupling schemes of the exchange term

1464: \[

1465: \big(\boldsymbol{J}_1 + \boldsymbol{J}_3\big)\cdot

1466: \big(\boldsymbol{J}_2 +  \boldsymbol{J}_4\big),

1467: \]

1468: into inter-bath single-particle operators. Among them, we just mention the inter-bath hopping,

1469: \[

1470: \sum_\sigma\, \big(c^\dagger_{1\sigma} + c^\dagger_{3\sigma}\big)

1471: \big(c^\pdag_{2\sigma} + c^\pdag_{4\sigma}\big) + H.c.,

1472: \]

1473: as well as the $d$-wave Cooper pairing,

1474: \[

1475: \big(c^\dagger_{1\uparrow}-c^\dagger_{3\uparrow}\big)

1476: \big(c^\dagger_{2\downarrow} - c^\dagger_{4\downarrow}\big) -

1477: \big(\uparrow \, \leftrightarrow \, \downarrow\big).

1478: \]

1479: They are all degenerate and, at $x\not = 0$, have dimension

1480: \[

1481: \frac{5}{8} + \frac{(1-2x)^2}{24}.

1482: \]

1483: At $x=2$ they become marginal, but, interestingly enough, their dimension is non-monotonic in $x$,

1484: although always greater than the staggered magnetization (\ref{tetramer-Ms}). Along this line, too,

1485: the $S$-matrices and the residual entropy remain constant and equal to  $(S_1,S_2,S_3,S_4) = (0,0,0,0)$ and $\ln 2$, respectively.  

1486: 

1487: 

1488: Like in the dimer and trimer examples, the relevance of the inter-bath hopping implies that 

1489: both critical lines are no more accessible if, instead of four impurities coupled by a spin-exchange,

1490: one considers four impurities coupled by a single particle hopping, which is the actual situation

1491: within cluster-DMFT. Unfortunately, in this case we cannot obtain reliable spectral functions by NRG because of numerical

1492: limitations. Therefore, we cannot verify whether, in spite of the fact that the critical point

1493: is washed out, a sizable critical region still survives. However we tend to believe that it is the case,

1494: just like in the previous examples.  Finally, since both critical lines are stable towards the conventional

1495: particle-hole symmetry breaking, the phase diagram for $J\not = J'$, as function of $U/\Gamma$ and of

1496: the average impurity-occupancy, still looks like the dimer one, see Fig.~\ref{dimerpdmu}.

1497: 

1498: 

1499: 

1500: 

1501: 

1502: \section{Discussion and conclusions}

1503: 

1504: In this Topical Review, we attempted to uncover the key features that distinguish Anderson impurity clusters from

1505: single-impurity models and that could play an important role within cluster dynamical mean-field theory as opposed to

1506: its original single-site formulation.

1507: All the examples that we have studied, namely 2-, 3- and 4-impurity clusters, share very similar properties.

1508: 

1509: In particular, if the impurities within the cluster are coupled to one another by a two-body spin-exchange 

1510: while each of them is hybridized with

1511: its own separate conduction bath, the phase diagrams as function of the average impurity occupancy and of

1512: the Hubbard $U$ are practically the same, see Fig.~\ref{dimerpdmu}. For $U/\Gamma$ less than a critical value, perfect

1513: Kondo screening occurs and the impurity-cluster spectral functions show the conventional Kondo resonance.

1514: Above that critical value, the inter-impurity exchange prevails instead and takes care of freezing out the impurity degrees

1515: of freedom. Here, the impurity spectral functions develop a pseudo-gap at 

1516: the chemical potential, which is gradually filled in by ``doping'', i.e., by

1517: moving the average impurity occupancy away from half-filling. These two regimes are separated by a critical line

1518: that is identified by several instability channels. In all cases, the instability channels correspond to

1519: all possible mean-field decoupling schemes of the spin-exchange into bilinear operators. They include the

1520: intra-bath magnetization, staggered according to the signs and relative strengths of the spin-exchange constants, 

1521: and all singlet inter-bath bilinear operators, like the inter-bath hoppings or singlet Cooper pairs.

1522: The trend from the dimer towards the tetramer is towards a prevailing instability in the staggered magnetization

1523: channel. The dynamics across the critical point is basically controlled by \underline{two} separate energy scales. One of them, which we

1524: denoted as $T_+$, is finite across the transition and roughly of the order of the spin-exchange.

1525: The other, $T_-$, is the scale generated by the deviation $X$ from the critical line. It vanishes 

1526: as $T_- \sim |X|^\alpha$, where, in the most interesting case of the tetramer, the exponent

1527: $1\leq \alpha \leq 3$, is

1528: non-universal and depends on the frustration. From the point of view of the impurity spectral functions,

1529: see e.g. Fig.~\ref{DOSexE},

1530: $T_+$ is the width of a broad incoherent peak within the Hubbard side-bands, smooth across the critical point.

1531: On the contrary, $T_-$ is the width of the Kondo-like resonance that develops on top of the broader one, in the

1532: Kondo screened phase. As the critical point is approached, the Kondo resonance becomes narrower, and

1533: eventually disappears right at the critical point, leaving behind only the incoherent peak. In the

1534: unscreened phase, $T_-$ controls the width of the pseudo-gap that opens inside the incoherent part.

1535: 

1536: If the impurities inside a cluster are coupled one another by a single-particle hopping, $t_\perp$, instead of a spin-exchange, the transition

1537: turns into a crossover from the Kondo-resonance behavior to the pseudo-gaped one. Indeed, the

1538: inter-impurity hopping plays a double role. On the one hand it generates, for large $U$, a spin-exchange, $J=4t_\perp^2/U$,

1539: that might drive the model across the critical point. On the other hand, it also induces

1540: a small inter-bath hybridization, $V\sim J_K\, t_\perp/U$, that is a relevant perturbation and destabilizes 

1541: the critical point. 

1542: Specifically, $V$ cuts off all critical point singularities below an energy scale $E_{cut-off}\sim V^\beta$, where

1543: $\beta\geq 3$ in the tetramer and $\beta=2$ in the dimer. Since $V$ is small

1544: and $\beta$ large, we expect that, in spite of the critical point

1545: being no longer accessible, a ``critical region'' should still survive if $E_{cut-off}\ll T_+\sim J$.

1546: We indeed found evidences in favor of this scenario both in the dimer as well as in the trimer, where

1547: the impurity spectral functions are numerically accessible.

