Spectral Properties of Elliptic Operators

In previous work we have replaced the strong version of an elliptic boundary value problem Lux = fx BCux = gx with the weak problem find u 5 V such that Bu, v = fv -v 5 V. x5U x5@

The connection between theweak problem and the strong problem, (and in particular the boundary conditions), was established by making use of the abstract Greens formula Bu, v = Lu, v H + T 0 v, T 1 u -u, v 5 V.

If the bilinear form B is V ? elliptic, Bu, u c 0 || u|| 2 -u 5 V. then it follows immediately V from the Lax-Milgram lemma that the weak problem has a unique solution for every f. This is the same thing as saying that L is an isomorphism from V onto H. On the other hand, if B is only V ? H ? coercive, Bu, u c 0 || u|| 2 ? c 1 ||u|| 2 -u 5 V, then V H we are not able to apply this lemma directly to the weak problem. Instead, we replace the weak problem with find u 5 V such that B W u, v = fv + Wu, v H -v 5 V.

For W sufficiently large, B W is V-elliptic and the Lax Milgram lemma implies the existence of a solution to this perturbed problem. Then L W is an isomorphism from V onto H, hence there is a bounded inverse, L ?1 from H to V. Then, since the embedding of V into H is W compact by the Rellich lemma, we have that i E L ?1 : H H is compact. That is, W L W u = f + Wu u = L ?1 f + WL ?1 u I ? Ku = F W W

where Ku = WL ?1 u, and F = L ?1 f. The operator I ? K is an operator of Fredholm type for W W which there is a theorem, known as the Fredholm alternative theorem, stating conditions under which there is a unique solution to I ? Ku = F. By translating this theorem into the notation and terminology of the weak boundary value problem, we obtain a similar existence-uniqueness theorem for the weak boundary value problem. Now we are going to apply a similar approach in considering the eigenvalue problem for an elliptic operator. We suppose that Lux = ?divAx 4ux + cx ux x5U

is uniformly elliptic on U, and we will consider the eigenvalue problem of finding values of the scalar V, such that Lux = V ux ux = 0 x5U x5@

has nontrivial solutions. Note that in the abscence of first order terms, Ld, f 0 = d, Lf 0 -d, f 5 C K U 0

and we say that L is symmetric. Then it follows that the associated bilinear form B is also symmetric, Bu, v = Bv, u -u, v 5 V = H 1 U. 0

Unless cx 0, the bilinear form B is not V-elliptic, but it is V-H-coercive, which is to say, B W is V-elliptic, for W sufficiently large. Then we recall that L W : H 1 U H 0 U is an 0 isomorphism for W sufficiently large. Then, instead of considering the eigenvalue problem for L, we will consider the eigenvalue problem for L ?1 which is a compact operator for which W there is a classical theorem. By translating the conclusions of the classical theorem into the terminology of our elliptic operator, we obtain the following theorem. Theorem Under the assumptions on L, there is a countable collection V k of scalars such that Lux = V ux ux = 0 has nontrivial solutions. Moreover, (a) the V k are all real and L < V 1 V 2 ` K (b) Let N k = u 5 H 1 U : Lu = V k u. Then dim N k < K all k and 0 u, v 0 = 0 if u 5 Nk, v 5 Nj jk x5U x5@ x5U x5@

(c) The weak solutions d k , for

Ldx = V dx dx = 0 ||d|| 0 = 1,

form an orthonormal basis for H 0 U, and at the same time, an orthogonal (but not orthonormal) basis for H 1 U. 0

Proof- Let L denote the smallest value such that L L : H 1 U H 0 U is an isomorphism. 0 Then for any W L, there is a bounded linear operator S W : H 0 U H 1 U, which is 0 inverse to L W ; i.e., L W S W f = f for all f 5 H 0 U. Let K = i E S W : H 0 U H 1 U H 0 U. 0 Then K is a compact linear operator on H 0 U. Note that SWf = u B W u, v = f, v 0 -v 5 H 1 U 0 LWu = f

For arbitrary f, g 5 H 0 U, let Kf = u, Kg = v. Then f, Kg 0 = f, v 0 = B W u, v = B W v, u = g, u 0 = u, g 0 = Kf, g 0 . Thus the symmetry of the bilinear form implies that K is symmetric, and since K is bounded, in fact compact, it follows that K is self adjoint. In addition, Kf, f 0 = u, f 0 = B W u, u c 0 ||u|| 2 , 1 so K is a compact self adjoint, positive operator on the Hilbert space H = H 0 U. We have the following theorem about such operators, Theorem (Fredholm-Riesz-Schauder) For K a compact self adjoint, positive operator on the Hilbert space H, (a) K has a countable collection of positive eigenvalues, S k , accumulating at zero; i.e., ||K|| LH S 1 S 2 ` 0 (b) Let N k = u 5 H : Ku = S k u. Then dim N k < K all k and u, v H = 0 if u 5 Nk, v 5 Nj jk

