-
Georgian Mathematical JournalVolume 14 (2007), Number 1,
145–167
ON BVPS FOR STRONGLY ELLIPTIC SYSTEMS WITHHIGHER ORDER BOUNDARY
CONDITIONS
FLAVIA LANZARA
Dedicated to the memory of Professor G. Fichera
Abstract. We consider BVPs for strongly elliptic systems of
order 2l withthe boundary conditions of order l + n, n > 0. By
representing the solutionby means of a simple layer potential of
order n and by imposing the boundaryconditions, we get a singular
integral system which is of regular type if andonly if the boundary
operator satisfies the Lopatinskĭı condition and whichcan be
solved if suitable compatibility conditions are satisfied. An
explicitformula for computing the index of the BVP is given.
2000 Mathematics Subject Classification: Primary: 35J55.
Secondary:45F15, 35G15.Key words and phrases: Strongly elliptic
system, simple layer potential,higher order boundary conditions,
Lopatinskĭı condition.
1. Introduction
Let us consider the Dirichlet problem for the Laplace equation
in a boundeddomain A ⊂ R2 with a Lyapunov boundary
∆2u = 0 in A, u = f on ∂A. (1.1)
The classical way of solving this problem is to represent the
solution by meansof a double layer potential
u(z) =
∫
∂A
ϕ(ζ)∂
∂νζlog |z − ζ| dsζ (1.2)
( ∂∂νζ
denotes differentiation along the inward normal at the point ζ
of ∂A; sζis the arc length) where ϕ has to be determined. By
imposing the boundarycondition u = f on ∂A we get a Fredholm
integral equation
−π ϕ(z) +∫
∂A
ϕ(ζ)∂
∂νζlog |z − ζ| dsζ = f(z), z ∈ ∂A,
which has one and only one solution ϕ in C0(∂A) for any
right-hand side f inthe same space. We remark that in this case
(1.2) belongs to C0(A).
The double layer potential approach has been extended, in two
different ways,in [1] and in [4] to higher order elliptic operators
with constant coefficients.
ISSN 1072-947X / $8.00 / c© Heldermann Verlag
www.heldermann.de
-
146 F. LANZARA
A second method of solving (1.1), via integral equations,
consists in repre-senting u by a simple layer potential
u(z) =
∫
∂A
ϕ(ζ) log |z − ζ| dsζ . (1.3)
When we impose the boundary condition u = f on ∂A we get the
followingintegral equation of first kind on ∂A∫
∂A
ϕ(ζ) log |z − ζ| dsζ = f(z) . (1.4)
If we suppose that f ∈ C1+λ(∂A), differentiating (1.4) with
respect to the arclength leads to the integral equation∫
∂A
ϕ(ζ)∂ log |z − ζ|
∂szdsζ =
∂f(z)
∂sz.
Here the integral has to be understood as a singular integral,
due to the strongsingularity of the kernel. As it was shown by
Muskhelishvili, this equation canbe regularized to a Fredholm
equation. The solution of the Dirichlet problemobtained by this
method belongs to C1+λ(A).
In [6] the extension of this method to higher order elliptic
operators has beenconsidered.
Representation (1.3) can also be used for solving boundary value
problemswith boundary conditions of first order
b1(z)ux + b2(z)uy + c(z)u = f on ∂A (1.5)
satisfying the condition b21(z)+b22(z) 6= 0 (equivalent to the
so-called Lopatinskĭı
condition). Of course, in this case f has to satisfy a finite
number of compati-bility conditions. If we impose the boundary
condition (1.5) to the simple layerpotential (1.3), we get a
singular integral equation which is of regular type andwhich can be
solved for any f satisfying the above-mentioned
compatibilityconditions.
Suppose we look for a more regular solution u ∈ C1+n+λ(A), n
> 0, of aboundary value problem with higher order boundary
conditions, e.g.,
∆2u = 0 in A,2∑
j=0
dj(z)∂2u
∂x2−j∂yj+ b1(z)ux + b2(z)uy + c(z)u = f on ∂A.
(1.6)
Problem (1.6) contains, as a very particular case, the Laplace
equation with theboundary condition
∂2u
∂ν2ζ+ p
∂u
∂νζ= f on ∂A,
p(z) ∈ Cλ(∂A), which was solved by means of a different kind of
potentials in[13]. Also, the Ventssel boundary conditions fall into
(1.6). This problem was
-
ON BVPS FOR STRONGLY ELLIPTIC SYSTEMS 147
originally stated in [22] and considered by many authors in the
recent years(see, e.g., [5], [3] and the references therein).
In order to solve problem (1.6) we use the concept of simple
layer potentialof order n. These potentials were introduced in
[10]. There the authors statedthat a solution of class Cn+l+λ(A) of
a general elliptic system of order 2l in theplane could be
represented by means of such potentials. This result was
laterproved in [11].
In the present paper we study the boundary value problem
Eu = 0 in A; Bu = f on ∂A, (1.7)
where E is a q×q matrix differential operator of order 2l with
constant complexcoefficients and B is a matrix differential
operator of order l+n, n > 0, on ∂A. Eis supposed to be strongly
elliptic. The function f belongs to the space Cλ(∂A)and we look for
a solution u in the space Cn+l+λ(A). In order to investigateproblem
(1.7) we represent u by a simple layer potential of order n
uν(z) =
q∑µ=1
l−1∑j=0
∫
∂A
ϕµj(ζ)∂l−1
∂ξl−1−j∂ηjF (n)νµ (z, ζ)dsζ , ν = 1, . . . , q, (1.8)
where {F (n)νµ (z, ζ)} is defined in Section 2.The paper is
organized as follows. In Section 2 we recall the definition of
a fundamental solution of order n, the definition of a simple
layer potential oforder n in the sense of Fichera and Ricci and the
Representation Theorem forsolutions of the strongly elliptic system
Eu = 0 in A. By imposing, on thefunction u given by (1.8), the
boundary conditions we get a system of singularintegral equations.
In Section 3 we prove that this system is of regular type ifand
only if the operator B satisfies the Lopatinskĭı condition with
respect to theoperator E. Our proof rests on some results in the
theory of matrix polynomialsobtained in [23] and [18]. The
regularity of the singular integral system impliesits solvability
provided that suitable compatibility conditions are satisfied.
Thismeans that we obtain the solutions of (1.7) by means of a
simple layer potentialsof order n. In Section 4 an explicit formula
for computing the index of the BVP(1.7) is given.
2. A Simple Layer Potential of order n
Let A be a simply connected open set of the plane of the complex
variablez = x+ iy such that ∂A has a uniformly Hölder continuous
normal field of someexponent λ, (0 < λ 6 1), (∂A ∈ C1+λ).
Let l and q be positive integers and m = lq. In the following we
consider thespaces:
• Cλ(∂A) formed by all complex functions satisfying, on ∂A, a
uniformHölder condition with some Hölder exponent λ : 0 < λ 6
1;
• C l+λ(A) formed by all complex functions which are
continuously differ-entiable up to the order l with respect to the
real variables x and y, and
-
148 F. LANZARA
each partial derivative of order l is uniformly Hölder
continuous in Awith some exponent λ;
• L2(∂A) formed by all measurable complex functions such that
|u|2 isintegrable over ∂A;
• [Cλ(∂A)]q, [C l+λ(A)]q and [L2(∂A)]q formed by all vector
functions Φ =(ϕ1, . . . , ϕq) such that ϕi ∈ Cλ(∂A), ϕi ∈ C l+λ(A)
and ϕi ∈ L2(∂A)respectively;
• H = [L2(∂A)]m × Cσn , σn = q(2ln + l(l+1)
2
), formed by the pairs (Φ; c)
such that Φ ∈ [L2(∂A)]m and c ∈ Cσn . We set < (Φ; c), (Ψ; d)
>=(Φ, Ψ) +
σn∑i=1
cidi, where (Φ, Ψ) =m∑
i=1
∫∂A
ϕiψids.