1548: 

1549: \bigskip

1550: 

1551: Coming back to our original scope, let us imagine that we implement a cluster-DMFT simulation of a Hubbard model

1552: using a dimer, a trimer or a tetramer as representative clusters. As $U$ increases driving the model towards

1553: the Mott transition, the effective impurity cluster must necessarily go through the

1554: above mentioned critical region (even if a true criticality is, rigorously speaking, not accessible since

1555: the impurities as well as the baths are coupled by a single-particle hopping).

1556: In this region, the instability channels of the avoided critical point will be amplified and, after the DMFT self-consistency,

1557: can induce a true bulk instability in the lattice model. At half-filling, our results on the impurity clusters suggest

1558: that most likely magnetism appears, even in the presence of frustration. However, since instabilities in particle-hole channels

1559: are weakened or removed by doping away from commensurate fillings, while that weakening does not happen in particle-particle channels, 

1560: a superconducting dome may emerge near half-filling. Indeed, recent cluster DMFT simulations of the

1561: Hubbard model on a square lattice~\cite{Jarrell-2005,Capone-2006} found evidence of a $d$-wave superconducting phase

1562: away from half-filling, in close analogy with the phase diagram of high-$T_c$ superconductors.

1563: 

1564: Nevertheless, irrespectively of which symmetry-broken phase actually occurs at low temperatures, the physics of 

1565: impurity-clusters suggests that, in the normal phase above a critical temperature, the transition to the Mott insulator is accompanied

1566: by the gradual opening of a pseudo-gap in the single-particle spectral function. In this pseudo-gaped region, Fermi-liquid

1567: behavior, i.e. $\Sigma(i\omega) \sim i\omega$, is recovered only at extremely low energies, suggesting 

1568: the existence of a finite temperature non-Fermi liquid behavior. Evidence in favor of this scenario has been found in a 

1569: recent cluster-DMFT simulation of a paramagnetic two-dimensional Hubbard model.~\cite{imada-2007} 

1570: 

1571: Another aspect worth emphasizing concerns the behavior of the Drude weight in the metal away from half-filling 

1572: across the symmetry-breaking phase-transition. From the point of view of the effective impurity model, a 

1573: symmetry breaking in the conduction baths opens up new screening channels that can rid the impurity of its residual entropy

1574: at the critical point. This in turns leads to an increase of screening energy gain that,

1575: translated back into the lattice model, implies 

1576: an \underline{increase} of band-energy gain, i.e. of the Drude weight. This behavior is actually the fingerprint of this

1577: kind of instability that reflects the underlying impurity critical point,

1578: as opposed to the conventional Stoner- or BCS-instability that are accompanied by a \underline{decrease} of Drude weight.

1579: The increase of Drude weight has been indeed observed in DMFT simulations of the two-dimensional

1580: Hubbard model~\cite{Jarrell-2005} as well as of a two-band Hubbard model~\cite{Capone04} that maps, within DMFT,

1581: onto the impurity dimer.

1582: 

1583: 

1584: 

1585: 

1586: 

1587: 

1588: \ack

1589: We acknowledge helpful discussions with E. Tosatti. We are very grateful to A. Georges, who pointed to our attention

1590: Ref.~\cite{Georges&Sengupta} that has been enlightening for our study.

1591: 

1592: \appendix

1593: 

1594: 

1595: \section*{Appendix: CFT at work}

1596: \addcontentsline{toc}{section}{Appendix: CFT at work}

1597: \setcounter{section}{1}

1598: 

1599: 

1600: In this Appendix, we show how {\sl conformal embedding} works in a few simple examples.

1601: We do it through the identification of the partition function, 

1602: the so-called {\sl character decomposition}.

1603: 

1604: The first step of bosonization in one dimension is the linearization of the free-electron spectrum around the 

1605: Fermi momentum~\cite{Sasha}.

1606: This linearization is not expected to affect the low energy behavior provided the perturbations are weak compared 

1607: to the band-width. Therefore let us consider, instead of a tight-binding model, free spinless Dirac fermions, 

1608: which have indeed a linear spectrum, on a chain of length $L$ with anti-periodic boundary conditions. 

1609: We are going to consider Dirac fermions moving only in one direction, namely with 

1610: a single chirality, because this is the case relevant to a semi-infinite chain where  

1611: the single-particle wave-functions with negative 

1612: momenta are not independent from those with positive ones. 

1613: 

1614: The single-particle wave-functions for a chiral Dirac fermion are plane waves with momentum 

1615: \[

1616: k = \frac{\pi}{L}\,(2n-1),

1617: \]

1618: $n$ being integer. The Hamiltonian in momentum space reads

1619: \be

1620: \mathcal{H} = v_F\,\sum_k\, k\, c^\dagger_k\,c^\pdag_k,

1621: \ee

1622: where $v_F$ has to be identified with the Fermi velocity of the original tight-binding model. 

1623: Let us define, for positive $k$,

1624: \ba

1625: \alpha^\pdag_k &=& c^\pdag_k,\\

1626: \beta^\pdag_k &=& c^\dagger_{-k},

1627: \ea

1628: so that, apart from an (actually infinite) constant, the Hamiltonian becomes

1629: \be

1630: \mathcal{H} = v_F\,\sum_{k>0}\, k\, \left(\alpha^\dagger_k\,\alpha^\pdag_k + \beta^\dagger_k\,\beta^\pdag_k\right). 

1631: \ee

1632: The partition function at temperature $T$ is simply

1633: \be

1634: Z_{Dirac} = \prod_{k>0}\,\bigg[1+\exp\left(-\beta\,v_F\,k\right)\bigg]^2 = 

1635: \prod_{n\geq 1}\, \left(1 + q^{n-1/2}\right)^2,

1636: \ee

1637: where conventionally~\cite{DiFrancesco} $q$ is defined as 

1638: \[

1639: q = \exp\left(-\beta\,\frac{2\pi v_F}{L}\right) \equiv {\rm e}^{2\pi i \tau}.

1640: \]  

1641: One can show that 

1642: \be

1643: Z_{Dirac}(q) = \frac{\theta_3(q)}{\varphi(q)},

1644: \label{NRG-CFT:ZDirac}

1645: \ee

1646: where $\theta_3$ is the third Jacobi theta-function 

1647: \[

1648: \theta_3(q) = \sum_{n=-\infty}^\infty\, q^{n^2/2},

1649: \]

1650: and $\varphi$ the Euler function

1651: \[

1652: \varphi(q) = \prod_{n\geq 1}\,\left(1-q^n\right).

