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

2: %documentstyle[aps,prl,multicol,fancyheadings,epsfig,amsbsy,amssymb,amstex]{revtex}

3: 

4: \begin{document}

5: % \draft command makes pacs numbers print

6: \draft

7: %inerst title

8: \title{Itinerant Ferromagnetism for Mixed Valence Systems}

9: % repeat the \author\address pair as needed

10: \author{C. D. Batista,$^1$ J. Bon\v ca,$^2$ and J. E. Gubernatis$^1$}

11: \address{$^1$Center for Nonlinear Studies and Theoretical Division

12: Los Alamos National Laboratory, Los Alamos, NM 87545\\

13: $^2$ Department of Physics, FMF, University of Ljubljana and J.

14: Stefan Institute, Ljubljana, Slovenia}

15: \date{\today}

16: \maketitle

17: \begin{abstract}

18: 

19: We introduce a novel mechanism for the unusual itinerant

20: ferromagnetism found in mixed valence systems like

21: Ce(Rh$_{1-x}$Ru$_x$)$_3$B$_2$ , La$_x$Ce$_{1-x}$Rh$_3$B$_2$ , US,

22: USe, and UTe.

23: %We solve the Periodic Anderson model by means of the Constrained

24: %Path Monte Carlo method and a mean field theory. To derive the

25: %experimental consequences of the new mechanism.

26: %Using the Constrained-Path Monte Carlo method and mean-field

27: %theory, we show this mechanism is modeled by the periodic Anderson

28: %model.

29: With it we can provide an explanation for the long-unexplained

30: large value of $T_c$ ($\sim$ 100$^\circ$K) value and the maximum

31: in the magnetization below $T_c$ found experimentally.

32: %, and  to complete

33: %%the characterization of this phase and confirmation of the

34: %mechanism we propose experiments.

35: We also show that this novel

36: itinerent ferromagnetism can be continuously connected with the

37: localized case for which the energy scale is much smaller

38: ($J_{RKKY} \sim$ 1$^\circ$K).

39: 

40: 

41: \end{abstract}

42: % insert suggested PACS numbers in braces on next line

43: \pacs{}

44: 

45: %\vspace*{-0.4cm}

46: \begin{multicols}{2}

47: 

48: \columnseprule 0pt

49: 

50: \narrowtext

51: %\vspace*{-0.5cm}

52: % body of paper here

53: %\section{Introduction}

54: {\it Introduction.} Understanding the mechanisms for

55: ferromagnetism usually involves a reconciliation of a localized

56: electron picture, traced back to Heisenberg \cite{heisenberg}, and

57: an itinerant electron picture, traced back to Bloch \cite{bloch}.

58: For many insulating materials, adding itinerant features, such as

59: indirect exchange, to the localized picture brings satisfactory

60: agreement with basic experimental features. For many metals,

61: adding localized features, such as spin waves, to the itinerant

62: picture has a similar effect. On the other hand it is now

63: appreciated that for some novel classes of materials, such as the

64: heavy Fermion and mixed valence materials, doctoring one picture

65: or the other is often a questionable procedure. The electrons can

66: display both localized and itinerant properties as these materials

67: are neither good insulators nor good metals.

68: 

69: In this letter we propose a simple mechanism for ferromagnetism

70: (FM) in heavy fermion mixed valence materials. The mechanism

71: relies on several specific energy scales set by the band structure

72: of these materials even though the electron correlation energy is

73: the dominant energy scale. The simplicity and generality of the

74: mechanism however should provide a useful framework for discussing

75: ferromagnetism in broad classes of heavy fermion mixed valence

76: materials often found in 4$f$ and 5$f$ materials and their compounds.

77: In fact we will show that it seems to explain some of the very

78: unusual long unexplained ferromagnetic properties of such Ce based

79: compounds as Ce(Rh$_{1-x}$Ru$_x$)$_3$B$_2$ \cite{malik}

80: and the uranian monochalcogenides \cite{santini}.

