Invariant subspaces of the periodic Navier–Stokes and magnetohydrodynamics equations: Symmetries and inverse cascades

Invariant subspaces of the periodic Navier–Stokes and magnetohydrodynamics equations: Symmetries and inverse cascades Manuel Núñez Citation: Journal of Mathematical Physics 41, 6193 (2000); doi: 10.1063/1.1287920 View online: https://doi.org/10.1063/1.1287920 View Table of Contents: http://aip.scitation.org/toc/jmp/41/9 Published by the American Institute of Physics Invariant subspaces of the periodic Navier–Stokes and magnetohydrodynamics equations: Symmetries and inverse cascades Manuel Nu ´n ˜ez Departamento de Ana ´lisis Matema ´tico, Universidad de Valladolid, 47005 Valladolid, Spain 共Received 14 February 2000; accepted for publication 24 May 2000兲 It is shown that when the initial condition and the forcing term of the periodic Navier–Stokes or magnetohydrodynamics equations have Fourier coefficients which vanish outside a certain semigroup of frequencies, the same happens to the solutions for all time. Subgroups of frequencies correspond to solutions possessing certain symmetries. By taking as a semigroup the frequencies whose Fourier components are non-negative integers, we get a class of solutions for which the higher modes do not influence the evolution of the lower ones; therefore, the phenomenon of inverse cascading cannot occur for them. © 2000 American Institute of Physics. 关S0022-2488共00兲03609-4兴 I. INTRODUCTION So far there exists a wide gulf between what is accepted in fluid mechanics and what can be rigorously deduced from the Navier–Stokes equations. This is specially true in the field of turbulence, where the standard view of the transfer of energy in fully developed turbulent flows remains analytically unproved. Let use remember that energy is typically introduced into the turbulent fluid by a large-scale forcing, is transported to smaller scales through an intermediate 共called inertial兲 range, and finally it is dissipated into heat at the dissipative range. For periodic problems, these ranges are identified with certain subsets of Fourier modes: the scales decrease as the spatial frequency grows. For dimension two, however, the above picture of a direct cascade is not entirely correct: Kraichnan and others1–5 showed that energy injected at a certain intermediate scale may be transferred to the largest scales and fairly obvious large structures appear in the flow. Those inverse cascades are believed to occur also for a number of magnetohydrodynamic magnitudes: the mean-square magnetic potential in two-dimensional and the magnetic helicity in threedimensional magnetohydrodynamics 共MHD兲共see Ref. 6 for the relevant definitions兲. The standard modern presentation7–10 of these evolution equations involves a number of subspaces of the space of square-integrable functions L2(⍀), plus Galerkin approximations and compactness theorems. These techniques work as well for closed subspaces of the standard ones, provided these subspaces are invariant for the relevant operators. For the periodic case, a number of possibilities are apparent. By choosing a certain class of invariant subspaces, we will show that when the initial condition and the forcing term satisfy a certain type of symmetry, the same happens to the solutions for as long as they are defined. A second example will yield whole families of solutions for which there is no inverse cascade at all: the large scales ignore the small ones in their evolution, irrespective of the value of the Reynolds number. These results, however, do not represent a threat for our picture of inverse cascading, because our velocity and magnetic fields are complex functions and therefore do not have any clear-cut physical meaning. The nonlinear character of the Navier–Stokes and MHD equations precludes the classical trick of taking the real part of a complex solution to find a real one. Hopefully this construction may be helpful to understand which analytic techniques may work in order to prove rigorously the existence