1653: \]

1654: 

1655: \bigskip

1656: 

1657: 

1658: On the other hand, it is known by bosonization~\cite{Sasha} that, for positive $p=2\pi n/L>0$, the operators

1659: \[

1660: b^\pdag_p = -i\sqrt{\frac{2\pi}{pL}}\,\rho(p) =-i\sqrt{\frac{2\pi}{pL}}\,\sum_k \,c^\dagger_k\,c^\pdag_{k+p}, \qquad 

1661: b^\dagger_p = i\sqrt{\frac{2\pi}{pL}}\,\rho(-p),

1662: \]

1663: satisfy bosonic commutation relations. In addition their equation of motion can be reproduced by the 

1664: Hamiltonian 

1665: \be

1666: \mathcal{H} = \frac{\pi v_F}{L} \,\sum_p\, \rho(p)\,\rho(-p) = 

1667: v_F\,\sum_{p>0}\, p\,b^\dagger_p\,b^\pdag_p + \frac{\pi v_F}{L}\,\Delta\,N^2,

1668: \ee

1669: where it is assumed that $\rho(p=0)$ is the variation $\Delta\, N$ of the electron number with respect to a reference value. 

1670: The partition function of this bosonic model is the product of the free-boson term 

1671: \[

1672: \prod_{p>0}\,\left[1 - \exp\left(-\beta\,v_F\,p\right)\right]^{-1} = \prod_{n>0}\,\left(1-q^n\right)^{-1} 

1673: = \varphi(q)^{-1}

1674: \]

1675: plus the contribution of $\Delta\,N$ which is, assuming an infinite reference number,  

1676: \[

1677: \sum_{n=-\infty}^\infty\, \exp\left(-\beta\,\frac{v_F\pi}{L}\,n^2\right) = 

1678: \sum_{n=-\infty}^\infty\, q^{n^2/2} = \theta_3(q).

1679: \]

1680: One immediately recognizes that the bosonic partition function coincides with the fermionic one. 

1681:    

1682: 

1683: \bigskip

1684: 

1685: We can proceed further on, and consider spinful Dirac fermions. Obviously, the partition function is the square of 

1686: $Z_{Dirac}$ in Eq.~(\ref{NRG-CFT:ZDirac}). As the simplest example of {\sl conformal embedding} and 

1687: {\sl character decomposition}, we consider a perturbation that only preserves 

1688: independently the spin $SU(2)$ symmetry and the charge isospin $SU(2)$, defined through the generators $\mathbf{I}$, 

1689: which are the $q=0$ components of the so-called isospin current operators   

1690: \ba

1691: I_z(q) &=& \frac{1}{2}\,\sum_{k \sigma}\, \Big(c^\dagger_{k\sigma}c^\pdag_{k+q\sigma} - \delta_{q 0 }\Big),\\

1692: I^+(q) &=& \sum_{k}\, c^\dagger_{k\uparrow}\,c^\dagger_{-k-q\downarrow},\\

1693: I^-(q) &=& \Big(I^+\Big)^\dagger.

1694: \ea

1695: The spin $SU(2)$ current operators are instead 

1696: \be

1697: \mathbf{S}(q) = \frac{1}{2}\sum_{k \alpha\beta}\, c^\dagger_{k\alpha}\,\boldsymbol{\sigma}_{\alpha\beta}\, 

1698: c^\pdag_{k+q\beta},

1699: \ee

1700: where $\boldsymbol{\sigma}$ are the Pauli matrices. In real space these current operators, $J_a=I_a,\,S_a$, 

1701: satisfy the commutation relations

1702: \[

1703: \bigg[J_a(x),J_b(y)\bigg] 

1704: = i\epsilon_{abc}\,\delta(x-y)\,J_c(x) - i\,k\,\frac{1}{4\pi}\,\delta_{ab}\, \frac{\partial \delta(x-y)}{\partial x},

1705: \]

1706: with $k=1$. For generic $k\geq 1$, the above commutation relations identify 

1707: an $SU(2)_k$ CFT~\cite{DiFrancesco},

1708: where the label $k$ may be regarded as the number of channels which are used to build up the generators. 

1709: An $SU(2)_k$ CFT has primary fields $\boldsymbol{\phi}^{(k)}_{2j}$ with spin quantum numbers 

1710: $j$, such that $2j=0,1,\dots,k$. Their dimensions are $x_j = j(j+1)/(k+2)$. 

1711: The product of two primary fields with spin $j$ and $j'$ yields all primary 

1712: fields with spin between $|j-j'|$ and $\mathrm{min}(k-j-j',j+j')$. The Hilbert space of the theory is obtained by applying the 

1713: primary fields on the reference vacuum state and, from this ancestor state, by generating all descendant 

1714: states applying the current operators 

1715: with $q<0$. This is what is called a {\sl conformal tower}~\cite{DiFrancesco}.

1716: The energy difference between the descendant states and 

1717: their ancestor is an integer multiple of the fundamental level spacing $\Delta = 2\pi v_F/L$. 

1718: The character $\chi^{(k)}_{2j}$ represents the 

1719: contribution to the partition function of the conformal tower generated by the primary field 

1720: $\boldsymbol{\phi}^{(k)}_{2j}$~\cite{DiFrancesco}.

1721: 

1722: This construction may look abstruse but actually has a simple physical interpretation. Let us consider again a single 

1723: spinful fermion, $k=1$. 

1724: Let us further assume that on average the number of electrons is equal to the number of sites, and the latter is even. 

1725: In this case the ground state is obtained by filling with two 

1726: electrons of opposite spin all single-particle levels below the chemical potential, which lies in the middle of 

1727: two consecutive single-particle levels separated by the fundamental spacing $\Delta$, from now on our energy unit.   

1728: With this definition, the Hilbert space can be constructed as follows. One can start from the vacuum and act on 

1729: it with particle-hole excitations, namely with $I_z(q)$ or $\mathbf{S}(q)$ with $q<0$. In addition, one can apply 

1730: the operators $I^+(q)$ or $I^-(q)$, again with $q<0$, 

1731: to change by an even multiple the number of electrons, and then consider all particle-hole excitations 

1732: on top of these states. In this way, one obtains all states which have even number of electrons, like the vacuum state. 

