1: \documentclass[prl,twocolumn,showpacs,nobibnotes]{revtex4}

2: 

3: \usepackage{bm}

4: \usepackage[dvips,final]{graphics}

5: 

6: \begin{document}

7: 

8: \title{Staggered Flux Phase in a Model of Strongly Correlated Electrons}

9: 

10: \author{J. B. Marston} 

11: \author{J. O. Fj{\ae}restad}

12: \affiliation{Department of Physics, Brown University, Providence, RI

13: 02912-1843}

14: 

15: \author{A. Sudb{\o}} 

16: \affiliation{Department of Physics, Norwegian University of Science

17: and Technology, N-7491 Trondheim, Norway}

18: 

19: \date{February 11, 2002}  

20: 

21: \begin{abstract}

22: We present numerical evidence for the existence of a staggered flux (SF) 

23: phase in the half-filled two-leg t-U-V-J ladder, with true long-range order 

24: in the counter-circulating currents.  The density-matrix renormalization-group 

25: (DMRG) / finite-size scaling approach, generalized to describe complex-valued 

26: Hamiltonians and wavefunctions, is employed.  The SF phase exhibits robust 

27: currents at intermediate values of the interaction strength.

28: \end{abstract}

29: 

30: \pacs{71.10.Fd, 71.10.Hf, 71.30.+h, 74.20.Mn}

31: 

32: \maketitle

33: 

34: The zero-temperature phase diagram of the two-dimensional Hubbard

35: model and its various extensions remains poorly understood.  An

36: intriguing possibility that has been the focus of considerable

37: attention is known variously as the ``orbital

38: antiferromagnet''\cite{halperin,OAF1,OAF2}, the ``staggered flux''

39: (SF) phase\cite{AM,MA,ted,ivanov,leung,didier}, or the ``d-density

40: wave''\cite{chetan,nayak}.  The state breaks both time-reversal 

41: and lattice translation symmetries; another phase, 

42: the ``circulating current phase''\cite{varma}, is similar but

43: preserves translational symmetry.  Such ``hidden'' forms

44: of order could arise in the pseudogap phase of the cuprate

45: superconductors, implying the existence of a quantum critical point

46: and possibly also non-Fermi liquid behavior\cite{nayak,varma}. The SF

47: phase competes against other better-known phases such as

48: charge-density waves (CDW), spin-density waves (SDW),

49: superconductivity (BCS), stripes, and Fermi liquids.  It is important

50: to ascertain whether or not such order can really occur in sensible

51: microscopic models of correlated electrons.

52: In this Letter we study numerically the

53: simplest possible system that could support such a phase, namely an

54: extended version of the Hubbard model on a two-leg ladder (see Fig.

55: \ref{fig:ladder}).  Our approach extends to arbitrary interaction strength

56: earlier analytical work by two of us that showed that the SF phase arises

57: at weak-coupling\cite{john}.  

58: 

59: Ladder systems are of interest both because they

60: are easier to understand theoretically than full two-dimensional lattices,

61: and also because they are realized in nature\cite{dagotto}. Although

62: ladders are one-dimensional, true long-range order (LRO) in the

63: orbital currents is possible at zero temperature because the currents

64: break discrete symmetries. This is in contrast to spin-density wave or

65: BCS superconducting order which break continuous symmetries and which

66: therefore can at most exhibit quasi-LRO with power-law decay. 

67: 

68: \begin{figure}

69: \resizebox{8cm}{!}{\includegraphics{ladder.eps}}

70: \caption{Upper diagram: A two-leg ladder of length $L = 8$.

71: The left and right blocks each retain Hilbert spaces of dimension $M$.  

72: To implement the DMRG / finite-size algorithm, the two sites belonging

73: to each rung are paired together into a single site with a Hilbert space

74: of complex dimension $D = 16$.  Lower diagram: Source current $j_{SF}$ applied to the

75: left-most rung induces currents in the interior of the ladder

76: (dashed circles). A source of Cooper pairs ($j_{BCS}$), a local chemical

77: potential ($j_{CDW/SP}$), and a local magnetic field ($j_{SDW}$) are added 

78: at the ladder ends to induce BCS, SP, CDW, and SDW order in the central

79: region of the ladder, which is then monitored as the ladder length increases.} 

80: \label{fig:ladder}

81: \end{figure}

82: 

83: Away from half-filling, numerical studies of the t-J

84: model\cite{SWA,Asle} and various extensions of it\cite{troyer} on the

85: two-leg ladder found no evidence for a SF phase. However, for weak

86: interactions, tendencies towards SF ordering (not true LRO, but

87: rather power-law decay of current-current correlation functions) 

88: have been found analytically

89: both for the spinless\cite{ners-luther-kus} and spinful\cite{schulz96}

90: cases (see also Ref.~\onlinecite{orignac}).

