Haantjes Algebras of Classical Integrable Systems
aa r X i v : . [ n li n . S I] F e b HAANTJES MANIFOLDS OF CLASSICAL INTEGRABLESYSTEMS
PIERGIULIO TEMPESTA AND GIORGIO TONDO
Abstract.
A general theory of classical integrable systems is proposed, basedon the geometry of the Haantjes tensor. We introduce the class of symplectic-Haantjes manifolds (or ω H manifold), as the natural setting where the notionof integrability can be formulated. We prove that the existence of suitableHaantjes structures is a necessary and sufficient condition for a Hamiltoniansystem to be integrable in the Liouville-Arnold sense.We also prove theorems ensuring the existence of a large family of com-pletely integrable systems, constructed starting from a prescribed Haantjesstructure.Furthermore, we propose a novel approach to the theory of separation ofvariables, intimately related to the geometry of Haantjes manifolds. A specialfamily of coordinates, that we shall call the Darboux-Haantjes coordinates, willbe introduced. They are constructed from the Haantjes structure associatedwith an integrable system, and allow the additive separation of variables ofthe Hamilton-Jacobi equation.Our analysis is performed in n degrees of freedom. We prove that some ofthe most classical examples of multidimensional Hamiltonian systems, as forinstance the Gantmacher class, possess a natural Haantjes structure. Finally,we present an application of our approach to the study of some models, as astationary reduction of the KdV hierarchy and a Drach-Holt type system; theseparability properties of the latter were not known. Contents
1. Introduction 22. Nijenhuis and Haantjes operators 53. The geometry of Haantjes operators 84. The theory of symplectic-Haantjes manifolds 144.1. Lenard-Haantjes chains 154.2. Symplectic-Nijenhuis manifolds and Lenard-Nijenhuis chains 185. Complete Integrability and Haantjes structures 185.1. Haantjes theorem for integrable systems 195.2. The analysis of Brouzet 216. New integrable models from Haantjes geometry 226.1. Harmonic functions and integrable systems 226.2. Waves and integrable systems 237. The theory of separation of variables in ω H manifolds 247.1. Darboux-Haantjes coordinates 247.2. Generators of a ω H manifold 257.3. Existence of DH coordinates 26 Date : February 9, 2016.
Introduction
Integrable systems play a relevant role in modern science: they are ubiquitousin many branches of modern mathematics and theoretical physics. This is the rea-son motivating in the last decades the search for intrinsic mathematical structuresunderlying the notion of integrability. In particular, the investigation of the prop-erties of exactly solvability of integrable systems led to the discovery of importantanalytic and geometric techniques. Finite-dimensional integrable models comingfrom classical or quantum mechanics, and the infinite-dimensional ones describedin terms of soliton equations, share indeed many geometric and algebraic properties.The study of the geometry of classical integrable systems has a long history,dating back to the works of Liouville, Jacobi, St¨ackel, Eisenhart, Arnold, etc. Inthis context, the bi-Hamiltonian approach has probed to be crucial.Essentially, a bi-Hamiltonian manifold is a differentiable manifold endowed witha pencil of Poisson structures [34]. In particular, the special class of ωN manifolds ,introduced in [43, 29], is characterized by a non-degenerate Poisson bivector (whoseinverse provides a symplectic structure ω ), and a compatible (1 ,
1) tensor field N with vanishing Nijenhuis torsion. Such a tensor, also called hereditary operator, hasa vanishing Nijenhuis torsion, as a consequence of the underlying bi-Hamiltonianstructure. The class of ωN manifolds offers a coherent approach to the constructionof separation variables, and has been successfully applied, for instance, to the studyof Gelfand-Zakharevich systems [24, 30, 18, 19, 20].The purpose of this paper is to present a new formulation of the notion of classicalintegrability, based on the theory of the Haantjes tensor . This tensor has beenintroduced in 1955 by Haantjes in [27], as a natural generalization of the Nijenhuistensor. Quite surprisingly, the relevance of the Haantjes differential-geometric workin the realm of integrable systems has not been recognized for a long time, withthe exception of some applications to Hamiltonian systems of hydrodynamic type[21, 11, 22].We shall define a new family of manifolds, called symplectic–Haantjes manifolds .We shall prove that integrability of a finite-dimensional system can be characterizedin terms of a set of commuting Haantjes operators, whose spectral and geometricproperties turn out to be particularly rich. The notion of
Lenard-Haantjes chain ,defined in this framework, is a natural extension in the context of Haantjes geometryof previous similar notions known in the literature, as that of Lenard-Magri chain
AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 3 [35] and of generalized Lenard chain [60, 17, 37], for quasi-bi-Hamiltonian systemsand their generalizations.There is a neat relation between the Haantjes geometry developed here and theknown Nijenhuis geometry. In fact, a subfamily of symplectic-Haantjes manifoldsis provided by the class of symplectic-Nijenhuis manifolds. Precisely, we shall showthat given a ω N manifold, one can construct, under mild assumptions, a ω H struc-ture by taking ( n −
1) independent powers of N . In this case, N will play the roleof a generator of a ω H structure.The notion of ω H manifolds is inspired by the very recent definition of Haantjesmanifold due to Magri [38, 39, 40, 41]. Our theory mainly differs from the fact that,besides the existence of n independent commuting Haantjes operators, we also allowa symplectic form ω compatible with the Haantjes operators to exist. Moreover,the Lenard-Haantjes chains of our theory are shorter than the ones defined in therecent Magri’s theory [39]. This is due to a weaker assumption that allows us todeal with both integrable and separable systems. This fact is an important noveltyof the present work that is not considered in the cited papers.Our main result concerning integrability is a theorem establishing that the exis-tence of a ω H manifold is a necessary and sufficient condition for a non-degenerateHamiltonian system to be integrable in the Liouville-Arnold sense. As a byproductof this theorem, we will be able to define new general classes of integrable mod-els possessing an assigned Haantjes geometry. Quite interestingly, the systems soobtained are related to analytic functions and to the wave equation.Another advantage of the present formulation `a la Haantjes (which also repre-sents the main motivation for our study), is its generality: Haantjes tensors areindeed a larger class of tensors than those of Nijenhuis. The proposed theory in-corporates essentially all the known results on integrability and separation of vari-ables of finite-dimensional systems that have been developed in a bi-Hamiltonianframework up to date, i.e. all the approaches based on Lenard chains and theirgeneralizations (as quasi-bi-Hamiltonian systems [8], etc).A noteworthy aspect is that the Haantjes operators appearing in the theory are not necessarily diagonalizable . This aspect represents a significant generalization ofthe ωN approach, where the operator N is diagonalizable by hypothesis. Moreover,our theory keeps the intrinsic simplicity enjoyed by the standard approach to theLenard-Magri chains for soliton hierarchies.At the same time, the theory of ω H manifolds is motivated by the crucial prob-lem of the construction of coordinate systems allowing the additive separation ofthe associated Hamilton-Jacobi (HJ) equation (the separation variables). This is,perhaps, the most important problem in the theory of classical integrable systems,to which many important contributions have been made.In 1904, Levi-Civita proposed in [33] a test for verifying whether a given Hamil-tonian is separable in an assigned coordinate system. Another important result,due to Benenti [4], states that a family of Hamiltonian functions { H i } ≤ i ≤ n is sep-arable in a set of canonical coordinates ( q , p ) if and only if they are in separableinvolution, i.e. if and only if they satisfy the relations(1) { H i , H j } | k = ∂H i ∂q k ∂H j ∂p k − ∂H i ∂p k ∂H j ∂q k = 0 , ≤ k ≤ n, where no summation over k is understood. However, such a theorem as well asthe Levi–Civita test are not constructive, since they do not help to find a complete PIERGIULIO TEMPESTA AND GIORGIO TONDO integral of the Hamilton–Jacobi equation. By contrast, a constructive definition ofseparation of variables (SoV) was given by Sklyanin [57] within the framework ofLax systems. In this setting, the Hamiltonian functions { H i } ≤ i ≤ n are separablein a set of canonical coordinates ( q , p ) if there exist n equations, called separationrelations, of the form(2) Φ i ( q i , p i ; H , . . . , H n ) = 0 det (cid:20) ∂ Φ i ∂H j (cid:21) = 0 , for i = 1 , . . . , n . They are named the Jacobi-Sklyanin separation equations for { H i } ≤ i ≤ n . The Jacobi-Sklyanin equations allow to construct a solution ( W, E ) ofthe HJ equation. In fact, by solving (2) with respect to p k = ∂W k ∂q k , one gets(3) W = X Z p k ( q ′ k ; H , . . . , H n ) | Hi = ai dq ′ k . However, the three above-mentioned criteria of separability are not intrinsic, sincethey require the explicit knowledge of the local chart ( q , p ) in order to be applied.To overcome such a drawback, in the last decades several approaches based on sym-plectic and Poisson geometry have been designed; they have offered a fundamentalgeometric insight into the theory of integrable systems. At the same time, a geo-metric version of integrability on differentiable manifolds can share new light onthe multiple connections among integrability, topological field theories, singularitytheory, co-isotropic deformations of associative algebras, etc.The problem of SoV can be recast and completely solved, in principle, in ourapproach. With respect to this problem, the main achievement is a theorem en-suring the existence, under mild hypotheses, of a set of distinguished coordinatesfrom the Haantjes structure associated with an integrable system, that we shall callthe Darboux-Haantjes coordinates . They represent separation coordinates for theHamilton-Jacobi equation associated with the system.Besides, we shall prove that a huge class of very general, multidimensional sepa-rable systems indeed possess a Haantjes geometry. The so-called telescopic systemsand the Gantmacher systems are all examples of very basic integrable models in n arbitrary dimensions possessing a ω H structure. Conversely, by using the un-derlying Haantjes structure, we will be able to determine separation coordinatesfor many integrable systems, including a family of Drach-Holt type systems, whoseseparability properties were not known.We also mention that the fundamental class of generalized St¨ackel systems andthe relevant example of the Jacobi-Calogero model have been studied in [63] as anapplication of the theory proposed in the present work.The structure of the paper is the following. In Section 2, we review the mainalgebraic properties of Nijenhuis and Haantjes tensors. In Section 3, the spectralproperties of Haantjes operators are discussed. In Section 4, we introduce themain geometrical structures needed for the discussion of integrability, i.e. the ω H manifolds; also, we clarify their relation with ωN manifolds. Section 5 containsthe theorem that characterizes complete integrability via the Haantjes geometry.In Section 6, new integrable models related to analytic functions and to the waveequation are deduced from suitable Haantjes structures. Section 7 is devoted to theproblem of separability of the Hamilton-Jacobi equation in the context of Haantjesgeometry. In particular, a theorem guaranteeing the existence of the DH coordinatesis proved. In Section 8, a procedure for the construction of Haantjes structures AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 5 for a given integrable system with two degrees of freedom is proposed. Also, therelevant example of the superintegrable Post-Winternitz system, whose separationcoordinates are still not known, is worked out. Some applications of our theoryof separation of variables are proposed in Section 9. Open problems and futureresearch plans are sketched in the final Section 10.2.
Nijenhuis and Haantjes operators
The integrability of a dynamical system defined over a manifold M essentiallyamounts to find privileged coordinate webs in which the equations of motion decou-ple. The natural frames of such webs, being obviously integrable, can be character-ized in a tensorial manner as eigen-distributions of a suitable class of (1 ,
1) tensorfields, i.e. the ones with vanishing Nijenhuis or Haantjes tensor. In this section,we review some basic algebraic results concerning the theory of such tensors. Fora more complete treatment, see the original papers [27, 52] and the related ones[53, 23].Let M be a differentiable manifold and L : T M → T M be a (1 ,
1) tensor field,i.e., a field of linear operators on the tangent space at each point of M . Definition 1.
The Nijenhuis torsion of L is the skew-symmetric (1 , tensor fielddefined by (4) T L ( X, Y ) := L [ X, Y ] + [ L X, L Y ] − L (cid:16) [ X, L Y ] + [ L X, Y ] (cid:17) , where X, Y ∈ T M and [ , ] denotes the commutator of two vector fields. In local coordinates x = ( x , . . . , x n ) , the Nijenhuis torsion can be written in theform(5) ( T L ) ijk = n X α =1 (cid:18) ∂ L ik ∂x α L αj − ∂ L ij ∂x α L αk + (cid:16) ∂ L αj ∂x k − ∂ L αk ∂x j (cid:17) L iα (cid:19) , amounting to n ( n − / Definition 2.
The Haantjes tensor associated with L is the (1 , tensor fielddefined by (6) H L ( X, Y ) := L T L ( X, Y ) + T L ( L X, L Y ) − L (cid:16) T L ( X, L Y ) + T L ( L X, Y ) (cid:17) . The skew-symmetry of the Nijenhuis torsion implies that the Haantjes tensor isalso skew-symmetric. Its local expression is(7) ( H L ) ijk = n X α,β =1 (cid:18) L iα L αβ ( T L ) βjk +( T L ) iαβ L αj L βk − L iα (cid:16) ( T L ) αβk L βj +( T L ) αjβ L βk (cid:17)(cid:19) . We shall first consider some specific cases, in which the construction of the Nijenhuisand Haantjes tensors will be particularly simple.
Example 3.
Let L be a field of operators that assumes a diagonal representation (8) L = n X i =1 l i ( x ) ∂∂x i ⊗ d x i , PIERGIULIO TEMPESTA AND GIORGIO TONDO in some local chart x = ( x , . . . , x n ) . Its Nijenhuis torsion is given by (9) ( T L ) ijk = ( l j − l k ) (cid:18) ∂l j ∂x k δ ij + ∂l k ∂x j δ ik (cid:19) . It is evident that ( T L ) ijk = 0 if i = j and i = k or if j = k . Thus, we can limitourselves to analyze the n ( n − components (10) ( T L ) jjk = ( l j − l k ) ∂l j ∂x k , j = k . If ∂l j ∂x k = 0 , each component vanishes if and only if l j ( x ) ≡ l k ( x ) . Therefore, wecan state the following Lemma 4.
