ON SYSTEMS OF DIFFERENTIAL EQUATIONS IN THE SPACE OF C -FUNCTIONS ON SMOOTH MANIFOLDS G.R.Belitskii* and I.L.Lerman** * Dept. of Mathematics, Ben-Gurion University of the Negev, Beer-Sheva, Israel ** Dept. of Mathematics, Ben-Gurion University of the Negev, Beer-Sheva, Israel Abstract. Problem of global solvability of the systems of linear differential equations with singularities is considered. In particular the conditions of surjectivity and of normal solvability in some special cases are obtained. 1. INTRODUCTION We consider linear differential operators of the form ( L )( x) (v )( x) A( x) ( x) (1) in the space of C -functions on a smooth manifold M . Here v is a C - vector C M , field on is a -complex-valued matrix, A : M End (C m ) m : M C is an unknown C - vector-function. The differential operator L acts on a topological space C -vector-functions, endowed with a standard topology of convergence with all its derivatives on each compact subset of M . The properties of the operator L from the point of view of a general theory of linear operators such as normal solvability, Fredholm or semi-Fredholm have been researched. Recall that an operator is called normally solvable if its image is closed. In this case the equation ( L )( x) ( x), where (x) is any complex-valued C - function on M , is also called normally solvable. A normally solvable operator and the corresponding equation are called Fredholm if both of subspaces Ker and Co ker are finite-dimensional. A normally solvable operator and the corresponding equation are called semiFredholm if min (dim Ker, dim Co ker) . The main subject of this work is research of normal solvability, Fredholm and semi-Fredholm of the operator L for vector fields with non-degenerated singular points. Our conjecture is that for a vector field with hyperbolic singularities on a manifold M the operator L is normally solvable. This was proven in [1] for one-dimensional manifold M . Let us recall the notion of glued fields, which was introduced in [2]. A vector field v on a manifold M is said to be glued from vector fields v on M if there exists an open covering M U such that v U are conjugate to v . The latter means that there exist diffeomorphisms : U M such that v U v . In these terms, the main aim of this work is a proving of normal solvability of the operator L for fields that are glued from hyperbolic linear fields on Rn . Further 7-36 the examples confirm this assertion will be considered. In particular, we will consider the vector fields on the sphere, the cylinder and the torus. For instance, a vector field on the cylinder T 1 R1 is presented in the form (2) v( x) (w(u), z ), x (u, z ) T 1 R1 , 0, 1 where w is a vector field on T with hyperbolic singular points. It is easy to see that this field is glued from linear hyperbolic fields. It will be proven in Theorem 10, that the operator L with the vector field (2) on the cylinder is normally solvable. Although the considered examples confirm our hypothesis, the question of normal solvability for such kind of fields still stays open in a general case. We start this work with research a one-dimensional version of the problem, by considering, in Section 2, the simplest case of a manifold, M R1 . Section 3 we begin with the definition of absorber and nozzle, which were introduced in [2], as tools of expanding of a local solution to a global one, in a form relevant to our research. In Section 4, the conditions of normal solvability of the operator L in the absence of invariant compact subsets for a vector field v is obtained. In Section 5, normal solvability of the operator with linear hyperbolic field on M R n is proved. Further, in Section 6, the case of the operator with two singular hyperbolic points on the sphere are researched. Section 7 concerns of normal solvability of the operator L with singular hyperbolic points on the cylinder and the torus. 2. One-dimensional case We start from the simplest case M R1 . Then a vector field v has the following form v( x) a( x) dxd with a real-valued function a(x) . If a(x) does not vanish then the operator L is surjective and dim KerL m . Hence, L is Fredholm. If a point x0 is singular for v , i.e. a( x0 ) 0 , then the problem of formal solvability arises. Indeed, let be a solution of the equation (3) a( x)´x A( x) ( x) ( x). Substitution of x x0 into equation (3) yields the condition ( x0 ) Im A( x0 ) . Furthermore, by differentiating equation (3) with respect to x and substituting x x0 , an infinite system of linear algebraic equations with respect to derivatives ( ( x0 ), ´( x0 ),..., ( k ) ( x0 ),... ) is obtained: x0 Ax0 x0 ´( x ) a´( x )´( x ) A´( x ) x Ax ´( x ) 0 0 0 0 0 0 0 (4) ... i x0 Pi x0 ,..., i x0 where i 0, 1,..., k ,.. and Pi x0 ,..., i x0 is a polynomial which is a linear combination of the derivatives { ( k ) ( x0 )} for k i . Equation (3) is called formally solvable at x0 if there exists series ˆ c k x x0 k , k! whose coefficients ck ( k ) ( x0 ) satisfy all equations of system (4). In this case the series is the formal Taylor series one of supposed solutions (x ) . According to the Leng theorem (see[1]), a sufficient and, evidently, necessary condition