81: 

82: The mechanism is based upon a two-band structure. The lower band

83: near the zone center is dispersive and nearly free of electron

84: correlation effects. Away from the zone center, this band is very

85: flat and doubly occupied states experience strong correlation

86: effects. The upper band is separated from the lower band by a gap

87: that is smaller than the correlation energy. When the chemical

88: potential is at the flat band level, the system becomes unstable

89: to the formation of a ferromagnetic state. There are compounds

90: satisfying this mixed valence requirement under normal conditions

91: or  by either adding additional electrons (doping), applying

92: pressure, or changing the temperature. This ferromagnetic state is

93: a manifestation of a Hund-like rule among electrons in band states as

94: opposed to usual case of electrons in localized states (atomic

95: orbitals).

96: 

97: We will show that the periodic Anderson model (PAM) is a simple

98: microscopic realization of this picture. Using quantum Monte Carlo

99: simulations, we will demonstrate that the model admits a

100: ferromagnetic ground state in the mixed valence regime with

101: electron occupancies reflecting our physical picture. We then

102: develop a mean field picture that reproduces the ground state

103: properties of the PAM and also allows us to compute finite

104: temperature properties. Under certain conditions we find the

105: finite temperature mean-field approximation predicts the same highly unusual

106: behavior as a peak, below $T_c$, in the temperature dependence of

107: the magnetization \cite{malik,santini} and the deviation of 

108: the inverse susceptibility from Curie-Weiss behavior \cite{cheche}.

109: 

110: {\it Model Hamiltonian.} The PAM is described by the Hamiltonian

111: \begin{eqnarray}

112:   H &=& -t\sum_{\langle i,j \rangle,\sigma} (d_{i\sigma}^\dagger

113:   d^{}_{j\sigma}+d_{j\sigma}^\dagger d^{}_{i\sigma})

114:   +V\sum_{i,\sigma} (d_{i\sigma}^\dagger

115:   f^{}_{i\sigma}+f_{i\sigma}^\dagger d^{}_{i\sigma}) \nonumber \\ & &

116:   \quad\quad +\epsilon_f\sum_{i,\sigma}n_{i\sigma}^f +\frac{U}{2}

117:   \sum_{i,\sigma}n_{i\sigma}^fn_{i\bar {\sigma}}^f

118: \label{eq:pam}

119: \end{eqnarray}

120: where $d_{i\sigma}^\dagger$ and $f_{i\sigma}^\dagger$ create an

121: electron with spin $\sigma$ in $d$ and $f$ orbitals at lattice

122: site $i$ and $n^f_{i\sigma}=f^{\dagger}_{i\sigma}f^{}_{i\sigma}$

123: is the number operator for the $f$-electrons of spin ${\sigma}$ at

124: site $i$. The hopping amplitude $t$

125: between $d$-orbitals is only to nearest-neighbor sites. The

126: hopping amplitude $V$ hybridizes different orbitals on the same

127: site.

128: 

129: When $U=0$, the resulting non-interacting Hamiltonian $H_0$ has

130: two dispersive bands:

131: \begin{equation}

132:  E_\sigma^\pm({\bf k})=\frac{1}{2} \Biggl[

133:    e_{\bf k}+\epsilon_f \pm \sqrt{(e_{\bf k}

134:                                             -\epsilon_f)^2+4V^2}

135:    \Biggr]

136: \end{equation}

137: separated by an hybridization gap. 

138: For a chain, the band structure is illustrated in Fig.~1a.

139: Irrespective of spatial dimension, each band has regions where the

140: ${\bf k}$-states are dominantly of $d$ or $f$-character. A mixed valence

141: condition arises when the Fermi energy $E_F$ sits in a crossover

142: region which is around the energy $\epsilon_f$. Three energies are

143: important: the hybridization gap $\Delta$, the $f$-state

144: dispersion $\delta_f=\partial E^{-}/\partial k|_{k_F}=\hbar v_F$ ($v_F$

145: is the Fermi velocity), 

146: and the Coulomb (correlation) energy $U$.

147: $\delta_f$  is measuring  the band dispersion close to the Fermi level. In the 

148: interacting problem we will

149: assume that $U>\delta_f,\Delta$. As we shall now discuss, the

150: competition is between $\delta_f$ and $\Delta$ which will be a

151: competition between a ferromagnetic and paramagnetic state.

