GRADIENT OF SOLUTION OF THE POISSON EQUATION

IN THE UNIT BALL AND RELATED OPERATORS

arXiv:1702.00929v1 [math.CV] 3 Feb 2017

DAVID KALAJ AND DJORDJIJE VUJADINOVIĆ

Abstract. In this paper we determine the L1 → L1 and L∞ → L∞

norms of an integral operator N related to the gradient of the solution
of Poisson equation in the unit ball with vanishing boundary data in
sense of distributions.

1. Introduction and Notation

We denote by B = B n and S = S n−1 the unit ball and the unit sphere
in Rn respectively. We will assume that n > 2 (the case n = 2 has been
already treated in [10, 11]). By the vector norm |·| we consider the standard
1
Euclidean distance |x| = ( ni=1 x2i ) 2 .
P
The norm of an operator T : X → Y defined on the normed space X with
image in the normed space Y is defined as
kT k = sup{kT xk : kxk = 1}.
Let G be the Green function, i.e., the function
 
1 1
G(x, y) = cn − ,
|x − y|n−2 [x, y]n−2
where
1
cn = ,
(n − 2)ωn−1
where ωn−1 is the Hausdorff measure of S n−1 and
[x, y] := |x|y| − y/|y|| = |y|x| − x/|x||.
The Poisson kernel P is defined
1 − |x|2
P (x, η) = , |x| < 1, η ∈ S n−1 .
|x − η|n
We are going to consider the Poisson equation

△u(x) = g, x ∈ Ω
(1.1)
u|∂Ω = f
where f : S n−1 → R is bounded integrable function on the unit sphere S n−1 ,
and g : B n → R is a continuous function.
2000 Mathematics Subject Classification. Primary 35J05; Secondary 47G10.
Key words and phrases. Möbius transformations, Poisson equation, Newtonian poten-
tial, Cauchy transform, Bessel function.
1
2 DAVID KALAJ AND DJORDJIJE VUJADINOVIĆ

The solution of the equation in the sense of distributions is given by

(1.2) Z Z
u(x) = P [f ](x) − G[g](x) := P (x, η)f (η)dσ(η) − G(x, y)g(y)dy,
S n−1 Bn
|x| < 1. Here dσ is the normalized Lebesgue n − 1 dimensional measure of
the unit sphere S = S n−1 .
Our main focus of observation is related to the special case of Poisson
equation with Dirichlet boundary condition

△u(x) = g, x ∈ Ω
(1.3)
u|∂Ω = 0
where g ∈ L∞ (B n ). The weak solution is then given by
Z
(1.4) u(x) = −G[g](x) = − G(x, y)g(y)dy, |x| < 1.
B
The problem of estimating the norm of the operator G in case of various
p
L − spaces was established by both authors in [12].
Since
x−y |y|2 x − y
 
∇x G(x, y) = cn (2 − n) − ,
|x − y|n [x, y]n
this naturally induces the differential operator related to Poisson equation
x−y |y|2 x − y
Z  
1
(1.5) D[g](x) := ∇u(x) = − g(y)dy.
ωn−1 B |x − y|n [x, y]n
Related to the problem of estimating the norm of the operator D, we are
going to observe the operator N : L∞ (B n ) → L∞ (B n ) defined by
|y|2 x − y 
Z  
1  x−y
(1.6) N [f ](x) =  − f (y)dy.
ωn−1 B  |x − y|n [x, y]n 
The main goal of our paper is related to estimating various norms of the
integral operator and N . Then we use those results to obtain some norm
estimates of the operator D. The compressive study of this problem for n = 2
has been done by the first author in [10, 11] and by Dostanić in [5, 6]. For
related results we refer to the papers [3, 4].