Let L be the diagonal field of operators (8) , and suppose that its Ni-jenhuis torsion vanishes. Let us denote with ( i , . . . , i j , . . . , i r ) , r ≤ n a subset of (1 , , . . . , n ) . If the j -th eigenvalue of L depends on the variables ( i , . . . , i j , . . . , i r ) ,then (11) l j ( i , . . . , i j , . . . , i r ) ≡ l i ≡ l i ≡ . . . = l i r Thus, apart when each eigenvalue is constant, we can distinguish several cases,ensuring that the Nijenhuis torsion of a diagonal operator vanishes. For instance,i) l j ( x ) = λ j ( x j ) , j = 1 , . . . , n ⇒ n simple eigenvalues ii) l j ( x ) = λ ( x ) j = 1 , . . . , n ⇒ , represent the extreme cases. An exhaustive analysis of all intermediate possibilitiesis left to the reader. Example 5.
Let dimM = 2 . Then, it easy to prove by a straightforward compu-tation that the Haantjes tensor of any field of smooth operators vanishes.
Example 6.
Let L be the diagonal operator of Example 3. Its Haantjes tensorreads (12) ( H L ) ijk = ( l i − l j )( l i − l k )( T L ) ijk , where ( T L ) ijk is given by eq. (9) . The following proposition is a direct consequence of eqs. (9) and (12).
Proposition 7.
Let L a smooth field of operators. If there exists a local coordinatechart { ( x , . . . , x n ) } , where L assumes the diagonal form (8) , then the Haantjestensor of L vanishes. Due to the relevance of the Haantjes (Nijenhuis) vanishing condition, we proposethe following definition.
Definition 8.
A Haantjes (Nijenhuis) field of operators is a field of operators whoseHaantjes (Nijenhuis) tensor identically vanishes.
As usual, the transposed operator L T : T ∗ M T ∗ M is defined as the trans-posed linear map of L with respect to the natural pairing between a vector spaceand its dual space < L T α, X > = < α, L X > α ∈ T ∗ M, X ∈ T M.
A relevant property of Nijenhuis operators, which is a direct consequence of Theor.17 below and eq. (11), and usually is not satisfied by Haantjes operators, is thefollowing
AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 7
Proposition 9.
The differentials of the eigenvalues λ i ( x ) of a diagonalizable Ni-jenhuis operator N are eigenforms of its transposed operator N T (13) N T d λ i = λ i d λ i . Very similar statements can be found in [52] and [25].The product of a Nijenhuis operator with a generic function is no longer a Ni-jenhuis operator, as is proved by the following identity(14) T f L ( X, Y ) = f T L ( X, Y )+ f (cid:16) ( L X )( f ) L Y − ( L Y )( f ) L X + Y ( f ) L X − X ( f ) L Y (cid:17) , which holds for any function f ∈ C ∞ ( M ). Instead, the differential and algebraicproperties of a Haantjes operator are much richer, as follows from these remarkableresults. Proposition 10. [9] . Let L be a field of operators. The following identity holds (15) H f I + g L ( X, Y ) = g H L ( X, Y ) , where f, g : M → R are C ∞ ( M ) functions, and I denotes the identity operator in T M .Proof.
See Proposition 1, p. 255 of [9]. (cid:3)
Proposition 11. [10] . Let L be an operator with vanishing Haantjes tensor in M . Then for any polynomial in L with coefficients a j ∈ C ∞ ( M ) , the associatedHaantjes tensor vanishes, i.e. (16) H L ( X, Y ) = 0 = ⇒ H ( P j a j ( x ) L j ) ( X, Y ) = 0 . Proof.
See Corollary 3.3, p. 1136 of [10]. (cid:3)
Propositions 10 and 11 imply that the powers of a single Haantjes operator (bycontrast with the case of a Nijenhuis operator) generate a module over the ring ofsmooth functions on M .Let us introduce an interesting example of Nijenhuis and Haantjes operatorsdrawn from the realm of Rational Mechanics. Example 12.
Let M = { ( P γ , m γ ) ∈ ( E n , R ) } be a finite system of mass points(possibly with m γ < ) in the n -dimensional affine Euclidean space E n . Let usconsider the (1 , tensor field defined by (17) E P ( ~v ) = X γ m γ (cid:0) ( P γ − P ) · ~v (cid:1) ( P γ − P ) ~v ∈ T P E n ≡ E n , called the planar inertia tensor (or Euler tensor in Continuum Mechanics), and theinertia tensor field, given by (18) I P ( ~v ) = X γ m γ (cid:18) | P γ − P | ~v − (cid:0) ( P γ − P ) · ~v (cid:1) ( P γ − P ) (cid:19) . They are related by the formulas (19) I P = trace ( E P ) I n − E P , E P = trace ( I P ) n − I n − I P , PIERGIULIO TEMPESTA AND GIORGIO TONDO where I n is the identity operator in E n . Both of them are symmetric w.r.t. theEuclidean scalar product, so that they are diagonalizable at any point of E n . Fur-thermore, by virtue of (19) they commute; consequently, they can be simultaneouslydiagonalized.If G is the center of mass of M , defined by G − P = 1 m X γ ( P γ − P ) m := X γ m γ m ∈ R \ { } , the following Huygens-Steiner transposition formulas hold E P ( ~v ) = E G ( ~v ) + m (cid:0) ( P − G ) · ~v (cid:1) ( P − G ) , (20) I P ( ~v ) = I G ( ~v ) + m | P − G | − m (cid:0) ( P − G ) · ~v (cid:1) ( P − G ) . (21) From eqs. (20) and (21) it follows that in the Cartesian coordinates ( x , . . . , x n ) with origin in G , defined by the common eigendirections of E G and I G , we have ( E P ) ij = λ i ( G ) δ ij + m x i x j , (22) ( I P ) ij = l i ( G ) δ ij + m (cid:18) n X α =1 x α − x i x j (cid:19) , (23) where λ i ( G ) and l j ( G ) denote, respectively, the eigenvalues of the tensor fields E and I , both evaluated at the point G . In [5, 6] it has been proved that the Nijenhuistorsion of E vanishes; then its Haantjes tensor also vanishes. Furthermore, weobserve that the torsion of I reads (24) ( T I ) ijk = 2 m n X α =1 (cid:18) x α I α (cid:0) δ ik − δ ij (cid:1) + x k I ij − x j I ik (cid:19) , i.e. it is not identically zero, although its Haantjes tensor vanishes as a consequenceof the identity (15) , applied to the relation (19) . Other relevant examples of Haantjes operators that are also Killing tensors in aRiemannian manifold can be found in [63].3.
The geometry of Haantjes operators
As we noted in Proposition 7, the Haantjes tensor H L of an operator L has arelevant geometrical meaning: its vanishing is a necessary condition for the eigen-distributions of L to be integrable. To clarify this point, let us recall that a referenceframe is a set of n vector fields { Y , . . . , Y n } such that, at each point x belongingto an open set U ⊆ M , they form a basis of the tangent space T x U . Two frames { X , . . . , X n } and { Y , . . . , Y n } are said to be equivalent if n nowhere vanishingsmooth functions f i do exist such that X i = f i ( x ) Y i , i = 1 , . . . , n . A natural frame is the frame associated to a local chart { ( x , . . . , x n ) } and denotedas ( ∂∂x , . . . , ∂∂x n ). Definition 13. A holonomic frame is a reference frame equivalent to a naturalframe. AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 9
In other words, to say that a frame { Y , . . . , Y n } is holonomic there must exist n nowhere vanishing functions f i and a local chart ( x , . . . , x n ) such that(25) Y i = f i ( x ) ∂∂x i , i = 1 , . . . , n . Proposition 14. [7]
A reference frame in a manifold M is a holonomic frame ifand only if it satisfies one the two equivalents conditions: • each two-dimensional distribution generated by any two vector fields Y i , Y j is Frobenius integrable; • each ( n − -dimensional distribution E i generated by all the vector fieldsexcept Y i is Frobenius integrable. Definition 15.
A field of operators L is said to be semisimple (or diagonalizable)if there exists a reference frame formed by (proper) eigenvector fields of L . Thisframe will be called an eigen-frame of L . Moreover, L is said to be simple if all itseigenvalues are pointwise distinct, namely if l i ( x ) = l j ( x ) , i, j = 1 , . . . , n , ∀ x ∈ M . Proposition 7 amounts to say that if an operator admits a local chart in which ittakes a diagonal form, then its Haantjes tensor necessarily vanishes, therefore theassociate coordinate frame is an eigenframe that is trivially holonomic. In 1955,Haantjes proved in [27] that the vanishing of the Haantjes tensor of a semisimple operator L is also a sufficient condition to ensure the integrability of each of itseigen-distributions (with constant rank) and the existence of local coordinate chartsin which L takes a diagonal form. We call such coordinates Haantjes coordinates .Furthermore, he stated that the vanishing of the Haantjes tensor of an operator L is also a sufficient (but not necessary) condition to ensure the integrability of eachof its generalized eigen-distributions (with constant rank) and the existence of localcoordinate charts in which L takes a Jordan form. An equivalent statement of theabove-mentioned results is that a Haantjes field of operators admits a generalizedeigen-frame that is a holonomic frame.Let us denote with
Spec ( L ) := { l ( x ) , l ( x ) , . . . , l s ( x ) } the set of the distincteigenvalues of an operator L , which we always assume real in all the forthcomingconsiderations, and with(26) D i = Ker (cid:16) L − l i ( x ) I (cid:17) ρ i , i = 1 , . . . , s the i -th generalized eigen-distribution, that is the distribution of all the generalizedeigenvector fields corresponding to the eigenvalue l i ( x ). In eq. (26), ρ i denotes theRiesz index of l i , namely the minimum integer such that(27) Ker (cid:16) L − l i ( x ) I (cid:17) ρ i ≡ Ker (cid:16) L − l i ( x ) I (cid:17) ρ i +1 . When ρ i = 1, D i is a proper eigen-distribution. Definition 16.
A generalized eigen-frame (or a Jordan eigen-frame) of a field ofoperators L is a frame of generalized eigenvectors of L . Theorem 17. [27] . Let L be a field of operators, and assume that the rank of eachgeneralized eigen–distribution D i is independent of x ∈ M . The vanishing of theHaantjes tensor (28) H L ( X, Y ) = 0 ∀ X, Y ∈ T M is a sufficient condition to ensure the integrability of each generalized eigen–distribution D i and of any direct sum D i ⊕ D j ⊕ . . . ⊕ D k (where all indices i, j, . . . , k are dif-ferent). In addition, if L is semisimple, the condition (28) is also necessary. In the original paper by Haantjes, the proof of Theorem 17 is explicitly madeonly for the case of a semisimple operator. Below, we present the proof for themore general case of an operator admitting generalized eigenvectors with arbitraryRiesz index.Here, for the sake of simplicity, we focus only on two eigenvalues of L , µ and ν ,possibly coincident. Let us denote by X α , Y β , two fields of generalized eigenvectors,with index α , respectively β , corresponding to the eigenvalues µ , resp. ν , andbelonging to a Jordan chain in D µ , resp. D ν ,(29) L X α = µX α + X α − , L Y β = νY β + Y β − , ≤ α ≤ ρ µ , ≤ β ≤ ρ ν , where X and Y are the null vector fields. Then, it holds true that(30) X α ∈ Ker (cid:16) L − µ I (cid:17) ρ µ , Y β ∈ Ker (cid:16) L − ν I (cid:17) ρ ν . Evaluating the Nijenhuis torsion on such eigenvector fields, we get T L ( X α , Y β ) = (cid:16) L − µ I (cid:17)(cid:16) L − ν I (cid:17) [ X α , Y β ] + ( µ − ν ) (cid:16) X α ( ν ) Y β + Y β ( µ ) X α (cid:17) − (cid:16) L − µ I (cid:17) [ X α , Y β − ] − (cid:16) L − ν I (cid:17) [ X α − , Y β ] + [ X α − , Y β − ] − (cid:16) X α ( ν ) Y β − + Y β − ( µ ) X α (cid:17) + (cid:16) X α − ( ν ) Y β + Y β ( µ ) X α − (cid:17) , where X α ( ν ) denotes the Lie derivative of the eigenvalue ν ( x ) with respect to thevector field X α . The analogous relation for the Haantjes tensor is(31) H L ( X α , Y β ) = X i,j =0 ( − i + j (cid:18) i (cid:19)(cid:18) j (cid:19)(cid:16) L − µ I (cid:17) − i (cid:16) L − ν I (cid:17) − j [ X α − i , Y β − j ] . Lemma 18.
Let L be a field of operators and X α , Y β , be two of its fields ofgeneralized eigenvectors in D µ , belonging to possibly different Jordan chains. If (32) H L ( D µ , D µ ) = 0 , then their commutator satisfies the relation (33)[ X α , Y β ] ∈ Ker (cid:16) L − µ I (cid:17) α + β +2 ≡ Ker (cid:16) L − µ I (cid:17) min( α + β +2 ,ρ µ ) ⊆ Ker (cid:16) L − µ I (cid:17) ρ µ , where min( , ) means the minimum of its arguments.Proof. If α = β = 1 and µ = ν , eq. (31) implies that [ X , Y ] ∈ (cid:16) L − µ I (cid:17) . Byinduction over ( α + β ), and applying the operator (cid:16) L − µ I (cid:17) α + β − to both membersof eq. (31) it follows that [ X α , Y β ] ∈ Ker (cid:16) L − µ I (cid:17) α + β +2 . (cid:3) AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 11
Proposition 19.
Let L be a field of operators. An eigen-distribution D µ with Rieszindex ρ µ is integrable if (34) H L ( D µ , D µ ) = 0 . In particular, if ρ µ = 1 , the converse is also true.Proof. Lemma 18 immediately implies that the Frobenius integrability conditionfor D µ (35) [ D µ , D µ ] ⊆ D µ is fulfilled. In particular, if ρ µ = 1, every µ -eigenvector of L is a proper eigenvector,and from eq. (31) one infers that H L ( D µ , D µ ) = 0 ⇐⇒ [ X , Y ] ∈ Ker (cid:16) L − µ I (cid:17) ≡ Ker (cid:16) L − µ I (cid:17) = D µ . (cid:3) Lemma 20.