• B = [Cλ(∂A)]m×Cσn formed by the pairs (Φ; c) such that Φ ∈
[Cλ(∂A)]mand c ∈ Cσn .
We consider systems of linear differential equations E u = 0 in
A where u isa q-vector function and E = {Eµν} is a q × q matrix
such that the elementsEµν are linear differential operators
Eµν = Eµν(
∂
∂x,
∂
∂y
)=
2l∑
k=0
aµνk∂2l
∂x2l−k∂yk, µ, ν = 1, . . . , q,
with constant complex coefficients ak = {aµνk }µ,ν=1,...,q.We
denote by
{Eµν} ={
Eµν
(∂
∂x,
∂
∂y
)}, µ, ν = 1, . . . , q,
the matrix obtained by taking the matrix of the co-factor of the
matrix E and
by transposing it. Let L(w) =2l∑
k=0
akw2l−k be the q × q matrix polynomial of
order 2l associated to the differential operator E and let
Lµν(w) and Lµν(w)be the characteristic polynomials associated to
the differential operators Eµν
and Eµν , respectively , i.e., Lµν(w) = Eµν(w, 1) and Lµν(w) =
Eµν(w, 1). If
L̃(w) = {Lµν(w)}µ,ν=1,...,q we have L(w) L̃(w) = L̃(w) L(w) =
det L(w)I.Hypothesis 1: The operator E is elliptic in the sense of
Petrovskii, i.e.,
det L(w) 6= 0, ∀w ∈ R and det a0 6= 0.Hypothesis 2: The
polynomial det L(w) of degree 2m (m = lq) has m zeroes
with the positive imaginary part and m zeroes with the negative
imaginary part.Let us denote by Γ a rectifiable closed Jordan curve
in the complex plane
w, lying in the half plane =w < 0, such that the bounded
domain which hasΓ as contour contains all the zeroes of det L(w)
which belong to the half plane=w < 0. Let Γ∗ be the symmetric of
Γ with respect to the real axis. Supposethat the bounded domain
which has Γ∗ as contour contains all the zeroes of
-
ON BVPS FOR STRONGLY ELLIPTIC SYSTEMS 149
det L(w) which belong to the half plane =w > 0. Consider the
function
Pα(z, ζ) = −14π2α!
∫
+Γ
[(x− ξ)w + (y − η)]α log[(x− ξ)w + (y − η)]det L(w)
dw
+1
4π2α!
∫
+Γ∗
[(x− ξ)w + (y − η)]α log[(x− ξ)w + (y − η)]det L(w)
dw,
α denoting a non-negative integer. The determination of log[xw +
y] on Γ andΓ∗ is chosen as specified in [10].
Let n be a non-negative integer. Consider the following q × q
matrix:F (n)(z, ζ) = {F (n)µν (z, ζ)}µ,ν=1,...,q ,
where
F (n)µν (z, ζ) = Eµν(∂
∂x,
∂
∂y)P2m+n−2(z, ζ), µ, ν = 1, . . . , q.
We call F (n)(z, ζ) the fundamental solution matrix of order n
for the operator E.Since Eµν is a differential operator of order
less than or equal to 2m− 2l, by
using the same technique as employed in [6, pp. 65–66],
essentially based on thePlemelij formula, we get
Theorem 2.1. If ω ∈ Cλ(∂A), for any z0 ∈ ∂A the following limit
relationholds
limz→z0
∫
∂A
ω(ζ)∂2l+n−1
∂x2l+n−k−1∂ykF (n)νσ (z, ζ) dsζ
=iω(z0)
4π
{ ∫
+Γ
Lνσ(w)w2l+n−1−k
det L(w)(ẋ0w + ẏ0)dw +
∫
+Γ∗
Lνσ(w)w2l+n−1−k
det L(w)(ẋ0w + ẏ0)dw
}
− 14π2
{ ∫
∂A
ω(ζ)dsζ
[ ∫
+Γ
Lνσ(w)w2l+n−1−k
det L(w)[(x0 − ξ)w + (y0 − η)] dw
−∫
+Γ∗
Lνσ(w)w2l+n−1−k
det L(w)[(x0 − ξ)w + (y0 − η)] dw]}
, k = 0, . . . , 2l + n− 1,
as z tends to z0 in the interior of A (the dot denotes
differentiation with respectto the arc length on ∂A, oriented in
the counterclockwise sense). The integralover ∂A must be understood
as a singular Cauchy integral.
If Φ = {ϕσj}, (σ = 1, . . . , q; j = 0, . . . , l − 1) belongs
to [Cλ(∂A)]m, definev0[Φ] = (v01[Φ], . . . , v
0q [Φ]):
v0ν [Φ](z) =
q∑µ=1
l−1∑j=0
∫
∂A
ϕµj(ζ)∂l−1
∂ξl−1−j∂ηjF (n)νµ (z, ζ)dsζ , ν = 1, . . . , q. (2.1)
-
150 F. LANZARA
The function v0[Φ] is called a simple layer potential of order n
and
v0[Φ] ∈ V (A) = {u ∈ [C2l(A)]q ∩ [C l+n+λ(A)]q : Eu = 0 in
A}
(see [10] and, for a complete proof, [11, Theorem 2.5]).
The matrix differential operator E is said to be a strongly
elliptic operator if,for any real w and for every non-zero complex
vector η = (η1, . . . , ηq), we have
<(
1,q∑i,j
Lij(w)ηj η̄i
)> 0; <
(1,q∑i,j
aij0 ηj η̄i
)> 0.
We shall assume E to be strongly elliptic. Then hypotheses 1),
2) are satisfied(see [2, p. 43], [8, p. 101], [14, p. 275], [16, p.
669],[23, p. 425]).
For any integer n > 0, consider the followingProblem (Pn):
find u ∈ [Cn+l(A)]q such that
Eu = 0 in A,
∂
∂s
∂n+l−1
∂xl−1−h∂yh+nu = 0 on ∂A, h = 0, . . . , l − 1.
Theorem 2.2. For a fixed n > 0, there exist
σn = q
[2ln +
l(l + 1)
2
]
linearly independent solutions of problem (Pn).
Before proving Theorem 2.2, we start with a preliminary
result:
Proposition 2.1. Let ω = (ω1, . . . , ωq) be a polynomial
solution of the systemEω = 0 in A. Then there exist polynomials ω̃
= (ω̃1, . . . , ω̃q) such that
Eω̃ = 0 ;∂ω̃
∂y= ω. (2.2)
Proof. If ω is a polynomial solution of the system Eω = 0,
then
χ(x) := E
( y∫
y0
ω(x, η)dη
)(2.3)
is a polynomial in x. Indeed,[E
( y∫
y0
ω(x, η)dη
)]
µ
=
q∑ν=1
{aµν0
y∫
y0
∂2l
∂x2lων(x, η)dη +
2l∑
h=1
aµνh∂2l−1
∂x2l−h∂yh−1ων(x, y)
}
-
ON BVPS FOR STRONGLY ELLIPTIC SYSTEMS 151
=
q∑ν=1
2l∑
h=1
aµνh
{−
y∫
y0
∂2l
∂x2l−h∂ηhων(x, η)dη +
∂2l−1
∂x2l−h∂yh−1ων(x, y)
}
=
q∑ν=1
2l∑
h=1
aµνh∂2l−1
∂x2l−h∂yh−1ων(x, y)
∣∣∣y=y0
, µ = 1, . . . , q.
