Variational Poisson-Nijenhuis structures for partial differential equations

We explore variational Poisson-Nijenhuis structures on nonlinear PDEs and establish relations between Schouten and Nijenhuis brackets on the initial equation with the Lie bracket of symmetries on its natural extensions (coverings). This approach allo…

Authors: Valentina Golovko, Iosif Krasilshchik, Alex

V ARIA TIONAL POISSON–NIJENHUIS STR UCTURES F OR P AR TIAL DIFFERENTIAL EQUA TIONS V. GOLO VKO, I. KRASIL ′ SHCHIK, AND A. VERBOVETSKY Abstract. W e explore v ariational Poisson–Nijenh uis structures on nonlinear PDEs and establish relations bet ween Schout en and Nijenhuis brac kets on the initial equation with the Lie brack et of symmetries on i ts natural extensions (co ve ri ngs). This approac h allows to construct a f ramework f or the theo ry of nonlocal structures. Introduction Poisson–Nijenh uis structures [1 6] play an imp ortant role b oth in classica l dif- ferential geometry (se e , for e x ample [1 , 10]) a nd in geometry of par tial differential equations, se e [11, 1 4]. In the la tter case existence of a Poisson–Nijenh uis s tructure virtually amo un ts to complete integrabilit y of the equation under co nsideration. Infinite-dimensional P ois son–Nijenhuis structures are well descr ib ed for the cas e of jets and for evolutionary different ial equatio ns rega rded as flows on the jet spac e. As fo r general differential equatio n, the corres p o nding theory was not introduced for a lo ng time. In our r elatively r ecent w or ks [7, 6] we outlined a n appr oach to the theory in a pplication to evolution equation in geometric a l setting. This appr o ach is based on the notion of ∆- c ov erings and reduces co ns truction of bo th r ecursion op erator s a nd Hamiltonian structures to solution of the linearised equation ℓ E (Φ) = 0 (1) on sp ecia l extensio ns of the initial equation E . W e call thes e extensio ns the ℓ - and ℓ ∗ -c overings and they play the ro le of tangent and cotangent bundles in the category of differ ent ial e quations. The ab ov e men tioned appro ach seems to work for general equations as well and we exp o s e its g eneralisatio n b elow. In Section 1 we define Poisson–Nijenh uis structur es in the “absolute case”, i.e., for the manifold of infinite jets. T o this end, w e r edefine the Schouten and F r¨ olicher– Nijenh uis bra ck ets (cf. with [7, 12], see als o [17]). W e als o expres s the compatibility condition betw een a Poisson bi-v ector and a Nijenhuis o per ator in ter ms of a sp e - cial bracket closely r elated to Vinog radov’s unifie d br acket , see [3]. The main r esult of this section is Theorem 2 that states the existence of infinite families of pair - wise compatible Hamiltonian structures related to the initial Poisson–Nijenhuis structure. In Section 2 Poisson–Nijenhuis str uctures on evolution equations are int ro duce d. W e show that an inv aria n t (with resp ect to the flow deter mined by the eq ua tion) Nijenhuis tensors are r e cu rsion op er ators for the symmetries, while inv ar iant Poisson bi- vectors amount to Hamiltonian structur e s. W e also define the ℓ - and ℓ ∗ -cov erings and r educe constructio n of rec ur sion op erato r s and Hamiltonian 2000 Mathematics Subje ct Classific ation. 37K05, 35Q53. Key wor ds and phr ases. Poisson–Nijenh uis structures, symmetries, conserv ation law, cov eri ngs, nonlocal structures. This work w as suppor ted in part b y the NW O–RFBR gran t 047.017.015 and RFBR– Consortium E.I.N.S.T. E.I.N grant 06-01-92060. 1 structures to solution of equatio n (1). The Schouten a nd F r¨ olicher–Nijenh uis brack- ets as well as the co mpa tibilit y conditions are reformulated in terms o f the Jaco bi brack ets of the corresp onding solutions and explicit for m ulas for these brackets ar e obtained. Section 3 genera lises the results to arbitrar y nonlinear partial differen- tial equatio n. Finally , in Section 4 we outline an approa ch to deal with nonlo c al Poisson–Nijenh uis structures. 