91: 

92: Here we will consider the half-filled two-leg ladder. We study an

93: extended Hubbard model, the t-U-V-J model, defined on the two-leg

94: ladder as follows:

95: \begin{widetext}

96: \begin{eqnarray}

97: H &=& \sum_{i=1}^{L-1} \sum_{\lambda=t,b} \bigg{\{}

98: -t_\parallel (c^{\dagger \sigma}_{i+1,\lambda} c_{i,\lambda,\sigma} + H.c.)

99: + V_\parallel (n_{i+1,\lambda} - 1) (n_{i,\lambda} - 1)

100: + J_\parallel \vec{S}_{i+1,\lambda} \cdot \vec{S}_{i,\lambda}

101: \bigg{\}}

102: \nonumber \\

103: &+& \sum_{i=1}^L \bigg{\{}

104: -t_\perp (c^{\dagger \sigma}_{i,t} c_{i,b,\sigma} + H.c.)

105: + V_\perp (n_{i,t} - 1) (n_{i,b} - 1) 

106: + J_\perp \vec{S}_{i,t} \cdot \vec{S}_{i,b}

107: \bigg{\}} + \frac{U}{2}~ \sum_{i=1,\lambda}^L (n_{i,\lambda} - 1)^2\ . 

108: \label{Hamiltonian}

109: \end{eqnarray}

110: \end{widetext}

111: Here $c^{\dagger \sigma}_{i \lambda}$ creates an electron of spin

112: $\sigma = \uparrow, \downarrow$ on site $i$ of leg $\lambda = t, b$ of

113: the ladder.  Operators $n_{i,\lambda} \equiv c^{\dagger

114: \sigma}_{i,\lambda} c_{i,\lambda,\sigma}$ and $\vec{S}_{i,\lambda}

115: \equiv \frac{1}{2} c^{\dagger \alpha}_{i,\lambda}

116: \vec{\sigma}_\alpha^\beta c_{i,\lambda,\beta}$ are respectively the

117: number and spin at site $(i,\lambda)$.  There is an implicit sum over

118: repeated raised and lowered spin indices. Particle-hole symmetry

119: implies that the chemical potential $\mu = 0$ when the system is

120: half-filled, with $\langle n_{i,\lambda} \rangle = 1$ on each site.

121: 

122: For $V_\parallel = J_\parallel = 0$ and

123: $J_\perp = 4 (U + V_\perp)$ the model has SO(5)

124: symmetry\cite{SZH}.  The phase diagram of the SO(5)-symmetric model 

125: with $t_\perp = t_\parallel$ was investigated in Ref.~\onlinecite{LBF98}

126: in the weak-coupling limit using a perturbative renormalization-group (RG)

127: analysis combined with bosonization. One of the phases of the SO(5) model, 

128: located in the $U-V_{\perp}$ plane between the lines $V_{\perp} = -2 U$ and 

129: $V_{\perp} \approx -5.7 U$ with $V_{\perp}>0$ and $U<0$, was identified as

130: having spin-Peierls (SP) order. However, a recent bosonization study

131: by two of us\cite{john} revealed that this phase in actuality exhibits

132: no dimerization, but instead is the SF phase with true LRO in the

133: orbital currents. For a correct understanding of the problem it was

134: essential to treat carefully the ``Klein factors'' that must be

135: introduced to maintain anti-commuting statistics of the (bosonized)

136: fermionic degrees of freedom.

137: 

138: Previous RG studies have shown that in weak coupling a rather generic

139: ladder model flows to a manifold with SO(5) symmetry for a range of

140: values of the model parameters\cite{LBF98,arrigoni}. In agreement with

141: these results, we have shown that there are RG flows towards the SF phase

142: also when not too large non-SO(5)-invariant terms (such as $J_{\parallel}$ and

143: $V_{\parallel}$) are added to the SO(5)-model\cite{in_preparation}. 

144: 

145: Weak-coupling RG is unreliable at intermediate interaction strengths.  