of an existence of a formal solution is compatibility of every finite k 0 7-37 subsystem of system (4). Let ˆ k x x0 k , k! be the formal Taylor series of the given function (x) and k 0 ˆ k x x 0 k , k! be the formal Taylor series of a supposed solution (x ) . Then system (4) may be written in a recursive form (5) k kI Ak Rk 0 ,..., k 1 , k 0,1,... where a´( x0 ) , A Ax0 and Rk 0 ,..., k 1 are linear functions that are determined by the given matrix A(x ) and the function (x) . Hence, if it holds that (6) k specAx0 , k 0,1,... then equation (3) has a solution. Condition (6) is one of versions of so-called the absence of resonances from the theory of vector fields. Note that if 0 a singular point x0 is called hyperbolic. In this case condition (6) holds for every large enough k. Thus, in the absence of resonances, equation (3) is formally solvable at x0 for an k 0 arbitrary C -function (x) . Conversely, condition (6) is necessary for equation (3) to be formally solvable at x0 for every C -function (x) . In other words, the following assertion is true. Proposition 1. Let x0 be a singular point of the vector field v . Then equation (3) is formally solvable at x0 for every vector-function (x) if and only if condition (6) holds. Proof. The part "only if" remains to be proven. Assume k N to be the smallest number such that k specAx0 . Consider a vector u Im Ax0 kI and set ( x) x x0 u . Show that (3) for the given (x) is not formally solvable at x0 . Indeed, the first k equations of system (4) have the form: 0 Ax0 x0 0 a´( x0 )´( x0 ) A´( x0 ) x0 Ax0 ´( x0 ) ... k -1 0 a´( x0 ) ( x0 ) Ax0 k -1 ( x0 ) ... uk! kI Ax0 k ( x0 ) Rk 0 x0 ,..., k 1 x0 k Since j specAx0 for j k 1 all equations have only trivial solutions x0 0 ,..., ( k 1) ( x0 ) 0 . Hence, the first non-trivial in the left side equation is uk! kI Ax0 k x0 . This equation has no solution since u ImkI Ax0 . However, an existence of a formal solution may not imply local solvability. Example 1. Consider the simplest equation a( x) ´( x) b( x), 7-38 (7) where exp 1 / x , x 0 a( x) 0, x 0 and 2 exp 1 / .2 x , x 0 b( x ) . 0, x 0 2 Equation (7) is formally solvable at x 0 and an arbitrary C - vector-function is its formal solution. But for x 0 it holds that ba(( xx)) . Hence the equation has no local smooth solutions in a neighborhood of x 0 . In this example the point x0 is a singular point of "an infinite type" of the vector field v( x) a( x) dxd . Note that the point x0 is a singular point of a finite type if the function a(x) has at the point a root of a finite order. This means that the function a(x) may be presented as a( x) ( x x0 ) r b( x) with r 0 and b( x0 ) 0 . Theorem 1. (Kyznezhov,[4])Let x0 be a singular point of a finite type for the vector field v (x ) . If equation (3) has a formal solution ̂ then there exists its C - local solution whose formal Taylor series at x0 coincides with ̂ . It is clear that if equation (3) has only a singular point then local solvability implies a global one. But this does not occur if the equation has more than one singularity. Example 2. Consider the equation: x( x 1) ´( x) t ( x) ( x) (8) where (x) is a given C - smooth complex-valued function on R1 and t const . The equation has two hyperbolic singular points x1 0 and x2 1 . The corresponding open intervals are U1 (, x2 ) and U 2 ( x1 , ) . In this case the condition of the absence of resonances is fulfilled at x1 iff t Z and at x2 iff t 2Z . Choosing t {Z 2Z } it is obtained that the equation is formally solvable at the points x 0 and x 1 and besides Ker( L U i ) {0}, i 1, 2 . A solution of equation (8) may be written in the form: x Cx z( x), x 0,1, where t x z ( x) , x 0,1 1 x satisfies the corresponding homogeneous equation. For the function C (x) is obtained: ( x) C´( x) , x 0,1. z ( x) x( x 1) Let the function (x) have the form: ( x) z ( x) x( x 1) ( x), (0) (1) 0, where (x ) is flat at the points x 0 and x 1 . Show that a global solution exists if and only if: 7-39 1 (s)ds 0. (9) 0 On each of the intervals U i for the given function (x) there exists a local solution i . Indeed, since the function (x ) is flat at x 0 , the function x 1 ( x) z1 ( x) (s)ds, 1 (0) 0, 0 where z1 xz x , x U1 is a C -local solution on U1 , while the function x 2 ( x) z2 ( x) (s)ds, 2 (1) 0 1 with z2 ( x) z( x), x U 2 is a C -local solution on U 2 , since (x ) is flat at x 1. Assume that there exists a global solution of equation (7). Then in the absence of resonances on each of intervals Ui it holds that i KerL U i {0} . Consequently the local solutions 1 and 2 have to coincide on the intersection U1 U 2 (0,1) , if a global solution does exist: 1 ( x) 2 ( x) , x (0,1), implying x (s)ds 0. 