Fluctuations near Quantum Critical Points of Valence Transition and Superconductivity Shinji WATANABE Department of Applied Physics, University of Tokyo Hongo 7-3-1, Bunkyo-ku Tokyo 113-0033 1 Introduction The superconductivity emerging on the verge of the magnetically ordered phase has been observed in various materials such as d- and felectron systems and organic compounds. The relation between the superconductivity and the magnetism has been intensively studied and roles of spin fluctuations in terms of mechanisms of the superconductivity which appears near the magnetic-instability point at zero temperature T = 0, i.e., the quantum critical point (QCP) have been argued. Recently, the superconductivity whose mechanism seems to be different from these types has been observed in Ce compounds and attracts much attention: In CeCu2 Ge2 [1], the superconducting transition temperature TSC has a maximum in the pressure regime far from the antiferromagnetic (AF) QCP, where the coefficient A in the resistivity ρ = ρ0 +AT n drops rapidly with n ∼ 2. Since A is proportional to the effective mass [2] (m∗ /m)2 which is related to the f-electron number as m∗ /m = (1 − nf /2)/(1 − nf ) [3], this implies that TSC has a maximum around the sharp valence change of Ce. Similar behavior has been observed in the isostructural compound CeCu2 Si2 [4] and in CeCu2 (Si1−x Gex )2 [5]. Recently, an NQR measurement has revealed that TSC increases under hydrostatic pressure in CeIrIn5 although AF spin fluctuation is suppressed [6], suggesting the different mechanism for the superconductivity from ordinary AF spin fluctuation. As known as the γ-α transition [7], Ce metal shows the first-order valence transition (FOVT) in its temperature-pressure phase diagram. The valence change of Ce is due to the 4f level located near the Fermi level, which can hybridize easily with the conduction band by applying pressure. At the critical point, the valence susceptibility diverges as diverging density fluctuation in the liquidgas transition. When the critical temperature is suppressed by controlling the chemical substitution and applying pressure, and enters the Fermi-degeneracy regime, diverging valence fluctuation is considered to be coupled with the Fermi-surface instability. This multiple instability in the quantum-degeneracy regime seems to be a key mechanism for understanding the instabilities observed in the above Ce compounds. However, theoretical understanding of the mechanism has not been fully achieved although a pioneering work on the valence-fluctuation-mediated superconductivity [8] and later discussions [9] have been made. Here, we report our recent studies on the nature of the QCP of the FOVT as well as the effects of fluctuations near the QCP. We show that the superconducting correlation becomes developed in the Kondo regime near the sharp valence crossover on the basis of the densitymatrix renormalization group (DMRG) [10] calculation for the one-dimensional periodic Anderson model which will be introduced in Sec.2. In Sec.3, the FOVT as well as the emergence of the QCP will be demonstrated and the ground-state phase diagram for the valence transition will be presented in Sec.4. In Sec.5 we will discuss the effects of valence fluctuations and will show that the superconducting correlations develop near the QCP. The detailed analysis of the Tomonaga-Luttinger parameter will be performed in two ways in Sec.6 and Sec.7, respectively. Finally, we summarize this report in Sec.8. 2 specially mentioned. Here, we consider infinite U for simplicity and basic results are unchanged if we set a finite value of U as far as U is dominant among the parameters in eq. (1). The number of states up to 1500 taken in our DMRG calculation allows well convergence for the open-boundary condition up to N = 80. Model We consider the periodic Anderson model with Coulomb repulsion between f and conduction electrons: H = −t † (ciσ ci+1σ + c†i+1σ ciσ ) + εf iσ +V fiσ ciσ + c†iσ fiσ + U +Ufc 3 nfiσ iσ † iσ N N nfi↑ nfi↓ Figure 1 shows the system size dependence of the f-electron number for Ufc = 7.0. We see that nf ∼ 1 is realized for εf = −6.0. As εf increases, nf at each system size decreases and we see that the