Let L be a Haantjes operator. The commutator of two generalizedeigenvector fields of L , with different eigenvalues µ , ν , fulfills the relation [ X α , Y β ] ∈ Ker (cid:16) L − µ I (cid:17) α +1 ⊕ Ker (cid:16) L − ν I (cid:17) β +1 (36) ≡ Ker (cid:16) L − µ I (cid:17) min( α +1 ,ρ µ ) ⊕ Ker (cid:16) L − ν I (cid:17) min( β +1 ,ρ ν ) ⊆ Ker (cid:16) L − µ I (cid:17) ρ µ ⊕ Ker (cid:16) L − ν I (cid:17) ρ ν , with ≤ α ≤ ρ µ , ≤ β ≤ ρ ν .Proof. If α = β = 1 and µ = ν , eq. (31) implies that [ X , Y ] ∈ Ker (cid:16) L − µ I (cid:17) ⊕ Ker (cid:16) L − ν I (cid:17) . By induction over ( α + β ), applying the operator (cid:16) L − µ I (cid:17) α − (cid:16) L − ν I (cid:17) β − to both members of (31) the assertion follows. (cid:3) It is immediate to ascertain that the above Lemma implies [ D µ , D ν ] ⊂ D µ ⊕ D ν , so that the following result holds Proposition 21.
Let L be a Haantjes operator, and D µ , D ν be two distributionswith Riesz indices ρ µ and ρ ν , respectively. Then, the distribution D µ ⊕ D ν ≡ Ker (cid:16) L − µ I (cid:17) ρ µ ⊕ Ker (cid:16) L − ν I (cid:17) ρ ν , µ = ν is integrable. The Haantjes Theorem 17 is an immediate consequence of Propositions 19 and21.In [16] and [26], the integrability of the eigendistributions of a Nijenhuis operatorwith generalized eigenvectors of Riesz index 2 was proved. However, the case ofHaantjes operators was not considered. On the other hand, to the best of ourknowledge, the proofs of the Haantjes theorem available in the literature (see forinstance [23], [25]) are based on the more restrictive assumption that the Haantjesoperator be diagonalizable.
Let us show in detail how to determine a coordinate system that, under theassumption of Theorem 17, provides a Jordan form for L . Denote by(37) E i := Im (cid:16) L − l i I (cid:17) ρ i = s M j =1 , j = i D j , i = 1 , . . . , s the ( n − r i )-dimensional distribution spanned by all the generalized eigenvectors of L , except those associated with the eigenvalue l i . Such a distribution will be calleda characteristic distribution of L . Let E ◦ i denote the annihilator of the distribution E i . Since L has real eigenvalues by assumption, the tangent and cotangent spacesof M can be locally decomposed as(38) T M = s M i =1 D i , T ∗ M = s M i =1 E ◦ i . Moreover, each characteristic distribution E i is integrable by virtue of Theorem17. We shall denote by E i the associated foliation and by S i ( x ) the connectedleave through x belonging to E i . Thus, the set ( E , E , . . . , E s ) generates as manyfoliations ( E , E , . . . , E s ) as the eigenvalues of L . Such a set of foliations will bereferred to as the characteristic web of L and the leaves S i ( x ) of each foliation E i as the characteristic fibers of the web. Definition 22.
A collection of r i smooth functions will be said to be adapted to afoliation E i of the characteristic web of L if the level sets of such functions coincidewith the characteristic fibers of the foliation. Definition 23.
A parametrization of the characteristic web of an operator L is anordered set of n independent smooth functions ( f , . . . , f n ) such that each orderedsubset ( f i , . . . , f i r ) is adapted to the i -th characteristic foliation of the web: (39) f i k |S i ( x ) = const ∀S i ( x ) ∈ E i , k = 1 , . . . , r, i r = i + r i . In this case, we shall say that the collection of functions is adapted to the web andthat each of them is a characteristic function . Corollary 24.
Assume that L has real eigenvalues. Then, the vanishing of theHaantjes tensor of L is sufficient to assure that it admits an equivalence class ofholonomic generalized eigenframes. Furthermore, if L is semisimple the vanishingof the Haantjes tensor is also a necessary condition. In addition, if L is simple each eigenframe is holonomic.Proof. Since each characteristic distribution E i is integrable by virtue of Haantjes’sTheorem 17, in the corresponding annihilator E ◦ i one can find r i exact one-forms(d x i , . . . , d x i r ) that provide functions ( x i , . . . , x i r ) adapted to the characteristicfoliation E i . Collecting together all these functions, one can construct a set of n independent coordinates ( x , . . . , x n ) adapted to the characteristic web.The natural frame ( ∂∂x , . . . , ∂∂x n ) turns out to be an eigenframe. In fact, as(40) D ◦ i = s M j =1 , j = i E ◦ j , the components of any generalized eigenvector W ∈ D i satisfy the conditions(41) W j = W ( x j ) = 0 , j = i , . . . , i r . AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 13
Thus, we have that W = P rk =1 W ( x i k ) ∂∂x ik , therefore(42) D i = Span (cid:8) ∂∂x i , . . . , ∂∂x i r (cid:9) , and each frame equivalent to ( ∂∂x , . . . , ∂∂x n ) is a holonomic eigenframe.Moreover, if L is semisimple Prop. 7 holds true and each eigenframe fulfills theconditions of Prop. 14. (cid:3) A local chart adapted to the characteristic web of L can be computed by usingthe transposed operator L T . Let us denote with(43) Ker (cid:16) L T − l i ( x ) I (cid:17) ρ i the i -th distribution of the generalized eigen 1-forms with eigenvalue l i ( x ), whichfulfills the property(44) Ker ( L T − l i I ) ρ i = (cid:18) Im (cid:16) L − l i I (cid:17) ρ i (cid:19) ◦ = E ◦ i Such a property implies that each generalized eigenform of L T annihilates all gen-eralized eigenvectors of L with different eigenvalues. Moreover, it allows to provethat Proposition 25.
Let L be a Haantjes operator. The differentials of the char-acteristic coordinate functions are exact generalized eigenforms for the transposedoperator L T . Conversely, each (locally) exact generalized eigenform of L T providesa characteristic function for the Haantjes web of L . The characteristic functions of a Haantjes operator are characterized by thefollowing simple property.
Proposition 26.
A function h on M is a characteristic function of a Haantjesoperator associated with the eigenvalue l i if and only if, given a set of local coor-dinates adapted to the characteristic web ( x , . . . , x n ) , h depends, at most, on thesubset of coordinates ( x i , . . . , x i r ) that are constant over the leaves of the foliation E i .Proof. If h = h ( x i , . . . , x i r ), it is constant on the leaves of E i , then d h ∈ E ◦ i .Viceversa, if we assume that d h ∈ E ◦ i , then it can be expressed in terms of a linearcombination of { d x i , . . . , d x i r } only. The thesis follows from the exactness of d h . (cid:3) Remark 27.
In particular, if L is a semisimple Nijenhuis operator, its eigenvaluesare characteristic functions for the Haantjes web, according to Proposition 9. Remark 28. [26]
Let us suppose that a generic operator L admits a symmetry,i.e. a vector field X such that (45) L X ( L ) = 0 . In this case, the operator L will be called a recursion operator for X . Then, theeigenvalues of L as well are invariant along the flow of X and the correspondinggeneralized eigen-distributions are stable, i.e. (46) L X ( l i ) = 0 , L X ( D i ) ⊆ D i , L X ( E ◦ i ) ⊆ E ◦ i ∀ i = 1 , , . . . , s . The theory of symplectic-Haantjes manifolds
In this section, we shall introduce the new class of symplectic-Haantjes manifolds,that we shall call the ω H manifolds by analogy with the known ωN ones [43]. Themain reason to define these manifolds is that they provide a natural setting in whichthe theory of Hamiltonian integrable systems can be properly formulated. Definition 29.
A symplectic-Haantjes or ω H manifold ( M, ω, K , K , . . . , K n − ) is a symplectic manifold of dimension n , endowed with n endomorphisms of thetangent bundle of M K α : T M T M α = 0 , . . . , n − , which satisfy the following conditions: • K = I . • Their Haantjes tensor vanishes identically, i.e. (47) H K α ( X, Y ) = 0 ∀ X, Y ∈ T M, α = 0 , . . . , n − . • The endomorphisms are compatible with ω (or equivalently, with the corre-sponding symplectic operator Ω := ω ♭ ): (48) K Tα Ω = Ω K α , α = 0 , . . . , n − , that is, the operators Ω K α are skew symmetric. • The endomorphisms are compatible with each others, namely they form acommutative ring K (49) K α K β = K β K α , α, β = 0 , . . . , n − , and also generate a module over the ring of smooth functions on M : (50) H (cid:0) P n − α =0 a α ( x ) K α (cid:1) ( X, Y ) = 0 , ∀ X, Y ∈ T M , where a α ( x ) are arbitrary smooth functions on M.The ( n + 1) -ple ( ω, K , K , . . . , K n − ) will be called the ω H structure associatedwith the ω H manifold and K the Haantjes module (ring). The above conditions can be re-cast by saying that, essentially, the endomor-phisms K α are Haantjes operators, compatible both with ω and with each others.Moreover, the assumption (49) ( respectively (50)) assures that each operator be-longing to the ring (respectively the module) K is a Haantjes operator compatiblewith ω and with the original Haantjes operators { K , K , . . . , K n − } .As a consequence of the above conditions, we get the following simple propositionthat turns out to be crucial for many results of the present theory. Proposition 30.
Given a ω H structure, let P = Ω − be the Poisson operatorassociated with the symplectic structure ω . Any composed operator Ω K α , P K Tβ , K Tβ Ω K α , K α P K Tβ , Ω (cid:0) K α − f ( x ) I (cid:1) m , α, β = 0 , . . . , n − , m ∈ N is skew sym-metric. For instance, it has important consequences on the spectrum of the Haantjesoperators.
Corollary 31.
Given a n -dimensional ω H manifold M , we will suppose that thenumber of distinct eigenvalues of each operator K α , as well as the dimension ofthe related eigenspaces, do not depend on the point x , at least for x in a dense open AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 15 subset of M . Then, every generalized eigen–distribution Ker ( K α − l ( α ) i I ) m i ( α ) , m ( α ) i = 1 , . . . , ρ ( α ) i , is even-dimensional. Therefore each eigenvalue l ( α ) i ( x ) has bothits geometric multiplicity ( dim (cid:0) Ker ( K α − l ( α ) i I ) (cid:1) ) and its algebraic multiplicity( dim (cid:0) Ker ( K α − l ( α ) i I ) (cid:1) ρ i ( α ) (cid:1) ) even.Proof. In a ω H manifold every generalized eigen–distribution Ker ( K α − l ( α ) i I ) m i has the same dimension of the kernel of the operator Ω ( K α − l ( α ) i I ) m i , which isskew-symmetric by virtue of Proposition 30. (cid:3) Due to the above corollary, the number of the eigenvalues of a Haantjes operator K α is less or equal to n , therefore their algebraic multiplicity is at least 2. Definition 32.
If the number of distinct eigenvalues of a Haantjes operator K ∈ K is exactly n , we say that such an operator is maximal. Let us denote with m K ( λ ) the minimal polynomial of K . Lemma 33.
A Haantjes operator of a ω H structure is maximal if and only if itsminimal polynomial is the product of n linear factors (51) m K ( λ ) = n Y i =1 (cid:16) λ − λ i (cid:17) . We shall also consider a particular class of ω H manifold, especially relevant for theapplications. Definition 34. A ω H manifold ( M, ω, K , K , . . . , K n − ) endowed with a vectorfield X ∈ T M such that (52) L X ( ω ) = 0 , L X ( K α ) = 0 , α ∈ { . . . , n − } , will be called a symplectic-Haantjes manifold with a (locally) Hamiltonian symmetryvector field X . According to Remark 28, each operator K α is a recursion operator for the Hamil-tonian symmetry X and their eigenvalues are integrals of motion for X .4.1. Lenard-Haantjes chains.
The theory of Lenard chains is a fundamentalpiece of the geometric approach to soliton hierarchies. Lenard chains have beenintroduced in order to construct integrals of motion in involution for infinite-dimensional Hamiltonian systems [34, 35] (see also [56], for a brief history aboutthe origin of the name “Lenard chains”). However, only recently some non trivialgeneralizations of Lenard chains have probed to be useful in the study of separationof variables for finite-dimensional Hamiltonian systems (see [48, 49, 62, 17, 19] andreference therein).Hereafter, as a byproduct of our approach `a la Haantjes, we propose a furthergeneralization of the standard notions of the theory, which has the advantage to beboth simple and directly connected to the theory of classical integrable systems.
Definition 35.
Let ( M, ω, K , K , . . . , K n − ) be a n -dimensional ω H manifoldand { H j } ≤ j ≤ n be n independent functions which satisfy the following relations (53) d H j = ˜ K Tα d H , j = α + 1 , α = 0 , . . . , n − , H := H , where (54) ˜ K α := n − X β =0 a ij ( x ) K β , i = α + 1 , j = β + 1 , α, β = 0 , . . . , n − , and a ij ( x ) are suitable smooth functions on M , that are assumed to satisfy theproperties i) a k = δ k , k = 1 , , . . . , n ; ii) a jk are elements of an invertible matrix-valued function A , called the tran-sition matrix between { K , K , . . . , K n − } and { ˜ K , ˜ K , . . . , ˜ K n − } .Under these conditions, we shall say that the functions { H j } ≤ j ≤ n form a Lenard–Haantjes chain generated by the function H . Remark 36.
The operators ˜ K α belong to the Haantjes module K generated by theoperators K α , so they are also Haantjes operators as a consequence of the assump-tion (50) . Moreover, they are compatible with ω and commute with each other. Con-sequently, the operators ˜ K α endow M with the ω H structure ( M, ω, ˜ K , ˜ K , . . . , ˜ K n − ) ,adapted to the function H , that we shall call a modified ω H structure. To enquire about the existence of Lenard-Haantijes chains, we have to considerthe co-distribution D ◦ H generated by H through the (transposed of) the Haantjesoperators K α (55) D ◦ H := Span { dH, K T d H, . . . , K Tn − d H } , and to compare the distribution D H of the vector fields annihilated by D ◦ H , with thedistribution of the vector fields symplectically orthogonal to D H . We shall denoteit by D ⊥ H ; it can be represented as(56) D ⊥ H = P ( D ◦ H ) = Span { X H , K X H , K X H , . . . , K n − X H } , where X H = P d H is the Hamiltonian vector field with Hamiltonian function H .Indeed, from Proposition 30 we deduce the following result. Proposition 37.
