RG-Whitham dynamics and complex Hamiltonian systems

Inspired by the Seiberg-Witten exact solution, we consider some aspects of the Hamiltonian dynamics with the complexified phase space focusing at the renormalization group(RG)-like Whitham behavior. We show that at the Argyres-Douglas(AD) point the number of degrees of freedom in Hamiltonian system effectively reduces and argue that anomalous dimensions at AD point coincide with the Berry indexes in classical mechanics. In the framework of Whitham dynamics AD point turns out to be a fixed point. We demonstrate that recently discovered Dunne-\"Unsal relation in quantum mechanics relevant for the exact quantization condition exactly coincides with the Whitham equation of motion in the Omega - deformed theory.


Introduction
The holomorphic and complex Hamiltonian systems attract now the substantial interest partially motivated by their appearance in the Seiberg-Witten solution to the N = 2 SUSY YM theories [1]. They have some essential differences in comparison with the real case mainly due to the nontrivial topology of the fixed energy Riemann surfaces in the phase space. Another subtle issue concerns the choice of the quantization condition which is not unique.
The very idea of our consideration is simple -to use some physical intuition developed in the framework of the SUSY gauge theories and apply it back to complex or holomorphic Hamiltonian systems which are under the carpet. The nontrivial phenomena at the gauge side have interesting manifestations in the dynamical systems with finite number degrees of freedom. There are a few different dynamical systems in SUSY gauge theory framework. In the N = 2 case one can define a pair of the dynamical systems related with each other in a well defined manner (see [2] for review). The second Whitham-like Hamiltonian system [3] is defined on the moduli space of the first Hamiltonian system. Note that there is no need for the first system to be integrable while the Whitham system is certainly integrable. It can be considered as the RG flow in the field theory framework [4]. One more dynamical system can be defined upon the deformation to N = 1 SUSY where the chiral ring relation plays the role of its energy level. In this case one deals with the Dijkgraaf-Vafa matrix model [5,6] in the large N limit. It is known that matrix models in the large N limit give rise to one-dimensional mechanical system, with the loop equation playing the role of energy conservation and 1-point resolvent playing the role of action differential pdq. The degrees of freedom in all cases can be attributed to the brane coordinates in the different dimensions and mutual coexistence of the dynamical systems plays the role of the consistency condition of the whole brane configuration. We shall not use heavily the SUSY results but restrict ourselves only by application of a few important issues inherited from the gauge theory side to the Hamiltonian systems with the finite number degrees of freedom. Namely we shall investigate the role of the RG-flows, anomalous dimensions at AD points and condensates in the context of the classical and quantum mechanics.
First we shall focus at the behavior of the dynamical system near the AD point. It is interesting due to the following reason. It was shown in [7] that the AD point in the softly broken N = 2 theory corresponds to the point in the parameter space where the deconfinement phase transition occurs. The field theory analysis is performed into two steps. First the AD point at the moduli space of N = 2 SUSY YM theory gets identified and than the vanishing of the monopole condensate which is the order parameter is proved upon the perturbation. The consideration in the complex classical mechanics is parallel to the field theory therefore the first step involves the explanation of the AD point before any perturbation. We argue that the number of degrees of freedom at AD point gets effectively reduced which is the key feature of the AD point in classical mechanics. Moreover we can identify the analog of the critical indices at AD point in Hamiltonian system as the Berry indexes relevant for the critical behavior near caustics. Also, we propose a definition for a "correlation length" for a mechanical system so that corresponding anomalous dimensions coincide with the field-theoretical ones. From the Whitham evolution viewpoint the AD point is the fixed point. However the second step concerning the perturbation and identification of the condensates is more complicated and we shall restrict ourselves by the few conjectures. Note that the previous discussion of the Hamiltonian interpretation of the AD points can be found in [8] however that paper was focused at another aspects of the problem.
Quantization of complex quantum mechanical systems is more subtle and we consider the role of the Whitham dynamics in this problem. The progress in this direction concerns the attempt to formulate the exact energy quantization condition which involves the non-perturbative instanton corrections. It turns out that at least in the simplest examples [9] the exact quantization condition involves only two functions. Later the relation between these two functions has been found [10]. We shall argue that the Dunne-Ünsal relation [10] which supplements the Jentschura-Zinn-Justin quantization condition [9] can be identified as the equation of motion in the Whitham theory. To this end we derive the Whitham equations in the presence of Ω-deformation, which has not been done in a literature before.
The paper is organized as follows. Whitham dynamics is briefly reviewed in Section 2. In Section 3 we shall consider the different aspects of the AD points in the classical mechanics. Section 4 is devoted to the clarification of the role of the Dunne-Ünsal relation and to the derivation of Whitham equations in the Ω-deformed theory. Also, we discuss various quantization conditions for complex systems and elucidate the role of the curve of marginal stability. The key findings of the paper are summarized in the Conclusion. In the Appendix we show how the Bethe ansatz equations are modified by the higher Whitham times.