1733: This is nothing but the conformal towers  

1734: obtained by the ancestor fields $\boldsymbol{\phi}^{(1)}_{0}$  

1735: in both the charge and spin $SU(2)_1$ sectors, which should therefore contribute to the 

1736: partition function with the product of characters $\left(\chi^{(1)}_0\right)_{charge}\,\left(\chi^{(1)}_0\right)_{spin}$. 

1737: 

1738: The rest of the Hilbert space includes all states with odd number of electrons. Since a single electron carries 

1739: isospin and spin 1/2, all these states have half-odd integer values of $I_z$ and $S_z$. 

1740: One realizes that they can all be obtained by applying the 

1741: product of the isospin and spin primary fields $\phi^{(1)}_{1\, charge}\times \phi^{(1)}_{1\, spin}$, 

1742: which is nothing but the single electron operator, and construct 

1743: out of it all descendant states. Their contribution to the partition function should then be 

1744: $\left(\chi^{(1)}_1\right)_{charge}\,\left(\chi^{(1)}_1\right)_{spin}$. The expression of the $SU(2)_k$ 

1745: characters~\cite[page 586]{DiFrancesco} is 

1746: \[

1747: \chi^{(k)}_l(q) = \frac{1}{\eta(q)^3}\, \sum_{n=-\infty}^\infty\, \bigg[2n\left(k+2\right)+l+1\bigg]\, 

1748: q^{(2n(k+2)+l+1)^2/4(k+2)},

1749: \]

1750: where 

1751: \[

1752: \eta(q) = q^{1/24}\,\varphi(q),

1753: \]

1754: is the Dedekind function. In the specific case of $k=1$,  

1755: \ba

1756: \chi^{(1)}_0(q) &=& \sqrt{\frac{\theta_3(q)^2 + \theta_4(q)^2}{2\eta(q)^2}},\\

1757: \chi^{(1)}_1(q) &=& \sqrt{\frac{\theta_3(q)^2 - \theta_4(q)^2}{2\eta(q)^2}},\\

1758: \ea

1759: where 

1760: \[

1761: \theta_4(q) = \sum_{n=-\infty}^\infty \, (-1)^n\, q^{n^2/2},

1762: \]

1763: is the fourth Jacobi theta-function. Hence we find that 

1764: \bea

1765: \left(\chi^{(1)}_0\right)_{charge}\,\left(\chi^{(1)}_0\right)_{spin} + 

1766: \left(\chi^{(1)}_1\right)_{charge}\,\left(\chi^{(1)}_1\right)_{spin}

1767: &=& \frac{\theta_3(q)^2}{\eta(q)^2} \nonumber \\

1768: &=& q^{-1/12}\, Z_{Dirac}(q)^2,

1769: \label{CFT-Z-single-channel}

1770: \eea

1771: which, apart from the vacuum polarization contribution $q^{-1/12}$, is exactly the partition function 

1772: of two species of Dirac fermions. 

1773: The spectrum, and correspondingly the partition function, can be represented as in Table~\ref{single-channel}.

1774: \begin{table}[htb]

1775: \caption{The spectrum of spinful electrons when the ground state contains an even number of particles.

1776: \label{single-channel}}

1777: \begin{indented}

1778: \item[] \begin{math}

1779: \begin{array}{lll}\br

1780: I & S & x \\ \mr

1781: 0 & 0 & 0 \\ \mr

1782: 1/2 & 1/2 & 1/2 \\ \br

1783: \end{array}

1784: \end{math}

1785: \end{indented}

1786: \end{table}

1787: In that and all forthcoming tables, we identify each conformal tower by the quantum numbers of the primary fields 

1788: which generate the ancestor states. $x$ is the energy in units of $\Delta$ of the ancestor state with respect to the 

1789: chemical potential. The descendant levels of an ancestor 

1790: have energies $x + n$, with $n$ a positive integer. Notice that an important consequence of conformal invariance is that 

1791: the energy of each state in units of $\Delta$ coincide with the dimension of the operator which, 

1792: applied to the vacuum, yields that state.    

1793: 

1794: Following the same reasoning, one can easily show that the table of the energy spectrum in the case of 

1795: odd chains at half-filling (i.e. odd average number of electrons)  

1796: is the one of Table~\ref{single-channel-odd}. 

1797: \begin{table}[htb]

1798: \caption{The spectrum of spinful electrons when the ground state contains an odd number of particles. 

1799: \label{single-channel-odd}}

1800: \begin{indented}

1801: \item[] \begin{math}

1802: \begin{array}{lll}\br

1803: I & S & x - 1/4 \\ \mr

1804: 1/2 & 0 & 0 \\ \mr

1805: 0 & 1/2 & 0  \\ \br

1806: \end{array}

1807: \end{math}

1808: \end{indented}

1809: \end{table}

1810: The ground state is fourfold degenerate, the chemical potential coinciding with a single-particle 

1811: level. One can readily realize that the even and odd chain spectra can be turned one into the other 

1812: by {\sl fusion} with a charge or a spin primary field $\boldsymbol{\phi}^{(1)}_{1}$.  

1813: This is the physics of the single-channel spin-1/2 Kondo impurity model. Indeed, when Kondo effect is established, 

1814: the impurity site becomes effectively a new site of the chain, thus changing the parity of the number of sites, i.e. 

1815: the boundary conditions.    

1816: 

1817: \paragraph{The two channel model}

1818: 

1819: 

1820: Let us consider a more involved {\sl conformal embedding} in the case of two channels of spinful fermions. 

1821: The partition function is the square of the partition function (\ref{CFT-Z-single-channel}) of a single channel, 

1822: and can be written for even chains as 

1823: \be

1824: Z= \sum_{n_1,n_2=0,1}\, \Bigg(\chi^{(1)}_{n_1}\,\chi^{(1)}_{n_2}\Bigg)_{charge}\; 

1825: \Bigg(\chi^{(1)}_{n_1}\,\chi^{(1)}_{n_2}\Bigg)_{spin},

1826: \ee

1827: where $n_1$ refers to channel 1 and $n_2$ to channel 2. For odd chains, one readily finds that 

1828: \be

1829: Z= \sum_{n_1,n_2=0,1}\, \Bigg(\chi^{(1)}_{n_1}\,\chi^{(1)}_{n_2}\Bigg)_{charge}\; 

1830: \Bigg(\chi^{(1)}_{1-n_1}\,\chi^{(1)}_{1-n_2}\Bigg)_{spin}.

