1: \documentstyle[aps,prb,epsfig,twocolumn]{revtex}

2: \begin{document}

3: \newcommand{\vektor}[1]{\mbox{\boldmath $#1$}}

4: \newcommand{\taubold}{\mbox{\boldmath$\tau$}}

5: \twocolumn[\hsize\textwidth\columnwidth\hsize\csname

6: @twocolumnfalse\endcsname

7: \title{Localized spin ordering in Kondo lattice models}

8: \author{I. P. McCulloch${}^{\dagger}$,

9: A. Juozapavicius${}^{\ddagger}$, A. Rosengren${}^{\ddagger}$

10: and M. Gulacsi${}^{\dagger}$}

11: \address{${}^{\dagger}$Department of Theoretical Physics,

12: Institute of Advanced Studies, \\

13: The Australian National University,

14: Canberra, ACT 0200, Australia}

15: \address{${}^{\ddagger}$Department of Theoretical Physics,

16: Royal Institute of Technology, \\

17: SE-100 44 Stockholm, Sweden}

18: \date{\today} 

19: \maketitle

20: \begin{abstract}

21: Using a non-Abelian density matrix renormalization

22: group method we determine the phase diagram of the Kondo lattice model

23: in one dimension, by directly measuring the magnetization of the

24: ground-state. This allowed us to discover a second ferromagnetic phase

25: missed in previous approaches. The phase transitions are found to be

26: continuous. The spin-spin correlation function is studied in detail,

27: and we determine in which regions the large and small Fermi surfaces

28: dominate. The importance of double-exchange ordering and its

29: competition with Kondo singlet formation is emphasized in

30: understanding the complexity of the model.

31: \end{abstract}

32: \pacs{PACS No. 75.50 Mb, 71.10 Hf, 05.10.Cc} 

33: ]

34: 

35: The Kondo lattice model (KLM) describes the interaction between

36: a conduction electron (CE) band and a half-filled narrow impurity,

37: e.g. $f$ electron, band and is thought to capture the essential

38: physics of the rare earth compounds. Although intensively studied

39: for two decades, the KLM is still far from being completely

40: understood. Recently, after the discovery of Kondo insulators

41: and the non-Fermi liquid behavior, interest in this field

42: has been greatly renewed, especially due to the non-Fermi

43: liquid behavior discovered in most of the heavy fermion compounds,

44: which resembles a Griffiths phase.\cite{neto}

45: 

46: In order to understand the role of the impurity spin in

47: determining the properties of KLM we must develop a better

48: understanding of the magnetic correlations.

49: The Griffiths phase in the one dimensional

50: KLM occurs naturally;\cite{graeme} it is, therefore, the

51: prototypical model for heavy fermion compounds.

52: Hence, this is an ideal system to study since we

53: have the bosonized solution \cite{graeme,boso} and we

54: know the behavior of the CEs  in both the paramagnetic (PM)

55: and ferromagnetic (FM) phases. However, less attention has

56: been given to understand the correlations between the

57: impurity spins. This is the focus of our study.

58: 

59: The Hamiltonian for the KLM is

60: \begin{equation}

61: H \: = \: -t \sum_{j=1, \sigma}^{L} ( c^{\dagger}_{j, \sigma}

62: c^{}_{j + 1, \sigma} + {\rm h.c.} ) \: + \:

63: J \sum_{j=1}^{L} {\bf S}^{c}_{j} {\bf \cdot} {\bf S}^{}_{j} \; ,

64: \label{klm}

65: \end{equation}

66: where $t > 0$ is the CE hopping parameter,

67: ${\bf S}_{j}$ are spin $1/2$ operators for the

68: localized spins, e.g. $f$, and

69: ${\bf S}^{c}_{j}= \frac{1}{2} \sum_{\sigma^{},\sigma^{\prime}}

70: c^{\dagger}_{j, \sigma^{}} {\vektor{\sigma}}_{\sigma^{},\sigma^{\prime}}

71: c^{}_{j, \sigma^{\prime}} $

72: with ${\vektor{\sigma}}$ the Pauli spin matrices and

73: $c^{}_{j, \sigma}$, $c^{\dagger}_{j, \sigma}$ the electron

74: annihilation and creation site operators. The Kondo coupling

75: $J$ is measured in units of the hopping $t$ and partial conduction

76: band filling, $n = N/L < 1$, is assumed throughout.

