EXPLICIT REPRESENTATION OF THE SOLUTION TO SOME BOUNDARY VALUE PROBLEM
S. MAZOUZI AND R. N. PEDERSON
Abstract.
In the half ray the unique solution to the boundary value problem L Dtu t f t;t>0 Bj Dtu 0 j;forj0; 1; :::;pÿ1
rapidly decreasing at in¢nity is shown to be explicitly represented in terms of Green's function and some boundary kernels, namely,
u t Xpÿ1
k0hk tkZ 1
0g t;sf sds
1. Introduction.
LetL zbe a polynomial of degreem1, where the coe¤cient ofzmis equal to 1, and letfBj zgpÿ1j0 beppolynomials of degreesfmjgpÿ1j0 respectively, so that mj<m for 0jpÿ1: We assume that L z has at most p roots having positive imaginary parts (counting multiplicities).
Throughout this paper we denote byDt the di¡erential operatorÿidtd;and ifJ Rthen we denote bys Jthe subspace ofC1 Jcontaining all the functionsu tsuch that
1 jtjpju q tj are bounded for allpandqinN.
Consider the following boundary value problem L Dtu t f t;t>0 2:1
Bj Dtu 0 j; forj0; 1; :::;pÿ1 2:2
Received October 19, 1995.
where thej are complex constants. Set
Lp z zÿ0 zÿ1::: zÿpÿ1 2:3
where thej's are theproots ofL zhaving positive imaginary parts.
Firstly, we assume thatmj<pforj0; 1; :::;pÿ1, then Bj z
Xpÿ1 k0
bjkzk 2:4
for j0; 1; :::;pÿ1:Now, if the matrix bjkis nonsingular, one can solve Eq. (2.4) for the unknown variableszk(for ¢xedz).
Hence, if we denote by bjkthe inverse matrix of bjk, we get zj
Xpÿ1 k0
bjkBk z
2:5
forj0; 1; :::;pÿ1:So that, Eq. (2.2) is equivalent to Dktu 0
Xpÿ1 j0
bkjj 2:6
fork0; 1; :::;pÿ1:Thus, any solution of the Cauchy problem (2.1), (2.2) is a solution of the problem
L Dtu t f t;t>0 2:7
Dktu 0 Xpÿ1
j0
bkjj
2:8
fork0; 1; :::;pÿ1:Conversely, any solution of (2.7), (2.8) is a solution of (2.1), (2.2).
Remarks. If the matrix bjkis singular, then there are constants cj not all zero so that
Xpÿ1 j0
cjBj z 0 2:9
Consequently, a necessary condition for the problem (2.1), (2.2) to have a solution is the following
Xpÿ1 j0
cjj0 2:10
This shows that we can not hope to get a solution for all choices of thej's.
Furthermore, even when we can solve, the solution is generally not unique.
Now if we allow to themj's to be greater thanp, then, by partial fractions we can write
Bj z Qj zLp z B0j z
2:11
Where the degree ofB0jis less thanp.
Whenp<mwe de¢ne the polynomial Lÿ z L z
Lp z
2:12
We conclude by the following Lemma [5]:
Lemma. Let L Dtbe any constant coefficient differential operator, and let f tbe any function ins(R).Then, there exists a function u t 2s Rwhich satisfies the differential equation
L Dtu t f t; for t>0 2:13
that the di¡erential equation
Lÿ Dtv t f t;t>0 2:14
has always a solutionv tbelonging tos(R) for any choice off tins R.
Thus, the boundary value problem (2.1), (2.2) is equivalent to the following LP Dtu t v t; t>0
2:15
Bj0 Dtu 0 jÿQj Dtv 0; j0; 1; :::;pÿ1 2:16
which is the form just treated. If we write Bj0 z
Xpÿ1 k0
b0jkzk; 0jpÿ1 2:17
then the problem (2.15), (2.16) has a unique solution for any given v t in s(R) and {j}Cif and only if the matrix b0jkis nonsingular.
We observe from Eq. (2.9) that the matrix bijis nonsingular if and only if the polynomials Bj zare linearly independent. Similarly, the matrix b0ij is nonsingular if and only if the polynomialsB0j zare linearly independent.