On the other hand, the condition det a0 6= 0 ensures the
existence of a poly-nomial solution v(x) of the system Ev = χ,
i.e., a0
∂2l
∂x2lv(x) = χ(x). We obtain
that the polynomial
ω̃(x, y) =
y∫
y0
ω(x, η)dη − v(x)
satisfies (2.2). ¤
Proof of Theorem 2.2. First we prove, by induction, that (Pn)
has only polyno-mial solutions. It is known that (P0) has as
solutions q-valued polynomials of
degree 6 l − 1: u(x, y) =0,l−1∑j+k
cj,kxj yk, cj,k ∈ Cq. Since the number of linearly
independent monomials of the form cj,kxjyk, 0 6 j + k 6 l − 1,
is l(l + 1)/2,
it follows that σ0 = ql(l+1)
2. Let us suppose that (Pn) has only polynomial so-
lutions. If u is a solution of (Pn+1), then ω :=∂u∂y
is a polynomial solution of
(Pn). Hence u has the representation
u(x, y) =
y∫
y0
ω(x, η)dη + ϕ(x).
In the proof of Proposition 2.1 we showed that χ in (2.3) is a
polynomial in xand, since Eu = Eϕ + χ = 0, i.e., a0ϕ
(2l)(x) + χ(x) = 0, we obtain that ϕ, too,is a polynomial.
Let us suppose that (Pn) has σn linearly independent solutions
{ω(i)}. If uis a solution of (Pn+1),
∂u∂y
is solution of (Pn) and∂u∂y
=σn∑i=1
ciω(i). It implies
(Proposition 2.1)
u =σn∑i=1
ciω̃(i) + Φ(x), Φ = (ϕ1, . . . , ϕq),
where Φ is a q-valued polynomial because u is a q-valued
polynomial. Moreover,since
Eu = EΦ = a0∂2l
∂x2lΦ(x) = 0 in A
and det a0 6= 0, ϕi(x) is a polynomial such that deg ϕi 6 2l −
1.
-
152 F. LANZARA
Let us denote by {p(h)}h=1,...,2lq the polynomials(1, 0, . . . ,
0)︸ ︷︷ ︸
q
; . . . ; (0, . . . , 0, 1)︸ ︷︷ ︸q
; (x, 0, . . . , 0)︸ ︷︷ ︸q
; . . . (0, . . . , 0, x2l−1)︸ ︷︷ ︸q
.
Then Φ(x) =2lq∑h=1
γhp(h)(x), γh ∈ C. We claim that {ω̃(h)}h=1,...,σn and
{p(h)}h=1,...,2lq are linearly independent. Ifσn∑i=1
ciω̃(i)(x, y) +
2lq∑
h=1
γhp(h)(x) = 0
we haveσn∑i=1
ci∂
∂yω̃(i)(x, y) =
σn∑i=1
ciω(i)(x, y) = 0
which implies c1 = · · · = cσn = 0. Then γ1 = · · · = γ2lq =
0.We deduce that σn+1 = σn + 2lq. ¤Theorem 2.3 (Representation
Theorem). Suppose that the operator E is
strongly elliptic and ∂A is of class C1+λ with λ > 1/2. u
belongs to V (A) if andonly if there exists (Φ; c) ∈ B such
that
u = v0[Φ] + p[c], (2.4)
where v0[Φ] is defined in the right-hand side of (2.1) and
p[c] =σn∑i=1
ciω(i), (2.5)
{ω(j)}j=1,...,σn being linearly independent polynomial solutions
of (Pn).A detailed proof of Theorem 2.3 is contained in [11]. Here
we recall some
results which will be used in the next sections.Consider a
function u ∈ V (A) and set
gh(z) =∂
∂sz
∂l−1
∂xl−1−h∂yh∂n
∂ynu(z), 0 6 h 6 l − 1.
We want to determine Φ in such a way that
∂
∂sz
∂l−1
∂xl−1−h∂yh∂n
∂ynv0[Φ](z) = gh(z), h = 0, . . . , l − 1, z ∈ ∂A. (2.6)
Using Theorem 2.1 the system (2.6) may be rewritten in the
canonical form as
(−1)l−14πgνh(z) =q∑
σ=1
l−1∑j=0
{Aνh,σjϕσj(z) + Bνh,σj
πi
∫
+∂A
ϕσj(ζ)
ζ − z dζ
+
∫
+∂A
ϕσj(ζ)Mνh,σj(z, ζ)dζ}
, ν = 1, . . . , q; h = 0, . . . , l − 1, z ∈ ∂A, (2.7)
-
ON BVPS FOR STRONGLY ELLIPTIC SYSTEMS 153
where gh = {gνh}ν=1,...,q;
Aνh,σj = i[∫
+Γ
Lνσ(w)
det L(w)w2l−(h+j)−2 dw +
∫
+Γ∗
Lνσ(w)
det L(w)w2l−(h+j)−2 dw
];
Bνh,σj = i[∫
+Γ
Lνσ(w)
det L(w)w2l−(h+j)−2 dw −
∫
+Γ∗
Lνσ(w)
det L(w)w2l−(h+j)−2 dw
];
Mνh,σj(z, ζ) = 1π
[ ∫
+Γ
Lνσ(w)H(w, z, ζ)
det L(w)w2l−(h+j)−2 dw
−∫
+Γ∗
Lνσ(w)H(w, z, ζ)
det L(w)w2l−(h+j)−2 dw
]= O (|z − ζ|λ−1) , 0 < λ 6 1.
Here H(w, z, ζ) = K(w, z, ζ)− 1, where
K(w, z, ζ) =
ζ̇(ẋw + ẏ)
x− ξz − ζ w +
y − ηz − ζ
, z 6= ζ,
1, z = ζ
defined for w ∈ Γ ∪ Γ∗ and (z, ζ) ∈ ∂A× ∂A.System (2.7) is a
system of singular integral equations of the kind
g(z) = AΦ(z) + B(SΦ)(z) + (MΦ)(z), z ∈ ∂A, (2.8)where g = {gνh};
A, B and M are the matrices of order lq: {Aνh,σj}, {Bνh,σj}and
{Mνh,σj} (h = 0, . . . , l − 1; ν = 1, . . . , q), (j = 0, . . . ,
l − 1; σ = 1, . . . , q);SΦ is the singular integral operator
(SΦ)(z) =1
πi
∫
+∂A
Φ(ζ)
ζ − z dζ
and MΦ is the Fredholm integral operator
(MΦ)(z) =
∫
+∂A
M(z, ζ) Φ(ζ) dζ.
System (2.8) is of regular type, according to Muskhelishvili
theory (see [11,p.263]). It follows that, for system (2.8), the
same Fredholm theorems of regularintegral systems hold.
Let us consider the associated homogeneous system
0 = AΦ(z) + B(SΦ)(z) + (MΦ)(z), z ∈ ∂A, (2.9)
-
154 F. LANZARA
and the adjoint homogeneous system, which we can write as
follows:
0 =
q∑ν=1
l−1∑
h=0
∫
∂A
ψνh(ζ)∂
∂sζ
∂l−1
∂ξl−1−h∂ηh∂l−1
∂xl−1−j∂yjF (0)νσ (ζ, z)dsζ , (2.10)
j = 0, . . . , l − 1; σ = 1, . . . , q, z ∈ ∂A.System (2.10) has
ql(l + 1)/2 linearly independent solutions. Denoting by
χ the index of system (2.8), i.e., the difference between the
dimensions ofthe spaces of solutions of (2.9) and (2.10),
respectively, we have χ = 0 ([15,p.419],[17, p.14],[9, p.5]).