0 The latter is a necessary and sufficient condition for global solvability. Hence, if is positive on (0,1) then (8) has no C -solution. For example, if , 0 1 0, 2 2 x 1 x then C -solutions are absent. Assume that the vector field v has an arbitrary number of isolated singular points under the condition of local solvability of the equation at each one of them. Denote {xi } the collection of all singular points and consider open intervals {U i ( xi 1 , xi 1 )} . The criterion of global solvability of the equation is presented by the following theorem. Theorem 2. Assume that equation (3) is locally solvable in each of open intervals U i . If the equation has a global solution then for every collection of local solutions ( x) exp 1 { i } C (U i ) there exist vector-functions i KerL U i such that i1 ( x) i ( x) i1 ( x) i ( x), x U i U i 1. (10) Conversely, if condition (10) is fulfilled for a collection of local solutions { i } then the equation has a global solution. Proof. Assume that equation (3) has a global C -solution . Let { i } be a set of C - local solutions on U i . Then the differences i i homogeneous equation 7-40 satisfy the Thus the functions i L i x 0 , x U i . satisfy condition (10). Conversely, suppose that condition (10) is fulfilled for some of C -local solutions { i } and some of functions i KerL U i . Consider the vector function (x ) which is defined by x i x i x , x U i . The function is well defined since the differences i i and i 1 i1 coincide on the intersection U i U i 1 . The function (x ) on every interval U i satisfies the equation: L x L i x L i x x . Consequently (x ) is a global solution of equation (3). Corollary 1. Let all singularities of the vector field v be of finite orders. Then the operator L is normally solvable. Proof. Let {xi } be the collection of all singular points of the operator L . Suppose that all the singularities of the operator are of finite types. Then these singular points are isolated. Consider intervals {U i ( xi 1 , xi 1 )} forming an open covering of R 1 . To prove the statement of Corollary 1 it is sufficient to prove normal solvability of the operator on each of intervals U i and finite-dimensionality of Ker ( L U i ) (see Corollary 3.2 in [8]). For an interval U i consider a sequence { n ( x)} Im( L U i ) which converges to a vector function ( x) C (U i ) . For the limit function (x) it needs to be proven that ( x) Im( L U i ) . Since { n ( x)} Im( L U i ) equation (3) is formally solvable at xi for each of functions n (x) . The formal solvability of equation (3) for the function n (x) means that every finite subsystem of system (4) has a solution. It follows from ni xi i xi , that it holds also for the limit function (x) that every finite subsystem of system (4) has a solution. According to the mentioned above the Leng Theorem, there exists a general solution which satisfies all the equations of the system. Thus equation (3) at xi with the limit function (x) has a formal solution. Further, by Theorem 1 for (x) there exists a local solution which may be extended dynamically on the whole interval U i . Hence, U i Im L U i and the restrictions L U i are normally solvable. It remains to ensure that dim Ker ( L U i ) for every i . In every open interval ( xi , xi 1 ) Since the operator has no singular points. Therefore dim Ker( L U i U i 1 ) n . Ker( L U i ) Ker( L U i U i 1 ) it holds dim Ker( L U i ) dim Ker( L U i U i 1 ) . Consequently dim Ker ( L U i ) n . 7-41 that Hence the operator L is normally solvable. In fact, the situation of Example 2 is general in the following sense. Theorem 3. For every collection of vector-functions {ci } KerL U i U i 1 there exist a vector-function (x) and local solutions { i } C (U i ) of equation (3) such that ci i 1 i . Proof. Consider vector-functions {ci } KerL U i U i 1 . There exist vector- functions i C that are defined on open intervals U i and satisfy the condition: ci i 1 i . The vector-function x L i x , x U i is well defined since L i 1 x L i x , x U i U i 1 . It is evident that i are local solutions of equation (3). Summarizing the previous facts we obtained the following criterion of surjectivity of the operator L . Corollary 2. Let all singularities {xi } of the vector field v be of finite types. Then the operator L is surjective if and only if the following conditions are fulfilled: (a) i k specAxi , k 0,1,... where i a( xi ) ; (b) each collection of vector-functions {ci } KerL U i U i 1 permits of a representation ci i 1 i where i KerL U i . If v has a unique singular point then the condition of surjectivity has been reduced to (a). But if the vector field has even two singular points condition (b) may not be fulfilled (see Example 2). The criterion of surjectivity of L with two singular points has the following form: Corollary 3. Let the vector field v have two singular points x1 and x2 of finite types. Then the operator L is surjective if and only if condition (a) is fulfilled and it holds that (11) dim Ker( L U1 ) dim Ker( L U 2 ) n. Indeed, if condition (11) holds then each of vector-functions c KerL U1 U 2 may be written in the form: c 1 2 , where i KerL U i , i 1, 2 . Thus condition (b) of Corollary 2 is also fulfilled. Conversely, if every vector function c KerL U1 U 2 may be represented in the mentioned form, then KerL U1 U 2 Ker( L U1 ) dim Ker( L U 2 ), implying (11). For instance, in Example 2 condition (a) holds at both of the singular points x1 0 and x2 1 . However, condition (b) is not satisfied. Indeed, in this case Ker( L U i ) {0}, i 1, 2 while dim KerL U1 U 2 1 . Example 3. Consider the equation: x 3 (1 x) 3 ´( x) 2t (2 x 1) ( x) ( x). 