Let ( M, ω, K , K , . . . , K n − ) be a n -dimensional ω H mani-fold, and H be a smooth function on M . The relation (57) D ⊥ H ⊆ D H holds true. Therefore, D H is a co-isotropic distribution and D ⊥ H is an isotropic one.Moreover, if dim ( D H ( x )) = n ∀ x ∈ M they coincide and form a Lagrangian distribution.Proof. Each vector field belonging to D ⊥ H is annihilated by the one-form belongingto D ◦ H as < K Tα d H, K β X H > = < d H, K α K β P d H > (30) = 0 (cid:3)
The following theorem states the necessary and sufficient conditions to ensurethe existence of a Lenard-Haantijes chain generated by a function H . AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 17
Theorem 38.
Let ( M, ω, K , K , . . . , K n − ) be a n -dimensional ω H manifold,and H be a smooth function on M . Let D ◦ H be the co-distribution spanned by theset of 1-forms (58) β α +1 := K Tα d H α = 0 , . . . n − , that we assume to be of rank n (independent on x ) and let D H be the distributionof the vector fields annihilated by them. Then, the function H generates a Lenard-Haantijes chain (53) if and only if D ◦ H ( D H ) is integrable in the sense of Frobenius.Under this assumption, the foliation F H of D ◦ H is a Lagrangian foliation.Proof. By definition, the Lenard-Haantijes chain (53) contains n exact 1-forms.Therefore they generate the integrable Lagrangian distribution(59) D ◦ = Span { d H , . . . , d H n } which coincides with D ◦ H , by virtue of the linear relation (54) among the K α andthe ˜ K α .Viceversa, let D H be integrable and F H be the associated foliation. Then, thereexist n independent functions ( H , H , . . . , H n ) which are constant on the leaves of F H . Their differentials belong to D ◦ H , hence(60) d H i = n X j =1 a ij ( x ) β j (58) = n X j =1 a ij ( x ) K Tj − d H j = 1 , . . . , n.
The choice H = H implies that ˜ K = I and, consequently, property i) in theDefinition 35. The fact that { β , . . . , β n } and { dH , . . . , dH n } pointwise are twodifferent basis of D ◦ H ( x ), implies the property ii) . In this manner, we can con-struct the modified ω H structure (54), equivalent to the original one but adaptedto the Lenard-Haantjes chain (53) generated by the function H . Furthermore,the functions ( H , H , . . . , H n ) are in involution by virtue of Theorem 41 below,therefore they generate a Lagrangian foliation w.r.t. the symplectic form ω . Sucha foliation coincides with the foliation F to which the Hamiltonian vector fields( X H , X H , . . . , X H n ) are tangent. (cid:3) Corollary 39.
Let ( M, ω, K , K , . . . , K n − ) be a n -dimensional ω H manifold,and H be the Hamiltonian function of a Hamiltonian vector field integrable in thesense of Liouville. Also, let ( H , H , . . . , H n ) be a set of independent integrals ininvolution and D ◦ the co-distribution spanned by their differentials. Such integralsform a Lenard-Haantjes chain generated by H = H if and only if H satisfies thefollowing conditions (61) K Tα d H ∈ D ◦ , α = 0 , , . . . , n − . Proof.
Conditions (61) are equivalent to say that D ◦ H ⊆ D ◦ . Therefore, D ◦ H ≡ D ◦ as they both have, by assumption, the same rank n . Thus, D H is integrable and,by virtue of the previous theorem, it follows that the function H is the generatorof the Lenard-Haantjes chain formed by ( H , H , . . . , H n ).Conversely, if ( H , H , . . . , H n ) form a Lenard-Haantjes chain generated by H ,by Definition (53) it follows that ˜ K Tα d H ∈ D ◦ , for α = 0 , , . . . , n −
1. Then, sinceformula (54) is invertible by hypothesis, conditions (61) are fulfilled. (cid:3)
Symplectic-Nijenhuis manifolds and Lenard-Nijenhuis chains.
A par-ticular but especially relevant class of ω H structures is represented by the ωN man-ifolds [43, 36]. They are symplectic manifolds endowed with a single endomorphismof the tangent bundle, N : T M T M that satisfies the following conditions: • its Nijenhuis torsion (4) vanishes identically, i.e. ∀ X, Y ∈ T M (62) T N ( X, Y ) = 0; • it is compatible with ω , that is, the tensor P = N Ω − is also a Poissontensor and is compatible with P := Ω − , i.e. the Schouten bracket of P and P vanishes.The above conditions, that amount to say that N is a Nijenhuis (or hereditary)operator compatible with ω , in turn ensure that the ω N structures represent aspecial class of bi-Hamiltonian structures . Remark 40.
The requirement that the Nijenhuis torsion of N vanishes impliesthat the Haantjes tensor (6) of N vanishes as well. Thus Nijenhuis operators area special class of Haantjes operators. Then, given a ω N manifold, one can construct directly a ω H structure by choos-ing as Haantjes operators the first ( n −
1) powers of N (63) K α ≡ N α , α = 0 , , . . . , n − , provided that they are linearly independent. It is easy to prove that the Haantjesoperators so constructed are compatible with ω , due to the algebraic compatibilitybetween N and ω . Moreover, they are compatible with each other since they arepowers of the same operator.In a ωN manifold one can construct a special class of Lenard-Haantjes chains.In fact, in this context Theorem 38 amounts to say that the co-distribution(64) D ◦ H = Span { d H, N T d H, . . . , ( N T ) n − d H } which is supposed to be of rank n , is integrable if and only if H generates thefollowing Lenard-Haantjes chain(65) d H j = ˜ K α dH = p α ( N T ) dH, with(66) p α ( N ) = n X k =1 a jk ( x ) N k − , α = 0 , , . . . , n − , j = α + 1 . Here a jk ( x ) are suitable coefficients fulfilling the requirements of Definition 35. Asthe modified Haantjes operators ˜ K α are generated by a unique Nijenhuis operator N , the chain (65) will be called a Lenard-Nijenhuis chain. Note that this type ofchains have been called Nijenhuis chains in [17], and generalized Lenard chains in[58, 61]. The particular chains with ˜ K α ≡ K α = N α are the classical Lenard-Magrichains. 5. Complete Integrability and Haantjes structures
The aim of this Section is to prove one of the main results of this paper. Also,we shall show in a specific example how the Haantjes formulation overcomes, forthe vector field under scrutiny, an obstruction to the existence of a classical Lenardchain pointed out by Brouzet.
AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 19
Haantjes theorem for integrable systems.
We propose a characteriza-tion of the notion of integrability in the sense of Liouville–Arnold in terms of ω H structures. Theorem 41.
Let M be a n -dimensional ω H manifold and { H , H , . . . , H n } besmooth functions forming a Lenard-Haantjes chain. Then, the foliation generatedby these functions is Lagrangian. Consequently, each Hamiltonian system, withHamiltonian functions H j , ≤ j ≤ n is integrable by quadratures.Conversely, let { H , . . . , H n } be a completely integrable system in n dimensions,defined by a Hamiltonian H = H and a set of independent integrals of motion H , . . . , H n . Let { ( J k , φ k ) } , k = 1 , . . . , n , denote a set of action-angle variables,with associated frequencies ν k := ∂H∂J k . If H is non degenerate [3] , that is (67) det (cid:18) ∂ν k ∂J i (cid:19) = det (cid:18) ∂ H∂J i ∂J k (cid:19) = 0 , then M admits, in any tubular neighborhood of an Arnold torus, an ω H structuregiven by (68) K α = n X i =1 ν ( α +1) i ν i (cid:18) ∂∂J i ⊗ d J i + ∂∂φ i ⊗ d φ i (cid:19) α = 0 , . . . , n − , where ν ( α +1) i are the frequencies of the ( α + 1) linear flow.Proof. By virtue of the classical Arnold-Liouville theorem, it is sufficient to provethat the functions H j belonging to a Lenard-Haantjes chain are in involution w.r.t.the Poisson bracket defined by the symplectic form ω . In fact,(69) { H j , H k } = < dH j , P dH k > = < ˜ K Tα dH, P ˜ K Tβ dH > = < dH, ˜ K α P ˜ K Tβ dH > P rop.
30= 0Let us prove the converse statement. The integrals of motion { H , . . . , H n } are allassumed to be smooth functions on an open dense subset of the phase space andin involution among each others. Due to the celebrated Arnold theorem [3], the2 n -dimensional phase space is foliated by leaves whose connected components, ifcompact, are invariant tori. Also, there exists at least in any tubular neighborhoodof each torus a set of action-angle (AA) variables { ( J i , φ i ) } , in which the symplectic2-form reads(70) ω = n X i =1 d J i ∧ d φ i . Owing to condition (67), the set { H , . . . , H n } depends on the action variables only.Then the functions H i take the generic form(71) H i = H i ( J , . . . , J n ) , i = 1 , . . . , n. With these data, we shall construct a semi-global ω H structure associated with S .We can take as Haantjes operators the following diagonal operators in the action-angle coordinates(72) K α = n X i =1 l ( α ) i ∂∂J i ⊗ d J i + n X i =1 l ( α ) n + i ∂∂φ i ⊗ d φ i α = 0 , . . . , n − , so that, at the same time, they have their Haantjes tensor vanishing, as requiredin eq. (47), they commute with each others, according to eq. (49), and fulfill thedifferential compatibility condition (50).Once we have chosen the natural vector fields ( ∂∂J i , ∂∂φ i ) as a basis of sharedeigenvectors fields for the Haantjes operators, it remains to determine their eigen-values ( l ( α ) i , l ( α ) n + i ).Notice that the algebraic compatibility conditions (48) of the operators (72) withthe symplectic form (70) are equivalent to the conditions(73) l ( α ) n + i = l ( α ) i i = 1 , . . . , n α = 1 , . . . , n − . Thus, the eigenvalues of our Haantjes operators (72) must be at least double. Fi-nally, we impose that the integrals of motion ( H , H , . . . , H n ) form a Lenard Haan-tjes chain generated by H = H , i.e.(74) K Tα dH = dH α +1 , α = 0 , . . . , n − . Being K α diagonal in the AA variables, such conditions are equivalent to thefollowing system of 2 n algebraic equations in the n indeterminate functions l ( α ) i ,(75) l ( α ) i ∂H∂J i = ∂H α +1 ∂J i , i = 1 , . . . , n. (76) l ( α ) i ∂H∂φ i = ∂H α +1 ∂φ i Obviously, eqs. (76) are trivially satisfied, so that only eqs. (75) should be takeninto account. Assuming, without loss of generality, that ν i = 0, i = 1 , . . . , n , theequations (75) imply that the elements in the diagonal of the k -th operators mustbe the ratios between the corresponding frequencies associated to the k -th integraland to the Hamiltonian, respectively. It is easy to prove that the Haantjes operatorsso obtained are independent, due to the independence of the integrals of motion.Consequently, the Haantjes operators that provide a Lenard-Haantjes chain formedby { H , . . . , H n } read as in eq. (68). (cid:3) Remark 42.
The Haantjes operators (68) exist without any restriction on theform of the Hamiltonian functions (71) , except for the condition (67) . However, ifone wishes to construct a Nijenhuis recursion operator N for H , i.e. a Nijenhuisoperator that, at the same time, provides a classical Lenard chain (77) d H j = ( N T ) ( j − d H , and has the natural vector fields ( ∂∂J i , ∂∂φ i ) as eigenvectors, then the Hamiltonianfunction must be necessarily of the separated form (78) H ( J , J , . . . , J n ) = n X i =1 H i ( J i ) , where H i ( J i ) is a smooth function of the single action variable J i (see [43] , [46] ). Remark 43.
The eigenvalues l ( α ) i of the Haantjes operators K α , α = 1 , , . . . , n − , depending only on action variables, are integrals of motion for the Hamiltonianvector field X H , i.e. their Lie derivatives along the flow of X H vanishes: L X H l ( α ) i = 0 . AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 21
However, this property does not imply that the Haantjes operators are recursionoperators for X H as (79) L X H K α = n X i,k =1 (cid:0) l ( α ) i − l ( α ) k (cid:1) ∂ν i ∂J k ∂∂φ i ⊗ d J k . The analysis of Brouzet.
In [12], R. Brouzet studied the existence for acompletely integrable system of a Nijenhuis recursion operator, that is a Nijenhuisoperator compatible with ω and fulfilling eq. (45). He proved that the existenceof a Nijenhuis recursion operator for X H requires very strong conditions on theform of its Hamiltonian function. Accordingly, he presented an example of anintegrable system with two degrees of freedom that does not admit a recursionoperator compatible with the original symplectic structure. Here we show thatsuch example does admit a simple formulation in the context of the ω H geometry.In his analysis, Brouzet considered the symplectic manifold M = R × T , withthe action variables ( J , J ) ∈ R , the angles ( φ , φ ) on the bi-dimensional torus T , and the Hamiltonian function(80) H = J (1 + J ) , which is not of the form (78) and is non degenerate according to (67), in the denseopen submanifold M ′ := { m ∈ M : J = 0 } . The corresponding Hamiltonianvector field(81) X H = (1 + J ) ∂∂φ + 2 J J ∂∂φ is completely integrable, since any smooth function depending only on the actionsis an integral of motion for it. For instance, let us take(82) H = J , which is functionally independent of H in M ′ . One can easily verify that the twoHamiltonian functions in involution ( H = H, H ) form a Lenard-Haantjes chainw.r.t. the ω H structure given by the standard symplectic form(83) ω = d J ∧ d φ + d J ∧ d φ and by the Haantjes operators(84) K = I , K = 1 J (cid:18) ∂∂J ⊗ d J + ∂∂φ ⊗ d φ (cid:19) , constructed in the open submanifold of M ′ where J = 0, according to the pre-scriptions (68).It is interesting to observe that the authors of [32] have by-passed the Brouzetobstruction to the definition of a Njenhuis recursion operator for the Hamiltonian(80) (and for other examples presented in [13]) by using a different strategy. Thealternative approach consists in allowing a Njenhuis recursion operator compatiblewith a symplectic structure different from the original one. By contrast, in ourtheory, the Haantjes operators are compatible with the very original symplecticstructure. New integrable models from Haantjes geometry
Once we have stated the conceptual equivalence between complete integrabilityof a Hamiltonian system and the existence of an associated Haantjes structure,we can use this equivalence in both ways: to construct integrable models froma given Haantjes geometry (the direct problem ) or conversely to determine theHaantjes geometry of a given integrable system (the inverse problem ). In thissection, we will adopt the first point of view, in order to show the flexibility of theHaantjes approach in applicative contexts. Indeed, by imposing the existence of aLenard-Haantjes chain generated by a specific Haantjes operator, we define classesof associated integrable models.6.1.