Moreover, (2.8) has solutions if and only if the datumg is
orthogonal to the eigensolutions of (2.10). These compatibility
conditionsare always satisfied as it is shown in [11]. If Φ is any
solution of (2.8), the simplelayer potential (2.1) gives a solution
of Eu = 0 in A determined up an arbitrarysolution of (Pn).
3. Boundary Value Problem
Let us denote by B = {Bµν}, µ = 1, . . . , m, ν = 1, . . . , q,
a m × q matrixboundary operator. It is taken in the form
B = B0 + B1, B0 = {Bµν0 }, B1 = {Bµν1 },where
Bµν0 =l+n∑
h=0
bµνh (z)∂l+n
∂xl+n−h∂yh; Bµν1 =
l+n−1∑s=0
l+n−1−s∑i=0
bµνis (z)∂l+n−1−s
∂xl+n−1−s−i∂yi.
The functions bµνh (z) and bµνis (z) belong to C
λ(∂A).For a strongly elliptic operator E, we consider the
following
BVP: Given f ∈ [Cλ(∂A)]m, find u ∈ [Cn+l+λ(A)]q ∩ [C2l(A)]q:Eu =
0 in A; Bu = f on ∂A.
Denote by B0(z; w) = {Bµν0 (z; w)} the following lq × q matrix
polynomial:
B0(z; w) =l+n∑
h=0
bh(z)wl+n−h, bh(z) = {bµνh (z)}µ=1,...,lq;ν=1,...,q.
Consider the simple layer potential of order n: v0[Φ] defined in
(2.1).From Theorem 2.1 we obtain, for any z0 ∈ ∂A,
limz→z0
(B0v0[Φ])µ(z)
= (−1)l−1q∑
σ=1
l−1∑j=0
{iϕσj(z0)
4π
[ ∫
+Γ
q∑ν=1
Bµν0 (z0; w)Lνσ(w)wl−1−j
(ẋ0w + ẏ0)
dw
det L(w)
+
∫
+Γ∗
q∑ν=1
Bµν0 (z0; w)Lνσ(w)wl−1−j
(ẋ0w + ẏ0)
dw
det L(w)
]
-
ON BVPS FOR STRONGLY ELLIPTIC SYSTEMS 155
− 14π2
∫
∂A
ϕσj(ζ)dsζ
[ ∫
+Γ
q∑ν=1
Bµν0 (z0; w)Lνσ(w)wl−1−j
[(x0 − ξ)w + (y0 − η)]dw
det L(w)
−∫
+Γ∗
q∑ν=1
Bµν0 (z0; w)Lνσ(w)wl−1−j
[(x0 − ξ)w + (y0 − η)]dw
det L(w)
]}.
If H(w, z, ζ) is the function defined in Section 2, w ∈ Γ∪Γ∗;
(z, ζ) ∈ ∂A×∂A,since
dsζ(x− ξ)w + (y − η) =
ζ̇dζ
(x− ξ)w + (y − η)=
1
ẋw + ẏ
dζ
z − ζ +1
ẋw + ẏ
H(w, z, ζ)
z − ζ ζ̇ dsζ
we have, for z ∈ ∂A,
(B0v0[Φ])µ(z) =
(−1)l−14π
q∑σ=1
l−1∑j=0
{αµ,σj(z)ϕσj(z)
+1
iπβµ,σj(z)
∫
+∂A
ϕσj(ζ)
ζ − z dζ +1
π
∫
∂A
γµ,σj(z, ζ)ϕσj(ζ)dsζ
}, µ = 1, . . . , m,
where
αµ,σj(z) = i
[ ∫
+Γ
q∑ν=1
Bµν0 (z; w)Lνσ(w)wl−1−j
(ẋw + ẏ)
dw
det L(w)
+
∫
+Γ∗
q∑ν=1
Bµν0 (z; w)Lνσ(w)wl−1−j
(ẋw + ẏ)
dw
det L(w)
];
βµ,σj(z) = i
[ ∫
+Γ
q∑ν=1
Bµν0 (z; w)Lνσ(w)wl−1−j
(ẋw + ẏ)
dw
det L(w)
−∫
+Γ∗
q∑ν=1
Bµν0 (z; w)Lνσ(w)wl−1−j
(ẋw + ẏ)
dw
det L(w)
];
γµ,σj(z, ζ) =
∫
+Γ
q∑ν=1
Bµν0 (z; w)Lνσ(w)wl−1−j
det L(w)(ẋw + ẏ)ζ̇H(w, z, ζ)
ζ − z dw
−∫
+Γ∗
q∑ν=1
Bµν0 (z; w)Lνσ(w)wl−1−j
det L(w)(ẋw + ẏ)ζ̇H(w, z, ζ)
ζ − z dw.
-
156 F. LANZARA
For any w ∈ Γ ∪ Γ∗, H(w, z, ζ)[det L(w)(ẋw + ẏ)]−1 is
uniformly Höldercontinuous with respect to (z, ζ) ∈ ∂A× ∂A.
Moreover we have, for z ∈ ∂A,
(B1v0[Φ])µ(z) = (−1)l−1
q∑σ=1
l−1∑j=0
∫
∂A
ϕσj(ζ)
×q∑
ν=1
Bµν1
(z;
∂
∂x,
∂
∂y
)∂l−1
∂xl−1−j∂yjF (n)νσ (z, ζ)dsζ , µ = 1, . . . ,m.
Define the matrices having m rows: µ = 1, . . . , m and m
columns: σ =1, . . . , q; j = 0, . . . , l − 1
A(z) = {αµ,σj(z)}; B(z) = {βµ,σj(z)}; Q(z, ζ) = {Qµ,σj(z,
ζ)},where
Qµ,σj(z, ζ) =1
πγµ,σj(z, ζ) + 4π
∂l−1
∂xl−1−j∂yj
q∑ν=1
Bµν1
(z;
∂
∂x,
∂
∂y
)F (n)νσ (z, ζ).
Qµ,σj(z, ζ) are weakly singular kernels.If Φ = {ϕσj} ∈
[Cλ(∂A)]m, we represent
Bv0[Φ](z) =(−1)l−1
4π
{A(z)Φ(z) + 1
iπB(z)
∫
+∂A
Φ(ζ)
ζ − z dζ
+
∫
∂A
Q(z, ζ)Φ(ζ)dsζ}
, z ∈ ∂A.
Introduce the operator S : [L2(∂A)]m → [L2(∂A)]m
SΦ(z) = A(z)Φ(z) + 1iπB(z)
∫
+∂A
Φ(ζ)
ζ − zdζ +∫
∂A
Q(z, ζ)Φ(ζ)dsζ .
S is a singular integral operator.Let u be a solution of BVP.
For the Representation Theorem 2.3 there exists
(Φ; c) ∈ H such that u = v0[Φ] + p[c] where v0[Φ] is a simple
layer potential oforder n defined in the right-hand side of (2.1)
and p[c] is given in (2.5).
Let L be the operator defined on Cσn as follows:Lc =
(−1)l−14πBp[c].
L : Cσn → [L2(∂A)]m is a continuous operator.By imposing the
boundary conditions to the function u in (2.4), we obtain
T (Φ; c) = (−1)l−14πf, (3.1)where T (Φ; c) = SΦ + Lc. T : H →
[L2(∂A)]m is a continuous operator.
Conversely, let (Φ; c) ∈ H be a solution of (3.1). Since Lc ∈
[Cλ(∂A)]m, wehave SΦ ∈ [Cλ(∂A)]m and this implies Φ ∈ [Cλ(∂A)]m
(see [7, p.264, TheoremXXII]).
