SISSA 33/2004/FM Integrals of motion from TBA and lattice-conformal dictionary Giovanni Feverati , Paolo Grinza ∗ ∗∗ International School for Advanced Studies (SISSA) via Beirut 2-4, 34014 Trieste, Italy INFN, sezione di Trieste 5 Abstract 0 0 TheintegralsofmotionofthetricriticalIsingmodelareobtainedbyThermodynamicBethe 2 Ansatz (TBA) equations derived from the A4 integrable lattice model. They are compared n with those given by the conformalfield theory leading to a unique one-to-one lattice-conformal a correspondence. They can also be followed along the renormalizationgroup flows generated by J the action of the boundary field ϕ1,3 on conformal boundary conditions in close analogy to the 0 usual TBA description of energies. 2 3 1 Introduction v 0 1 The study of two-dimensional integrable quantum field theories (QFTs) and integrable lattice 1 models has lead to the discovery of a large number of deep analogies between them. It is worth 5 0 to recall here that, in both cases, integrability is equivalent to an algebraic relation (Yang-Baxter 4 equation)forthefundamental“evolution”operatorinthetwolanguages, theS-matrixintheformer 0 / caseandthetransfermatrixinthelattercase. Furthermore,apossiblelinkbetweenthemisgivenby h the so-called conformally invariant field theories (CFTs) because they play a fundamental unifying t - role as being themselves integrable and describing the critical properties of lattice models. Hence, p e adopting this point of view, one can consider integrable QFTs obtained by adding a perturbation h which preserves the integrable structure of the conformal field theory. : v The relevant case, here, is the ϕ boundary perturbation of the c = 7/10 unitary minimal i 1,3 X model(it describesthe universality class of thetricritical Isingmodel(TIM)). Indeed,if this theory r is defined on a cylinder of finite length, the operator ϕ acting on the boundaries generates a set a 1,3 of integrable flows [1] between different sectors of the model. On the other hand, many results indicate that the A Andrews-Baxter-Forrester (ABF) inte- 4 grable lattice model [6] belongs to the same universality class of TIM. In particular, the boundary flows of TIM were extensively studied in [2, 3] using the continuum scaling limit of this model (see [4, 5] and references in [2] for the derivation of the TBA equations). Moreover, this correspondence has been made stronger by showing that the scaling limit of both the energy and the associated degeneracies of the lattice model exactly matches with the corresponding quantities given by the Virasoro characters of the c = 7/10 minimal unitary CFT [7]. The aim of this paper is to establish a lattice-conformal dictionary (the wording is borrowed from[8])betweenthelatticemodelandthe(possiblyperturbed)underlyingCFT.Inotherwordswe will find a correspondence between the states on the lattice, given by the eigenvalues/eigenvectors of the transfer matrix, and the states which belong to the Virasoro irreduciblemodules. In order to achieve thisresultwewilllookatasuitablefamilyofinvolutivelocalintegrals ofmotion (IM).From the point of view of CFT such integrals of motion can be constructed as appropriate polynomials of the Virasoro generators whose actual expression can be obtained in a constructive way (since ∗ [email protected] ∗∗[email protected] 1 a general expression is not known to exist), see [11]. Within the lattice model, such integrals of motion can be accessed by usual TBA techniques showing a remarkable correspondence between their eigenstates and those of the