Harmonic functions and integrable systems.
Even the case of uniformHaantjes operators (i.e. independent of x ∈ M ) provides an interesting class ofintegrable models directly related to the theory of analytic functions. Theorem 44.
Let M be a symplectic manifold of dimension , ( x, y, p x , p y ) aset of Darboux coordinates in M , ϕ ( p x , p y ) = ϕ ( p x , p y ) + iϕ ( p x , p y ) and ψ = ψ ( x, y ) + iψ ( x, y ) be two analytic functions ( i = − ). Then, the Hamiltonian (85) H ( x, y, p x , p y ) = ϕ ( p x , p y ) + ψ ( x, y ) admits the first integral of motion (86) H ( x, y, p x , p y ) = ϕ ( p x , p y ) + ψ ( x, y ) . Proof.
Consider the uniform Haantjes operator(87) K = − ∂∂x ⊗ d y + ∂∂y ⊗ d x − ∂∂p x ⊗ d p y + ∂∂p y ⊗ d p x . We construct the Lenard-Haantjes chain(88) K T dH = dH . It amounts to solve the systems of equations ( ∂H ∂p y = ∂H ∂p x , ∂H ∂p x = − ∂H ∂p y , (89)and ( ∂H ∂y = ∂H ∂x , ∂H ∂x = − ∂H ∂y , (90)for the functions ϕ , ϕ and ψ , ψ respectively. These relations are nothing butthe classical Cauchy-Riemann equations. Hence, these functions and consequently H and H are all harmonic functions. (cid:3) Example 45.
A simple class of integrable models arises when both analytic func-tions ϕ ( p x , p y ) and ψ ( x, y ) are chosen to have a polynomial structure. For instance,choosing a third degree homogeneous polynomial in the momenta ( p x , p y ) , we havethe system (91) H = p x − p x p y + x − y , H = 3 p x p y − p y + 2 xy . AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 23
Waves and integrable systems.
Solutions of the wave equations also defineintegrable systems, via the Haantjes geometry.
Theorem 46.
Let ξ = x + y √ , η = x − y √ , p ξ = p x + p y √ , p η = p x − p y √ characteristiccoordinates and momenta in an open set of M . The Hamiltonian (92) H ( ξ, η, p ξ , p η ) = f ( η ) + g ( ξ ) + F ( p η ) + G ( p ξ ) where f , g , F , G are arbitrary functions of their arguments, is integrable and admitsthe first integral of motion (93) H ( ξ, η, p ξ , p η ) = − f ( η ) + g ( ξ ) − F ( p η ) + G ( p ξ ) . Proof.
Consider the uniform Haantjes operator in cartesian coordinates and mo-menta(94) K = ∂∂x ⊗ d y + ∂∂y ⊗ d x + ∂∂p x ⊗ d p y + ∂∂p y ⊗ d p x . We construct the Lenard-Haantjes chain(95) K T dH = dH . This chain is defined by the differential relations ( ∂H ∂p y = ∂H ∂p x , ∂H ∂p x = ∂H ∂p y , ( ∂H ∂y = ∂H ∂x , ∂H ∂x = ∂H ∂y , (96)These equations can be combined to provide the wave equations H i,p x p x − H i,p y p y = 0 , H i,xx − H i,yy = 0 , i = 1 , . Therefore the Hamiltonian functions H ( x, y, p x , p y ) = F ( p x − p y ) + G ( p x + p y ) + f ( x − y ) + g ( x + y )(97)and H ( x, y, p x , p y ) = − F ( p x − p y ) + G ( p x + p y ) − f ( x − y ) + g ( x + y ) , (98)where F, G, f, g are arbitrary smooth functions of their arguments, define a com-pletely integrable system, separable in the coordinates ( ξ, η, p ξ , p η ). (cid:3) Example 47.
Choosing the functions
F, G, f, g as a power of their argument, weget the interesting class of models H = ( p x − p y ) n + ( p x + p y ) n + ( x − y ) m + ( x + y ) m , (99) H = − ( p x − p y ) n + ( p x + p y ) n − ( x − y ) m + ( x + y ) m . (100) For n = 2 , the Hamiltonian function H is quadratic in the momenta and cor-responds to a class of separable systems that can be found in [54] (page 81). Inparticular, for n = 2 , m = 3 one gets the Sawada-Kotera system [1] . For n > onegets, to the best of our knowledge, a new family of integrable systems. The inverse method outlined in this section can be widely adopted to generatenew models from known Haantjes operators. However, an exhaustive analysis ofthis approach is out of the scopes of the present paper. The theory of separation of variables in ω H manifolds Darboux-Haantjes coordinates.
The simple form admitted by the Haan-tjes operators (68) in AA variables suggests the search for a set of distinguishedlocal coordinates in M that, at the same time be symplectic and diagonalize everyHaantjes operator of the underlying geometric structure. Definition 48.
Let ( M, ω, K , K , . . . , K n − ) be a ω H manifold. A set of localcoordinates ( q , . . . , q n ; p , . . . , p n ) will be said to be a set of Darboux-Haantjes (DH)coordinates if in this set the symplectic form ω assumes the Darboux form(101) ω = n X i =1 d p i ∧ d q i and each Haantjes operator diagonalizes: (102) K α = n X i =1 l ( α ) i (cid:16) ∂∂q i ⊗ d q i + ∂∂p i ⊗ d p i (cid:17) α = 0 , . . . , n − , with l (0) i = 1 , i = 1 , . . . , n . Remark 49.
The set of semisimple operators is closed under C ∞ ( M ) –linear com-binations and under the product of operators. Thus, if a set of DH coordinatesexists in a given ω H manifold, the compatibility conditions (49) and (50) are au-tomatically satisfied, thanks to Proposition 7. There is a natural relation between AA variables and DH coordinates in theHaantjes geometry, as clarified below.
Proposition 50.
Any set of AA variables for a completely integrable system is aset of DH coordinates for the Haantjes structure ( M, ω, K , K , . . . , K n − ) givenby the symplectic form ω and the Haantjes operators (68) . In the subsequent considerations, under mild hypotheses we solve the problemof the existence of DH coordinates for a given ω H manifold, possibly different fromthe ones of Theorem 41.First, we present some properties of algebraic and differential nature that holdunder the assumption that the ω H structure is semisimple. Definition 51.
Let ( M, ω, K , K , . . . , K n − ) be a ω H manifold. It will be saidto be semisimple if each operator K α is semisimple, that is each K α admits areference frame of proper eigenvectors (according to Definition 16). Proposition 52.
In a semisimple ω H manifold, the relations Ω ( D αj ) ≡ ( E αj ) ◦ ⇔ D αj ≡ P (( E αj ) ◦ ) = ( E αj ) ⊥ , (103) Ω ( E αj ) ≡ ( D αj ) ◦ ⇔ E αj ≡ P (( D αj ) ◦ ) = ( D αj ) ⊥ , (104) hold true.Proof. Property (103) follows from the compatibility condition (48) and from theinvertibility of the symplectic operator Ω . In fact, for each eigenvector Y j ∈ D αj ,the one-form Ω Y j is an eigenform of K Tα , as one infers from K Tα Ω Y j (48) = Ω K Y j = l ( α ) j Ω Y j . AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 25
Then, by taking into account eq. (44), we deduce that Ω Y j belongs to ( E αj ) ◦ . Sinceit has the same dimension of D αj , we get eq. (103). The relation (104) follows fromeq. (103) and from the observation that, by construction, E j is a complementarysubspace of D αj in T M . (cid:3) Proposition 53.
In a semisimple ω H manifold, the distributions D αj are integrableand even dimensional. In addition, if the ω H structure is semisimple, their integralleaves are symplectic submanifolds of M and are symplectically orthogonal to eachother, namely ω ( D αj , D αj ) = symplectic (105) ω ( D αj , D αk ) = 0 j = k (106) Proof.
The distributions D αj are integrable due to Theorem 17 and are even-dimensionalby virtue of Corollary 31. Moreover, they are symplectic as D αj ∩ ( D αj ) ⊥ (104) = D αj ∩ E αj = { } . Finally, property (106) follows from the fact that D αk ⊆ E αj (104) ≡ ( D αj ) ⊥ , if j = k . (cid:3) Generators of a ω H manifold. We shall investigate here the possibility thatthe Haantjes operators of a ω H manifold can be generated by a single Haantjes (orNijenhuis) operator. Definition 54.
Let ( M, ω, K , K , . . . , K n − ) be a n -dimensional ω H manifold.An operator K will be called a generator of the ω H structure if the set (107) B c := { I , K , K , K n − } is a basis of the Haantjes module K . Such a basis will be called a cyclic basis of K and allows one to represent each Haantjes operator K α as a polynomial field in K of degree at most (n-1), i.e. (108) K α = p α ( x , K ) = n − X i =0 a ( α ) i ( x ) K i , where a ( α ) i ( x ) are smooth functions in M such that det( a ( α ) i ) = 0 . Proposition 55.
The set of the generators of an ω H structure coincides with theset of the operators that belong to the module K generated by the original Haantjesoperators { K , K , . . . , K n − } , and have their minimal polynomial of degree n .In addition, if the ω H structure is semisimple, any generator of K is maximal,according to Definition 32.Proof. If K is a generator, together with any of its powers, it belongs to K thanksto Proposition 11. Therefore, all of its powers are Haantjes operators compatiblewith the ω H structure. Moreover, being B c defined by (107) a basis of K , the first( n −
1) powers of K must be linearly independent; thus its minimal polynomial hasdegree not lesser then n . Precisely, it has just degree n , as the powers of K greaterthan ( n − K as well, can be written as linear combinations of thecyclic base.Conversely, if K is an operator belonging to the module K , we deduce that K ,together with its powers, is a Haantjes operator compatible with the ω H structure. Since its minimal polynomial is of degree n by hypothesis, its first ( n −
1) powers { I , K , K , K n − } are linearly independent; they form a basis for the module K and consequently generate K .Finally, if the ω H structure is semisimple, the Haantjes operators K α admit acommon reference frame (thanks to their commutativity (49)) where they take adiagonal form, and each element of the module K is semisimple as well in the samereference frame. Therefore, any generator is semisimple and, having the minimalpolynomial of degree n , is maximal by virtue of Lemma 33. (cid:3) Let us denote with D j the eigen-distributions of a generator K , and recall that D αj denote the eigen-distributions of the Haantjes operators K α . Proposition 56.
Let ( M, ω, K , K , . . . , K n − ) be a n -dimensional, semisimple ω H manifold. If the Haantjes module K admits a generator K , then the followingalternative conditions hold (109) D j ∩ D αk = { } or D j ⊆ D αk , α = 0 , . . . , n − , k = 1 , . . . , n . Proof.
Let us consider the eigen-distributions D j := Ker (cid:16) K − l j ( x ) I (cid:17) , for j =1 , . . . , n , which are two-dimensional by the maximality property of any generator.As K and K α commute by the hypothesis (49), D j is K α -invariant, therefore K α can be restricted to D j and such a restriction is semisimple. Consequently, in D j there exist two independent eigenvector fields Y , Y of K α . Let us show that theycorrespond to the same eigenvalue of K α , say l ( α ) k . Indeed, if by absurd should theycorrespond to different eigenvalues of K α , the property (106) would imply that ω ( Y , Y ) = 0 . Nevertheless, being D j two-dimensional, this should imply that D j is isotropic, incontradiction with the property to be symplectic (105). Hence the thesis. (cid:3) Existence of DH coordinates. We shall prove that, in a given semisimple ω H manifold, the existence of a Haantjes (or Nijenhuis) generator is equivalent tothe existence of a set of DH coordinates.The main result of this section is the following theorem, establishing the existenceof DH coordinates. Theorem 57.
Let ( M, ω, K , K , . . . , K n − ) be a n -dimensional, semisimple ω H manifold. If a set of DH coordinates and { ( q , p ) } does exist, then each operator ofthe form (110) K = n X i =1 λ i ( x ) (cid:16) ∂∂q i ⊗ d q i + ∂∂p i ⊗ d p i (cid:17) , ( λ i ( x ) being arbitrary smooth functions on M fulfilling the maximality condition λ i ( x ) = λ j ( x ) , ∀ x ∈ M ) is a generator of the given Haantjes structure through therelations (111) K α = n X i =1 l ( α ) i Π j = i ( K − λ j I )Π j = i ( λ i − λ j ) α = 0 , . . . , n − . Conversely, if the module K generated by { K , K , . . . , K n − } contains a max-imal operator, then there exists locally a set of Darboux-Haantjes coordinates. AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 27
Proof.