Generalities
Let us define some notations which will be used later. Hyper-elliptic curve is defined by where P N (x) -is polynomial of degree N. There are 2g cycles A i , B i , i = 1...g which can be chosen as follows (A i , A j ) = 0, (B i , B j ) = 0, (A i , B j ) = δ ij . For genus g hyper-elliptic curve there are exactly g holomorphic abelian-differentials of the first kind ω k : which are linear combinations of dx/y, ..., x (N −1)/2 dx/y. Period matrix is given by: Define dΩ j -meromorphic abelian differential of the second-kind by the following requirements: normalization: and behavior near some point(puncture):

5)
dΩ 0 is actually abelian differential of the third kind -with two simple poles with residues +1, −1.
We will also need the so-called Seiberg-Witten(SW) differential dS. Throughout the paper we will extensively use its periods: It is useful to introduce vectors U (j) : which obey the identity Below we will use Riemann bilinear identity for the pair of meromorphic differentials ω 1 ,ω 2 : however sometimes it is more convenient to work with non-normalized differentials dv k = x k dx/y: Recall now some general facts concerning Whitham dynamics. In classical mechanics, action variables a i are independent of time. However, sometimes it is interesting to consider a bit different situation when some parameters of the system become adiabatically dependent on time. Then the well-known adiabatic theorem states that unlike other possible integrals of motion, a i still are independent (with exponential accuracy) on times.
While considering finite-gap solutions to the integrable system one deals with a spectral curve and a tau-function If one introduces "slow"(Whitham) times t i = ǫT i , ǫ → 0, Whitham hierarchy equations tell us how moduli can be slowly varied provided (2.12) still gives the solution to the leading order in ǫ [11], [3]. These equations have zero-curvature form [3] This guarantees the existence of dS such that which results in the adiabatic theorem: The full Whitham-Krichever hierarchy (2.14) has a variety of solutions. Every dS satisfying (2.15) generates some solution. In particular, dS can be chosen to be the Seiberg-Witten meromorphic form T 1 dS SW , T 1 = log Λ is the first Whitham time.
One can introduce several times where dΩ i obey the following requirements: and ≈ means that they have the same periods and behavior near the puncture It was argued in [12] that higher times correspond to the perturbation of the UV Lagrangian by single-trace N = 2 vector superfield operators: The first Whitham time T 1 is just a shift of UV coupling. In Appendix we will discuss the spectral curve when higher times are switched on and derive generalized Bethe equations for this case, which hitherto has not been discussed in a literature.

Whitham dynamics in the real case
For completeness, let us recall the analogue of the Whitham hierarchy for the case of the real phase space. It means that we consider a real dimension one curve on a two dimensional real plane instead of a complex curve. Let us introduce complex coordinatesz, z then the curve is determined by the equation We shall assume thatz, z pair yields the phase space of some dynamical system and the curve itself corresponds to its energy level. With this setup it is clear that Poisson bracket betweenz and z is fixed by the standard symplectic form. Let us remind the key points from [13] where the Whitham hierarchy for the plane curve was developed. The phase space interpretation has been suggested in [14]. The Schwarz function S(z) is assumed to be analytic in a domain including the curve. Consider the map of the exterior of the curve to the exterior of the unit disk where ω is defined on the unit circle. Introduce the moments of the curve t n = 1 2πin z −n S(z)dz, n < 0 (2.23) which provide the following expansion for the Schwarz function Let us define the generating function where One can derive the following relations ∂ tn Ω(z) = (z n (ω)) + + 1/2(z n (ω)) 0 (2.31) ∂t n Ω(z) = (S n (ω)) + + 1/2(S n (ω)) 0 (2.32) Therefore we identify log ω as angle variable and the area inside the curve t 0 as the action variable. Let us denote by (S(ω)) + the truncated Laurent series with only positive powers of ω kept and the (S(ω)) 0 is the constant term in the series. The differential dΩ dΩ = Sdz + log ωdt 0 + (H k dt k −H k dt k ) (2.33) yields the Hamiltonians and Ω itself can be immediately identified as the generating function for the canonical transformation from the pair (z,z) to the canonical pair (t 0 , log ω).
The dynamical equations read ∂t n S(z) = ∂ zHn (z) (2.35) and the consistency of (2.34), (2.35) yields the zero-curvature condition which amounts to the equations of the dispersionless Toda lattice hierarchy. The first equation of the hierarchy reads as follows where ∂ t 0 φ = 2 log r. The Lax operator L coincides with z(ω) LΨ(z, t 0 ) = zΨ (2.37) and its eigenfunction -Baker-Akhiezer(BA) function looks as follows Ψ = e Ω h . Hamiltonians corresponding to the Whitham dynamics are expressed in terms of the Lax operator as follows Now it is clear that the BA function is nothing but the coherent wave function in the action representation. Indeed the coherent wave function is the eigenfunction of the creation operatorb Ψ = bΨ (2.39) From the equations above it is also clear that Ω it is the generating function for the canonical transformations from the b, b + representation to the angle-action variables.
Having identified the BA function for the generic system let us comment on the role of the τ function in the generic case. To this aim it is convenient to use the following expression for the τ function where the bra vector depends on times while the ket vector is fixed by the point of Grassmanian This representation is convenient for the application of the fermionic language where ∆(z) is Vandermonde determinant. The consideration above suggests the following picture behind the definition of the τ function. The fixing the integrals of motion of the dynamical system yields the curve on the phase space. Then the domain inside the trajectory is filled by the coherent states for this particular system. Since the coherent state occupies the minimal cell of the phase space the number of the coherent states packed inside the domain is finite and equals N. Since there is only one coherent state per cell for the complete set it actually behaves like a fermion implying a kind of the fermionic representation.
Therefore we can develop the second dynamical system of the Toda type based on the generic dynamical system. The number of the independent time variables in the Toda system amounts from the independent parameters in the potential in the initial system plus additional time attributed to the action variable. Let us emphasize that the choice of the particular initial dynamical system amounts to the choice of the particular solution to the Toda lattice hierarchy.