146: Instead we use the ``infinite-size'' version of the density-matrix 

147: renormalization-group

148: (DMRG) algorithm\cite{white} to search for the ordered currents.  The

149: half-filled ladder is expected to be fully gapped over a large portion 

150: of the phase diagram; consequently (as shown

151: below) the infinite-size algorithm is sufficiently accurate to obtain

152: well-converged results.  Each site of the ladder has a Hilbert space

153: of dimension $4$, as the site can either be unoccupied, have a single

154: electron of either up or down spin, or be doubly occupied. To employ

155: the DMRG algorithm, we group pairs of sites connected by a rung into a

156: single site of Hilbert space dimension $D = 4 \times 4 = 16$ 

157: (see Fig. \ref{fig:ladder}).  Errors

158: in the calculation of observables introduced by the DMRG truncation of

159: the Hilbert space can be systematically reduced by increasing the size

160: $M$ of the Hilbert space retained in each of the two outer blocks up

161: to limits imposed by computer memory and speed.  The  

162: largest Hilbert space we use has $M = 150$ with corresponding complex

163: dimension $M \times D \times D \times M = 5,760,000$.  

164: 

165: Questions of the spontaneous formation of order are addressed by the

166: combined DMRG / finite-size scaling approach described in some detail

167: in Ref.~\onlinecite{tsai}.  Quantum-critical points have been studied with

168: the method, and critical exponents have been obtained at percent-level

169: accuracy\cite{senthil,tsai}.  In the present case we induce

170: symmetry breaking by applying a source current $j_{SF}$ to the left end

171: rung of the ladder.  Terms that induce Cooper-pair formation by

172: breaking U(1) particle-number symmetry ($j_{BCS}$), induce CDW and

173: SP order by breaking lattice reflection symmetries ($j_{CDW/SP}$), 

174: and induce a SDW through the application of a local magnetic field ($j_{SDW}$), 

175: are also added to the right end of the ladder (see Fig. \ref{fig:ladder}):

176: \begin{widetext}

177: \begin{eqnarray}

178: H \rightarrow H &+& 

179: j_{SF} * i t_\perp (c^{\dagger \sigma}_{1,t} c_{1,b,\sigma} - H.c.)

180: + j_{BCS} * (c^{\dagger \uparrow}_{L,t} c^{\dagger \downarrow}_{L,b} 

181: - c^{\dagger \downarrow}_{L,t} c^{\dagger \uparrow}_{L,b} + H.c.)

182: \nonumber \\

183: &+& j_{CDW/SP} * n_{L,t} + j_{SDW} * S^z_{L,b}\ .

184: \end{eqnarray}

185: \end{widetext}

186: In addition to these explicit source terms, 

187: we note that the open boundary conditions on the ladder also

188: act as O(1) source terms for a columnar dimer pattern\cite{tsai}. 

189: As the Hamiltonian is complex-valued when $j_{SF} \neq 0$, we

190: generalize the DMRG code to handle complex-valued wavefunctions and

191: reduced density matrices.  This generalization comes at the cost of

192: doubling the required memory, and slowing down both the sparse and the

193: dense matrix diagonalization, but the cost is offset by the advantage

194: that now we can access fully ordered ground states, well beyond linear

195: response theory.

196: 

197: We calculate the expectation value of the current operators $2

198: t_\perp~ {\rm Im}\{c^{\dagger \alpha}_{i,t} c_{i,b,\alpha}\}$ and $2

199: t_\parallel~ {\rm Im}\{c^{\dagger \alpha}_{i+1,\lambda}

200: c_{i,\lambda,\alpha}\}$ respectively on the central rungs and links of

201: the ladder, checking that current is conserved (Kirchhoff's junction

202: rule) at the vertices in the central region; deviations are typically of

203: order $10^{-11}$ in units where $t_\perp = t_\parallel = 1$.  We also monitor the 

204: Cooper pair amplitude, $\langle c^{\dagger \uparrow}_{i,t} c^{\dagger \downarrow}_{i,b} 

205: - c^{\dagger \downarrow}_{i,t} c^{\dagger \uparrow}_{i,b} \rangle$; 

206: the average magnitude of deviations in the electron hopping amplitudes 

207: $\langle c^{\dagger \sigma}_{i, \lambda} c_{j, \lambda^\prime, \sigma} 

208: + H.c. \rangle$ from their mean value; deviations in the electron occupancy from one, 

209: $\langle n_{i,\lambda} - 1\rangle$; 

210: and the local spin density, $\langle S^z_{i,\lambda} \rangle$.  