In each set of DH coordinates, the Haantjes operators { K , . . . , K n − } as-sume a diagonal form. Let us take as possible generators the family of diago-nal operators (110). They are Haantjes operators compatible with ω and with { K , K , . . . , K n − } thanks to the Remark 49. In order to assure that K α is gen-erated by K , according to eq. (108), it suffices to show that there exist ( n − p α ( x , λ ), λ ∈ R , such that l ( α ) i ( x ) = p α ( x , λ i ) = n − X k =0 c ( α ) k ( x ) λ ki ( x ) , α = 1 , · · · , n − . These algebraic equations are solved by means of the Lagrange interpolation poly-nomials of degree ( n − π i ( λ ) = Π j = i ( λ − λ j )Π j = i ( λ i − λ j ) , which yield the expressions p α ( x , λ ) = n X i =1 l ( α ) i ( x ) π i ( λ ) . Therefore,(112) K α = n X i =1 p α ( x , λ i ) (cid:16) ∂∂q i ⊗ d q i + ∂∂p i ⊗ d p i (cid:17) = p α ( x , K ) α = 0 , . . . , n − . Conversely, if the module K contains a maximal operator K , such an operatoris a generator of the ω H structure thanks to Proposition 55. Moreover, being K a maximal Haantjes operator, we show that K admits a set of symplectic coordi-nates, in which it takes the diagonal form (110). Indeed, we need to prove that aparametrization of the characteristic web of K exists with 2 n coordinate functions( q , . . . , q n ; p , . . . , p n ) that are Darboux coordinates for ω . Under the maximalityassumption, the characteristic web is of type ( n, n, E j , ( E j ) ◦ are integrable;we shall see that their foliations ( E , E , . . . , E s ) form the characteristic web of the ω H manifold. Equivalently, any parametrization of the web always diagonalizessimultaneously each operator K α and K Tα .Each characteristic distribution E k is integrable and is of co-dimension 2. Onecan find in each annihilator ( E k ) ◦ the differentials of two functions ( f k , g k ) thatspan ( E k ) ◦ . Collecting together these functions, one obtains a parametrization( f , . . . , f n ; g , . . . , g n ) of the characteristic web of K . By virtue of Corollary 24,such parametrization enables us to put the operator K in diagonal form, whilst thesymplectic form reads(113) ω = n X i =1 b i ( f i , g i ) d g i ∧ d f i . In fact, as ∂∂f j , ∂∂g j ∈ D j due to relation (42), eq. (106) implies that0 = ω (cid:16) ∂∂f j , ∂∂f k (cid:17) = ω (cid:16) ∂∂g j , ∂∂g k (cid:17) = ω (cid:16) ∂∂f j , ∂∂g k (cid:17) , j = k . Furthermore, the closure of ω implies that the component functions b i dependonly on the pair ( f i , g i ). Then, by means of eq. (105), inside the subspace ( E j ) ◦ , where the restriction of K T is a multiple of the identity thanks to the condition(109), one can perform a Darboux transformation q i = ˜ f i ( f i , g i ) , p i = ˜ g i ( f i , g i ),involving only the pairs ( f j , g j ). After these transformations, one obtains a localchart { ( q , . . . , q n ; p , . . . , p n ) } in which the symplectic form ω takes the Darbouxform (101). Therefore, we have proven the existence of a coordinate system ( q ; p )which still diagonalizes the generator K and simultaneously reduces ω to a canonicalform.Finally, even the original Haantjes operators K α , being generated by K accord-ing to eq. (108), in such coordinates take the diagonal form (112). We concludethat these coordinates are DH coordinates in the considered ω H manifold, for eachHaantjes operator belonging to the module K . (cid:3) As an immediate consequence of the previous Theorem, of Proposition 26 andof Proposition 9, we have the following
Corollary 58.
In any set of DH coordinates, a function f i is a characteristicfunction of the Haantjes web, related to the eigenvalues { l ( α ) i } ≤ α ≤ ( n − , if andonly if it depends on the single pair ( q i , p i ) only, that is (114) f i = f i ( q i , p i ) . Therefore, ( f i , f j ) satisfy the involution relations (115) { f i , f j } = 0 i = j . Moreover, a family of generators of the Haantjes module does exist, formed byNijenhuis operators. They are the operators that in DH coordinates take the form (110) , with eigenvalues (116) λ i ( x ) = λ i ( q i , p i ) i = 1 , . . . , n , where λ i is an arbitrary smooth function depending only on the single conjugate pair ( q i , p i ) . Therefore, such eigenvalues are characteristic functions of the Haantjes weband are in mutually involution. The Haantjes generators (110) are parametrized by the arbitrary functions λ i ( x );therefore, they form an infinite family of operators. By contrast, given a set of n functions in involution ( H , . . . , H n ), the Haantjes operators providing the as-sociated Lenard-Haantjes chain (53) are uniquely determined, once a holonomicframe of common eigenvectors has been chosen. Thus, the Haantjes operators { ˜ K , ˜ K , . . . , ˜ K n − } appear to play a more fundamental role than their genera-tors K .7.4. Haantjes theorem for separable systems.
Without loss of generality, weassume, that we have already exchanged the original operator K α with ˜ K α and,for the sake of simplicity, we drop off the tilde over ˜ K α from now on.The next theorem is the main result concerning the existence of separation vari-ables for ω H manifolds. It states that such structures characterize each separableHamiltonian system. Theorem 59.
Let ( M, ω, K , K , . . . , K n − ) be a semisimple ω H manifold and ( H , H , . . . , H n ) be a set of n functions belonging to a Lenard–Haantjes chaingenerated by H = H . If the Haantjes module K contains a maximal operator, theneach set ( q , p ) of DH coordinates provides separation variables for the Hamilton–Jacobi equations associated with each function H j . AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 29
Conversely, if ( H , H , . . . , H n ) are a set of n vertically independent functions,that is (117) det (cid:18) ∂H i ∂p j (cid:19) = 0 , and are separable in a set of Darboux coordinates ( q , p ) , then they belong to aLenard–Haantjes chain w.r.t. the ω H structure ( M, ω, K , K , . . . , K n − ) givenby (118) K α = n X i =1 ∂H α +1 ∂p i ∂H∂p i (cid:18) ∂∂q i ⊗ d q i + ∂∂p i ⊗ d p i (cid:19) α = 0 , . . . , n − . Proof.
Theorem 57 guarantees the existence of sets of DH coordinates. Thereforeit suffices to show that the functions H j are in separable involution in these coordi-nates, according to eq. (1). To this aim, let us note that, due to the diagonal form(102) of K Tα in a DH local chart, the relations ∂H j ∂q k = l ( j − k ∂H∂q k , (119) ∂H j ∂p k = l ( j − k ∂H∂p k , (120)hold. Here l ( j − k denotes the eigenvalues of the Haantjes tensors K Tα , α = j − { H i , H j } | k = l ( i − k ∂H∂q k l ( j − k ∂H∂p k − l ( j − k ∂H∂q k l ( i − k ∂H∂p k = 0 . We prove the converse statement in a way analogous to that of Theorem 41.Without loss of generality, we can assume that ∂H∂p i = 0 , i = 1 , . . . , n . The opera-tors (118), being diagonal in the separated coordinates, have their Haantjes tensorvanishing. Also, they commute with each others and fulfill the differential compati-bility condition (50). The algebraic compatibility relations (48) with the symplecticform are equivalent to the conditions(121) l ( α ) n + i = l ( α ) i i = 1 , . . . , n . Thus, the Haantjes operators (118) must possess eigenvalues that are at least dou-ble.Finally, we impose that the integrals of motion ( H , H , . . . , H n ) form a LenardHaantjes chain generated by H ≡ H . Being K α diagonal in the ( q , p ) variables,such conditions are equivalent, for each α , to the overdetermined system of 2 n algebraic equations in the n indeterminate functions l ( α ) i l ( α ) i ∂H∂q i = ∂H α +1 ∂q i , (122) l ( α ) i ∂H∂p i = ∂H α +1 ∂p i , (123) i = 1 , . . . , n . However, the above equations are compatible, because the Benenticonditions (1) of separate involution assure that ∂H∂q i ∂H α +1 ∂p i = ∂H∂p i ∂H α +1 ∂q i , ≤ i ≤ n . Consequently, the equations (123) provide the unique solution (118), where theHaanties operators K α are independent thanks to the condition (117). (cid:3) Constructing explicitly a set of DH coordinates is a difficult task that entailsto integrate the eigen-distributions of a Haantjes generator K for the Haantjesmodule generated by { K , K , . . . , K n − } . We will use this method in Subsect.9.2. However, it can be simplified when a Nijenhuis generator N is at our disposal.In fact, the non constant eigenvalues of such N are just characteristic functionsof the Haantjes web by virtue of Corol. 58. Therefore, if all the eigenvalues arenot constant and functionally independent, being N maximal, one can start withhalf of the DH coordinates ( λ ( x ) , . . . , λ n ( x )). Moreover, it has been proved in[42] that they can be complemented by quadratures with n conjugated momenta( µ ( x ) , . . . , µ n ( x )) satisfying(124) N T d µ i = λ i d µ i . Therefore, in such canonical coordinates ( λ , µ ), the Nijenhuis generator N takes adiagonal form, consequently they are DH coordinates for the given ωH structure.They have been called Darboux-Nijenhuis (DN) coordinates in [17] and specialDN coordinates in [19]. Let us note that, in such a situation, the tensor field P := N Ω − turns out to be a Poisson bivector compatible with Ω, endowing the ωH manifold M also with a bi-Hamiltonian structure.7.5. Classical separable Hamiltonian systems.
In this Section, given a genericHamiltonian system whose Hamilton-Jacobi equation can be solved by separationof variables, we construct an associated Haantjes structure.Hamiltonian systems solvable by separation of variables are obviously integrableby quadrature. Theorem 41 holds for non degenerate Hamiltonian systems which,consequently, admit the Haantjes structure (68). However, we will prove that aHaantjes structure can also be constructed starting with separation variables dif-ferent from the AA variables. Moreover, such a structure does not necessarilycoincide with the one coming from (68).As an application of the defining relations (118), we get explicitly the Haantjesstructure for three large classes of separable Hamiltonian systems (see [3], Propo-sition 5, page 125).
Remark 60.
For the families of n -dimensional Hamiltonian systems treated belowevery Haantjes operator for n > is not maximal. However, the associated Haantjesmodule K contains, by construction, at least a maximal generator of the form (110) . The proof of the statements presented hereafter is direct and is left to the reader.
Proposition 61 ( “Functionally separated” systems ) . Let us consider the Hamil-tonian function (125) H (cid:0) f ( q , p ) , f ( q , p ) , . . . , f n ( q n , p n ) (cid:1) , together with the integrals of motion (126) I j = f j ( q j , p j ) , j = 1 , . . . , n − . This system admits the Haantjes structure given by the Haantjes operators (127) K = I , K j := (cid:18) ∂H∂f j (cid:19) − (cid:18) ∂∂q j ⊗ d q j + ∂∂p j ⊗ d p j (cid:19) , j = 1 , . . . , n − . AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 31
The spectrum of K j is (128) Spec ( K j ) = ( , (cid:18) ∂H∂f j (cid:19) − ) . Such Haantjes operators provide the Lenard–Haantjes chain K Tj d H = d I j j = 1 , . . . , n − , I = H .
Example 62.
The expression of the natural Hamiltonian function additively sep-arated in cartesian coordinates reads H = n X j =1 (cid:18) p j m j + V ( x j ) (cid:19) . It admits the integrals of motion I j = p j m j + V ( x j ) j = 1 , . . . , n − . The Haantjies operators are K = I , and the uniform ones K j := (cid:18) ∂∂q j ⊗ d q j + ∂∂p j ⊗ d p j (cid:19) j = 1 , . . . , n − . Proposition 63 ( Telescopic systems ) . The system defined by the Hamiltonianfunction (129) H (cid:18) f n (cid:16) f n − (cid:0) . . . f ( f ( q , p ) , q , p ) , . . . , q n − , p n − (cid:1) , q n , p n (cid:17)(cid:19) together with the integrals of motion (130) I = f ( q , p ) , I = f (cid:0) f ( q , p ) , q , p (cid:1) , I n := H , admits the Haantjes structure given by the Haantjes operators K = I and (131) K j := (cid:18) ∂f n ∂f n − ∂f n − ∂f n − . . . ∂f j +1 ∂f j (cid:19) − X i ≤ j (cid:18) ∂∂q i ⊗ d q i + ∂∂p i ⊗ d p i (cid:19) j = 1 , . . . , n − . The spectrum of these operators is given by (132)
Spec ( K j ) = ( , (cid:18) ∂f n ∂f n − ∂f n − ∂f n − . . . ∂f j +1 ∂f j (cid:19) − ) . The operators K j form the Lenard–Haantjes chain K Tj d H = d I j j = 1 , . . . , n − . Example 64.
The most general Hamiltonian function separable in spherical coor-dinates { ( q = φ, q = θ, q = r ) } is given by (133) H = 12 m (cid:18) p r + p θ r + p φ r sin θ (cid:19) + g ( r ) + h ( θ ) r + s ( φ ) r sin θ , where g ( r ) , h ( θ ) , s ( φ ) are arbitrary smooth functions of their argument. The inte-grals of motion are I = p φ m + s ( φ ) , I = 12 m (cid:18) p θ + p φ + s ( φ )sin θ (cid:19) + h ( θ ) , I = H .
According to eqs. (131) , the Haantjies operators are K = I , and K = r sin θ (cid:18) ∂∂φ ⊗ d φ + ∂∂p φ ⊗ d p φ (cid:19) (134) K = r (cid:18) ∂∂φ ⊗ d φ + ∂∂p φ ⊗ d p φ + ∂∂θ ⊗ d θ + ∂∂p θ ⊗ d p θ (cid:19) . (135) Remark 65. If g = − k/r, h ( θ ) = s ( φ ) = 0 , one gets the classical Coulomb-Keplersystem. For such model, the authors of [44] proved that it does not exist a Nijenhuisrecursion operator compatible with the original symplectic structure. Indeed, theNijenhuis structure they found provides a Lenard chain of dependent integrals ofmotion.Accordingly, we stress the fact that a Haantjes structure is needed to constructthe Lenard–Haantjes chain formed by H = H , H = I , H = I . Proposition 66 ( Gantmacher systems ) . Let us consider the Hamiltonian systemdefined by the Hamiltonian function (136) H = P nk =1 f k ( q k , p k ) P nk =1 g k ( q k , p k ) , with the integrals of motion (137) I k = f k ( q k , p k ) − Hg k ( q k , p k ) , k = 1 , . . . , n − . The system admits the Haantjes structure given by the Haantjes operators K = I ,and (138) K j := − g j I + G (cid:18) ∂∂q j ⊗ d q j + ∂∂p j ⊗ d p j (cid:19) j = 1 , . . . , n − , with G := P nk =1 g k ( q k , p k ) , whose spectrum is given by (139) Spec ( K j ) = {− g j , G − g j } . They provide the Lenard–Haantjes chain K Tj d H = d I j j = 1 , . . . , n − . Example 67 ( Liouville systems ) . Choosing in eq. (136) f k := 12 a k ( q k ) p k + V k ( q k ) , g k := g k ( q k ) , one gets the class of systems that Liouville showed to be separable [54] . For instance,they include each natural Hamiltonian system in the plane, separable in cartesian ( x, y ; p x , p y ) , polar ( r, θ ; p r , p θ ) , elliptic or parabolic coordinates ( ξ, η ; p ξ , p η ) . Theassociated Haantjes operators read, respectively AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 33 K car = − (cid:18) ∂∂y ⊗ d y + ∂∂p y ⊗ d p y (cid:19) , (140) K pol = − r (cid:18) ∂∂θ ⊗ d θ + ∂∂p θ ⊗ d p θ (cid:19) , (141) K par = η (cid:18) ∂∂ξ ⊗ d ξ + ∂∂p ξ ⊗ d p ξ (cid:19) − ξ (cid:18) ∂∂η ⊗ d η + ∂∂p η ⊗ d p η (cid:19) , (142) K ell = − η (cid:18) ∂∂ξ ⊗ d ξ + ∂∂p ξ ⊗ d p ξ (cid:19) − ξ (cid:18) ∂∂η ⊗ d η + ∂∂p η ⊗ d p η (cid:19) . (143) Remark 68.