We say that thefBj zg are linearly independent modulo Lp z if the fBj0g are linearly independent.
2. Main results.
Now we are able to give explicitly the solution to the boundary value pro- blem (2.1), (2.2) in terms of Green's function; we shall follow closely Pederson's work after rede¢ning in a convenient manner the functionsuj t
(see [3], [4]).
Theorem 1. Let L zbe a polynomial of degree m, having at most p roots 0,1,...,pÿ1with positive imaginary parts and no real roots. LetfBj zgpÿ1j0be p polynomials of degrees fmjg with mj<m, which are linearly independent modulo
Lp z zÿ0 zÿ1::: zÿpÿ1
Then, for any f 2s Rand for any choice of the constants0; 1; :::; pÿ1; there exists a unique solution u t 2s Rsatisfying the boundary value pro- blem(2.1), (2.2).Furthermore, this solution can be represented as follows
u t Xpÿ1
j0
hj tj Z 1
0 g t;sf sds 2:18
Proof. Let us first consider the casef 0; then, the general solution of the equation
L Dtu t 0 2:19
has the form
u t mÿ1X
k0
kexp ik
where thek's are the roots ofL z 0. It is worth to recall that the coe¤- cientskbecome polynomials in twhenever there are multiple roots. Now, in order foru tto be inL2 0; 1;the coe¤cientskmust vanish for any k such thatIm k0 otherwiseu tcould not be inL2 0;1. Therefore, the solution of (2.19) which belongs to the spaceL2 0;1is
u t Xpÿ1
k0
kexp ik 2:20
withImk>0.
Following AGMON, DOUGLIS and NIRENBERG [1], we de¢ne Lk
Xk j0
aj kÿj; k0; 1; :::;pÿ1 2:21
where the constantsaj are de¢ned through the expression Lp
Xp j0
aj pÿj 2:22
Letÿandÿÿbe recti¢able Jordan contours in the upper and the lower half plane enclosing the roots ofLp zandLÿp zrespectively.
De¢ne the functions uj t 1
2i I
ÿ
Lpÿjÿ1
Lp exp itd 2:23
We claim that
L Dtuj t 0; t>0 2:24
and
Dktuj 0 jk 2:25
for j; k0; 1; :::;pÿ1, where jk is the Kronecker Delta. Indeed, by dif- ferentiation under the integral sign (which is of course allowed), we get
Dkt uj t 1 2i
I
ÿ
Lpÿjÿ1
Lp kexp itd 2:26
Now, if we takeÿ to be a large circle about the origin with radiusn 2N*
so thatÿ encloses0; 1; :::; pÿ1;then, Dkt uj 0 1
2 Z 2
0
Lpÿjÿ1 nei!
Lp nei! nk1 ei! k1d!
1 2
Z 2
0
Q nei! Lp nei!d!
Where Q is a polynomial of degreepkÿj in n. Since Lp is of degree p, then, by lettingngo to in¢nity, we obviously get
Dktuj 0 jk; for kÿj0
For the case kÿj>0, we note that the polynomial kLpÿjÿ1 di¡ers from kÿjÿ1Lp by a polynomialQof degree at most equals tokÿ1. Thus,
Dktuj 0 1 2i
I
ÿ
Lpÿjÿi Lp kd
1 2i
I
ÿ
kÿjÿ1Lp Q Lp d
1 2i
I
ÿkÿjÿid 1 2i
I
ÿ
Q Lp d
1 2i
I
ÿ
Q Lp d
By the same argument as before, since the degree of Q z is equal to kÿ1<pÿ1, we can observe that the last integral is zero. As a con- sequence, we obtain
Dktuj 0 jk; j;k0;1; :::;pÿ1 On the other hand we have
L Dtuj t 1 2i
I
ÿ
Lpÿjÿi Lp Lÿ
LP eitd 0:
It follows that the setfuj tgspans the negative exponential solutions of the homogeneous boundary value problem associated to (2.1), (2.2). Now, in order to obtain a solution to the inhomogeneous boundary value problem (2.1), (2.2), we de¢ne the functions
v t 1 2
I
ÿ
eit L d 2:27
It follows from the fact that the contour ÿ[ÿÿ can be deformed into a large circle that
Dkt v 0 vÿ 0 imÿ1;k
2:28
wherek0; :::;mÿ1. As a consequence, the function Z t
0 v tÿs vÿ tÿsf sds 2:29
is a solution of the Eq. (2.1) with zero Cauchy Data.
