Solutions of Inverse Problems for Variational Calculus

Mircea Neagu
Abstract In 1 the author presents a short history of the problem studied in this paper. 2 introduces the notion of harmonic map between a Riemannian space and a generalized Lagrange space, in a natural way. In 3 is proved that for certain systems of dierential or partial dierential equations, the solutions are harmonic maps, in the sense denite in 1. 4 describes the main properties of the generalized Lagrange spaces constructed in 3.

Mathematics Subject Classication: 53C60, 49N45, 35R30 Key words: generalized Lagrange spaces, harmonic maps, geodesics, dierential equations, partial dierential equations.

Introduction

The problem of nding a geometrical structure of Riemannian type on a manifold M such that the orbits of an arbitrary vector eld X should be geodesics, was analysed by Sasaki [5]. Doing with this problem, Sasaki creates the well known almost contact structures on a manifold of odd dimension, although the initial problem rests open. After the introduction of generalized Lagrange spaces by Miron [2], the same problem is resumed by Udrite [6, 7, 8]. This succeded to discover a Lagrange structure on M , s dependent of the vector eld X and a (1,1)-tensor eld, such that the orbits of class C 2 should be geodesics. Moreover, he formulates a more general problem [8], namely 1) There exist the structures of Lagrange type such that the solutions of certain PDEs of order one should be harmonic maps? 2)What is a harmonic map between two spaces endowed with structures of this type? In this paper, the author tries to solve these problems. He will use the new notion of harmonic map on a direction, oering thus a partial answer to the Udrites questions. s

Harmonic maps on a direction

Let (M m , (a)) be Riemannian manifold of dimension m and (N n , hij (x, y)) generalized Lagrange space of dimension n, where a = (a )=1,m are coordinates on M and (x, y) = (xk , y k )k=1,n are coordinates on T N . 1

Remark. On M , the coordinates are indexed by , , , . . . and, on N , respectively T N , the coordinates are indexed by i, j, k, . . .. Also, on M N , the rst m coordinates will be indexed by , , . . . and the last n coordinates by i, j, k, . . .. Let A X (M ) be an arbitrary vector eld on M . If the manifold M is connected, compact and orientable, we can dene the (, A, h)-energy functional or the energy functional on the direction A taking
A Eh : C (M, N ) R, A Eh (f ) = i i i f

1 2

i j (a)hij (f (a), f (A))f f da,

where f = x (f ), the map f . Denition. The map f C (M, N ) is (, A, h)-harmonic i f is a critical point A for the (, A, h)-energy functional Eh . Remarks. i) If hij (x, y) = hij (x) is Riemannian metric, it recovers the classical denition of a harmonic map between two Riemannian manifolds [1]. ii) If we consider M = [a, b] R, 11 = 1 and A = d/dt, we obtain not C (M, N ) = {x : [a, b] N | x C dierentiable} = a,b (N ) and the (1, d/dt, h)energy functional should be E1h (x) =
d/dt

f i = , = det( ) and f : T M T N is the dierential of a

1 2

hij (x(t), x(t))

a

dxi dxj dt, x a,b (N ). dt dt

In conclusion, the (1, d/dt, h)-harmonic curves are exactly the geodesics of the generalized Lagrange space (N, hij (x, y)) [2].

The geometrical interpretation of the solutions of certain PDEs of order one

Let f C (M, N ) be a smooth map and let be the global section not i f = f da |a i y (T M f 1 (T N )).

f (a)

On M N , let T be one tensor of type (1, 1) with all components equal to zero i excepting (T ) i=1,n . Let be the system of partial dierential equations
=1,m

(E)

f = T expressed locally by

f i i = T (a, f (a)). a

If (M, ) and (N, ij ) are Riemannian manifolds we can build a scalar proi j duct on (T M f 1 (T N )) putting < T, S >= (a)ij (f (a))T S , where j i T = T da i and S = S da j . y y 2

In these conditions we can prove the following Theorem. If (M, ), (N, ) are Riemannian manifolds and f C (M, N ) is solution of the system (E), then f is solution of the variational problem asociated to the functional LT : C (M, N )\{f | a M such that < f, T > (a) = 0} R+ , LT (f ) = 1 2 f 2 T 2 1 da = 2 < f, T > 2 T 2 i j ij f f da. 2 < f, T >