We state this result in the following
-
ON BVPS FOR STRONGLY ELLIPTIC SYSTEMS 157
Theorem 3.1. If u is solution of BVP, then u can be represented
by meansof (2.4) where (Φ; c) ∈ H is a solution of the singular
integral system (3.1).
Conversely, if (Φ; c) ∈ H is a solution of (3.1), the function u
defined in (2.4)is a solution of the BVP.
Now we study the solvability of system (3.1) that is equivalent,
by Theorem3.1, to the solvability of the BVP.
According to Muskhelishvili theory, the singular integral
operator S is ofregular type if, ∀z ∈ ∂A, det{A(z)± B(z)} 6= 0.
These conditions can be written in the equivalent way as
det
{ ∫
+Γ
B0(z; w)L−1(w)[I, . . . , wl−1 I]
dw
ẋw + ẏ
}6= 0, z ∈ ∂A;
det
{ ∫
+Γ∗
B0(z; w)L−1(w)[I, . . . , wl−1 I]
dw
ẋw + ẏ
}6= 0, z ∈ ∂A.
Let us factor the polynomial det L(w) as L+(w)L−(w) where L+(w)
[L−(w)]is a polynomial of degree m which has as zeroes all the
zeroes of det L(w) suchthat =w > 0 [=w < 0].
We shall say that the boundary operator B satisfies the
Lopatinskĭı condi-tion (or the complementing boundary condition)
with respect to the differentialoperator E if, for every z ∈ ∂A,
the rows of the m × q matrix polynomialB0(z; w)L̃(w) are linearly
independent mod L
−(w) and mod L+(w) ([2, p.42],[23,p.422]).
This condition means that if, for z ∈ ∂A, there exists d ∈ Cm
such thatdB0(z; w)L̃(w) = L
−(w)M(z; w)
for some 1 × q matrix polynomial M(z; w), then d = 0.
Analogously, if wereplace L−(w) by L+(w).
Theorem 3.2. Let E be strongly elliptic. The singular integral
operator S isof regular type if and only if the boundary operator B
satisfies the Lopatinskĭıcondition with respect to the
differential operator E.
In order to prove Theorem 3.2 we shall use the method introduced
in [23,Ch.10,pp.416-427] and in [18], where the authors reformulate
the Lopatinskĭıcondition in various equivalent forms. The main
tool is the spectral theory ofmatrix polynomials. We start with
some preliminaries and definitions.
In the following we denote by X a q × lq matrix; by T an lq × lq
matrix; byY an lq × q matrix and by I the identity matrix of order
lq. By hypothesis,the contours Γ and Γ∗ do not intersect the roots
of det L(w) = 0 and the zeroesof det L(w) such that =w < 0 [=w
> 0] lie inside Γ [Γ∗]. In the following wealways refer to Γ. In
the same way it is possible to consider Γ∗.
Denote by σ(T ) the spectrum of T , i.e., σ(T ) = {w ∈ C :
det(Iw − T ) = 0}.Definition 3.1. We say that (X, T, Y ) is a
Γ-spectral triple for L(w) if
-
158 F. LANZARA
i) σ(T ) lies inside Γ;ii) L−1(w) − X(Iw − T )−1Y has an
analytic continuation inside Γ as a
matrix function of w;iii) the 2lq × lq matrix
X...
X T 2l−1
is injective;iv) the lq×2lq matrix [Y, . . . , T 2l−1Y ] is
surjective or, equivalently, has rank
lq.
The fact that the matrix (Iw − T )−1, as a function of w, is
analytic, exceptat the points of σ(T ), enables us to obtain some
important results by the useof contour integrals in the complex
w-plane.
Let U(T ) be the class of complex-valued analytic functions f
such that itsdomain is an open set in the complex plane containing
σ(T ).
If D is any bounded (regular) domain such that σ(T ) ⊂ D; D ⊂
dom(f), wedefine the operator matrix f(T ) corresponding to f
as
f(T ) =1
2πi
∫
+∂D
f(w)(Iw − T )−1dw.
The integral has a value independent of the particular choice of
D.Moreover, for any f, g ∈ U(T ),
f(T )g(T ) = g(T )f(T ) =1
2πi
∫
+∂D
f(w)g(w)(Iw − T )−1dw,
where D is any bounded (regular) domain such that σ(T ) ⊂ D; D ⊂
dom(f)∩dom(g) (see [20, p. 289]).
In particular, since σ(T ) lies inside Γ, we have
T j(aT + bI)−1 = (aT + bI)−1T j =1
2πi
∫
+Γ
wj
aw + b(Iw − T )−1dw, (3.2)
j = 0, 1, . . . ; (a, b) ∈ R2 − {(0, 0)}.Proposition 3.1.
Property ii) in Definition 3.1 can be replaced by
ii′)1
2πi
∫
+Γ
wj
aw + bL−1(w)dw = XT j(aT + bI)−1Y,
j = 0, 1, . . . ; (a, b) ∈ R2 − {(0, 0)}.Proof. In view of
(3.2), condition ii′) is equivalent to
1
2πi
∫
+Γ
wj
aw + b
[L−1(w)−X(Iw − T )−1Y ] dw = 0, (3.3)
-
ON BVPS FOR STRONGLY ELLIPTIC SYSTEMS 159
j = 0, 1, . . . ; (a, b) ∈ R2 − {(0, 0)}.If ii) holds true,
(3.3) immediately follows. Conversely, from (3.3) we deduce
that1
2πi
∫
+Γ
wj[L−1(w)−X(Iw − T )−1Y ] dw = 0, j = 0, 1, . . . .
Since σ(T ) lies inside Γ, this is equivalent to ii). ¤In [23,
p. 27] and [18, p. 3115], the following theorem has been
proved:
Theorem 3.3. If Γ is a closed contour not intersecting the roots
of det L(w),then there exists a Γ-spectral triple for L(w).
If (X, T, Y ) is a Γ-spectral triple for L(w), then σ(T )
coincides with the rootsof det L(w) inside Γ, i.e., with the roots
of L−(w) (see [18, p.3116]).
By virtue of Proposition 3.1 the lq × lq matrix∫
+Γ
I...
wl−1 I
L−1(w)[I, . . . , wl−1 I] dw (3.4)
can be written as
X...
X T l−1
[Y, . . . , T l−1Y ]. (3.5)
If E is strongly elliptic, (3.4) is invertible [11, p. 263]
which implies that bothlq × lq matrices in (3.5) are
nonsingular.
Theorem 3.4. Let us suppose that the lq× lq matrix (3.4) is
invertible. Thefollowing statements are equivalent:
a) If (X,T, Y ) is a Γ-spectral triple for L(w) then the lq × lq
matrix
∆B(z) =l+n∑j=0
bj(z)XTl+n−j, z ∈ ∂A,
is invertible, i.e., det ∆B(z) 6= 0, z ∈ ∂A.b) The lq × lq
matrix
1
2πi
∫
+Γ
B0(z; w)L−1(w)[I, . . . , wl−1 I]
dw
ẋw + ẏ, z ∈ ∂A, (3.6)
is invertible.c) The rows of the lq×q matrix B0(z; w)L̃(w) are
linearly independent modulo
L−(w).
Proof. a) ⇔ b) Let (X, T, Y ) be a Γ-spectral triple for L(w).