The Haantjes operators (141) , (142) , (143) , obtained through theconstruction (118) do not coincide with the Haantjes operators of the same systemthat one could construct through AA variables according to eq. (68) . In fact, as aconsequence of Remark 43, the eigenvalues of (118) must be integrals of motion,whilst the eigenvalues of the operators (141) , (142) , (143) are not invariant alongthe flow of the Hamiltonian vector fields. This fact implies that Haantjes structuresfor integrable systems are usually not unique. The inverse problem for systems with two degrees of freedom
In this Section, we are concerned with the inverse problem. In other words, givena set of independent functions in involution, we will construct a Haantjes structurefor them, that is we shall determine the Haantjes structures of an assigned integrablesystem.8.1.
A general procedure.
Let us consider the simplest case of Hamiltonian sys-tems with two degrees of freedom. We propose a general procedure to compute aHaantjes operator adapted to the Lenard chain formed by two integrals of motion.We search for a generator of the Haantjes module, that is, a Haantjes operator K whose minimal polynomial should be of degree two, namely, the maximum degreeallowed by our assumptions:(144) m K ( λ ) := λ − c ( x ) λ − c ( x ) . Let us note that such a request does not imply the semisimplicity property for K ,unless the existence of two distinct roots of m K ( z ) is also supposed. Remark 69.
In the case n = 2 , every pair of Haantjes operators { I , K } generatesa Haantjes module (see Proposition 10) and a ring w.r.t. the operator product(Proposition 11). Any other generator ˜ K has the form (145) ˜ K = f I + g K , where f is an arbitrary smooth function and g a nowhere vanishing smooth function.Thus, (146) det ( ˜ K − λ I ) = det ( f I + g K − λ I ) = g n det (cid:18) K − λ − fg I (cid:19) . Therefore, the eigenvalues λ i of ˜ K and l i of K are related by the affine equation (147) λ i = f + g l i i = 1 , consequently, λ ≡ λ is equivalent to l ≡ l . Then, we can conclude that theHaantjes module contains a maximal generator if and only if all generators aremaximal. The procedure entails two steps.(1) Given two independent integrals of motion in involution ( H , H ), find aHaantjes operator that provide a Lenard-Haantjes chain for them, thereforefulfilling the conditions (61). Consequently, find an operator K such that K T Ω = Ω K (148) K T d H = d H (149) ( K T ) d H = (cid:0) c K T + c I (cid:1) d H (150) H K ( X, Y ) = 0 ∀ X, Y ∈ T M (151) The algebraic compatibility condition (148) allows to reduce the unknowncomponents of the K operator from 16 to 6. Therefore(152) K = l l l l l − l l l l − l l l , where l ij are arbitrary functions on M . The conditions (149) and (150) pro-vide a system of 8 algebraic equations in the 6 unknown functions, being c = l + l and c = − l l + l l − l l . As such equations are not indepen-dent, we are left with 3 unknown functions. The vanishing of the Haantjestensor of K (151) provides an over-determined system of 24 PDEs of firstorder, which can be managed with some suitable ansatz. For instance, somehomogeneity properties for the components of K can be assumed.The next step allows to construct DH coordinates.(2) If the Haantjes generator K found is maximal, it does exist also a Nijenhuisgenerator N . Then, find two smooth functions ( f, g ), with g nowherevanishing, such that N = f I + g K (153) T N ( X, Y ) = 0 ∀ X, Y ∈ T M (154) The vanishing of the Nijenhuis torsion furnishes an over-determined systemof 24 equations with the two unknown functions ( f, g ).If the above procedure is successful, the eigenvalues (147) of N are half of the DHcoordinates that we are looking for, by virtue of Rem. 27. A set of conjugatemomenta can be found as characteristic functions of the Haantjes web as we shalldo in Subsect. 9.2.2. Remark 70.
For a large class of systems, i.e. the so called quasi-bi-Hamiltoniansystems [14, 48, 49, 62] , step 2 can be simplified. In fact, thanks to the results of [63] , we can prove that if eqs. (153) , (154) admit the solution (155) f = 12 trace ( K ) , g = − , AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 35 then the Hamiltonian system under scrutiny has a quasi-bi-Hamiltonian formula-tion. Therefore, the eigenvalues of the Haantjes generator K , found in step 1, arethemselves characteristic functions of the web since, by plugging the solution (155) into eq. (147) , we obtain that (156) l = λ , l = λ . This property will be useful in Section 8.
Let us note that eq. (19b), for n = 2, coincides with the projection of eq. (153),with the solution (155), from the tangent bundle T ∗ E onto E .8.2. On the super-integrable Post-Winternitz system.
In this section, bymeans of the procedure described above we face the inverse problem for a systemwhich recently has attracted much attention: the Post-Winternitz (PW) system[55]. Indeed, it is a maximally superintegrable system [47] with integrals of motioncubic and quartic in the momenta. As a consequence, its bounded orbits are closedand periodic. Thus, as well as every superintegrable system, it does not fulfill thenon degeneracy condition (67) and Theorem 41 cannot be applied. Despite its regu-larity properties, the separability structures of the PW system are not known. Sinceit does not belong to the St¨ackel class, the PW system is certainly not separableby an extended point transformation.Let us consider the Hamiltonian system(157) H = 12 ( p x + p y ) + a xy a ∈ R , with the two independent integrals of motion(158) H = 2 p x + 3 p y p x + a (cid:18) y p y + 6 xy p x (cid:19) , (159) H = p y − ay p x p y + 4 a xy p y − a (cid:18) y / − x y (cid:19) . We shall prove that they form two different Lenard-Haantjes chains (
H, H ) and( H, H ), each of them being sufficient to assure the complete integrability of thePW system.By performing the extended-point canonical transformation(160) q = y , q = xy , p = 2 xy p x + 3 y p y , p = y p x , we reduce the Hamiltonian functions to a rational form from which we infer theweights of the three components of (152), still unknown after having imposed theconditions (149), (150). As a result of the previous approach, we get the Haantjesstructure ( ω, I , K (2) P W ) for the Lenard-Haantjes chain (
H, H ), where(161) K (2) P W = 3 p x p y y p x − y
00 0 2 p x
00 0 p y p x . Similarly, we obtain the Haantjes structure ( ω, I , K (3) P W ) for the Lenard-Haantjeschain (
H, H ), where(162) K (3) P W = 4 p y + 2 a xy / − ( p x p y + 3 ay / ) 0 − yp x p y + 2 a xy / yp x
00 0 p y + 2 a xy /
00 0 − ( p x p y + 3 ay / ) p y + 2 a xy / . Although both K (2) P W and K (3) P W have their minimal polynomial of degree 2, theyare not semisimple, since each of them has only one eigenvalue of algebraic multi-plicity equal to 4, with two proper eigenvectors and two generalized eigenvectors.Therefore, they do not fulfill the assumptions of Theor. 57. Moreover, neither ofthe two Haantjes modules generated by { I , K (1) P W } and { I , K (2) P W } possesses a max-imal generator by Remark 69. However, their two Lenard–Haantjes chain ensurethe superintegrability of the PW model. Thus, the existence of K (2) P W and K (3) P W shows that the Haantjes theory can be naturally applied to non-St¨ackel systemsnot possessing any evident separability structure, even when they do not satisfythe nondegeneracy condition (67).9.
Applications to separable Hamiltonian systems
In this section, in order to show the large range of applicability of the theorypreviously developed, we will discuss two important examples of integrable systems.The first one concerns a Hamiltonian system on a six-dimensional symplectic man-ifold, which is obtained as a stationary reduction of the seventh order equationof the Korteweg de Vries (KdV) hierarchy. The second one is a Drach-Holt typesystem, considered to be an example of nonseparable system till now.9.1.
The stationary reduction of the seventh order KdV flow revisited.
In [60], a method to obtain the Poisson pencil P − λP of the stationary flows of theKdV hierarchy was presented. In [48], this method was applied to get the stationaryreduction of the seventh order equation of the hierarchy. The restricted Poissonpencil turns out to be a degenerate pencil of co-rank one in a seven dimensionalmanifold M (7) , being therefore a Gelfand-Zakarevich system [24]. It possesses apolynomial Casimir function of length four, starting with a Casimir of P andending with a Casimir of P . Then, a Marsden-Ratiu reduction procedure [45],similar to the one used in other cases [60, 50, 51, 18, 20], was performed to eachsix-dimensional symplectic leaf S of the Poisson tensor P , in order to get rid ofthe Casimir of P . Furthermore, by restricting the polynomial Casimir function to S , one of the authors was able to obtain in [48] three Hamiltonian functions ininvolution in the Darboux chart ( q , q , q , p , p , p ) H = p p + 12 p − q + 52 q q + 12 q q − q , (163) H = 12 p + p p q + p q − p q − p p q − q − q q + 12 q q + 2 q q ,H = 12 p q + p q − p p q − p p q q + 12 p q + 12 q q − q q q − q . AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 37
However, as typically happens in the cases of Gelfand-Zakarevich systems [19],the reduced integrable Hamiltonian systems on S do not allow a bi-Hamiltonianformulation. Nevertheless, they can be described in the context of our new theory.In fact, proceeding analogously to the case of two degrees of freedom, we searchfor a Haantjes operator K whose minimal polynomial be of the maximum degreeallowed by our assumptions, in this case 3:(164) m K ( λ ) := λ − c ( x ) λ − c ( x ) λ − c ( x ) , and that satisfy K T Ω = Ω K (165) K T d H = d H (166) ( K T ) d H = (cid:0) c ( K T ) + c K T + c I (cid:1) d H (167) H K ( X, X ′ ) = 0 ∀ X, X ′ ∈ T M . (168)Under the simplest ansatz that its elements be linear in the Darboux coordinates( q , q , q , p , p , p ), we find the unique solution(169) K = q − q q − q − q q − p − p q − q − q p q p − q q . Since the minimal polynomial of K is of degree 3, by virtue of Proposition 55 K is a generator of the Haantjes ring K . Thus, we search for another Haantjesoperator K such that K = f I + g K + h K , (170) K T d H = d H , (171)where f , g , h are suitable smooth functions on M . The unique solution is K =( q + 2 q ) I − q K + K , therefore(172) K = − q q − q q − q q − q q + 2 q p q − p q q − q q − p − q − ( p q − p q ) p − q − q q q + 2 q . Since the ω H structure ( M, ω, I , K , K ) admits a maximal generator, i.e., K ,the Hamiltonian functions (163) have a set of DH coordinates as separation vari-ables. A set of such coordinates can be computed finding a Nijenhuis generator ofthe Haantjes module generated by K . This is what one of the present authors didin [48] by a complete different method, that is to say by means of the Marsden-Ratiu reduction procedure above mentioned. In that paper, separation coordinateswere called Darboux-Nijenhuis coordinates. In [28], the same coordinates have beenconsidered as orthogonal separation variables for the Hamiltonian function (163),in the cotangent bundle of a three–dimensional Minkowski space. Separation of variables for a Drach-Holt type system.
The approachproposed in this paper offers an effective procedure to construct algorithmicallyseparation variables. As a paradigmatic example, we shall study the case of a systemshowing an irrational dependence on its coordinates, namely a three-parametricdeformation of the Holt potential, that has been introduced in [15]:(173) H = 12 (cid:0) p x + p y (cid:1) + a x + 3 y y / + a xy / + a y / a , a , a ∈ R . This system is integrable in the manifold M = T ∗ ( E \ { y = 0 } ), with a third-orderintegral H = 2 p x + 3 p x p y + 12 a (cid:18) xy / p y + 2 x − y y / (cid:19) +(174) + a (cid:18) y / p y + 6 xy / p x (cid:19) + 6 a y / p x . When a →
0, the Hamiltonian H converts into a one-parametric deformationof the Post-Winternitz superintegrable potential (157). A crucial aspect is that H has been considered in the literature to be an example of nonseparable sys-tem , since, not being of classical St¨ackel type, it is not separable by means ofan extended–point transformation. A natural question is whether there is a fullcanonical transformation (in general difficult to find) redeeming its separability.Our theory of integrability `a la Haantjes enables us to solve this problem, since itprovides a set of DH coordinates by means of the procedure outlined in Subsection8.1.9.2.1. The Haantjes structure.
The prescription in step 1 provides with a Haantjesoperator linear in the momenta, which reads(175) K DH = 3 p x p y y p x − y − a y / p x a y / p y p x . It endows the manifold M with the ωH structure ( ω, I , K DH ). The Lenard-Haantjes chain is defined by d H = K TDH d H . According to the general theory developed above, the potential functions of theexact one-forms belonging to the eigen-distributions of K TDH provide the separationcoordinates. In particular, as stated in Remark 70, it can be checked that thissystem is quasi-bi-Hamiltonian. Therefore the eigenvalues of the Haantjes operator(175) are just characteristic functions of the web of K DH , due to eq. (156).9.2.2. Separation Coordinates.
