9 9 9 1 n a J Critical and off-critical studies of the 2 1 Baxter-Wu model with general toroidal ] h boundary conditions c e m - F. C. Alcaraz and J. C. Xavier t a Departamento de F´ısica t s . Universidade Federal de S˜ao Carlos t a m 13565-905, S˜ao Carlos, SP, Brasil - d PACS numbers : 05.50+q,64.60Cn, 75.10Jn n o c [ 1 Abstract v 9 TheoperatorcontentoftheBaxter-Wumodelwithgeneraltoroidal 9 boundaryconditions is calculated analytically and numerically. These 0 1 calculations were done by relating the partition function of the model 0 withthegeneratingfunctionofasite-colouringprobleminahexagonal 9 lattice. Extending the original Bethe-ansatz solution of the related 9 / colouring problem we are able to calculate the eigenspectra of both t a models by solving the associated Bethe-ansatz equations. We have m also calculated, by exploring the conformal invariance at the critical - point, the mass ratios of the underlying massive theory governing the d n Baxter-Wu model in the vicinity of its critical point. o c : v 1 Introduction i X r The Baxter-Wu model is the simplest non-trivial spin model with three-spin a interactions. Its Hamiltonian is defined on a triangular lattice by H = J σ σ σ , (1) i j k − <ijk> X 1 where the sum extends over the elementary triangles, J is the coupling con- stant and σ = 1 are Ising variables located at the sites. Historically this i ± model was introduced by Wood and Griffiths in 1972 [1], as an example of a model exhibiting an order-disorder phase transition and not having a global up-down spin reversal symmetry. This model is self-dual [1, 2] with the same critical temperature as that of the Ising model on a square lattice. In 1973 Baxter and Wu [3] related the partition function of this model, in the ther- modynamic limit, with the generating function of a site colouring problem on an hexagonal lattice. Solving this colouring problem through a generalized Bethe-ansatz they calculated the leading exponents [3] α = 2/3, µ = 2/3 and η = 1/4 of the Baxter-Wu model. The equality of these exponents with those of the 4-state Potts model [4-6] added to the fact that both models have the same four-fold degeneracy of the ground state, induce the conjec- ture that they share the same universality class of critical behaviour. This conjecture is interesting since contrary to the Baxter-Wu model the 4-state Potts model is exactly integrable only at its critical point. However from nu- merical studies of these models on a finite lattice it is well known that both models show different corrections to finite-size scaling (at the critical point). Whereas in the 4-state Potts models [7-10] these corrections are governed by a marginal operator, producing logarithmic corrections with the system size, which brings enormous difficulty to extract reliable results from finite-size studies of the model, this is not the case in the Baxter-Wu model [11-13]. Nowadays with the developments of conformal invariance applied to criti- calphenomena [13]theclassificationofdifferent universality classes ofcritical behaviour becomes clear. Two models belong to the same universality class only if they share the same operator content, not only the leading critical exponents. The operator content of the 4-state Potts model was already conjectured from finite-size studies in its Hamiltonian formulation [14] and can be obtained by a Z(2) orbifold of the Gaussian model in a special com- pactification radius (see [15] for a review). In this paper we numerically and analytically calculate the operator content of the Baxter-Wu model with sev- eral boundary conditions. Part of our numerical calculation was announced in [12]. In order to do this calculation we generalize the original Bethe- ansatz solution of the related site-colouring problem with