7-42 (12) The equation has two non-hyperbolic singular points x1 0 and x2 1 . Therefore, if it is assumed that t 0 then condition (a) is satisfied at both of the singular points. Furthermore, a general solution of the corresponding homogeneous equation has the form: t , x 0,1. x C exp 2 2 x 1 x Therefore if then while dim Ker( L U i ) 1, i 1, 2 Re t 0 dim KerL U1 U 2 1 . Hence, condition (b) is also satisfied, implying that for Re t 0 the equation has a global solution for every C -function (x) . For Re t 0 condition (b) fails and operator (12) is not surjective. Of course, all previous considerations may be immediately carried out on the unit circle T . In order to formulate conditions of Fredholm for the operator L consider system (4) at a singular point xi of the vector field v . This is an infinite linear system with respect to an infinite vector w ( w0 ,..., wi ,...) . The system may be rewritten in the form Ti w where Ti is the corresponding linear operator in the space of infinite sequences w . Denote d i dim Co ker Ti in an algebraic sense. For example, if condition (6) holds then d i 0 . Denote by S the space of all collections c {ci } such that ci KerL U i U i1 . Let representation ci i 1 i , S0 S consists of elements of S permits of the with i Ker( L U i ) . It is obvious that S 0 is subspace. Denote b co dim S 0 and s dim KerL . Note that s n . Theorem 4. Let all the singularities of the vector field v be of finite types and such that i d i , b are satisfied. Then the operator L is Fredholm of indL b d i s. i In particular, if all singularities are hyperbolic (a´( xi ) 0) , then d i , and we obtain Corollary 4. Let the vector field v on R1 have a finite number of hyperbolic singularities. Then equation (3) is Fredholm. Since the number of hyperbolic singularities on the unit circle T is finite, the operator L on T is Fredholm as well. 3. Multi-dimensional case. Absorber and Nozzle It is clear that a multi-dimension case of M is more complicated. We start this part of our work with some additional relevant definitions. Let F t (x) be a flow of the vector field v via a point x . Denote by (t ( x), t ( x)) a maximal open interval of an existence of the flow. For a compact manifold M such interval for all x coincides with the real line. Recall that a subset S is called v -invariant if t and for all x S and t 0 it holds that 7-43 F t ( x) S . Let W be a subset of the manifold M and consider the closed interval t (W ), t W : t x , t x . xW In particular, it is possible that the interval is reduced to zero. But if it has inner points, then the corresponding open interval (t (W ), t (W )) is a maximal interval of an existence of the trajectory F t (x) for all x W . Definition 1. A v -invariant subset M M is called an absorber if for every x M there exists a neighborhood W and a number t 0 0 such that t t 0 , F t u M , u W . Subset M M is called a nozzle if it becomes an absorber after the substitution t t . In other words there exists t 0 0 such that t t 0 , F t u M , u W . Example 4. Let us consider an example of a constant vector field xe n where e is a vector in R . Its flow is F t ( x) x te . Let f (x) be a linear functional such that f (e) 0 . Then the subset { A {x : f ( x) 1}} is an absorber and the subset {N {x : f ( x) 1}} is a nozzle. Example 5. Consider the vector field x x m , m 0 on the real line. If m is odd then the point is a node. Then every closed neighborhood of this point is a nozzle and R1 is an absorber. If m is even then the point is a point of overflow. In this case, every interval , is an absorber and , is a nozzle. Theorem 5. Let M , M be an absorber and a nozzle for a vector field v , respectively. If the equation ( L )( x) ( x) (13) has a solution in a neighborhood of the intersection M M then it has a global solution. In particular, if M M is empty set, then the equation is solvable for every vector-function (x) . Proof. Assume that 0 : M C p is a local C -solution of equation (13) in a neighborhood U of the intersection M M , i.e. 0 is defined on M and satisfies the equation L 0 x x , x U . A supposed global solution of (13) will be searched in the form 0 where the vector-function is a solution of the equation (14) v A ~ ~ Here L 0 and does vanish in U . It remains to prove that equation (14) has a solution for every C -vector-function ~x that equals identically to zero in a neighborhood of the intersection 7-44 M M . For this aim will be used Lemma 1 [see below]. The lemma claims the existence of a decomposition of the vector-function ~x in the form ~ . Here M 0 and 0 . Since equation (13) is linear, it follows that suffice M it to find solutions of the equations (15) L x x . The vector-functions are obtained in the same way. Therefore we will describe in detail only a solution of the equation for the function . Consider the following system of differential equations: x v x . (16) y A x y x Let ( F t ( x), G t ( x, y)) be the corresponding flow. The projection G t ( x, y ) may be presented in the form: t G t x, y t , x y 1 s, x F s x ds , (17) 0 where (t , x) is a fundamental matrix of the corresponding linear homogeneous system. It is easy to check that if a function is a solution of the equation (18) F t x G t x, x , t R1 , then satisfies the equation (15). Since M 0 it is true that for every x there exists a neighborhood W such that for s t 0 it holds F s u 0 for all u W . Hence, the integral x 1 s, x F s x ds (19) 0 converges and presents a C -mapping on M . Immediate verification shows that (x) is a solution of equation (18). Indeed, F x 1 s, x F s t x ds t 0 1 u t , F t x F u x du t t 1 u t , x u , x F x du 1 u, x F u x du G t x, x , 0 0 Since t , x 1 u, x 1 u t , F t x The latter follows from the following flow property G t s x, y G t F t x, G s x, y . Setting in (18) 0 , we arrive at the following equality u, x u t , F t x t , x 7-45 (20) that is equivalent to (20). Thus, a solution of equation (15) is obtained. Similarly, the vector-function is obtained in the form x 1 s, x F s x ds . (21) 0 A vector-function is a C -solution of equation (14). Lemma 1. Let us consider two closed subsets S1 and S 2 of a manifold M . If a C -function equals identically to zero in a neighborhood of the intersection S1 S 2 then there exists the decomposition , where S1 0 and S2 0 . Proof. Consider the complements U1 M \ S1 and U 2 M \ S 2 . The intersection S1 S 2 is closed while the subsets U1 ,U 2 are open on M . It is obvious that the intersection U1 U 2 M \ (S1 S 2 ) is an open subset as well. The submanifold U1 U 2 is a para-compact set. Therefore there exists a decomposition of the unity as a sum of C -functions 1 ( x) and 2 ( x) . The functions 1 ( x) and 2 ( x) must satisfy the following conditions: they are determined on U1 U 2 , supp i U i and 1 ( x) 2 ( x) 1 . Let us set x x , x U 1 x 1 0 , x U1 and x . It is easy to see that the functions and are well defined everywhere on the manifold M and it holds that S1 0 , S2 0 and 0 . Let us return to Examples 4 and 5. In Example 4 A N is empty set. Hence, according to Theorem 1 the operator L is surjective. In Example 5, the intersection of an absorber and a nozzle contains a neighborhood of the origin. As a consequence, local solvability implies a global one. 4. Equations in the absence of invariant compact subsets The typical example of the vector field without invariant compact subsets in R n is a constant vector field v ( x ) e , where e is some vector. It is well known that on the real line in the absence of invariant compact subsets a vector field is conjugate with a constant one and the corresponding operator has the form (22) ( L )( x) ´x A( x) ( x). The operator L on R1 is surjective and dim KerL n . In contrast to the real line, in the plane, even in the absence of invariant compact subsets, the vector field may be not conjugate with a constant field. As a consequence, the operator may be not surjective and even not normally solvable. The following example demonstrates this in the case A( x) 0 . 7-46 Example 6. Let us consider the manifold M R 2 \ , 0 that is evidently diffeomorphic to the plane. Consider the vector field v , x , M . It is obvious that the flow F t x e t x where 1 0 0 1 has no invariant compact subsets. However, for instance, the Abelian equation v x 1 has no continuous solutions. Indeed, if is a solution then et , ( , et ) t. If , 0 , then we arrive at a contradiction as t . Thus the operator L v is not surjective. Moreover, the operator is not normally solvable. In order to prove that it suffices to check that Im L is dense in the space of C - functions on M . That means that for every C -function (x) there exists a sequence { n ( x)} Im L that converges to (x) . To show that consider a sequence of functions { n ( x)} such that each of them equals to zero inside of a circle of the radius 1 / n and equals to 1 outside of the circle of the radius 2 / n . Easy to show that the sequence { n ( x)} converges to the unity in C -topology on M . Hence the sequence { n } { n ( x) ( x)} converges to the function (x) . It remains to prove that n ( x) Im L . For any n 0 , let us consider the subsets 1 M n x ( , ) : 2n and 1 M n x ( , ) : . 2n The subset M (n) is an absorber for the vector field v , and M (n) is a nozzle. Their intersection is contained in the neighborhood of the radius 2 / n of the point x0 0 , where n ( x) 0 . Therefore, according to Theorem 1 the equation ( L )( x) n ( x), has a solution. Thus n ( x) Im L . The same arguments show that Im L is dense in the space of C - functions on M for every matrix A(x ) . In the considered example the field was not conjugate to a constant one because of an existence of so-called non-wandering compact subsets. Definition 2. A subset K M is called wandering for a vector field v if t (K ) and t (K ) and there exists a number t 0 0 such that t t 0 , F t K K is empty set . This means that every trajectory F t (x) leaves the subset K in a finite period of time. If K is not wandering, then it is called non-wandering subset. 7-47 Theorem 6. If all compact subsets K M are wandering for a vector field v , then the operator L is surjective. Proof. Note that if all compact subsets are wandering then the field is full, i.e. its flow F t (x) is defined for every t R and x M . It was proven in [3] that in this case the flow is conjugative with a flow of shifts. This means that there exists a manifold B such that dim B dim M 1 . There also exists a diffeomorphism : M R1 B such that F t ( s, z ) ( s t , z ) where ( s, z ) R1 B . Hence the operator L , in an obvious sense, is equivalent to the operator ~ ~ L s, z As, z s, z . s that is surjective. 