211: 

212: First consider the pure minimal Hubbard t-U model with $t_\perp =

213: t_\parallel = 1$, $U = 4$, and $V_\perp = V_\parallel = J_\perp =

214: J_\parallel = 0$.  As shown in the semilog plot of Fig. \ref{fig:currents}(a), 

215: applied source currents of $j_{SF} = 1$ and $0.01$ induce currents on the four

216: central rungs of the ladder that decrease exponentially  

217: as the ladder is enlarged via the DMRG algorithm, in agreement with the

218: weak-coupling RG calculation\cite{LBF98}.  Likewise we find no

219: instabilities towards BCS, SP, CDW or SDW order.  Instead the phase is a

220: fully-gapped insulator qualitatively the same as that found in the 

221: Heisenberg antiferromagnet. 

222: 

223: \begin{figure}

224: \resizebox{8cm}{!}{\includegraphics{rung-log.eps}}

225: \resizebox{8cm}{!}{\includegraphics{rung-linear.eps}}

226: \caption{Semilog (a) and linear (b) plots of the magnitude of the induced current 

227: averaged over the four central rungs of the ladder. 

228: Dotted lines are for calculations with $M = 50$,

229: dashed lines are for $M = 100$, and solid lines are for $M = 150$.

230: Results are shown for the case of the SO(5) ladder with 

231: $t_\perp = t_\parallel = 1$, $U = -0.4$, $V_\perp = 0.9$, $J_\perp = 2$, and $V_\parallel =

232: J_\parallel = 0$.  

233: Also shown on the semilog plot is the case of the minimal Hubbard model with 

234: $U = 4$ and $j_{SF} = 0.01$ (lower line that drops exponentially) and 

235: $j_{SF} = j_{BCS} = j_{CDW/SP} = j_{SDW} = 1$ (upper lines that drop exponentially). 

236: In both plots, the lower set of SO(5) curves is for 

237: a small source current $j_{SF} = 0.01$ applied to the ladder's left edge. 

238: The upper set is for large source currents $j_{SF} = j_{BCS} = j_{CDW/SP} = j_{SDW} = 1$. 

239: Note the convergence to the same asymptotic value of current 

240: as the thermodynamic limit of long ladder length is approached.  The asymptotic current

241: for $M=100$ differs by 0.6\% from that at $M=50$; the value for $M=150$

242: differs by only $0.1\%$ from that at $M=100$ demonstrating good convergence.}  

243: \label{fig:currents}

244: \end{figure}

245: 

246: Next consider the effect of turning on interactions $J_\perp$ and

247: $V_\perp$ along the rungs of the ladder.  According to the phase diagram in 

248: Ref. \onlinecite{LBF98}, for the case of exact SO(5) symmetry with $J_\perp = 4 (U + V_\perp)$

249: the SF phase should occur\cite{john} in the weak-coupling limit 

250: for $U < 0$ at $V_\perp / U = -9/4$. 

251: Fig. \ref{fig:currents} shows that the SF phase does indeed arise at 

252: these ratios for intermediate interaction strength:  For

253: $U = -0.4$, $V_\perp = 0.9$, and $J_\perp = 2$, a small source current induces

254: orbital currents in the central region of the ladder that grow with increasing

255: ladder length.   The currents saturate to the same non-zero asymptotic 

256: value in the limit of large ladder length, regardless of the size of the 

257: source terms.  There is good convergence with increasing block Hilbert

258: space dimension $M$; the asymptotic value of the rung current 

259: for the most accurate truncation of $M=150$ is 0.8047 and differs by only 0.1\%

260: from the value obtained with $M=100$.  The currents alternate in sign from

261: rung to rung as expected in the SF phase.  There are no other

262: instabilities: As shown in Fig. \ref{fig:otherorder} 

263: the BCS, SP, CDW, and SDW order parameters all vanish exponentially 

264: as the ladder is enlarged via the DMRG algorithm, consistent with the

265: weak-coupling prediction.

266: The excitation spectrum remains fully gapped; consequently the SF phase

267: exists over a range of parameter space well beyond the region of 

268: exact SO(5) symmetry.  For example, robust currents arise 

269: for $U = -0.5$, $V_\perp = 0.75$, and $J_\perp = 2$,

270: and also when interactions along the links are turned on, $U = -0.5$, 

271: $V_\perp = V_\parallel = 0.75$, $J_\perp = 2$, and $J_\parallel = 0.5$. 