77: 

78: The method that we use is density matrix renormalization group (DMRG)

79: which, however, is extended to explicitly preserve $SU(2)$ spin and

80: pseudospin symmetry. Hence we can measure the magnetization directly

81: and determine rigorously the PM - FM phase boundary. The obtained

82: result is in excellent agreement with a recent bosonized solution

83: \cite{graeme} and contradicts common view that this phase boundary

84: goes to infinite Kondo coupling $J$ as the CE density approaches

85: half-filling.\cite{RevModPhys,JPhys} We also determine the regions of

86: the phase diagram where large and small Fermi surfaces are dominant,

87: which has been a central issue for much of the research in this area

88: for some years.

89: 

90: In addition, we have discovered a second FM region

91: not seen before. For most dopings, this region of FM

92: separates the regions of large and small Fermi surface. This

93: most likely resolves the question as to the applicability

94: of the Luttinger theorem to the KLM, shown by Yamanaka {\it et al.},

95: \cite{Luttinger} since the Fermi points are not expected to

96: remain constant across a phase transition. 

97: 

98: To accelerate the computation, we make use of several operators that

99: commute with the Hamiltonian, $S^+, S^-, S^z, I^+, I^-, I^z$,

100: respectively the generators of the spin $SU(2)$ and pseudospin $SU(2)$

101: algebras.\cite{PseudoSpin} Combined, the generators form the algebra

102: $SO(4)$. All of the states in our DMRG calculation transform as

103: irreducible representations of this algebra. Since $SO(4)$ is

104: non-Abelian these representations have, in general, degree $> 1$,

105: which implies that a single basis state in the $SO(4)$ representation

106: is equivalent to multiple states of the purely Abelian representation

107: of most previous DMRG calculations. This is the origin of the dramatic

108: performance improvements of the non-Abelian DMRG.  The states are

109: labeled by the eigenvalues of the Casimir operators of $SO(4)$, which

110: are $S^2 = s(s+1)$ and $I^2 = i(i+1)$. Hence we can label all

111: irreducible representations by $[s,i]$, which has degree

112: $(2s+1)(2i+1)$. In this construction, a chemical potential would appear

113: as a term in the Hamiltonian proportional to $I^z$, acting in an

114: identical way a magnetic field coupled to $S^z$.

115: Although the basis states in the calculation are eigenstates of

116: $S^2$ and $I^2$, rather than $S^z$ and $I^z$, all these operators

117: mutually commute so it is possible to simply replace $S^z$ and

118: $I^z$ by the chosen eigenvalues in this case.

119: A single site of the Kondo lattice contains just three

120: such states. The simplest is the Kondo singlet state, transforming as

121: the $[0,0]$ representation of degree 1. The Kondo triplet state

122: transforms as the $[1,0]$ representation of degree 3, and encapsulates

123: the three projections $\vert \Uparrow \uparrow \rangle$, $\sqrt{1/2} (

124: \vert \Uparrow \downarrow \rangle + \vert \Downarrow \uparrow

125: \rangle)$, $\vert \Downarrow \downarrow \rangle$ in a single

126: state. Here, $\Uparrow$ denotes localized $f$, and $\downarrow$ the

127: conduction electron spins, respectively. Finally, the holon state

128: (actually, the tensor product of a holon and a $f$ spin) transforms as

129: the $[1/2,1/2]$ representation of degree 4 and has the projections

130: $\vert \Uparrow 0 \rangle$, $\vert \Downarrow 0 \rangle$, $\vert

131: \Uparrow \uparrow \downarrow \rangle$, $\vert \Downarrow \uparrow

132: \downarrow \rangle$.  The single-site operators are $3 \times 3$

133: matrices over this basis.  The matrix elements can be determined by

134: the Wigner-Eckart theorem, which specifies the relationship between

135: the 3 dimensional reduced basis and the full 8 dimensional basis. For

136: a comprehensive description of the new algorithm, see

137: Ref.\ \onlinecite{totalspinDMRG}. At half filling (where the ground state is