1. V aria tional Poisson –Nijenhuis structures on J ∞ ( π ) 1.1. Geometrical structures. Let us r ecall definitions and res ults we sha ll use. F or details w e r efer to [2]. Let π : E → M be a vector bundle o ver an n -dimensional ma nifold M and π ∞ : J ∞ ( π ) → M be the infinite jet bundle of lo cal sections o f the bundle π . If x 1 , . . . , x n are lo cal co ordina tes in the bas e a nd u 1 , . . . , u m are co or dinates alo ng the fib er of π the c anonic al c o or dinates u j σ arise in J ∞ ( π ) defined by j ∞ ( s ) ∗ ( u j σ ) = ∂ | σ | s j ∂ x σ , where s = ( s 1 , . . . , s m ) is a loca l sec tio n of π , j ∞ ( s ) is its infinite jet and σ = i 1 i 2 . . . i | σ | , i α = 1 , . . . , n , is a multi-index. Denote by F ( π ) the algebra of smo oth functions on J ∞ ( π ). The ba sic geometrica l structure o n J ∞ ( π ) is the Cartan distribution C tha t is spanned by total derivatives D i = ∂ ∂ x i + X j,σ u j σi ∂ ∂ u j σ . Different ial o p erators on J ∞ ( π ) in total deriv a tives will be ca lled C - differ en t ial op er ators . In lo cal c o ordinates, they hav e the form k P σ a σ ij D σ k , where a σ ij ∈ F ( π ). Let P and Q be F ( π )-mo dules of sectio ns of some v ector bundles over J ∞ ( π ). All C - differential op erators fro m a P to Q for m an F ( π )-mo dule denoted by C Diff ( P , Q ). The adjoint op er ator to a C -differential opera tor ∆ : P → Q is denoted by ∆ ∗ : ˆ Q → ˆ P , where ˆ P = Hom F ( π ) ( P, ¯ Λ n ( π )) and ¯ Λ n ( π ) is the F ( π )-mo dule of horizo nt al n - form on J ∞ ( π ), i.e., forms ω = a dx 1 ∧ · · · ∧ dx n . Denote by C Diff sk ew ( k ) ( P, Q ) the mo dule o f k -linear skew-symmetric C -differential op erator s P × · · · × P → Q and by C Diff sk-ad ( k ) ( P, ˆ P ) ⊂ C Diff sk ew ( k ) ( P, ˆ P ) the subset of op erator s skew-adjoin t in each argument. A π ∞ -vertical vector field on J ∞ ( π ) is called evolutionary if it preser ves the Car - tan distribution. There is a o ne-to-one cor resp ondence b etw een e volutionary vector fields and sectio ns o f the bundle π ∗ ∞ ( π ). Denote b y κ ( π ) the corr esp onding mo dule of sections. In lo cal co o r dinates, the evolutionary vector field that co rresp onds to a section (the gener ating s ection, or function) ϕ = ( ϕ 1 , . . . , ϕ m ) is of the form З ϕ = X j,σ D σ ( ϕ j ) ∂ ∂ u j σ . The comm utator of evolutionary vector fields induces in κ ( π ) a Lie algebr a struc- ture that is given b y the higher Jac obi br acket defined a s a unique se c tio n { ϕ, ψ } satisfying [ З ϕ , З ψ ] = З { ϕ, ψ } . The bracket { ϕ, ψ } is express ed b y { ϕ, ψ } = З ϕ ( ψ ) − З ψ ( ϕ ) . F ollowing [7, 8], we s hall call elements of κ ( π ) variational ve ctors , elements of the module C Diff sk-ad ( k − 1) ( ˆ κ , κ ) will be called variatio nal k -ve ctors , while elemen ts of ˆ κ will b e called variational 1 -forms and elements o f the mo dule C Diff sk-ad ( k − 1) ( κ , ˆ κ ) 2 v aria tional k - forms , r esp ectively . The Lie deriv ative on v ariationa l vectors L ϕ : κ → κ takes the form L ϕ = З ϕ − ℓ ϕ , (2) where the line arization op er ator ℓ ϕ is defined by the equalit y ℓ ϕ ( α ) = З α ( ϕ ), α ∈ κ . Un lo cal c o ordinates, it has the form ℓ ϕ ( α ) = X j,σ ∂ ϕ ∂ u j σ D σ ( α j ) , α = ( α 1 , . . . , α m ) . The Lie der iv ative on v ar iational forms L ϕ : ˆ κ → ˆ κ is of the form L ϕ = З ϕ + ℓ ∗ ϕ . (3) 1.2. V ariational P o isson–Ni jenh uis structures. Recall (see [7]) that the vari- ational Schouten br acket of t wo ope rators A , B ∈ C Diff sk-ad ( ˆ κ , κ ) is defined by [ [ A, B ] ]( ψ 1 , ψ 2 ) = − ℓ A, ψ 1 ( B ψ 2 ) + ℓ A, ψ 2 ( B ψ 1 ) − A ( ℓ ∗ B , ψ 1 ( ψ 2 )) − ℓ B , ψ 1 ( Aψ 2 ) + ℓ B , ψ 2 ( Aψ 1 ) − B ( ℓ ∗ A, ψ 1 ( ψ 2 )) , ψ 1 , ψ 2 ∈ ˆ κ . (4) An o pe rator A is called H amiltonian if the [ [ A, A ] ] = 0 and tw o Ha miltonian oper - ators A a nd B are c omp atible if their Schouten brack et v anishes. Remark 1. Here and b elow the notation ℓ ∆ ,p 1 ,...,p n ( ϕ ), ϕ ∈ κ , for a C - differential op erator ∆ : P × · · · × P → Q means ℓ ∆ ,p 1 ,...,p n ( ϕ ) = З ϕ (∆)( p 1 , . . . , p n ) , p 1 , . . . , p n ∈ P. F or tw o op er ators R , S ∈ C Diff ( κ , κ ) their F r¨ olic her–Nijenhuis br acket (cf. with [12], see a lso [17]) is defined by [ R, S ] F N ( ϕ 1 , ϕ 2 ) = { R ϕ 1 , S ϕ 2 } + { S ϕ 1 , R ϕ 2 } − R ( { S ϕ 1 , ϕ 2 } + { ϕ 1 , S ϕ 2 } − S { ϕ 1 , ϕ 2 } ) − S ( { Rϕ 1 , ϕ 2 } + { ϕ 1 , R ϕ 2 } − R { ϕ 1 , ϕ 2 } ) , ϕ 1 , ϕ 2 ∈ κ . (5 ) If [ R, R ] F N = 0 we shall refer to R as a Nijenhuis op er ator . F o r particula r co mpu- tations it is conv enient to use the equality [ R, S ] F N ( ϕ 1 , ϕ 2 ) = − ℓ R, ϕ 1 ( S ϕ 2 ) − ℓ S, ϕ 1 ( Rϕ 2 ) + ℓ R, ϕ 2 ( S ϕ 1 ) + ℓ S, ϕ 2 ( Rϕ 1 ) + R ( ℓ S, ϕ 1 ( ϕ 2 ) − ℓ S, ϕ 2 ( ϕ 1 )) + S ( ℓ R, ϕ 1 ( ϕ 2 ) − ℓ R, ϕ 2 ( ϕ 1 )) . (6) Definition 1 (cf. with [11]) . A Hamiltonia n op erato r A ∈ C Diff sk-ad ( ˆ κ , κ ) and a Nijenh uis op er ator R ∈ C Diff ( κ , κ ) constitute a variational Poisson–Nijenhuis structur e ( A, R ) on J ∞ ( π ) if the following compatibilit y co nditions ho ld (i) R ◦ A = A ◦ R ∗ , (7) (ii) C ( A, R )( ψ 1 , ψ 2 ) = L Aψ 1 ( R ∗ ψ 2 ) − L Aψ 2 ( R ∗ ψ 1 ) + R ∗ L Aψ 2 ( ψ 1 ) − R ∗ L Aψ 1 ( ψ 2 ) + E h ψ 1 , AR ψ 2 i − R ∗ E h ψ 1 , Aψ 2 i = 0 , (8) where E : ¯ H n ( π ) → ˆ κ is the Euler op er ator and ¯ H n ( π ) is the n th ho rizontal de Rham co homology gr oup, while h . , . i : ˆ κ × κ → ¯ H n ( π ) is the natural pa iring. In terms o f linearizatio n oper ators condition (ii) has the form C ( A, R )( ψ 1 , ψ 2 ) = − ℓ R ∗ , ψ 1 ( Aψ 2 ) + ℓ R ∗ , ψ 2 ( Aψ 1 ) + ℓ ∗ A, ψ 1 ( R ∗ ψ 2 ) + ℓ ∗ R ∗ , ψ 1 ( Aψ 2 ) − R ∗ ( ℓ ∗ A, ψ 1 ( ψ 2 )) = 0 . Similarly to the finite-dimensional c a se, we ha ve the following Prop ositio n 1. L et a Hamiltonian op er ator A ∈ C Diff sk-ad ( ˆ κ , κ ) and a Nijenhuis op er ator R ∈ C Diff ( κ , κ ) define a Poisson–Nijenhuis structur e on J ∞ ( π ) . Then the c omp osition R ◦ A is a Hamiltonian op er ator c omp atible with A . 3 Pr o of. By straightforward computations one can prov e that [ [ RA, RA ] ]( ψ 1 , ψ 2 ) − 2 R [ [ A, R A ] ]( ψ 1 , ψ 2 ) + R 2 [ [ A, A ] ]( ψ 1 , ψ 2 ) − [ R, R ] F N ( Aψ 1 , Aψ 2 ) = 0 (9 ) and 2[ [ A, RA ] ]( ψ 1 , ψ 2 ) − [ [ A, A ] ]( R ∗ ψ 1 , ψ 2 ) − [ [ A, A ] ]( ψ 1 , R ∗ ψ 2 ) − 2 A ( C ( A, R )( ψ 1 , ψ 2 )) = 0 , (10) from where the statement follo ws immediately .  Theorem 2 . L et a Hamiltonian op er ator A ∈ C Diff sk-ad ( ˆ κ , κ ) and a Nijenhuis op er ator R ∈ C Diff ( κ , κ ) define Poisson–Nijenhuis structur e on J ∞ ( π ) . Then on J ∞ ( π ) ther e is a hier ar chy of iter ate d Hamiltonian op er ators , that is a se- quenc e of Hamiltonian op er ators R i A , i ≥ 0, which ar e p air-wise c omp atible , i.e. , [ [ R i A, R j A ] ] = 0, i, j ≥ 0 . Pr o of. The pro of is by inductio n on n = max( i, j ). F or n = 1 the statement follows from the pr op osition ab ov e. Assume now that [ [ R i A, R j A ] ] = 0, i, j = 0 , . . . , n , and A ( C ( R i A, R j )) = 0, i + j ≤ n , and let us prov e that [ [ R i A, R n +1 A ] ] = 0, i = 0 , . . . , n + 1, and A ( C ( R i A, R j )) = 0 , i + j ≤ n + 1. First, note that b y direct computations one ca n prove the follo wing formulas [ [ RA, RB ] ]( ψ 1 , ψ 2 ) − R [ [ RA, B ] ]( ψ 1 , ψ 2 ) − R [ [ A, R B ] ]( ψ 1 , ψ 2 ) + R 2 [ [ A, B ] ]( ψ 1 , ψ 2 ) − [ R, R ] F N ( Aψ 1 , B ψ 2 ) − [ R, R ] F N ( B ψ 1 , Aψ 2 ) = 0 , (11) [ [ RA, B ] ]( ψ 1 , ψ 2 ) + [ [ A, RB ] ]( ψ 1 , ψ 2 ) − [ [ A, B ] ]( R ∗ ψ 1 , ψ 2 ) − [ [ A, B ] ]( ψ 1 , R ∗ ψ 2 ) − A ( C ( B , R )( ψ 1 , ψ 2 )) − B ( C ( A, R )( ψ 1 , ψ 2 )) = 0 , (12) C ( RA, R )( ψ 1 , ψ 2 ) + C ( A, R 2 )( ψ 1 , ψ 2 ) − C ( A, R )( R ∗ ψ 1 , ψ 2 ) − C ( A, R )( ψ 1 , R ∗ ψ 2 ) − R ∗ ( C ( A, R )( ψ 1 , ψ 2 )) = 0 . (13) Let us substitute R n A a nd R n − l A , l = 1 , . . . , n , for A and B in (12), r esp ectively . Then we get [ [ R n +1 A, R n − l A ] ] − R n − l A ( C ( R n A, R )) = 0 . (14) Now let us take R n − 1 A , R n − l A , l = 1 , . . . , n , and R 2 for A , B and R in (12) , resp ectively . Then w e have [ [ R n +1 A, R n − l A ] ] − R n − l A  C ( R n − 1 A, R 2 )  = 0 . (15) If we ha ve in (13) R n − 1 A for A w e ge t C ( R n A, R ) + C ( R n − 1 A, R 2 ) = 0 . (16) Therefore, ta king the sum o f (14) and (15) we obtain tha t [ [ R n +1 A, R n − l A ] ] = 0 for l = 1 , . . . , n . Let us substitute now R n − 1 A a nd R n A for A and B in (11), res pe c tively , and then R n A for A in (9). Thus we get [ [ R n +1 A, R n A ] ] = 0 a nd [ [ R n +1 A, R n +1 A ] ] = 0. In order to pr ove that A ( C ( R i A, R j )) = 0 for i + j ≤ n + 1 , one has to put B = R n − l A , l = 0 , . . . , n and ta ke R l +1 for R in (12).  