272:  

273: In the large-$J_\perp$ limit, with ratios $U / J_\perp = -1/4$ 

274: and $V_\perp / J_\perp = 3/8$ kept fixed, 

275: the ground state of the ladder consists of a direct product

276: of spin-singlet dimers on each of the rungs, with no broken

277: symmetries\cite{SZH}.  We have verified that the orbital currents 

278: cease when the interaction strength is increased further to

279: $J_\perp > 2.5$.  We are carrying out a more extensive investigation

280: of the phase diagram of model Eq. \ref{Hamiltonian} by using the

281: weak-coupling RG equations in combination with the DMRG / finite-size scaling

282: method\cite{in_preparation}.

283: 

284: \begin{figure}

285: \resizebox{8cm}{!}{\includegraphics{other.eps}}

286: \caption{Semilog plot of various order parameters in the SF phase (see text).

287: Edge source terms $j_{BCS} = j_{CDW/SP} = j_{SDW} = j_{SF} = 1$ 

288: induce modulations in corresponding observables at the center of the ladder.  

289: Dotted, dashed, and solid lines correspond to $M = 50$, $100$ and $150$ respectively.

290: Plotted in the figure, from highest to lowest in magnitude, are the order parameters 

291: for CDW, SP, BCS, and SDW tendencies.  See text for the definition of these

292: order parameters.  

293: Each of the expectation values decays exponentially with increasing ladder length.  

294: (The SP order parameter eventually levels off, but this residual order vanishes

295: as the block size $M$ is increased, showing that it is an artifact of the Hilbert

296: space truncation.) 

297: Similar results are obtained upon applying either large or small source terms, one

298: at a time. 

299: Thus there is no tendency towards BCS, SP, CDW or SDW order in the SF phase.}

300: \label{fig:otherorder}

301: \end{figure}

302: 

303: It would be interesting to study the effects of doping away from half-filling.

304: Numerical study of the doped SF phase would require the use

305: of the more accurate ``finite-size'' DMRG algorithm, 

306: as either gapless excitations, or rich 

307: spatial structures such as stripes, are expected to occur\cite{WAS}.  

308: One question to be answered is whether or not LRO in the currents can 

309: still arise at commensurate hole concentrations. 

310: 

311: In summary we have shown that a phase of strongly correlated electrons

312: exists in which currents form spontaneously in the half-filled two-leg ladder.

313: At half-filling the SF phase is fully gapped, and thus is an insulator of

314: the Mott-Hubbard type.  Our results, suitably generalized to two or three 

315: spatial dimensions, may have application to several stoichiometric compounds that were

316: recently proposed to be in the SF phase\cite{stoichiometric}.

317: It is also intriguing that the SF phase lies in between a checkerboard CDW phase, 

318: and the D-Mott phase\cite{LBF98,john,in_preparation}, as

319: charge segration into stripes and d-wave superconductivity are two phases

320: that occur in the high-T$_c$ cuprates.

321: 

322: {\it Note added:} U. Schollw{\"o}ck has recently used the ``finite-size'' version

323: of the DMRG algorithm, generalized to complex-valued wavefunctions, to study the

324: SO(5) ladder with the same parameters as Fig. \ref{fig:currents}. 

325: The value of the saturated rung current that he finds agrees quantitatively with our result. 

326: 

327: %%%%%%%%%%%%%%%%%%%%%%

328: We thank I. Affleck, L. Balents, S. Chakravarty, S. Kivelson,

329: D.-H. Lee, P. Lee, D. Scalapino, U. Schollw{\"o}ck, M. Troyer, 

330: S.-W. Tsai, C. Varma, and S. White for helpful discussions.  This

331: work was supported in part by the NSF under grant Nos. DMR-9712391,

332: CDA-9724347, and PHY99-07949.  

333: J.O.F. was supported by the Norwegian Research Council,

334: Grant No 142915/432, and A.S. by Grant No. 115541/410

335: Some computational work was carried out at Brown University's

336: Technology Center for Advanced Scientific Computing and Visualization.

337: 

338: \begin{thebibliography}{}

339:   

340: \bibitem{halperin} B. I. Halperin and T. M. Rice, Solid State Phys.

341: {\bf 21}, 115 (1968).

342: 

343: \bibitem{OAF1} H. J. Schulz, Phys. Rev. B{\bf 39}, 2940 (1989). 

344:   

345: \bibitem{OAF2} A. A. Nersesyan and G. E. Vachnadze, J. Low Temp. Phys.

346: {\bf 77}, 293 (1989).

347: 

348: \bibitem{AM} I. Affleck and J. B. Marston, Phys. Rev. B{\bf 37}, 3774 (1988). 