system-size dependence of nf drastically changes between εf = −5.641 (solid triangle) and -5.639 (open diamond). For εf = −5.641, by taking the linear extrapolation using the data of N = 56, 64, 72 and 80, nf is estimated at 0.930 in the N → ∞ limit. For εf = −5.639, by using the data of N = 24, 32, 40, 48, 56 and 64, the linear extrapolation to the N → ∞ limit gives nf = 0.486. This sudden jump in nf indicates the firstorder transition between the Kondo state and the mixed-valence (MV) state. Since the linear dependence on 1/N holds quite well as seen in Fig. 1, we obtain nf extrapolated linearly to the N → ∞ limit, which is shown in Fig. 2(a). We see that nf jumps between εf = −5.641 and -5.639 as shown in Fig. 1, indicating the FOVT. The origin of the first-order transition is understood as follows: When εf is deep, the Kondo state with nf = 1 is realized to get the energy gain of the f-level. When εf is larger, the f electron is moved into the conduction band to avoid the energy loss by the shallow εf . As Ufc becomes large, the first-order transition takes place between the Kondo and the mixed valence (MV) states, since the large Ufc forces the electrons to pour exclusively into either the f level or the conduction band. An important point here is that in the MV state the f electrons are moved into the conduction band to avoid the energy loss by Ufc as minimum as possible, so that nf = 1/4 is realized for large Ufc . This is in striking contrast to i=1 nfi nci , (1) i=1 where ciσ (c†iσ ) and fiσ (fi†σ ) are the annihilation (creation) operators of the conduction and f electrons, respectively, at the ith site with spin σ in the N -lattice sites. Here, t is the transfer integral for the conduction electrons and εf is the one-body f -level energy. The third term describes the hybridization between f and conduction electrons. The term with U denotes the Coulomb repulsion for f electrons and the term with Ufc the Coulomb repulsion between f and conduction electrons. Here, the number operator is defined by naiσ = a†iσ aiσ and nai = σ naiσ for a = f and c. The filling n is defined by n = (nf + nc )/2 with na = First-order valence transition N 1 na N i=1 i for a = f and c. We note that the half filling is realized when n = 1 in this definition. For U = Ufc = 0, the f orbital and the conduction band are hybridized and the chemical potential is located at the lower hybridized band for n < 1. Then, the Fermi volume contains the f -electron number and the large Fermi surface is realized. Throughout this report, the transfer integral for conduction electrons t is taken as the energy unit. To discuss typical heavy-electron states, we consider the filling near half, n = 7/8 with U = 100 and V = 0.1 to simulate the infinite on-site Coulomb repulsion, unless 2 f 1 nf 0.8 0.6 0.4 0.2 0 0.05 0.1 -6.000 -5.800 -5.750 -5.700 -5.650 -5.641 -5.640 -5.639 -5.630 -5.600 -5.500 -5.300 nf 1/N phase -5 -4.5 -4 nf the MF theory which predicts the transition between nf = 1 and nf = 0 for large Ufc [12]. As Ufc is set to be smaller, the jump in nf becomes smaller. Figure 2(b) shows nf extrapolated to the N → ∞ limit for Ufc = 6.0, which still shows a jump in nf . On the other hand, we see that no jump appears in nf in Fig. 2(c) for Ufc = 5.0. These results indicate that the QCP of the valence transition, at which the valence susceptibility χf ≡ −∂nf /∂εf diverges is located between Ufc = 5.0 and 6.0. The accurate determination of the QCP concludes , UfcQCP ) = the location of the QCP as (εQCP f (−4.520, 5.846) [11]. We see that nf shows the crossover from the Kondo to the MV states for Ufc = 4.0 in Fig. 2(d) and the gradual crossover for Ufc = 0.0 as seen in Fig. 2(e). Ground-state gram -5.5 nf Figure 1: System-size dependence of f-electron number for t = 1.0, V = 0.1, U = 100 and Ufc = 7.0 at n = 7/8. 