152: 

153: {\it Mechanism for Ferromagnetism.} In Fig.~1a-c we illustrate the

154: physical mechanism for the ferromagnetic state which emerges when

155: the electron filling $\rho$ is arround 1/4 

156: (one electron per lattice site) 

157: and the energy $\epsilon_f$ of the $f$-orbitals is

158: close to $E_F$. The latter condition defines the intermediate

159: valence regime.

160: 

161: To explain the mechanism, we first consider the non-interacting

162: case (Fig.~1a). States in the lower band with mainly $f$-character

163: have a small dispersion $\delta_f$ because of the absence of

164: direct hopping between the $f$-orbitals. In the interacting case,

165: because the Coulomb interaction only affects the electrons in the

166: $f$-orbitals, the electrons which were doubly occupying the states

167: in the lower band with mainly $d$-character are practically

168: unaffected. On the other hand, the electrons in the many states

169: which are close to the Fermi level and have mainly an

170: $f$-character will be strongly affected. These electrons spread to

171: higher unoccupied ${\bf k}$-states in the $f$-part of the band and

172: polarize by an in-band Hund's rule (Fig.~1b). To see the analogy

173: to Hund's rule more clearly, we consider the limiting case,

174: represented in Fig.~1c, where $\delta_f=0$ and $\Delta \neq 0$,

175: reducing Fig.~1b to a two level system in momentum space with each

176: level being strongly degenerate (we only plot the states with predominant $f$

177: character). If we add electrons to this two levels system 

178: it is easy to show that the ferromagnetic solution has the lowest energy and is

179: therefore the ground state. By polarizing, the spatial part of the 

180: wave function becomes antisymmetric and there is no double occupancy 

181: in the real space. In this way, the Coulomb repulsion is reduced 

182: to zero. The kinetic energy has also the lowest possible value if the electrons

183: are occupying the lower energy (degenerated) levels (see Fig.~1c).  

184: 

185: 

186: 

187: We now describe the conditions for the stability of the 

188: ferromagnetic state, and we will see how this 

189: mechanism departs from the one of Stoner \cite{stoner}.

190: To this end we state the condition $U \gg \Delta$. From Fig.~1b

191: it is clear that the cost of the spreading of the ferromagnetic 

192: solution is an increase in the kinetic energy proportional to $\delta_f$.

193: On the other hand, if we build up a nonmagnetic solution only

194: using the states of the lower band, there will be a minimum delocalization

195: for each particle due to the fact that a finite set of $f$-character 

196: states is in the upper band (see Fig.~1). This can be seen by 

197: constructing an $f$ Wannier function with the lower band states.

198: This Wannier function will have a delocalization lenght which depends 

199: on the ${\bf k}$ wave vector where the two bands are crossing.  

200: Therefore, by localizing in real space to avoid the double occupancy, 

201: the electrons will have a finite

202: probability of occupying the $f$-states in the upper band. The

203: energy cost per electron of occupancy is proportional to $\Delta$.

204: If the hybridization gap is much larger than the dispersion of the

205: states, that is, $\Delta \gg \delta_f$, the ferromagnetic state lies

206: lower in energy than the nonmagnetic state. The energy of the

207: excited nonmagnetic state is proportional to $\Delta$. This can

208: be easily seen from the limiting (and non-realistic) case $\delta_f=0$ 

209: ploted in Fig.~1c, for which the analogy with the atomic shells and

210: the Hund's rule is evident. 

211: 

212: 

213: \begin{figure}[tbp] 

214: \begin{center}

215: \vspace{-1.0cm}

216: \epsfig{file=Fig1ab.ps,width=70mm,angle=-0}

217: \end{center}

218: \end{figure}

219: \begin{figure}[tbp]

220: \begin{center}

221: \vspace{-3.0cm}

222: \epsfig{file=Fig1cprl.ps,width=68mm,angle=-0}

223: \vspace{-5.1cm}

224: \end{center}

225: \caption{Illustration of the ferromagnetic mechanism.}

226: \label{fig1}

227: \end{figure}

228: 

229: 