1.1. Gauss hypergeometric function. Through the paper we will often

use the properties of the hypergeometric functions. First of all, the hy-
pergeometric function F (a, b, c, t) = 2 F1 (a, b; c; t) is defined by the series
expansion
∞
X (a)n (b)n n
t , for |t| < 1,
n!(c)n
n=0
and by the continuation elsewhere. Here (a)n denotes shifted factorial, i.e.
(a)n = a(a + 1)...(a + n − 1) and a is any real number.
The following identity will be used in proving the main results of this paper:
 GRADIENT OF SOLUTION TO POISSON EQUATION AND RELATED OPERATORS 3

π
sinµ−1 t
   
µ 1 1−µ 1+µ 2
Z
(1.7) dt = B , F ν, ν + ; ,r
0 (1 + r 2 − 2r cos t)ν 2 2 2 2
(see, e.g., Prudnikov, Brychkov and Marichev [14, 2.5.16(43)]), where B is
the beta function.
By using the Chebychev’s inequality one can easily obtain the following
inequality for Gamma function (see [8]).
Proposition 1.1. Let m, p and k be real numbers with m, p > 0 and p >
k > −m: If
(1.8) k(p − m − k) > 0(6 0)
then we have
(1.9) Γ(p)Γ(m) > (6)Γ(p − k)Γ(m + k).
1.2. Möbius transformations of the unit ball. The set of isometries of
the hyperbolic unit ball B n is a Kleinian subgroup of all Möbius transforma-
n
tions of the extended space R onto itself denoted by Conf (B) = Isom(B).
We refer to the Ahlfors’ book [2] for detailed survey to this class of important
mappings.
(1 − |x|2 )(y − x) − |y − x|2 x
(1.10) Tx y = ,
[x, y]2
and
 
x − y 
(1.11) |Tx y| = 
 
[x, y] 
n
1 − |x|2

(1.12) dy = dz.
[z, −x]2

2. The L∞ norm of the operator N

In this section we are going to find the norm of the operator N , defined
in (1.6), and by using this we estimate the norm of operator D.
Theorem 2.1. Let N : L∞ (B) → L∞ (B) be the operator defined in (1.6).
Then
2nπ n/2
kN kL∞ →L∞ = .
(n + 1)Γ(n/2)
Proof. At the beginning, let us note that
|y|2 x − y 
Z  
 x−y
kN kL∞ →L∞ = sup 
n
− dy.
x∈B B |x − y|
 [x, y]n 
So, we need to find
sup K(x),
x∈B
4 DAVID KALAJ AND DJORDJIJE VUJADINOVIĆ

where Z  2x − y 

 x−y |y|
K(x) = 
n
−  dy.
B |x − y|
 [x, y]n 
Now we are going to use the change of variables y = T−x z(Tx y = z), where
T−x : B → B is the Möbius transform defined by
(1 − |x|2 )(y + x) + x|z + x|2
T−x (z) = .
[z, −x]2
 
 x−y 
Now we use the following relations |Tx (y)| =  [x,y]  and
(1 − |x|2 )(−x|z|2 − z) 1 − |x|2
x − T−x (z) = , |x − T−x (z)| = |z| .
[z, −x]2 [z, −x]
We have that
|y|2 x − y 
 
 x−y
 |x − y|n − [x, y]n 


 x − y n 
   
1  2
= (x − y) − (|y| x − y)    
|x − y|n  [x, y]  
1 (x − y) − (|y|2 x − y)|z|n 
 
= n
|x − y|
1 (x − T−x z) − (|T−x z|2 x − T−x z)|z|n 
 
(2.1) = n
|x − T−x z|
[z, −x]n  (1 − |x|2 )(−x|z|2 − z) (1 − |x|2 )(z + x) n 
 
=  + |z| 
(1 − |x|2 )n |z|n  [z, −x]2 [z, −x]2
[z, −x]n (1 − |x|2 )  n−1

= |z| (−x|z| − z/|z|) + |z| (z + x)
(1 − |x|2 )n |z|n [z, −x]2
[z, −x](n−2)  n−1
|z|

= (z + x) − (x|z| + z/|z|) .
(1 − |x|2 )(n−1) |z|n−1
According to the identity (2.1), we have
I = sup K(x)
x∈B n
(2.2) Z 1 Z 
rx(r n−2 − 1) + ξ(r n − 1)

2
= sup (1 − |x| ) dr dξ.
x∈B n 0 S |rx + ξ|n+2
Further we have the following simple inequality
rx(r n−2 − 1) + ξ(r n − 1) 6 (1 − r n−2 )|rx + ξ| + r n−2 − r n ,
 

and thus
rx(r n−2 − 1) + ξ(r n − 1)
 
(1 − r n−2 )|rx + ξ| r n−2 − r n
6 + .
|rx + ξ|n+2 |rx + ξ|n+2 |rx + ξ|n+2
So
(1 − r n−2 )|rx + ξ| + r n−2 − r n
Z 1 Z

2
I 6 maxn (1 − |x| ) dr dξ.
x∈B 0 S |rx + ξ|n+2
 GRADIENT OF SOLUTION TO POISSON EQUATION AND RELATED OPERATORS 5

Then we have
π
dξ ωn−1 sinn−2 tdt
Z Z
a
= R π n−2 .
S |rx + ξ| 0 sin tdt 0 (1 + r 2 |x|2 + 2r|x| cos t)a/2
By (1.7) we obtain
Z π
sinn−2 tdt
0 (1 + r 2 |x|2 + 2r|x| cos t)a/2
√
πΓ 21 (−1 + n)


n n 2 2
= F a/2, 1 − + a/2, ,r x .
Γ n2


2 2
In view of
2π n/2
ω Γ[n/2] 2π n/2
R π n−1n−2 = √ =√ ,
πΓ[1/2(−1+n)] πΓ[1/2(−1 + n)]
0 sin tdt
Γ[n/2]
we then infer
dξ π n/2 n n 2 2
Z 
= 2 F a/2, 1 − + a/2, ,r x .
|rx + ξ|a Γ n2


S 2 2
Hence
I 6 Cn sup J(x)
x∈B
with
1  
3 1+n n 2 2
Z
2 n−2
J(x) = (1 − |x| ) (1 − r )F , , , r x dr
0 2 2 2
Z 1 2 2 Z 1
2 n−2 n n − (n − 4)r |x|
+ (1 − |x| ) (r −r ) dr = Kr (x)dr.
0 n (1 − r 2 |x|2 )3 0
Here
∞
X
Kr (x) = Am (r)|x|2m
m=0
where A0 (r) = 1 − rn , r = |x|, and for m > 1
r 2m−4
r n 1 − r 2 −2m(−2 + n + 2m) + 2(1 + m)(n + 2m)r 2



Am (r) =
2n
+ r 2m−4 r 2 − r n −2m(−2 + 2m + n) + (1 + 2m)(−1 + 2m + n)r 2

Γ[ n ]Γ[ 1+2m
2 ]Γ[
n+2m−1
]
× √2 1+n

n
2
.
2 πm!Γ 2 Γ 2 + m
Thus
∞
X
I 6 a0 + am |x|2m
m=1
where
2nπ n/2
a0 = ,
(n + 1)Γ(n/2)
6 DAVID KALAJ AND DJORDJIJE VUJADINOVIĆ

and
2(−3 + n)n + 4(−2 + n)m
am =
n(−3 + n + 2m)(−1 + n + 2m)(1 + n + 2m)
(−2 + n)(−3 + n + 4m)Γ n2 Γ − 12 + m Γ 21 (−3 + n) + m




− √ .
8 πΓ 1+n Γ(1 + m)Γ n2 + m



2

Then am < 0 if and only if

√
2((−3 + n)n + 2(−2 + n)m) πm!Γ 1+n n



2 Γ 2 +m

bm := n

1

3+n < 1.
(−2 + n)n(−3 + n + 4m)Γ 2 Γ − 2 + m Γ 2 +m

Then by (1.9) we have

     
1 3+n 2+n
Γ − +m Γ + m > Γ (m) Γ +m
2 2 2
and so
√
2m((n − 3)n + 2(n − 2)m) πΓ 1+n


2
bm 6 c(m) :=
.
(−2 + n)(n + 2m)(−3 + n + 4m)Γ 1 + n2

Since the last expression increases in m because

√ 1+n
2 (n − 3)2 n2 + 4(n − 3)(n − 2)nm + 4(6 + (n − 3)n)m2


′ πΓ 2
c (m) = >0
(n − 2)(n + 2m)2 (n + 4m − 3)2 Γ 1 + n2

we have
√
πΓ 1+n


2

bm 6 lim c(m) = < 1.
m→∞ 2Γ 1 + n2
So
2nπ n/2
sup K(x) = K(0) = ,
(n + 1)Γ(n/2)
what we needed to prove.