Generalities
Here we review the Argyres-Douglas phenomena [15] and following [16] demonstrate how one can compute some anomalous dimensions in the superconformal theory.
The key element of the Seiberg-Witten solution is the spectral curve which is (N-1)-genus complex curve for SU (N ) gauge theory. In case of pure gauge SU (N ) theory it is given by(Λ is dynamical scale) In SU (2) case it is torus: where u = h 2 , which at u 2 = 1 degenerates -one of its cycles shrinks to zero.
Recalling BPS-mass formula, this can be interpreted as monopole/dyon becomes massless and the description of the low-energy theory as U(1) gauge theory breaks down. Much more interesting situation is possible in SU (3) case [15]: then for u = 0, v 2 = Λ 6 , the curve becomes singular: In this case, two intersecting cycles shrink -it means that mutually non-local particles (monopole and dyon charged with respect to the same U(1)) become massless.
In [15] it was conjectured that at this point the theory is superconformal. This result was generalized to SU (2) gauge theory with fundamental multiplets in [16].
In brief, the argument goes as follows: Let us denote ρ is dimensionless, ǫ has a dimension of mass and sets an energy scale. Then the genus two curve degenerates to the "small" torus with modular parameter and periods ω s , ω s D ∼ 1/a → ∞. The modular parameter of the "large" torus . Below we will often use "s" and "l" indices to denote small and large tori. The period matrix becomes diagonal (again up to O(δu, δv) non-diagonal terms): The crucial observation is that modulus of the "small" torus is independent of scale ǫ. Due to the diagonal form of the period matrix, the "small" U(1) factor (with masses ≈ ǫ 5/2 /Λ 3/2 ) decouples from the "large" U(1) factor (with masses ≈ Λ) and we are left with the RG fixed point with the coupling constant τ s = e 2πi/3 Anomalous dimensions can be restored as follows [16]. Kähler potential Im(aa D ) has dimension 2, so a and a D have dimension 1. From (3.7) we infer that relative dimensions are D(x) : D(δu) : D(δv) = 1 : 2 : 3 -it could be seen either as the R−charge condition or as a requirement for a cubic singularity. From

Toda chain: Argyres-Douglas point
In this subsection we comment on the behavior of the solutions to the equations of motion of Toda chain near the Argyres-Douglas point and show how the number of effective degrees of freedom get reduced . In the case of a periodic Toda chain it is possible to write down an explicit solution using the so-called tau-function [11]: ζ -is just a constant, U (k) are defined in the section [2.1]. Then coordinates of particles q n can be expressed in terms of τ -functions Since at the AD point the period matrix is diagonal, the theta function factorizes into the product of two theta functions corresponding to small torus and large torus: Moreover, since U (1) = a D − τ a and τ s = e 2πi/3 , a s , a s D → 0 corresponding theta function completely decouples and the solution is determined up to the relative shift in terms of the large torus only. This is the reduction of degrees of freedom mentioned in the Introduction.
In the case of N-particles it is possible to degenerate several pairs of intersecting cycles. In this case several small tori will appear. The period matrix will be blockdiagonal and respective masses a, a D tend to zero. So we can conclude that small tori will again decouple and corresponding degrees of freedom get frozen.