230: It is important to emphasize that the hybridization gap $\Delta$

231: and $f$-state dispersion $\delta_f$ are the two basic ingredients

232: of our mechanism: $\Delta$ only appears if there is more than one

233: band and $\delta_f$ is small only if the hybridization is weak. If

234: $U>\Delta,\delta_f$, then $\Delta$ is the energy scale of the

235: paramagnetic state, while $\delta_f$ is the energy scale of the

236: ferromagnetic solution. We can estimate the magnitude of these

237: scales for the PAM: for $V \lesssim t/2$, we get $\delta_f \sim

238: \partial e_k/\partial k|_{k_F} V^2/(e_{k_F}-\epsilon_f)^2$ 

239: and $\Delta \sim V^2/(2dt-\epsilon_f)+

240: \sqrt{(2dt+\epsilon_f)^2+4V^2}-2dt$. Therefore, if $\epsilon_f$ is

241: close to the bottom of the lower band ($\epsilon_f+2dt \sim t$)

242: and the Fermi level is close to the top ($E_F \sim \epsilon_f-V^2/(2dt-\epsilon_f)$),

243: $\Delta$ is considerably larger than $\delta_f$ and the ground

244: state is ferromagnetic. We can see that the ferromagnetic

245: solution is stable for comfortably realistic values of the

246: parameters. It is clear from this analysis the particle density $\rho$

247: for a ferromagnetic solution must be close to quarter filling.

248: 

249: {\it Quantum Monte Carlo Method.} Using the constrained-path Monte

250: Carlo method (CPMC), we computed the ground-state properties of

251: the PAM on a square lattice. The CPMC method projects the ground

252: state from an initial state $|\psi_T\rangle$ by converting the

253: iterative procedure

254: \begin{equation}\label{eq:cpmc}

255:   |\psi_{i+1}\rangle =e^{-\tau H}|\psi_{i}\rangle

256: \end{equation}

257: into a branched random walk. The details of the method are

258: described elsewhere \cite{zhang} as are our prior uses of it on

259: the square PAM \cite{bonca,batista}.

260: 

261: The defining characteristic of the CPMC method is its elimination

262: of the fermion sign problem by excluding random walkers

263: $|\phi\rangle$ that violate $\langle \psi_T|\phi\rangle>0$. If

264: $|\psi_T\rangle$ were the exact ground state $|\psi_0\rangle$,

265: this constraint would generate an exact elimination of the sign

266: problem, and hence an exact solution. Extensive benchmarking

267: indicates the CPMC method provides accurate estimates of the

268: energy and various correlation functions.

269: 

270: We prepared the trial state in specific values of the total spin

271: and z-component of spin \cite{batista}. Because the

272: Hamiltonian(\ref{eq:pam}) conserves these quantum numbers, the

273: iterative process (\ref{eq:cpmc}) produces a ground state with the

274: same $S$ and $S_z$. In the ferromagnetic regime, when the number of

275: lattice sites $N$ was greater than the number of electrons $N_e$,

276: the resulting energy as a function of $S$ almost always showed a

277: minimum at a value of $S$ satisfying

278: $0<S<\frac{1}{2}(2N-N_e)$. Hence we typically found a partially

279: polarized ferromagnetic ground state.

280: 

281: For square lattices we also computed dependence of the static spin structure

282: factor and the electron occupancies both on position and wave

283: number space \cite{bonca}. Most notable for the present work is

284: the wave vector dependence of the different spin components of the

285: upper and lower band electron occupancies shown in Fig.~2. It has

286: the same quantitative features as Fig.~1b.

287: 

288: 

289: {\it Mean Field Theory.} The quantum Monte Carlo simulations are

290: limited to chains and square lattices of relatively small sizes at

291: zero temperature. However, we have found that their predictions

292: are well described by a simple spin-polarized mean-field

293: approximation. We can easily extend this approximation to larger

294: systems sizes in higher dimensions and at finite temperatures. We

295: can use large lattices in three-dimensions at finite temperatures

296: to compare with the results of experiments.

297: 

298: %There are two reasons to expect mean field theory to give a good

299: %description of the ferromagnetic phase: One, the condition

300: %$U>\Delta>\delta_f$ for the ferromagnetic state is satisfied even

301: %if $U<t$ (weak coupling). Two, the ferromagnetic ground state

302: %suggested above is very similar to a partially polarized

303: %non-interacting eigenstate. (Of course the picture described in

304: %Fig.1 only depicts the essential features of the mechanism.)