By formula ii′)matrix (3.6) is equal to(
l+n∑j=0
bj(z)XTl+n−j
)(ẋT + ẏI)−1[Y, . . . , T l−1 Y ] (3.7)
-
160 F. LANZARA
and we know that [Y, . . . , T l−1 Y ] is invertible because
(3.4) is invertible. Hence,since σ(T ) lies inside Γ, matrix (3.7)
has a nonzero determinant if and only if
∆B(z) =l+n∑j=0
bj(z)XTl+n−j has a nonzero determinant.
b) ⇒ c) If d ∈ Cm is such thatdB0(z; w)L̃(w) = L
−(w)M(z; w) (3.8)
for some 1 × q matrix polynomial M(z; w), then dividing (3.8) by
det L(w)(ẋw + ẏ) gives
dB0(z; w)L−1(w)(ẋw + ẏ)−1 =
M(z; w)
L+(w)(ẋw + ẏ)
and the right-hand side is holomorphic inside Γ. Whence
d
∫
+Γ
B0(z; w)L−1(w)
wj
ẋw + ẏdw = 0
and then
d
∫
+Γ
B0(z; w)L−1(w)[I, . . . , wl−1 I]
dw
ẋw + ẏ= 0.
Thus b) implies d = 0.c) ⇒ b) Conversely, let d ∈ Cm such
that
d
∫
+Γ
B0(z; w)L−1(w)[I, . . . , wl−1 I]
dw
ẋw + ẏ= 0.
In view of (3.7), since the lq× lq matrix [Y, . . . , T l−1 Y ]
is invertible we see thatd∆B(z)(ẋT + ẏI)
−1 = 0.
Multiplying both sides of this equation by T kY for k = 0, 1, .
. . and making useof the formula ii′), we obtain
d
∫
+Γ
B0(z; w)L−1(w)
wkdw
ẋw + ẏ= 0, k = 0, 1, . . . .
Hence dB0(z; w)L−1(w) has an analytic continuation inside Γ as a
matrix func-
tion of w. This means that dB0(z; w)L̃(w) vanishes at the roots
of L−(w) = 0.
Hence (3.8) holds for some 1×q matrix polynomial M(z; w) and
then c) impliesd = 0. ¤
If we repeat the same arguments for Γ∗, we deduce Theorem 3.2
from Theo-rem 3.4.
Set
Hµσ(z; w) =
q∑ν=1
Bµν0 (z; w)Lνσ(w), µ = 1, . . . ,m; σ = 1, . . . , q.
Let us regard Hµσ(z; w) as polynomials in the indeterminate
w.
-
ON BVPS FOR STRONGLY ELLIPTIC SYSTEMS 161
For z ∈ ∂A, we reduce the polynomials Hµσ(z; w) mod L−(w) (or
modL+(w)):
Hµσ(z; w) ≡m−1∑
β=0
qσβµ (z)wm−1−β mod L−(w) (mod L+(w)).
Then consider the matrix Q(z) = {qσβµ (z)} having m rows: µ = 1,
. . . ,m and mqcolumns: β = 0, . . . , m−1; σ = 1, . . . , q. A
method for checking the Lopatinskĭıcondition is
Proposition 3.2. The boundary operator B satisfies the
Lopatinskĭı condi-tion with respect to the differential operator E
if and only if, for every z ∈ ∂A,the rank of Q(z) is m.
Proof. Since, for z ∈ ∂A and σ = 1, . . . , q,m∑
µ=1
dµHµσ(z; w)
≡m−1∑
β=0
(m∑
µ=1
dµqσβµ (z)
)wm−1−β mod L−(w) (mod L+(w)), (3.9)
we havem∑
µ=1
dµHµσ(z; w) ≡ 0 if and only ifm∑
µ=1
dµqσβµ (z) = 0, σ = 1, . . . , q; β = 0, . . . ,m− 1.
The above-considered system has nontrivial solutions if and only
if the rank ofQ(z) is less than m. Then, for (3.9), the
Lopatinskĭı condition holds true if andonly if the rank of Q(z) is
m. ¤
Let T ∗ : [L2(∂A)]m → H be the adjoint operator of T . Then, if
(Φ; c) ∈ Hand Ψ ∈ [L2(∂A)]m, we have
〈(Φ; c), T ∗Ψ〉 = (T (Φ; c), Ψ) = (SΦ, Ψ) + (Lc, Ψ)
= (Φ,S∗Ψ) + (−1)l−14πσn∑i=1
ci(Bω(i), Ψ),
S∗ : [L2(∂A)]m → [L2(∂A)]m denoting the adjoint operator of S.
ThenT ∗Ψ = (S∗Ψ; (−1)l−14π(Bω(1), Ψ), . . . , (−1)l−14π(Bω(σn), Ψ))
.
Consider the following homogeneous transposed system of
(3.1):
T ∗Ψ = 0. (3.10)Theorem 3.5. Suppose that the boundary operator
B satisfies the Lopatinskĭı
condition with respect to the differential operator E. Then the
BVP admits asolution if and only if
(f, Ψ) = 0, ∀Ψ ∈ Ker(T ∗) (3.11)
-
162 F. LANZARA
where Ker(T ∗) is the space formed by all the solutions of
(3.10).Proof. If u is a solution of the BVP, then there exists (Φ;
c) ∈ H such that(3.1) holds true (see Theorem 3.1) and
(f, Ψ) =(−1)l−1
4π(T (Φ; c), Ψ) = (−1)
l−1
4π〈(Φ; c), T ∗Ψ〉 = 0, ∀Ψ ∈ Ker(T ∗).
Conversely, suppose that conditions (3.11) hold true. Because of
Theorem 3.1,we will prove Theorem 3.5 if we study the solvability
of system (3.1). Since theoperator S is of regular type, there
exists a reducing operator S ′ : [L2(∂A)]m →[L2(∂A)]m (see [15, p.
419], [21, p. 73]) such that S ′S = I + K where I is theidentity
operator and K is a compact operator of the space [L2(∂A)]m.
Consider the operator T ′Ψ = (S ′Ψ; 0). T ′ is a continuous
operator from[L2(∂A)]m → H and T ′T (Φ; c) = (Φ+KΦ+S ′Lc; 0). T ′
is a reducing operatorfor T because T ′T − I is a compact operator
of the space H. Then (see [7,p. 138],[15, p. 420]) the Fredholm
theorems hold and system (3.1) admits asolution if and only if
conditions (3.11) are satisfied. ¤
4. An Index Formula for the BVP
Let χP be the index of the BVP, i.e., the difference between the
dimensionof the space Ker(BV P ) of eigensolutions of the
homogeneous problem associ-ated to BVP and the maximum number of
linearly independent compatibilityconditions. By Theorem 3.5, we
have
χP = dim Ker(BV P )− dim Ker(T ∗). (4.1)The homogeneous system
associated to (2.6), i.e., (2.9), can be written as
S0Φ = 0. S0 is an operator of the space [L2(∂A)]m and
dim Ker(S0) = q l(l + 1)2
. (4.2)
Φ ∈ Ker(S0) if and only if v0[Φ] is a solution of (Pn). It
follows that there existsc ∈ Cσn such that v0[Φ] = −p[c], where
p[c] is given in (2.5). This shows thatu = v0[Φ] + p[c] is zero and
then representation (2.4) of the solution of BVP isnot unique.
Theorem 4.1. In the hypotheses of the Representation Theorem
2.3, if ubelongs to V (A), there are uniquely determined (Φ⊥; d) ∈
Ker(S0)⊥×Cσn, suchthat
u = v0[Φ⊥] + p[d]. (4.3)
Proof. Let u be in V (A). For the Representation Theorem 2.3,
there exists(Φ; c) ∈ H such that u = v0[Φ] + p[c]. If [L2(∂A)]m =
Ker(S0) ⊕ Ker(S0)⊥, wehave Φ = Φ⊥⊕Φ0 with Φ⊥ ∈ Ker(S0)⊥ and Φ0 ∈
Ker(S0) uniquely determined.Then u = v0[Φ⊥]+v0[Φ0]+p[c]. Since Φ0 ∈
Ker(S0), v0[Φ0] is a solution of (Pn)and there exists b ∈ Cσn such
that v0[Φ0] = p[b]. Hence follows the validity of(4.3) by assuming
d = b + c.
