Magnetism in One-Dimensional Systems—The Heisenberg Model for Infinite Spin Mhewart E, Fisien* ‘The Rockefeller Institute, New York 21, New York (Received 25 October 1963) Is observed that the free-energy, susceptibility, snd correlation functions for a hain of JV spins with nearest-neighbor isotropic Heisenberg coupling can be calculated explicitly in ‘the (classical) limit of inf “Heisenberg chains of spin 3, I, INTRODUCTION ‘UCH of the theoretical study of magnetism has been based on the Heisenberg and Ising spin Hamiltonians which, for a system of LN magnetic atoms in an external field H, may be combined in the form + Jui (SPSF+ S252) a EHS, (1.1) where Si, Si” and S\* are the components of the spin vector §; of the ith atom. The case Ji corresponds to pure Ising coupling while Jij* = Ji; corresponds to spatially isotropic Heisen- berg coupling. In general, one can calculate the free energy of the corresponding magnetic sys- tem and thence its other physical properties, only asa series in powers of 1/T valid at high temperatures or as an asymptotic (spin wave) series valid at low temperatures. Exact solu- tions, when available, enable one to gauge the convergence of such series and to judge the accuracy of approximate results derived by ex- trapolating the series, or from variational pri ciples, Green's-function decoupling procedures, etc. ‘The only exact results for two- or three- dimensional systems are Onsager’s famous cal- culation for the S=4 plane square lattice Ising model (J#=0) with nearest-neighbor interac- tions' and its extensions to other plane lattices.? Even for one-dimensional systems, however, * On leave of absence from the Wheatstone Physice Laboratory, King’s College, London W. C. 2, England. TL, Onsager, Phys. Rev. 68, 117 (1944), 12 For references see M. E, Fisher, J. Math, Phys, 4, 278, (1963). ‘The results are compared briefly with those for Tsing and exact solutions are of interest both because of the mathematical difficulty of the general theo- retical problem and because in some real crystals the dominant spin interactions occur only be- tween atoms arranged in linear chains* The solution of the one-dimensional S=4 Ising model with nearest neighbor coupling is simple and well known? Recently, however, exact results have also been obtained for nearest neighbor purely transverse coupling (i.e., J!!=0) in S=4 linear chains. Lieb, Schultz, and Matis have made a detailed study of the spectrum and eigenstates of this Hamiltonian while Katsura® calculated the exact partition function and in- vestigated the thermal and magnetic properties of an infinite chain, It seems not to have been previously observed that an explicit solution for the linear chain can also be obtained for pure isotropic Heisenberg coupling in the limit of infinite spin (S—> %). As we show in this note the mathematics is quite clementary—the Hamiltonian being es- sentially classical—and is familiar in connection with the problem of the configuration of a linear polymer molecule with restricted bond angles.* The results are, however, quite instructive and the comparison with other cases is interesting. II, FREE ENERGY IN ZERO FIELD For nearest-neighbor isotropic interactions in an open linear chain of N-+1 atoms of spin S the *See, for example, T, Haseda and A. R. Miedema, hoses 27, 1102 (96), "t Watanabe and Haseda, I Chem, Ps. 28, 325 (i988), and KB. Grit, tbo publi Cgfen T Sels and D. Mats, Am, Phys. 6,407 {Sten Phys. Rev, 1, 508 (1960), Sc, for example, Pej. Flos Panels of Polymer Chemises (Corvall Vniveriry Bre, Tehace, 1983), Chap. {0p 40859 an pp. sie 5. 343344 Hamiltonian (1-1) becomes x x 23s x SiSia-gsd EES. (24) and the ground state energy for ferromagnetic coupling (J>0) is Ey=—2NJsS—(N+1)gs8SH. (2.2) To effect a reasonable comparison between different spin values (in particular with S=3) and to enable the limit S— % to be taken we set 2sS=3T, g5= 4g, and introduce the unit vector operators ;= with commutation relations SP51— 5/5 = (1/Spiss*. (24) In the limit S— © these operators commute and (2.1) goes over to the classical Hamiltonian x x “HN zy SiS 3g8 EAs (2.5) with ferromagnetic ground state energy Eo= -}NJ-HN41g6H, (2.6) obtained by aligning all spins parallel. (With the normalization (2.3) the corrections for large, finite S with respect to the limiting results for S=n is of order 1/S. If, however, S is replaced by (S(S+1)} the correction terms are only of order 1/S*) Setting K=4J/kT the partition function in zero field is simply tun ff E.. PE expt Ese (2.7) (2.8) where dQ; is the clement of solid angle for the vector s,. This expression is quite analogous to the configurational partition function for a linear molecular chain with restricted rotation between adjacent bonds which are directed along the vectors §:. If polar coordinates 6i, ¢: for s, G=1, 2, ++-N) are referred to s--1 as polar axis the integrals separate yielding Ly f on f ” exp(K cost.) sindsld, MICHAEL EB, FISHER so that 2y(K)=((sinhK)/K]*. Owing to the symmetry of this expression under K+ =K the thermal properties of the model are the same for ferromagnetic as for anti- ferromagnetic interactions (ie., for J positive or negative). The free energy per spin in zero field for a ong chain is then given by (2.9) —Fy/NkT=InsinhK—InK. (2.10) ‘The magnetic internal energy is Uv/ANI=(1/K)—cothK, (2.11) which has the high-temperature expansion, 1 Ux/ANI=-=K+—K + 345 (oyna Qn)! + ByKet +++ (2.12) and the low-temperature expansion (J>0) Uy /}NI= 14287 /T Ie HT Ag BIRT (2.13) For the specific heat we have Cy/Nk=1—KY/sinh’K, (2.14) so that [as can also be seen from (2.13)] Cx(T) does not approach zero as T—>0. Correspond- ingly the entropy Sy(T) diverges to — = as T—r0 (assuming Sy(T) 0 as T+ ~). These are, of course, unrealistic features of any clas- sical spin model. For systems with large but finite spin the specific heat would deviate from (2.14) at very low temperatures and fall to zero. At higher temperatures, however, the devia- tions would only be of order 1/S. ‘The specific heat is plotted in Fig. 1 [curve (@)] and seen to rise monotonically as the tempera~ ture falls, to a maximum value at T'=0. For comparison are shown the specific heats for (b) the S=§ Ising chain (J+=0), (c) the S=3 anti- ferromagnetic and (d) ferromagnetic isotropic Heisenberg chains (J!'=J}!) and (e) the S=4 pure transverse chain (J!'=0).* The results for the S= Heisenberg chains are accurate esti-MAGNETISM IN ONE-DIMENSIONAL SYSTEMS. ‘mates recently derived from numerical studies.? It is evident that the system with infinite spin is relatively much less stable. The most stable system, as judged from the position of the specific heat maximum, is the S=} antiferro- magnetic Heisenberg chain. III. CORRELATION FUNCTIONS AND SUSCEPTIBILITY ‘The spin pair (and higher order) correlation functions for a finite system in zero field can be evaluated in a similar way to the partition function. Defining a pair correlation function by soy.s(D) = (Se-8e41)=3(si*5640", A) =3hy4 f o Cond Xexp[—paelsit], (3.2) the integrals for j>i+1 (120) may be factored off as before. Similarly by referring s; to s;as polar axis the integrals for j0) the suscepti- bility rises monotonically and diverges as 1/7? at T=0, The antiferromagnetic susceptibility, however, is finite at T=0 as might have been anticipated since it is a consequence of the spatial isotropy. The low-temperature expansion is, using (3.6) and (2.13), New RIT XaaulT) HARTY [J|-+@T/[ TP} (3.12) The susceptibility then rises to a rounded maxi- mum of relative height Xeex/X(0) =1.2045, MICHAEL FE, FISHER at ED max/ | J) = 0.2382, (3.14) as shown in Fig. 2 [curve (a)]. For comparison the susceptibility of (b) the S=3 antiferromag- netic Heisenberg chain? and (c) the perpendicular susceptibility for an Ising chain** of S=3 are also plotted. The Ising perpendicular suscepti- bility has been reduced by a factor of three since it is “anomalously” large as the antiferro- magnetic spin coupling does not operate be- tween spins perpendicular to the axis in this The result (3.11) can immediately be gen- eralized to yield the susceptibility of a “Bethe lattice” of coordination number g. (A Bethe lattice is an infinite Cayley tree on which Bethe’s statistical approximation is exact since there are no closed loops.) The denominator in (3.11) should merely be changed to [1~(q—1)u(K)1. ‘This model displays a transition at a critical point T, given by cothK.—(1/K.)=1/(g—1), where the ferromagnetic susceptibility diverges as 1/(TT.) while the anti ceptibility remains finite, The modified formula for x(7) is, of course, valid only for temperatures above the critical point. It does not seem possible by simple methods to solve the infinite-spin problem explicitly in the presence of a finite magnetic field although a formal solution could be given by expressing all the derivatives with respect to field in terms of the higher order spin correlations in zero field which can be calculated. (3.15) sromagnetic sus- ACKNOWLEDGMENTS I am grateful to Professor C. Domb for originally stimulating my interest in the prob- lem for infinite spin and to Miss J. C. Bonner for some of the numerical calculations. "M. E, Fisher, Physica 26, 618 (1960); J. Math, Phys. 4, 124 (1968). #M. E. Fisher and J. W. Essam, J. Math, Phys. 2, 609 (ase