305: 

306: The mean field Hamiltonian $H_{MF}$ is

307: \begin{equation}

308:   H_{MF} = H_{0} + \frac{U}{2}

309: \sum_{i,\sigma} (\langle n_{i\sigma}^f \rangle n_{i\bar

310: {\sigma}}^f + n_{i\sigma}^f \langle n_{i\bar {\sigma}}^f \rangle-

311: \langle n_{i\sigma}^f \rangle  \langle n_{i\bar {\sigma}}^f

312: \rangle) \label{eq:mf}

313: \end{equation}

314: $\langle n_{i \uparrow}^f \rangle$ and $\langle n_{i \downarrow}^f

315: \rangle$ are the determined in a self-consistent way within the

316: mean-field theory. The translational symmetry of $H$ implies that

317: we can always find a translationally invariant ground state. This

318: means that $\langle n_{i\uparrow}^f \rangle$ and $\langle n_{i

319: \downarrow}^f \rangle$ will not depend on the site index $i$.

320: Therefore there are two variational parameters to be determined

321: from the mean-field equations: $\langle n_{\uparrow}^f \rangle$

322: and $\langle n_{\downarrow}^f \rangle$.

323: 

324: Even though the mean field approximation can give a good

325: description of the FM ground state, it is well known that such

326: approximations overestimate the energy of the paramagnetic phase

327: because they improperly estimate the real space correlations that

328: are very important for the paramagnetic solution. Therefore it is

329: crucial to check that the ground state is the FM one by using a

330: more accurate method to evaluate the energy of the PM state. To

331: this end we calculated the ground state of the PAM with $6 \times

332: 6$ unit cells using the CPMC technique. We

333: found excellent agreement with the energy and as seen in Fig.~2

334: the electron occupancies.

335: 

336: \begin{figure}[tbp]

337: \begin{center}

338: \vspace{-1.2cm}

339: \epsfig{file=Fig2prl.ps,width=80mm,angle=-90}

340: \vspace{-2.3cm}

341: \end{center}

342: \caption{Comparison between Mean Field and CPMC values of the 

343: occupation numbers of the non-interacting band states

344: for each spin polarization in a $6\times6$ cluster

345: ($\rho=7/24$,$V=0.5t$, $U=t$, and $\epsilon_f=-3t$). On the 

346: $x$-axis we only include the nonequivalent wave vectors ordered

347: according to increasing non-interacting energies.}

348: \label{fig2}

349: \end{figure}

350: 

351: {\it Consequences.} Bolstered by this agreement we used the

352: mean-field approximation to calculate various thermodynamic

353: properties of the ferromagnetic phase for the three-dimensioanl

354: cubic PAM. In Fig.~3a we show the temperature dependence of the

355: magnetization for different values of $\epsilon_f$. From the

356: arguments given above it is clear that the magnetization comes

357: from the electrons which are occupying the states with

358: $f$-character. By increasing $\epsilon_f$ we are reducing the

359: number of $f$-electrons and therefore the magnitude of the zero

360: temperature magnetization. Some critical value of $\epsilon_f$

361: exists where the chemical potential starts departing from

362: $\epsilon_f$ and a new energy scale $\epsilon_f-E_F$ emerges in

363: the system. This new energy scale is reflected in the appearance

364: of a magnetization peak. (See the $\epsilon_f=-t$ case in

365: Fig.~3a.) To understand the last statement one first has to

366: realize that the zero temperature magnetization is small due to

367: the reduction of the number of $f$-electrons. When the temperature

368: is of the order of $\epsilon_f-E_F$, electrons are promoted

369: electrons from the double occupied band states to the

370: $f$-character states which have a large entropy (large density of

371: states). The $f$-electrons will be polarized due to the energy

372: considerations above explained. In this way we can explain the

373: origin of the magnetization peak below $T_c$. It is important to

374: note that the source for the large entropy is associated with

375: charge and not with spin degrees of freedom. This fact explains

376: why a state with larger magnetization has a higher entropy. The

377: magnetization curves shown in Fig. 3a are in good qualitative

378: agreement with the magnetization versus temperature data measured

379: in Ce(Rh$_{1-x}$Ru$_{x}$)$_3$B$_2$ for different values of $x$

380: \cite{malik}.