138: a pseudospin singlet) 400 block states are equivalent to around 2500

139: states of a calculation using $N$ and $S^z$ quantum numbers, although

140: the relative advantage of $SO(4)$ decreases as the system is doped

141: away from half filling. We used the new DMRG algorithm to obtain the

142: ground state energy, magnetization and different correlation

143: functions, i.e., the momentum distribution, density-density,

144: conduction electron spin-spin and the $f$ spin structure factor,

145: $S(k)$. The obtained results can be summarized with the phase diagram

146: presented in Fig.\ 1, which will be analyzed in detail hereafter. The

147: main properties of the phase diagram have been confirmed on chains of

148: 120 or more sites. Results for the magnetization were calculated on

149: smaller chains, 40 - 60 sites, where the energies can be calculated

150: more accurately.  We found no finite size effects that would

151: affect the properties of Fig.\ 1. In all cases, we extrapolate

152: to zero truncation error based on well-converged sweeps of

153: between 200 and 500 $SO(4)$ states kept.

154: 

155: As it can be seen from Fig.\ 1, the main feature dominating the KLM is

156: $f$ spin FM ordering. The FM ordering is due to the

157: double-exchange (DE) interaction which appears as a consequence

158: of an excess of localized spins over CEs: \cite{zener}

159: each CE has to screen more than one localized spin, and

160: since hopping is energetically most favorable for CEs

161: which preserve their spin, this tends to align the localized

162: spins. This element was missing in the early approaches,

163: which concentrated on the competition between

164: Kondo singlet formation at large $J$ and the RKKY interaction

165: in the weak coupling limit.\cite{Jullien} This picture

166: is borrowed from the single impurity Kondo model and is

167: inadequate for the lattice case.\cite{RevModPhys,JPhys} 

168: 

169: Starting the analysis of the phase diagram for large $J$,

170: we see that all CEs form singlets with the localized

171: $f$ spins \cite{largeJ} which become inert. The uncoupled $f$ spins

172: order FM in a mechanism similar to the $J < 0$ case.\cite{zener}

173: Here, there is no competition between Kondo singlet

174: formation and DE. The fully polarized state [with $S = (L-N)/2$]

175: appears for any value of $n < 1$ \cite{graeme,largeJ}

176: contrary to the suggestion of Refs.\ \onlinecite{RevModPhys,JPhys}

177: that close to half filling the PM phase extends to $J \rightarrow \infty$.

178: As $J$ is lowered, KLM can be rigorously mapped into a random transverse

179: field Ising model,\cite{graeme} hence the phase transition (the

180: solid curve in Fig.\ 1) is identical to the quantum order - disorder

181: transition. It should be emphasized that this is also true for

182: the second FM phase, as will be shown later on.

183: 

184: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

185: 

186: \begin{figure}[h]

187: \epsfxsize=7cm

188: \centerline{\epsfbox{KLM_fig1.eps}}

189: \caption{

190: The obtained phase diagram of KLM. The two shaded areas are

191: the FM phases. The open circles and triangles correspond to

192: points at which the FM energy level crosses the $S=0$ level.

193: The dashed curves are the derived phase transition lines

194: (the solid curve was already obtained in

195: Ref.\ \protect\onlinecite{graeme}).}

196: \label{fig1}

197: \end{figure}

198: 

199: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

200: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

201: 

202: \begin{figure}[h]

203: \epsfxsize=7cm

204: \centerline{\epsfbox{KLM_fig2.eps}}

205: \caption{

206: Normalized magnetization curves (relative to the

207: ground state energy, $E_0$) across the phase transition

208: at quarter filling for a 40 site lattice.}

209: \label{fig2}

210: \end{figure}

211: 

212: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

213: 

214: The phase transition obtained via DMRG fits exceptionally well this

215: picture, confirming the bosonization result of Ref.\ \onlinecite{graeme}.