4 1 0.8 0.6 0.4 (a)U =7.0 fc 0.2 0 -6 1 0.8 0.6 0.4 (b) U =6.0 fc 0.2 0 -5 1 0.8 0.6 0.4 (c) U =5.0 fc 0.2 0 -4 1 0.8 0.6 0.4 (d) U =4.0 fc 0.2 0 -3.5 -3 1 0.8 0.6 0.4 (e) U =0.0 fc 0.2 0 -1 -0.5 -3 nf -3.5 nf -2.5 εf 0 diaFigure 2: The f-electron number in the bulk limit for t = 1, V = 0.1, U = 100 at n = 7/8 for (a)Ufc = 7.0, (b)6.0, (c)5.0, (d)4.0 and (e)0.0. In (b), symbols are the same as those in (a). The ground-state phase diagram in the plane of εf and Ufc is summarized in Fig. 3. The FOVT (the solid line with the filled diamonds) with the QCP (open circle) is determined by the DMRG method. The dashed line with open squares represent the crossover point between the Kondo and MV states, at which χf has a maximum as a function of εf for Ufc in the DMRG calculation. The FOVT (the solid line with the filled triangles) with the QCP QCP (εQCP f MF , Ufc MF ) = (0.189, 0.978) (open 3 diamond) determined by the slave-boson MF theory is also shown in Fig. 3. In contrast to the MF result (not shown in Fig. 3), we have found no evidence of the phase separation signaled by the diverging compressibility and the inhomogeneous distribution of the charge density by the DMRG calculation [12]. Furthermore, a remarkable result is that the QCP identified by the DMRG (open circle) is shifted from the QCP by the MF theory (open diamond) to a substantially larger |εf | and Ufc . This indicates that quantum fluctuations and correlation effects beyond the MF theory destabilizes the FOVT and terminates the phase separation. As a result, the crossover between the Kondo and MV states is realized in a wide region of the phase diagram. As shown in Fig. 2, the sharp valence crossover from the Kondo to MV states appears for Ufc = 4.0 and 5.0. In connection to the Ce compounds such as CeCu2 (Ge/Si)2 , this sharp crossover regime with strong valence fluctuation is most interesting, since the superconductivity is observed in this regime. This point will be discussed in the next section. 5 Enhancement correlation of 7 Mixed Valence 6 Ufc 5 QCP Kondo slave boson MF 4 3 2 1 0 -6 -5 -4 -3 εf -2 -1 0 pairing Figure 3: Ground-state phase diagram in the plane of Ufc and εf for t = 1, V = 0.1, U = 100 at n = 7/8. First-order valence transition line (solid line with solid diamonds) with quantum critical point (open circle) is determined by the DMRG. Dashed line with open squares represents crossover determined by the point where χf shows the maximum as a function of εf for several Ufc . First-order valence transition line (solid line with solid triangles) with QCP (open diamond) determined by the slaveboson mean-field theory, is also illustrated for comparison. We have calculated the superconducting correab (i)P cd † (i + x) with lation functions Pm± m± 1 ab (i) = √ (ai↑ bi+m↓ ± ai↓ bi+m↑ ) Pm± 2 (2) for the singlet(-) and triplet(+) pairing. Here, a, b, c and d denote the f and/or conduction electron operators. We have calculated the nearest-neighbor (NN) pairing (m = 1) and also the onsite pairing (m = 0). Figure 4 shows the superconducting correlations for Ufc = 5.0 where the sharp valence crossover occurs (see Fig. 2(c)) for (a)εf = −3.71, (b)−3.9 and (c)−4.2. To avoid the effect of the open boundary, we here fix the position of the operator at the central site, i.e., i = N/2 for the onsite pairing and i = N/2 − 1 for the NN pairing. For εf = −3.71 at which the sharp valence crossover takes place, we see that the pairing correlations decay rapidly as ∼ x−5 in 4 |<Pabm-(i) Pcdm-(i+x)+>| |<Pabm-(i) Pcdm-(i+x)+>| |<Pabm-(i) Pcdm-(i+x)+>| 10-2 10-3 10-4 10-5 cccc ffff 10-6 ffcf -7 10 cfcf ffcc 10-8 cfcc -9 10 0 10-210 10-3 10-4 10-5 10-6 10-7 -2 10-8 x -9 10 0 10-210 10-3 10-4 10-5 10-6 10-7 10-8 (c) 10-9 0 10 Fig. 4(a). A remarkable result in Fig. 4(b) is that the power of the decay of the correlations for εf = −3.9 is slightly smaller than 2 as clearly seen, for example, in the data of cf P cc † | and |P cf P cf † | with dashed lines |P0− 1− 0− 0− indicating x−2 . We have confirmed that all ab (i)P cd † (i + x)| show the correlations of |Pm− m− nearly the same power which is smaller than 2. On the other hand, for εf = −4.2, i.e., in the Kondo regime, it is found that the pairing correlations decay faster than the dashed lines with ∼ x−2 in Fig. 4(c). We confirmed that these results are not changed in the calculations for all the other system sizes we performed. These results indicate that the superconducting correlation is enhanced near the sharp valence increase. We also note here the general remarks on the pairing correlations: The NN singletpairing correlation for the conduction