-
ON BVPS FOR STRONGLY ELLIPTIC SYSTEMS 163
Suppose that there exists (Ψ; d) 6= (0, 0), Ψ ∈ Ker(S0)⊥, such
that, in A,v0[Ψ] + p[d] = 0. Then v0[Ψ] is a solution of (Pn). It
follows that Ψ ∈ Ker(S0).Since Ψ ∈ Ker(S0) ∩Ker(S0)⊥, we have Ψ ≡
0. Then d = 0. ¤
Let T̃ = T |Ker(S0)⊥×Cσn .Theorems 3.1 and 4.1 lead to a
correspondence between the solutions of
T̃ (Φ; d) = 0. (4.4)and the solution of the homogeneous problem
associated to BVP. Specifically,if u ∈ Ker(BV P ), then there
exists (Φ; d) ∈ Ker(S0)⊥ × Cσn such that u =v0[Φ] + p[d] and T̃ (Φ;
d) = 0. Conversely, if (Φ; d) ∈ Ker(S0)⊥ × Cσn satisfiesthe
equation (4.4) then v0[Φ] + p[d] belongs to Ker(BV P ).
Moreover, if Ker(T̃ ) denotes the space of eigensolutions of
(4.4), thenProposition 4.1. We have
dim Ker(BV P ) = dim Ker(T̃ ). (4.5)Proof. Let u(1), . . . ,
u(s) ∈ Ker(BV P ) be linearly independent. By Theorem 4.1there
exist {(Φ(i); d(i))} ∈ Ker(S0)⊥×Cσn such that T̃ (Φ(i); d(i)) = 0
and u(i) =v0[Φ(i)] + p[d(i)], i = 1, . . . , s. If {c1, . . . , cs}
are complex constants such that
s∑i=1
ciΦ(i) = 0,
s∑i=1
cid(i) = 0,
we deduce thats∑
i=1
ciu(i) = v0
[ s∑i=1
ciΦ(i)
]+ p
[ s∑i=1
cid(i)
]= 0
which implies c1 = · · · = cs = 0. Hence dim Ker(BV P ) 6 dim
Ker(T̃ ).Conversely, if {(Φ(i); d(i))}i=1,...,s ∈ Ker(S0)⊥ × Cσn
are linearly independent
solutions of (4.4), by assuming u(i) = v0[Φ(i)]+p[d(i)] we have
u(i) ∈ Ker(BV P ).Therefore, if
s∑i=1
ciu(i) = 0, we deduce that
v0[ s∑
i=1
ciΦ(i)
]= p
[−
s∑i=1
cid(i)
]. (4.6)
Hence the function on the left-hand side of (4.6) is a solution
of (Pn) i.e.s∑
i=1
ciΦ(i) ∈ Ker(S0). On the other hand, we have
s∑i=1
ciΦ(i) ∈ Ker(S0)⊥. Then
s∑i=1
ciΦ(i) = 0. From (4.6) we deduce that
s∑i=1
cid(i) = 0 and then c1 = · · · = cs =
0. Hence dim Ker(BV P ) > dim Ker(T̃ ). ¤From (4.1) and
Proposition (4.5) we deduce that:
χP = dim Ker(T̃ )− dim Ker(T ∗).
-
164 F. LANZARA
Proposition 4.2. We have
dim Ker(T ) = dim Ker(T̃ ) + ql(l + 1)2
. (4.7)
Proof. Let s = dim Ker(T̃ ). Denote by {(Φ(i); d(i))}i=1,...,s a
basis for Ker(T̃ ).Since Ker(T̃ ) ⊂ Ker(T ), let us assume {(Ψ(h);
c(h))}h=1,...,r ⊂ Ker(T ) such that{(Φ(i); d(i))}i=1,...,s ∪
{(Ψ(h); c(h))}h=1,...,r is a basis for Ker(T ).
Now we will prove that r = ql(l+1)2
. Suppose that r > ql(l+1)2
. For a fixed h =
1, . . . , ql(l+1)2
+ 1, consider the function (Ψ(h); c(h)) and the corresponding
u(h) =
v0[Ψ(h)] + p[c(h)]. Since u(h) ∈ V (A) and T (Ψ(h); c(h)) = 0,
u(h) ∈ Ker(BV P )and there exists (χ(h); b(h)) ∈ Ker(S0)⊥ × Cσn
such that u(h) = v0[χ(h)] + p[b(h)]with (χ(h); b(h)) ∈ Ker(T̃ ). It
follows that
(χ(h); b(h)) =s∑
i=1
λ(h)i (Φ
(i); d(i)), {λ(h)1 , . . . , λ(h)s } 6= {0, . . . , 0}.
Then
v0[Ψ(h)] + p[c(h)] = v0[χ(h)] + p[b(h)] =s∑
i=1
λ(h)i v
0[Φ(i)] +s∑
i=1
λ(h)i p[d
(i)]
and therefore
p
[ s∑i=1
λ(h)i d
(i) − c(h)]
= v0[Ψ(h) −
s∑i=1
λ(h)i Φ
(i)
]. (4.8)
The function on the left-hand side of (4.8) is a solution of
(Pn) and
Ψ(h) −s∑
i=1
λ(h)i Φ
(i) ∈ Ker(S0), h = 1, . . . , ql(l + 1)2
+ 1.
In view of (4.2), there exists µ = {µ1, . . . , µ ql(l+1)2
+1} 6= 0 such that
ql(l+1)2
+1∑
h=1
µh
(Ψ(h) −
s∑i=1
λ(h)i Φ
(i)
)= 0. (4.9)
(4.8) and (4.9) lead to
ql(l+1)2
+1∑
h=1
µh
(c(h) −
s∑i=1
λ(h)i d
(i)
)= 0. (4.10)
From (4.9) and (4.10) we deduce that {(Ψ(j); c(j))}j=1,...,
ql(l+1)2
+1and
{(Φ(i); d(i))}i=1,...,s are linearly dependent. This is
impossible. It follows thatr 6 ql(l+1)
2.
Let {Ψ(h)}h=1,...,
ql(l+1)2
∈ Ker(S0) be linearly independent functions. Sincev0[Ψ(h)] is
solution of (Pn), there exist b
(h) ∈ Cσn such that v0[Ψ(h)] = −p[b(h)].This implies that (Ψ(h);
b(h)) ∈ Ker(T ), h = 1, . . . , ql(l+1)
2.
-
ON BVPS FOR STRONGLY ELLIPTIC SYSTEMS 165
We will prove that (Ψ(h); b(h))h=1,...,
ql(l+1)2
and (Φ(j); d(j))j=1,...,s are linearly
independent functions in Ker(T ). Indeed, suppose thats∑
j=1
λj(Φ(j); d(j)) +
ql(l+1)2∑
h=1
µh(Ψ(h); b(h)) = 0
Since Ψ(h) ∈ Ker(S0) and Φ(j) ∈ Ker(S0)⊥ we deduce thats∑
j=1
λjΦ(j) =
ql(l+1)2∑
h=1
µhΨ(h) = 0,
which implies µh = 0, h = 1, . . . ,ql(l+1)
2. Then
s∑j=1
λj(Φ(j); d(j)) = 0
which implies λj = 0, j = 1, . . . , s, because {(Φ(j);
d(j))}j=1,...,s ⊂ Ker(T̃ ) arelinearly independent.
As a consequence we have that
dim Ker(T ) > s + ql(l + 1)2
.