216: The open circles correspond to points at which the energy of the FM

217: state crosses the energy of the singlet state. Since the phase

218: transition is second order, this is only an upper bound on the

219: true transition line. However the partially polarized region is

220: very small, of the order of $J/t \sim 0.01$, which is why

221: this phase transition has not previously been observed to be

222: continuous. A typical example of the energy versus the

223: magnetization ($M$) is presented in Fig.\ 2. This shows that in

224: the transition regime, $\partial^2 E / \partial M^2$ is positive.

225: We have accounted for all known random errors, these are errors

226: arising from the tolerance of the matrix diagonalization,

227: variations in the energy across the DMRG sweep, and error

228: arising from the extrapolation to zero truncation error.

229: These errors are of the order of the symbol size in this figure.

230: 

231: Below the solid curve, Fig.\ 1, the Kondo singlets are not

232: inert anymore and they greatly contribute to the properties of

233: KLM. Excluding the Kondo triplet states, the CE wave function in the

234: continuum limit satisfies a nonlinear Schr\"{o}dinger equation

235: \cite{polaron} which has finitely delocalized solitonic solutions.

236: \cite{holstein} This corresponds to a dressing of the CE by a cloud

237: of antiparallel local spins, i.e., spin polarons are formed. The

238: polaronic length scale competes with the length scale set by the free

239: CE mean free path and introduces competing time scales: slow motion of

240: the polarons with low energy dynamics and fast motion of the free CEs

241: with high energies. This scenario resembles a two-fluid picture with

242: intrinsic inhomogeneities which involves spin fluctuations and

243: short-range spin correlations, which we call a {\sl polaronic liquid}.

244: 

245: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

246: 

247: \begin{figure}[h]

248: \epsfxsize=7cm

249: \centerline{\epsfbox{KLM_fig3.eps}}

250: \caption{

251: Typical $J$ dependence of the spin structure factor, $S(k)$,

252: and the momentum distribution, $n(k)$ ($n = 0.6$).}

253: \label{fig3}

254: \end{figure}

255: 

256: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

257: 

258: Finite temperature DMRG \cite{shibata} confirmed the presence

259: of short-range $f$ spin correlations in the van-Hove singularities.

260: Consequently the structure factor peaks at $2 k_F - \pi$,

261: where $k_F$ is the Fermi point determined by the filling of the CE

262: band. This means that the localized $f$ spins, even though they

263: are completely immobile, contribute to the volume of the Fermi sea.

264: This conventionally is called a {\sl large} Fermi surface, the

265: effect of which is also seen in the momentum distribution

266: function, see Fig.\ 3. As the polarons are formed the peak

267: of $S(k)$ shifts from the small $J/t$ value of $2 k_F$:

268: the slow motion of the spin polarons will dominate the

269: low energy dynamics of the quasiparticles. This proves

270: that the appearance of the large Fermi surface is a

271: dynamical effect since it involves local inhomogeneities,

272: impurity spin fluctuation, and short-range correlations

273: of the $f$ spins. This phase is related \cite{graeme} to a

274: Griffiths phase, suggesting that the small - large Fermi

275: surface crossover is a Griffiths singularity.

276: 

277: The large Fermi surface is conventionally explained by

278: reference to the periodic Anderson model (PAM) ancestry.