5. Operator with only a hyperbolic singular point in R n The main result of this section is the following statement. Theorem 7. Let v (x ) be a linear hyperbolic field in R n . Then equation (3) has a global solution iff it is formally solvable. First, let us discuss the notion of formal solvability for the multi-dimensional case. Let x0 be a singular point and : R n C m -a supposed solution of the equation (v )( x) A( x) ( x) ( x), x M (23) Substituting x x0 in (23) we get the condition ( x0 ) Im A( x0 ) . Furthermore, differentiating equation (23) with respect to x and substituting x x0 , the infinite system of linear algebraic equations with respect to derivatives ( ( x0 ), 1 ( x0 ),..., k ( x0 ) ( k ) ( x0 ),... ) is obtained: (v )( x0 ) A( x0 ) ( x0 ) ( x0 ) (v1 )( x0 ) A( x0 )1 ( x0 ) 1 ( x0 ) R1[ ( x0 )] (24) ... (v k )( x0 ) A( x0 ) k ( x0 ) k ( x0 ) Rk ( x0 ), 1 ( x0 ),..., k 1 ( x0 ) where k ( x0 ) ( k ) ( x0 ) and Ri x0 ,...,i x0 are linear combinations of the derivatives { x0 ,..., k 1 ( x0 )} . As in one-dimensional case equation (23) is called formally solvable at x0 if there exists formal series I x x0 ˆ C I , I! I whose coefficients CI C m satisfy all equations of system (24). Here I I1 ,..., I n I! I1!... I n ! is a integer-valued multi-index, and x x0 I I1 In x1 x10 ... xn xn0 . In this case the series is the formal Taylor series of a supposed solution (x ) . By the Leng Theorem [5] a necessary and sufficient condition for an existence of a formal solution is compatibility of every finite subsystem of system (24). By 1 ,..., n denote the eigenvalues of v( x0 ) and 1 ,..., m the eigenvalues of A( x0 ) . It is well known that the numbers 7-48 n j I i i i 1 are the obstacles to solvability of the system. The condition n j I i i (25) i 1 guarantees formal solvability at x0 for every (x) . Condition (25) is one of versions of a so-called the absence of resonances the vector fields theory. [Proof of Theorem 7] . Remind that if equation (31) with hyperbolic point x0 is formally solvable at x0 then it has a local C solution 0 [6]. Without loss of generality we assume that x0 0 . By S denote the invariant subspaces that corresponding to spec with Re 0 and Re 0 , respectively. Then R n S S , and for all x R n it holds that x x x where x S . The subset M x : x , 0 is an absorber for the vector field v , while the subset M x : x , 0 is a nozzle. Let a local solution 0 be determined in a neighborhood U of x0 . For small enough it holds that M M U . Hence, according to Theorem 5, the equation has a global solution. Corollary 6. Let v be a linear hyperbolic field in R n . Then the operator L is normally solvable. Proof. Consider a sequence { n ( x) Im L} that converges to a vector-function ( x) C . For the limit function (x) we have to prove that ( x) Im L . Since n ( x) Im L , it follows that equation (23) is formally solvable at x 0 for each of functions n (x) . Formal solvability of equation (23) for a function n (x) means that each of finite subsystems of system (24) has a solution. From ni x0 i x0 Follows that every finite subsystem of system (24) has a solution for the limit function as well. According to above mentioned Leng`s Theorem, there exists a general solution which satisfies all equations of the system. Thus equation (23) at x0 with the limit function (x) has a formal solution. Then, by Theorem 7, for (x) there exists a global solution. Hence, Im L and the operator L is normally solvable. If a singular point of the operator is a hyperbolic saddle, then dim KerL [7]. If the number of resonances is finite then dim CoKerL and the operator is semiFredholm. But if the singular point is a node, then both of spaces KerL and CoKerL are finite-dimensional implying Corollary 7. If a linear hyperbolic field in R n is a node, then the operator L is Fredholm. 7-49 6. Operator on the sphere with two nodes Consider a vector field on the n-dimensional sphere S n , having two fixed points. Theorem 8. The operator L x v xA x x on the sphere with the vector field that is glued from two linear hyperbolic nodes is normally solvable. Proof. Denote by L U i : C (U , C p ) C (U , C p ) the operator L that acting on any open subset U . Furthermore, we assume that the field v is glued from linear nodes at points- x1 , x2 . Let us show that the subspace Im L is closed. Indeed, the field v has the invariant open covering U1 Sn \ x1 , U 2 Sn \ x2 . The operator L U i , i 1, 2 is normally solvable and it holds that Ker( L U i ) are finitedimensional. Then, according to Corollary 3.2 [8] the operator L is normally solvable. Note that in contrast to the one-dimensional case [1], for n 2 the operator L on the sphere is semi-Fredholm, even if the vector field is glued from two linear hyperbolic nodes. Indeed, in spite of the fact that Ker( L U i ) are finite-dimensional, dim KerL U1 U 2 . In order to show that, recall that all compact subsets of the vector field v on the