1831: \ee

1832: These expressions manifestly show that the two-channel free conduction electrons are invariant under independent 

1833: spin or isospin $SU(2)$ transformations within each channel, namely under a large 

1834: symmetry $SU(2)\times SU(2) \times SU(2) \times SU(2)$. 

1835: 

1836: Let us suppose that the impurity couples only to the spin-current operators in such a way that only the overall 

1837: SU(2) spin symmetry is preserved. Therefore, while the charge degrees of freedom can still be represented 

1838: by two $SU(2)_1$ CFT's, the appropriate conformal embedding for the spin sectors should involve an $SU(2)_2$ CFT, 

1839: since the total spin current is made up of two channels, times the coset CFT, namely 

1840: \[

1841: SU(2)_1 \times SU(2)_1 \rightarrow SU(2)_2\,\times\, \frac{SU(2)_1 \times SU(2)_1}{SU(2)_2}.

1842: \]

1843: Since the central charge is conserved and each $SU(2)_k$ has a central charge $3k/(k+2)$, the coset theory should 

1844: have $c=1/2$, which corresponds to the central charge of an Ising CFT. This can be proved rigorously by the 

1845: {\sl character decomposition}. 

1846: 

1847: The Ising CFT has three primary fields, the identity $I$, with dimension 0, the energy field $\epsilon$, with dimension 1/2, 

1848: and the spin field $\sigma$ with dimension 1/16. The fusion rules are~\cite{DiFrancesco}

1849: \be

1850: I\times I = I,\;\; \epsilon\times \epsilon =I,\;\;

1851: \sigma\times\sigma = I + \epsilon,\;\; I\times\epsilon =\epsilon,\;\;

1852: I\times \sigma = \sigma,\;\; \epsilon\times \sigma = \sigma.

1853: \label{Ising-fusion}

1854: \ee

1855: The characters $\chi^I_x$, where $x$ is the dimension of the primary field, are given by 

1856: (all functions are assumed to depend on the variable $q$, even when not indicated) 

1857: \ba

1858: \chi^I_0 &=& \frac{1}{2}\, \Bigg[ \sqrt{\frac{\theta_3}{\eta}} + \sqrt{\frac{\theta_4}{\eta}}\Bigg],\\

1859: \chi^I_{1/2} &=& \frac{1}{2}\, \Bigg[ \sqrt{\frac{\theta_3}{\eta}} - \sqrt{\frac{\theta_4}{\eta}}\Bigg],\\

1860: \chi^I_{1/16} &=& \sqrt{\frac{1}{2}}\, \sqrt{\frac{\theta_2}{\eta}},\,

1861: \ea

1862: where the second Jacobi function is defined by

1863: \[

1864: \theta_2 = \sum_{n=-\infty}^\infty\, q^{(n-1/2)^2/2}.

1865: \]

1866: 

1867: One can show that the product of characters of two 

1868: $SU(2)_1$ CFTs, $\chi^{(1)}_{2j}\,\chi^{(1)}_{2j'}$, can be related to the product of characters 

1869: of an $SU(2)_2$ and an Ising CFTs, $\chi^{(2)}_{2l}\,\chi^I_x$, by

1870: \ba

1871: \chi^{(1)}_{0}\,\chi^{(1)}_{0} &=& \chi^{(2)}_{0}\,\chi^I_{0} + \chi^{(2)}_{2}\,\chi^I_{1/2},\\

1872: \chi^{(1)}_{0}\,\chi^{(1)}_{1} &=& \chi^{(1)}_{1}\,\chi^{(1)}_{0} = \chi^{(2)}_{1}\,\chi^I_{1/16},\\

1873: \chi^{(1)}_{1}\,\chi^{(1)}_{1} &=& \chi^{(2)}_{0}\,\chi^I_{1/2} + \chi^{(2)}_{2}\,\chi^I_{0}.

1874: \ea

1875: By means of the above decomposition one can write the tables of the spectra for even and odd chains, as given in 

1876: Table~\ref{k=2-even-odd}.

1877: \begin{table}[thb]

1878: \caption{The spectra of two channels on even chains, left table, and odd chains, right table. 

1879: $I_1$ and $I_2$ are the isospin value of each channel, $S$ the value of the total spin 

1880: and $\mathrm{Ising}$ refers to the Ising primary fields.

1881: \label{k=2-even-odd}}

1882: \begin{indented}

1883: \item[]\begin{math}

1884: \begin{array}{lllll}\br

1885: I_1 & I_2 & S & \mathrm{Ising} & x \\ \mr

1886: 0 & 0 & 0 & 0 & 0 \\ \mr

1887: 0 & 0 & 1 & 1/2 & 1 \\ \mr

1888: 1/2 & 0 & 1/2 & 1/16 & 1/2 \\ \mr

1889: 0 & 1/2 & 1/2 & 1/16 & 1/2 \\ \mr

1890: 1/2 & 1/2 & 0 & 1/2 & 1 \\ \mr

1891: 1/2 & 1/2 & 1 & 0 & 1 \\ \br

1892: \end{array}

1893: \qquad 

1894: \begin{array}{lllll}\br

1895: I_1 & I_2 & S & \mathrm{Ising} & x-1/2 \\ \mr

1896: 0 & 0 & 0 & 1/2 & 0 \\ \mr

1897: 0 & 0 & 1 & 0 & 0 \\ \mr

1898: 1/2 & 0 & 1/2 & 1/16 & 0 \\ \mr

1899: 0 & 1/2 & 1/2 & 1/16 & 0 \\ \mr

1900: 1/2 & 1/2 & 0 & 0 & 0 \\ \mr

1901: 1/2 & 1/2 & 1 & 1/2 & 1 \\ \br

1902: \end{array}

1903: \end{math}

1904: \end{indented}

1905: \end{table}

1906: By these tables one easily realizes that the spectrum of even chains can be turned into that of odd chains, and vice versa,

1907: either by fusion with the $SU(2)_2$ primary field of spin 1, or by fusion with the Ising field $\epsilon$.  