Critical indexes in superconformal theory and Berry indexes
Since the superconformal theory actually pertains to the small torus, lets look closely at the vicinity of the Argyres-Douglas point. Near the AD point two tori are almost independent, so we can concentrate solely on the part of the tau function which corresponds to the small torus -we will drop subscript s for brevity. The key observation above was that a, a D → 0 and τ → exp(2πi/3), hence we can expand (3.12) in Taylor series: We denote θ ′ (2πinU (0) + ζ|τ ) = B n for brevity, then (3.15) For general ζ, the coefficient in front of (a D − τ a)t is not zero. Let us recall that the modular parameter τ is independent of ǫ in the leading order. The same is true for the U (0) since it equals to L D − Lτ , where L D , L are periods of third kind Abelian differential x 2 dx/y We can define "correlation length" δq as the distance travelled by particles over the time 1/Λ (here Λ = 1). Usually, correlation length tends to infinity near a conformal point. Here, in classical mechanical system, it tends to zero. We can obtain "anomalous dimensions" by deriving the integrals of motion in terms of δq: Equation (3.15) tells us that δq is proportional to a, that is it has a field-theoretical anomalous dimension 1. Therefore, we can conclude that the mechanical "anomalous dimensions" defined by (3.17) coincide with the field-theoretical anomalous dimensions (3.10). Surprisingly, counterparts of these superconformal dimensions also arise in the context of caustics in optics (see [17] for a review). In optics, one is interested in the wave function: where k is an inverse wavelength. One can define singularity indices β, σ j as The function W determines the shape of the caustic. Classification of all possible W has been intensively studied in the catastrophe theory framework. Equation (3.19) reminds the wave function of Lagrangian brane, with W playing the role of the superpotential. From the SW theory viewpoint, W defines the spectral curve with C playing the role of moduli. Let us consider the standard AD point in SU (3). Then: In the notation of [17]: and singularity indices read as However, we have to identify variables properly. In optics, or equivalently, classical mechanics everything is measured in terms of k, whereas in the field theory everything is measured in terms of a. Obviously, x = s and s has its own scaling properties: one requires the highest term ks 6 to be scale invariant [17]. Therefore D(k) = 6D(s) = 12/5 (recall that D(x) = 2/5 -eq. (3.10)). Therefore, in the field theoretical normalization which are exactly the anomalous dimensions in the equation (3.10). Similar analysis can be carried out for the case of N c = 2, N f = 1 where we have found a perfect agreement too.

Argyres-Douglas point via Whitham flows
In this section we specify the Whitham equations to the case of the pure SU (3) gauge theory -3 particle Toda chain. Then we consider the SU (2) case with fundamental matter. We demonstrate that the AD point is a fixed point for the Whitham dynamics. From the point of view of N-particle closed Toda chain, higher times T l , l > N −1 just do not exist and corresponding flow should be trivial. From the field-theoretical viewpoint, it reflects the fact that for a N × N matrix A, A N = a 1 A N −1 + .. + a N −1 A + a N however the trivialization of the flow is not obvious. In order to prove that we will use Riemann bilinear identity and equation (2.15). In case of Toda-chain Abelian differentials of the first kind are linear combinations of dx/y, ..., x N −2 dx/y, therefore they have at most (N-2)-degree zero at infinity, d −1 dΩ j has pole of order j at infinity.
therefore the flow in indeed trivial.
In case of N-particle Toda chain all first-and second-derivatives of prepotential were calculated in [4].
Since Argyres-Douglas point is RG fixed point for one of U(1) factors, we conjecture that for this U(1) factor (i.e. "small" torus) Whitham dynamics should be also trivial at least in T 1 . In case of SU (3) we can consider only T 1 and T 2 . In this case H 2 = u, H 3 = v and using the fact that where near Argyres-Douglas point [15,18], Applying Riemann bilinear identity to dS and dv k : and for dv k and dΩ 1 Generalization of (3.40),(3.41) for SU (N ) with non-zero times T i is straightforward. This form clearly shows that if some U(1) factors decouple, they decouple in the Whitham dynamics as well. Whitham equations depend on the choice of A-and B-cycles, in other words they are not invariant under modular group. AD point is significant because it is modular invariant. It means that whatever basis of cycles we choose, AD will be stationary point.
Let us compare the AD point with other possible degenerations, for example to the case when all B-cycles vanish [19]. For simplicity take T 2 = 0 then the period matrix: where Λ m -are some constants. Due to the diagonal form of τ , two U(1) factors again decouple. Since a D n → 0 and a n do not vanish, Whitham dynamics is nontrivial.
If all A-cycles vanish, a n → 0, so dynamics is again nontrivial. Now let us consider the SU (2) theory with fundamental matter. If N f < 4, beta-function is not zero and RG dynamics is not trivial. In [20] the case with only two non-zero Whitham times was considered. The result is as follows: we have two non-zero times from the very beginning: and derivative of prepotential with respect to T 1 : According to general philosophy, ∂a/∂T 1 = 0, ∂a/∂T 0 = 0 , hence for N f = 1 We see that the right hand side is proportional to the charge condensate (see section [3.5]). It was proved in [7] that both monopole and charge condensate vanish at the AD point in the theory with N f = 1. Therefore, we conclude that the statement that the AD point is a fixed point for the Whitham dynamics holds when fundamental matter is switched on.