electrons cc P cc † | has the largest amplitudes and the |P1− 1− second largest is the correlation between onsite f-c singlet pair and the NN c-c singlet paircf P cc † | as seen in Fig. 4. When the ing |P0− 1− sharp valence-crossover point is approached, cf P cf † | correlation is eni.e., εf → −3.71, |P0− 0− hanced prominently as seen in Fig. 4(a). This indicates that valence fluctuations affects this pairing correlation remarkably. We have confirmed that the valence transition takes place by keeping the same Fermi volume in the MF theory [12]. In the DMRG calculation, the large Fermi surface is signaled by the peak structure of the Friedel oscillation in the both sides of the FOVT [12]. We have also confirmed that the charge gap and the spin gap close in the Kondo, the MV and the crossover regimes (see Fig. 6(a)). These results suggest that the system is described by the single-band Tomonaga-Luttinger (TL) liquid, which is adiabatically connected from the U = Ufc = 0 state where the f level and the conduction band are hybridized and the chemical potential is located at the lower hybridized band. In the vicinity of the QCP, it is nontrivial whether this assumption holds, or not. To clarify this point, we have tried to obtain the central charge c by calculating the system-size dependence of the ground-state energy. However, it is difficult to estimate the -5 x (a) 101 (b) 101 -2 x x 101 Figure 4: Singlet pairing correlation funcab (i)P cd † (i + x)| with P ab (i) = tions |Pm− m− m− √ (ai↑ bi+m↓ − ai↓ bi+m↑ )/ 2 (see text) for (a)εf = −3.71, (b)−3.90 and (c)−4.20 at Ufc = 5.0 with t = 1, V = 0.1, U = 100 and n = cc P cc † | (open circle), |P ff P ff † | 7/8: |P1− 1− 1− 1− cf P cc † | (shaded diamond), (solid circle), |P0− 1− † ff P cc † | (shaded triangle), |P cf P cf | |P1− 1− 0− 0− † ff P cf | (shaded (shaded triangle) and |P1− 0− square). Here, i = N/2 is set for m = 0 and i = N/2 − 1 for m = 1. The dashed line ∼ x−2 is drawn for comparison. 5 value of c with sufficient accuracy under the open-boundary condition. Hence, we here perform the analysis assuming that the system is described by the single-component TL liquid. Actually, under this assumption the consistent values of Kρ by the independent evaluations of the spin, charge and superconducting correlation functions are obtained at least in the regime not very close to the QCP, which will be shown below. For the single component TL liquid without the spin and charge gaps, the singlet superconducting correlation decays as x−1−1/Kρ [13]. Then, each paring correlaab (i)P cd † (j)| should decay with the tion |Pm− m− same power with −1 − 1/Kρ in the distance large enough, which is actually seen in Fig.4. Since the charge and spin correlations decay as x−1−Kρ [13], the above results shown in Fig. 4(b) indicate Kρ > 1, namely, that the superconducting correlation becomes dominant in the Kondo regime just near the sharp valence crossover. 6 the maximum (minimum) value. The Kρ estimated in this manner is summarized in Fig. 5. We see that the error bars obtained from independent correlation functions are rather small. This seems to imply that the correct power-law decay of the TL liquid may be captured. Here, it should be noted that at Ufc = 0.0 for εf ≥ −1.0, Kρ is plotted by using the result from the SS correlation, since the SS, CDW and SDW correlations do not give the consistent Kρ in the TL form, which has been also reported in the 2-leg ladder system [14]. For εf ≥ −3.75 at Ufc = 5.0 and for εf ≥ −5.60 at Ufc = 7.0, Kρ is estimated from 4kF CDW and SS correlations. For Ufc = 5.0, we see that Kρ exceeds 1 around εf = −3.9 in Fig. 5(b), indicating that the superconducting correlation is dominant, which is consistent with the result presented in Fig. 4(b). For Ufc = 7.0, Kρ jumps at the firstorder transition point as in Fig. 5(a). For Ufc = 0, Kρ remains smaller than 1, indicating that the paring correlation does not develop. These results indicate that the superconducting correlation is enhanced in the Kondo regime near sharp valence crossover close to the QCP of the valence transition. This behavior seems to be consistent with CeCu2 Ge2 [1], CeCu2 Si2 [4] and CeCu2 (Si1−x Gex )2 [5]. Evaluation of Kρ by correlation functions In order to estimate the TL parameter Kρ systematically, we have made a least-square fit for the charge, spin and superconducting correlations defined by Ōi Ōj† for Ōi ≡ Oi − Oi and cc (i), respectively, since Oi = nci , Sicz and Pm− the conduction electrons have a large amplitude in the pairing correlation as seen in Fig. 4, which seem to give a reliable estimate. By assuming the asymptotic forms of the correlation functions of the single-component TL liquid [13], we have performed the least-square † with fixing i at fit to the data for Ōi Ōi+x the central site and the i + x sites, whose values are common in several system sizes. In this way, we eliminate the effect from the open boundary and obtain reliable TL parameters. We have obtained three Kρ ’s from the singletsuperconducting (SS), CDW and SDW correlation functions. Then we employ the middle point between their maximum and minimum values as a mean value, with the error defined by the difference between the middle point and 7 Evaluation of Kρ by charge compressibility and charge velocity To get further insight into the non-monotonic behavior of Kρ around εf = −3.9 for Ufc = 5.0 in Fig. 5(b), we have calculated the charge compressibility κ ≡ ∂(2n)/∂µ, since the TL parameter is expressed by the charge velocity vc and κ as Kρ = πvc κ/2. The charge compressibility is expressed by κ = 2/(N∆c ) with the charge gap ∆c ≡ [E(2nN + 2, 0) +E(2nN − 2, 0) − 2E(2nN, 0)] /2, (3) where E(Ne , S) is the ground-state energy of the subspace specified by the electron number Ne and total spin S. Figure 6 shows the system-size dependence of (a)∆c and (b)κ for 6 0.4 ∆c 0.3 0.2 0.1 εf -4.20 -4.00 -3.90 -3.85 -3.80 -3.71 -3.65 0 0 1 1 (a) 0.1 0.2 π/(N+1) 1.5 Kρ nf 0.5 (a) 0 -6.5 -6 -5.5 -5 1 κ 0.5 0 1 (b) 0.5 0 1 0.02 0.04 1/N 0.06 Kρ nf 0.5 0 0.5 (b) -4.5 -4 Figure 6: System-size dependence of (a) charge gap ∆c and (b) charge compressibility for t = 1, V = 0.1, U = 100 at n = 7/8. 0 -3 -3.5 1 nf Kρ 1 0.5 0 0.5 (c) -1 -0.5 εf 0 0 Figure 5: nf vs. εf (open square) and Kρ vs. εf (solid diamond) for t = 1, V = 0.1 and U = 100 at n = 7/8 for (a)Ufc = 7.0, (b)5.0 and (c)0.0. various εf at Ufc = 5.0. We see that the metallic phase is realized in the sharp valence crossover regime in Fig. 6(a). We extrapolate κ to the N → ∞ limit assuming the linear dependence on 1/N . The resultant κ in the bulk limit is shown in Fig. 7. We see that κ increases as εf increases, especially around εf = −3.71. This implies that the total charge fluctuation becomes large when the f electron is moved into the conduction band and this can be understood if we recall that the compressibility is expressed by the total charge correlations in the form of the thermal average: κ = i,j (nfi + nci )(nfj + ncj )/(NT ). A remarkable result here is that the compressibility stays constant around εf = −3.9 where Kρ estimated from the least-square fit of the correlation functions has the maximum. This indicates that the enhancement of Kρ is caused by the enhancement of the charge velocity, but not by the compressibility. In order to get further insight into this, we have calcu- 7 1.0 K* ρ κ 1.5 κ , v*c nf , Kρ , Κ∗ρ 2.0 v*c nf Kρ 1.0 0.5 0.5 0 -4.2 -4 -3.8 εf -3.6 0 -3.4 Figure 7: Compressibility (solid triangle) obtained by κ = limN →∞ 2/(N∆c ) and charge velocity (solid circle) obtained by vc∗ = limN →∞ ∆c /∆q for t = 1, V = 0.1, U = 100 and Ufc = 5.0 at n = 7/8. Kρ (shaded diamond) obtained by the least-square fit of correlation functions, Kρ∗ = πvc∗ κ/2 (open diamond) and nf (open square) are shown. lated the charge velocity defined by vc∗ ≡ lim N →∞ ∆c ∆q (4) with ∆q = π/(N + 1) instead of vc , since the numerical accuracy is not enough for the second-lowest eigenvalue in the Lanczos diagonalization [17]. We see that vc∗ (filled circles) takes maximum around εf = −3.9 and this suggests that the enhancement of vc is the origin of the enhancement of Kρ . We also note that Kρ∗ = πκvc∗ /2 (open diamonds) is consistent with Kρ (shaded diamonds) obtained by the least-square fit to the correlation functions for εf ≤ −3.85 within the error bars. Since Kρ∗ is obtained by vc∗ and κ after the extrapolation to N → ∞, this suggests that the estimate of Kρ made by the