Thus, the general solution of the inhomogeneous boundary value problem (2.1), (2.2) which is bounded must have the form:
u t Xpÿ1
j0
juj t Z t
0 v tÿs vÿ tÿs f sds 2:30
ÿ Z 1
0 vÿ tÿsf sds
This is a consequence of the facts that v t is a sum of the negative ex- ponentials when t>0, and vÿ t is a sum of negative exponentials when t<0. Now, by virtue of the complementing condition (linear independence of theB0js, we conclude that
Bk Dt Z t
0 v tÿs vÿ tÿsf sds 0
t0
Since the function (2.30) is a formal solution of (2.1), (2.2) then, it must sa- tisfy the following
kBk Dtu 0 Xpÿ1
j0
jBk Dtuj 0
Bk Dt Z t
o v tÿs vÿ tÿsf sds ÿ
t0
ÿ Z 1
0 f sBk Dtvÿ tÿsds
t0
Xpÿ1
j0
jfQk DtLp Dtuj 0 B0k Dtuj 0gÿ
ÿ Z 1
0 f sBk Dtvÿ tÿsds
t0
Xpÿ1
j0
j bkjÿ Z 1
0 f sBk Dtvÿ tÿsds
t0
whereBk z Qk zLp z B0k z Bk0 zmod Lpand
B0k z Xpÿ1
j0
bkjzj
We get an algebraic system of equations with unknown variablesj
Xpÿ1
j0
bkljk Z 1
0 f sBk Dtvÿ tÿsds
t0
; k0;1; :::;pÿ1
We deduce from the complementing condition that the determinant of the matrix bkjis not zero; so, as a consequence, the above set of equations has a unique solutionf0; ::: ; pÿ1g:
De¢ne the inverse matrix
bkj bkjÿ1 Hence,
jXpÿ1
k0
bjkk Z 1
0
Xpÿ1
k0
bjkf sBk Dtvÿ tÿs
t0
; j0;1; :::;pÿ1
and upon substitution of thej's into (2.30) we obtain the bounded solution of the given BVP,
u t Xpÿ1
j0
Xpÿ1
k0
bjkkuj t
Z 1
0
Xpÿ1
j0
Xpÿ1
k0
bjkBk Dtvÿ tÿs f sds
( )
uj t
t0
Z t
0 v tÿsf sdsÿ Z 1
t vÿ tÿsf sds If we set
hk t 1 2i
I
ÿ
Xpÿ1
j0
bjkLpÿjÿ1 Lp eitd fork0; ::: ;pÿ1
g1 t 1 2
I
ÿ
eit
L d; if t>0
ÿ1 2 I
ÿÿ
eit
L d; if t<0
g2 t;s ÿi 42
I
ÿ
I
ÿÿ
Xpÿ1
j0
Xpÿ1
k0
Lpÿjÿi Bk
Lp L ei tÿsd d and
g t;s g1 tÿs g2 t;s
then, the solution of the boundary value problem (2.1), (2.2) takes the ¢nal form
u t Xpÿ1
j0
hj tj Z 1
0 g t;sf sds
An immediate computation shows that the above kernels satisfy the follow- ing estimates
jDkthj tj C0exp ÿr0t; 8t>0;8k0; 1; :::
jDktg1 tj C1 exp ÿr1jtj; 8t2R;8k0; 1; :::
jDktg2 t;sj C2 exp ÿr2 ts; 8t>0;8s>0;8k0; 1; :::
for some positive constantsC0,C1,C2,r0,r1, andr2, depending only onL z
andfBjg.