On confinement in the classical mechanics
Since the main purpose of the paper is to understand the reincarnation of the field theory phenomena in the complex classical dynamics we are to make some comment on the confinement phenomena. The rigorous derivation of the confinement in the softly broken N=2 SUSY YM theory in [1] was the first example in the strongly coupled gauge theory. Although it is a kind of abelian confinement irrelevant for QCD it is extremely interesting by its own. The non-vanishing order parameter is the monopole condensate which provides the confinement of the electric degrees of freedom. It is proportional to the parameter of microscopic perturbation by N = 1 superpotential which breaks N = 2 to N = 1. In the IR one has the following exact superpotential [1]: At the monopole point, where a D = 0, one arrives at the monopole condensate [1]: 3.49) and the charge condensate of matter in the fundamental representation [7]: One more piece of intuition comes from the consideration of the AD point in the softly broken SQCD [7]. Since at the AD point both monopole and matter condensates vanish the AD point is the point of deconfinement phase transition. Note that the gluino condensate does not vanish at the AD point. These results have been obtained using the interpolation between N = 2 and N = 1 theories via the Konishi anomalies.
We would like to ask a bit provocative question: is it possible to recognize all condensates and the deconfinement phase transition in the framework of the classical mechanics? We shall not answer these questions completely but make some preliminary discussion on this issue. First of all, consider the pure SU (2) case which corresponds to the cosine potential. Upon the perturbation added the monopole condensate (3.49) gets developed and due to the Konishi anomaly relation the gluino condensate is proportional to the scalar condensate Therefore, as the first step we could ask about the meaning of the Konishi anomaly relation in the Hamiltonian framework. Two dynamical systems are involved. The scalar condensate u plays the role of the energy in the N = 2 Hamiltonian system with V = Λ cos q while upon deformation to N = 1, the gluino condensate plays the role of the action(period of 1-point resolvent) in the Dijkgraaf-Vafa matrix model [5,6]. Potential for this system reads as: where f n−1 is polynomial of degree n − 1, if W U V has degree n + 1. For the simplest deformation µΦ 2 it is nothing but the complex oscillator. Actually we have to make the second step. At the first one the meaning of the AD point as the decoupling of the small torus has been found. Now the question concerns the very precise identification of the soft breaking of SUSY in the framework of the complex Hamiltonian system. The analogy with the Peierls model mentioned in [21] can be useful here. It describes the one-dimensional superconductivity of electrons propagating on the lattice. The key point is that the Riemann surface which is the solution to the equation of motion in the Toda system simultaneously plays the role of the dispersion law for the Lax fermions. Therefore the degeneration of the surface at AD point corresponds to the degeneration of the Fermi surface for the fermions. Therefore the deconfinement phase transition at AD points presumably corresponds to the breakdown of superconductivity in the Peierls model. We hope to discuss this issue in details elsewhere.
Also, note that eqs. (3.49),(3.50) strongly resemble Whitham equations of motion from the previous section. It is not a coincidence -Whitham dynamics is useful for softly breaking N = 2 → N = 0 [22,23,24]: we can promote the first time T 1 = log Λ to background N = 1 spurion chiral multiplet. After that, we can switch on the other scalar component of this multiplet: This deformation preserves all holomorphic properties of the original theory, so we are able to write down the exact prepotential for this new theory: Since θ explicitly enters the prepotential, the theory has no supersymmetry. Additional terms in the IR Largangian are [22](G * = G): where F ′ = ∂F/∂a and τ = Im(F ′′ ) -is a coupling constant, ψ is a fermion in the N = 1 chiral multiplet. In the UV we have: Note that G gives masses to both fermions and imaginary part of the Higgs field, whereas deformation to N = 1 by the superpotential (3.47) gives usual Higgs mass term µ 2φ φ and µψψ and does not give mass to the gluino λ. In [23,24] various monopole and dyon condensates were calculated. Here, we find gluino condensate, that is we derive an analogue of the Konishi anomaly using Whitham equations. Let us emphasize once more that we deal with not N = 1 theory, but with the N = 0 one obtained by a very special deformation of the N = 2 theory. So we do not expect that the final expression would be the same as in the N = 1 theory. However, as we will see in a moment, the result naturally generalizes the Konishi anomaly. Varying (3.55) with respect to φ and λλ, ψψ (for simplicity we consider real φ) we get: and taking into account that ∂F/∂T 1 = 2u Since W IR = µu, the last equation looks very natural and to some extend is an analogue of (3.51).
4 On the quantization procedure