Corollary 2.2. Let D be the mapping defined in (1.5) and v = ∇u = Dg,

g ∈ L∞ (B n ). Then

2nπ n/2
kvk∞ ≤ kgk∞ .
(n + 1)Γ(n/2)
Proof. At the beginning, let us note that
|y|2 x − y
 
x−y
∇x G(x, y) = cn (2 − n) − .
|x − y|n [x, y]n
 GRADIENT OF SOLUTION TO POISSON EQUATION AND RELATED OPERATORS 7

For x ∈ B we have
Z 
 
k∇u(x)k = sup   ∇G(x, y)g(y)dy, ξ 
|ξ|=1 B
Z 
 
= sup  h∇G(x, y), ξi g(y)dy 

|ξ|=1 B

|y|2 x − y
Z   
 x−y 
(2.3) = (n − 2)cn sup  
n
− , ξ g(y)dy 
|ξ|=1 B |x − y| [x, y]n 

|y|2 x − y
 
 x−y
Z

≤ (n − 2)cn sup  n
− n
, ξ  |g(y)|dy
B |ξ|=1 |x − y| [x, y]
Z  2x − y 

 x−y |y|
= (n − 2)cn 
n
−  |g(y)|dy.
B |x − y|
 [x, y]n 
So, we obtain the upper estimate for the gradiente of u, i.e.
k∇uk∞ ≤ kN kkgk∞ .


3. The L1 norm of the operator N

In the sequel let us state a well-known result related to the Riesz potential.
Let Ω be a domain of Rn , and let |Ω| be its volume. For µ ∈ (0, 1] define
the operator Vµ on the space L1 (Ω) by the Riesz potential
Z
(Vµ f )(x) = |x − y|n(µ−1) f (y)dy.
Ω

The operator Vµ is defined for any f ∈ L1 (Ω), and Vµ is bounded on L1 (Ω),

or more generally we have the next lemma.
Lemma 3.1. ([9], pp. 156-159]). Let Vµ be defined on the Lp (Ω) with p > 0.
Then Vµ is continuous as a mapping Vµ : Lp (Ω) → Lq (Ω), where 1 ≤ q ≤ ∞,
and
1 1
0 ≤ δ = δ(p, q) = − < µ.
p q
Moreover, for any f ∈ Lp (Ω)
 1−δ 
1 ωn−1 1−µ µ−δ
kVµ f kq ≤ |Ω| kf kp .
µ−δ n
1
Theorem 3.2. The norm of the operator N : L1 → L1 is n−2 .

Corollary 3.3. Let g ∈ L1 (B) and v = ∇u = D[g]. Then

1
kvk1 ≤ kgk1 .
n−2
In order to prove Theorem 3.2 we need the following lemma
8 DAVID KALAJ AND DJORDJIJE VUJADINOVIĆ

y−x |x|2 y−x

Lemma 3.4. Let H(x, y) = |y−x|n − [y,x]n , and let

1
Z
H[g] = |H(x, y)|g(y)dy.
(n − 2)ωn−1 B

Then

1 1
Z
(3.1) kHkL∞ →L∞ = |H(0, y)|dy = .
(n − 2)ωn−1 B n−2

Proof of Lemma 3.4. We need to find

Z
sup |H(x, y)|dy.
x B

We will show that its supremum is achieved for x = 0. We first have

|y|(1 − |x|2 )
 
1 1
|H(x, y)| 6 K(x, y) + L(x, y) = |x − y| n
− + .
|x − y| [x, y]n [x, y]n

Further we have

  n 
1 1 −  x − y   dy.
Z Z
   
(3.2) sup |K(x, y)| dy = sup
x∈B n Bn x∈B n Bn |x − y|n−1   [x, y]  

We use the change of variables z = Tx y, i.e. T−x z = y, where Tx y is the

Möbius transform

(1 − |x|2 )(y − x) − |y − x|2 x

 
x − y 
Tx y = , |Tx y| = 
 .
[x, y]2 [x, y] 

