BULLETIN OF THE TRANSILVANIA UNIVERSITY OF BRAŞOV ON THE SOLVABILITY OF NAVIER-STOKES EQUATIONS CRISTINA SBURLAN Abstract: In this paper we study the solvability of non-stationary Navier-Stokes equations for the flow of an incompressible viscous fluid, using the abstract Fourier Method. Both linear and nonlinear cases are treated. By this method, we obtain the existence and the uniqueness of the strong solution of the problem, which is the classical solution of the abstract problem. We construct effectively the solution as a Fourier series, using the sequence of orthonormal eigenfunctions of the duality mapping between a Hilbert space and its topological dual, which, in our case, is Stokes operator. Keywords: Fourier Method, Navier-Stokes Equations 1. Fourier Method for Stokes System Let R N , 2 ≤ N ≤ 3, be a bounded domain with enough smooth boundary to apply Green’s formula and Sobolev-Kondrashov embedding theorem, and denote Q (0, ) , (0, ) . Consider the Navier-Stokes system for the flow of an incompressible fluid ( u )( x, t ) 0 (the incompressibility condition) ~ ut ( x, t ) (u )u( x, t ) u( x, t ) p( x, t ) f ( x, t ), ( x, t ) Q u 0 on . ~ Denoting f f p, equation (1) becomes ut ( x, t ) (u )u ( x, t ) u ( x, t ) f ( x, t ) 1 where is the dynamic viscosity ( is the inverse of the Reynolds number). Re (1) (2) The velocity u = (u1, ..., uN) and the pressure p are not known, and they must be determinated from the system (1). We will study the case of dynamical viscous fluids with big . In this case we have u ( x, t ) (u )u ( x, t ) , and we can approximate equation (2) with ut ( x, t ) u( x, t ) f ( x, t ) (the Stokes system) (3) __________ Denote C0, () { y (C0 ()) N | y 0} and H C0, () L2 . We have (see [5], p. 83) that H { y ( L2 ()) N | y 0, n y 0 on } . We call this Hilbert space the space of incompressible fluids and let V { y ( H 01 ()) N | y 0} be Faculty of Mathematics and Informatics, Ovidius University of Constantza, Romania On the Solvability of Navier-Stokes Equations 322 subspace of H. Let P : ( L2 ()) N H be the Helmholz projection. From [5], we have ~ ~ ~ that P( f ) P( f p) P( f ) P(p) P( f ) , because P (p ) 0 . Denote by A L(V , V ) the Stokes operator N ( Ay, w) yi ( x) wi ( x)dx, y, w V . i 1 Then we can formulate problem (3) as an evolution equation ~ dy Ay P( f ) dt (4) where A is symmetric ( (Ay,w) (y,Aw) ) and strongly monotone ( ( Ay, w) y ), and 2 V is the energetic space of this operator. Suppose that we have also the initial u (0) u 0 H , which becomes y(0) y 0 , where y 0 P(u 0 ) . We have the following theorem (see [7], p. 67) Theorem 1.1 Let X be a real Hilbert space and L : D( L) X X symmetric and strongly monotone operator, with the domain D (L ) dimensional. Consider the energetic space E of the operator L A : D( A) X X be the Friedrichs extension of the operator L, Au := Ju, u D ( A) , D( A) {u E | Ju X } (where J : E E * , condition a linear, infinitely and let Ju, v (u, v) E , u, v E is the duality mapping from E to E * ). If the embedding E X is compact, then there exist the sequences (en ) nN E and (n ) nN (0, ) , which are the eigenfunctions and the eigenvalues of A, i.e., (Aen, v) = λn(en, v), v X , n N (5) and they have the properties i) (en ) nN is an orthonormal basis in E; ii) ( n en ) nN is an orthonormal basis in X; iii) (n en ) nN is an orthonormal basis in E * ; iv) (n ) nN is an increasingly sequence to +∞. Consider en V and n (0, ) from the above theorem (applied for the Hilbert spaces V H ) and bn (t ) y 0 n e n t t Pn ( s)e n ( s t ) ds , where y 0 n , Pn (s ) are the 0 ~ Fourier coefficients of y 0 and, respectively, P ( f ) , in V . Now we can apply for the problem (4) the Fourier method presented in [1] and we can formulate the following theorem ~ Theorem 1.2 If P( f ) L2 (0, T ; H ) , then the function y (t ) b (t )e n 1 the space C (0, T ;V ) H 1 (0, T ;V * ) , problem (4), i.e., y(0) = y0 and n n belongs to and this is the unique weak solution of the ~ ( y ' (t ), v) ( Ay(t ), v) ( P( f )(t ), v), v V . Bulletin of the Transilvania University of Braşov Vol 13(48) - 2006 323 ~ Moreover, if P ( f ) H 1 (0, T ; H ) , then y C (0, T ;V ) C 1 (0, T ; H ) is the classical solution of the Cauchy problem (4). 