Different quantizations of complex Hamiltonian systems
There are some new points in the quantization of complex integrable systems. First of all, the essential part of a quantization concerns a choice of Hilbert space. In the pioneer work [25], in the case of one degree of freedom the following quantization was suggested: Hilbert space consists of analytic functions on a complex plane with possible irregular singularity at infinity, and a scalar product is given by: where C is some contour on a complex plane. Hamiltonian is taken to be a standard one:Ĥ =p 2 /2 + U (q), withp = i∂/∂q. Then the Schrödinger equation is just the standard Schrödinger equation analytically continued to a complex plane. If U (q) is an entire function then the equation is consistent with the definition of the Hilbert space. When the curve C coincides with the real axis this construction gives the standard quantization.
In the real case the quantization condition for the energy levels comes from the requirement that the wave function is normalizable. In [25] an analogue of the WKB quantization was suggested: where integral should be taken along the line where integrand is real. Note that since everything is complex now, it is actually two real conditions on a complex energy u: Re a = 2π n (4.4) Im a = 0 Perfect agreement with numerical computations has been found. It worths mentioning that the same condition was proposed in [26] for studying complex nonhermitian Hamiltonians. However, if the potential is not holomorphic, one can impose different quantization condition: wave function is not required to be holomorphic. Instead, one imposes its single-valuedness. At least one such example is known in literature [27]: spectrum of XXX chain with complex spin emerging in high energy QCD for describing effective interaction between Reggeons [28,29]. In brief, the problem is as follows: complex spin chain has a non-holomorphic Hamiltonian: Actually z andz are complex coordinates on a real plane of Reggion coordinates. Requirement that the ψ has no monodromy around cycles yields a bit different WKB quantization condition [27]: Re a = 2π n (4.6) Re a D = 2π n D which coincides with the conventional WKB condition when n D = 0. Returning to the SW theory, in [30] it was shown that in the Nekrasov-Shatashvili(NS) limit ǫ 2 = 0 of Ω deformation, underlying integrable systems get quantized. The following quantization condition was proposed for theories without matter (Toda chain) or with adjoint matter (Calogero system): Quantization condition (4.3) looks exactly the same as Nekrasov-Shatashvili quantization. Nevertheless they are different: in (4.3) the integral can be taken along the finite number of paths on a complex plane(to ensure convergence), whereas in Nekrasov-Shatashvili quantization (4.7) one can choose arbitrary element of SL(2, Z): the choice a l = 2πǫ 1 n l is called type A quantization condition, while a D = 2πǫ 1 n l -type B. It was conjectured [30] that the type A condition fixes the wave function to be normalizable on the real axis and type B corresponds to the wave function, which is 2π periodic along the imaginary axis.The conjecture about the type A was proven in [31]. We do not know what conditions are imposed on the wave function by other elements of SL(2, Z). The case with fundamental matter was considered in [32], where it was shown that the conventional algebraic Bethe ansatz with polynomial Baxter function implies a l = m l − ǫ 1 n l , n l ∈ N. In the Appendix we will show how this quantization condition is modified by the non-zero Whitham times.
It is in order to make a comment concerning the place of the curve of marginal stability in the quantum spectrum. In the Seiberg-Witten theory with the gauge group SU (2) a BPS particle with electric and magnetic charges (q, p) has mass M = Z = |qa + pa D |. A BPS particle can decay into a BPS particle iff a and a D are collinear, that is Im a D a = 0 (4.8) This equation defines the curve of marginal stability on the moduli space. On a quantum mechanical side, energy level crossing occurs when there are two different cycles with the same allowed energy level. Let us denote these cycles a and na + ma D . Nekrasov-Shatashvili quantization conditions: na + ma D = k 2 (4.10) k 1 , k 2 ∈ N, but = ǫ 1 is not necessary real. If we divide the second equation by the first one If the original cycles are different, m = 0 and Im a D /a = 0. So we conclude that the level crossing can happen on the curve of marginal stability only.