1908: As we mention in Section~\ref{The impurity dimer}, the unstable fixed point of the two spin-1/2 Kondo impurity model 

1909: was found by Affleck and Ludwig~\cite{Affleck92PRL,Affleck95} to correspond to fusion with the Ising field $\sigma$. If one  

1910: performs such a fusion in the spectra of Table~\ref{k=2-even-odd}, one obtains, by means of the fusion rules 

1911: (\ref{Ising-fusion}), a new spectrum 

1912: \begin{table}[hbt]

1913: \caption{Left table: The spectrum of the unstable fixed point of the two spin-1/2 Kondo impurity model.

1914: Right table: The boundary-operator dimensions at the unstable fixed point. The single and double asterisks identify, 

1915: respectively, the symmetry-invariant and symmetry-breaking relevant physical operators.

1916: \label{k=2-fixed-content}}

1917: \begin{indented}

1918: \item[] \begin{math}

1919: \begin{array}{lllll}\br

1920: I_1 & I_2 & S & \mathrm{Ising} & x-1/16 \\ \mr

1921: 0 & 0 & 0 & 1/16 & 0 \\ \mr

1922: 1/2 & 0 & 1/2 & 0 & 3/8 \\ \mr

1923: 0 & 1/2 & 1/2 & 0 & 3/8 \\ \mr

1924: 0 & 0 & 1 & 1/16 & 1/2 \\ \mr

1925: 1/2 & 1/2 & 0 & 1/16 & 1/2 \\ \mr

1926: 1/2 & 0 & 1/2 & 1/2 & 7/8 \\ \mr

1927: 0 & 1/2 & 1/2 & 1/2 & 7/8 \\ \mr

1928: 1/2 & 1/2 & 1 & 1/16 & 1 \\ \br

1929: \end{array}

1930: \end{math}

1931: \qquad 

1932: \begin{math}

1933: \begin{array}{lllll}\br

1934: I_1 & I_2 & S & \mathrm{Ising} & x \\ \mr

1935: 0 & 0 & 0 &       0 & 0 \\ \mr

1936: 0 & 0 & 0 &       1/2 & 1/2^{(*)} \\ \mr

1937: 1/2 & 0 & 1/2 &   1/16 & 1/2 \\ \mr

1938: 0 & 1/2 & 1/2 &   1/16 & 1/2 \\ \mr

1939: 0 & 0 & 1 &       0 & 1/2^{(**)} \\ \mr

1940: 1/2 & 1/2 & 0 &   0 & 1/2^{(**)} \\ \mr

1941: 0 & 0 & 1 &       1/2 & 1 \\ \mr

1942: 1/2 & 1/2 & 0 &   1/2 & 1 \\ \mr

1943: %1/2 & 0 & 1/2 &   1/16 & 1/2 \\ \mr

1944: %0 & 1/2 & 1/2 &   1/16 & 1/2 \\ \mr

1945: 1/2 & 1/2 & 1 &   0 & 1 \\ \mr

1946: 1/2 & 1/2 & 1 &   1/2 & 3/2 \\ \br

1947: \end{array}

1948: \end{math}

1949: \end{indented}

1950: \end{table}

1951: that was shown to reproduce the NRG spectrum obtained by Jones and Varma~\cite{Jones89}.

1952: By performing a further fusion (so-called {\sl double fusion}~\cite{Cardy,affleck-1995-26}) 

1953: with $\sigma$, one determines the dimensions of the 

1954: boundary operators at the unstable fixed point, shown in the same Table~\ref{k=2-fixed-content}. Their physical meaning 

1955: is discussed in Section~\ref{The impurity dimer}. 

1956: 

1957: %\bibliographystyle{unsrt}

1958: %\bibliography{jpc}

1959: 

1960: \begin{thebibliography}{99}

1961: 

1962: \bibitem{Mott1949}

1963: N.F. Mott.

1964: \newblock The basis of the electron theory of metals, with special reference to

1965:   the transition metals.

1966: \newblock {\em Proc. Phys. Soc. (London)}, 62:416, 1949.

1967: 

1968: \bibitem{Mott}

1969: N.F. Mott.

1970: \newblock {\em Metal Insulator Transition}.

1971: \newblock Taylor and Francis, London, 1990.

1972: 

1973: \bibitem{misguich-2005-}

1974: Gregoire Misguich and Claire Lhuillier.

1975: \newblock Two-dimensional quantum antiferromagnets.

1976: \newblock In H.T. Diep, editor, {\em Frustrated Spin Systems}, page 981.

1977:   World-Scientific, Singapore, 2005.

1978: \newblock and references therein.

1979: 

1980: \bibitem{Hewson}

1981: A.C. Hewson.

1982: \newblock {\em The Kondo Problem to Heavy Fermions}.

1983: \newblock Cambridge University Press, 1997.

1984: 

1985: \bibitem{millis-1993}

1986: A.J. Millis.

1987: \newblock Effect of a nonzero temperature on quantum critical points in

1988:   itinerant fermio systems.

1989: \newblock {\em Phys. Rev. B}, 48:7183, 1993.

1990: 

1991: \bibitem{coleman-2001-13}

1992: P.~Coleman, C.~Pepin, Q.~Si, and R.~Ramazashvili.

1993: \newblock How do fermi liquids get heavy and die?

1994: \newblock {\em J. Phys.: Condens. Matter}, 13:723, 2001.

1995: 

1996: \bibitem{si-2004-}

1997: Qimiao Si.

1998: \newblock Quantum critical metals: Beyond the order parameter fluctuations.

1999: \newblock {\em Advances in Solid State Physics}, 44:253, January 2004.

2000: 

2001: \bibitem{Jones87}

2002: B.~A. Jones and C.~M. Varma.

2003: \newblock Study of two magnetic impurities in a fermi gas.

2004: \newblock {\em Phys. Rev. Lett.}, 58(9):843, 1987.

2005: 

2006: \bibitem{Jones88}

2007: B.~A. Jones, C.~M. Varma, and J.~W. Wilkins.

2008: \newblock Low-temperature properties of the two-impurity kondo hamiltonian.

2009: \newblock {\em Phys. Rev. Lett.}, 61(1):125, 1988.

2010: 

2011: \bibitem{Jones89}

2012: B.~A. Jones and C.~M. Varma.

2013: \newblock Critical point in the solution of the two magnetic impurity problem.

2014: \newblock {\em Phys. Rev. B}, 40(1):324, 1989.

2015: 

2016: \bibitem{Affleck92PRL}

2017: I.~Affleck and A.~W.~W. Ludwig.

2018: \newblock Exact critical theory of the two-impurity kondo model.

2019: \newblock {\em Phys. Rev. Lett.}, 68(7):1046, 1992.