Now to see that the expression (2.18) is rapidly decreasing at in¢nity it su¤ces to show that the function
vj t tj Z 1
0 jf sj C1 exp ÿr1jtÿsj C2 exp ÿr2 tsds is bounded for each nonnegative integerj. Since f 2s Rthere is a con- stantC>0 such that
jf sj C
1sj2;8s>0 it then follows that
vj t C0tj 2 t
jZ t 2
0
ds
1sj2C1 2tj 2tj
Z t
t2
ds 1s2
C1 t 1t jZ 1
t
ds
1s2C20tj Z 1
0 exp ÿr2 tsds C j
Z 1
0
ds
1s2C3tjexp ÿr2t<1
Hencevj tis bounded inRand consequentlyu t 2s R:
Finally, using classical techniques we can easily prove the uniqueness of this solution. This establishes the proof of the given theorem.
Let us denote byHk R;k0 the completion of the space s Rwith respect to the norm
kuk2kXk
j0
Z 1
0 ju k tj2dt 2:31
and we de¢ne the subspace
VkHk R \Ck0;1
As a consequence of the previous representation theorem and Theorem 6^9 [5] we obtain the estimate of the solution to the problem (2.1)^(2.2) in terms of the Dataf and0; :::; pÿ1:
Theorem 2. Under the same assumptions of Theorem 1, we conclude that for each k2N;there is a constant C>0(depending only on L z;Bj zand k) such that, for each f 2Vkand0; ::: ; pÿ12Cp;the solution u2Vmkto the BVP 2:1 ÿ 2:2satisfies the estimate
kukmkC Xpÿ1
j0
jjj kf kk
!
and has the representation u t Xpÿ1
j0
hj tj Z 1
0 g t;sf sds 2:32
(wherehj tandg t;sare the same as in Theorem 1.).
Proof. We deduce form the density of s R in Vk that there is a se- quence fn s R converging to f in Vk: On the other hand there corresponds to eachfn at most one solutionun 2s Rsatisfying:
L Dtun t fn t; t>0
Bj Dtun 0 j; j0; ::: ;pÿ1 and given by
un t Xpÿ1
j0
hj tj Z 1
0 g t;sfn sds 2:33
We conclude by Theorem 6ÿ9 [5] that there is a constantC0>0 depending only onL zandksuch that
kunÿXpÿ1
j0
hj tjkmkC0kfnkk
2:34
and
kunÿukmkC0kfnÿfkk
2:35
Hence,
kukmkXpÿ1
j0
khjkmkjjj C0kfkkC Xpÿ1
j0
jjjkfkk
! 2:36
whereCmaxC0;khjkmk; j0; ::: ; pÿ1 : The estimate (2.36) shows that the isomorphism
P:ÿ0; ::: ; pÿ1; f
!u
is continuous fromCpVkontoVmk:Consequently, by lettingn! 1in (2.33) we obtain
u t Xpÿ1
j0
hj tj Z 1
0 g t;sf sds; t>0
Pÿo; ::: ; pÿ1;f This proves the theorem.
Remarks. 1) If L z admits a real root then we cannot hope to get an
estimate of the form (2.36) even under smooth Data as shows the following example:
du dt 1
t12L2 R \C1 R;
u 0 0
whose unique solution is
u t 0ln 1t; t>0
which is not inL2 0;1whatsoever the value of the constant0:
2) The best constantC in (2.36) is equal to the norm of the isomorphism Pde¢ned by
SupjXpÿ1
j0
hj tj Z 1
0 g t;sh sdsj
where the supremum is taken over allh2Vkand0; ::: ; pÿ12Csuch that Xpÿ1
j0
jjj khkk1
REFERENCES
1. S. Agmon, A. Douglis and L. Nirenberg,Estimate near the boundary for solutions of elliptic P.D.E., Comm. Pure Appl. Math. 12 (1959).
2. S. Mazouzi,Linear elliptic Boundary Value Problem in the half space, Master's Thesis, Car- negie Mellon Univ., PA., USA (1985).
3. R. N. Pederson,Explicit formulae for complementary boundary operators of linear elliptic problems, J. Differential Equations 46 (1982).
4. R. N. Pederson,Green's function and boundary kernels for elliptic boundary value problems, Unpublished class notes.
5. M. Schechter,Modern Methods in Partial Diff. Eq., McGraw Hill (1977).
INSTITUTE OF MATHEMATICS P.O. BOX 12, EL-HADJAR ANNABA
ALGERIA
DEPARTMENT OF MATHEMATICS CARNEGIE MELLON UNIVERSITY PITTSBURGH, PA. 15213 USA