Proof. In the space (T M f 1 (T N )), the Cauchy inequality for the scalar product <, > holds. It follows that we will have < T, S >2 T 2 S 2 , T, S (T M f 1 (T N )), with equality i there exists K F(M ) such that T = KS. Consequently, for every f C (M, N ), we will obtain L(f ) = 1 2 f 2 T 2 1 da < f, T >2 2
M

da =

1 V ol (M ). 2

1 Now, if f is solution of the system (E), we will conclude LT (f ) = V ol (M ). This 2 means that f is a global minimum point for LT 2. Remarks. i) In certain particular cases of the system (E), the functional LT becomes exactly a functional of type (, A, h)-energy. ii) The global minimum points of the functional LT are solutions of the system f = KT , where K F(M ) not necessarily with K = 1. Examples. 1. Orbits For M = ([a, b], 1) and T = (x1 (T N )), the system (E) becomes (E1 ) dxi = i (x(t)), x : [a, b] N , dt 2 dxi dxj ij dt. < , x (d/dt) >2 dt dt

that is the system of orbits for , and the functional L is 1 L (x) = 2

b a

Hence the functional L is a (1, d/dt, h)-energy (see ii of rst remarks of this paper). The fundamental metric tensor hij : T N \{y| b(y) = 0} R is dened by 2 hij (x, y) = 2 ij (x) = ij (x) exp 2 ln | < , y > | . < , y > This case is studied, in other way, by Udrite in [6]-[8]. s 2. Pfa systems For N = (R, 1) and T = A 1 (T M ), the system (E) will become (E2 ) df = A, f F(M ),

that is a Pfaan system, and the functional LT reduces to 1 LA (f ) = 2 A 2 f f da, [f (A )]2 3

where A = A

. Hence the functional LA is a (g, A , h)-energy, where a 1 1 (a) and h : T R\{0} R, h(x, y) = 2 = e2 ln |y| . g (a) = A 2 y

3. Pseudolinear functions k We suppose that T (a, x) = k (x)A (a), where k is vector eld on N and A is 1-form on M . In this case the system (E) will be f k (E3 ) = k (f )A (a) a and the functional LT is expressed by 1 LT (f ) = 2 = 1 2 2 A 2 i j ij f f da = < , f (A ) >2

i j g (a)hij (f (a), f (A ))f f da, where

1 2 (a) and hij (x, y) = ij (x) = exp (2 ln | < , y > |)ij (x). g (a) = A 2 < , y >2 It follows that the functional LT becomes a (g, A , h)-energy. Remark. Taking M an open subset in (Rn , = ) and N = (R, = 1), the system from the third example is (P L) f = (a)A (f (a)) , = 1, m. a

Supposing that (grad f )(a) = 0 , a M , the solutions of this system are the well known pseudolinear functions [4]. These functions have the following property for every xed point x0 M , the hypersurface of constant level Mf (x0 ) = {x M |f (x) = f (x0 )} is totally geodesic [4] (i. e. the second fundamental form vanishes identically). In conclusion, these pseudolinear functions are examples of (g, A , h)-harmonic 1 functions on the Riemannian space M, g (a) = with values into the geA 2 (x) . For example, we have the following neralized Lagrange space R, h(x, y) = y pseudolinear functions: 3. 1. f (a) = e<v,a>+w , where v M, w R, is solution for the system (P L) with (a) = 1 and A(f (a)) = f (a)v. < v, a > +w , where v, v M ; w, w R, is solution for (P L) 3. 2. f (a) = < v , a > +w 1 and A(f (a)) = v f (a)w. with (a) = < v , a > +w Remark. A geometrical interpretation of the solutions of the system (E), in the i i general case, when the tensor T is expressed by T (a, x) = t r (x)Ar (a), where r=1 r 1 {r }r=1,t X (N ) is a family of vector elds on N and {A }r=1,t (T M ) is a family of covector elds on M , will be treated, in other sense, in a subsequent paper. 4