Quantization and the Dunne-Ünsal relation
Seiberg-Witten solution to the Whitham-Krichever hierarchy can be thought of as a non-autonomous Hamiltonian system with the Hamiltonian 2u(a, Λ)/iπ and canonical pair {a j , a k D } = δ jk [33]. For 2-particle Toda chain: The last equation follows from the Matone relation [34]: which, in turn, can be thought of as a Hamilton-Jacobi equation, where the prepotential is playing the role of the mechanical action.
In what follows we will need to know how Whitham dynamics is affected by the Ω deformation. The prepotential involves two contributions [35]: and it was shown in [36] that the log Λ derivative of the instanton part is unchanged by the Ω-deformation: l =n +∞ 0 ds s exp(−s(a l − a n )) sinh(sǫ 1 /2) sinh(sǫ 2 /2) (4.17) The integral is divergent at the lower bound. The prescription is that one should keep only non-singular part -this is the origin of the scale Λ. Expanding the integrand near s = 0, one obtains the following Λ-dependent terms: n a 2 n log( a n Λ ) − ǫ 2 1 + ǫ 2 2 24 log a n Λ (4.18) Combining together perturbative and instanton contributions: Upon differentiating w.r.t. a, we conclude that Whitham equations of motion (4.12) still hold even in the case of general ǫ 1 , ǫ 2 . The natural question is what happens with the full Whitham hierarchy (2.14). One can try to attack this problem using beta-ensemble approach [37,38]. This approach is based on the AGT conjecture, since conformal blocks are equal to Dotsenko-Fateev beta-ensemble with finite N [39]. Actually, AGT conjecture in the NS limit(ǫ 2 → 0 which implies N → ∞ in the beta-ensemble) is equivalent to the statement [40]: WKB approximation allows to expand the phase of the wave function in powers of = ǫ 1 In [40] it was conjectured that the prepotential given by quantum periods of the differential p quant dq coincides with the Nekrasov prepotential in the NS limit. This statement was checked [40,41] up to o( 6 , log Λ) however no conceptual proof is known so far.
On the other hand, in [42] the large N limit of the beta-ensemble was thoroughly considered, and it was proven that the large N limit corresponds to the quantization of some mechanical system. One point resolvent plays the role of the Seiberg-Witten meromorphic differential, moreover it equals to dψ/ψ, where ψ is wave-function of the quantum mechanical system. We see that the AGT conjecture, the beta-ensemble approach and the conjecture about the exact WKB periods are all tightly related. Strikingly, after an appropriate deformation of Abelian meromorphic differentials, equations (2.15) and (2.15) still hold [42]. Therefore if we believe in either the conjecture about the exact WKB periods (4.20) from [40] or the AGT conjecture [43], we can conclude that in the Nekrasov-Shatashvili limit the Whitham dynamics is not quantized but only deformed.
Moreover, using this conjecture we will show now that the Whitham equations in the form (4.12) are quite general and are not affected by the quantization. For simplicity we will concentrate on genus one case. Let us consider Hamiltonian V (q) is polynomial of degree 2d, d > 1. For the exact WKB phase p quant = f we have the Riccati equation: f has a representation in power series in : f = f 0 + f 1 + 2 f 2 + ... Several first terms are: Again, since we require ∂a/∂c = 0, we have Now we apply Riemann bilinear identities for differentials ∂f ∂c dq and ∂f ∂E dq: The idea is that only f 0 contributes to the residue. Indeed, it is not difficult to show that f n behaves as O(x −(1+(n−1)(d+1)) ) at infinity and ∂f n /∂E = O(x −(1+2d+(n−1)(d−1)) ).
Then the contribution of order n+k is given by a differential which behaves at most as O(x −(1+2d+(n+k−2)(d+1)) ). The "classical" part behaves as O(x) and therefore can contribute, quantum corrections are suppressed by powers of x. The first quantum correction behaves as O(1/x d ) so has a zero residue. Higher quantum corrections have a zero even of higher degree at infinity. So we conclude that ∂a D ∂c = const ∂E ∂a (4.26) and const depends on a normalization and does not receive quantum corrections.
Recently, there was much progress in studying the relation between perturbative and non-perturbative expansions (see [9,44,10,45] and references therein) in both quantum mechanics and quantum field theory. In [9] Zinn-Justin and Jentschura using resurgence in multi-instanton expansion have conjectured the exact quantization condition for several quantum mechanics potentials. Amazingly it involves only two functions B(E, g) and A(E, g), where E is an energy(u in our notation) and g is a coupling constant. In [10] Dunne andÜnsal have found a relation between these two functions. We shall demonstrate that this relation is nothing but Whitham equation of motion.
The most simple example concerns a double-well potential: The first Whitham time is the coupling constant c which stands in front of the whole potential cV (q). In case of the double-well potential (4.27) c coincides with 1/g and the rescaling E → 2E/g is needed. In genus one, we have usual definitions for periods: The electric period a corresponds to classically allowed region near the bottom of the well, whereas a D is an instanton factor corresponding to the barrier penetration between two wells. Let us recover coefficients in Whitham equations. If we impose the constraint ∂a/∂g = 0 then we have for the dual period: Taking into account the Picard-Fuchs relation: we get and exact quantization condition reads as [9]: One should understand this relation in a sense that after finding the energy in series of g (including non-perturbative factors) it will be possible to resum the resulting series using Borel method. Moreover, all the ambiguities will cancel each other [9]. The Dunne-Ünsal relation [10] reads as where the function B(E, g) is easy to calculate Originally, calculation of the function A(E, g) involved tedious multi-instanton calculation. Note the arguments of A(B, g): derivative w.r.t. g is taken keeping B constant. Since B = a/2π we discern here the first Whitham equation ∂a/∂g = 0.
The second equation turns out to be the Dunne-Ünsal relation itself. Let us compare (4.32) with WKB quantization condition for a double-well potential [46]: According to [9], exact quantization condition reads as where φ is Bloch phase -we are dealing with the periodic potential which possesses band structure. Note the mismatch in the factor 1/2 with the quantization condition obtained in [10] using uniform WKB method [45] instead of resurgence in instanton calculus. We argue that the right choice is We will show in a moment, that this analytical continuation agrees with the Whitham equations, like in the double-well case.
To this end we can make use of the WKB quantization condition for a generic periodic potential [46]: 2 exp(ia D /2) cos(a/2) + 1 2 exp(−ia D /2) cos(a/2) = 2 cos(φ) (4.43) where a and a D are electric and magnetic quantum periods as before. Since there is a very simple relation We would like to emphasize that this derivation strongly relies upon the conjecture that quantum WKB periods give Nekrasov prepotential (see the discussion below eq. (4.20)). If we believe in either this conjecture or equivalently in the AGT conjecture, we can consider the above derivation as a proof of the Dunne-Ünsal relation, since we have shown that the Whitham equations are unaffected by the Ω-deformation.
Moreover, we claim that the Dunne-Ünsal relation holds for every genus one potential. For higher genera exact quantization condition has not even been conjectured yet. However the Whitham equations are the same so we can conjecture that they play the role of Dunne-Ünsal relations again. Note that we have used the Whitham dynamics for Riemann surfaces, that is for holomorphic dynamical systems. However we could use the real version described above as well. In this case the appropriate technique for the multi-regions in the phase space has been developed in [47]. We hope to consider the higher genus potentials elsewhere.

Conclusion
In this paper we make some observations concerning properties of the complex Hamiltonian systems. We have argued that the AD point can be considered as the fixed point from the Whitham dynamics viewpoint and it was shown that anomalous dimensions at AD point coincide with the Berry indexes in the classical mechanics. Also, we have defined a "correlation length" for the mechanical system near the AD point. We have derived Whitham equations for the Ω-deformed theory. Moreover we have made the useful observation that the Dunne-Ünsal relation relevant for the exact quantization condition can be considered as the equation of motion in the Whitham dynamics.
Certainly there is a lot to be done to treat the complex Hamiltonian systems properly both classically and quantum mechanically. In particular it would be important to clarify the fate of the Whitham hierarchy in the case of non-zero ǫ 1 , ǫ 2 and develop its own quantization. It seems that this issue has a lot in common with the generalization of the classical-quantum duality from [48,49] to the quantumquantum case.
The work of A.G. and A.M. was supported in part by grants RFBR-12-02-00284 and PICS-12-02-91052. The work of A.M. was also supported by the Dynasty fellowship program. A.G. thanks SCGP at Stony Brook University during the Simons Summer Workshop where the part of this work has been done for the hospitality and support. We would like to thank G. Basar, K. Bulycheva, A. Kamenev, G.  In this section we will consider Seiberg-Witten theory with the gauge group SU (N c ) with N f fundamental matter hypermultiplets in the NS limit of Ω deformation. We will switch on higher Whitham times and explicitly show how they deform spectral curve and Baxter equation.
Without higher Whitham times and Ω-deformation, the case of N f = 2N c cor-responds to the XXX spin chain with twist h = − 2q q + 1 , q = exp(2πiτ uv ) and inhomogeneities θ l , J l . The spectral curve reads as [2]: where A(x), D(x), t(x) are the following polynomials: Note that q corresponds to ultraviolet coupling, S and T act as

Masses of hypermultiplets correspond to parameters
In the hyperelliptic parametrization the curve looks as NS limit ǫ 1 = 0, ǫ 2 = ǫ corresponds to the quantization of the XXX chain. Spectral curve (6.1) promotes to the Baxter equation, since w becomes operator w = exp(iǫ∂ x ): The case of N f < 2N c can be obtained by taking some of the masses to infinity, while keeping the product constant. It leads to the following spectral curve (6.10) with A N f (x) = M -is a magnon number, x k -Bethe roots. Now, we consider non-zero Whitham times, which are coupling constants for the single-trace N = 2 vector superfields (see eq. (2.20)). Our considerations are close to those in [32,50].

R(x) =
A(x)D(x) P (x)P (x + ǫ 1 ) Since ρ is constant, variation over y li can be thought of as a variation of ρ. Therefore, we end up with the following saddle point equation: (6.20) or using the explicit expression for the x 0 li : Indeed, we see that T 1 is responsible only for the shift of τ uv . This is the generalized Bethe ansatz equation we have mentioned before and one can derive the following Baxter equation: −h exp t ′ (x + ǫ 1 ) ǫ 1 A(x)Q(x+ǫ 1 )+(2+h) exp − t ′ (x) ǫ 1 D(x)Q(x−ǫ 1 ) = 2T (x)Q(x) (6.22) In the classical limit ǫ 1 → 0, the spectral curve reads as y 2 = T (x) 2 + (h + 2)hA(x)D(x) exp(t ′′ (x)) (6.23) Several comments are in order. First of all, note that in (6.20) products are infinite. It was argued in [32], that if the following quantization condition is imposed a l = m l − ǫ 1 n l , n l ∈ Z, n l > 0 (6.24) the most of the factors decouple x li = x 0 li = a l + (i − 1)ǫ, i ≥ n l (6.25) and we are left with the polynomial Baxter function, that is with the algebraic Bethe ansatz. However, it is apparent from the (6.22) that Q could not be polynomial because of the exponential factors. Nonetheless, we can get rid of them by looking for a solution in the form Q(x) = F (x) exp(C(x)/ǫ 1 ) (6.26) where F (x), C(x)-polynomials. For C(x) we have the following equations t ′ (x + ǫ 1 ) + C(x + ǫ 1 ) − C(x) = 0 (6.27) −t ′ (x) + C(x − ǫ 1 ) − C(x) = 0 which are dependent. Therefore, we can always construct C(x) from t(x) unambiguously. For F (x) we have the standard algebraic Bethe ansatz equations. One can repeat all considerations from the [32] and a that the quantization condition (6.24) is not modified.