F or subsequen t constr uctions we need the op era tor C ∗ ( A, R )( ψ , ϕ ) = − ℓ A, ψ ( Rϕ ) + ℓ R, ϕ ( Aψ ) + R ( ℓ A, ψ ( ϕ )) + A ( ℓ ∗ R, ϕ ( ψ ) − ℓ R ∗ , ψ ( ϕ )) defined by h C ( A, R )( ψ 1 , ψ 2 ) , ϕ i = h ψ 2 , C ∗ ( A, R )( ψ 1 , ϕ ) i . (17) 4 2. V aria tional P oisson–Nijenhuis structure s on evolution equa tions 2.1. Symmetries and cosymmetries. Cons ider a system of e volution equatio ns E = { F = u t − f ( x, t, u, u 1 , . . . , u k ) = 0 } , (18) where b oth u = ( u 1 , . . . , u m ) and f = ( f 1 , . . . , f m ) ar e vectors a nd u t = ∂ u/ ∂ t , u k = ∂ k u/∂ x k . F or simplicity , w e consider the case of one space v ariable x , though everything works in ge ne r al situa tio n a s well. Denote by F ( E ) the algebra o f smo o th functions on E . Recall tha t equation (18) E can be unders to o d as the space J ∞ ( π ) × R with the Cartan distribution generated by the fields D x and D t = ∂ /∂ t + З f . H ere t the co ordinate along R . In lo cal co ordinates , these fields are of the for m D x = ∂ ∂ x + m X j =1 X k ≥ 0 u j k +1 ∂ ∂ u j k , D t = ∂ ∂ t + m X j =1 X k ≥ 0 D k x ( f j ) ∂ ∂ u j k . A symmetry of the equation E is a π ∞ -vertical vector field on E that preserves the Cartan distr ibution. The set o f a ll symmetr ie s for ms a Lie algebra over R denoted b y sym( E ) and there is a one-to- one corr esp ondence b etw een sym( E ) a nd smo oth sections ϕ ∈ Γ( π ∗ ∞ ( π )) = κ ( E ) satisfying the equa tion ℓ E ( ϕ ) = 0 , (19) where ℓ E = D t − ℓ f is the linea rization op er ator of E . A c onservation law fo r the equation (18) is a horizontal 1-form η = X dx + T dt closed with resp e ct to ho rizontal de Rham differen tial ¯ d : ¯ Λ 1 ( E ) → ¯ Λ 2 ( E ), i.e ., such that D t ( X ) = D x ( T ) , where X , T ∈ F ( E ). A co nserv a tion law η is trivial if it is of the for m η = ¯ dh , h ∈ F ( E ). The space of equiv alence classes of conser v ation laws coincides with the first horizontal de Rham co homology group and is denoted b y ¯ H 1 ( E ). T o any conserv ation law η = X dx + T dt there co r resp ond its g enerating function ψ η = E ( η ) that satis fies the equation ℓ ∗ E ( ψ η ) = 0 . (20) Solutions of the last equa tion are called c osymmetries of e q uation E and the space of cosymmetries of E will b e denoted by sym ∗ ( E ). 2.2. In v arian t P oiss o n–Nijenhuis structures. Consider the mo dules κ and ˆ κ on the space J ∞ ( π ) × R of e x tended jets, i.e., we admit explicit dep endence of their elements on t . Since v ector fields on E act on κ and ˆ κ by Lie deriv atives, w e can give the following Definition 2. An op erator O acting from κ to κ (or from κ to ˆ κ , etc.) is called inv ar iant if L D t ◦ O = O ◦ L D t . Prop ositio n 3. A n op er ator A : ˆ κ → κ is invariant iff ℓ F ◦ A + A ◦ ℓ ∗ F = 0 . (21) An op er ator R : κ → κ is invariant iff ℓ F ◦ R − R ◦ ℓ F = 0 . (22) Pr o of. Indeed, from (2 ) one has for the action on κ L D t = L ∂ ∂ t + З f = ∂ ∂ t + З f − ℓ f = ℓ F , while from (3) for the action on ˆ κ we obtain L D t = L ∂ ∂ t + З f = ∂ ∂ t + З f + ℓ ∗ f = − ℓ ∗ F 5 and this pr ov es bo th (21) a nd (22).  Remark 2. F rom (22) we see that an in v ariant C -differential oper atos takes sym- metries of equation E to s y mmetries, i.e., is a r e cursion op er ator for symmetries o f E . On the other hand, a C - differential op erator A that e njoys (21) tak es co symmetries to symmetries. As it was shown in [7], if A is skew-adjoint and satisfies [ [ A, A ] ] = 0 it is a Hamiltonian structu re for E and a ll Hamiltonian structures can b e found in such a w ay . Definition 3. Let ( A, R ) b e a Poisson–Nijenhuis structure on J ∞ ( π ) × R . W e say that it is a Poisson–Nijenhuis structure on E if b oth A and R ar e inv a riant op erator s . Thu s, a Poisson–Nijenhuis o n E c o nsists of a Hamiltonian structur e A and a recursion op era tor which is a Nijenhuis ope rator compatible with A . 2.3. The ℓ ∗ - and ℓ -co v erings. W e shall now lo ok a t Poisson–Nijenhuis str uctures from a differen t p o in t of view. T o this end, consider the following extension of the equation E . Let us add to E new o dd v ar iable p = ( p 1 , . . . , p m ) that satisfies p t = − ℓ ∗ f ( p ) . (23) The system consisting of the initial equation E a nd equation (23) is ca lled the ℓ ∗ -c overing of E and is denoted by L ∗ E . The extended total deriv a tives on L ∗ E are ˜ D x = D x + m X j =1 X k ≥ 0 p j k +1 ∂ ∂ p j k , ˜ D t = D t − m X j =1 X k ≥ 0 ˜ D k x ( ˜ ℓ ∗ f ( p j )) ∂ ∂ p j k . Note that a ny C -differential op erato r O o n E can b e lifted to an o p erator ˜ O on L ∗ E by sup er s cribing tildes o ver co rresp onding total de r iv atives. Let A : ˆ κ → κ be a C -differential op erator o f the form k P i ≥ 0 a l ij D i x k . Consider a p -linear vector function H A = ( H 1 A , . . . , H m A ), H l A = X i,j a l ij p j i , a l ij ∈ F ( E ) . (24) Theorem 4 (see [7]) . A skew-adjoint op er ator A satisfies (21) iff ˜ ℓ E ( H A ) = 0 . (25 ) In the terminolo gy o f the cov ering theory [1 3], a vector function satisfying (25) is nothing but a shadow of s y mmetry o f E in the ℓ ∗ -cov ering or, mor e pr ecisely , H A is genera ting section of the shadow ˜ З H A = X k ≥ 0 ˜ D k x ( H j A ) ∂ ∂ u j k . In co or dinate-free terms, a s ha dow is a der iv ation of F ( E ) with v alues in the function algebra on L ∗ E that pr eserves the Cartan distribution. Lemma 5. L et A ∈ C Diff sk-ad ( ˆ κ , κ ) and ˜ З H A b e the c orr esp onding shadow. Then ther e exists a symmet ry of the e quation L ∗ E su ch that its r estriction to F ( E ) c oin- cides with ˜ З H A . Pr o of. Consider the v ector function α = ( α 1 , . . . , α m ) defined b y α = − 1 2 ˜ ℓ ∗ H A ( p ) and set ˜ З H A , ¯ H A = ˜ З H A + X k,j ˜ D k x ( α j ) ∂ ∂ p j k . (26) 6 It is ea sily chec ked that the field ˜ З H A , ¯ H A is the desired symmetry .  Consider now t wo oper ators A , B ∈ C Diff sk-ad ( ˆ κ , κ ) and using Lemma 5 define the Jac obi br acket of H A and H B by {H A , H B } = ˜ З H A , ¯ H A ( H B ) + ˜ З H B , ¯ H B ( H A ) (27) Remark 3. The plus sign in the r ight-hand side of equation (27) is due to the fact that b oth the fields ˜ З H A , ¯ H A , ˜ З H B , ¯ H B and the functions H B , H A are o dd. Remark 4. In more explicit ter ms the Jaco bi bracket ca n b e rewritten in the for m {H A , H B } = − ˜ ℓ H A ( H B ) − ˜ ℓ H B ( H A ) − A ( ˜ ℓ ∗ H B ( p )) + B ( ˜ ℓ ∗ H A ( p )) 2 . (28) Note that ther e exists a one-to-one cor r esp ondence b etw een C -differential op era- tors from the mo dule C Diff sk ew ( k ) ( ˆ κ , κ ) and functions on L ∗ E tha t are ( k − 1)-linear with resp ect to the o dd v aria bles. F or ∆ ∈ C Diff sk ew ( k ) ( ˆ κ , κ ) denote b y H ∆ the corres p o nding function. Then b y dire ct computations one can prove the following Theorem 6. L et H A and H B b e shadows of symmetries in the ℓ ∗ -c overing to which ther e c orr esp ond op er ators A, B ∈ C Diff sk-ad ( ˆ κ , κ ) . Then {H A , H B } = H [ [ A,B ] ] . (29) If ψ 1 and ψ 2 are sections of the ℓ ∗ E -cov ering such that ( dψ i )( C ) ⊂ ˜ C , i = 1 , 2, where C and ˜ C are the Cartan distributions o n E and L ∗ E , resp ectively , then equation (29) can b e r ewritten as follows ψ ∗ 1 ψ ∗ 2 {H A , H B } = [ [ A, B ] ]( ψ 1 , ψ 2 ) . W e now intro duce the ob ject dual to the ℓ ∗ -cov ering. Namely , c o nsider the extension of E b y new o dd v ar iables q = ( q 1 , . . . , q m ) that satisfy the equatio n q t = ℓ f ( q ) . (30) The system consisting o f the initial equation and equation (30) will b e called the ℓ -c overing o f E a nd denoted by LE . The extended total deriv a tives on LE are ˜ D x = D x + m X l =1 X k ≥ 0 q l k +1 ∂ ∂ q l k , ˜ D t = D t + m X l =1 X k ≥ 0 ˜ D k x ( ˜ ℓ f ( q l )) ∂ ∂ q l k . T o any C -differential op erato r R : κ → κ of the for m P i ≥ 0 a l ij D i x let us put into corres p o ndence a q -linear vector function N R = ( N 1 R , . . . , N m R ), where N l R = X i,j a l ij q j i , a l ij ∈ F ( E ) . Similar to Theorem 4 we hav e the following Theorem 7 (see [7, 6]) . An op er ator R satisfies r elation (22), i.e , is a r e cursion op er ator for the e quation E iff ˜ ℓ E ( N R ) = 0 , (31) i.e. , N R is a shado w of a symmetry in the ℓ - c overing. The corre sp onding deriv atio n is of the fo r m ˜ З N R = X k ≥ 0 ˜ D k x ( N j R ) ∂ ∂ u j k . In par a llel to Lemma 5 w e have the followin g auxilia ry result. 7 Lemma 8. L et R ∈ C Diff ( κ , κ ) and ˜ З N R b e the c orr esp onding shadow. Then t her e exists a symmet ry of the e quation LE su ch t hat its r estriction t o F ( E ) c oincides with ˜ З N R . Pr o of. Consider the v ector function β = ( β 1 , . . . , β m ) with β j = X k,l ∂ N j R ∂ u l k q l k and set ˜ З N R , ¯ N R = ˜ З N R + X k,j ˜ D k x ( β j ) ∂ ∂ q j k . (32) This is the symmetry we are lo oking for .  Using Lemma 8 we define the Jac o bi brack et o f vector functions N R and N S by {N R , N S } = ˜ З N R , ¯ N R ( N S ) + ˜ З N S , ¯ N S ( N R ) . (33) In explicit terms the Ja cobi brack et has the for m {N R , N S } = − ˜ ℓ N R ( N S ) − ˜ ℓ N S ( N R ) + R ( ˜ ℓ N S ( q )) + S ( ˜ ℓ N R ( q )) . (34) Denote b y N [ R,S ] F N the function o n LE bilinear with resp ect to the v a riables q and corresp onding to the op era tor [ R, S ] F N ∈ C Diff sk ew (2) ( κ , κ ). Then b y the dir e c t computations we obtain Theorem 9. L et N R and N S b e shadows of symmetries in the ℓ -c overing to which ther e c orr esp ond op er ators R and S ∈ C Diff ( κ , κ ) . Then {N R , N S } = N [ R,S ] F N . If ϕ 1 , ϕ 2 are sectio ns of the ℓ -cov ering preserv ing the Ca rtan distributions then ϕ ∗ 1 ϕ ∗ 2 {N R , N S } = [ R, S ] F N ( ϕ 1 , ϕ 2 ) . 