of inverse cascades. Another caveat is that our results depend on very specific arithmetical JOURNAL OF MATHEMATICAL PHYSICS VOLUME 41, NUMBER 9 SEPTEMBER 2000 61930022-2488/2000/41(9)/6193/5/$17.00 © 2000 American Institute of Physics properties of the complex exponentials and therefore cannot be extended in any obvious way to other boundary value problems. II. THE MATHEMATICAL FORMALISM Consider a viscous incompressible fluid filling 共for notational simplicity兲the box ⍀ ⫽关0,2 ␲ 兴N, possessing a velocity uand pressure p, under the action of a time-independent forcing f. We will assume that all the magnitudes are spatially periodic, and that the mean of uand 共necessarily兲fare zero, 具 u 典 ⫽ 冕 ⍀u共t,x兲dV⫽0.共1兲 After normalization of constants, uwill satisfy the incompressible Navier–Stokes equation ⳵ u ⳵ t⫽ ␯ ⌬u⫺u•ⵜu⫺ⵜp⫹f, ⵜ•u⫽0, 共2兲 u共0兲⫽u0, where ␯ stands for the fluid viscosity. If the fluid is a conducting one with resistivity ␩ , and a magnetic field Bis present, it acts upon the velocity through the Lorentz force while satisfying the induction equation. The resulting magnetohydrodynamic 共MHD兲equations, after normalization of constants, are ⳵ u ⳵ t⫽ ␯ ⌬u⫺u•ⵜu⫹B•ⵜB⫺ⵜ 冉 p⫹B2 2 冊 ⫹f1, ⳵ B ⳵ t⫽ ␩ ⌬B⫺u•ⵜB⫹B•ⵜu⫹f2, ⵜ•u⫽0, ⵜ•B⫽0, 共3兲 具 u 典 ⫽ 具 B 典 ⫽0, u共0兲⫽u0, B共0兲⫽B0. Both systems may be cast in the same form: after projection in the space of fields with zero divergence 共killing in this way all the gradients兲, one gets ⳵ w ⳵ t⫽Dw⫹C共w,w兲⫹f,共4兲 where w⫽u共Navier–Stokes兲or (u;B)共MHD兲;Dis an elliptic operator 共either v⌬or (v⌬; ␩ ⌬)兲, and Cis a bilinear form with the antisymmetric property (C(w1,w2),w2)⫽0. For the Navier– Stokes equations, C(u,v)⫽⫺P(u•ⵜv), where as mentioned before Pis the orthogonal projection upon the space of solenoidal fields, with mean zero. In the MHD case, C共共u1,B1兲;共u2;B2兲兲⫽共P共⫺u1•ⵜu2⫹B1•ⵜB2兲;P共⫺u1•ⵜB2⫹B1•ⵜu2兲兲. 6194 J. Math. Phys., Vol. 41, No. 9, September 2000 Manuel Nu ´n ˜ez From now on, we will restrict ourselves to the Navier–Stokes equations to simplify the notation. The proofs for the MHD case are the same with obvious modifications. The main working spaces are H⫽ 兵 u苸L2共⍀兲N:ⵜ•u⫽0, 具 u 典 ⫽0,u•nanti⫺periodic 其 ,共5兲 V⫽H1共⍀兲N艚H.共6兲 The divergence of uis to be understood in the sense of distributions. It may be shown that the boundary values u•nhave in this case a precise mathematical sense.7H1(⍀) is the usual Sobolev space of functions whose gradient is square-integrable. The projection P:L2(⍀)N→His easily expressed in terms of the Fourier series of a periodic function, P 冉 兺 k苸ZNakeik•x 冊 ⫽兺 k苸ZNbkeik"x,共7兲 where b0⫽0, and bkis the projection within CNof akon the hyperplane k•x⫽0. It is known that Ctakes V⫻Vinto the dual space V⬘. III. ALTERNATIVE SETTINGS AND CONSEQUENCES Let H0be a closed subspace of Hsuch that the Laplacian ⌬takes H2(⍀)N艚H0into H0, and if u,v苸H0are such that P(u•ⵜv)苸H, then P(u•ⵜv)苸H0. Let V0⫽V艚H0, and assume u0,f 苸H0. Then all the standard proofs of existence work with H0,V0instead of H,Vand we get the same results: the solution u苸C(关0,T兴,H0)艚L2(关0,T兴,V0) up to some T(T⫽⬁for N⫽2兲and, if u0苸V0,u苸C(关0,T兴,V0)艚L2(关0,T兴,H0艚H2(⍀)N). Thus the solution remains within H0for all time. The only difficulty is to find a subspace H0satisfying all the requirements. Let us look at the Fourier representation of u.⌬ucorresponds to 兺⫺ 兩 k 兩 2uˆkeik"x. On the other hand, the k-th Fourier coefficient of ujvlis 具 ujvl 典 共k兲⫽兺 n苸ZNuˆ j共n兲v ˆl共k⫺n兲.共8兲 Take tentatively as H0the space of functions of Hwhose Fourier coefficients vanish outside a subset A傺ZN. Then H0is invariant for ⌬. For the product of two functions of H0to lie within H0, we need that whenever k苸A,(ujvl)ˆ(k)⫽0. Since we may choose at will ujand vl, this happens if and only if all the summands within 共8兲vanish, i.e., if k苸A,n苸A, then k⫺n苸A. This means that Amust be a semigroup of ZN: i.e., if m,n苸A, then m⫹n苸A. The projection Pkeeps such functions within H0, since it does not extend the support of the Fourier transform. Hence any subspace H(A) formed by the functions whose Fourier transform