transfer matrix (a similar analysis was done in [12] by means of the nonlinear integral equations method in the context of KdV theory.). Therefore, as we will show in the sequel, the desired state-by-state correspondence can be found by comparing the eigenvalues of the integrals of motion in the CFT with those of the lattice model. In this sense, our approach is algebraic-analytic while in [13] it is combinatorial. Finally, onecanwonderaboutthefateofsuchintegralsofmotionwhenarelevantperturbation is switched on. In the present case, where the ϕ boundary perturbation is concerned, the answer 1,3 is that they can be chosen to be preserved, i.e. to remain involutive integrals of motion under such a perturbation. Indeed, among the possible families of involutive integrals of motion allowed in the CFT, we will consider that one whose spins are predicted to be preserved, for the present perturbation, by the counting argument in [14]. This implies that the perturbed theory is still integrable and it is remarkable that the eigenvalues of the integrals of motion can be followed along the Renormalization Group flow in close analogy to the usual TBA description of energies. As a final consideration, we also find a connection between our results and a recent series of papers by Bazhanov, Lukyanov and Zamolodchikov (BLZ) [16, 17]. We will see that, up to normalizations, the eigenvalues of the transfer matrix (and hence those of the integrals of motion) of the lattice model coincides with the eigenvalues of the so-called “T-operators” introduced in the above mentioned papers. The outline of the paper is as follows. In sect. 2 we describe the A ABF integrable lattice 4 model; in sect. 3 we carefully discuss both the continuum scaling limit of the lattice transfer matrix and the emergence of the integrals of motion in such a context; sect. 4 is devoted to a brief presentation of the integrals of motion in CFT. We give our results in sect. 5. The connection between our paper and the BLZ papers is elucidated in sect. 6, and finally we draw our conclusions in sect. 7. 2 A lattice model 4 This model was exhaustively discussed in [7, 2] and we summarize here the properties that will be useful later. For convenience, we will adopt the notations introduced in [2]. At the critical temperature, onacylinder of finitelength, itis definedby thefollowing doublerowtransfer matrix 1 σ1′ σ2′ σN′ −1 2 2 λ u λ u λ u DN(u,ξ)σ,σ′ ≡ DN1,1|2,1(u,ξlatt)σ,σ′ = 2 −τ1 −τ2 τN−1 −τN u,ξlatt τ1X,...,τN u u u 1 σ1 σ2 σN−1 2 2 i ξ = λ+ ξ, ξ R (2.1) latt − 5 ∈ which has been specialized to the boundary flow that we are going to examine: χ χ (note 1,2 1,1 7→ that it is a constant r flow). The left boundary is kept fixed and the boundary interaction, shown with dashed lines, acts on the right boundary only and it is tuned by the parameter ξ. The heights σ ,σ ,τ 1,2,3,4 j j′ j ∈ { } on adjacent vertexes are constrained to follow the A adjacency rule which also forces the number 4 N of faces on a row to be odd. In the following we will always restrict to this case, in particular 2 when the limit lim will be considered. The bulk weights are fixed to their critical values N →∞ d c d c sin(λ u) sinu S S a c W a b u = u = sin−λ δa,c+ sinλ S S δb,d (2.2) (cid:18) (cid:19) r b d a b where the physical range of the spectral parameter is 0 < u < λ, and λ = π is the crossing 5 parameter. The crossing factors are defined as S = sinaλ/sinλ. a Among the general expressions for integrable boundaries given in [9], we only need to consider the following case: 2 2 S