2.4. Compatibility condition. W e shall now e xpress the compatibility conditio n of a Hamiltonian structure A and a recur sion oper ator R on equa tion (18) in simila r geometric terms. T o this end, r ecall that b oth LE and L ∗ E are fib ere d over the equation E and denote the corr esp onding fiber bundles b y τ : LE → E and τ ∗ : LE → E , r esp ectively . Consider the Whitney pro duct τ ⊗ τ ∗ : LE × E L ∗ E → E o f these bundles. One can extend the total deriv atives to the s pace LE × E L ∗ E by setting ˜ D x = D x + m X l =1 X k ≥ 0  q l k +1 ∂ ∂ q l k + p l k +1 ∂ ∂ p l k  , ˜ D t = D t + m X l =1 X k ≥ 0  ˜ D k x ( ˜ ℓ f ( q l )) ∂ ∂ q l k − ˜ D k x ( ˜ ℓ ∗ f ( p l )) ∂ ∂ p l k  . Thu s the equation LE × E L ∗ E amounts to E ex tended bo th by (23) and (30). Due to the natural pr o jections LE × E L ∗ E → L ∗ E and LE × E L ∗ E → LE , the vector fields ˜ З H A , ¯ H A and ˜ З N R , ¯ N R (see equalities (26) and (32)) may b e c onsidered as deriv ations fro m F ( L ∗ E ) and F ( LE ) to F ( LE × E L ∗ E ), res p ectively . Lemma 10 . Ther e exist symmetries ˜ З H A , ¯ H A ,α and ˜ З N R , ¯ N R ,ρ of e quation LE × E L ∗ E such that their r estrictions to the function algebr as F ( L ∗ E ) and F ( LE ) c oincide with ˜ З H A , ¯ H A and ˜ З N R , ¯ N R , r esp e ctively. 8 Pr o of. Let us set ˜ З H A , ¯ H A ,α = ˜ З H A , ¯ H A + X k,j ˜ D k x ( α j ) ∂ ∂ q j k , (35) where the vector function α = ( α 1 , . . . , α m ) is defined by α = ˜ ℓ H A ( q ). Let also consider the field ˜ З N R , ¯ N R ,ρ = ˜ З N R , ¯ N R + X k,j ˜ D k x ( ρ j ) ∂ ∂ p j k , (36) where ρ = ( ρ 1 , . . . , ρ m ) is defined b y ρ = − ˜ ℓ ∗ N R ( p ) − ˜ З q ( R ∗ )( p ) . By direct computations one ca n chec k that the fields (35) and (36) p ossess the needed pro pe rties.  Using Lemma 10 let us define the brack et {H A , N R } = ˜ З H A , ¯ H A ,α ( N R ) + ˜ З N R , ¯ N R ,ρ ( H A ) , or, in explicit terms, {H A , N R } = − ˜ ℓ N R ( H A ) − ˜ ℓ H A ( N R ) − A  ˜ ℓ ∗ N R ( p ) + ˜ З q ( R ∗ )( p )  + R ( ˜ ℓ H A ( q )) . (37) Denote by C ∗ A,R the bilinear with resp ect to the v aria bles p and q function o n LE × E L ∗ E c orresp onding to the C -differ e n tial op era tor C ∗ ( A, R ) : ˆ κ × κ → κ (see equality (17)). Then the following res ult holds: Theorem 11. L et H A b e a shadow in the ℓ ∗ -c overing to which ther e c orr esp onds an op er ator A ∈ C Diff sk-ad ( ˆ κ , κ ) and N R b e a shadow in the ℓ -c overing to which ther e c orr esp onds an op er ator R ∈ C Diff ( κ , κ ) . Then {H A , N R } = C ∗ A,R . F rom the ab ov e said we get the following Theorem 12 . L et a Hamiltonian op er ator A ∈ C Diff sk-ad ( ˆ κ , κ ) and a r e cursion op er ator R ∈ C Diff ( κ , κ ) define Poisson–Nijenhuis stru ct ur e on evolution e qua- tion E , while H A and N R b e t he c orr esp onding shadows in the ℓ ∗ - and ℓ -c overing over E , r esp e ctively. Then (i) {H A , H A } = 0 , (38) (ii) {N R , N R } = 0 , (39 ) (iii) {H A , N R } = 0 . (4 0) 3. V aria tional P oisson–Nijenhuis structures in general case Consider now the infinite pr olonga tio n E ⊂ J ∞ ( π ) of a genera l differe ntial equa - tion as a submanifold in the space J ∞ ( π ) of infinite jets o f some lo cally triv ia l bundle π : E → M . In loc a l co ordina tes E is given b y the system F j ( x 1 , . . . , x n , u 1 , . . . , u m , . . . , u j σ , . . . ) = 0 , j = 1 , . . . , r . W e assume that F = ( F 1 , . . . , F r ) ∈ P , wher e P is the mo dule of sections of some vector bundle ov er J ∞ ( π ). Co nsider the linear ization op er ator ℓ E : κ → P and its a djoint ℓ ∗ E : ˆ P → ˆ κ . Similar to the evolutionary case, we can construct ℓ - a nd ℓ ∗ -cov erings by ex tending E with ℓ E ( q ) = 0 and ℓ ∗ E ( p ) = 0. 