These functions , for a >
0, read(176) λ = l = 6( p x + 3 √ a y / ) , λ = l = 6( p x − √ a y / ) . In order to get a system of DH coordinates, they can be completed with a pair ofconjugate momenta that have the non trivial expressions
AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 39 µ = 1576 r a (cid:18) p x + 12 √ a p x y / + 108 a p x y / + 216 p a p x y +12 p y y / − √ a x + 324 a y / (cid:19) , (177) µ = 1576 r a (cid:18) p x − √ a p x y / + 108 a p x y / − p a p x y − p y y / − √ a x + 324 a y / (cid:19) , and have been computed as potential functions of two exact 1-forms belonging tothe two characteristic eigen-distributions ( E ◦ , E ◦ , ) of L TDH , that locally decompose T ∗ M = E ◦ ⊕ E ◦ . The set of coordinates (176), (177) are separation variables forthe Hamiltonian functions H and H , due to Theorem 59.9.2.3. Separation Equations of Jacobi–Sklyanin.
In order to solve the Hamilton-Jacobi equation of the Drach-Holt model, one has to write down the Jacobi–Sklyanin equations that read b µ − (cid:0) b λ + b (cid:1) µ + b λ + b λ + b λ − λ H + H + b = 0 ,b µ − (cid:0) b λ + b (cid:1) µ + b λ + b λ − b λ + λ H − H + b = 0 , where b i , i = 1 , . . . , b = 10368 q a , b = a , b = 216 a √ a , b = √ a ,b = a , b = 1216 , b = 18 a √ a . We arrive therefore at the separated solutions of the Hamilton-Jacobi equation W ( λ ; h , h ) = b R λ (cid:18) − ( b λ + b ) ± p ( b λ + b ) − b P ( λ ) (cid:19) d λ ,W ( λ ; h , h ) = b R λ (cid:18) − ( b λ + b ) ± p ( b λ + b ) − b Q ( λ ) (cid:19) d λ , where h , h are the values of H , H on the lagrangian tori, and P ( λ ; h , h ) := b λ + b λ + b λ − λ h + h + b ,Q ( λ ; h , h ) := b λ + b λ − b λ + λ h − h + b . Future Perspectives
The extension of the present theory to the case of quantum integrable systemsis a nontrivial task. This research line would pave the way to an algebraic in-terpretation of the notion of Haantjes integrability developed here, in terms ofinfinite-dimensional commuting operators on a Hilbert space.Also, it would be interesting to compare the geometric structures underlying thevision offered here with the intrinsic, purely algebraic structures developed in [31],in the context of nilpotent integrability .A natural extension of the present theory to the case of superintegrable systems [59], especially maximally superintegrable ones, is in order. Along these lines, wealso wish to construct a generalization of our approach to the study of the geometry of certain classical systems, as the Post-Winternitz model of Section 8.2, that do notpossess any simple system of separation coordinates. We believe that our theory canoffer a proper language in which the study of the relation between superintegrabilityand separability indeed can be carried out.
Acknowledgement
The authors wish to thank heartily F. Magri for having drawn our attention tothe theory of Haantjes manifolds.P. T. acknowledges the support of the research project FIS2011–22566, Ministeriode Ciencia e Innovaci´on, Spain.G. T. acknowledges the financial support of the research project PRIN 2010-11“Geometric and analytic theory of Hamiltonian systems in finite and infinite di-mensions”. Moreover, he thanks many useful discussions with his colleagues andfriends S. Logar and D. Portelli.The authors also thank the Universidad Complutense de Madrid and the Universit`adi Trieste for providing financial support in the context of the scientific interchangeprogram between the two institutions.
References [1] Y. Aizawa, N. Saito,
On the stability of isolating integrals, I. Effect of the perturbation inthe potential function , J. Phys. Soc. Jpn, , 1636–1640 (1972).[2] M. A. Akivis, V. V. Goldberg, Differential Geometry of Webs . In: F.J.E. Dillen, L.C.A.Verstraeelen, eds. Handbook of Differential Geometry Vol. I, pp. 1–152, Elsevier, Amsterdam2000.[3] V.I. Arnold, V.V. Kozlov, A.I. Neishtadt,
Mathematical Aspects of Classical and CelestialMechanics , Springer, Berlin (1997).[4] S. Benenti,
Separability structures on Riemannian manifolds , Lect. Notes Math.
Inertia tensors and Stackel systems in the Euclidean spaces , Rend. Sem. Mat.Univ. Politec. Torino , 315–341 (1992).[6] S. Benenti, Orthogonal separable dynamical Systems , Math. Publ., Silesian Univ. Opava ,163–184 (1993).[7] S. Benenti, C. Chanu and G. Rastelli, Remarks on the connection between the additive sepa-ration of the Hamilton–Jacobi equation and the multiplicative separation of the Schr¨odingerequation. I. The completness and Robertson conditions , J. Math. Phys. , 5183–5222 (2002).[8] M. B laszak, Multi-Hamiltonian Theory of Dynamical Systems , Texts and Monographs inPhysics, Springer-Verlag, Berlin (1998).[9] O. I. Bogoyavlenskij,
Necessary Conditions for Existence of Non-Degenerate HamiltonianStructures , Commun. Math. Phys , 253-290 (1996).[10] O. I. Bogoyavlenskij,
General algebraic identities for the Nijenhuis and Haantjes tensors ,Izvestya Mathematics , 1129-1141 (2004).[11] O. I. Bogoyavlenskij, Schouten tensor and bi–Hamiltonian systems of hydrodynamic type , J.Math. Phys , paper n. 023504, pp. 14 (2006).[12] R. Brouzet, Systemes bihamiltoniens et complete integrabilite en dimension 4 , C.R. Acad.Sci. Paris , 895–898 (1990).[13] R. Brouzet,
About the existence of recursion operators for completely integrable Hamiltoniansystems near a Liouville torus , J. Math. Phys. , 1309–1313 (1993).[14] R. Brouzet, R. Caboz, J. Rabenivo, V. Ravoson, Two degrees of freedom quasi–bi–Hamiltonian systems , J. Phys. A , 2069–2076 (1996).[15] R. Campoamor-Stursberg, J. F. Cari˜nena & M. F. Ra˜nada, Higher-order superintegrabilityof a Holt related potential , J. Phys. A ,435202, 6 pp. (2013).[16] M. De Filippo, G. Vilasi, and M. Salerno A geometrical approach to the integrability of solitonequations , Lett. Math. Phys. , 85-91 (1985). AANTJES MANIFOLDS OF CLASSICAL INTEGRABLE SYSTEMS 41 [17] G. Falqui, F. Magri and G. Tondo,
Bi–Hamiltonian systems and separation of variables: anexample from the Boussinesq hierarchy , Theor. Math. Phys. , 176–192 (2000).[18] G. Falqui, M. Pedroni,
On a Poisson reduction for Gelfand-Zakharevich manifolds , Rep.Math. Phys. Separation of variables for bi–Hamiltonian systems , Math. Phys.,Anal. and Geom. , 139–179 (2003).[20] G. Falqui, M. Pedroni, Poisson Pencils, Algebraic Integrability, and Separation of Variables ,Regul. Chaotic. Dyn. On the matrix Hopf equation and integrable hamiltonian system of hy-drodynamic type, which do not possess Riemann invariants , Phys. Lett. A , 391-397(1993).[22] E. V. Ferapontov and D. G. Marshall,
Differential-geometric approach to the integrability ofhydrodynamics chains: the Haantjes tensor , Mat. Ann. , 61–99 (2007).[23] A. Frolicher and A. Nijenhuis,
Theory of Vector-Valued Differential Forms. Part I , Indag.Mathematicae , 338–359 (1956).[24] I. M. Gelfand, I. Zakharevich, Webs, Lenard schemes, and the local geometry of biHamilto-nian Toda and Lax structures , Selecta Math. , 131-183 (2000).[25] Gerdjikov V.S., Vilasi G. and Yanovski A.B.: Integrable Hamiltonian Hierarchies.
Lect. Not.Phys., Vol.
Berlin Heidelberg: Springer, 2008[26] D. Gutkin, Λ-
Cohomologie et op´erateurs de r´ecursion , Ann. Inst. Henri Poincar´e , 355-366(1987).[27] J. Haantjes, On X n − -forming sets of eigenvectors , Indag. Mathematicae , 158–162(1955).[28] J. T. Horwood, R. G. McLenaghan, R. G. Smirnov, Hamilton-Jacobi theory in three-dimensional Minkowski space via Cartan geometry , J. Math. Phys. paper 053507, 40pp. (2009).[29] Y. Kosmann-Schwarzbach, F. Magri, Poisson-Nijenhuis structures , Ann. Inst. H. PoincarPhys. Thor. , 3581 (1990).[30] A. Ibort, F. Magri and G. Marmo, Bi–Hamiltonian structures and St¨ackel separability , J.Geom. Phys., , 210–228 (2000).[31] A. Ibort, G. Marmo, M. A. Rodr´ıguez and P. Tempesta, Nilpotent integrability, reductionof dynamical systems and a third-order Calogero-Moser system , Preprint arXiv: 1501.05875(2014),[32] G. Landi, G. Marmo and G. Vilasi
Recursion operators: Meaning and existence for completelyintegrable systems , J. Math. Phys., , 800–815 (1994).[33] T. Levi–Civita, Sulla integrazione dell’equazione di Hamilton–Jacobi per separazione di vari-abili , Math. Ann. , 383–397 (1904).[34] F. Magri, A simple model of the integrable Hamiltonian equation , J. Math. Phys. , no. 5,1156–1162 (1978).[35] F. Magri, A Geometrical approach to nonlinear solvable equations , in Lecture Notes in Physics , (M. Boiti, F. Pempinelli, G. Soliani eds.), Springer–Verlag, Berlin (1980), pp. 233–263.[36] F. Magri,
Geometry and soliton equations . In: Atti Accad. Sci. Torino Suppl. physics ,(1990), pp. 181–209.[37] F. Magri,
Lenard chains for classical integrable systems , Theor. Math. Phys. , 1716–1722(2003).[38] F. Magri,
Recursion operators and Frobenius manifolds , SIGMA , paper 076, 7 pp. (2012).[39] F. Magri, Haantjes manifolds . Journal of Physics: Conference Series , paper 012028, 10pp. (2014).[40] F. Magri,
WDVV Equations , Preprint arXiv:1510.07950 (2015).[41] F. Magri,
Haantjes manifolds and Veselov Systems , Preprint arXiv:1510.07951 (2015).[42] F. Magri, T. Marsico,
Some developments of the concept of Poisson manifolds in the sense ofA. Lichnerowicz . In Gravitation, Electromagnetism and Geometric Structures, (G. Ferrarese,ed.), pp. 207–222, Pitagora editrice, Bologna 1996.[43] F. Magri, C. Morosi,
Characterization of Integrable Systems through the Theory of Poisson–Nijenhuis Manifolds , Quaderno S , Universit`a di Milano (1984).[44] G. Marmo and G. Vilasi, When do recursion operators generate new conservation laws? ,Phys. Lett. B , 137–140 (1992). [45] J. E. Marsden, T. Ratiu,
Reduction of Poisson Manifolds , Lett. Math. Phys. , 161–169(1986).[46] H.P. McKean, Compatible Brackets in Hamiltonian Mechanics . In Important Developmentin Soliton Theory, 344–354, Springer Ser. Nonlinear Dynamics, Springer, Berlin, (1993).[47] W. Miller Jr, S. Post and P. Winternitz,
Classical and Quantum Superintegrability withApplications , J. Phys. A, , 423001 (2013).[48] C. Morosi, G. Tondo, Quasi–bi–Hamiltonian systems and separability , J . Phys. A , 2799–2806 (1997).[49] C. Morosi, G. Tondo, On a class of dynamical systems both quasi–bi–Hamiltonian and bi–Hamiltonian , Phys. Lett. A , 59–64 (1998).[50] C. Morosi, G. Tondo,
The quasi-bi-Hamiltonian formulation of the Lagrange top , J. Phys. A , 1741–1750 (2002).[51] C. Morosi, G. Tondo, Separation of Variables in multi–Hamiltonian systems: an applicationto the Lagrange top , Theor. Math. Phys. , 1550–1560, (2003).[52] A. Nijenhuis, X n − -forming sets of eigenvectors , Indag. Mathematicae , 200-212 (1951).[53] A. Nijenhuis, Jacobi–type identities for bilinear differential concomitants of certain tensorfields I,II , Indag. Math , 390-397, 398-403 (1955).[54] A.M. Perelomov, Integrable Systems of Classical Mechanics and Lie Algebras , Birk¨auser,Basel (1990).[55] S. Post, P. Winternitz,
A non separable quantum superintegrable system in 2D real Euclideanspace , J. Phys. A , 162001, 8pp. (2011).[56] J. Praught, R. G. Smirnov, Andrew Lenard: a mystery unraveled , SIGMA , 005 (2005).[57] E. K. Sklyanin, Separation of variables: new trends , Prog. Theor. Phys. Suppl. , 35–60(1995).[58] P. Tempesta, G. Tondo,
Generalized Lenard chains, separation and Superintegrability , Phys.Rev. E , paper n. 046602, pp. 11 (2012).[59] P. Tempesta, P. Winternitz, J. Harnad, W. Miller, Jr, G. Pogosyan, M. A. Rodr´ıguez (eds), Superintegrability in Classical and Quantum Systems , Montr´eal, CRM proceedings and Lec-ture Notes, AMS, vol. (2004).[60] G. Tondo, On the integrability of stationary and restricted flows of the KdV hierarchy , J.Phys. A , 5097-5115, (1995).[61] G. Tondo, Generalized Lenard chains and multi–separability of the Smorodinsky-Winternitzsystem , J. Phys.: Conference Series , 012042, 10 pp. (2014).[62] G. Tondo, C. Morosi,
Bi-Hamiltonian manifolds, quasi–bi–Hamiltonian systems and Sepa-ration of Variables , Rep. Math. Phys. , 255–266 (1999).[63] G. Tondo, P. Tempesta, Haantjes structures for the Jacobi-Calogero model and the BenentiSystems , Preprint arXiv: 1511.00234 (2015), submitted to SIGMA.
Departamento de F´ısica Te´orica II, Facultad de F´ısicas, Universidad Complutense,28040 – Madrid, Spain and Instituto de Ciencias Matem´aticas, C/ Nicol´as Cabrera, No13–15, 28049 Madrid, Spain.
E-mail address : [email protected], [email protected] Dipartimento di Matematica e Geoscienze, Universit`a degli Studi di Trieste, piaz.leEuropa 1, I–34127 Trieste, Italy.
E-mail address ::