2020: 

2021: \bibitem{Affleck95}

2022: I.~Affleck, A.~W.~W. Ludwig, and B.~A. Jones.

2023: \newblock Conformal-field-theory approach to the two-impurity kondo problem:

2024:   Comparison with numerical renormalization-group results.

2025: \newblock {\em Phys. Rev. B}, 52(13):9528, 1995.

2026: 

2027: \bibitem{paul-1996-}

2028: Bruce~C. Paul and Kevin Ingersent.

2029: \newblock Frustration-induced non-fermi-liquid behavior in a three-impurity

2030:   kondo model, 1996.

2031: 

2032: \bibitem{ingersent-2005-95}

2033: Kevin Ingersent, Andreas W.~W. Ludwig, and Ian Affleck.

2034: \newblock Kondo screening in a magnetically frustrated nanostructure: Exact

2035:   results on a stable, non-fermi-liquid phase.

2036: \newblock {\em Phys. Rev. Lett.}, 95:257204, 2005.

2037: 

2038: \bibitem{Affleck:1990by}

2039: Ian Affleck and Andreas W.~W. Ludwig.

2040: \newblock The kondo effect, conformal field theory and fusion rules.

2041: \newblock {\em Nucl. Phys.}, B352:849--862, 1991.

2042: 

2043: \bibitem{Affleck:1990iv}

2044: Ian Affleck and Andreas W.~W. Ludwig.

2045: \newblock Critical theory of overscreened kondo fixed points.

2046: \newblock {\em Nucl. Phys.}, B360:641--696, 1991.

2047: 

2048: \bibitem{Jamneala2001mc}

2049: T.~Jamneala, V.~Madhavan, and M.~F. Crommie.

2050: \newblock Kondo response of a single antiferromagnetic chromium trimer.

2051: \newblock {\em Phys. Rev. Lett.}, 87(25):256804, Nov 2001.

2052: 

2053: \bibitem{Kudasov2002u}

2054: Yu.~B. Kudasov and V.~M. Uzdin.

2055: \newblock Kondo state for a compact cr trimer on a metallic surface.

2056: \newblock {\em Phys. Rev. Lett.}, 89(27):276802, Dec 2002.

2057: 

2058: \bibitem{savkin-2005-94}

2059: V.~V. Savkin, A.~N. Rubtsov, M.~I. Katsnelson, and A.~I. Lichtenstein.

2060: \newblock Correlated adatom trimer on metal surface: A continuous time quantum

2061:   monte carlo study.

2062: \newblock {\em Phys. Rev. Lett.}, 94:026402, 2005.

2063: 

2064: \bibitem{lazarovits-2005-95}

2065: Bence Lazarovits, Pascal Simon, Gergely Zarand, and Laszlo Szunyogh.

2066: \newblock Exotic kondo effect from magnetic trimers.

2067: \newblock {\em Phys. Rev. Lett.}, 95:077202, 2005.

2068: 

2069: \bibitem{George96}

2070: A.~Georges, G.~Kotliar, W.~Krauth, and M.~J. Rozenberg.

2071: \newblock Dynamical mean-field theory of strongly correlated fermion systems

2072:   and the limit of infinite dimensions.

2073: \newblock {\em Rev. Mod. Phys.}, 68(1):13, 1996.

2074: 

2075: \bibitem{Kotliar2001spb}

2076: Gabriel Kotliar, Sergej~Y. Savrasov, Gunnar P\'alsson, and Giulio Biroli.

2077: \newblock Cellular dynamical mean field approach to strongly correlated

2078:   systems.

2079: \newblock {\em Phys. Rev. Lett.}, 87(18):186401, Oct 2001.

2080: 

2081: \bibitem{Maier2005}

2082: T.~Maier, M.~Jarrell, T.~Pruschke, and M.~H. Hettler.

2083: \newblock Quantum cluster theories.

2084: \newblock {\em Rev. Mod. Phys.}, 77:1027, 2005.

2085: 

2086: \bibitem{senechal-2000-84}

2087: D.~Senechal, D.~Perez, and M.~Pioro-Ladriere.

2088: \newblock The spectral weight of the hubbard model through cluster perturbation

2089:   theory.

2090: \newblock {\em Phys. Rev. Lett.}, 84:522, 2000.

2091: 

2092: \bibitem{potthoff-2003-91}

2093: M.~Potthoff, M.~Aichhorn, and C.~Dahnken.

2094: \newblock Variational cluster approach to correlated electron systems in low

2095:   dimensions.

2096: \newblock {\em Phys. Rev. Lett.}, 91:206402, 2003.

2097: 

2098: \bibitem{Lichtenstein2000k}

2099: A.~I. Lichtenstein and M.~I. Katsnelson.

2100: \newblock Antiferromagnetism and d-wave superconductivity in cuprates: A

2101:   cluster dynamical mean-field theory.

2102: \newblock {\em Phys. Rev. B}, 62(14):R9283--R9286, 2000.

2103: 

2104: \bibitem{Wilson75}

2105: K.~G. Wilson.

2106: \newblock The renormalization group: Critical phenomena and the kondo problem.

2107: \newblock {\em Rev. Mod. Phys.}, 47(4):773, 1975.

2108: 

2109: \bibitem{Krishnamurthy80i}

2110: H.~R. Krishnamurthy, J.~W. Wilkins, and K.~G. Wilson.

2111: \newblock Renormalization-group approach to the anderson model of dilute

2112:   magnetic alloys. i. static properties for the symmetric case.

2113: \newblock {\em Phys. Rev. B}, 21(3):1003, 1980.

2114: 

2115: \bibitem{Krishnamurthy80ii}

2116: H.~R. Krishnamurthy, J.~W. Wilkins, and K.~G. Wilson.

2117: \newblock Renormalization-group approach to the anderson model of dilute

2118:   magnetic alloys. ii. static properties for the asymmetric case.

2119: \newblock {\em Phys. Rev. B}, 21(3):1044, 1980.

2120: 

2121: \bibitem{DiFrancesco}

2122: P.~{Di Francesco}, P.~Mathieu, and D.~S\'en\'echal.

2123: \newblock {\em Conformal Field Theory}.

2124: \newblock Springer, 1996.

2125: 

2126: \bibitem{Capone02}

2127: M.~Capone, M.~Fabrizio, C.~Castellani, and E.~Tosatti.