finite sizes taken in this work gives reliable values for the thermodynamic limit. A more accurate determination of Kρ is left for future study. The increase in vc∗ toward εf − 3.9 in the Kondo regime arises from the decrease of the density of states at the Fermi level, i.e., the broadening of the Kondo peak. As εf further increases, vc∗ is suppressed, since Ufc interrupts the coherent motion of the electrons in the MV state. This is the reason why vc∗ has a max- imum near the sharp valence increase. This suppression of vc∗ and coherence with increasing εf in the MV regime is not captured in slave-boson MF theory [12]. The enhanced vc∗ implies the enlargement of the effective band width. Therefore the coherent motion of electrons together with enhanced valence fluctuation is interpreted as the origin of the enhanced pairing correlation. The enhancement of Kρ around εf = −3.9 agrees with the following naive expectation: In the deep Kondo regime, the SDW correlation is dominant by the RKKY interaction and the sharp valence crossover point where CDW correlation becomes dominant with enhanced χf . Between the two regimes, the superconducting correlation can be dominant, since the regime midway between Kondo and sharp valence crossover is unfavorable for both the SDW and CDW correlations. Here it should be stressed that the development of the pairing correlation is caused by the enhanced charge velocity, but not by the compressibility. This fact is in striking contrast to the ordinary models which have shown Kρ > 1 such as the 1D t-J model [15] and 1D t-V model at n = 1/4 [16]. In these models the Kρ > 1 regime appears just next to the phase-separated regime in the ground-state phase diagram. Then, enhanced Kρ is caused by the enhanced κ reflecting the instability toward the phase separation. In contrast, in the present system, the enhancement of Kρ is caused by the enhanced charge velocity, i.e., the increase of the effective bandwidth of electrons without diverging κ. 8 Summary We have reported our recent studies on the nature of the QCP of the valence transition. The DMRG calculations for the ground state show that the first-order valence transition emerges with the quantum critical point with diverging valence susceptibility. Instead of the phase separation in the mean-field result, quantum fluctuations generate a wide region of crossover between the Kondo and mixed valence states. It is found that the superconducting correlation is developed in the Kondo regime near 8 [15] M. Ogata, et al.: Phys. Rev. Lett. 66 (1991) 2388. the quantum critical point of the valence transition. The origin is ascribed to the enhanced coherent motion of electrons with valence fluctuation. This report is based on the recent works done with M. Imada and K. Miyake. [16] K. Penc and F. Mila: Phys. Rev. B 49 (1994) 9670. [17] When the charge velocity vc is defined by vc ≡ lim∆q→0 ∆∗c /∆q for ∆∗c ≡ E1 (2nN, 0)−E(2nN, 0) with E1 being the first-excited state, Kρ = πvc κ/2 holds. If the symmetric condition εf + U/2 = 0 and Ufc = 0 are satisfied in eq. (1), ∆c = ∆∗c ; i.e., vc = vc∗ holds by charge SU(2) symmetry. It is expected that vc ∼ vc∗ holds for eq. (1) in the Kondo regime. [D class; 20000K (A), 16000K (B)] References [1] D. Jaccard, et al.: Physica B 259-261 (1999) 1. [2] K. Miyake, et al.: Solid State Commun. 71 (1989) 1149. [3] T. M. Rice and K. Ueda: Phys. Rev. B 34 (1986) 6420. [4] A. T. Holmes, et al.: Phys. Rev. B 69 (2004) 024508. [5] H. Q. Yuan, et al.: Science 302 (2003) 2104. [6] S. Kawasaki, et al.: Phys. Rev. Lett. 94 (2005) 037007. [7] D. C. Koskenmaki and K. A. Gschneidner, “Handbook on the Physics and Chemistry of Rare Earths”, edited by K. A. Gschneidner and L. Eyring (North-Holland, Amsterdam, 1978) Vol. 1, Chap. 4. [8] Y. Onishi and K. Miyake: J. Phys. Soc. Jpn. 69 (2000) 3955. [9] P. Monthoux and G. G. Lonzarich: Phys. Rev. B 69 (2004) 064517. [10] S. R. White: Phys. Rev. Lett. 69 (1992) 2863, S. R. White: Phys. Rev. B 48, (1993) 10345. [11] S. Watanabe, M. Imada and K. Miyake: J. Phys. Soc. Jpn. 75 (2006) 043710. [12] S. Watanabe, M. Imada and K. Miyake: in preparation. [13] H. Schulz: Int. J. Mod. Phys. B 5 (1991) 57. [14] R. M. Noack, et al.: cond-mat/9601047. 9