intersection U1 U 2 are wandering. Therefore the vector field v on U1 U 2 is conjugate to a shift on R1 Sn1 . It means that there exists a diffeomorphism T : Sn \ x1 , x2 R1 Sn1 such that (Tv)( s, z) ( / (s), 0) , where s R 1 and z Sn1 . Hence the homogeneous equation (23) is equivalent to the equation s, z ~ As, z s, z . s The general solution of this equation has the for ( s , z ) ( s , z )c ( z ) where ( s, z ) is the fundamental matrix of the system that depends on variable z as a parameter while c : Sn1 C p is an arbitrary C vector-function. Therefore dim CoKerL , implying dim CoKerL . Now we state a criterion of global solvability. Theorem 9. Assume that equation (23) has a C -solution on each of open subsets U1 ,U 2 , where U i Sn \ xi , i 1, 2 . a) If the equation has a global solution then for every collection of "local" solutions i C (U i ), i 1, 2 there exist vector-functions i KerL U i such that 2 ( x) 1 ( x) 2 ( x) 1 ( x), x U1 U 2 . (26) Conversely, if condition (26) is fulfilled for a collection of local solutions { i } , then the equation has a global solution. b) For every vector function c KerL U1 U 2 there exist a vector-function (x) 7-50 and local solutions { i } C (U i ) of equation (23) such that c 2 1 . Proof. Assume that equation (23) has a global C -solution . Let 1 and 2 be C - local solutions on U1 and U 2 , respectively. Then the differences i i satisfy the homogeneous equation L i x 0 , x U i . Thus the functions i , i 1, 2 satisfy condition (26). Conversely, suppose that condition (26) is fulfilled for some of C -local solutions { i } and some of functions i KerL U i . Consider a vector-function (x ) that is defined by x i x i x , x U i , i 1,2. Since the differences 1 1 and 2 2 coincide on the intersection U1 U 2 , the function is well defined. The function (x ) on every set U i satisfies the equation: L x L i x L i x x . Consequently (x ) is a global solution of equation (23). To prove the second part of the theorem, consider vector-function c KerL U1 U 2 . There exist vector functions i C , i 1, 2 which are defined on the open subsets U i and satisfy the condition: c 2 1 (see [9]). Define a vector-function (x) by the formula: x L i x , x U i , i 1,2. Since L 2 x L1 x , x U1 U 2 . the function (x) is well defined. Thus, this is evident that i are local solutions of equation (23). 7. Examples on the Cylinder and the Torus Recall that we consider a vector field v(u, z ) ( w(u ), z ) where u T 1 and z R1 , on the standard two-dimensional cylinder M T 1 R1 . Here w is a vector field on the circle with "non-degenerated singularities", i.e. has the form T1 1 d w(u ) a(u ) du , u T where a(u ) is a C -function with roots of finite order. For such fields on T 1 normal solvability was proven in [1]. Moreover, conditions of surjectivity and of Fredholm were obtained. Let, as above, A : M End (C m ) be a C -complex-valued matrix. Theorem 10. If v is glued from vector fields of the form (2) on M , then the operator ( L )( x) (v )( x) A( x) ( x), x M is semi-Fredholm with dim KerL . Before the proving of the theorem, let us state conditions of global solvability of the equation L on M T1 R1 . At first, note that if 0 , then T 1 is an absorber for the vector field v , while for 0 it is a nozzle. Therefore, to prove solvability of the equation, suffice it to prove its solvability in a neighborhood of the 7-51 circle T 1 . Let us assume that 0 . By differentiating the equation (v )(u.z ) A(u, z ) (u, z ) (u, z ) (27) with respect to z and substituting z 0 the infinite system of linear equations with respect to derivatives ( (u ), 1 (u ),..., k u z 0 (zuk , z ) k z 0 ,...) , which is similar to the system (24), is obtained: (v )(u ) A(u ) (u ) (u ) (v1 )(u ) A(u )1 (u ) 1 (u ) R1[ (u )] ... (v k )(u ) A(u ) k (u ) k (u ) Rk (u ), 1 (u ),..., k 1 (u ) where k u z 0 (zuk , z ) k (28) . z 0 Obviously, if Im L , then the system has C -solutions k (u) k z k , k 0,1,... z 0 Note that from the Fredholm property of the operator L on T 1 follows that for every finite subsystem, the space of solutions is finite-dimensional. Hence the system is solvable iff every its finite subsystem is solvable. Lemma 2. For any C - complex-valued function on the manifold 1 1 M T R with a vector field of the form (2), the equation L has a global C -solution if and only if system (28) is solvable. Proof. The part "if" remains to be proven. Let us assume that a set of the functions { k (u ), u T 1} forms solutions of system (28). Using the Borel Lemma, may be constructed a function ~ on the manifold M such that k ~ k u k z 0 , k 0,1,... z and consequently it holds that L~ , where is a flat function on T 1 . A supposed global solution of the equation L will be searched in the form ~ where is a solution of the equation (29) L ( x) x that is flat on the circle. Repeating the above considered arguments regarding to equations (15)-(18) we can represent a solution of (29) in the form x 1 s, x F s x ds , (30) 0 where is the fundamental matrix and F t is the flow of v . It remains to check that this formula presents a C -vector-function on M . Indeed, by fixing a compact neighborhood U {(u, z), z a} M of the circle 7-52 T 1 , an estimate 1 ( x, s) Ce Ns , x U , s R1 is valid. Since the function is flat on the circle, it holds that p k F1 s u , es z C kp esp z , z a , k , p 0,1,2... (31) Therefore, for under the integral sign expression takes place the estimate p 1 x, s F s x C0 p Ce s p N z . (32) We obtain convergence of the integral by choosing p N / . Thus it holds that C 0 . In the same may be obtained an estimate for under the integral expression sign of the derivative of expression (30) 1 x, s F s x k C~e sp N z p , x (u, z). Convergence of the corresponding integrals implies that C . Proof of Theorem 10. Consider a sequence of vector-functions { n ( x)} such that n Im L and that converges to a vector-function C . For the limit function we have to prove that Im L as well. In order to prove this we will show that system (28) is solvable for the limit vector function . Each of the equations of the system is a linear equation on the circle with a vector field whose singularities are hyperbolic of finite order. The Fredholm property of equations of this type was proven in [1]. Since n Im L and n k k , it holds also for the limit function that for every finite index k system (28) has a solution. Therefore, by the Lemma 2, Im L . It remains to ensure that the space of solutions of the corresponding homogeneous equation (33) v x Ax x , x M is finite-dimensional. Indeed, every C -vector-function may be written in the form p0 1 f k (u ) z k (u, z ), k ! k 0 ( x) with an arbitrary index p0 , where o( z p0 ) and f k (u ) (34) k ( u , z ) z k z 0 . Furthermore, by applying the operator L to (34) we arrive to the equality L ( x) L f k (u) z k L (u, z). p0 For KerL it is obtained that (35) k 0 L ( x) L 1 (36) . z 0 k 0 k! Equation (36) for every fixed choice of f k has a unique solution o( z p0 ) . Indeed, for every two solutions 1 , 2 , 1 2 it holds that L( 2 1 ) 0 . It follows from (29)-(32) that the function satisfies the equality x 1 (t , x) F t x . p0 7-53 f k (u ) z k and the estimate u, z C (u, z )C p e t N p z , 0 0 p0 (37) where index p0 satisfies the inequality N p0 0 , and consequently, as t yields 0 . Furthermore, backing to (32), we see that a definite choosing f k (u) yields a unique representation for the vector-function . The functions f k satisfy system (28) for k p0 and 0 . The system is homogeneous system of linear equations on the circle, with the vector field whose singularities are of hyperbolic type. It was proven that the space of solutions of such system has a finite dimension [1]. For the given index p0 , the space of solutions of the restricted to k p0 system (28) coincides with the dimension of the KerL . Hence dim KerL . Under conditions of Theorem 10, the operator L may be not Fredholm because a dimension of CoKerL may be is infinite. Example 7. Let a vector field on the cylinder T 1 R1 have the form v( x) ( w(u ), z ), x (u, z ) T 1 R1 , (38) z where w(u ) sin u u . Consider the equation: (v )( x) ( x) ( x) (39) where is an irrational constant. System (28) for the operator L in this case has the form (40) (w k )(u) (k ) k (u) k (u), k 0,1,2,... with k (u ) (zuk , z ) k z 0 . Here the field w is glued from two hyperbolic fields. The corresponding covering of is U1 U 2 where U1 T 1 \ {0} and T1 1 U 2 T \ { } . Since is irrational, i.e. in the absence of resonances, it holds that dim Ker ( L U i ) 0 , i 1,2,... . A solution of every equation of (40) may be written as: u Ck u sk (u), u 0, , where k u s k (u ) tan , u 0, 2 satisfies the corresponding homogeneous equation. For the function C k (u ) is obtained: k (u ) C k ´(u ) , u 0, . s k (u ) sin u Assuming k (u ) to be flat at 0 and , we arrive to the following criterion of solvability of equation L : 0 k (u ) k (u ) (41) 0 sk (u ) sin u du 0, sk (u ) sin u du 0, k 0,1, 2,.. From an infinity of a number of conditions follows that dim CoKerL . 7-54 References [1] G. Belitskii, I.Lerman, Linear systems of differential equations in the space of infinitely differentiable functions on the real line, J. Functional Differential Equations, Israel, 14, N2-3-4(2007), pp 147--159. [2] G. Belitskii, Functional invariants of diffeomorphisms of smooth manifolds, Dokl.Akad.Nauk Ukrainy, 11(1985), pp 6--8. [3] G. Belitskii, Yu. Lyubich, The locally translation structure of flows on the wanderingpoints set. Ergod. Th. & Dynam.Sys..20, (2000), pp 633--639. [4] A.N. Kyznezhov, Differentiable solutions to degenerate systems of ordinary equations. Functional Analysis and Its Applications.. Springer, New York, 6,N2, (1972),pp 119--127. [5] S. Leng, Hilbert's nullestensatz in infinite-dimensional space, Proc. Amer. Math. Soc., 3(1952), 291--293. [6] G. Belitskii, Functional equations, and local conjugacy of mappings of class C , Math USSR-Sb, 20(1973), pp 587--602. [7] L.P.Kuchko, On linear functional equations [Russian], Vestnik Harkovsk. Uni.,Ser.Meh.-Mat., 39(1974), pp 3--14. [8] G. Belitskii, N.Bykov, Spaces of cohomologies associated with linear functional equations. Ergod. Th. & Dynam.Sys..18, (1995), pp 343--356. [9] R.Godement, Topologie Algebrique et des Faisceaux.Hermann, Paris(1958). [10] P. Hartman, Ordinary Differential Equations, New York-LondonSydney,(1964). 7-55