sin(ξ u) sin(2λ+ξ u) B2,1 2±1 2 u, ξlatt = 2±1 u, ξlatt = S2±1 lasttin±λ cosh2Im(ξ la)tt∓ (cid:18) (cid:19) r 2 latt 2 2 1 u − = 2 1 −ξlatt 2 √2cosλ −sinλ . (2.3) ± cos2Im(ξ ) u λ latt − −ξlatt 2 1 The second line of the equation shows that the same boundary interaction can be obtained by adding a single column to the right of the lattice; the faces of this column depend upon ξ but the new right boundary is fixed to the sequence 1,2,1,.... Let us now consider the right boundary of the lattice defined in (2.1). We can see that, for ξ = it is formed by the quasi-free sequence 2,2 1,2,2 1,..., while for ξ = 0 it is formed −∞ ± ± by the fixed sequence 1,2,1,2,.... In the continuum scaling limit, the former case corresponds to the ultraviolet (UV) critical point χ (q) and the latter to the infrared (IR) critical point χ (q) 1,2 1,1 of the boundary renormalization group flow. Byconstruction,thegivenbulkandboundaryweightssatisfytheYang-Baxter andBoundaryYang- Baxter equations leading to an integrable model. Consequently, the double row transfer matrix forms a one parameter family of commuting operators: [DN(u,ξ),DN(u,ξ)] = 0 (2.4) ′ for arbitrary complex values of the spectral parameters u,u. An important symmetry becomes ′ apparent when one defines the following normalized transfer matrix 2N sin(u+2λ)sinλ tN(u,ξ) = DN(u,ξ)S (u,ξ )S(u)√2cosλ (2.5) 2 latt sin(u+3λ)sin(u+λ) (cid:20) (cid:21) where sin2(2u λ) S(u) = − , (2.6) sin(2u+λ)sin(2u 3λ) − sinλ sin(u ξ +λ)sin(u+ξ +3λ) cosh2Im(ξ ) latt latt latt S (u,ξ ) = − − . (2.7) 2 latt sin(u ξ )sin(u ξ +2λ)sin(u+ξ λ)sin(u+ξ +2λ) latt latt latt latt − − − Indeed, one can show that the following functional equation is satisfied: tN(u,ξ)tN(u+λ,ξ) = 1+tN(u+3λ,ξ). (2.8) 3 0 λ 2λ 3λ 4λ first strip second strip Fig. 1. Schematic example of zeros of the eigenvalue of the transfer matrix labeled (21100010). Here: | m1 =6, m2 =2, n1 =3, n2 =1. The normalized transfer matrix t also forms a family of commuting operators [tN(u,ξ),tN(u,ξ)] = 0 (2.9) ′ so that the eigenstates are independent of u, and the functional equation holds true also for the eigenvalues. We would like to stress that it is a crucial point since the TBA approach is precisely based on such an evidence. In the continuum scaling limit, the large N corrections to the eigenvalues1 of the double row transfer matrix are related to the excitation energies of the associated perturbed conformal field theory by 1 2πsin5u 1 logDN(u,ξ logN) = Nf (u)+f (u,ξ)+ E(ξ)+o( ) (2.10) b bd −2 − N N where f is the bulk free energy, f is the surface (i.e. boundary dependent) free energy and E(ξ) b bd is a scaling function given by (3.20). Since both the bulk and surface free energy contributions are the same for all the eigenvalues, they can be removed from the previous expression by subtracting the largest eigenvalue of DN (it plays the role of a hamiltonian ground state). At the boundary critical points, the scaling energy E(ξ) reduces to c +n, IR: ξ + −24 → ∞ E(ξ) = (2.11) ( c + 1 +n, UV: ξ −24 10 ′ → −∞ where n,n N are certain excitation levels. ′ ∈ Thederivation of both theTBAequations andthe scaling energy (2.11) is obtained by looking at the analytic structure of the eigenvalues of the transfer matrix: DN(u,ξ) is an entire function of u characterized by simple zeros only, moreover it is periodic on the real axis u u+π. In the ≡ large N limit, the zeros areorganized in two sortof strings and theyare distributedin two different strips, defined by 1λ 6 Re(u) 6 3λ and 2λ 6 Re(u) 6 4λ respectively. An example of such a −2 2 situation is schematically shown in Fig. 1. The zeros are bound either to appear in the center of each strip (located at Re(u) = λ or 3λ) and 2 we will refer to them as 1-strings, or they can appear in pairs (u,u) with the same imaginary part ′ Im(u) = Im(u) on the two edges of a strip, and we will call them 2-strings. In order to label the ′ number of such strings in the upper-half plane, we will use m and m for 1-strings located in 1 2 1Weuse D, t to indicate eigenvalues of D, t. 4 the first and second strip respectively; analogously n and n will be used for 2-strings2. These 1 2 numbers form the so-called (m,n)-system and are constrained to be n = N+m2 m > 0 1 2 − 1 m ,m even. (2.12) 1 2 n2 = m21 −m2 > 0 )⇒ (j) The imaginary part of the position of the 1-string is indicated by v k 1 first strip: DN( λ+iv(1),ξ) =0, second strip: DN(3λ+iv(2),ξ) = 0; (2.13) 2 k k (j) (j) where the k’s are ordered from the bottom to the top in the upper half plane, 0 < v < v < 1 2 (j) (j) ... < v . Analogously, w is used for the 2-strings. m1 k Finally, the states (eigenvalues-eigenstates) both on the lattice and in the continuum scaling (j) limit are uniquely characterized by the non-negative quantum numbers I . Their topological k (j) meaning is that when we refer to the k-th 1-string in strip j, I is the number of 2-strings with k larger imaginary part. The notation which indicates the content of zeros works as in the following example as shown in Fig. 1: (21100010) is a state with 6 zeros in the first strip and 2 in the | (1) (1) second; their quantum numbers are respectively I = 2, I = 1,.... In the sector (1,2) a 1 2 frozen zero can occur [7], so a parity σ = 1 is added to the previous notation (namely ( ) , see ± ·|· ± Tables 4 and 5). 3 Continuous transfer matrix from the lattice Since our task is to compare lattice computations with CFT results, we have to define a suitable continuumlimitoflatticequantities. Inthefollowingwewillfirstdefinethecontinuumscalinglimit for the lattice transfer matrix, and then the corresponding integrals of motion will be introduced. 3.1 Continuum scaling limit of the transfer matrix In [2] the finitesize (order 1/N) corrections to the lattice energies (scaling energies) were computed with standard TBA techniques that involve a continuum scaling limit on the lattice model. This becomes fully apparent when the bulk of the lattice is off the critical temperature, t = 0 (t = 6 (T T )/T ), and the continuum limit (N ,a 0, a is the lattice spacing) is sensible only c c − → ∞ → if the temperature is also scaled, t 0 as N t ν = µ = const. (ν = 5/4) [10]. Physically, this → | | corresponds to a huge magnification of the critical region. The bulk critical point itself is reached when the dimensionless mass µ goes to zero: µ 0. In the present case, the bulk is at the critical → temperature (µ = 0) and we use the boundary perturbation to move away from the boundary critical points. The aim of this section is to show how to define a meaningful continuum scaling limit of the lattice transfer matrix. Let us recall that in [2], from the derivation of the TBA equations, it was shown that the following scaling property for the normalized transfer matrix holds i ˆt(u,ξ) = lim tN(u+ logN,ξ logN) (3.1) N 5 − →∞ and hence a proper definition of the continuous transfer matrix (i.e., a transfer matrix for the continuum theory) is Dˆ(u,ξ) = lim DN(u+ i logN,ξ logN) 2sinλei(u−λ2) 2N (3.2) N 5 − " N1/5 # →∞ 2We do not need to keep track of the lower half plane because it is the mirror image of the upper half plane, thanksto thecomplex conjugation symmetry of the transfer matrix. 5 where we can explicitly observe the periodicity property Dˆ(u,ξ) = Dˆ(u+π,ξ). Since DN is real in the center of each strip (as shown in [2]) and the square of the factor in brackets is real for Re(u) = λ/2, 3λ, we deduce that also Dˆ is real in the center of each strip. Themainmotivationtointroducethepreviousdefinitionistohaveanobjectforthecontinuum theory with the same analytic structure of the lattice transfer matrix. Hence, let us analyze such a structure starting from the string content of the eigenvalues of the double-row transfer matrix. A given eigenvalue DN(u,ξ) is characterized by its content of 1-strings. Itis important tonotice that, increasing N, the number of 1-strings, of 2-strings in the second strip and the relative position of 1- and 2-strings does not change. Instead, it is observed the appearance of new 2-strings close to the real axis in the first strip, moreover eq. (2.12) tells us that n grows as N/2 (notice that N 1 grows in steps of 2 so that n grows in steps of 1). Such new 2-strings push away the remaining 1 zeros from the real axis and since the extension of the region occupied by them grows as logN, (j) then the following limit is finite (v is a function of N): k (j) def (j) y = lim (5v logN). (3.3) k N k − →∞ (j) In terms of the scaled coordinate y , the real axis located at Im(u) = 0 is shifted to i . From k − ∞ (3.2) and (3.3) one is immediately led to the following observation: 1 i i 1 i lim DN( λ+ (5v(1) logN)+ logN,ξ logN) = 0 = Dˆ( λ+ y(1),ξ) (3.4) N 2 5 k − 5 − 2 5 k →∞ (where a similar equation holds for the zeros in the second strip). The remarkable consequence of such an equation is that the content of zeros of an eigenvalue survives the continuum limit, the only difference being the infinite number of 2-strings that appear in the first strip, for u i , → − ∞ because of the growth of n N/2. 1 ∼ Inordertoshowthatthelimitin(3.2)isindeedmeaningful,inthesensethatitisfinite,wehave to take into account the following ingredients. Firstly, the derivation of the TBA equations shows the existence of the limit in (3.1); secondly, examining the normalization coefficients S (u,ξ), S(u) 2 in eq. (2.5) we have i i Sˆ (u,ξ) = lim S (u+ logN, λ+ (ξ logN)) (3.5) 2 2 N 5 − 5 − →∞ sinλ sin(u+ iξ+2λ) = − 5 ei(u−12λ) sin(u+ iξ 2λ)sin(u+ iξ+λ) 5 − 5 i Sˆ(u) = lim S(u+ logN)= 1 (3.6) N 5 →∞ which are finite quantities. The factor in square brackets in (2.5) gives: sin(u+ 5i logN +2λ)sinλ 2N 2sinλei(u−λ2) 2N (3.7) "sin(u+ i logN +3λ)sin(u+ i logN +λ)# N∼ " N1/5 # 5 5 →∞ that exactly cancels the corresponding factor in (3.2). Therefore we get the equation ˆt(u,ξ) =Dˆ(u,ξ)Sˆ (u,ξ) (3.8) 2 which shows that Dˆ(u,ξ) is finite because ˆt(u,ξ) and Sˆ are finite. It also shows that, in the 2 continuum theory, ˆt and Dˆ are almost equivalent, the difference being in zeros and poles explicitly dependent upon ξ that are introduced in ˆt by the factor Sˆ . On the other hand, the lattice tN 2 possesses some non physical zeros and poles of order 2N contained in the square bracket term of 6 (2.5). Moreover, for a generic A ABF model, the zeros have to be understood as zeros of DN p while their occurrence in tN is much more involved. Finally, it can be shown that the functional equation (2.8) keeps the same form even after the continuum scaling limit ˆt(u,ξ)ˆt(u+λ,ξ) = 1+ˆt(u+3λ,ξ). (3.9) This equation can besolved for the eigenvalues ofˆt and the solution is given by the TBA equations obtained in [2]. Itisimportanttonoticethatthenormalization oftN(u,ξ)isuniquelyfixedbytherequirement that the functional equation holds in the form (2.8), namely no multiplicative or additive constants are allowed (in particular, this fixes the relation (2.5) between DN(u,ξ) and tN(u,ξ)). Sincesuch a functional equation remains unchanged after the continuum scaling limit, the lattice normalization uniquelyfixesthecontinuumone. Thisfactwillhaveimportantconsequencesforacorrectdefinition of the integrals of motion. 3.2 TBA equations and integrals of motion Let us discuss the TBA equations for the eigenvalues of ˆt. First of all, we define the functions tˆ (x,ξ) and tˆ (x,ξ) which are the eigenvalues of ˆt referred to the center of each strip: 1 2 tˆ (x,ξ) = tˆ(λ + ix,ξ) = s e ǫ1(x), (3.10) 1 1 − 2 5 tˆ (x,ξ) = tˆ(3λ+ ix,ξ) = s e ǫ2(x). (3.11) 2 2 − 5 where the pseudoenergies ǫ (x) are introduced, and the quantities s = 1 play the role of inte- j j ± gration constants and will be fixed later. Since the functions tˆ(x) are real for real x and satisfy j tˆ(x) > 1, we can define the real functions L (x) as j j − L (x) = log(1+tˆ(x,ξ)). (3.12) j j In terms of the functions tˆ and tˆ , the functional equation (3.9) becomes 1 2 π π tˆ (x+i )tˆ (x i ) = 1+tˆ (x) 1 1 2 2 − 2 π π tˆ (x+i )tˆ (x i ) = 1+tˆ (x). (3.13) 2 2 1 2 − 2 Consequently the TBA equations can be written as m1 y(1) x + dy L (y) tˆ (x,ξ) = s gˆ (x,ξ) tanh k − exp ∞ 2 , (3.14) 1 1 1 2 2π cosh(x y) kY=1 (cid:18)Z−∞ − (cid:19) tˆ (x,ξ) = e 4e−xs gˆ (x,ξ) m2 tanh yk(2)−x exp +∞ dy L1(y) (3.15) 2 − 2 2 2 2π cosh(x y) kY=1 (cid:18)Z−∞ − (cid:19) and the positions of the zeros are fixed by the quantization conditions (note the inversion of the indices: ψ is for strip 2 and ψ for strip 1): 1 2 (1) (1) (1) ψ (y ) = n π = 2(I +m k)+1 m π, (3.16) 2 k k k 1− − 2 (cid:16) (cid:17) (2) (2) (2) ψ (y ) = n π = 2(I +m k)+1 m π. (3.17) 1 k k k 2− − 1 (cid:16) (cid:17) 7 (j) The same equations hold for the 2-string locations z so, each time ψ (x) = nπ is satisfied for an l j integer n with the appropriate parity3, x is a 1-string or a 2-string in the strip 3 j. The explicit − expressions for ψ are given by: j π π ψ (x) Re iǫ (x i ) = Re ilog(s tˆ (x i ,ξ)) (3.18) 1 1 1 1 ≡ − − 2 − 2 (cid:16) π (cid:17) m1 (cid:16) y(1) x π(cid:17) + dy L (y) = iloggˆ (x i ,ξ)+i logtanh k − +i ∞ 2 , 1 − 2 2 4 −PV 2π sinh(x y) π Xk=1 (cid:0) π (cid:1) Z−∞ − ψ (x) Re iǫ (x i ) = Re ilog(s tˆ (x i ,ξ)) (3.19) 2 2 2 2 ≡ − − 2 − 2 (cid:16) (cid:17) π (cid:16) m2 y(2) (cid:17)x π + dy L (y) = 4e x +iloggˆ (x i ,ξ)+i logtanh k − +i ∞ 1 − 2 − 2 2 4 −PV 2π sinh(x y) Xk=1 (cid:0) (cid:1) Z−∞ − where the integral around the singularity x = y is understood as a principal value. Moreover, for numerical convenience, we take the fundamental branch for each logarithm so that in general loga+logb cannot be identified with log(ab). The energy predicted for each state is given by: E(ξ) = π2 m1 e−yk(1) − ∞ dπy2 e−yL2(y) (3.20) Xk=1 Z−∞ and reduces to (2.11) at both the critical points (compare also with (2.10)). Since we are interested in the flow χ χ , we have 1,2 1,1 7→ s = s = 1; 1 2 x+ξ (3.21) gˆ (x,ξ) = 1; gˆ (x,ξ) = tanh . 1 2 2 It is useful to recall that the functions tˆ(x,ξ) are eigenvalues of a commuting family of operators: j [ˆt(u,ξ),ˆt(u,ξ)] = 0. (3.22) ′ Hence we are in the position to show how the local integrals of motion emerge from the continuum scaling limit of the transfer matrix ˆt(u,ξ). First of all we expand the quantity logtˆ for x 1 → −∞ where tˆ is given by (3.14). In order to do so we need the following formulæ 1 logtanh yk(1) −x = log 1−e−yk(1)+x 2 1+e−yk(1)+x 1 t t3 t5 log − = 2 t+ + +... , t < 1 1+t − 3 5 | | (cid:16) (cid:17) 1 ∞ = 2 ( 1)n 1e(2n 1)x, x < 0. − − coshx − n=1 X Hence, taking advantage of the previous results, we arrive to the following expressions for the continuum scaling limit of the logarithm of the transfer matrix ∞ logtˆ (x,ξ) = C I (ξ)e(2n 1)x, (3.23) 1 n 2n 1 − − − n=1 X CnI2n 1(ξ) = 2 m1 e−(2n−1)yk(1) +( 1)n +∞ dye−(2n−1)yL2(y), (3.24) − 2n 1 − π − Xk=1 Z−∞ 3There is the following constraint: n(j)+ s3−j+1 =0 mod 2. k 2 8 where the quantities4 I are the eigenvalues of the infinite family of integrals of motion in 2n 1 involution I . It is−important to stress that the TBA calculation provides only the value of 2n 1 n { − } the product C I (ξ) as it is clear from eq. (3.23) and (3.24). The coefficients C will be fixed n 2n 1 n − later by imposing appropriate normalization conditions (see sect. 5). As pointed out in sect. 3.1, the normalization of tˆ is fixed once forever on the lattice and 1 survives the continuum scaling limit hence its largest eigenvalue, being the lattice groundstate, in that limit is directly mapped to the true vacuum of the continuum field theory. This forbids the presence of additive constants to I in (3.23). Such a thing is in agreement with the general 2n 1 − properties one expects for the above integrals of motion in the continuum limit: since they are given by integrals over local densities, and the local densities are components of Lorentz vectors, there is no freedom to add any arbitrary scalar contribution to them. Hence we expect to find, at the fixed points of the massless flow, the same values for I given by the conformal field theory 2n 1 − (see next section) once the proper normalization for the corresponding C has been chosen. n As an example, it is interesting to note that comparing (3.24) with (3.20) we obtain C = π 1 and I (ξ) = E(ξ): as expected, the first integral of motion is the energy. 1 Now, our purpose is to use the TBA equations to compute the eigenvalues of I and 2n 1 − compare them with their analog from the conformal field theory. As we will see, the two sets of integrals of motion precisely correspond to one another. 4 Integrals of motion in conformal field theory Conformal field theories are integrable, in the sense that they possesses an infinite number of mutually commuting integrals of motion. Their construction was given in [11] by canonically quantizing the family of classical integrals of motion of the Modified Korteweg–de Vries equation (that also coincide with the integrals of motion of the classical sine-Gordon equation). A general expression for such integrals of motion is not known but they can all be obtained in a constructive way. They are local expressions (polynomials) in the Virasoro generators and the first few of them are given by5 (see [16] for their explicit expressions) c I = L 1 0 − 24 ∞ 2+c c(5c+22) I = 2 L L +L2 L + (4.1) 3 −n n 0− 12 0 2880 n=1 X 3 ∞ I = :L L L :δ + L L 5 n m p 0,m+n+p 1 2n 2n 1 2 − − m,n,p n=1 X X ∞ 11+c c 4+c + n2 1 L L L2 6 − 4 − −n n− 8 0 n=1(cid:18) (cid:19) X (2+c)(20+3c) c(3c+14)(7c+68) + L 0 576 − 290304 (the symbol : : indicates normal order, obtained by arranging the operators L in an increasing n sequence with respect to their indices). Their simultaneous diagonalization is, in principle, a straightforward but lengthy calculation of linear algebra. Actually, the rapidly growing complexity of the involved expressions makes the 4We hope that the symbols I(j), with j = 1,2 introduced for the quantum numbers, are not confused with the k similar symbols I2n−1 used for theintegrals of motion. 5It is convenientto indicate each integral of motion with I2n−1 and the corresponding eigenvalues with I2n−1. 9 computation possible for the first few levels only. This becomes clear if we look at the characters of the Virasoro algebra: χr,s(q) = q−2c4 Tr qL0 where the trace is taken in the irreducible module of the highest weight state (r,s). For example, for the TIM one has: χ1,1(q) = q−2c4(1+q2+q3+2q4+2q5+4q6+4q7+7q8+...) (4.2) χ1,2(q) = q−2c4+∆1,2(1+q+q2+2q3+3q4+4q5+6q6+8q7+...). (4.3) In the previous expression, the coefficient α in the monomial α qn gives the degeneracy of the n n subspace at level n. At a given level of truncation, the expression for the vacuum sector χ (q) 1,1 holds for all the unitary minimal models except the c = 1/2 Ising model. Up to level 3 there is no degeneracy and the eigenstate is given by the unique linearly independent vector. At level 4 and 5 the double degeneracy leads to an algebraic quadratic equation that can be easily solved. At level 6 the degeneracy of order 4 leads to a quartic algebraic equation and, in general, one realizes that the complexity in managing this problem rapidly grows with the degeneracy of the level. More details on this topic can be found in Tables A1 and A2 of the Appendix where we present a list of eigenstates/eigenvalues for the vacuum sector of all unitary minimal models (Table A1) and for the sector (1,2) of TIM (Table A2). From Tables A1 and A2, we notice that I alone is enough to 3 completely remove the degeneracy, at least up to level 6. The list of integrals of motion discussed here survives the off critical perturbation by ϕ [14]. 1,3 More precisely, off critical involutive integrals of motion I (λ) exist for the perturbed CFT6 2n 1 − +λϕ and at criticality they reduce to the expression (4.1). p,q 1,3 M 5 TBA data versus CFT As anticipated in the introduction, we are interested in studying the model both at criticality and along the boundary flow χ χ generated by the ϕ integrable perturbation of TIM. 1,2 1,1 1,3 7→ Let us first describe the sector (1,1) of the theory, i.e. the IR critical point. Let us recall that TBA computations provide the product C I (ξ), see eq. (3.24). The constants C are precisely n 2n 1 n − introduced to make contact with the integrals of motion of the CFT. As already pointed out at the end of sect. 3.2, the first of them is exactly know, C = π. Since we are interested in I and I we 1 3 5 can numerically fix the corresponding constants C and C using the vacuum state7 2 3 C = 2.1838434, C = 3.7555032. (5.1) 2 3 The general expression was computed in [16] and reads8: (10n 7)!! Cn = 22−n 31−2n 51−n − π. (5.2) n!(4n 2)! − In particular, this gives: C = π, C = 1001 π 2.1838432 and C = 7436429 π 3.7555026. 1 2 1440 ∼ 3 6220800 ∼ After that, we are ready to turn the crank of TBA equations in order to obtain the other eigenval- ues of the operators I and I . The comparison between them and the corresponding eigenvalues 3 5 coming from CFT can be found in Table 1. We can immediately notice that the agreement is quite remarkable: the solutions of the TBA equations are given with seven significant digits of precision and perfectly match the CFT results. 6With Mp,q we refer to theminimal CFT with central charge c=1− 6(pp−qq)2. 7Actually, onereaches higher numerical precision using the first excited state and so we did. 8Wewould like to thanksthereferee of Nucl. Phys. B for having pointed out to usthis equation. 10