2. Fourier Method for Navier-Stokes system Consider the Navier-Stokes nonlinear system (1) and the Hilbert spaces H { y ( L2 ()) N | y 0, n y 0 on } and V { y ( H 01 ()) N | y 0} . Consider again the Stokes operator A L(V , V ) , N ( Ay, w) yi ( x) wi ( x)dx, y, w V , i 1 and define the threelinear form b : V × V × V → R, b( y, z, w) N y D z w dx , i , j 1 i i j j which determines the nonlinear operator C : V → V , C ( y, w) b( y, y, w), y, w V . We have (u )u (u1 D1 u 2 D2 u N DN )u (u1 D1u k u 2 D2 u k u N DN u k )1k N . Remark 2.1 The scalar product in H between h H and v V is in fact the product between h and v in the duality from V * and V . Consider the weak formulation of the problem, which for f L2 (0, T ;V * ) and u 0 H (given) asks for finding the function u L2 (0, T ;V ) such that u t L2 (0, T ;V * ) and du Au C (u ) f on (0, T ) dt u (0) u 0 (6) (7) A weak solution which belongs to C (0, T; V ) ∩ C1 (0, T;H) is named strong solution. Remark 2.2 From the Remark 2.1, we have that equation (6) can be reformulated as follows d u, v dt H u, v H (u )u , v H f , v , v V (8) Remark 2.3 We can write d u, v dt H f Au C (u ), v H , v V . Because A is linear and continuous from V to V * , if u L2 (0, T ;V ) we have that Au L2 (0, T ;V * ) . On the other side, C (u ) L1 (0, T ;V * ) (see [8] p. 281). So f Au C (u ) L1 (0, T ;V * ) and then u t L1 (0, T ;V * ) . From here we have (see [8], p. 250) that u is a.e. equal with a continuous function from [0, T) to V * , so, after a modification on a set of null measure, u will be a continuous function from [0, T) to V * . On the Solvability of Navier-Stokes Equations 324 ~ ~ ~ Consider now f L2 (0, T ; H ) , f t L1 (0, T ; H ) , u 0 H given, where T 0 . We ~ have (see [8] p. 250) that f is a.e. equal with a continuous function from [0, T ] to H. We will search for the weak solution of the Navier-Stokes equation as the Fourier series in V u ( x, t ) bn (t ) en ( x) lim n n 0 n b k 0 k (t ) ek ( x) (9) where (en ) nN is the orthonormal basis in V formed with eigenfunctions of the duality mapping J : V V * ( J , see [3], p. 127-128), obtained from Theorem 1.1, n are the corresponding eigenvalues, and (bn ) nN are the corresponding Fourier coefficients. Denoting en (enm )1 m N and vn ( x, t ) n b p 0 p (t ) e p ( x) , n N , the kth component (1 ≤ k ≤ N) of the vector (v n )vn will be N n n v1n D1vnk vn2 D2 vnk vnN DN vnk b p e mp b p Dm e kp . m 1 p 0 p 0 We have (using Mertens theorem) (u )u lim (v n )v n ; n ((vn )vn )(, t ), e j H N N n n b p (t )e mp ( x) b p (t )Dm e kp ( x) e kj ( x)dx k 1 m 1 p 0 p 0 p n N N b p l (t )bl (t ) e mp l ( x) Dm elk ( x) e kj ( x)dx p 0 l 0 k 1 m1 denote c j (b0 (t ), b1 (t ), , bn (t )), j 0 . Taking in (8) v = ej ( j N ), replacing formally (9) in this relation and using the fact that < ei, ej >V = δij , i N , Je j e j j e j and < ej , v >V =λj < ej , v >H , v V , we obtain that bj(t) verify the scalar nonlinear Cauchy problem b 'j j b j j c j (b0 , b1 ,, bn ) f j (t ), b j (0) u0 j ,0 j n, (10) where u 0 j and fj(t) are the corresponding Fourier coefficients in V for u0 and f (, t ) . n N . We have that the solution of homogeneous linear system t b j b j 0,0 j n , has the form b j (t ) e j k j , k j R,0 j n . We apply Let ' j the variation of constants method and we search the components of the solution b(t ) (b0 (t ), b1 t ,, bn (t )) of system (10) to be of the form b j (t ) e j t h j (t ),0 