279: \cite{JPhys,Luttinger,largeFS} Our results imply that

280: even for PAM, this simple picture is inadequate. In

281: particular, we see no reason why a small - large

282: Fermi surface crossover, marked by a FM phase,

283: should not also appear in PAM. However, the

284: behavior of the Fermi surface crossover close to quarter

285: filling is numerically difficult to determine (dotted

286: line in Fig.\ 1), hence we are not yet able to rule

287: out the possibility that the large and small Fermi

288: surface regions are adiabatically connected. Even prior

289: to the current calculation, the nature of the Fermi surface 

290: in the weak-coupling regime was not clear, with the suggestion 

291: from Ref.\ \cite{JPhys} that the Fermi surface vanishes at 

292: a point in proximity to where we find the ferromagnetic phase.

293: For $n < 0.5$

294: the width of the polarons is over several lattice spacings

295: (diverging for $n \rightarrow 0$, \cite{polaron}) hence the

296: energy needed to excite these polarons is too large for this

297: effect to happen. The polarons will not contribute to the low

298: energy dynamics and the system behaves as an RKKY liquid, as we

299: explain below. 

300: 

301: An interesting phenomenon appears as we further lower $J$.

302: The residual weight attached to the Kondo singlets vanishes,

303: hence all CEs which participated in the formation of these

304: singlets, become delocalized. The distance between these CEs

305: is much larger than the lattice spacing, and below

306: $J \le 2 {\sqrt{n}} \sin ( \pi n )$ their continuum

307: limit takes the regular quantum sine-Gordon form.\cite{boso}

308: In the bosonization language of Ref.\ \onlinecite{graeme}, this

309: means that the spin Bose fields, $\Phi_{\sigma}$ cannot be

310: approximated by their noninteracting expectation values,

311: rather by their expectation value corresponding to a

312: sine-Gordon (sG) model, $\Phi_{\sigma} \approx \langle

313: \Phi_{\sigma} \rangle_{\rm{sG}}$. However, the charge degrees

314: of freedom not being affected by the sine-Gordon spin gap,

315: their corresponding Bose fields, $\Phi_{\rho}$ may be still

316: approximated by their noninteracting values. Extending the

317: bosonized results of Ref.\ \onlinecite{graeme} to a finite

318: $\langle \Phi_{\sigma} \rangle_{\rm{sG}}$, we obtain

319: the critical Hamiltonian governing the PM - FM

320: phase transition at intermediate $J$ values in the form:

321: $H_{\rm crit.} = - J^2 {\cal{A}} / (2 \pi^2 v_F) \sum_{j} 

322: {\bf S}^{z}_{j} {\bf S}^{z}_{j + 1} + 2 J {\cal{B}} \sum_{j} 

323: \{ 1 - ( \langle \Phi_{\sigma} \rangle^{2}_{\rm{sG}} / 2 ) 

324: [1 + J / (2 \pi v_F) ]^2 + \cos ( 2 k_F j ) \} {\bf S}^{x}_{j}$,

325: where ${\cal{A}}$ and ${\cal{B}}$ are functions which depend only

326: on the cutoffs introduced by the bosonization

327: scheme.\cite{graeme,boso} Following closely previous

328: bosonization approaches,\cite{graeme,boso} we can prove that

329: the critical behavior of the FM transition for the intermediate

330: this $J$ case is of a random transverse-field Ising

331: model type, where the transverse field

332: $h_j = 2 J {\cal{B}} \{ 1 - ( \langle \Phi_{\sigma} \rangle^{2}_{\rm{sG}} / 2 ) 

333: [1 + J / (2 \pi v_F) ]^2 + \cos ( 2 k_F j ) \}$

334: is driven by a displaced cosine distribution

335: of the form: 

336: $\rho (h) = [1 / (2 \pi J {\cal{B}} ) ] \{ 1 - [ h / (2 J {\cal{B}} ) + 

337: ( \langle \Phi_{\sigma} \rangle^{2}_{\rm{sG}} / 2 ) 

338: [1 + J / (2 \pi v_F) ]^2 - 1 ]^2 \}^{-1/2}$.