has support contained within a semigroup A傺ZNis an invariant one for the Navier–Stokes equation. If Ais a subgroup, we may say a good deal about the possible solutions. Subgroups of ZNare free and generated by at most Nelements. This means that there are a number of basic frequencies, k1,...kp,p⭐N, such that H(A) is formed by the functions whose Fourier modes correspond to frequencies which are a linear combination with integer coefficients of k1,...kp. Since any complex exponential eink"xis a function of k•x, such solutions depend only on the variables k1 •x,...kp•x. To see the meaning of this, consider N⫽2, k1⫽(1,0). All such velocities are functions of the first spatial variable x. Take now k1⫽(2,0), k2⫽(0,2). Then the velocities depend on 2x,2y, i.e., u(x,y)⫽u( ␲ ⫹x,y)⫽u(x, ␲ ⫹y). We see that those are functions satisfying a certain number of symmetries. It is not unexpected for the Navier–Stokes and MHD equations to keep for all time the symmetries of the initial condition and the forcing, although it is not entirely trivial given their nonlinear character. 6195J. Math. Phys., Vol. 41, No. 9, September 2000 Invariant subspaces of the periodic Navier–... The analysis for general semigroups is made difficult by its possible complicated structure. We will concentrate on the simplest and most important: NN, i.e., the set of k苸ZNwith kj⭓0 for all j. Of course, if Ais a semigroup, so is ⫺A, which means that we could have taken as well negative frequencies. Let QMdenote the projection from L2(⍀)Ninto the finite-dimensional space of trigonometric polynomials with frequencies bounded by M, QM共f兲⫽兺 兩 k 兩 ⭐Mf ˆ共k兲eik•x, 兩 k 兩 ⫽ 兩 k1 兩 ⫹•••⫹ 兩 kN 兩 .共9兲 Let ube a solution of the Navier–Stokes equations such that u0,f苸H(NN). Then QMusatisfies ⳵ QMu ⳵ t⫽ ␯ ⌬QMu⫺PQM共QMu•ⵜQMu兲⫹QMf, ⵜ•QMu⫽0, 共10兲 QMu共0兲⫽QMu0. The proof is simple: let us apply the projection QMto the Navier–Stokes equation. QMcommutes with ⳵ / ⳵ t,⌬, and P. For any product uv of functions within H(NN), and 兩 k 兩 ⭐M, the kth Fourier coefficient is 共uv兲ˆ共k兲⫽兺 luˆ共l兲v ˆ共k⫺l兲.共11兲 For any summand to be nonzero, we need lj⭓0, kj⫺lj⭓0, which means kj⭓0, plus 兺jkj⭐M; thus 兺jlj⭐M,兺jkj⫺lj⭐M. Hence only the coefficients uˆ(l), v ˆ(l) with 兩 l 兩 ⭐Moccur. In other words, QM共u•ⵜu兲⫽QM共QMu•ⵜQMu兲,共12兲 from which the result follows. Equation 共10兲is in fact an ordinary differential one, since QMuis finite dimensional: all the spatial derivatives are continuous linear functionals. It is known that it has a unique solution for all time 共even when N⫽3兲because the nonlinear term is a polynomial in Qmu: it is a finitedimensional reaction-diffusion equation. Standard energy inequalities show that Qmutends in L2((0,⬁),H) to a weak solution u, which of course coincides for N⫽2 with the smooth one. The important thing, however, is that QMusatisfies by itself an evolution equation with no implication whatsoever of the remaining modes of u. Thus, although smaller frequencies 共larger scales兲do have an effect on the behavior of a certain mode, larger frequencies 共smaller scales兲do not. This clearly precludes inverse cascades. As for the kind of forcing within H(NN), they can be indeed very complicated functions, but they cannot be real ones: for real functions one has f ˆ(⫺k)⫽f ˆ(k). Hence uitself must be complex, which is hard to interpret as a physically realistic velocity field. And, as told before, the real part of uis not a solution of a Navier–Stokes equation with time-independent forcing: certainly one could add a term ⫺v•ⵜvto f, with vthe imaginary part of u, but of course any function is a solution of the Navier–Stokes equation with a suitable forcing term, so this is not meaningful. Moreover, this trick would yield a time-dependent forcing term. It is worth noting that when fsatisfies a Gevrey condition (兺 兩 f ˆ(k) 兩 2e ␴ 兩 k 兩 ⬍⬁for some ␴ ⬎0兲, the solution uis itself Gevrey for as long as it remains bounded in the H1-norm, which mean always for N⫽2共see Ref. 11; the proofs there work without substantial modification for the 6196 