j n, (11) and replacing in (10) we obtain, for all 0 ≤ j ≤ n, j t h 'j (t ) e j t f j (t ) j e Denoting h = (hj)0≤j≤n and c j (e 0t h0 (t ), e 1t h1 (t ),, e nt hn (t )) (12) Bulletin of the Transilvania University of Braşov Vol 13(48) - 2006 j t F (t , h) (e j t f j (t ) j e 325 c j (e 0t h0 , e 1t h1 ,, e nt hn )) 0 j n , the system (10) leads us to h' F (t , h), h(0) (u 00 , u 01 , u 02 ,u 0 n ) (13) We have that cj is the sum of quadratic terms in h0, h1, ...hn, and from here it results that F is continuous with respect to (t, h), and F is Lipschitzian in h uniformly with respect to t (being of class C1 with respect to h) on a relative compact neighborhood of the point (0, u 00 , u 01 , u 02 , u 0 n ) R n 2 . Applying now Picard’s theorem, it results that the ~ system (10) has a unique solution h =h(t), h C 1 ([0, T ]), 0 T T . We will prove that the series b (t )e ( x) ,with n 0 n bn given by (11) and (13), converges uniformly to n u C (0, T ;V ) , the series b (t )e n 0 ' n n ( x) converges uniformly to u t C (0, T ;V * ) , and, moreover, this is the unique strong solution of the Navier-Stokes system (1) (for a fixed ). For this, we need the following Lemma: Lemma 2.1 The series n 0 ( N 1) n is convergent in R. Proof. We have 0 0 1 2 n . For R , denote by N(λ) the cardinal number of the set { j N | j } . N(λ) is given by Weyl’s law (see [9]): N ( ) () N N R ( ) , (2 ) N where R(λ) = O(λN−1), ( ) is the volume of the domain , and ωN is the volume of the unity sphere from RN. As R(λ) = O(λN−1), there exist m > 0 and λ′ > 0 such that () N N mN 1 , ' , so N (2 ) () N N () N N mN 1 N ( ) mN 1 , ' . N (2 ) (2 ) N M n { j N | n j n 1} For all n N , consider the sets N ( ) ={ j N | We 1 1 1 } (finite sets) and denote by mn the cardinal number of this set. n 1 j n have that M n { j N | j n 1} \ { j N | j n}, n N * , so mn N ( n 1) N ( n ), n N * and m0 = N(1). On the other hand, there exists n0 N such that N 1 N N 1 N () N () N m n 1 n 1 N ( n 1 ) m n 1 n 1 (2 ) N (2 ) N N 1 N N 1 N () N () N and m n n N( n) m n n , n n0 . N N (2 ) (2 ) From here we obtain that On the Solvability of Navier-Stokes Equations 326 mn m n N 1 2 N 1 N denote N 2 2 ( ) 1 1 N 2 1 1 n 1 1 An . n n (2 ) N Furthermore, we have that N 1 An () N 1 1 2 1 m N 3 1 1 N 2 N 1 N n n (2 ) n 2 n 2 1 Because for N ≥ 2 the series n 1 N 2 1 1 1 . n An N 1 0 , applying the is convergent and lim n n 1 n N 2 2 N 2 2 n An limit comparison criterion it results that the series N 1 is convergent. We have now n 1 n m A 1 1 . But M0 is finite, N 1 N n1 Nn1 , N 1 N 1 N 1 n n jM n j n 0 n jM 0 j n 1 jM n j A for all n N * , and because the series Nn1 is convergent, by the comparison n 1 n 1 that 1 1 criterion it results that the series n 0 N 1 converges. □ n Coming back to our problem, denote M 1 sup sup hn (t ) and M 2 sup sup hn' (t ) . nN t[ 0 ,T ] nN t[ 0 ,T ] We have that M1 < ∞ and M2 < ∞ (because hn, h′n are continuous on the compact interval [0, T], for all n N ). 2 t 2 t We have bn (t ) e n hn (t ) e n hn2 (t ) . But for all δ > 0, δ < T, using the 2 fact that λn → ∞, we have that there exists n N such that e n n , t [ , T ] , so bn (t ) M 12 1 2 N 1 n 2n t Then we have n 0 V bn (t )en bn (t ) , so n 0 2 b (t ) n 0 for all t [ , T ] , where δ is arbitrary in (0, T). 2 1 N 1 n , , n n , t [ , T ] . From here, using the comparison criterion and Lemma 2.1, we obtain that the series e 2n n 2 is convergent b (t )e n 0 n n converges uniformly for t [0, T ] to u C ([0, T ]; V ) . t t Furthermore, for n N we have bn' (t ) n e n hn (t ) e n hn' (t ) ,and we obtain 2 bn' (t ) 2 2 M 12 2n e 2nt 2 M 22 e 2nt . Bulletin of the Transilvania University of Braşov Vol 13(48) - 2006 2 bn' (t ) From here it results that 2 2 M 12 e 2nt 2M 22 n 327 e 2nt . 