The geometry asociated to PDEs of order one

For beginning, we remark that, in all above cases, the solutions of class C 2 of the system f = T becomes (, A, h)-harmonic maps, in the sense denite in this paper. Moreover, the generalized Lagrange structures constructed above are of type (M n , e2(x,y) ij (x)), where : T M \{Hyperplane} R is a smooth function. Now, we assume that a generalized Lagrange space (M n , gij (x, y)) satises the following axioms: a. 1. The fundamental tensor eld gij (x, y) is of the form gij (x, y) = e2(x,y) ij (x). a. 2. The space is endowed with the non-linear connection
i Nj (x, y) = i (x)y k , jk

where i (x) are the Christoel symbols for the Riemannian metric ij (x). jk Under these assumptions, our space veries a constructive-axiomatic formulation of General Relativity due to Ehlers, Pirani and Schild [2]. This space represents a convenient relativistic model, since it has the same conformal and projective properties as the Riemannian space (M, ij ). Developing the formalism presented in [2], [3] and denoting the curvature tensor i eld of the metric ij (x) by rjkl , the following Maxwells equations hold Fij|k + Fjk|i + Fki|j =
h gip rqjk (ijk)

p q y y , y h

Fij |k + Fjk |i + Fki |j = (fij|k + fjk|i + fki|j ), fij |k + fjk |i + fki |j = 0, where the electromagnetic tensors Fij and fij are Fij = gip gjp i xj x y p , fij = gip gjp i y j y yp.

k We will use the following notations: rij = rijk , r = ij rij ,

= Nij j , xi xi y

H = kl

, V = ab a b , = ij ij , = ab ab , xk xl y y ij = |j + 1 ij H xi xi xj 2 where ab = + 1 ab V . y a b y a y b 2

The Einsteins equations of the space (M, gij (x, y)) take the form rij 1 rij + tij = KT H ij 2 V (2 n)(ab ab ) = KTab , 5

H V where Tij and Tab are the h- and the v-components of the energy momentum tensor eld, K is the gravic constant and

tij = (n 2)(ij ij ) + ij rst y s tp

s a + i rtja y t is ap p rtja y t . p y y y

Consequently, in certain particular cases, it is posible to build a generalized Lagrange geometry (in Mirons sense) naturally attached to a system of partial dierential equations. This idea was suggested by Udrite in private discussions. s Open problem. Because the generalized Lagrange structure constructed in this paper is not unique, it arises a natural question: -Is it possible to build a unique generalized Lagrange geometry naturally asociated to a given PDEs system?

References
[1] J. Eells, L. Lemaire, A report on harmonic maps, Bull. London Math. Soc. 10 (1978), 1-68. [2] R. Miron, M. Anastasiei, The Geometry of Lagrange Spaces: Theory and Applications, Kluwer Academic Publishers, 1994. [3] R. Miron, R. K. Tavakol, V. Balan, I. Roxburgh, Geometry of Space-Time and Generalized Lagrange Gauge Theory, Publ. Math. , Debrecen, 42, 3-4 (1992), 215-224. [4] T. Rapcsk, Smooth Nonlinear Optimization in Rn , Kluwer Academic Publishers, a 1997. [5] S. Sasaki, Almost Contact Manifolds, I, II, III, Mathematical Institute Tohoku University, 1965, 1967, 1968. [6] C. Udrite, A. Udrite, Electromagnetic Dynamical Systems, Balkan Journal of s s Geometry and Its Applications, Vol. 2, No. 1, 1997, 129-140. [7] C. Udrite, Electromagnetic Dynamical Systems as Hamilton-Poisson systems, s Workshop on Di. Geom. , Global Analysis, Lie Algebras, Aristotle University of Thessaloniki, June 25-29, 1997. [8] C. Udrite, Geometric dynamics, Second Conference of Balkan Society of Geomes ters, Aristotle University of Thessaloniki, Greece, June 23-26, 1998, Southeast Asian Bulletin of Mathematics, Springer-Verlag, 24(2000), 1-11.

University POLITEHNICA of Bucharest Department of Mathematics I Splaiul Independentei 313 77206 Bucharest, Romania e-mail:mircea@mathem.pub.ro