339: Accordingly, the FM transitions emerging at

340: intermediate values of $J$ are of a quantum order -

341: disorder type. These transitions are driven by spin polarons,

342: contrary to the FM phase emerging at high $J$ values, which

343: is given by the uncoupled $f$ spins (in a mechanism similar

344: to the $J < 0$ case).  The new critical line is 

345: $J_c = \alpha ({\cal {A}}, {\cal {B}}) \sin (\pi n /2) / 

346: [1 - \beta ({\cal {A}}, {\cal {B}})] - \gamma ({\cal {A}}, {\cal {B}}, 

347: \langle \Phi_{\sigma} \rangle^{2}_{\rm{sG}})$.

348: The bosonization (conformal field-theory) arguments

349: does not determine the magnitude of $\alpha$, $\beta$

350: and $\gamma$, accordingly these constants are used as

351: fitting parameters to the numerically obtained points.

352: The best fits are the dashed curves in Fig.\ 1.

353: 

354: This is the second FM phase in Fig.\ 1, which has proven difficult to

355: detect with conventional (Abelian) DMRG.\cite{exactFM,numerics} Previous

356: DMRG calculations did show a weak FM signal at $n = 0.8$ and $J = 1.6$

357: and $1.8$,\cite{largeFS} but the results were discarded in later papers

358: by the same authors.\cite{JPhys,RevModPhys} Likewise an

359: exact diagonalization of a very small system gave FM for $n=0.75$ and

360: $J=1.5$.\cite{exactFM} Using the non-Abelian DMRG algorithm we could

361: also check the energy of each total spin state, shown in Fig.\ 4,

362: which clearly shows a second ferromagnetic region although we have

363: not yet confirmed numerically the order of the phase boundaries.

364: For the FM Kondo lattice model, $J < 0$, a phase separated regime

365: was observed in numerical approaches.\cite{dagotto} However for

366: $J > 0$ we found no change in sign of the inverse charge

367: compressibility. Thus, this phase is a true FM rather than a

368: phase separated region. 

369: 

370: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

371: 

372: \begin{figure}[h]

373: \epsfxsize=7cm

374: \centerline{\epsfbox{KLM_fig4.eps}}

375: \caption{

376: The gap, $\Delta E$, from the fully polarized ferromagnetic

377: state to every other spin state {\it vs.} $J$,  for $n=0.8$,

378: and 60 site lattice, measured along intervals of $J \pm 0.05$.

379: For most data points the error bars are of order $\sigma_{\Delta E}

380: \sim 10^{-5}$ or less, except for the $S=0$ curve for very low

381: and very high $J$, where the errors are of order

382: $\sigma_{\Delta E} \sim 5 \times 10^{-4}$. The inset 

383: shows the second ferromagnetic region.}

384: \label{fig4}

385: \end{figure}

386: 

387: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

388: 

389: Below the second FM region the KLM reduces to a system of free

390: localized spins in fields determined by CE scattering: dominant

391: $2k_{F}$ modulations are manifest, see Fig.\ 3, superimposed on

392: an incoherent background. This reflects the momentum transferred

393: from the CE band to the spin chain in backscattering interactions,

394: together with incoherent forward scattering. This case is referred

395: to as an RKKY liquid as the scattering processes give an RKKY-like

396: correlation for the $f$ spins, even though the RKKY interaction

397: strictly diverges in one dimension.

398: 

399: In conclusion, using a non-Abelian DMRG method a most comprehensive

400: analysis of the short- and long-range ordering of the localized moments

401: in KLM is presented.  We show that DE ordering and its competition with

402: Kondo singlet formation is the dominant feature of the phase diagram.

403: The non-Abelian DMRG method allowed us to discover that FM does not

404: only appear at large $J$ but also at intermediate values. This second

405: FM phase was missed in previous approaches. We also show that at large $J$

406: FM is due to ordering of uncoupled $f$ spins, while for intermediate $J$,

407: i.e., the second FM region, FM is due to ordering of the spin polarons.

408: The inhomogeneous polaronic state between these two FM phases

409: is analogous to a two-fluid system and it exhibits a large Fermi

410: surface.