periodic boundary condition. This is necessary because this solution only gives part of the eigenspectrum of the associated transfer matrix. We also extend our Bethe- ansatz solution in other cases where the site-colouring problem is not on a 2 periodic lattice. This extension enables us to obtain the operator content of this model and the Baxter-Wu model for more general toroidal bound- ary conditions. Our numerical study was done by numerically solving the Bethe-ansatz equations and the analytical work was done by studying these equations using standard techniques based on the Wiener-Hopf method [16]. We believe that beyond the Ising model the Baxter-Wu model is the simplest spin model that can be solved exactly for arbitrary temperatures. We explore this solution in order to obtain the mass spectrum of the massive theory describing fluctuations near the critical point. The layout of this paper is as follows. In section 2 we introduce a site- colouring problem on an hexagonal lattice whose generating function is ex- actly related, in the bulk limit, with the Baxter-Wu model. Our construc- tion is valid for some toroidal boundary conditions, generalizing the original construction [3] for periodic lattices. In section 3 we present the transfer matrix associated to the colouring problem and calculate its eigenvalues by the Bethe-ansatz. In section 4 the operator content of the Baxter-Wu model and the related site-colouring problem is obtained. The mass spectrum of the massive field theory governing the thermal and magnetic perturbations is calculated in section 5. Finally our conclusions are present in section 6 and the analytical calculation of some of the conformal dimensions of the site-colouring problem is presented in an appendix. 2 The Baxter-Wu model and the related site- colouring problem InthissectionwerelatetheBaxter-Wumodelwithgeneraltoroidalboundary conditionswithasite-colouring problem(SCP)onthehexagonallattice. The construction presented here generalizes those presented by Baxter and Wu [3] for the periodic case. Let us consider a triangular lattice with L (M) rows (columns) along the horizontal (vertical) direction, respectively. For convenience we take L as a multiple of 3, and decompose the lattice in three triangular sublattices 2 formed by the points denoted by , and in Fig. 1. We attach at each site of a sublattice Ising variables◦σ◦ , σ2△and σ△ , and the Baxter-Wu { } { } { } model, with the Hamiltonian (1) is given in terms of the simplest three-body 3 interactions with one spin in each sublattice. The model has a non-local Z(2) Z(2) symmetry corresponding to the × global change of variables in two of the sublattices, i.e., σ△ aσ△ σ◦ bσ◦ σ2 abσ2 (a = b = 1). → → → ± This fact implies that at sufficiently low temperatures the model have a four- fold degenerate ground state. Creating, at low temperatures, domain walls connecting those different ground states we see that in fact the symmetry of those long-range excitations is D(4), like in the 4-state Potts model. These reasonings suggest that critical fluctuations of both models are described in terms of the same quantum field theory. In order to do a detailed study of several sectors of this underlying field theory we consider the Baxter- Wu model with general toroidal boundary conditions compatible with its Z(2) Z(2) symmetry: × σ△ = aσ△ σ◦ = bσ◦ σ2 = abσ2 , (2) i,j i+L,j i,j i+L,j i,j i+L,j 2 where a,b = 1. In Fig.1 the simbols , and denote the bordering sites ± ◦ △ which are related by (2) to the bulk ones (filled simbols). Imposing periodic boundary condition along the vertical direction the partition function of the model ZBW can be written as L×M ZBW = Tr TˆBW M , (3) L×M (a,b) (cid:16) (cid:17) where TˆBW is the associated row-to-row transfer matrix. Its elements (a,b) σ ,...,σ TˆBW σ′,...,σ′ are given by the Boltzmann weights generated by h 1 N| (a,b)| 1 Ni the spin configurations σ ,...,σ and σ′,...,σ′ of adjacent rows and are { 1 N} { 1 N} given by L σ ,...,σ TˆBW σ′,...,σ′ = exp K σ′σ (σ +σ′ ) , (4) h 1 N| (a,b)| 1 Ni  j j+2 j+1 j+1  j=1 X   with K = J and on the right side the appropriate boundary condition kBT (a,b) is taken into account. Following Baxter andWu [3]we canrelate ZBW to thepartitionfunction L×M orgenerating functionZSCP ofaSCP onanhoneycomb latticewithN = 2L N×M 3 4 rows and M columns. In order to do this, it is convenient to define link variables λ at the links of the hexagonal lattice formed by the sublattices { } 2 and (heavy lines in Fig.1). These variables are given by the product of t◦he site variables at the ends of the link (λ = σ2σ◦). In terms of these new variables the Hamiltonian is given by H = J σ△(λ +λ +λ ), (5) − i 1 2··· 6 i∈△ X where the summation is only over the sublattice and λ ,λ ,...,λ are the 1 2 6 link variables surrounding a given site variable△σ△. Taking into account i the boundary conditions (2) the partition function, in terms of these new variables, can be written as 1 ZBW = exp Kσ△(f λ +f λ + +f λ ) (1+λ λ ...λ ) , L×M 1 1 2 2 ··· 6 6 2 1 2 6 σX△,λ Y△ (cid:26) n o (cid:27) { } (6) where the product extends over all the elementary hexagons formed by the sites in sublattices 2 and , and surrounding a site variable σ△. The factor ◦ 1(1+λ λ ...λ )in(6)isnecessarysincethevariables λ arenotindependent. 2 1 2 6 { } The factors f (i = 1,2,...,6) in (6) are constants defined on the links of the i 2 hexagonal lattice ( - ) and depend on their relative location (see Fig. 1). ◦ If both ends of a link i (with corresponding variable λ ) belongs to the bulk i of the lattice ( - ) f = 1, if one of its ends is on the border ( - or 2) i • ◦ •− f = ab, and if both ends are on the border ( -2) f = a. To proceed it is i i ◦ convenient [3] to rewrite the last product in (6), surrounding each lattice site as △ 1 (1+λ λ ...λ ) = g(λ ,µ△)g(λ ,µ△) g(λ ,µ△), (7) 1 2 6 1 2 6 2 ··· µ△X=±1 where g(λ,µ) = 2−7/6(λ+µ+ λ µ ). (8) | − | Substituting the expression (7) into equation (6) the partition function will be expressed in terms of a single sublattice with site variables σ△ and △ { } µ△ , and link variables λ . Taking into account the boundary factors it is { } { } 5 straightforward to write ZBW = exp K(σ△ +σ△)λ g(λ ,µ△)g(λ ,µ△) , L×M  i j i,j i,j i i,j j  {σ△X,µ△}<Yi,j>λi,Xj=±1 (cid:16) (cid:17)  (9)   where < i,j > are links on the hexagonal lattice formed by the sublattices 2 and (see Fig. 1). In order to derive (9) we are forced to restrict ourselves ◦ only to boundary conditions (2) where a = b = 1. Fortunately this does ± not restrict our analysis since due to the D(4) symmetry of the model, the eigenspectra of the transfer matrices with boundary condition a = b and 6 a = b = 1 are degenerate. Summing over λ in (9) we obtain − { } ZBW = w(σ△,µ△;σ△,µ△), (10) L×M i i j j {σ△X,µ△}<Yi,j> where w(σ ,µ ;σ ,µ ) = 2−1/3[exp(K(σ +σ ))+µ µ exp( K(σ +σ ))]. (11) i i j j i j i j i j − Finally, following Baxter and Wu [3] we now associate the above partition function to the generating function of a SCP with 8 colours. We associate odd colours 1, 3, 5 and 7 at a given site according to the values of the △ variables (σ△,µ△) attached at the site: (++) 1 ( +) 3 ( ) 5 (++) 7. (12) → − → −− → → The relation (11) tell us that links connecting colours 3,7 and 1,5 should be forbidden in the SCP. This constraint can be easily implemented [3] by introducing the even colours 2, 4, 6 and 8 on the sublattice , forming ◦ with the sublattice an hexagonal lattice with N = 2L (M) sites in the △ 3 horizontal (vertical) direction. The constraint is that all nearest-neighbour colours on this hexagonal lattice must differ by 1 (modulo 8). ForM , ± → ∞ the partition function ZBW is then related to the generating function ZSCP, L,M N,M given by ZSCP = zn1zn2 zn8 = ZBW, (13) N,M 1 2 ··· 8 L,M {XC} where n ,n ,...,n is the number of sites coloured with colour 1,2,...,8 in a 1 2 8 given configuration C. The fugacities z (i = 1,2,...,8) are obtained from i 6 (11) z = z = z = z = 2sinh(4K), 1 3 5 7 (14) z−1 = z = z−1 = z = sinh(2K) t. 2 4 6 8 ≡ ThecriticalpointoftheBaxter-WumodelandoftheSCP isgivenbytheself- dual point t = t = 1 [3]. Concerning the boundary conditions the relations c are as follows. The ZBW with periodic boundary condition (a = b = 1 in L,M (2)) is related with ZSCP, also with periodic lattice. The boundary condition N,M a = b = 1 in (2) is related to a boundary condition in the SCP such that − if at site (1,j) we have a colour variable c = 1,3,5 or 7 at site (N +1,j) 1,j the colour variable should be c = 3,1,7 and 5 respectively. We do not N+1,j need to mention even colours since N is even (see Fig. 1). 3 The transfer matrix and the Bethe-ansatz of the SCP In this section we derive the row-to-row transfer matrix of the SCP and generalize its Bethe-ansatz solution presented by Baxter and Wu [3]. The solution presented in [3], as we shall see, only gives part of the eigenspectra of the transfer matrix of the SCP, with periodic boundary conditions. Here we extend the solutions forthe periodic case andalso derive the Bethe-ansatz equations for the more general boundary condition c = c +2κ (mod 8); κ = 0,1,2 or 3, (15) i,j i+N,j along the horizontal direction. The periodic case corresponds to κ = 0. Following [3], for a given configuration c of colours on the hexagonal i,j { } lattice, we say we have a dislocation on a given link of the lattice wherever the colour on the right end of the link is smaller (modulo 8) than that on the left end. In Fig. 2 we show two configurations with the corresponding dislocations (dotted lines) for a lattice with width N = 4. In Fig. 2a the lattice is periodic (κ = 0) and in Fig. 2b the boundary condition is given by (15) with κ = 1. The generating function of the SCP can be written as ZSCP = Tr TˆSCP M , (16) N,M (κ) (cid:16) (cid:17) 7 where TˆSCP is the associated row-to-row transfer matrix. This transfer (κ) matrix has elements C TˆSCP C′ given by the product of the Boltz- h{ }| (κ) |{ }i mann weights due to the colour configurations C = c ,c ,..,c and 1 2 N { } { } C′ = c′,c′,...,c′ of two adjacent rows. If the configuration produced { } { 1 2 N} by C and C′ contains only colours differing by 1 (modulo 8), we have { } { } ± N C TˆSCP C′ = (z(c )z(c′))1/2, (17) h{ }| (κ) |{ }i i i i=1 Y with fugacities z defined in (14). On the other hand if the configuration does i not satisfy this constraint, C TˆSCP C′ = 0. It is simple to see that h{ }| (κ) |{ }i this requirement implies, for arbitrary values of κ in (15), the conservation of the number n of dislocations along the vertical direction. Consequently the Hilbert space associated to TˆSCP can be separated into block-disjoint sectors (κ) labelled by the values of n. The possible values of n depend on the boundary condition (15). For κ = 0, 2 (1, 3) they are even (odd), and are given by n , j where 0 n = N κ 4j 2N j = 0, 1 2,... (18) j ≤ − − ≤ ± ± The colour configurations C and C′ , in a sector with n disloca- { } { } tions, can be conveniently expressed by the sets (m;X) = (m;x ,x ,...,x ) 1 2 n and (m′;X′) = (m′;x′,x′,...,x′ ), respectively. The odd numbers m and 1 2 n m′ give the colour at the first site and the sets X (x ,x ,...,x ) and 1 2 n ≡ X′ (x′,x′,...,x′ ) give the position of the dislocations on the row. The sets ≡ 1 2 n X and X′ should satisfy 1 x x x N , 1 x′ x′ x′ N, (19) ≤ 1 ≤ 2 ≤ ··· ≤ n ≤ ≤ 1 ≤ 2 ≤ ··· ≤ n ≤ and should have no more than one repeated value of x or x′ if they are odd, or more than three repeated values if they are even. In Fig. 2a we show the configurations (m;X) = (1;2,3,4,4) and (m′;X′) = (3;2,2,3,4) which belong to the sector with n = 4 and periodic boundary condition. In Fig 2b we show the configurations (m;X) = (1;2,3,4) and (m′;X′) = (3;2,2,2) belonging to the sector with n = 3 in the lattice with boundary condition κ = 1 in (15) and width N = 4. Two configurations X = (x ,x ,...,x ) and 1 2 n X′ = (x′,x′,...,x′ ) are connected through the operator TˆSCP if beyond (19) 1 2 n (κ) x′ = x 1, if x odd j j − j (20) x′ = x ,x 1,x 2 if x even. j j j − j − j 8 In the case x = x we have an additional constraint x′ = x′ . We should i i+1 i 6 i+1 notice, for arbitrary boundary condition, the identification X = (0,x ,x ,...,x ) = (x ,x ,...,x ,N). (21) 2 3 n 2 3 n The transfer matrix, in a given sector with n dislocations is now given by 1−(−1)κ n m,X TˆSCP m′,X′ = z 4 w(m+x +x′ 2j), w(m) = (z z )1/2, h | (κ) | i m+1 j j − m m+1 j=1 Y (22) if X and X′ satisfy (19) and (20), and is zero otherwise. In the sector with n dislocations the eigenvectors ψ(n) of TˆSCP, with eigenvalue Λ can be written (κ) as ψ(n) = z−1−(−81)κf (x ,x , ,x ) m;X , (23) m+1 m 1 2 ··· n | i {mX,X} where the summation is restricted to the configurations with n dislocations, and f (x ,x , ,x ) are unknown amplitudes. The eigenvalue equation for m 1 2 n TˆSCP is given b··y· (κ) ∗ n w(m+x +x′ 2j) f (X′) = Λf (X), (24)  j j −  m+2 m X′ j=1 X Y   where the asterisk indicates that X and X′ satisfy the conditions (19) and (20). The relation (21) implies that in (24) the amplitudes having x′ = 0 1 should be replaced by the boundary condition 1−(−1)κ f (0,x′,x′, ,x′ ) = z 4 f (x′,x′, ,x′ ,N). (25) m+2 2 3 ··· n m+1 m 2 3 ··· n Due to the values of the fugacities (14) it is simple to see that TˆSCP, (κ) besides conserving the number of dislocations, also has an additional Z(2) symmetry (eigenvalues ǫ = 1), since adding 4 (modulo 8) to all colours in ± a given configuration does not change its weight in the generating function, that is f (X) = ǫf (X). (26) m+4 m Following Baxter and Wu [3] we assume the following Bethe-ansatz for the amplitudes f (X) = a(P)φ (m 2,x ) φ (m 2n,x ), (27) m P1 − 1 ··· Pn − n P X 9 where the summation is over all the n! permutations P = P ,P ,...,P of 1 2 n { } integers 1,2,...,n . We require the existence of n wave numbers k (j = j { } 1,2,...,n) and signs ǫ (j = 1,2,...,n) such that j a exp(ik x), x odd φ (m,x) = ǫ φ (m+4,x) = j,m j (28) j j j b exp(ik x), x even. ( j,m j Observe that the Z(2)-parity eigenvalue of the wave function is given by ǫ = n ǫ , and it is even or odd depending on the numbers of negative j=1 j values of ǫ . The Bethe-ansatz solution presented by Baxter and Wu [3] only Q j gives the symmetric eigenvalues (ǫ = 1), for periodic boundary conditions (κ = 0). We can follow the same procedure as in [3] in order to derive the Bethe-ansatz equations. We have to consider various possible choices of X to determine the eigenvalue Λ. Firstly let us consider the case where all dislocations are located at dis- tinct positions, i.e., x = x = = x . Eq. (24) is then replaced by 1 2 n 6 6 ··· 6 w(m+x+x′)φ (m+2,x′) = λ φ (m,x), j = 1,...,n, (29) j j j x′ X where we have denoted Λ = λ λ . (30) 1 n ··· Actually Eq. (29) represents two equations corresponding to x odd or even and can be written as T V = λ V , (31) j,m+2 j,m+2 j j,m where 0 α a T = j,m+1 ; V = j,m (32) j,m+2 α A j,m b j,m−1 j,m ! j,m ! with α = w(m)exp( ik ); A = w(m)+w(m 2)exp( 2ik ). (33) j,m j j,m j − − − Using the fact that a a j,m−2 = ǫ j,m+2 , b j b j,m−2 ! j,m+2 ! it is simple to see that λ can be obtained from the eigenvalue equation j (T T )V = λ2ǫ V . j,m j,m+2 j,m+2 j j j,m+2 10