prev

next

out of 9

Published on

07-Apr-2017View

214Download

0

Transcript

Heavy-fermion formation in USn3: Static and dynamical propertiesS. Kambe,1 H. Sakai,1 Y. Tokunaga,1 T. D. Matsuda,1 Y. Haga,1 H. Chudo,1 and R. E. Walstedt21Advanced Science Research Center, Japan Atomic Energy Agency, Tokai-mura, Ibaraki 319-1195, Japan2Physics Department, University of Michigan, Ann Arbor, Michigan 48109, USAReceived 13 December 2007; revised manuscript received 14 February 2008; published 7 April 2008USn3 is a heavy-fermion system with an electronic specific heat coefficient 170 mJ /K2 mol. In order tofurther characterize the heavy-fermion phenomena for USn3, the Knight shift and spin-lattice relaxation timeT1 of119Sn NMR have been measured. The static specific heat and dynamical T1 properties in the heavy-fermion state can be described in a quantitatively consistent way in terms of a spin-fluctuation model with twoconstant energy scales. However, it is necessary to introduce a T-dependent effective RudermanKittelKasuyaYosida interaction JQ JQa+bT in order to describe the crossover from an incoherent, localizedstate to a coherent, heavy-fermion state. In addition, a universal scaling behavior is proposed for the crossoverregime. The parameters obtained are used to predict the T dependence of the thermal expansion coefficient.DOI: 10.1103/PhysRevB.77.134418 PACS numbers: 76.60.k, 75.30.MbI. INTRODUCTIONIn f-electron itinerant systems, heavy-fermion HF stateshave been observed at low temperatures in certaincompounds.1 In such HF systems, the f-electron momentsare localized above the effective Fermi temperature T*. AtTT*, a crossover from the incoherent localized state to thecoherent HF state takes place, driven by the Kondo interac-tion between the localized f moments and the conductionelectrons, which usually predominates over the RudermanKittelKasuyaYosida RKKY interaction between thef-electron moments. In the HF state, the static magnetic sus-ceptibility 0,0 and the Sommerfeld electronic specificheat coefficient =Cel /T become quite large compared withordinary metals.The T dependence of the dynamical susceptibilityIm q , also clearly reflects the formation of a HF state.For example, in CeRu2Si2,2 the spin-lattice relaxation rate1 /T1T Im q , increases with decreasing T and becomesconstant below TT*. In ordinary metals, 1 /T1T constantbehavior is observed for a wide temperature range Korringabehavior. In contrast, in HF systems, such behavior is ob-served when an HF state i.e., Fermi liquid is formed belowT*. At temperatures well above T*, T1 becomes constant,which is characteristic behavior for systems with localizedmoments.Up to now, the static uniform susceptibility, specific heat,etc. and dynamical spin-lattice relaxation time, inelasticneutron scattering, etc. properties have been discussed sepa-rately in HF systems. In this study, we have tried to repro-duce the T dependence of the specific heat and spin-latticerelaxation time of the HF system USn3 in a quantitativelyconsistent way using the spin-fluctuation model for itinerantmagnets by Moriya.3 This approach gives a good account ofboth quantities for temperatures below T*. However, T de-pendence of the effective RKKY interaction has to be in-troduced in order to describe the crossover regime.In this paper, data for the T dependences of the Knightshift and spin-lattice relaxation time at the Sn site in USn3are presented. Among UX3 compounds with the same AuCu3fcc structure, where X is a IVB element X :Si ,Ge,Sn,Pb,USn3 shows the largest4,5 170 mJ /K2 mol and 0,0=9.0103 emu /mol at T0 K. These parameters give aWilson ratio of 2, indicating that USn3 is a typical heavy-fermion compound.II. EXPERIMENTA powder USn3 sample was prepared for NMR measure-ments by crushing a well-characterized single crystal.6 Theresistivity of the sample showed good metallic behavior witha residual resistivity of 1.7 cm. 119Sn I=1 /2 NMRmeasurements were performed using a conventional pulsedspectrometer with a 12 T superconducting magnet. The /2 pulse sequence has been used to excite nuclear spin-echo signals. Field-sweep NMR spectra were taken at a fre-quency of 119 MHz using digital averaging of the nuclearspin-echo signals. Spin-lattice relaxation time T1 data werealso obtained with this method.III. EXPERIMENTAL RESULTSA. Static susceptibilityFigure 1 shows the cubic AuCu3-type crystal structureof USn3. The U site has cubic local symmetry, whereas forthe Sn site, it is tetragonal. The local principal axis for the SnnFIG. 1. Color online Crystal fcc structure of USn3. n indi-cates the local symmetry axis of the Sn site.PHYSICAL REVIEW B 77, 134418 20081098-0121/2008/7713/1344189 2008 The American Physical Society134418-1http://dx.doi.org/10.1103/PhysRevB.77.134418site indicated as n in Fig. 1 is important for analyzing theNMR results.Figure 2 shows the T dependence of the static susceptibil-ity 0,0 of the NMR sample. At high temperatures, 0,0shows CurieWeiss CW behavior see inset. Below 30 K,0,0 starts to saturate and finally becomes independent ofT below 6 K. The latter behavior signals the formation of aHF state. Thus, we consider that T*30 K in USn3. Thedata for 0,0 have been least-squares fitted to the CWfunction 0,0=const0,0+ef f23kBTabove 50 K, yieldingconst0,0=3.0105 emu /mol, an effective moment ef f=2.44B, and a Weiss temperature =58 K. The constantterm const0,0 usually represents the Van Vleck orbital anddiamagnetic susceptibilities. The value obtained forconst0,0 is quite small compared with 0,0, indicatingthat the dominant term in USn3 is the spin susceptibilityspin0,0. In the inset to Fig. 2, the T dependence of1 / 0,0const0,0 is presented. The straight line ob-tained above 50 K confirms an ideal CW behavior at hightemperatures. The effective moment eff=2.44B is rathersmaller than the U3+ ionic value 3.87B, indicating that the5f electrons have an itinerant nature below 300 K.In band calculations, the noninteractive Pauli spin suscep-tibility is estimated as 6.6104 emu /mol from the densityof states at the Fermi level,7 yielding a 14 enhancement of0,0 in the HF state.B. NMR spectrum and Knight shiftFigure 3 shows a field sweep spectrum at an NMR fre-quency of 119 MHz for the 119Sn at 1.6 K. This is a typicalpowder-pattern spectrum for I=1 /2 in a site of tetragonaluniaxial symmetry. The Knight shifts K for applied mag-netic field H n and K for H to the principal axis n of theSn site are determined based on fitting procedures usingan axially symmetric powder pattern with Gaussianbroadening.8 From K and K, the isotropic and anisotropiccomponents, KisoK +2K /3 and KaniK K /3,have been estimated.Figure 4 shows the T dependence of Kiso and Kani. SinceKisoKani, the overall shift is basically isotropic in this com-pound. Kiso and Kani are plotted vs 0,0 the so-called K-plots in Fig. 5, showing good linearity in both cases. Hyper-fine coupling constants A0iso=68 kOe /B for Kiso andA0ani=8.8 kOe /B for Kani are determined from the slopesof linear fits to these data Table I. Since hyperfine cou-plings of this magnitude cannot be explained by classicaldipolar-dipolar interactions, transferred hyperfine fields dueto hybridization between U 5f and Sn 5s, p orbitals are con-sidered to be the underlying mechanism. Since Kiso is mainlydriven by spin polarization transferred to the Sn 5s orbital,the point on the plot where Kiso=0 corresponds to where thestatic spin susceptibility spin0,0=0. As shown in Fig. 5,the extrapolation of the Kiso-0,0 plot very nearly inter-12x10-31086420(0,0)3002001000T (K)USn350040030020010001/((0,0) const(0,0))3002001000T (K)FIG. 2. Color online T dependence of the static susceptibility0,0. In the inset, a CurieWeiss plot of 1 / 0,0const0,0vs T is presented.FIG. 3. Color online 119Sn field sweep NMR spectrum at119 MHz in a powder sample of USn3 T=1.6 K. The edge posi-tions indicated by arrows correspond to shift parameters K H nand K Hn. Solid line is a fitted curve based on the axiallysymmetric powder pattern with Gaussian broadening.FIG. 4. Color online T dependence of the isotropic Kisoand anisotropic Kani Knight shifts KisoK +2K /3 and KaniK K /3.(0,0)FIG. 5. Color online Knight shift versus static susceptibility0,0 plot K- plot. The solid lines were obtained with least-squares fits. The slopes of the lines correspond to the hyperfinecoupling constants A0iso and A0ani.KAMBE et al. PHYSICAL REVIEW B 77, 134418 2008134418-2sects the origin, indicating that 0,0spin0,0 in USn3.This is consistent with the small nonspin susceptibility constwhich was estimated from the static susceptibility measure-ments. The origin of Kani is apparently hybridization with theSn 5p orbital, which gives an anisotropic, dipolar hyperfinefield. The large Kiso compared with Kani indicates that thehybridization between the U 5f and Sn 5s orbitals is compa-rable with that between the U 5f and Sn 5p orbitals since the5s hyperfine field is expected to be much larger than that ofthe 5p electrons.Since the K- plot is linear down to 1.6 K in the HF state,the hyperfine coupling constant at the Sn site is not modifiedby the HF effects in this compound. In some HF compoundsthe K- plot becomes nonlinear due to T-varying hyperfineinteraction contributions.9 In contrast, in USn3 there is nosuch complication from T-dependent couplings; thus, a quan-titative analysis is possible in detail as described below.C. Spin-lattice relaxation time T1Figure 6 shows the T dependence of 1 /T1 at the 119Snsite for Hn measured at the K position in Fig. 3.At high temperatures, the T dependence of T1 becomesweak. This regime corresponds to a crossover from the HF toa localized momentlike state. In the localized moment state,T1 is independent of T and is expressed as the exchange-narrowing limit case,101/T1ex = 2A0/2JJ + 13nex. 1Here, the effective total angular momentum J and ex-change frequency ex are estimated to be 0.82 and3.11012 s1, respectively, from the CW parametersef f =2.44B and =58 K, and n is the number of neigh-boring magnetic sites around Sn, i.e., n=4 in the presentcase. These parameters lead to 1 /T1ex1.9105 s1, whichis much larger than the observed value at 240 K. Even athigh temperatures, the 5f electrons still have some itinerantcharacter in USn3.At low temperatures, 1 /T1 is proportional to T below10 K, i.e., Korringa behavior appears in the HF state. Foritinerant systems with magnetic exchange enhancements,the usual Korringa relation between T1, K and Se /N2h /4kB: T1TK2=S is modified with K term,leading to an extended Korringa relation for ligand sites,11T1TK2 = nSK1. 2The estimated K0.6 is smaller than 1 in USn3, suggest-ing that ferromagnetic exchange enhancement seems to bepresent in this compound. However, an alternative explana-tion is the cancellation of antiferromagnetic fluctuations atthe 119Sn, as discussed below. The present case indicates thatwe should be careful in applying the extended Korringa re-lation to ligand-site NMR results when the hyperfine formfactor is a critical element. In fact, in the paramagnetic stateof the isostructural antiferromagnet UIn3 TN=88 K, K isfound to be less than 1 owing to a similar cancellation.12In order to estimate the anisotropy of hyperfine fluctua-tions, 1 /T1 for H n has also been measured at 20 and120 K not shown. Generally, 1 /T1, at tetragonal sites isexpressed as1/T1 =2N2 kBTB2 qAq2 Im q,nn2N2 kBB2 qfq2A02 Im q,nn,1/T1 =N2 kBTB2 qAq2 Im q,nn+ Aq2Im q,nnN2 kBB2 q fq2A02 Im q,nn+ fq2A02 Im q,nn , 3where N is the nuclear gyromagnetic ratio, Aq is the hy-perfine coupling constant, fq is the hyperfine form factorwhich reflects the local symmetry of Sn site, Im q ,n isthe dynamical susceptibility, and n=119 MHz is the NMRfrequency. The off-diagonal hyperfine coupling13 is ignoredhere since the hyperfine coupling is very nearly isotropic atthe Sn site. Thus, the hyperfine form factor fq is the samefor H n and Hn .Based on Eq. 3,TABLE I. Transferred hyperfine coupling constants A0 inkOe /B for the 119Sn in USn3 obtained from K- plots.A0 A0 A0iso A0ani862 592 682 8.80.5FIG. 6. Color online T dependence of 1 /T1 at the 119Sn site forHn .HEAVY-FERMION FORMATION IN USn3: STATIC PHYSICAL REVIEW B 77, 134418 2008134418-321/T1 1/T11/T1A02qfq2 Im q,nA02 qfq2 Im q,n. 4The experimental value of21/T11/T11/T120.2 at 20 and120 K then arises from the anisotropy of the transferred hy-perfine coupling constant A02 /A02 2.1 see Table I.From Eq. 4, this fact indicates that the dynamicalsusceptibility Im q , is isotropic i.e., Imq ,n=Im q ,n, in agreement with the local cubic symmetryof the U site, which makes the main contribution to the mag-netic susceptibilities i.e., 0,0 and Im q ,.IV. ANALYSIS AND DISCUSSION BASED ON THE SCRMODEL OF DYNAMIC MAGNETISMA. Analysis based on the usual self-consistentrenormalization modelIn this section, the electronic specific heat4 and the presentspin-lattice relaxation results are interpreted in a quantita-tively consistent way based on the framework of the self-consistent renormalization SCR model. Originally, the SCRmodel was developed by Moriya in order to interpret weakmagnetism in itinerant systems.3 More recently, this modelwas adapted to describe the HF state.14In the SCR model, the dynamical susceptibility is charac-terized by two energy scales, T0 and TA, which correspond tomagnetic fluctuation energy in and q spaces, respectively.The q dependence of the RKKY interaction JQ is expressedas JQJQ+q=2TAq / qB 2 around the antiferromagneticwave vector Q, where qB is the zone-boundary vector. Thus,the dispersion of the RKKY interaction can be defined asJQJQJ0=2TA. Since this JQ includes all q-dependenteffects such as s f mixing in addition to the original RKKYinteraction,14 JQ represents effective RKKY interaction. Usu-ally, T0 has a magnitude of T* and is connected with the localmagnetic susceptibility. The framework of the SCR modelfor application to HF systems has been described in previousworks.15,16Since no magnetic phase transition takes place in USn3,the nature of the magnetic correlations must be decided insome other way. In the present analysis, correlations havebeen presumed to be antiferromagnetic see below.For the case of antiferromagnetic correlations, the generalsusceptibility Q+q , has been formulated to be141Q + q,=1Q,0+ Bq2 iLL= 2TAyT + qqB2 i2T0 ,yT 12TAQ,0,TA BqB22,T0 TALL=TAqqQ q 0 ,JQ =L QLL, 5where B is the dispersion constant, L and q are the localand q-dependent susceptibilities, and L and q are the localand q-dependent relaxation rates, respectively. The T depen-dence of yT is determined based on the fluctuation-dissipation theorem in order to guarantee internal consis-tency on condition that the total spin fluctuation S2 isconstant.3In addition, two dimensionless parameters y0 and y1 areintroduced to characterize the state of the system,y0 yT = 0 K, y1 2JQ2TA=4JQ2JQ. 6Here, y0 is a measure of the deviation from the quantumcritical point, since y0=0 corresponds to a quantum phasetransition at T=0 K. y1 reflects the strength of dispersion ofthe effective RKKY exchange interaction JQ.In the usual SCR model, T0, TA, and JQ are assumed to beT independent. This can be justified below T*, and thus,interpretation of the electronic specific heat and spin-latticerelaxation rate at temperatures below 30 K are consideredfirst.Figure 7 shows the T-dependence of the electronic spe-cific heat Cel /T as estimated by Norman et al.4 It should benoted that specific heat results for our sample are in goodagreement with this previous result. Since Cel is obtained bysubtraction of the large phonon contribution from the totalspecific heat,4 there is a degree of uncertainty about the es-timated Cel above 10 K. The electronic specific heat Cel canbe expressed as a function of yT and T0.14 We have ob-tained y00.22, y11, T033 K Table II, and the T de-pendence of yT by fitting Cel /T TA cannot be determinedfrom the specific heat14. The deviation of Cel /T at high tem-peratures may be partly due to an overestimation of the pho-non contribution to the specific heat. Figure 8 shows the Tdependence of yT=1 / 2TAQ ,0. At high temperatures,yT exhibits CW behavior, i.e., Q ,01 /T. In contrast,FIG. 7. Color online Closed circles are the T dependence ofthe electronic specific heat Cel /T from Ref. 4, which has been esti-mated by subtraction of the phonon part from the total specific heat.The solid line is Cel /T calculated using the spin fluctuation param-eters in Table II and the equation for Cel /T Ref. 14.KAMBE et al. PHYSICAL REVIEW B 77, 134418 2008134418-4yT is nearly constant below 10 K inset of Fig. 8.The values obtained for T0, y0, and y1 are similar to thosefor other HF compounds with the same magnitude of i.e., for CeRu2Si2 Ref. 15: 350 mJ /K2 mol, T0=14 K,TA=16 K, y0=0.31, and y1=1.6. In the SCR model,14 therelaxation rate at q=Q is estimated as QT2T0yT.Actually, QT agrees with the T dependence of half widthof magnetic quasiparticle excitations spectra in USn3; Tdetermined in inelastic neutron scattering measurements17Fig. 9. On the other hand, the local relaxation rate is esti-mated as LT0.53T0y1+QT5.1102 K+QT5.6102 K at 0 K, which is considerably larger thanQ0 K46 K. Although such a high energy magnetic ex-citation has not been detected in the neutron scatteringmeasurements,17 this may be due to a weak and broad spec-trum for LT.It should be noted that T1 is found to be independent ofthe applied magnetic field H between 3 and 7 T. Althoughthe Zeeman splitting energy 2BH is comparable with thecharacteristic energy of the spin fluctuations T0, the spin-lattice relaxation is insensitive to H in the present measure-ments.Using the relation SelT=0TCel /TdT, the T dependence ofthe electronic entropy SelT can be estimated. For the HFstate of systems with the degeneracy of quasiparticlesN2J+1=2 i.e., J=1 /2 case, SelT becomes R ln 2 atTT*T0. Actually, SelT seems to reach R ln 2 around40 KT0 in USn3 not shown. Recently, the KadowakiWoods plot for highly degenerate HF systems N2 wasfound to be different from the usual behavior of N2systems.18 However, the point for USn3 is found on theKadowakiWoods plot for the N2 case.18 These facts in-dicate that the effect of degeneracy is not pronounced inUSn3 as well as other U-based HF compounds,18 althoughorbital degeneracy can affect physical properties because ofthe cubic symmetry. Since a N2 case is assumed in thepresent SCR model,14 the consistency of the present analysisis confirmed.From Eq. 5, the dynamical susceptibilityIm Q+q , can be expressed asIm Q + q, =12TA 2T0yT + qqB22 + 2T02.7Noting the isotropic nature of Im Q+q ,, 1 /T1Tbased on Eq. 3 then becomes1/T1T N2 kBB2 A02 + A02 qfQ + q2Im Q + q,nn. 8Figure 10 shows the q dependence of the hyperfine formfactor at the Sn site for an isotropic hyperfine field. Thehyperfine form factor is a function of the two x ,y axessince the Sn site is located in the U plane Fig. 1. As shownin Fig. 10, fQ+q2 for Q= , has a minimum at q= qx ,qy= 0,0, indicating that antiferromagnetic fluctua-tions are canceled at the Sn site.In the present analysis, an antiferromagnetic correlationQ= , is assumed in order to interpret the specific heatand T1 consistently. Figure 11 illustrates the q dependence ofIm Q+q ,n around Q= , calculated from the givenparameter values at several temperatures. As T decreases,FIG. 8. Color online T dependence of calculated yT fory00.22, y11, and T033 K. The inset shows the same plot ona semilogarithmic scale.FIG. 9. Color online T dependence of the relaxation rateQT2T0yT estimated from yT in Fig. 8. For comparison,the relaxation rate T determined in neutron inelastic scatteringmeasurements Ref. 17 is presented.q=(-,)q=(,)q=(-,-)f(Q+q)2q qx yq=(q ,q )x yFIG. 10. Color online q dependence of the hyperfine formfactor fQ+q2=sinqx /22+sinqy /22 at Sn sites in USn3 for thecase of Q= ,. The origin of the plot q= 0,0 corresponds tofQ2.HEAVY-FERMION FORMATION IN USn3: STATIC PHYSICAL REVIEW B 77, 134418 2008134418-5enhancement of Im Q+q ,n develops around q= 0,0,i.e., antiferromagnetic fluctuations develop at low tempera-tures. Although fQ+q2 has a minimum at q= 0,0, thespin-lattice relaxation at the Sn site probes the T dependenceof Im Q+q ,n because there is some overlap betweenfQ+q2 and Im Q+q ,n see Figs. 10 and 11.Based on Eq. 8 and the T dependence obtained forIm Q+q ,n from the specific heat result, the T depen-dence of 1 /T1T has been calculated by q integration. SinceA0 has already been obtained, TA is the only adjustableparameter for T1, which is estimated as 44 K. Fromthe SCR model,15 TA has a magnitude of1 / 21+y00,0T=0 K70 K, which is roughly consistentwith the present estimate.If a ferromagnetic correlation Q= 0,0 were assumed,the estimated value of 1 /T1T would become much largerthan the observed value due to the hyperfine form factor, orTA would become unnaturally large in order to reproduce theobserved T1. This indicates that antiferromagnetic correla-tions Q= , are the most likely internally consistentexplanation for the specific heat and spin-lattice relaxationrate in USn3. In fact, antiferromagnetic correlations are alsosuggested by inelastic neutron scattering measurements.17 Itshould be noted that only type II antiferromagneticQ= , , ordering, i.e., no ferromagnetic ordering, hasbeen found in UX3 compounds up to now.19Figure 12 shows the measured T dependence of 1 /T1Tcompared with calculated values. In the HF state below10 K, 1 /T1T has a weak T dependence. In the crossoverregime above 10 K, 1 /T1T decreases more rapidly withincreasing T. Here, the T dependence of 1 /T1T is wellreproduced by the calculations below 30 K, i.e., T0. Above30 K, the observed 1 /T1T decreases more rapidly thanthat predicted by the SCR model. This is not unexpectedsince the present SCR model cannot treat the crossover fromthe HF to the incoherent state. The T1 data show that withincreasing T, low energy fluctuations disappear abruptly, ow-ing to the collapse of the HF state.B. Analysis of the crossover regime with T-dependentJQ and TAWe now argue that, in the crossover regime, the charac-teristic parameters for the random phase approximationTABLE II. Obtained characteristic parameters for USn3 based on the SCR model.T0TALLTABqB22y012TAQ ,00Ky14JQ2JQJ0332 K 442 K 0.220.02 10.1q=(,)q=(-,)q=(-,-)q=(q ,q )x yqx yqFIG. 11. Color online Calculated T and q dependences ofIm Q+q ,n for the parameters given in Table II. Since Q= ,, the origin of the plot q= 0,0 corresponds to Im Q ,n.FIG. 12. Color online T dependence of 1 /T1T at 119Sn sitefor Hn . The solid line shows calculated values of 1 /T1T basedon the Eq. 8 and the parameters in Table II. The dashed line showsthe calculated values of 1 /T1T with T-dependent JQ shown inFig. 13.KAMBE et al. PHYSICAL REVIEW B 77, 134418 2008134418-6RPA-form magnetic susceptibility Eq. 5 will be modi-fied due to the formation of HF quasiparticles. However, T0is considered to be constant, while JQ and TA=2JQ / 2y1become T dependent.20 In addition, y1 is assumed to be Tindependent in the present analysis since the RKKY interac-tions have no strong T dependence in paramagnetic HF sys-tems. In order to reproduce the experimental results withreasonable modifications, one may simply introduceT-dependent JQ and TA on the condition of constant total spinfluctuations S2. The constant TA determined in the previoussection is noted as TA0 K in this section.In Fig. 12, the T dependence of 1 /T1T is shown, wherean optimized fit dotted line has been obtained with theT-dependent JQ plotted in Fig. 13. The other parameters y0,y1, and T0 have been fixed at the values presented in Table II.The experimental results for 1 /T1T can be accurately re-produced in this fashion. It should be noted that the elec-tronic specific heat Cel is unaffected by this modificationsince yT and T0 are unchanged.The T dependence of JQ is obtained by fitting the data for1 /T1T. Figure 13 shows that the JQT plot showsa very nearly straight line above 25 K, indicating thatJQTa+bT in the crossover regime. As a result of com-petition between the RKKY interaction and the Kondo effectin the crossover regime, the effective RKKY interaction de-creases with decreasing T.Such a T dependence may be connected with a similarcharacteristic T law for the line width of the quasielasticneutron scattering q observed in USn3 Ref. 17 andheavy-fermion compounds2124 since JQ= LQ / LLhere.14 Such behavior may be interpreted theoretically interms of crystal field effects in Kondo systems,25 although noclear crystal field excitation is detected in USn3.17 In anycase, the present analysis suggests that theoretical develop-ments which include a T-dependent effective RKKY interac-tion are necessary to describe the formation of the HF state.In the present analysis, TAT=2JQT / 2y1 increaseswith increasing T, while T0 remains constant. From Eq. 5,this indicates that qq=T0TATdecreases with increasing T inthe crossover regime. In order to characterize the crossoverregime to the HF state, it may be useful to define a hypo-thetical scaling function fT /T0 by means ofqq =T0TA0 Kftt T/T0. 