Thus r > ql(l + 1)2
and the theorem is proved. ¤
From (4.1), (4.5) and (4.7) we deduce that
χP = χT − ql(l + 1)2
, (4.11)
where χT denotes the index of the operator T , i.e. χT = dim
Ker(T ) −dim Ker(T ∗) (see [21, p. 63]).
Let us denote by T1 and T2 the following operators of the space
H →[L2(∂A)]m:
T1(Φ; c) = SΦ; T2(Φ; c) = Lc.Then T = T1 + T2. Since T2 is a
compact operator, we have (see [15, p. 118])
χT = χT1 (4.12)
The following two equations
T1(Φ; c) = f, f ∈ [L2(∂A)]m, (4.13)and
SΦ = f, f ∈ [L2(∂A)]m, (4.14)are equivalent in the sense that
they are both solvable or both unsolvable. SinceS is an operator of
regular type, the necessary and sufficient conditions for
thesolvability of equation (4.14), and hence of (4.13), are (f, Ψ)
= 0, ∀Ψ ∈ Ker(S∗).Hence
dim Ker(S∗) = dim Ker(T ∗1 ). (4.15)
-
166 F. LANZARA
Set p = dim Ker(S). Let {Φ(i)}i=1,...,p be a basis for
Ker(S).Then {(Φ(i); 0)}i=1,...,p belong to Ker(T1) and, together
with
(0; 1, 0, . . . , 0︸ ︷︷ ︸σn−pla
), (0; 0, 1, . . . , 0︸ ︷︷ ︸σn−pla
), . . . , (0; 0, 0, . . . , 1︸ ︷︷ ︸σn−pla
) ,
form a basis for Ker(T1). Thusdim Ker(T1) = dim Ker(S) + σn.
(4.16)
From (4.12), (4.15) and (4.16) we deduce that
χT = χS + σn. (4.17)
A final form for χP follows from (4.17), (4.11) and Theorem
2.2:
Theorem 4.2. The index of the BVP is given by
χP = χS + 2lnq.
Since S is of regular type, its index is given by the
Muskhelishvili’s formula
χS =1
2πi
[log
det (αµ,σj(z)− βµ,σj(z))det (αµ,σj(z) + βµ,σj(z))
]
+∂A
,
where [ ]+∂A denotes the jump of the function between brackets
after a coun-terclockwise tour along ∂A. This makes χP explicitly
computable.
References
1. S. Agmon, Multiple layer potentials and the Dirichlet problem
for higher order ellipticequations in the plane. I. Comm. Pure
Appl. Math 10(1957), 179–239.
2. S. Agmon, A. Douglis, and L. Nirenberg, Estimates near the
boundary for solutionsof elliptic partial differential equations
satisfying general boundary conditions. II. Comm.Pure Appl. Math.
17(1964), 35–92.
3. D. E. Apushkinskaya and A. I. Nazarov, A survey of results on
nonlinear Venttselproblems. Appl. Math. 45(2000), No. 1, 69–80.
4. A. Cialdea, A multiple-layer potential theory alternative to
Agmon’s. Arch. RationalMech. Anal. 120(1992), No. 4, 345–362.
5. A. Favini, G. R. Goldstein, J. A. Goldstein, and S.
Romanelli, C0-semigroupsgenerated by second order differential
operators with general Wentzell boundary condi-tions. Proc. Amer.
Math. Soc. 128(2000), No. 7, 1981–1989.
6. G. Fichera, Linear elliptic equations of higher order in two
independent variables andsingular integral equations, with
applications to anistropic inhomogeneous elasticity.Partical
differential equations and continuum mechanics, 55–80. Univ. of
WisconsinPress, Madison, Wis., 1961.
7. G. Fichera, Operatori di Riesz–Fredholm. Operatori
riducibili. Equazioni integralisingolari. Istituto Matematico “G.
Castelnuovo”, Università di Roma “La Sapienza”,1964.
8. G. Fichera, Linear elliptic differential systems and
eigenvalue problems. Lecture Notesin Mathematics, 8.
Springer-Verlag, Berlin–New York, 1965.
9. G. Fichera, Simple layer potentials for elliptic equations of
higher order. In: BoundaryIntegral Methods. Proceedings of the
IABEM Symposium, L.Morino, R. Piva (eds.),Roma, 1990. Springer
Verlag, 1990, 1–14.
-
ON BVPS FOR STRONGLY ELLIPTIC SYSTEMS 167
10. G. Fichera and P. E. Ricci, The single layer potential
approach in the theory ofboundary value problems for elliptic
equations. Function theoretic methods for partialdifferential
equations (Proc. Internat. Sympos., Darmstadt, 1976), 39–50.
Lecture Notesin Math., Vol. 561, Springer, Berlin, 1976.
11. F. Lanzara, A representation theorem for solutions of higher
order strongly ellipticsystems. Homage to Gaetano Fichera, 233–271,
Quad. Mat., 7, Dept. Math., SecondaUniv. Napoli, Caserta, 2000.
12. Ya.B. Lopatinskĭı, On a method of reducing boundary
problems for a system of dif-ferential equations of elliptic type
to regular integral equations. (Russian) Ukrain. Mat.Ž. 5(1953),
123–151; English transl.: Amer. Math. Soc. Transl. 2 89(1970),
149–183.
13. V. V. Luk’yanov and A. I. Nazarov, Solution of the Venttsel’
problem for theLaplace and the Helmholtz equations by means of
iterated potentials. (Russian) Zap.Nauchn. Sem. S.-Peterburg.
Otdel. Mat. Inst. Steklov. (POMI ) 250(1998), Mat. Vopr.Teor.
Rasprostr. Voln. 27, 203–218, 337–338; English transl.: J. Math.
Sci. (New York)102(2000), No. 4, 4265–4274.
14. C. Miranda, Partial differential equations of elliptic type.
Second revised edition.(Translated from the Italian) Ergebnisse der
Mathematik und ihrer Grenzgebiete, Band2. Springer-Verlag, New
York–Berlin, 1970.
15. N. I. Muskhelishvili, Singular integral equations. Boundary
problems of functionstheory and their applications to mathematical
physics. (Translation from the Russian)Wolters–Noordhoff
Publishing, Groningen, 1972.
16. L. Nirenberg, Remarks on strongly elliptic partial
differential equations. Comm. PureAppl. Math. 8(1955), 649–675.
17. P. E. Ricci, Sui potenziali di semplice strato per le
equazioni ellittiche di ordine supe-riore in due variabili. Rend.
Mat. (6) 7(1974), 1–39.
18. B. Rowley, Matrix polynomials and the index problem for
elliptic systems. Trans.Amer. Math. Soc. 349(1997), No. 8,
3105–3148.
19. B. Rowley, An index formula for elliptic systems in the
plane. Trans. Amer. Math.Soc. 349(1997), No. 8, 3149–3179.
20. A. E. Taylor, Introduction to functional analysis. John
Wiley & Sons, Inc., New York;Chapman & Hall, Ltd., London,
1958.
21. N. P. Vekua, Systems of singular integral equations.
(Translated from the Russian) P.Noordhoff, Ltd., Groningen,
1967.
22. A. D. Venttsel’, On boundary conditions for
multi-dimensional diffusion processes.Theor. Probability Appl.
4(1959), 164–177.
23. J. T. Wloka, B. Rowley, and B. Lawruk, Boundary value
problems for ellipticsystems. Cambridge University Press,
Cambridge, 1995.
(Received 13.07.2006)
Author’s address:
Dipartimento di MatematicaUniversità degli studi di Roma “La
Sapienza”,Piazzale Aldo Moro 2, 00185 RomaItalyE-mail:
[email protected]