2128: \newblock Strongly correlated superconductivity.

2129: \newblock {\em Science}, 296:2364, 2002.

2130: 

2131: \bibitem{Capone04}

2132: M.~Capone, M.~Fabrizio, C.~Castellani, and E.~Tosatti.

2133: \newblock Strongly correlated superconductivity and pseudogap phase near a

2134:   multiband mott insulator.

2135: \newblock {\em Phys. Rev. Lett.}, 93:047001, 2004.

2136: 

2137: \bibitem{DeLeo03f}

2138: M.~Fabrizio, A.~F. Ho, L.~{De Leo}, and G.~E. Santoro.

2139: \newblock Nontrivial fixed point in a twofold orbitally degenerate anderson

2140:   impurity model.

2141: \newblock {\em Phys. Rev. Lett.}, 91:246402, 2003.

2142: 

2143: \bibitem{DeLeo04f}

2144: L.~{De Leo} and M.~Fabrizio.

2145: \newblock Spectral properties of a two-orbital anderson impurity model across a

2146:   non-fermi-liquid fixed point.

2147: \newblock {\em Phys. Rev. B}, 69:245114, 2004.

2148: 

2149: \bibitem{Sasha}

2150: A.O. Gogolin, A.A. Nersesyan, and A.M. Tsvelik.

2151: \newblock {\em Bosonization and strongly correlated systems}.

2152: \newblock Cambridge University Press, 1998.

2153: 

2154: \bibitem{Zamo}

2155: A.A. Belavin, A.M. Polyakov, and A.B. Zamolodchikov.

2156: \newblock Infinite conformal symmetry in two-dimensional quantum field theory.

2157: \newblock {\em Nuclear Physics B}, 241:333--380, 1984.

2158: 

2159: \bibitem{Zamo1}

2160: V.G. Knizhnik and A.B. Zamolodchikov.

2161: \newblock Current algebra and wess-zumino model in two dimensions.

2162: \newblock {\em Nuclear Physics B}, 247:83--103, 1984.

2163: 

2164: \bibitem{Cardy1}

2165: J.L. Cardy.

2166: \newblock Operator content of two-dimensional conformally invariant theories.

2167: \newblock {\em Nuclear Physics B}, 270:186--204, 1986.

2168: 

2169: \bibitem{Cardy}

2170: John~L. Cardy.

2171: \newblock Boundary conditions, fusion rules and the verlinde formula.

2172: \newblock {\em Nucl. Phys.}, B324:581, 1989.

2173: 

2174: \bibitem{affleck-1995-26}

2175: Ian Affleck.

2176: \newblock Conformal field theory approach to the kondo effect.

2177: \newblock {\em Acta Phys. Polon. B}, 26:1869, 1995.

2178: 

2179: \bibitem{NozieresJLTP}

2180: Philippe Nozi\`eres.

2181: \newblock A fermi liquid description of kondo effect at low temperatures.

2182: \newblock {\em J. Low Temp. Phys.}, 17:31, 1974.

2183: 

2184: \bibitem{Blume}

2185: M.~Blume.

2186: \newblock Theory of the first-order magnetic phase change in {UO$_{2}$}.

2187: \newblock {\em Phys. Rev.}, 141:517, 1966.

2188: 

2189: \bibitem{Capel}

2190: H.~W. Capel.

2191: \newblock On possibility of first-order phase transitions in ising systems of

2192:   triplet ions with zero-field splitting.

2193: \newblock {\em Physica}, 32:866, 1966.

2194: 

2195: \bibitem{Affleck:1991tk}

2196: Ian Affleck and Andreas W.~W. Ludwig.

2197: \newblock Universal noninteger 'ground state degeneracy' in critical quantum

2198:   systems.

2199: \newblock {\em Phys. Rev. Lett.}, 67:161--164, 1991.

2200: 

2201: \bibitem{Affleck93}

2202: I.~Affleck and A.~W.~W. Ludwig.

2203: \newblock Exact conformal-field-theory results on the multichannel kondo

2204:   effect: Single-fermion green's function, self-energy, and resistivity.

2205: \newblock {\em Phys. Rev. B}, 48(10):7297, 1993.

2206: 

2207: \bibitem{Affleck92PRB}

2208: I.~Affleck, A.~W.~W. Ludwig, H.-B. Pang, and D.~L. Cox.

2209: \newblock Relevance of anisotropy in the multichannel kondo effect: Comparison

2210:   of conformal field theory and numerical renormalization-group results.

2211: \newblock {\em Phys. Rev. B}, 45(14):7918, 1992.

2212: 

2213: \bibitem{affleck-2001-594}

2214: Ian Affleck, Masaki Oshikawa, and Hubert Saleur.

2215: \newblock Quantum brownian motion on a triangular lattice and c=2 boundary

2216:   conformal field theory.

2217: \newblock {\em Nuclear Physics B}, 594:535, 2001.

2218: 

2219: \bibitem{Georges&Sengupta}

2220: A.~Georges and A.~Sengupta.

2221: \newblock Solution of the 2-impurity 2-channel {Kondo} model.

2222: \newblock {\em Phys. Rev. Lett.}, 74:2808, 1995.

2223: 

2224: \bibitem{Schotte}

2225: K.~Schotte and U.~Schotte.

2226: \newblock Tomonaga's model and the threshold singularity of x-ray spectra of

2227:   metals.

2228: \newblock {\em Phys. Rev.}, 182:479, 1969.

2229: 

2230: \bibitem{Jarrell-2005}

2231: T.~A. Maier, M.~Jarrell, T.~C. Schulthess, P.~R.~C. Kent, and J.~B. White.

2232: \newblock Systematic study of d-wave superconductivity in the 2d repulsive

2233:   hubbard model.

2234: \newblock {\em Physical Review Letters}, 95(23):237001, 2005.

2235: 

2236: \bibitem{Capone-2006}

2237: M.~Capone and G.~Kotliar.

2238: \newblock Competition between d-wave superconductivity and antiferromagnetism

2239:   in the 2d hubbard model, cond-mat/0603227, 2006.

2240: 

2241: \bibitem{imada-2007}

2242: Y.Z. Zhang and M.~Imada.

2243: \newblock Pseudogap and mott transition studied by cellular dynamical mean

2244:   field theory, cond-mat/0706.0444, 2007.

2245: 

2246: \end{thebibliography}

2247: 

2248: 

2249: 

2250: \end{document}

2251: