Characterization of ferrimagnetic Heisenberg chains according to the constituent spins Shoji Yamamoto Department of Physics, Okayama University, Tsushima, Okayama 700-8530, Japan Takahiro Fukui 0 Institut fu¨r Theoretische Physik, Universit¨at zu K¨oln, Zu¨lpicher Strasse 77, 50937 K¨oln, Germany 0 0 2 Tˆoru Sakai Department of Physics, Himeji Institute of Technology, Ako, Hyogo 678-1297, Japan n (Received 30 August 1999) a J Thelow-energy structureand thethermodynamicproperties of ferrimagnetic Heisenbergchains 1 of alternating spins S and s are investigated by the use of numerical tools as well as the spin- wave theory. The elementary excitations are calculated through an efficient quantum Monte Carlo ] techniquefeaturingimaginary-timecorrelation functionsandarecharacterizedintermsofinteract- h c ing spin waves. The thermal behavior is analyzed with particular emphasis on its ferromagnetic e and antiferromagnetic dual aspect. The extensive numerical and analytic calculations lead to the m classification of the one-dimensional ferrimagnetic behavior according to the constituent spins: the - ferromagnetic (S >2s), antiferromagnetic (S <2s), and balanced (S =2s) ferrimagnetism. t a PACS numbers: 75.10.Jm, 65.50.+m, 75.40Mg, 75.30Ds t s . t a I. INTRODUCTION ferromagnetic aspect, reducing the ground-state magne- m tization,formagaplessdispersionrelation,whereasthose - of antiferromagnetic aspect, enhancing the ground-state d One-dimensional quantum ferrimagnets are one of the magnetization, are gapped from the ground state. As n hot topics and recent progress [1–15] in the theoretical a result of the low-energy structure of dual aspect, the o understanding of them deserves special mention. Such specific heat shows a Shottky-like peak in spite of the c a vigorous argument a great deal originates in the pio- [ ferromagnetic low-temperature behavior and the mag- neeringeffortstosynthesizebinuclearmagneticmaterials netic susceptibility times temperature exhibits a round 1 including one-dimensional systems. The first ferrimag- minimum [3,10,18]. When the exchange coupling turns v netic chain compound [16], MnCu(dto) (H O) 4.5H O 4 2 2 3· 2 anisotropic, the dispersion of the ferromagnetic excita- (dto = dithiooxalato = S C O ), was synthesized by 0 2 2 2 tions is no more quadratic [1] and the plateau in the GleizesandVerdaguerandstimulatedthepublicinterest 0 ground-statemagnetizationcurve[12]due tothe gapped in this potential subject. The following example of an 1 antiferromagneticexcitationsvanishesviatheKosterlitz- 0 ordered bimetallic chain [17], MnCu(pba)(H2O)3·2H2O Thouless transition [15]. 0 (pba = 1,3-propylenebis(oxamato) = C7H6N2O6), ex- The quantum behavior of the model with higher spins 0 hibitingmorepronouncedonedimensionality,furtherac- at/ tvievsattigedattiohnes.phTyhsiecraela[1ls8o,1a9p],paesarwedellaansicdheeam[2ic1a]lt[h2a0t],tihne- TishnooutghyeDtrsilolocnleeatraal.s[1th8a]tmiandethtehe(Sfi,rss)t a=tte(m1,p12t)tcoarsee-. m veal the general behavior of the model, they fixed the alternatingmagneticcentersdonotneedtobemetalions smaller spin to 1 in their argument with particular em- - but may be organic radicals. 2 d phasis on the problem of the crystal engineering of a Apracticalmodelofaferrimagneticchainistwokinds n molecule-based ferromagnet the assembly of the highly o of spins S and s (S >s) alternating on a ring with anti- − magnetic molecular entities within the crystal lattice in c ferromagnetic exchange coupling between nearest neigh- : bors, as described by the Hamiltonian, a ferromagnetic fashion. There also exists an extensive v numerical study [3], but the leading attention was not i X N necessarily directed to the consequences of the variation r =J (Sj sj +sj Sj+1) . (1) of the constituent spins. So far the generic behavior, a H · · rather than individual features, of ferrimagnetic mixed- Xj=1 spin chains has been accentuatedor predicted, but there are few attempts to characterize or classify the typical Here N is the number of unit cells and we set the unit- one-dimensional ferrimagnetic behavior as a function of cell length equal to 2a in the following. The simplest case,(S,s)=(1,1),has sofarbeen discussedfairly well. (S,s). In such circumstances, we aim in this article at 2 elucidating which property of the model (1) is universal Thereliebothferromagneticandantiferromagneticlong- and which one, if any, is variable with (S,s). rangeordersinthegroundstate[2–4]. Thegroundstate, which is a multiplet of spin (S s)N, shows elementary It is true that numerical tools are quite useful in this − excitations of two distinct types [8]. The excitations of context, but they are not almighty. Although the low- 1 temperature ferromagnetic behavior is quite interesting S(q,τ) 1;k Sz l;k +q 2e−τ[El(k0+q)−E1(k0)], ≃ h 0| q| 0 i from the experimental point of view, it is hardly feasi- Xl (cid:12) (cid:12) (cid:12) (cid:12) ble to numerically take grand-canonical averages at low (3) enough temperatures. It is also unfortunate that with the increase of (S,s), the available information is more at a sufficiently low temperature, where k is the mo- and more reduced in both quality and quantity. Then 0 mentum at which the lowest-energy state in the sub- we have an idea of describing the model in terms of the space is located. Therefore we can reasonably approx- spin-wavetheory. The conventionalspin-wavetreatment imate E (k +q) E (k ) by the slope ∂ln[S(q,τ)]/∂τ oflow-dimensionalmagnetsmaydiscourageus. TheHal- 1 0 − 1 0 − in the large-τ region. When we take interest in the dane conjecture [22] revealeda qualitative breakdownof lower edge of the excitation spectrum, such a treatment the spin-wave description of one-dimensionalHeisenberg is rather straightforward. The elementary excitations of antiferromagnets. Neither quantum corrections [23] nor the Haldane antiferromagnets were indeed revealed thus additional constraints [24] to spin fluctuations end up [26]. Here, due to the two kinds of spins in a chain and with an overall scenario for the low-energy physics ap- theresultantdualaspectofthelow-energystructure,the plicable to general spins. However, Ivanov [7] has re- relevant subspace and operator O are not uniquely de- cently reported that the spin-wave description of one- j fined. Since the total magnetization, M = (Sz+sz), dimensional Heisenberg ferrimagnets is quite successful. j j j Though his calculations were restricted to the ground- is a conserved quantity in the present sysPtem, we con- sider calculating S(q,τ) independently in each subspace state energy and magnetization, the highly accurate es- with a given M [27]. The elementary excitation energies timates obtained there are surprising enough, consid- for the ferromagnetic branch are obtained from a single ering the diverging ground-state magnetization in the calculation,S(q,τ)inthesubspaceofM =0,whilethose one-dimensionalantiferromagneticspin-wavetheory. For for the antiferromagnetic branch from a couple of calcu- antiferromagnets, quantum fluctuations of domain-wall lations, S(q,τ) and the lowest energy in the subspace of type, connecting the two degenerate N´eel states, are im- M = (S s)N +1. We have taken Sz sz for O and portant, whereas for ferrimagnets, domain-wall excita- found tha−t O = Sz sz extracts thejei±genjvalues ojf the tions lead to magnetization fluctuations and are thus of j j − j bonding (lower-energy) states in both subspaces. The less significance. Thereforeitis likelythat the spin-wave choice of the scattering matrices is a profound problem approach is highly efficient for ferrimagnets. In an at- in itself and is fully discussed elsewhere [8,28]. tempt to demonstrate such an idea, we try to describe Althoughthechainlengthwecanreachwiththeexact- not only the low-energy structure but also the thermal diagonalization method is strongly limited, the diago- behavior of the Heisenberg ferrimagnetic spin chains (1) nalization results are still helpful in the present system withinthe frameworkofthe spin-wavetheory. Extensive whose correlation length is generally so small as to be numerical calculations, supplemented by the spin-wave comparable to the unit-cell length (see Fig. 5 bellow). analysis, fully reveal the one-dimensional ferrimagnetic Actually, the ground-state energies for N = 10 coincide behavior as a function of the constituent spins. with their thermodynamic-limit values within the first severaldigits. TheLanczosalgorithmgivesthemostpre- II. ELEMENTARY EXCITATIONS ciseestimatefortheground-stateenergyandtheantifer- romagnetic excitation gap, whereas the quantum Monte Carlotechniqueisnecessaryfortheevaluationofthecur- Inordertoinvestigatethelow-energystructure,weem- vature of the small-momentum dispersion. ployanewquantumMonteCarlotechnique[25]aswellas On the other hand, we consider a spin-wave descrip- theconventionalLanczosdiagonalizationalgorithm. The tion of the elementary excitations as well. We introduce idea is in a word expressed as extracting the low-lying the bosonic operators for the spin deviation in eachsub- excitations from imaginary-time quantum Monte Carlo lattice via data at a low enough temperature. The imaginary-time correlation function S(q,τ) is generally defined as S+ = 2S a†a a , Sz =S a†a , S(q,τ)= eHτO e−HτO , (2) j q − j j j j − j j (4) q −q s+ =b† 2s b†b , sz = s+b†b , (cid:10) (cid:11) j jq − j j j − j j where O = N−1 N O e2iaqj is the Fourier trans- q j=1 j form of an arbitrarPy local operator Oj, which may be whereweregardS ands asquantitiesofthe sameorder. an effective combination of the spins S and s, and the TheHamiltonian(1)isexpressedintermsofthe bosonic thermal average at a given temperature β−1 = k T, operators as B A Tr[e−βHA]/Tr[e−βH], is taken in a certain sub- hspaice≡. S(q,τ) can be represented in terms of the eigen- =Eclass+ 0+ 1+O(S−1), (5) H H H vectors and eigenvalues of the Hamiltonian, l;k (l = | i 1,2, ) and E (k) (E (k) E (k) ), and behaves where E = 2sSJN is the classicalground-stateen- l 1 2 class ··· ≤ ≤ ··· − like ergy, and and are the one-body and two-body 0 1 H H 2 terms of the order O(S1) and O(S0), respectively. We is the O(S1) quantum correctionto the ground-stateen- mayconsiderthesimultaneousdiagonalizationofH0 and ergy, and α†k and βk† are the creation operators of the 1 in the naivestattempt to go beyond the noninteract- ferromagnetic and antiferromagnetic spin waves of mo- H ing spin-wave theory. Indeed, the higher-order terms we mentum k whose dispersion relations are given by take into account, the better description of the ground- statepropertiesweobtain[7]. However,suchanidea,asa ω± =ω (S s), (8) k k± − whole, ends in failure, bringing a gap to the lowest-lying ferromagnetic excitation branch and thus qualitatively with misreading the low-energy physics. Therefore, we take )))) an alternative approachat the idea of first diagonalizing ωk = (S s)2+4Sssin2(ak). (9) acbd 0 andnextextractingrelevantcorrectionsfrom 1. 0 q − (((( H H H is diagonalized as [3,4] Using the Wick theorem, is rewritten as 1 H H0 =E0+JXk (cid:16)ωk−α†kαk+ωk+βk†βk(cid:17) , (6) H1 =E1−JXk (cid:16)δωk−α†kαk+δωk+βk†βk(cid:17) 8888 .... + irrel+ quart, (10) 0000 H H where where the O(S0) correction to the ground-state energy, E0 =J [ωk (S+s)] , (7) E1, and those to the dispersions, δωk±, are, respectively, Xk − given by 6666 .... 0000 pppp //// kkkk aaaa 2222 4444 .... 0000 ) 2222 2 .... ))/ 0000 211 /,,) 1221 ,//, 1332 (((( 0000 .... 0000 0000 000 00 00 0 0 000 0000 .... ... .. .. . . ... .... 5443 433 23 22 2 1 111 0000 JJJJ //// ))))kkkk((((EEEE E = 2JN Γ2+Γ2+ S/s+ s/S Γ Γ , (11) III. THERMODYNAMIC PROPERTIES 1 − h 1 2 (cid:16)p p (cid:17) 1 2i sin2(ak) Γ δω± =2(S+s)Γ + 2 [ω (S s)] , (12) The dual structure of the excitations leads to unique k 1 ωk √Ss k± − thermal behavior. In order to complement quantum MonteCarlothermalcalculations,especiallyatlowtem- with peratures, we consider describing the thermodynamics in terms of the spin-wave theory. Introducing the addi- 1 S+s tional constraint of the total magnetization being zero Γ = 1 , 1 2N1Xk (cid:18)√Sωskcos−2(a(cid:19)k) (13) isnutcocetehdeedcoinnveonbttiaoinnainlgspainn-wexavceelltehnetordye,scTraipktaihonashoif [t2h9e] Γ2 = , low-temperature thermal behavior of one-dimensional −N ω Xk k Heisenberg ferromagnets. The present authors have re- cently applied the idea to the spin-(1,1) ferrimagnetic 2 while the irrelevant one-body terms Heisenberg chain [10]. Here we develop the method for general spin cases and make a detailed analysis of its (S s)2 cos(ak) validity as a function of (S,s). The core idea of the so- = J − Γ (α β +α†β†), (14) Hirrel − √Ss 1Xk ωk k k k k criazleleddamsocdoinfiterdolslipnign-wthaevebothsoeonrynu[3m0b,3e1rs] cbaynibmepsousminmgaa- certainconstraintonthemagnetization. Fromthispoint and the residual two-body interactions are both Hquart ofview,thezero-magnetizationconstraint,whichisquite neglectedsoastokeeptheferromagneticbranchgapless. reasonable for isotropic magnets, plays a relevant role in The resultant Hamiltonian is compactly represented as ferromagnets. Theresultantlow-temperatureexpansions [29] of the specific heat and susceptibility of the spin-s E +J ω−α†α +ω+β†β , (15) Heisenberg ferromagnetic chain, H≃ g Xk (cid:16) k k k k k k(cid:17) e e C 3 ζ(3) 1 = 2 t12 t+O(t32), (18) with NkB 8s21 √π − 2s2 χJ 2s4 ζ(1) ω± =ω± δω±, (16) = t−2 s25 2 t−32 k k − k N(gµB)2 3 − √π E =E +E +E . (17) eg class 0 1 s ζ(1) 2 + 2 t−1+O(t−12), (19) 2(cid:20) √π (cid:21) Now we compare all the calculations in Fig. 1. The diagonalization results fully demonstrate the steady ap- with t = k T/J and Riemann’s zeta function ζ(z), in- plicability and good precision of our Monte Carlo treat- B deedcoincidewiththethermodynamicBethe-ansatzcal- ment. The spin-wave approach generally gives a good culations [32] for s= 1 within the leading few terms. description of the low-energy structure. Even the free 2 In the present system, the zero-magnetization con- spin waves allow us to have a qualitative view of the straint is explicitly represented as elementary excitations. The relatively poor description of the antiferromagnetic branch by the free spin waves Sz+sz =(S s)N σn−σ =0, (20) implies that the quantum effect is more relevant in the h j ji − − k Xj Xk σX=± antiferromagnetic branch. Here we may be reminded of e the spin-wave treatment of mono-spin Heisenberg mag- where n± = n±P (n−,n+) with P (n−,n+) be- nets. Fortheferromagneticchains,thespin-wavedisper- k n−,n+ k k ingtheprobaPbilityofn− ferromagneticandn+ antiferro- sions are nothing but the exact picture of the elemen- magneticspinwavesappearinginthek-momentumstate. tary excitations, while for the antiferromagnetic chains, Equation (20) straightforwardly proposes that the ther- thosearegenerallynomorethanaqualitativeview. The malspindeviationshouldcancelthe N´eel-statemagneti- point is that in the presentsystem the spin-wavepicture zation. By minimizing the free energy is efficient for both elementary excitation branches and the spin-wave series potentially lead to an accurate de- F =E + (n−ω−+n+ω+) scription. More specifically, the divergence of the boson g k k k k numbers, which plagues the antiferromagneticspin-wave Xk e e e e treatmentinonedimension,doesnotoccurinthepresent +k T P (n−,n+)lnP (n−,n+), (21) B k k system. This viewpoint is further discussed in the final Xk nX−,n+ section. We conclude this section by listing in Table I the spin-waveestimates ofafew interestingquantities in with respect to P (n−,n+) at each k under the k comparisonwiththenumericalsolutions,wherewedefine condition (20) as well as the trivial constraints the curvature v as ω− =v(2ak)2. P (n−,n+)=1, we obtain k→0 n−,n+ k P e 4 C = 3 S−s 21 ζ(32)t12 1 t+O(t32), (22) mferargonmetaigcnaetsipcecatspaetctSf/osr=S/s1, while. aAhnuonthderredcopnesricdeenrt- NkB 4(cid:18) Ss (cid:19) √2π − Ss → ∞ ation also leads us to such a picture. Since the pertur- χJ Ss(S s)2 e e e ζ(1) = − t−2 (Ss)21(S s)32 2 t−32 vation from the decoupled dimers [8] qualitatively well N(gµB)2 3 − − √2π describesthelow-lyingexcitationsofthesystem,wepro- eζ(1) 2 e pose in Fig. 2 an idea of decomposing ferrimagnets +(S s) 2 t−1+O(t−21), (23) intoferromagnetsandantiferromagnets,whereweletthe − (cid:20)√2π(cid:21) decoupled dimers and the Affleck-Kennedy-Lieb-Tasaki e e valence-bond-solidstates[34]symbolizeferrimagnetsand wheret=k T/Jγ withγ =1 Γ (S+s)/Ss Γ /√Ss. B − 1 − 2 integer-spin gapped antiferromagnets, respectively. Now The specific heat C has been obtained by differentiating we expect spin-(S,s) ferrimagnets to behave like combi- e the free energy F, whereas the susceptibility χ by calcu- nations of spin-(2s) antiferromagnets and spin-(S s) lating( M2 M 2)/3T,wherewehavesetthegfactors − of the shpinsiS−handis both equal to g, because the differ- ferromagnets. Since the antiferromagnetic excitations of the ferrimagnetic ground state are gapped, the low- ence between them amounts to at most several percent temperature behavior of ferrimagnets should be only of of themselves in practice [33]. ferromagnetic aspect. At low temperatures there is in- deed no effective contributionfrom the spin-(2s) antifer- (1,1/2) (3/2,1/2) romagneticchainwithanexcitationgap[22]immediately abovethegroundstate. Inthiscontext,wearesurprised (a) (a) but pleased to find that provided S = 2s, the expres- sions(22) and(23)coincide with those for ferromagnets, (18)and(19),exceptforthequantumrenormalizingfac- (b) (b) tor γ. The low-temperature thermodynamics should be dominatedbythesmall-momentumferromagneticexcita- tions. Within the linearized spin-wavetheory,the small- (c) (c) momentum dispersions of Heisenberg ferrimagnets and ferromagnets are characterized by the curvatures (3/2,1) (2,1) v(S,s)-ferri = Ss J, (24) 2(S s) (a) (a) − v(S−s)-ferro =(S s)J, (25) − (b) (b) respectively, where we find that they coincide with each other only when S = 2s. The criterion, S = 2s, is fur- ther convincing when we consider the high-temperature (c) (c) behavior. The paramagnetic behavior of the spin-(S,s) ferrimagnet is given as FIG. 2. Schematic representations of the M = (S −s)N F(S,s)-ferri = ln[(2S+1)(2s+1)] , (26) groundstatesofspin-(S,s)ferrimagneticchainsofN elemen- Nk T − B tarycellsinthedecoupled-dimerlimit(a),theAKLTground k Tχ(S,s)-ferri 1 states of spin-(2s) antiferromagnetic chains of 2N spins (b), B = [S(S+1)+s(s+1)] , (27) the M = 2N(S−s) ground states of spin-(S−s) ferromag- Ng2µ2B 3 netic chains of 2N spins (c). The arrow (the bullet symbol) whereas those of the spin-(S s) ferromagnet and the denotes a spin 1/2 with its fixed (unfixed) projection value. − The solid segment is a singlet pair. The circle represents an spin-(2s) antiferromagnet are as follows: operation of constructing spins S and s by symmetrizing the spin 1/2’s inside. The relation (a) ≈ [(b)+(c)]/2 may be F(S−s)-ferro+F(2s)-antiferro expected. Nk T B = ln[(2S 2s+1)(4s+1)] , (28) − − Theanalyticexpressions(22)and(23)giveusabird’s- kBT(χ(S−s)-ferro+χ(2s)-antiferro) eye view of the one-dimensional ferrimagnetic behavior. Ng2µ2 B The spin-(S,s) ferrimagnet turns into the spin-s antifer- 1 romagnet in the limit of S s, whereas it looks like = [(S s)(S s+1)+2s(2s+1)] . (29) → 3 − − the spin-S ferromagnet in the limit of S/s . In → ∞ this sense, the subtraction S s may be regarded as Theseasymptoticvaluesagreewitheachotheronlywhen − the ferromagnetic contribution, while the residual spin S = 2s. This is simply the consequence of the degrees amplitude 2s as the antiferromagnetic one. No ferro- offreedom. InferrimagnetsofS >2s,the ferromagnetic 5 spin degrees of freedom overbalance the antiferromag- numberofthe bosons nσ ratherthanthe subtrac- σ=± k netic ones, while in ferrimagnets of S < 2s, vice versa. tion σn−σ. InoPurnaivestattemptto improvethe OnlythebalancedferrimagnetwithS =2siswellapprox- theorPy,σw=e±replkace the constraeint (20) by imatedasthe simple combinationofthe spin-(S s)fer- e − romagnet and the spin-(2s) antiferromagnet. However, Sz sz =(S+s 2Γ )N we note that even in the case of S = 2s, the similarity h j − ji − 1 Xj between the ferrimagnetic behavior, (22) and (23), and nσ theferromagneticone,(18)and(19),doesnotgobeyond (S+s) k =0. (30) − ω theleadingfewtermsshownhere. Theferromagneticfea- Xk σX=± ek tures of ferrimagnets are thermally smearedout. On the )))) other hand, the ferrimagnetic behavior further deviates Hoacbdwever,the alternativecondition (30)changesthe low- (((( from the purely ferromagnetic one due to the quantum temperature description, (22) and (23), which should be effect characterized by γ. The low-temperature expan- keptunchangedunderanyartificialconstraint,aswellas sions (22) and (23) imply that as temperature goes to considerablyunderestimatestheSchottky-likecharacter- 0000 .... zero, the quantum effect is reduced for the specific heat istic peak of the specific heat. In an attempt to remove 5555 C,whereasenhancedforthesusceptibilityχ. Inthelimit Γ1 from Eq. (30), we reach a phenomenological modifi- of S/s , the quantum corrections Γ and Γ both cation 1 2 → ∞ vanish. nσ :Sz sz : =(S+s)N (S+s) k =0, Although the spin-wave theory combined with the ad- h j − j i − ω ditional constraint (20) is so useful, it should further be Xj Xk σX=± ek 0000 modifiedathighertemperaturessoastocontrolthetotal (31) 4.4.4.4. ) ))2 JJJJ 21/) /,11 0000//// 12 ,, ....TTTT ,/22 3333 13 BBBB /( ((3 kkkk ( 0000 .... 2222 0000 .... 1111 0000 .... 0000 8822 00 66 88 4466 44 22 22 0000 .... .. .. .. .... .. .. .. .... 0011 11 00 00 0000 00 00 00 0000 BBBB kkkkNNNN //// CCCC )))) acbd (((( 0000 .... 5555 0000 .... 4444 ) ))2 JJJJ 21/) /,11 0000//// 12,, ....T T T T ,/22 3333 13/( BBBB 3 kkkk (( ( 0000 .... 2222 0000 .... 1111 0000 .... 0000 5000 0 00 00 50 0 0000 .... . .. .. .. . .... 1642 3 14 21 02 1 0000 BBBB BBBB ggggNNNN //// TTTT kkkk mmmm cccc 2222 2222 antiferromagnetic coupling between nearest neighbors in common? The surprising efficiency of the present spin- wave treatment can be attributed to the orderedground state [2] of ferrimagnets and thus to the nondiverging sublattice magnetization. The key constant Γ is noth- 1 ing but the quantum spin reduction 1 1 a†a = b†b , (32) N h j jig N h j jig Xj Xj where A denotes the ground-state average of A. Γ g 1 h i monotonically decreases as S/s increases, diverging at S/s = 1 and vanishing for S/s . Since the spin re- )) duction Γ can be a measure for→th∞e validity of the spin- 11 wave desc1ription, the spin-wave theory in general works 2,2, better as S/s, as well as Ss, increases. (( Inthis context,it isinteresting to observethe ground- 88 state spin correlations, SzSz for r =2la, f (r)= j j+l (33) S (cid:26)Szsz for r =(2l+1)a, )) j j+l 11 szsz for r =2la, ,, f (r)= j j+l (34) 22 s (cid:26)szSz for r =(2l+1)a. // j j+l+1 33 (( 66 aa )) // 22 // rr 11 ,, 22 3/3/ 44 (( )) 22 // 11 22 ,, 11 (( )) ab (( 00 00 05 00 0 50 00 05 .. .. .. . .. .. .. 14 30 20 1 00 11 21 - -- -- Ss ))rr(( ff [8] S.Yamamoto,S.Brehmer,andH.-J.Mikeska,Phys.Rev. nard, M. Drillon, and P. Legoll, Inorg. Chem. 29, 3325 B 57, 13610 (1998). (1990). [9] S. Yamamoto and T. Sakai, J. Phys. Soc. Jpn. 67, 3711 [21] A.Caneschi,D.Gatteschi,P.Rey,andR.Sessoli, Inorg. (1998). Chem. 27, 1756 (1988); A.Caneschi, D.Gatteschi, J.-P. [10] S. Yamamoto and T. Fukui, Phys. Rev. B 57, 14008 Renard, P. Rey, and R. Sessoli, ibid. 28, 1976 (1989); (1998); S. Yamamoto, T. Fukui, K. Maisinger, and U. 2940 (1989). Schollw¨ock,J.Phys.: Condens.Matter10,11033(1998). [22] F. D. M. Haldane, Phys. Lett. 93A, 464 (1983); Phys. [11] K.Maisinger, U.Schollw¨ock,S.Brehmer,H.-J.Mikeska, Rev. Lett.50, 1153 (1983). and S. Yamamoto, Phys. Rev. B 58, 5908 (1998); [23] T. Oguchi, Phys.Rev. 117, 117 (1960). A. K. Kolezhuk, H.-J. Mikeska, K. Maisinger, and U. [24] S. M. Rezende, Phys.Rev.B 42, 2589 (1990). Schollw¨ock, Phys. Rev.B 59, 13565 (1999); [25] S. Yamamoto, Int.J. Mod. Phys.C 8, 609 (1997). [12] T. Kuramoto, J. Phys. Soc. Jpn. 67, 1762 (1998); 68, [26] S. Yamamoto, Phys.Rev.Lett. 75, 3348 (1995). 1813 (1999). [27] S. Yamamoto and S. Miyashita, Phys. Lett. A 235, 545 [13] S.Yamamoto, Phys.Rev. B59, 1024 (1999). (1997). [14] C.Wu,B.Chen,X.Dai,Y.Yu,andZ.-B.Su,Phys.Rev. [28] S. Yamamoto, Phys.Rev.B 51, 16128 (1995). B 60, 1057 (1999). [29] M. Takahashi, Phys. Rev.Lett. 58, 168 (1987). [15] T.SakaiandS.Yamamoto,Phys.Rev.B60,4053(1999); [30] M. Takahashi, Phys. Rev.B 40, 2494 (1989). S.YamamotoandT.Sakai,J.Phys.: Codens.Matter11, [31] J. E. Hirschand S.Tang, Phys.Rev.B 40, 4769 (1989); 5175 (1999). S.Tang,M.E.Lazzouni,andJ.E.Hirsch,Phys.Rev.B [16] A. Gleizes and M. Verdaguer, J. Am. Chem. Soc. 103, 40, 5000 (1989). 7373 (1981); 106, 3727 (1984). [32] M. Takahashi and M. Yamada, J. Phys. Soc. Jpn. 54, [17] Y.Pei,M.Verdaguer,O.Kahn,J.Sletten,andJ.-P.Re- 2808 (1985); M. Yamada and M. Takahashi, ibid. 55, nard,Inorg. Chem. 26, 138 (1987). 2024 (1986). [18] M.Drillon,J.C.Gianduzzo,andR.Georges,Phys.Lett. [33] M.Hagiwara, K.Minami, Y.Narumi,K.Tatani,andK. 96A, 413 (1983); M. Drillon, E. Coronado, R. Georges, Kindo, J. Phys.Soc. Jpn. 67, 2209 (1998). J. C. Gianduzzo, and J. Curely, Phys. Rev.B 40, 10992 [34] I. Affleck,T. Kennedy,E. H.Lieb, and H.Tasaki, Phys. (1989). Rev. Lett. 59, 799 (1987); Commun. Math. Phys. 115, [19] M. Verdaguer, A. Gleizes, J.-P. Renard, and J. Seiden, 477 (1988). Phys.Rev.B 29, 5144 (1984). [35] N. Fujiwara and M. Hagiwara, to be published in Solid [20] O. Kahn, Y. Pei, M. Verdaguer, J.-P. Renard, and J. State Commun.. Sletten, J. Am. Chem. Soc. 110, 782 (1988); P. J. van [36] S. Yamamoto, to be published in Phys. Rev. B 61, No. Koningsbruggen,O.Kahn,K.Nakatani,Y.Pei,J.-P.Re- 2 (2000). TABLE I. Comparison between the linear-spin-wave (LSW), interacting-spin-wave (ISW), and numerical estimates of the ground-state energy Eg, the antiferromagnetic excitation gap ∆, and the curvature of the ferromagnetic dispersion, v. The most accurate numerical estimates of Eg and ∆ are obtained by the exact diagonalization (Exact), whereas those of v by the quantumMonte Carlo (QMC). (S,s) Eg/NJ ∆/J v LSW ISW Exact LSW ISW Exact LSW ISW QMC (1,1) −1.4365 −1.4608 −1.4541(1) 1 1.6756 1.759(1) 1 0.3804 0.37(1) 2 2 (3,1) −1.9580 −1.9698 −1.9672(1) 2 2.8025 2.842(1) 3 0.3390 0.31(1) 2 2 8 (3,1) −3.8281 −3.8676 −3.861(1) 1 1.5214 1.615(5) 3 1.1319 0.90(3) 2 2 (2,1) −4.8729 −4.8973 −4.893(1) 2 2.6756 2.730(5) 1 0.8804 0.76(2) 9