381: 

382: % Brief description of $Ce(Rh$_{1-x}$Ru$_{x}$)$_3$B$_2$

383: 

384: \begin{figure}[tbp]

385: \begin{center}

386: \vspace{-1.2cm}

387: \epsfig{file=Fig3prl.ps,width=90mm,angle=-90}

388: \vspace{-2.5cm}

389: \end{center}

390: \caption{Mean field results ($\rho=1/4$,$V=0.5t$, $U=2t$) for (a) magnetization,

391: (b) inverse susceptibility,

392: (c) density of states and (d) specific heat.}

393: \label{fig3}

394: \end{figure}

395: 

396: 

397: Another interesting aspect of this ferromagnetism is the deviation

398: from linearity (Curie-Weiss behavior)

399: for the inverse magnetic susceptibility above

400: $T_c$ over a large temperature range (see Fig. 3b) \cite{cheche}. 

401: The behavior contrasts that predicted by theories based on 

402: localized magnetic moments.

403: 

404: From above  it is also clear that the change in entropy from the

405: magnetic to the paramagnetic phase depends on the number for $f$

406: electrons (spin degrees of freedom). This dependence is seen in

407: Fig.~3d where we plot the calculated specific heat for different

408: values of $\epsilon_f$.

409: 

410: A characteristic of this ferromagnetic state that has consequences

411: for the photoemission experiments is the implication that the

412: quasi-particle dispersion should be close to that of the electrons

413: in  the non-interacting case. We recall that the ferromagnetic

414: state described in Fig. 1 is very similar to a non-interacting

415: polarized state. As in the Stoner mechanism, the main role of the

416: Coulomb repulsion is to polarize states with well defined

417: momentum. The hybridization gap of the noninteracting solution is

418: most likely replaced by a pseudogap (see Fig.~3c). This pseudogap should be seen

419: in the optical conductivity measurements.

420: 

421: 

422: {\it Conclusions.} Our quantum Monte Carlo simulations and

423: mean-field calculations clearly show the existence of an itinerant

424: ferromagnetic phase in mixed valence materials that is supported

425: by the PAM. This new phase describes qualitatively  many experimental

426: features such as an unusually large value for $T_c$($\sim$

427: 100$^\circ$K) and the maximum in the magnetization below $T_c$

428: \cite{shaheen} that are found in such ferromagnetic compounds as

429: Ce(Rh$_{1-x}$Ru$_x$)$_3$B$_2$ \cite{malik},

430: Ce(Rh$_{1-x}$Os$_x$)$_3$B$_2$ \cite{malik}, and

431: La$_x$Ce$_{1-x}$Rh$_3$B$_2$ \cite{shaheen}.

432: 

433: Many previous theories of 4$f$ and 5$f$ electron materials treated

434: these materials as systems of localized moments in the

435: $f$-orbitals. These theories are thus unable to describe an

436: itinerant ferromagnetic phase and experimental consequences

437: \cite{santini} like the peak in the magnetization below $T_c$

438: observed in some of these systems \cite{malik,cheche}, the large

439: value of $T_c$, the deviation of $\chi^{-1}(T)$ from the

440: Curie-Weiss law above $T_c$ in the uranium monochalcogenides

441: \cite{cheche}, and the mixed valence behavior of these compounds

442: \cite{shaheen,kanter,erbudak}.

443: 

444: 

445: 

446: The physical picture just presented, when combined with our

447: previous results \cite{batista}, allows a reconciliation of the

448: localized and delocalized ferromagnetism pictures painted by

449: Heisenberg and Bloch seventy years ago in the sense that it is

450: possible to go continuously from the mixed valence FM state where

451: the $f$-electrons are delocalized to a FM state where there is one

452: localized electron in each $f$-orbital \cite{batista}. In our

453: picture one can do this by decreasing $\epsilon_f$ from the Fermi

454: level to a value near or below the bottom of the valence band. We

455: note that the energy scale of a localized ferromagnetic state is

456: that of the RKKY interaction, which according to de Gennes's rule

457: \cite{degennes} is of the order of 1$^\circ$K, while the scale for

458: the itinerant case introduced here is that of the hybridization

459: gap $\Delta$, which is of the order of 100$^\circ$K for heavy

460: fermion compounds \cite{note}. In going from the delocalized (mixed valence)

461: to the localized regime we thus expect a strong reduction of $T_c$

462: accompanied by an increase of the zero temperature magnetization.