9Since qq is expressed approximately as an -integrationof Imq , up to q: qq0qImq ,d using theKramers-Kronig relation, the scaling function ftqq isrelated with the spectral weight of Imq , for a wideenergy range.Figure 14 shows the estimated ft=TA0 K /TAT as afunction of the normalized temperature t=T /T0 in USn3. Forcomparison, ft=TA0 Kqq / T0 is also presented us-ing qq data for q=Q determined by neutron scattering26,27in CeRu2Si2 T0=14 K and TA0 K=16 K. Although a de-tailed formula of ft e.g., an exponent T0.5 for USn3 maydepend on other parameters such as y1 see below, the roughfeatures of ft may be universal for HF compounds. Theseeffects should be explained by a theory for the crossoverregime which goes beyond the SCR model. This kindof scaling function has been already proposed forCe1xLaxRu2Si2 systems.26In the present analysis, y1 is assumed to be T independent.If we assume that y1 is T dependent and TA is constant, wefind again JQT=0.52TA0y1Tc+dT in the crossoverregime. This fact suggests that the T behavior for JQ doesnot depend on T dependence of y1. In contrast, the T depen-dence of TA ft1 could be modified by a possible weak Tdependence of y1. Based on an improved model, the scalingfunction may be formulated as ft ,y1 which can character-ize the crossover regime of HF system.C. Prediction of thermal expansion coefficientFinally, we have predicted the T dependence of the elec-tronic thermal expansion coefficient = 1VdVdT based on the Tdependence obtained for y. In antiferromagnetic HF systems,the thermal expansion coefficient is one of the few quantitieswhich can detect the staggered magnetic susceptibilityQ ,0 owing to the large Grneisen parameter of HF sys-tems. For antiferromagnetic systems, may be expressedas28FIG. 13. Color online The T dependence of JQ estimated fromthe fitting of 1 /T1T Fig. 12 above 20 K using Eqs. 7 and 8.FIG. 14. Color online Hypothetical scaling functionftqq as function of t=T /T0 in USn3 T0=33 K. For compari-son, ft in CeRu2Si2 T0=14 K obtained from neutron scatteringdata Refs. 26 and 27 is presented.HEAVY-FERMION FORMATION IN USn3: STATIC PHYSICAL REVIEW B 77, 134418 2008134418-7 =32DQT0JQdydT, 10where DQ and are the magnetovolume constant at q=Qand the compressibility, respectively, which can be deter-mined experimentally. Figure 15 shows the T dependence of 1JQdydT calculated from the T dependence of y Fig. 8 forthe constant and T-dependent JQ cases.Since reflects the T dependence of JQ, as shown in Fig.15, it will be interesting to compare the calculated with anexperimental one in USn3, although no such data have beenreported up to now. For the related case of the heavy-fermionsystem Ce1xLaxRu2Si2, the experimental value28 issmaller than the calculated one for the fixed JQ case aboveT0, which can be explained by values of JQ which increasewith T.V. CONCLUSIONIn the heavy-fermion compound USn3, the hyperfine cou-pling constant at the Sn site is large and isotropic, indicatingthat the transferred hyperfine coupling is mainly due to hy-bridization between U 5f and Sn 5s orbitals. The spin-latticerelaxation rate 1 /T1T at the Sn site increases with decreasingT and finally becomes constant in the heavy-fermion state.The T dependence of 1 /T1T and the electronic specificheat Cel in the HF state can be described quantitatively interms of a spin-fluctuation model.14 However, in order todescribe the crossover from the heavy-fermion tolocalizedlike state above T*, a T-dependent effective RKKYinteraction JQa+bT and its dispersion