We obtain

n
1 − |x|2

dy = dz.
[z, −x]2
 GRADIENT OF SOLUTION TO POISSON EQUATION AND RELATED OPERATORS 9

Assume without loss of generality that x = |x|e1 . Further for ξ = (ξ1 , ..., ξn ),
(3.3)
2 n
1 n (1 − |x| )
Z Z
sup |H(x, y)| dy = sup |1 − |z| | dz
x∈B B x∈B B |x − T−x z|n−1 [z, −x]2n
(1 − |z|n−2 ) dz
Z
= sup(1 − |x|2 )n
 x[z,−x]2−(1−|x|2 )(x+z)−|x+z|2x n−1 [z, −x]2n
 
x∈B B  [z,−x]2 
(1 − |z|n ) dz
Z
2 n
= sup(1 − |x| ) n−1
[z, −x]2n
 2
x∈B B |z|n−1  1−|x| 
[z,−x]
1 − |z|n
Z  
2 n−(n−1)
= sup(1 − |x| ) [z, −x](n−1)−2n dz
x∈B B |z|n−1
Z 1
dξ
Z
2 n n−(n−1)−1
= sup(1 − |x| ) (1 − r )r dr 2n−(n−1)
x∈B 0 S |rx + ξ|
Z 1
dξ
Z
= sup(1 − |x|2 ) (1 − r n )dr n+1
x∈B 0 S (r 2 |x|2 + 2r|x|ξ1 + 1) 2
1
2π 2 (−1+n)
Z 1 Z π
2 n sinn−2 t
= 1  sup (1 − |x| ) (1 − r )dr n+1 dt,
Γ 2 (−1 + n) x∈B 0 0 (r 2 |x|2 + 2r|x| cos t + 1) 2
Z 1
2π n/2
 
3 1+n n 2 2
=  n  sup(1 − |x|2 ) (1 − r n )F , , , r |x| dr
Γ 2 x∈B 0 2 2 2
2π n/2
=   sup J(x) = ωn−1 sup J(x)
Γ n2 x∈B x∈B

where
∞
n X
J(x) = − em |x|2m 6 J(0),
1 + n m=1
with
n 8m2 + (n − 1)2 + 2m(3n − 5) Γ m − 12 Γ 32 + m + n2 Γ n2





em = √
.
(2m + n − 1)2 (2m + n + 1)2 πΓ[1 + m]Γ m + n2 Γ 1+n


2
In the first appearance of hypergemetric function we used (1.7).
On the other hand similarly we prove that
Z
L(x) = L(x, y)dy
B
 
′ 2

1+n 3+n 2
= Cn 1 − |x| F 1, , , |x|
2 2
∞
X 2(1 + n)
= Cn′ (1 − |x|2m )
−1 + 4m2 + 4mn + n2
m=1
6 L(0),
10 DAVID KALAJ AND DJORDJIJE VUJADINOVIĆ

where
n π n/2 ωn−1
Z
Cn′ = L(0) = |y|dy = = .
B n + 1 Γ[1 + n/2] n+1
Hence
Z Z
sup |H(x, y)|dy 6 |H(0, y)|dy = ωn−1 J(0) + L(0)
x B B
 
n 1
= + ωn−1 = ωn−1 .
n+1 n+1
This implies (3.1). 
Proof of Theorem 3.3 and Corollary 3.2. Since
kN kL1 →L1 = kN ∗ kL∞ →L∞ ,
where N ∗ is appropriate adjoint operator and
Z Z
∗
N f (x) = N (y, x)f (y)dy = |H(x, y)|f (y)dy, f ∈ L∞ (B),
B B
we have
kN ∗ kL1 →L1 = kHkL∞ →L∞ .
So Theorem 3.3 follows from Lemma 3.4.
On the other hand, Corollary 3.2 follows from the following inequality
kD[g]kL1 →L1 ≤ kN kL1 →L1 .

At this point let us point out the fact that
D : Lp (B, R) → Lp (B, Rn ),
where Lp (B, Rn ) is the appropriate Lebesgue space of vector-functions. By
kDkp we denote the norm of the operator D.
By using the Ries-Thorin interpolation theorem we obtain the next esti-
mates of the norm for the operators N and D
Corollary 3.5. Let us denote by kN ki := kN kLi →Li , i ∈ {1, ∞}. Then
1 p−1
kDkp < kN kp ≤ kN k1p kN k∞p , 1 < p < ∞.
Conjecture 3.6. We know that D, N maps Lp (B) into L∞ (B) for p > n.
We have that
kN gk∞ ≤ Ap kgkp ,
and
kDgk∞ ≤ Bp kgkp ,
where
Z  2 x − y q
 1/q
1  x−y |y|
Ap = sup  −  dy ,
ωn−1 x∈B B  |x − y|n [x, y]n 
and
 GRADIENT OF SOLUTION TO POISSON EQUATION AND RELATED OPERATORS11

q 1/q
|y|2 x − y
Z 
1  x−y 
Bp = sup 
 |x − y|n − , η  dy .
ωn−1 x∈B,|η|=1 B [x, y]n 

Then we conjecture that

q 1/q
 !1/q
Γ[1 + q]Γ 1 + −1 + n1 q
Z  
1 1 − p1
Ap = − |y| dy = ω n−1 ,
n−1 nΓ 2 + nq
 
ωn−1 B |y|

and

Z  q 1/q
1 q 1
Bp = sup | hy, ηi | − 1 dy
ωn−1 |η|=1 B |y|n

 !1/q
Γ n2 Γ[1 + q]Γ 21 (−1 + n + q) Γ 1 + −1 + n1 q
    
p
−1
= ωn−1 .
nΓ 21 (−1 + n) Γ n+q Γ 2 + nq
     
2

References
[1] S. Agmon, A. Douglis, and L. Nirenberg, Estimates near the boundary for solu-
tions of elliptic partial differential equations satisfying general boundary conditions.
I, Comm. Pure Appl. Math. 12 (1959), 623–727.
[2] L. V. Ahlfors: Möbius transformations in several dimensions University of Min-
nesota, School of Mathematics, 1981, 150 p.
[3] J. M. Anderson, A. Hinkkanen, The Cauchy transform on bounded domains. Proc.
Amer. Math. Soc. 107 (1989), no. 1, 179–185.
[4] J. M. Anderson, D. Khavinson; V. Lomonosov, Spectral properties of some inte-
gral operators arising in potential theory. Quart. J. Math. Oxford Ser. (2) 43 (1992),
no. 172, 387–407.
[5] M. Dostanić, Norm estimate of the Cauchy transform on Lp (Ω). Integral Equations
Operator Theory 52 (2005), no. 4, 465–475.
[6] M. Dostanić, Estimate of the second term in the spectral asymptotic of Cauchy
transform. J. Funct. Anal. 249 (2007), no. 1, 55–74.
[7] M. Dostanić, The properties of the Cauchy transform on a bounded domain, Journal
of the Operator Theory 36 (1996), 233–247
[8] S.S. Dragomir, R.P. Agarwal, N. S. Barnett, Inequalities for Beta and Gamma
functions via some classical and new integral inequalities. (English) J. Inequal. Appl.
5, No.2, 103-165 (2000).
[9] D. Gilbarg and N. Trudinger, Elliptic Partial Differential Equations of Second
Order, Second edition. Grundlehren der Mathematischen Wissenschaften [Fundamen-
tal Principles of Mathematical Sciences], 224. Springer-Verlag, Berlin, 1983. xiii+513
pp.
[10] D. Kalaj, On Some Integral Operators Related to the Poisson Equation, Integral
Equation and Operator Theory 72 (2012), 563-575.
[11] D. Kalaj: Cauchy transform and Poisson’s equation, Advances in Mathematics, Vol-
ume 231, Issue 1, 10 September 2012, Pages 213-242.
[12] D. Kalaj, Dj. Vujadinović: The solution operator of the inhomogeneous Dirichlet
problem in the unit ball, Proc. Amer. Math. Soc. 144 (2016), 623-635
[13] D. Kalaj, M. Pavlović: On quasiconformal self-mappings of the unit disk satisfying
the Poisson’s equation, Trans. Amer. Math. Soc. 363 (2011) 4043–4061.
12 DAVID KALAJ AND DJORDJIJE VUJADINOVIĆ

[14] A. P. Prudnikov, Yu. A. Brychkov and O. I. Marichev, Integrals and Series,

Vol. 1: Elementary Functions. Gordon and Breach Sci. Publ., New York, 1986.
[15] G. Thorin, Convexity theorems generalizing those of M. Riesz and Hadamard with
some applications. Comm. Sem. Math. Univ. Lund [Medd. Lunds Univ. Mat. Sem.]
9, (1948), 1–58.

University of Montenegro, Faculty of Mathematics, Dzordza Vašingtona

bb, 81000 Podgorica, Montenegro
E-mail address: [email protected]

University of Montenegro, Faculty of Mathematics, Dzordza Vašingtona

bb, 81000 Podgorica, Montenegro
E-mail address: [email protected]