2n But, similarly with the above method, for any δ > 0, δ < T, using the fact that λn → ∞, we have that there exist n' , n'' N such that 1 e 2nt e 2n e so 2 n 1 N 1 n N 1 n , n n' , t [ , T ] , 1 e 2nt e 2n 2n t N 1 n , n n'' , t [ , T ] , , n n'' , t [ , T ]. Then for n n max {n' , n'' } , we have 2 bn' (t ) 1 2 2 M 12 n N 1 n 1 2M 22 N 1 n , n n , t [ , T ] . Using again the comparison criterion and Lemma 2.1, it results that the series bn' (t ) 2 b (t )e is convergent. But n n 0 2 ' n n 0 so the series b (t )e ' n n 0 n V n 0 * bn' (t ) n 2 n en V * bn' (t ) n n 0 2 , converges to u~ C (0, T ;V * ) . Also, it is very easy to see that n u u~ (as distributions on (0, T)). t We will prove now that u b e is a weak solution for problem (1). For t = 0, we n n n 0 n 0 n 0 bn (0)en ( x) u 0n en ( x) u 0 ( x) . have u ( x,0) We obtain that n b e ,e ' j j 0 j k f j 0 j j 0 k n c j 0 H n J b j e j , ek j 0 H n j 0 j j j (b0 , b1 ,, bn ) e j , ek H H b 'j e j , ek f j ,0 k n , so, furthermore, using that Jej = λjej , we have e j , ek n j j 0 H n n b e , e j H n c j 0 j j (b0 , b1 ,, bn ) e j , ek H ,0 k n . e j , ek H For n → ∞, we obtain: u , eh t Ju, ek H H (u )u, ek f , ek H , k N . On the Solvability of Navier-Stokes Equations 328 But (ek ) kN is an orthonormal basis in V , therefore u ,w t Ju, w H H (u )u, w f , w H , w V , so u is a weak solution of problem (1). Remark 2.4 Under the above conditions, we have that the weak solution u belongs to C (0, T ;V ) C 1 (0, T ; H ) , so it is a strong solution of problem (1). Remark 2.5 We can prove the uniqueness of the strong solution. For this, if we suppose that u1 and u2 are two strong solutions of the Navier-Stokes system (1), and we consider the Fourier series developments in V n 0 n 0 u1 ( x, t ) p n (t )en ( x), u 2 ( x, t ) q n (t )en ( x) , we can prove that p n (t ) q n (t ), n N , so u1 u2 . Now we can formulate the following theorem ~ ~ ~ Theorem 2.1 If f L2 (0, T ; H ) , f t L1 (0, T ; H ) , u 0 H , T 0 , then the function u ( x, t ) bn (t )en ( x) given in (9) is the unique strong solution of the Navier-Stokes n 0 nonlinear system (6)-(7). References 1. Sburlan, C., Sburlan, S., Fourier Method for Evolution Problems, In: BAM2038CI/2002, Technical University Budapest, 2002, p. 27–35. 2. Sburlan, C., On the solvability of Navier-Stokes Equations, In: Mathematical Modeling of Environmental and Life Science Problems, Ed. Acad. Rom., Bucarest, 2006, p. 223–230. 3. Pascali, D., Sburlan, S., Nonlinear Mappings of Monotone Type, Ed. Acad. Rom. Sijthoff & Noordhoff Int.Publ., 1978. 4. Sburlan, C., Sburlan, S., Fourier Method for Stokes Equation, BAM2038-CI/2002, Technical University Budapest, 2002, p. 159–163. 5. Sohr, H., The Navier-Stokes Equations. An Elementary Functional Analytic Approach, Birkhauser Verlag, 2001. 6. Zeidler, E., Applied Functional Analysis. Applications to Mathematical Physics, Springer Verlag, 1995. 7. Sburlan. S., Morosanu, G., Monotonicity Methods for PDE’s, MB-11/PAMM, Budapest, 1999. 8. Temam, R., Navier-Stokes Equations, North-Holland Publ. Comp., 1977. 9. Hörmander, L., The Spectral Function of an Elliptic Operator, Acta Math., 1968, p. 93–218. Asupra rezolvării sistemului Navier-Stokes Rezumat: In aceastǎ lucrare studiem rezolvarea sistemului Navier-Stokes nestaţionar pentru curgerea unui fluid vâscos incompresibil, folosind metoda Fourier abstractǎ. Este tratat atât cazul liniar, cât şi cel neliniar. Prin aceastǎ metodǎ, obţinem existenţa şi unicitatea soluţiei tari a problemei, care este Bulletin of the Transilvania University of Braşov Vol 13(48) - 2006 soluţia clasicǎ a problemei abstracte.Vom construi efectiv soluţia ca o serie Fourier, folosind şirul ortonormal de funcţii proprii ale aplicaţiei de dualitate între un spaţiu Hilbert şi dualul său topologic, care, în cazul nostru, este operatorul lui Stokes. Cuvinte cheie: Metoda Fourier , Ecuaţiile Navier-Stokes 329