411: 

412: Work in Australia was supported by the Australian Research Council. In

413: Sweden by the Swedish Natural Science Research Council. 

414: 

415: \begin{references}

416: \bibitem{neto}M. C. de Andrade, {\sl et al.}, Phys. Rev. Lett.

417: {\bf 81}, 5620 (1998).

418: \bibitem{graeme}G. Honner and M. Gulacsi, Phys. Rev. Lett.

419: {\bf 78}, 2180 (1997); Phys. Rev. B{\bf 58}, 2662 (1998).

420: \bibitem{boso}O. Zachar, S. A. Kivelson, and V. J. Emery,

421: Phys. Rev. Lett. {\bf 77}, 1342 (1996).

422: \bibitem{RevModPhys}H. Tsunetsugu, M. Sigrist, and K. Ueda,

423: Rev. Mod. Phys. {\bf 69}, 809 (1997).

424: \bibitem{JPhys}N. Shibata and K. Ueda,

425: J. Phys. Cond. Matter. {\bf 11}, R1 (1999).

426: \bibitem{Luttinger}M. Yamanaka, M. Oshikawa and I. Affleck, Phys.

427: Rev. Lett. {\bf 79}, 1110 (1997).

428: \bibitem{PseudoSpin}B. A. Jones, C. M. Varma and J. W. Wilkins,

429: Phys. Rev. Lett. {\bf 61}, 125 (1988); T. Nishino and K. Ueda,

430: Phys. Rev. B{\bf 47}, 12451 (1993).

431: \bibitem{totalspinDMRG}I. P. McCulloch and M. Gulacsi, cond-mat/0012319.

432: \bibitem{zener}C. Zener, Phys. Rev. {\bf 82}, 403 (1951); P. W. Anderson

433: and H. Hasegawa, Phys. Rev. {\bf 100}, 675 (1955).

434: \bibitem{Jullien}R. Jullien, J. N. Fields, and S. Doniach,

435: Phys. Rev. B {\bf 16}, 4889 (1977).

436: \bibitem{largeJ}M. Sigrist, H. Tsunetsugu, K. Ueda and T. M. Rice,

437: Phys. Rev. B{\bf 46}, 13838 (1992).

438: \bibitem{polaron}M. Sigrist, H. Tsunetsugu and K. Ueda, Phys. Rev. Lett.

439: {\bf 67}, 2211 (1991).

440: \bibitem{holstein}T. Holstein, Ann. Phys. (N.Y.) {\bf 16}, 407 (1961).

441: \bibitem{shibata}N. Shibata and H. Tsunetsugu, J. Phys. Soc. Japan

442: {\bf 68}, 3138 (1999).

443: \bibitem{largeFS}N. Shibata, K. Ueda, T. Nishino and

444: C. Ishii, Phys. Rev. B{\bf 54}, 13495 (1996).

445: \bibitem{exactFM}H. Tsunetsugu, M. Sigrist and K. Ueda,

446: Phys. Rev. B{\bf 47}, 8345 (1993).

447: \bibitem{numerics}M. Troyer and D. W\"{u}rtz, Phys. Rev.

448: B{\bf 47}, 2886 (1993); S. Moukouri and L. G. Caron,

449: Phys. Rev. B{\bf 52}, R15723 (1995); S. Caprara and A. Rosengren,

450: Europhys. Lett. {\bf 39}, 55 (1997); I. P. McCulloch, M. Gulacsi,

451: S. Caprara, A. Juozapavicius, and A. Rosengren, J. Low Temp. Phys.

452: {\bf 117}, 323 (1999).

453: \bibitem{dagotto}S. Yunoki, {\sl et al.}, Phys.

454: Rev. Lett. {\bf 80}, 845 (1998); E. Dagotto, {\sl et al.}, Phys.

455: Rev. B{\bf 58}, 6141 (1998).

456: \end{references}

457: 

458: \end{document}

459: 

460: 

461: