POLONICI MATHEMATICI LXXIV (2000)
Harmonic functions in a cylinder with normal derivatives vanishing
on the boundary
by Ikuko Miyamoto and Hidenobu Yoshida (Chiba)
Dedicated to the memory of Professor Bogdan Ziemian
Abstract. A harmonic function in a cylinder with the normal derivative vanishing on the boundary is expanded into an infinite sum of certain fundamental harmonic functions.
The growth condition under which it is reduced to a finite sum of them is given.
1. Introduction. Let R
n(n ≥ 2) denote the n-dimensional Euclidean space. The solution of the Neumann problem for an infinite cylinder
Γ
n(D) = {(X, y) ∈ R
n: X ∈ D, −∞ < y < ∞},
with D a bounded domain of R
n−1, is not unique, because we can add to each solution harmonic functions in Γ
n(D) with normal derivatives vanishing on the boundary. Hence, to classify general solutions we need to characterize such functions. If D = (0, π) and Γ
n(D) is the strip
H = {(x, y) ∈ R
2: 0 < x < π, −∞ < y < ∞},
then by applying a result of Widder [6, Theorem 2] which characterizes a harmonic function in H vanishing continuously on the boundary ∂H of H, we can obtain the following result:
Theorem A. Let h(x, y) be a harmonic function in H such that ∂h/∂x vanishes continuously on ∂H. Then
h(x, y) = A
0y + B
0+
∞
X
k=1
(A
ke
ky+ B
ke
−ky) cos kx,
where the series converges for all x and y, and A
0, B
0, A
1, B
1, A
2, B
2, . . . are
2000 Mathematics Subject Classification: Primary 31B20.
Key words and phrases : Neumann problem, harmonic functions, cylinder.
[229]
constants such that
A
ke
ky+ B
ke
−ky= 2 π
π
\
0
h(x, y)kx dx (k = 1, 2, . . .).
Although this theorem is easily proved, we cannot proceed similarly in the case where Γ
n(D) is a cylinder in R
n(n ≥ 3). This kind of problem was originally treated by Bouligand [1] in 1914.
Theorem B (Bouligand [1, p. 195]). Let h(X, y) be a harmonic function in Γ
n(D) such that the normal derivative of h vanishes continuously on the boundary ∂Γ
n(D) of Γ
n(D). If h(X, y) tends to zero as |y| → ∞, then h(X, y) is identically zero in Γ
n(D).
In this paper we shall prove a cylindrical version of Theorem A (The- orem). As corollaries we shall obtain two results generalizing Theorem B (Corollaries 1 and 2).
2. Preliminaries. Let D be a bounded domain in R
n−1(n ≥ 3) having a sufficiently smooth boundary ∂D. For example, D can be a C
2,α-domain (0 < α < 1) in R
n−1bounded by a finite number of mutually disjoint closed hypersurfaces (see Gilbarg and Trudinger [3, pp. 88–89] for the definition of C
2,α-domain). Consider the Neumann problem
(2.1) (∆
n−1+ µ)ϕ(X) = 0
for any X = (x
1, . . . , x
n−1) ∈ D,
(2.2) lim
X→X′, X∈D
(∇
n−1ϕ(X), ν(X
′)) = 0 for any X
′∈ ∂D, where
∆
n−1= ∂
2∂x
21+ . . . + ∂
2∂x
2n−1, ∇
n−1=
∂
∂x
1, . . . , ∂
∂x
n−1and ν(X
′) is the outer unit normal vector at X
′∈ ∂D.
Let {µ
k(D)}
∞k=0be the non-decreasing sequence of non-negative eigen- values of this Neumann problem. In this sequence we write µ
k(D) the num- ber of times equal to the dimension of the corresponding eigenspace. If the normalized eigenfunction corresponding to µ
k(D) is denoted by ϕ
k(D)(X), the set of consecutive eigenfunctions corresponding to the same value of µ
k(D) in the sequence {ϕ
k(D)(X)}
∞k=0forms an orthonormal basis for the eigenspace of the eigenvalue µ
k(D). It is evident that µ
0(D) = 0 and
ϕ
0(D)(X) = |D|
−1/2(X ∈ D), |D| =
\
D
dX.
In the following we shall denote {µ
k(D)}
∞k=0and {ϕ
k(D)(X)}
∞k=0by
{µ(k)}
∞k=0and {ϕ
k(X)}
∞k=0respectively, without specifying D. For each D
there is a sequence {k
i} of non-negative integers such that k
0= 0, k
1= 1, µ(k
i) < µ(k
i+1),
µ(k
i) = µ(k
i+ 1) = µ(k
i+ 2) = . . . = µ(k
i+1− 1)
and {ϕ
ki, ϕ
ki+1, . . . , ϕ
ki+1−1} is an orthonormal basis for the eigenspace of the eigenvalue µ(k
i) (i = 0, 1, 2, . . .). Since D has a sufficiently smooth boundary, we know that
µ(k) ∼ A(D, n)k
2/(n−1)(k → ∞) and
X
µ(k)≤t
{ϕ
k(X)}
2∼ B(D, n)t
(n−1)/2(t → ∞)
uniformly with respect to X ∈ D, where A(D, n) and B(D, n) are constants depending on D and n (e.g. see Carleman [2], Minakshisundaram and Plei- jel [4], Weyl [5]). Hence there exist positive constants M
1, M
2such that
M
1k
2/(n−1)≤ µ(k) (k = 1, 2, . . .) and
|ϕ
k(X)| ≤ M
2k
1/2(X ∈ D, k = 1, 2, . . .).
3. Statement of our results. The gradient of a function f (P ) defined on Γ
n(D) is
∇
nf (P ) = ∂f
∂x
1(P ), . . . , ∂f
∂x
n−1(P ), ∂f
∂y (P )
(P = (x
1, . . . , x
n−1, y) ∈ Γ
n(D)). We first remark that
I
k(P ) = e √
µ(k)yϕ
k(X) and J
k(P ) = e
−√
µ(k)yϕ
k(X) (P = (X, y) ∈ Γ
n(D)) are harmonic functions on Γ
n(D) satisfying
P →Q, P ∈Γ
lim
n(D)(∇
nI
k(P ), ν(Q)) = 0 and
P →Q, P ∈Γ
lim
n(D)(∇
nJ
k(P ), ν(Q)) = 0, where ν(Q) is the outer unit normal vector at Q ∈ ∂Γ
n(D).
Theorem. Let h(P ) be a harmonic function on Γ
n(D) satisfying
(3.1) lim
P →Q, P ∈Γn(D)
(∇
nh(P ), ν(Q)) = 0 for any Q ∈ ∂Γ
n(D). Then
h(P ) = A
0y + B
0+
∞
X
k=1
(A
kI
k(P ) + B
kJ
k(P ))
for any P = (X, y) ∈ Γ
n(D), where the series converges uniformly and absolutely on any compact subset of the closure Γ
n(D) of Γ
n(D), and A
k, B
k(k = 0, 1, 2, . . .) are constants such that (3.2) A
ke √
µ(k) y
+ B
ke
−√
µ(k) y
=
\
D
h(X, y)ϕ
k(X) dX (k = 1, 2, . . .).
Corollary 1. Let p and q be non-negative integers. If h(P ) is a har- monic function on Γ
n(D) satisfying (3.1) and
(3.3) lim
y→∞
e
−√
µ(kp+1) y
M
h(y) = 0, lim
y→−∞
e √
µ(kq+1) y
M
h(y) = 0, where
M
h(y) = sup
X∈D
|h(X, y)| (−∞ < y < ∞), then
h(P ) = A
0y + B
0+
kp+1−1
X
k=1
A
kI
k(P ) +
kq+1−1
X
k=1
B
kJ
k(P )
for any P = (X, y) ∈ Γ
n(D), where A
k(k = 0, 1, . . . , k
p+1− 1) and B
k(k = 0, 1, . . . , k
q+1− 1) are constants.
Corollary 2. Let h(P ) be a harmonic function on Γ
n(D) satisfying (3.1) and
M
h(y) = o(e √
µ(1)|y|) (|y| → ∞).
Then h(P ) = A
0y + B
0for any P = (X, y) ∈ Γ
n(D), where A
0and B
0are constants.
4. Proofs of Theorem and Corollaries 1, 2. Let f (X, y) be a function on Γ
n(D). The function c
k(f, y) of y (−∞ < y < ∞) defined by
c
k(f, y) =
\
D
f (X, y)ϕ
k(X) dX
is simply denoted by c
k(y) in the following, without specifying f .
Lemma 1. Let h(P ) be a harmonic function on Γ
n(D) satisfying (3.1).
Then
c
0(y) = A
0y + B
0, (4.1)
c
k(y) = A
ke √
µ(k) y
+ B
ke
−√
µ(k) y
(k = 1, 2, . . .) (4.2)
with constants A
k, B
k(k ≥ 0) and
c
k(y) = {e √
µ(k) (y−y2)
− e √
µ(k) (y2−y)
}c
k(y
1) e √
µ(k) (y1−y2)
− e √
µ(k) (y2−y1)
(4.3)
+ {e √
µ(k) (y1−y)
− e √
µ(k) (y−y1)
}c
k(y
2) e √
µ(k) (y1−y2)
− e √
µ(k) (y2−y1)
for any y
1and y
2, −∞ < y
1< y
2< ∞ (k = 1, 2, 3, . . .).
P r o o f. First of all, we remark that h ∈ C
2(Γ
n(D)) (Gilbarg and Trudin- ger [3, p. 124]). Since
\
D
(∆
n−1h(X, y))ϕ
k(X) dX =
\
D
h(X, y)(∆
n−1ϕ
k(X)) dX (−∞ < y < ∞), from Green’s identity, (2.2) and (3.1), we have
∂
2c
k(y)
∂y
2=
\
D
∂
2h(X, y)
∂y
2ϕ
k(X) dX = −
\
D
∆
n−1h(X, y)ϕ
k(X) dX
= −
\
D
h(X, y)(∆
n−1ϕ
k(X)) dX
= µ(k)
\
D
h(X, y)ϕ
k(X) dX = µ(k)c
k(y)
from (2.1) (k = 0, 1, 2, . . .). With constants A
kand B
k(k = 0, 1, 2,. . .) these give
c
0(y) = A
0y + B
0and
c
k(y) = A
ke √
µ(k) y
+ B
ke
−√
µ(k) y
(k = 1, 2, . . .),
which are (4.1) and (4.2). When we solve for A
kand B
kthe equations c
k(y
i) = A
ke √
µ(k) yi
+ B
ke
−√
µ(k) yi
(i = 1, 2), we immediately obtain (4.3).
Remark . From (4.2) we have, for k = 1, 2, . . .
y→∞
lim c
k(y)e
−√
µ(k) y= A
kand lim
y→−∞
c
k(y)e √
µ(k) y= B
k.
Lemma 2. Let h(P ) be a harmonic function on Γ
n(D) satisfying (3.1).
Let y be any number and y
1, y
2be two any numbers satisfying −∞ < y
1<
y − 1, y + 1 < y
2< ∞. For two non-negative integers p and q,
∞
X
k=kp+q+1
|c
k(y)| · |ϕ
k(X)| ≤ L(p)M
h(y
1) + L(q)M
h(y
2),
where
L(j) = M
22|D|
∞
X
k=kj+1
k exp(− pM
1k
1/(n−1)).
P r o o f. From Lemma 1, we see that
c
k(y) = exp{−pµ(k)(y − y
1)} 1 − exp{2pµ(k)(y − y
2)}
1 − exp{2 pµ(k)(y
1− y
2)} c
k(y
1) + exp{pµ(k)(y − y
2)} 1 − exp{2 pµ(k)(y
1− y)}
1 − exp{2 pµ(k)(y
1− y
2)} c
k(y
2).
Hence (4.4)
∞
X
k=kp+q+1
|c
k(y)| · |ϕ
k(X)| ≤ I
1+ I
2, where
I
1=
∞
X
k=kp+1
exp{−pµ(k)(y − y
1)}|c
k(y
1)| · |ϕ
k(X)|
I
2=
∞
X
k=kq+1
exp{− pµ(k)(y
2− y)}|c
k(y
2)||ϕ
k(X)|.
For I
1, we have
I
1≤ M
22|D|M
h(y
1)
∞
X
k=kp+1
k exp(− pµ(k)) (4.5)
≤ M
22|D|M
h(y
1)
∞
X
k=kp+1
k exp(− pM
1k
1/(n−1)), because y − y
1> 1.
For I
2, we also have
(4.6) I
2≤ M
22|D|M
h(y
2)
∞
X
k=kq+1
k exp(− pM
1k
1/(n−1)).
Finally (4.4)–(4.6) give the conclusion of the lemma.
Proof of Theorem. Take any compact set T ⊂ Γ
n(D) and two numbers y
1, y
2satisfying
max{y : (X, y) ∈ T } + 1 < y
2, min{y : (X, y) ∈ T } − 1 > y
1.
Let (X, y) be any point in T . Since c
k(y) is the Fourier coefficient of the
function h(X, y) of X with respect to the orthonormal sequence {ϕ
k(X)}
∞k=0,
we have
h(X, y) =
∞
X
k=0