349: 

350: \bibitem{MA} J. B. Marston and I. Affleck, Phys. Rev. B{\bf 39}, 11538 (1989).

351:   

352: \bibitem{ted} T. C. Hsu, J. B. Marston, and I. Affleck, Phys. Rev.

353: B{\bf 43}, 2866 (1991).

354:   

355: \bibitem{ivanov} D. A. Ivanov, P. A. Lee, and X.-G. Wen, Phys. Rev.

356: Lett. {\bf 84}, 3958 (2000).

357: 

358: \bibitem{leung} P. W. Leung, Phys. Rev. B{\bf 62}, R6112 (2000). 

359: 

360: \bibitem{didier} K. Tsutsui, D. Poilblanc, and S. Capponi, Phys. Rev.

361: B{\bf 65}, 020406 (2001).

362: 

363: \bibitem{chetan} C. Nayak, Phys. Rev. B{\bf 62}, 4880 (2000).  

364: 

365: \bibitem{nayak} S. Chakravarty, R. B. Laughlin, D. K. Morr, and C. Nayak, 

366: Phys. Rev. B{\bf 63}, 094503 (2001).

367: 

368: \bibitem{varma} C. M. Varma, Phys. Rev. B{\bf 55}, 14554 (1997);

369: Phys. Rev. Lett. {\bf 83}, 3538 (1999); Phys. Rev. B{\bf 61}, 3804 (2000).

370: 

371: \bibitem{john} J. O. Fj{\ae}restad and J. B. Marston,

372: Phys. Rev. B{\bf 65}, 125106 (2002).

373: 

374: \bibitem{dagotto} E. Dagotto and T. M. Rice, Science {\bf 271}, 618 (1996).

375: 

376: \bibitem{SWA} D. J. Scalapino, S. R. White, and I. Affleck,

377: Phys. Rev. B{\bf 64} 100506 (2001).

378: 

379: \bibitem{Asle} J. B. Marston and A. Sudb{\o}, unpublished.

380: 

381: \bibitem{troyer} S. Chakravarty, U. Schollw{\"o}ck, and M. Troyer,

382: private communication (2001).

383: 

384: \bibitem{ners-luther-kus} A. A. Nersesyan, A. Luther, and F. Kusmartsev,

385: Phys. Lett. A \textbf{176}, 363 (1993).

386: 

387: \bibitem{schulz96} H. J. Schulz, Phys. Rev. B \textbf{53}, R2959 (1996).

388: 

389: \bibitem{orignac} E. Orignac and T. Giamarchi, Phys. Rev. B{\bf 56},

390: 7167 (1997).

391: 

392: \bibitem{SZH} D. Scalapino, S.-C. Zhang, and W. Hanke, Phys. Rev. B{\bf 58}, 

393: 443 (1998).

394: 

395: \bibitem{LBF98} H. H. Lin, L. Balents, and M. P. A. Fisher,

396: Phys. Rev B{\bf 58}, 1794 (1998).

397:  

398: \bibitem{arrigoni} E. Arrigoni and W. Hanke, Phys. Rev. Lett. 

399: {\bf 82}, 2115 (1999). 

400:  

401: \bibitem{in_preparation} J. O. Fj{\ae}restad, J. B. Marston, and A. Sudb{\o}, 

402: in preparation.

403: 

404: \bibitem{white} S. R. White, Phys. Rev. Lett. {\bf 69}, 2863 (1992);

405: Phys. Rev. B{\bf 48}, 10345 (1993).

406: 

407: \bibitem{tsai} S.-W. Tsai and J. B. Marston, Phys. Rev. B{\bf 62}, 5546 (2000). 

408: 

409: \bibitem{senthil} T. Senthil, J. B. Marston, and M. P. A. Fisher, 

410: Phys. Rev. B{\bf 60}, 4245 (1999). 

411: 

412: \bibitem{WAS} S. R. White, I. Affleck, and D. J. Scalapino, 

413: Phys. Rev. B{\bf 65}, 165122 (2002).

414: 

415: \bibitem{stoichiometric} P. Chandra, P. Coleman, and J. A. Mydosh,

416: Physica B 312-313, 397 (2002); P. Chandra, P. Coleman, J. A. Mydosh, and

417: V. Tripathi, cond-mat/0205003 (to appear in Nature); 

418: D. F. Schroeter and S. Doniach, cond-mat/0201524.

419: 

420: \end{thebibliography}

421: \end{document}

422: