PreprinttypesetinJHEPstyle-HYPERVERSION = 1 c-map as c string 2 1 0 2 n a Sergei Alexandrov J 0 2 Universit´e Montpellier 2 & CNRS, Laboratoire Charles Coulomb UMR 5221, F-34095, Montpellier, France ] h t e-mail: [email protected] - p e h Abstract: We show the existence of a duality between the c-map space describing the [ universal hypermultiplet at tree level and the matrix model description of two-dimensional 1 v string theory compactified at a self-dual radius and perturbed by a sine-Liouville potential. 2 It appears as a particular case of a general relation between the twistor description of four- 9 3 dimensional quaternionic geometries and the Lax formalism for Toda hierarchy. Furthermore, 4 . we give an evidence that the instanton corrections to the c-map metric coming from NS5- 1 0 branes can be encoded into the Baker-Akhiezer function of the integrable hierarchy. 2 1 : v i X r a Contents 1. Introduction 1 2. Toda lattice hierarchy 2 2.1 Lax formalism 3 2.2 String equations 4 2.3 Dispersionless limit 5 2.4 Example: MQM with sine-Liouville perturbation 6 3. Four-dimensional QK spaces and integrability 7 3.1 Toda ansatz 7 3.2 Twistor space and Lax formalism 8 4. c-map vs c = 1 string 10 4.1 Universal hypermultiplet at tree level 10 4.2 Duality 11 5. NS5-branes and the Baker-Akhiezer function 11 5.1 Inclusion of the one-loop correction 14 6. Discussion 14 1. Introduction Integrability isavery powerful tooltofindexact solutionsanditis always agreatachievement when a theory is proven to possess an integrable structure. A particularly interesting class of theories where various appearances of integrability have been observed during the last few years is characterized by the presence of N = 2 supersymmetry. Remarkably, it appears both in the gauge [1, 2, 3, 4, 5] and in the string theory context [6, 7]. However, an exhaustive explanation of these observations is missing so far. In this note we reveal one more instance of an integrable structure in theories with N = 2 supersymmetry. Namely, we observe that the four-dimensional quaternion-Ka¨hler (QK) space obtained by the local c-map construction [8, 9] is in a certain sense dual to a sine-Liouville perturbation of the c = 1 string theory compactified at the self-dual radius. The former represents the tree level moduli space of the so called universal hypermultiplet [10] arising in compactifications of type II strings on a rigid Calabi-Yau. On the other hand, the latter is known to be an exactly solvable theory which is described by Toda integrable hierarchy [11, 12]. In particular, the lowest equation of this hierarchy is given by the Toda equation, which is also known to parametrize four-dimensional quaternionic metrics with one isometry – 1 – [13]. The duality which we report here is based on the observation that the two systems are described by the same solution of the Toda equation. However, the duality goes much further as it allows to identify also various quantities characterizing either the c-map or the c = 1 string. These identifications originate in a general correspondence between a twistor description of quaternionic manifolds and the Lax formalism for integrable hierarchies [14]. The duality between the c-map and the c = 1 string is just its simplest example. The above mentioned correspondence is not at all surprising. First of all, the existence of integrable hierarchies governing the four-dimensional hyperka¨hler (HK) geometry is very well known (see, for instance, [15, 16, 17]). Moreover, recently it was realized that hyperka¨hler manifolds with a rotational isometry endowed with a hyperholomorphic line bundle can be associated with QK manifolds of the same dimension also possessing an isometry [18, 19]. In particular, this QK/HK correspondence identifies the rigid and local c-map spaces and makes a contact between QK geometry and integrability. Although, this might seem to imply that the duality we propose is rather trivial from the mathematical point of view, its physical realization is quite fascinating as it suggests that the hypermultiplet sector of N = 2 superstring compactifications and non-critical c = 1 string theory are closely related to each other. We are not aware of any examples where some perturbation of two-dimensional string theory was related to the hypermultiplet moduli space. Of course, this duality would become really interesting if it can be extended beyond the classical limit and used to extract quantum corrections on one side from the known results on the other side. We leave the systematic study of this problem for the future. Nevertheless, in this note we make the first step in this direction. Namely, we observe that the holomorphic functions on the twistor space of the c-map, which have been shown to generate NS5-brane instanton corrections to the tree level universal hypermultiplet, can be interpreted in terms of the Baker-Akhiezer wave function of the Toda hierarchy. Together with previous findings [3, 7], this suggests that the whole non-perturbative picture describing N = 2 string compactifications possesses an integrable structure where the relations discussed here will find their natural place. 2. Toda lattice hierarchy We start from a brief review of two-dimensional Toda lattice hierarchy [20] which provides the general framework for what we are going to discuss. This integrable hierarchy describes many physical phenomenawhich, inparticular, includeacertainclassofdeformationsofc = 1string theory. Moreprecisely, it encompasses all perturbationsgenerated by thespectrum oftachyon operators corresponding to the c = 1 string compactified on a circle [11, 12]. Furthermore, the Lax formalism of this hierarchy turns out be closely related to the matrix model description of the c = 1 string, which is provided by Matrix Quantum Mechanics (MQM). Therefore, first we present this formalism, then we explain its interpretation from the MQM point of view, and finally we give a particular solution of the Toda hierarchy which describes c = 1 string theory with the so-called sine-Liouville perturbation. Although we do not need the Toda hierarchy in its full generality to explain the relation with the c-map spaces, it is nevertheless useful to recall it as we want to discuss further extensions and generalizations of the proposed duality. – 2 – 2.1 Lax formalism Let us first give a formal definition of the Toda lattice hierarchy. To this end, introduce two semi-infinite series ∞ ∞ L = r(s)ωˆ + u (s)ωˆ k, L¯ = ωˆ 1r(s)+ ωˆku¯ (s), (2.1) k − − k X X k=0 k=0 where s is a discrete variable labeling the nodes of an infinite lattice, ωˆ = e~∂/∂s is the shift operator in s, and the Planck constant ~ plays the role of a spacing parameter. The operators (2.1) are called Lax operators. Their coefficients r, u and u¯ are taken to be functions of k k two infinite sets of “times” t . To each time variable one associates a Hamiltonian H { ±k}∞k=1 ±k generating an evolution along t by the usual rule k ± ~∂L = [H ,L], ~∂L¯ = [H ,L¯], ∂tk k ∂tk k (2.2) ~ ∂L = [H ,L], ~ ∂L¯ = [H ,L¯]. ∂t−k −k ∂t−k −k This system represents the Toda hierarchy if the Hamiltonians satisfy an additional require- ment, namely that they can be expressed through the Lax operators (2.1) as follows 1 1 H = (Lk) + (Lk) , H = (L¯k) + (L¯k) , (2.3) k > 0 k < 0 2 − 2 where the symbol ( ) means the positive (negative) part of the series in the shift operator ωˆ > < and ( ) denotes the constant part. The integrability follows from the existence of an infinite 0 set of commuting flows generated by ~ ∂ H whose commutativity isa consequence of (2.2). ∂tk− k This definition shows that the Toda hierarchy is a collection of non-linear equations of the finite-difference type in s and differential with respect to t for the coefficients r(s,t), u (s,t) k k and u¯ (s,t). Its hierarchic structure is reflected in the fact that one obtains a closed equation k on the first coefficient r(s,t) and its solution provides the input information for the following equations. It is easy to show that r(s,t) should satisfy a discrete version of the Toda equation ∂2logr2(s) ~2 = 2r2(s) r2(s+~) r2(s ~), (2.4) ∂t ∂t − − − 1 1 − which gives the name to the hierarchy. For our purposes it is important to introduce the following Orlov-Shulman operators [21] M = kt Lk +s+ v L k, M¯ = kt L¯k +s v L¯ k. (2.5) k k − k k − − − − − X X X X k 1 k 1 k 1 k 1 ≥ ≥ ≥ ≥ The coefficients v (s,t) are supposed to be found from the condition on the commutators of k M and M¯ with th±e Lax operators [L,M] = ~L, [L¯,M¯] = ~L¯. (2.6) − The importance of the Orlov-Shulman operators comes from the fact that they can be con- sidered as perturbations of the simple operators of multiplication by the discrete variable s, whereas the Lax operators appear to be a “dressed” version of the shift operator ωˆ. Indeed, if one requires that v vanish when all t = 0, then in this limit M = M¯ = s, L = ωˆ and k k ± ± – 3 – L¯ = ωˆ 1. Moreover, the commutation relations (2.6) can be recognized as a dressed version − of the trivial relation [ωˆ,s] = ~ωˆ. (2.7) Another application of the Orlov–Shulman operators is that they lead directly to the notion of τ-function. Indeed, one can show that their coefficients must satisfy [22] ∂v ∂v k l = . (2.8) ∂t ∂t l k This implies that there exists a generating function τ [t] of v such that s k ± ∂logτ [t] v (s,t) = ~2 s . (2.9) k ∂t k It is called the τ-function of Toda hierarchy. It encodes all information about a particular solution. For example, it allows to extract the first coefficient in the expansion of the Lax operators r2(s ~) = τs+~τs−~. (2.10) − τ2 s Plugging this relation into (2.4), one obtains that the τ-function is subject to the Toda equation. In physical applications the τ-function usually coincides with partition functions andthecoefficients v aretheone-point correlatorsoftheoperatorsgeneratingthecommuting k flows H . k Finally, note that the Toda hierarchy can be equivalently represented as an eigenvalue problem given by the following equations ∂Ψ ∂Ψ xΨ = LΨ, ~x = MΨ, ~ = H Ψ, (2.11) k ∂x ∂t k ¯ ¯ ¯ plus similar equations for Ψ where L,M should be replaced by L, M. The eigenfunction − Ψ(x;s,t)isknownastheBaker-Akhiezerfunction. Oftenithasadirectphysical interpretation and plays a very important role. 2.2 String equations The Toda hierarchy provides a universal description for many integrable systems because the equations of the hierarchy have many different solutions. There are two ways to select a particular solution. The most obvious one is to provide an initial condition which can be, for instance, the τ-function for vanishing times corresponding physically to the partition function of a non-perturbed system. However, the Toda equations involve partial differential equations of high orders and require to know not only the τ-function at t = 0 but also its derivatives. k ± Therefore, it is not always clear whether the non-perturbed function is sufficient to restore the full τ-function. Fortunately, there is another way to select a unique solution of the Toda hierarchy. It relies on the use of the so-called string equations which are some algebraic relations between the Lax and Orlov–Shulman operators. The advantage of the string equations is that they represent, in a sense, already a partially integrated version of the hierarchy equations. For example, instead of solving a second order differential equation on the τ-function, they reduce the problem to certain algebraic and first order differential equations. – 4 – It is important, however, that the string equations cannot be chosen arbitrary because they should be consistent with the commutation relations (2.6). For example, if they are given by two equations of the following type L¯ = f(L,M), M¯ = g(L,M), (2.12) the operators defined by the functions f and g must satisfy [f(ωˆ,s),g(ωˆ,s)] = ~f. (2.13) − 2.3 Dispersionless limit In this paper we are mostly interested only in the classical limit of the Toda hierarchy, which is obtained by taking the spacing parameter ~ to zero. In this limit one recovers what is known as dispersionless Toda hierarchy [22]. It is described by the same equations as above with the only difference that all commutators should be replaced by Poisson brackets and all operators are considered now as functions on the phase space parametrized by s and ω with the symplectic structure induced from (2.7) ω,s = ω. (2.14) { } As in the full theory, a solution of the dispersionless Toda hierarchy is completely char- acterized by a dispersionless τ-function. In fact, it is better to talk about free energy since it is the logarithm of the full τ-function that can be represented as a series in ~ logτ = ~ 2+2nF . (2.15) − n X n 0 ≥ The dispersionless limit is described by the first coefficient of this series, the dispersionless free energy F . It satisfies the classical limit of the Toda equation 0 ∂ ∂ F +e∂s2F0 = 0, (2.16) t1 t−1 0 and can be selected from all solutions by the same string equations (2.12) as in the quantum case. The quasiclassical asymptotics of the Baker-Akhiezer function is given by the usual WKB approximation. Since Ψdiagonalizes theLaxoperator, itcanbeconsidered asa wave function representing the quantum state of the system in the L-representation. The conjugate variable is provided by M which implies that 1 logx Ψ exp MdlogL , (2.17) ∼ (cid:26)~ Z (cid:27) where M is considered as a function of L, s and t . Using (2.5), one finds k 1 v Ψ = e~1S+O(~0), S = tkxk +slogx+ φ n x−n. (2.18) 2 − n X X k 1 k 1 ≥ ≥ The integration constant φ can be related to the free energy as φ(s,t) = ∂ F . s 0 − – 5 – 2.4 Example: MQM with sine-Liouville perturbation Two-dimensional or non-critical c = 1 string theory is known to be an example of integrable modeldescribedbytheTodahierarchy. However, thisisdifficulttoseeinitsCFTformulation. On the other hand, the integrability becomes transparent in its matrix model description in terms of Matrix Quantum Mechanics (see [23, 24] for a review). To see how it appears, one should restrict to the singlet sector of MQM where all angular degrees of freedom are integrated out. The remaining matrix eigenvalues describe a system of free one-dimensional fermions in the inverse oscillator potential. The target space description of string theory arises as an effective theory of collective excitations of the free fermions. In particular, different backgrounds correspond to different states of the Fermi sea: the simplest linear dilaton background is dual to the ground state, whereas various perturbed backgrounds with a non-trivial condensate of the tachyon field come from ceratin time-dependent states. The most convenient way to describe the dynamics of these free fermions is to use the light-cone coordinates [12] x = x p where (x,p) are the coordinates on the phase space of ± ± √2 one fermion. It is clear that x satisfy the canonical commutation relations ± x ,x = 1. (2.19) { + −} In the quasiclassical approximation the problem reduces to finding the exact form of the profile of the Fermi sea in the phase space, given its asymptotic form at x . Here we ± → ∞ are interested in deformations generated by the spectrum of the theory compactified on a circle of radius R. Then the profile of the Fermi sea is determined by the compatibility of the following two equations 1 1 x x = M (x ) = kt xk/R +µ v x k/R. (2.20) + − ± ± ±R ±k ± ± R ±k −± X X k 1 k 1 ≥ ≥ Here xk/R correspond to the tachyon vertex operators with momentum k/R, t are their k ± ± coupling constants, µ is the Fermi level, and v are to be found as functions of t and k k ± ± µ. For vanishing coupling constants, the equations (2.20) describe the hyperbola x x = µ + − which is the trajectory of a free fermion in the inverse oscillator potential. Comparing the r.h.s. of (2.20) with (2.5), one immediately sees that M are related to ± the Orlov-Shulman operators. In fact, one can establish a precise correspondence between all data of the Lax formalism and MQM quantities: the lattice parameter s is identified with the Fermi level µ; • the Lax operators coincide with the light-cone coordinates, L = x1/R, L¯ = x1/R; • + − the Orlov-Shulman operators encode the asymptotics of the Fermi sea and are given by • M = RM , M¯ = RM ; + − the Baker-Akhiezer function coincides with the perturbed one-fermion wave function; • the string equations are the same as the equations for the profile of the Fermi sea and • correspond to the choice f = (M/R)1/RL 1, g = M. − – 6 – The commutation relation (2.19) then follows from a combination of (2.6) with the string equations. Finally, the classical limit of the shift operator ω turns out to be a uniformization parameter for the Riemann surface described by (2.20).1 Note that the two Baker-Akhiezer functions, Ψ and Ψ¯, give one-fermion wave functions in the two chiral representations, x and x , respectively. The requirement that they represent + − the same quantum state is precisely what generates the profile equations (2.20). Due to this condition, the two wave functions are related by a Fourier transform. Restricting to a finite set of non-vanishing couplings t , it is straightforward to find an k ± exact solution, i.e. an explicit form of the Lax and Orlov-Shulman operators as well as the τ-function of this system. We are interested in the particular case R = 1 and t = 0 for k ± all k 2. The perturbation induced by the couplings t is known to correspond to the 1 ≥ ± sine-Liouville operator on the string worldsheet, and the compactification radius R = 1 is the self-dual point with respect to T-duality which leads to an enhanced symmetry. The corresponding solution of the Toda hierarchy is given by [12] 1 3 F = µ2 logµ µt t , x = √µω 1 t . (2.21) 0 2 (cid:18) − 2(cid:19)− 1 −1 ± ± ∓ ∓1 3. Four-dimensional QK spaces and integrability Now we turn to a seemingly completely different subject — quaternionic geometry, which is characterized by the existence of three (almost) complex structures Ji satisfying the algebra of quaternions and appears in two forms, hyperka¨hler and quaternion-Ka¨hler. In fact, it is well known that in four-dimensions HK spaces possess an integrable structure, which in the presence of anisometry is given by thedispersionless Toda hierarchy presented inthe previous section. Moreover, as we explained in the introduction, the QK/HK correspondence allows to transfer these results to the realm of four-dimensional QK geometry. In this section we establish precise relations between the quantities used to describe QK spaces with one Killing vector and the ones entering the Lax formalism. 3.1 Toda ansatz The most direct way to see the appearance of Toda in the description of QK spaces is to recall that in four dimensions QK manifolds coincide with self-dual Einstein spaces with a non-vanishing cosmological constant. Such spaces are known to be classified by solutions of a single non-linear differential equation [25]. In the presence of an isometry, it can be reduced to the three-dimensional continuous Toda equation ∂ ∂ T +∂2eT = 0, (3.1) z z¯ ρ which is identical to the equation (2.16) on the dispersionless free energy of Toda hierarchy. In terms of the solution T(ρ,z,z¯), the metric is given by [26] 3 P 1 ds2 = dρ2 +4eTdzdz¯ + (dθ+Θ)2 , (3.2) −Λ (cid:20)ρ2 Pρ2 (cid:21) Q (cid:0) (cid:1) 1In addition, to describe the theory in Lorentzian signature, the Planck constant should be continued analytically to imaginary values, ~ =i~ . Toda MQM – 7 – where the isometry acts as a shift in the coordinate θ. Here P 1 1 ρ∂ T, Λ is the ≡ − 2 ρ cosmological constant, and Θ is a one-form such that dΘ = i(∂ Pdz ∂ Pdz¯) dρ 2i∂ (PeT)dz dz¯. (3.3) z z¯ ρ − ∧ − ∧ The integrability condition for (3.3) follows from (3.1). Of course, not only the metric, but also all geometric data such as quaternionic two forms, Levi-Civita connection, etc., can be expressed through T. We refer to [27] for their explicit expressions. Comparing the Toda equations (3.1) and (2.16), one immediately finds the following identifications z,z¯ t , ρ s, T ∂2F . (3.4) ↔ ±1 ↔ ↔ s 0 In principle, one can go further and find similar relations for other quantities. However, as we already know, working with differential equations is not the best way to describe a solution of the hierarchy. It is much better to use string equations which turn out to have a certain twistorial interpretation [15, 14]. Similarly, QK manifolds have a much more powerful descriptionintermsoftheirtwistor spaces [28]. Aswewillseenow, thetwo becomeessentially identical. 3.2 Twistor space and Lax formalism The twistor space of a QK manifold is a CP1 bundle over , whose connection is given Z M M by the SU(2) part p~ of the Levi-Civita connection2 on and the CP1 fiber corresponds to M the sphere of almost complex structures Ji. The twistor space carries a so called complex contact structure [29, 30], which is represented by a holomorphic one-form such that the X associated top-form (d )n is nowhere vanishing. Locally it can be written as X ∧ X 2 [i] = eΦ[i] dt+p+ ip3t+p t2 , (3.5) − X t − (cid:0) (cid:1) where the index [i] labels open patches of a covering of , t parametrizes the CP1 fiber, Z and the function Φ[i](xµ,t) is known as the contact potential. In the case we are interested in, when features one Killing vector, the contact potential is actually real, globally well M defined, and t-independent so that one can put Φ[i] = φ(xµ). Locally, it is always possible to choose such coordinates that the contact form acquires X a standard form. Restricting to the case dimR = 4, this means that in a patch i M U ⊂ M one can write [i] = dα[i] +ξ[i]dξ˜[i], (3.6) X ˜ where (ξ,ξ,α) are analogous to holomorphic Darboux coordinates of complex symplectic geometry. Generically, these coordinates must be regular in . The only exception is the i U coordinates defined around the north ( t = 0 ) and the south ( t = ) poles of + CP1. In these patches the Darboux coo{rdinat}es∈arUe assumed to have t{he fol∞low}i∈ngUe−xpansions ∞ ∞ ξ[ ](t) = t 1 + ξ[ ]t n, ξ˜[ ](t) = ξ˜[ ]t n, ± R ∓ n± ± ± n± ± Xn=0 Xn=0 (3.7) ∞ α[ ](t) = α[ ]t n 2clogt, ± n± ± − Xn=0 2Recall that the holonomy group of a 4n-dimensional QK manifold is contained in Sp(n) SU(2). × – 8 – where isareal functionon andc isarealnumber known asanomalous dimension. Thus, R M ξ[ ] have a simple pole and α[ ] have a logarithmic singularity controlled by c. The latter has ± ± an important physical meaning generating the one-loop correction to the local c-map [28]. On the overlap of two patches , the two coordinate systems are related by a com- i j U ∩U plex contact transformation. Such transformations are generated by holomorphic functions H[ij](ξ[i],ξ˜[j],α[j]), which we call transition functions. Together with the anomalous dimen- sion, they encode all geometric information about and its twistor space . To extract the M Z metric from them, the most non-trivial step is to find the Darboux coordinates as functions of t and coordinates on . Although this cannot be accomplished generically, it is possible M to write certain integral equations on the Darboux coordinates in terms of H[ij]. We refrain from writing them explicitly and refer to [31, 32] for more details. This twistor description has been related to the one based on the Toda ansatz (3.2) in [27]. For our purposes it is sufficient to provide the relation between the coordinates, contact and Toda potentials, which read as follows 1 1 ρ = eφ, z = ξ˜[+], θ = Imα[+], T = 2log . (3.8) 2 0 2 0 R Note that the only effect of the anomalous dimension is to shift the contact potential intro- duced in (3.5), i.e. eφ = eφ c. The identification (3.8) nicely maps this shift to the c=0 | − ambiguity in the solution of the Toda equation, T˜(ρ,z,z¯) = T(ρ+c,z,z¯). Although the two ˜ solutions, T and T, are related by a coordinate change, due to the explicit dependence of the metric (3.2) on ρ, the geometry of the corresponding QK manifold is strongly affected by the parameter c. After this very brief review of the twistor description of QK spaces, one can easily see how it is embedded into the Lax formalism of the dispersionless Toda hierarchy. Indeed, comparing the expansions (3.7) with (2.1), one observes a close similarity between the Lax operators and the Darboux coordinates ξ[ ]: they are both given by Laurent series which ± contain one singular term. This similarity is further strengthen by the comparison of the coefficients of these singular terms. Using (3.8) and (2.10), one finds = eT/2 e∂s2F0/2 = r, (3.9) R ↔ in agreement with (3.4). Correspondingly, the phase space variable ω should be identified with the CP1 coordinate, ω 1 t. Besides, the symplectic structures derived from (2.6) − ↔ and d (3.6) suggest that the Orlov–Shulman operators should be related to the Darboux X coordinates ξ˜[ ]. Finally, it is well known (see, for instance, [15, 22]) that the string equations ± (2.12) can be considered as gluing conditions between the two patches around the north and southpolesofCP1. Thecondition(2.13)inturnrequires thatthegluingconditionsgeneratea symplectomorphism preserving the symplectic structure. We summarize all the identifications in the following table: ω 1 t − ←→ lattice variable s ρ = eφ ←→ ∂2F T = 2log s 0 ←→ R t z,z¯ (3.10) 1 L±,L¯ ←→ ξ[ ] ± ←→ M/L,M¯/L¯ ξ˜[ ] ± ←→ string equations gluing conditions ←→ – 9 –