(c) The normalized eigenfunctions for K form an orthonormal basis for H. The compact self adjoint, positive operators on the Hilbert space H are the analogues of symmetric positive definite matrices on R n . Now L W Sd = d Kd = S d B W Sd, v = d, v 0 -v 5 V = H 1 U 0 hence L W Sd = d L W d = 1 d S so the eigenfunctions of K are the eigenfunctions of L W but the eigenvalues of L W are the reciprocals of the eigenvalues of K. Thus

||K|| LH S 1 S 2 ` 0 In addition, with

implies

0<

1 < J 1 J 2 ` K. ||K|| LH

L W d k = J k d k Ld k = J k ? W d k = V k d k 0< 1 ? W < V 1 V 2 ` K. ||K|| LH

If d k denote the weak solutions of L W dx = J dx dx = 0 ||d|| 0 = 1, then for u 5 V = H 1 U, write 0 Define Note that x5U x5@

u = > k>0 u, d k 0 d k = > k>0 u k d k -u, v 5 V.

H

u, v V = B W u, v + u, v H

u, v V = B W > k>0 u k d k , > j>0 v j d j + > k>0 u k d k , > j>0 v j d j = > k > j u k v j B W d j , d k + > k u k v k .

But hence and, in particular,

B W d j , d k = L W d j , d k H = J j d j , d k H = V j N jk u, v V = > k J k + 1u k v k . d j , d k V = J k + 1 N jk .

To see that the eigenfunctions form a complete family for the inner product on V, suppose u, d k V = 0 Then i.e., But hence -k.

0 = u, d k V = B W u, d k + u, d k H = J k u, d k H + u, d k H J k + 1u, d k H = 0 Jk 1 >0 ||K|| LH -k. -k

u, d k H = 0

-k and since the family is complete in H, u = 0.n

Note that Similarly,

u 5 V = H 1 U iff ||u|| 2 = u, u V = > k J k + 1u 2 < K 0 k V u 5 H = H 0 U iff ||u|| 2 = u, u 0 = > k u 2 < K k H

For F 5 D v U, f 5 DU H, Ff = F> k f k d k = > k f k Fd k = > k f k F k and we have F 5 V v iff |Ff| = > k f k F k = > k 1 + J k 1/2 f k 1 + J k ?1/2 F k > k 1 + J k ?1 F 2 k i.e., In addition, F 5 V v iff u 5 H iff u 5 V iff
1/2

||f|| 2 V 5 2.

1 + J k ?1/2 F k u k 5 2 . 1 + J k 1/2 u k

5 2.

and we can define, more generally, u 5 H s U iff 1 + J k s/2 u k 5 2,

and we have, with this definition, H s U v = H ?s U. This scale of Hilbert spaces, each with the inner product u, v s = > k 1 + J k s u k v k is completely analogous to the Fourier spaces H s R n we have defined previously. Consider now the following examples. 1) Lux = ?ux, u0 = u^ = 0 0<x<^

Here cx = 0 so W = 0 and J k = V k . For this operator, it is easy to compute the eigenvalues and eigenfunctions d k x = 2 2 ^ sinkx and V k = k k = 1, 2, u

Then the function ux = has 2x ^ x 2 1? ^

k^ 2

0 < x < ^/2 ^/2 < x < ^ . for s 1;

u k = u, d k 0 = ? 4 2 sin ^k 1 + V k 1/2 ~ k u 5 H 1 = H 1 U. 0 so

Now i.e.

1 + V k s/2 u k ~ k s?2 5 2

On the other hand, for the function we find but i.e., Finally, for vk = 1 ? ?1 k 5 2 k

vx = 1, 0 < x < ^,

1 + V k s/2 v k ~ k s?1 5 2 v 5 H 0 = L 2 U. wx = Nx ? x 0 wk =

for s 0;

x 0 5 0, ^ wk 6 2 for s > 1/2;

and

2 ^ sinkx 0 5 K

1 + V k ?s/2 w k ~ k ?s 5 2 In particular, 2) N 5 H ?1 U.

Lux, y = ?4 2 ux, y u=0

in U = 0 < x, y < ^ on @

Then, again, c = 0 so J = V, and 2 d j,k x, y = ^ sinjx sinky In this case, for we have and V j,k = j 2 + k 2 , j, k = 1, 2, u

wx, y = Nx ? x 0 Ny ? y 0 x 0 , y 0 5 U 2 w j,k = ^ sinjx 0 sinky 0 1 + V j,k ?s/2 w k ~ 1 + j 2 + k 2 ?s/2 5 2 for s > 1

Then

w 6 H ?1 U. 2x ^ x 2 1? ^ 0 < x < ^/2, 0 < y < ^ ^/2 < x < ^, 0 < y < ^

For the function

ux, y =

we have

u j,k =

c j,k 5 2 j k2 1 + j 2 + k 2 s/2 6 2 if s = 1 j k2

but

1 + j 2 + k 2 s/2 u j,k ~

Then u 5 H but u 6 V. .