9 F ollowing the scheme exp os ed in Section 2, we are lo o king for C -differential op erator s such that the diagrams (i) κ ℓ E / / R   P ¯ R   κ ℓ E / / P , (ii) ˆ P ℓ ∗ E / / A   ˆ κ ¯ A   κ ℓ E / / P (41) are comm utative. Literally cop ying the co nstruction o f Section 2, we can put into corres p o ndence to an op era tor R from the firs t diag r am in (41) a q -linea r func- tion N R = ( N 1 R , . . . , N m R ) on LE , while for A from the second diagr am we construct a p -linea r function H A = ( H 1 A , . . . , H m A ) o n L ∗ E . W e ag ain use the gener al result prov ed in [7]: Theorem 13. An op er at or R fits diagr am (i) in (41) iff ˜ ℓ E ( N R ) = 0 , (42) while A fits dia gr am (ii) iff ˜ ℓ E ( H A ) = 0 . (43 ) Among the o p erators A let us distinguish those ones that enjoy the prop er ty similar to s kew-adjoin tness. Namely , we shall cons ide r oper ators such that A ∗ = − ¯ A. (44) Remark 5. This pro p erty means that the op era tor ( A, ¯ A ) : ˆ P ⊕ ˆ κ → κ ⊕ P is skew-adjoin t. Note now that explicit expressio ns (28), (34) and (37) for Jaco bi br ack ets do not r ely on the fact that they were given for evolutionary eq uations. Using this observ ation we g ive the following Definition 4. Let E ⊂ J ∞ ( π ) b e a different ial equation. (1) A C -differential op era tor A : ˆ P → κ is called a Hamiltonian structur e on E if it fits the left diagram (41), equation (44 ) holds and {H A , H A } = 0, where the brack et is given by (28). Two Hamilto nia n structures A and B are said to be c omp atible if {H A , H B } = 0. (2) A C -differential oper ator R : κ → κ is calle d a Nijenhuis op er ator for the equation E if it fits the right diagram (41) a nd {N R , N R } = 0, where the brack et is given b y (34). (3) A pair of C -differential op erator s ( A, R ) is called a Poisson–Nij enhuis st r u c- tur e o n E if R is a Nijenhuis op erator, A is a Ha milto nian structur e such that A ∗ ◦ R ∗ = ¯ R ◦ A ∗ and {N R , H A } = 0, where the bra ck et is given by (37). Remark 6 . If ( A, R ) is a P oiss on–Nijenhuis s tructure on E then R is a recursion op erator for symmetries of E . Mor eov er, a Hamiltonian structure A , similar to the evolutionary case, deter mines a Poisson brack et on the gro up of conserv atio n laws of E . Namely , if ω 1 , ω 2 are conserv ation laws then we set { ω 1 , ω 2 } A = L d 0 ,n − 1 1 ω 1 ( ω 2 ) , where d 0 ,n − 1 1 : E 0 ,n − 1 1 → E 1 ,n − 1 1 is the differential in Vinog r adov’s C - sp ectral s e- quence, se e [17]. F or evolutionary equatio ns this differential coinc ide s with the Euler op er ator. The following result genera lises Theorem 2: Theorem 14 . I f ( A, R ) is a Poisson–Nijenhuis stru ctur e on E then R i ◦ A , i = 0 , 1 , 2 , . . . , is a family of p air-wise c omp atible Hamiltonian structu r es on E . 10 4. Concl uding remarks: nonlocal P oisson–Nijenhuis structures Strictly sp ea king all co nstructions exp os e d ab ove a re v a lid for lo c al Poisson– Nijenh uis structures. In rea lit y , the o per ators A or R o r b o th ar e n onlo c al , i.e ., contain terms like D − 1 x . F or example, reca ll the recur sion op er ator for the KdV equation. It seems that our approach can b e extended to structur es of this type. The gener al scheme is as follows. Let E b e a differential equation and τ : f LE → LE b e a general co vering over LE in the sense of [13]. Then s olutions of the equation ˜ ℓ E ( N ) = 0 linear with resp ect to o dd v a riables g ive rise to nonlo cal C -differe ntial recurs ion op erator s R N with nonlo calities corresp onding to no nlo cal v ariables defined b y τ . These solutions are shadows of symmetries in this cov ering. The hardest problem lies in definition of the J acobi bra ck et for suc h shadows. Nontriviality o f this prob- lem is illustrated b y observ ation g iven in [15]. A wa y to commute shadows can b e derived from the results of [9] but the constructions g iven there are amb iguo us. In [5] we describ ed a cano nic a l co nstruction for the J acobi bra ck et { . , . } of shad- ows of a general nature. Given this construction a nd taking into account the ab ov e exp osed results, we can define nonlo cal Nijenh uis op erator s R N as the ones satis - fying {N , N } = 0 . In a similar ma nner, we can consider cov erings over τ ∗ : g L ∗ E → L ∗ E and, s olving the equatio n ˜ ℓ E ( H ) = 0 , lo ok for nonloc a l Hamiltonian op er ators A H corres p o nding to shadows H . The Hamiltonianity condition is given by {H , H} = 0 . Finally , the compatibility condition for R N and A H are expr essed by {H , N } = 0 , where the brack et is considered in the Whitney pro duct o f τ and τ ∗ . A detailed theo r y of nonlo cal Poisson–Nijenhuis structures will b e given else- where. Remark 7. As it was demonstrated in [7] and [6], a very efficient wa y to co ns truct nonlo cal recursio n op er ators and Hamiltonian structures for evolution equations is the use o f nonlo cal vectors (for the ℓ ∗ -cov ering) and cov ectors (for the ℓ -cov ering). In par ticula r, this method, b y its nature, leads to the so-called we akly nonlo c al op er ators . When this pap er was almost finished, Maria Clara Nucci indicated to us the work [4] whe r e the c onstruction of nonlo ca l vectors was reinv ented. The author of [4] exploits the Lagrang ian structur e of the ℓ ∗ -cov ering, thoug h the reason for existence of nonlo ca l vectors is mo re general (it suffices to compare the cons truction with the one for nonlo cal c ovectors on the ℓ -covering). References [1] J.V. Beltr´ an and J. Mont erde, Poisson–Nijenhuis structur es and t he Vino gr adov br ack e t , Ann. Global A nal. Geom., 12 (1994), no. 1, 65–78. [2] A.V. Bo charo v, V.N. Chetv eriko v, S.V. Duzhin, et al. Sy mmetries and Conservation L aws for Differ ential Equations of Mathematic al Physics . Amer. Math. So c., Providence, RI, 1999. Edited and with a preface by I. Kr asil ′ shc hik and A. Vinogradov. [3] A. Cabras and A.M. Vinogrado v, Ext ensions of the Poisson br acket to differ ential forms and multi-ve ctor fie lds , J. Geom. and Phys., 9 (1992), 75–100. 11 [4] N.H. Ibragimov, A new co nservation the or em , J. Math. Anal. Appl. 333 (2007), 311–328. [5] A. V erb ov etsky , V. Golovk o, I. Kr asil ′ shc hik, L ie br acket for nonlo c al shadows , Scien tific Bulletin of MSTUCA, 91 (2007), 13–21 (in Russian). [6] Kersten P .H.M., Krasil ′ shc hik I.S., V erb ov etsky A.M., A ge ometric study of the disp ersionless Boussinesq t yp e e quation , Acta Appl. Math., 90 (2006) no. 1, 143–178. arXiv: nlin/0511012 [7] P . Kersten, I. Kr asil’shchik, A. V erb ov etsky , Hamiltonian op e ra tors and ℓ ∗ -c overings , J. Geom. and P hys. 50 (2004), 273–302 . arXiv:math.DG/0304245. [8] P . Kersten, I. Krasil ′ shc hik, A. V erb ov etsky , On the i nte g r ability c onditions for some struc- tur es r elate d to evolution differ ential equa tions , Acta Appl. Math., 83 (2004) no. 1-2, 167–173. arXiv:math.DG/0310451. [9] N. G. Khor ′ k ov a, Conservation laws and nonlo c al symmetries , W orks of Mosco w State T ech- nical Universit y (1988), no. 512, 105–119 (in Russi an). [10] Y. Kosmann-Sch wa rzbach, The Lie bialgebr oid of a Poisson–Nijenhuis manifold , Lett. Math. Ph ys. 38 (1996), no. 4, 421–428. [11] Y. Kosmann-Sch warzbac h, F. M agri, Poisso n–Nijenhuis st ruct ur es , Ann. Inst. H. P oincar´ e Ph ys. Th´ eor., 53 (1990), no. 1, 35–81. [12] Krasil ′ shc hik I.S. , Some new c ohomolo gic al invariants of nonline ar differential equa tions , Different ial Geometry and Its Appl. 2 (1992) no. 4, 307–350. [13] I.S. Krasil’shchik, A.M. Vi nogrado v, Nonlo c al tr ends in the ge ometry of differ ent ial e quations: symmetries, c onservation laws, and B¨ acklund tra nsformations , Acta Appl. Math. 1 5 (1989) no. 1-2, 161–209. [14] F. Magri, C. M orosi, A ge ometrica l char acterization of inte gra ble Hamiltonian sy stems thr ough the the ory of Poisson–Nijenhuis manifolds , Uni v ersity of Mi lan, Quaderno, S 19 (1984), 20 p. [15] P .J. Olv er, J. Sanders, and J.P . W ang, Ghost symmetrie s , J. Nonli near Math. Phys. 9 (200 2) Suppl. 1, 164–172. [16] I. V aisman, A le ctur e on Poisson–Nijenhuis structur es, inte gr able syst e ms and foliations, F euil let ages et syst` emes int´ egr ables , (Mon tp ellier, 1995) , Progr. Math., Bir kh¨ auser Boston, Boston, M A, 145 (1997), 169–185. [17] A.M. Vinogradov , Cohomolo gic al Analysis of Partial Differ ential Equations and Se c ondary Calculus , AM S T ranslation of Mathematical Monographs series, 204 , 247 pp., 2001. V alentina Golovko, Lomonosov MSU, F acul ty of Physics, Dep ar tment of Ma thema t- ics, Vorob’evy Hills, Moscow 11990 2 Russia. E-mail addr ess : golovko@mccme. ru Iosif Krasil ′ shchik, Independent University of Mosco w, B. Vlasevsky 11, 119002 Moscow, Russ ia E-mail addr ess : josephk@diffie ty.ac.ru Alexander Verbovetsky, Independent University of Moscow, B. Vla sevsky 1 1, 1190 02 Moscow, Russ ia E-mail addr ess : verbovet@mccme .ru 12

Original Paper

Loading high-quality paper...

Comments & Academic Discussion

Loading comments...

Leave a Comment