strengthJQ=JQJ0, which were not included in the foregoingmodel, should be considered. In order to develop a realisticmodel for the formation of the heavy-fermion state, experi-mental determination of the RKKY dispersion in the para-magnetic state and a theoretical interpretation for it beyondthe usual spin-fluctuation model are suggested to be impor-tant.In the crossover regime, a universal behavior character-ized by a scaling function fT /T0 is proposed, which shouldbe deduced naturally from a realistic model.From the estimation of thermal expansion coefficientbased on the SCR model, a thermal expansion coefficientmeasurement is found to be a good test to confirm the pro-posed T dependence of JQ.It is proposed that the present quantitative analysis basedon T1 and Cel is a useful method for determining the type ofRKKY interaction antiferromagnetic in the present casewhich is present in a compound without magnetic phase tran-sition.ACKNOWLEDGMENTSWe thank K. Miyake and Y. Takahashi for useful discus-sions on the SCR model and S. Raymond for sending us dataof qq in Ce1xLaxRu2Si2 and for many important sugges-tions. We are grateful to K. Kubo, T. Hotta, K. Kaneko, N.Metoki, H. Kadowaki, H. Yasuoka, and G. H. Lander forstimulating discussions.1 J. Flouquet, in Heavy Fermion Matter, edited by W. Halperin,Progress in Low Temperature Physics Vol. XV Elsevier, Am-sterdam, 2005, p. 3.2 Y. Kitaoka, H. Arimoto, Y. Kohori, and K. Asayama, J. Phys.Soc. Jpn. 54, 3236 1985.3 T. Moriya, Spin Fluctuations in Itinerant Electron MagnetismSpringer, Berlin, 1985.4 M. R. Norman, S. D. Bader, and H. A. Kierstead, Phys. Rev. B33, 8035 1986.5 K. Sugiyama, T. Iizuka, D. Aoki, Y. Tokiwa, K. Miyake, N.Watanabe, K. Kindo, T. Inoue, E. Yamamoto, Y. Haga, and Y.nuki, J. Phys. Soc. Jpn. 71, 326 2002.6 Y. Haga, E. Yamamoto, T. Honma, N. Kimura, M. Hedo, H.Ohkuni, D. Aoki, M. Ito, and Y. nuki, JJAP Ser. 11, 2691999.7 H. Harima unpublished.8 A. Abragam, Principles of Nuclear Magnetism Oxford Univer-sity, New York, 1961.9 N. J. Curro, B.-L. Young, J. Schmalian, and D. Pines, Phys. Rev.B 70, 235117 2004.10 T. Moriya, Prog. Theor. Phys. 16, 641 1956.11 A. Narath and H. T. Weaver, Phys. Rev. 175, 373 1968.12 H. Sakai, S. Kambe, and Y. Tokunaga unpublished.13 N. J. Curro, New J. Phys. 8, 173 2006.14 T. Moriya and T. Takimoto, J. Phys. Soc. Jpn. 64, 960 1995.15 S. Kambe, J. Flouquet, and T. E. Hargreaves, J. Low Temp.Phys. 108, 383 1997.16 S. Kambe, S. Raymond, L. P. Regnault, J. Flouquet, P. Lejay,and P. Haen, J. Phys. Soc. Jpn. 65, 3294 1996.17 M. Loewenhaupt and C.-K. Loong, Phys. Rev. B 41, 92941990.18 N. Tsujii, H. Kontani, and K. Yoshimura, Phys. Rev. Lett. 94,057201 2005.19 S. Demuynck, L. Sandratskii, S. Cottenier, J. Meersschaut, andFIG. 15. Color online The calculated T dependence of theelectronic thermal expansion coefficient based on the T depen-dences of y Fig. 8 and JQ Fig. 13 for T0=33 K. The solid anddashed lines correspond to the constant JQ and T-dependent JQcases, respectively.KAMBE et al. PHYSICAL REVIEW B 77, 134418 2008134418-8M. Root, J. Phys.: Condens. Matter 12, 4629 2000.20 K. Miyake private communication.21 L. P. Regnault, J. L. Jacoud, J. M. Mignot, J. Rossat-Mignod, C.Vettier, P. Lejay, and J. Flouquet, Physica B 163, 606 1990.22 J. Rossat-Mignod, L. P. Regnault, J. L. Jacoud, C. Vettier, P.Lejay, J. Flouquet, E. Walker, D. Jaccard, and A. Amato, J.Magn. Magn. Mater. 76, 376 1988.23 N. B. Brandt and V. V. Moschalkov, Adv. Phys. 3, 373 1984.24 S. Horn, E. Holland-Moritz, M. Loewenhaupt, F. Steglich, H.Scheuer, A. Benoit, and J. Flouquet, Phys. Rev. B 23, 31711981.25 S. Maekawa, S. Takahashi, S. Kashiba, and M. Tachiki, J. Phys.Soc. Jpn. 54, 1955 1985.26 W. Knafo, S. Raymond, J. Flouquet, B. Fk, M. A. Adams, P.Haen, F. Lapierre, S. Yates, and P. Lejay, Phys. Rev. B 70,174401 2004.27 S. Raymond, W. Knafo, J. Flouquet, and P. Lejay, J. Low Temp.Phys. 147, 215 2007.28 S. Kambe, J. Flouquet, P. Lejay, P. Haen, and A. de Visser, J.Phys.: Condens. Matter 9, 4917 1997.HEAVY-FERMION FORMATION IN USn3: STATIC PHYSICAL REVIEW B 77, 134418 2008134418-9