J. Math. Phys., Vol. 41, No. 9, September 2000 Manuel Nu ´n ˜ez MHD case兲. Taking as new variables zj⫽eixj, this means that fand u(t) may be analytically extended to a neighborhood of the torus TN傺CN. If moreover fand u0belong to H(NN), the Fourier expansion becomes a Taylor one, and what we have proved is that u(t) remains analytic in a neighborhood of the polidisk B ¯ (0,1)傺CN. The wide variety of analytic functions shows that the forcing functions fmay be made to satisfy many properties. Concerning other magnitudes for which the phenomenon of inverse cascading is claimed, we find much the same result. The mean-square magnetic potential in two-dimensional MHD is found in the following way: let B⫽兺B ˆ(k)eik"xbe the Fourier series of the plane magnetic field B ⫽(B1,B2,0), satisfying as always k•B ˆ(k)⫽0. Take A ˆ(k) such that ik⫻(0,0,A ˆ(k))⫽B ˆ(k). Then A(x)⫽兺A ˆ(k)eik"xis the scalar potential vector. Its mean-square potential is 兺 兩 A ˆ(k) 兩 2. Since the Fourier components B ˆ(k), 兩 k 兩 ⭐M, depend for all time only on the remaining B ˆ(l), 兩 l 兩 ⭐M, the same happens for A ˆ(k). Thus the mean potential of QMAdepends only on the previous history of the modes up to size M, and there is no inverse cascade. As for the magnetic helicity Hin three dimensions, it has no interest in H(NN) because H⫽兺A ˆ(k)•B ˆ(⫺k)⫽0. Since the magnetic helicity measures in a sense the knottedness of the magnetic field, these fields are in a sense topologically simple; but this has not much to do with turbulence, which may be present at dimension two, where there is no possibility of a field to become entangled. IV. CONCLUSIONS By analyzing the invariant subspaces of the operators occurring in the periodic Navier–Stokes and magnetohydrodynamic equations, we have found that when the Fourier frequencies of the forcing and initial conditions lie within a certain semigroup of ZN, the same happens to the solution for as long as it is defined in the classical sense. When we take as a semigroup a subgroup of ZN, we find that these solutions correspond to fields satisfying a certain number of symmetries. By taking as a semigroup the set of positive frequencies, we find that the truncated Fourier series of the solution satisfies an evolution equation by itself and therefore it ignores the remaining Fourier modes. As a consequence there cannot be inverse cascading for this class of solutions. These turn out to be always complex functions and not realistic physical fields, but these results may be helpful to understand the analytic setting where existence of inverse cascades may eventually be rigorously proved. 1R. H. Kraichnan, ‘‘Inertial ranges in two-dimensional turbulence,’’ Phys. Fluids 10, 1417–1423 共1967兲. 2R. H. Kraichnan, ‘‘Eddy viscosity in two and three dimensions,’’ J. Atmos. Sci. 33, 1521–1536 共1976兲. 3E. D. Siggia and H. Aref, ‘‘Point-vortex simulation of the inverse cascade in two-dimensional turbulence,’’ Phys. Fluids 24, 171–173 共1981兲. 4U. Frisch and P.-L. Sulem, ‘‘Numerical simulation of the inverse cascade in two-dimensional turbulence,’’ Phys. Fluids 27, 1921–1923 共1984兲. 5V. Borue, ‘‘Inverse energy cascade in stationary two-dimensional turbulence,’’ Phys. Rev. Lett. 72, 1475–1478 共1994兲. 6D. Biskamp, Nonlinear Magnetohydrodynamics 共Cambridge University Press, Cambridge, 1993兲. 7R. Temam, Navier-Stokes Equations, Theory and Numerical Analysis 共North-Holland, Amsterdam, 1984兲. 8R. Temam, Navier-Stokes Equations and Nonlinear Functional Analysis, CBMS-NSF Regional Conference Series in Applied Mathematics, SIAM, Philadelphia, 1983. 9M. Sermange and R. Temam, ‘‘Some mathematical questions related to the MHD equations,’’ Commun. Pure Appl. Math. 36, 635–664 共1983兲. 10R. Temam, Infinite Dimensional Dynamical Systems in Mechanics and Physics 共Springer, New York, 1988兲. 11C. Foias and R. Temam, ‘‘Gevrey class regularity for the solutions of the Navier-Stokes equations,’’ J. Funct. Anal. 87, 359–369 共1989兲. 6197J. Math. Phys., Vol. 41, No. 9, September 2000 Invariant subspaces of the periodic Navier–...