463: (The $T=0$ magnetization is proportional to the $f$ occupation

464: number.) This expectation is consistent with the observed behavior

465: of the magnetization in La$_x$Ce$_{1-x}$Rh$_3$B$_2$ as function of

466: $x$ \cite{shaheen}.

467: 

468: 

469: 

470: {\it Acknowledgements.} This work was sponsored by the US DOE

471: under contract W-7405-ENG-36. We acknowledge useful discussions

472: with A. J. Arko, B. Brandow, M. F. Hundley, J. J. Joyce,   

473: J. M. Lawrence, S. Trugman, and J. L. Smith. In particular, 

474: we thank J. M. Lawrence 

475: for pointing out the experimental work on the Ce compounds.
J. B. 

476: acknowledges the support Slovene Ministry of Education, Science and

477: Sport, Los Alamos National Laboratory and FERLIN.

478: 

479: 

480: 

481: 

482: %              BIBLIOGRAPHY

483: 

484: \begin{thebibliography}{99}

485: 

486: \bibitem{heisenberg}

487: W. Heisenberg, Z. Physik {\bf 49}, 619 (198).

488: 

489: \bibitem{bloch}

490: F. Bloch, Z. Physik {\bf 57}, 545 (1929).

491: 

492: \bibitem{malik}

493: S. K. Malik, A. M. Umarji, G. k. Shenoy, and M. E. Reeves, Phys. Rev. B

494: {\bf 31}, 4728 (1985).

495: 

496: \bibitem{santini}

497: P. Santini, R. L\'emanski, and P. Erdos, Advan. in Phys. {\bf 48}, 537

498: (1999).

499: 

500: \bibitem{cheche}

501: V. I. Chechernikov, A. V. Pechennikov, E. I. Yarembash, L. F. Martinova,

502: and V. K. Slavyanskikh,

503: Sov. Phys. JETP {\bf 26}, 328 (1968).

504: 

505: 

506: \bibitem{stoner}

507: E. C. Stoner, Phil. Mag. {\bf 15},1018 (1933).

508: 

509: \bibitem{zhang}

510: Shiwei Zhang, J. Carlson and J. E. Gubernatis, Phys. Rev.

511: Lett., {\bf 74}, 3652 (1995); Phys. Rev. B {\bf 55}, 7464 (1997);

512: J. Carlson, J. E. Gubernatis, G. Ortiz, and Shiwei Zhang, Phys.

513: Rev. B {\bf 59}, 12788 (1999).

514: 

515: \bibitem{bonca}

516: J. Bon\v ca and J. E. Gubernatis, Phys. Rev. B {\bf 58}, 6992

517: (1998).

518: 

519: \bibitem{batista} C. Batista, J. Bon\v ca, and J. E. Gubernatis,

520: Phys. Rev. B, to appear (cond-mat/0009128).

521: 

522: %Experiments

523: 

524: 

525: \bibitem{shaheen}

526: S. A. Shaheen, J. S. Schilling, and R. N. Shelton, Phys. Rev. B {\bf 31},

527: 656 (1985).

528: 

529: \bibitem{degennes} P.-G. de Gennes, Comm. Energie At. (France)

530: Rappt. {\bf 925}, (1959).

531: 

532: \bibitem{note}

533: This change in the energy scale of both ferromagnetic states

534: is obtained with Monte Carlo by taking the energy difference between 

535: the PM and the FM solutions at $T=0$. The mean field theory

536: is not appropiate to describe this transition because it overestimates

537: the PM energy.

538: 

539: \bibitem{kanter}

540: M. A. Kanterand C. W. Kazmierowicz, J. Appl. Phys. {\bf 35}, 1053 (1954).

541: 

542: \bibitem{erbudak}

543: M. Erbudak and J. Keller, Z. Phys. B {\bf 23}, 281 (1979).

544: 

545: \end{thebibliography}

546: 

547: \end{multicols}

548: 

549: \end{document}

550: 

551: 

552: