Hiroshima Math. J.

36 (2006), 1–21

Sharp estimates of the Green function, the Poisson kernel and

the Martin kernel of cones for symmetric stable processes

Krzysztof Michalik
(Received March 23, 2004)
(Revised June 28, 2005)

Abstract. We investigate the Green function, the Poisson kernel and the Martin
kernel of circular cones in the symmetric stable case. We derive their sharp esti-
mates. We also investigate properties of the characteristic exponent of these estimates.
We prove that this exponent is a continuous function of the aperture of the cone.

1. Introduction
In recent years many results in potential theory of a-stable processes have
been obtained ([2], [9], [10], [12]–[17], [19], [20], [22]–[24], [27]). In particular,
the behaviour of a-harmonic functions, the Green function and the Poisson
kernel in smooth domains ([19], [23]) and Lipschitz domains ([9], [10], [22]) has
been investigated. In case of smooth domains the main tool is the Green
function and the Poisson kernel of a ball. The estimates of the Green function
and the explicit formula of the Poisson kernel are known ([8], [19], [23], [25]).
In case of Lipschitz domains the situation is more complicated and it is cones
that seem to be a proper tool. On the other hand, cones are Lipschitz do-
mains themselves and they are regular enough to obtain more detailed results
than those in [10]. In fact, properties of the Green function, the Poisson kernel
and the exit time in bounded and unbounded cones has been studied both in
the classical case ([1], [3], [18]) and a-stable case ([14], [21], [23], [26]). The
latest results for a < 2 are [2] about the so-called generalized cones and [14]
about the Poisson kernel in ‘smoothed’ bounded cones.
The estimates presented in [23] are proved only for cones of acute
aperture. Besides, they are not sharp at the vertex of the cone. The aim of
this paper is to improve these estimates and to extend them to all circular
cones. The basic tool is the so-called Martin kernel with pole at infinity
(introduced in [2]) and its degree of homogeneity. We also use the estimates of
the Green function in smooth domains ([19], [24]) and of the Poisson kernel in

2000 Mathematics Subject Classification. Primary 31B25, 60J45.

Key words and phrases. Stable process, a-harmonic function, Green function, Poisson kernel,
Martin kernel, circular cone.
2 Krzysztof Michalik

Lipschitz domains ([22]). Another important tool is a version of the Boundary

Harnack Principle ([27]).
In section 2 we present basic notation and terminology. Section 3
contains the most important results. First we prove some properties of the
Martin kernel with pole at infinity and its degree of homogeneity. They seem
to be useful in other applications. Next we obtain estimates of the function
f (truncated Green function) used in [22]. With the use of this function we
prove sharp estimates of the Green function and the Poisson kernels of
bounded cones (Theorems 3.6 and 3.7). With some extra arguments we extend
our estimates to unbounded cones (Theorem 3.11). As a consequence of
Theorem 3.6 we obtain estimates of the Martin kernel in cones (Theorems 3.12
and 3.13). We also prove the existence of Martin representation in unbounded
cones (cf. [9]) and we show the consistence between the Martin kernel with pole
at infinity and the classical Martin kernel with the boundary point at infinity.

2. Preliminaries

We will denote by j
j the Euclidean norm of vectors. For B H Rd , d b 2

we denote its complement by B c and its characteristic function by 1B . For
x A Rd , Bðx; rÞ will denote the open ball centered at x of radius r. For a Borel
set B and r > 0 we define rB ¼ frx : x A Bg and B þ x ¼ fx þ y : y A Bg. For
x A Rd let dx ðBÞ ¼ distðx; qBÞ.
Let D denote a bounded open set in Rd . We say D is a Lipschitz domain
if there exist constants R 0 , l > 0 such that for every z A qD there is a function
F : Rd1 ! R and an orthonormal coordinate system y ¼ ð y1 ; . . . ; yd Þ such that
D V Bðz; R 0 Þ ¼ fy : yd > F ð y1 ; . . . ; yd1 Þg V Bðz; R 0 Þ:
Moreover, F is Lipschitz with the Lipschitz constant not greater than l.
Furthermore, if F is di¤erentiable and ‘F is Lipschitz with the Lipschitz
constant not greater than l then D is called a C 1; 1 domain.
Let ðXt ; P x Þ be the rotation invariant(‘symmetric’) a-stable Levy motion
(i.e. homogeneous with independent increments) on Rd with its index a A ð0; 2Þ
([7]). For a Borel subset B of Rd let TB and tB be the first entry time and the
first exit time respectively i.e. TB ¼ infft b 0 : Xt A Bg and tB ¼ TB c .
In this paper constants are always positive numbers. In equations and
inequalities they may change under arithmetic transformations but they will be
denoted by the same symbols. The notation of the form c ¼ cða; b; . . .Þ means
that the constant c depends only on a; b; . . . .
A nonnegative Borel function h on Rd is said to be a-harmonic on D if for
each bounded open set B with B H D and for x A B we have
hðxÞ ¼ E x hðXtB Þ < y: ð1Þ
 Estimates in cones for stable processes 3

If h 1 0 on D c then it is called singular a-harmonic on D. If B can be replaced

by D in (1) then h is called regular a-harmonic on D.
The following theorem ([27, Theorem 3.1]) will be one of the basic tools in
our paper.

Theorem 2.1 (Boundary Harnack Principle). Let D be an open set,

z A qD, r > 0, k A ð0; 1Þ and BðA; krÞ is a ball in D V Bðz; rÞ. Then there exists
some constant C ¼ Cðd; aÞ > 1 such that for any two functions u, v, which
are positive regular a-harmonic in D V Bðz; 2rÞ and vanish in D c V Bðz; rÞ, we
have

vðxÞ uðxÞ vðxÞ

C 1 k dþa a a Ckda ; x A D V Bðz; r=2Þ:
vðAÞ uðAÞ vðAÞ

For x A Rd we define the a-harmonic measure of D as oDx ðBÞ ¼ P x ðXtD A BÞ.

As a function of x it is regular a-harmonic on D if B is fixed. If D is
Lipschitz, its harmonic measure is concentrated on ðDÞ c and has the density
with respect to the Lebesgue measure, which is called the Poisson kernel (see
[10]). This kernel will be denoted by PD ðx; yÞ, x A D, y A ðDÞ c . It satisfies the
following scaling property

PD ðx; yÞ ¼ ð1=r d ÞPð1=rÞD ðx=r; y=rÞ; r > 0: ð2Þ

When D ¼ Bð0; rÞ, r > 0, the Poisson kernel is given by the following explicit
formula
!a=2
r 2  jxj 2 1
Pr ðx; yÞ ¼ Cd; a 2
; jxj < r; jyj > r;
jyj  r 2 jx  yj d

where Cd; a ¼ Gðd=2Þpd=21 sinðpa=2Þ (see [8], [25]).

For all nonnegative Borel measurable functions f we define the Riesz
potential of f by
ðy ð
Uf ðxÞ ¼ E x
f ðXt Þdt ¼ Ad; a jx  yj ad f ð yÞdy;
0

where Ad; a ¼ 2a pd=2 Gððd  aÞ=2Þ=Gða=2Þ (see [7]).

For a Borel set B we define the Green potential of f by
ð tB ð
GB f ðxÞ ¼ E x f ðXt Þdt ¼ GB ðx; yÞ f ð yÞdy;
0

where GB ðx; yÞ is the Green function of B defined by

GB ðx; yÞ ¼ Ad; a ðjx  yj ad  E x jx  XtB j ad Þ; x; y A B; x 0 y; ð3Þ
4 Krzysztof Michalik

GB ðx; xÞ ¼ y and GB ðx; yÞ ¼ 0 otherwise. This function is symmetric (i.e.

GB ðx; yÞ ¼ GB ðy; xÞ), positive in int B and if B1 H B2 then GB1 a GB2 . Fur-
thermore, GB satisfies the following scaling property
GB ðx; yÞ ¼ ð1=r da ÞGð1=rÞB ðx=r; y=rÞ; r > 0: ð4Þ
For other properties of the Green function see [24] and [19].
Every nonnegative function which is singular a-harmonic on a bounded
Lipschitz domain D has a unique representation (called the Martin repre-
sentation)
ð
f ðxÞ ¼ MD ðx; zÞmðdzÞ; ð5Þ
qD

where m is a finite Borel measure on qD. The kernel function MD ðx; zÞ, called
the Martin kernel, may be defined by

GD ðx; yÞ
MD ðx; zÞ ¼ lim ; x A D; z A qD: ð6Þ
D C y!z GD ðx0 ; yÞ

The existence of this limit follows from the Boundary Harnack Principle (see
[9], [20]).
We will also use the following estimates for the Green function of C 1; 1
domains ([19], [24]).

Theorem 2.2. If D is a C 1; 1 domain, then there exist constants c ¼ cðD; aÞ

and C ¼ CðD; aÞ such that for x; y A D,
!
Ad; a dxa=2 ðDÞdya=2 ðDÞ
c min ; a GD ðx; yÞ
jx  yj da jx  yj d
!
Ad; a dxa=2 ðDÞdya=2 ðDÞ
a min ;C : ð7Þ
jx  yj da jx  yj d

From now on we will assume that r > 0, x A D, y A ðqDÞ c , z; Q A qD.

x0 A D will be a fixed reference point. From now on we take R 0 so small that
dx0 ðDÞ b R 0 =2. For r a R 0 =32 and Q A qD we denote by AQ; r a point for
which BðAQ; r ; krÞ H BðQ; rÞ V D for a certain absolute constant k ¼ kðDÞ ¼
pffiffiffiffiffiffiffiffiffiffiffiffiffi
1=ð2 1 þ l 2 Þ. The set of such points is nonempty and AQ; r is not unique.
For r > R 0 =32 we set AQ; r ¼ x1 , where x1 A D is another fixed point such that
jx0  x1 j ¼ R 0 =4. In particular, dx1 ðDÞ b R 0 =4. Furthermore, let r ¼ rðx; yÞ
¼ maxðdx ðDÞ; dy ðDÞ; jx  yjÞ. For r a R 0 =32 we denote by A ¼ Aðx; yÞ a point
for which BðA; krÞ H D V Bðx; 3rÞ V Bðy; 3rÞ. The set of such points is also
nonempty and if S A qD is such that dy ¼ jy  Sj, we may take A ¼ AS; dy . If
r > R 0 =32, we set Aðx; yÞ ¼ x1 . See [22] and [11] for details.
 Estimates in cones for stable processes 5

Let fD ðxÞ ¼ minðGD ðx; x0 Þ; Cd; a ðR 0 =4Þad Þ. We will use the following
estimates for the Poisson kernel and the Green function of Lipschitz domains
([22])
Theorem 2.3. There are constants c ¼ cðd; a; l; R 0 ; diamðDÞÞ, C ¼ Cðd; a;
l; R 0 ; diamðDÞÞ such that for x A D, y A ðDÞ c we have
fD ðxÞfD ð yÞ
c a GD ðx; yÞ
jx  yj da fD2 ðAðx; yÞÞ
fD ðxÞfD ðyÞ
aC : ð8Þ
jx  yj da fD2 ðAðx; yÞÞ
Theorem 2.4. There are constants c ¼ cðd; a; l; R 0 ; diamðDÞÞ, C ¼ Cðd; a;
l; R 0 ; diamðDÞÞ such that for x A D, y A ðDÞ c we have
fD ðxÞfD ð y 0 Þ
c a PD ðx; yÞ
jx  yj da fD2 ðAðx; y 0 ÞÞdya ðDÞð1 þ dy ðDÞÞa

fD ðxÞfD ðy 0 Þ
aC ; ð9Þ
jx  yj da fD2 ðAðx; y 0 ÞÞdya ðDÞð1 þ dy ðDÞÞa
where y 0 ¼ AS; dy and S A qD satisfies dy ¼ jy  Sj.
We will also use the following version of ([10, Lemma 17]), which can be
proved in an analogous way.
Lemma 2.5. Let Bn and B be open sets such that Bn % B. Assume
that for x A B we have P x ðXtB A qBÞ ¼ 0 and P x ðtB < yÞ ¼ 1. Then
limn!y P x ðfXtBn ¼ XtB gÞ ¼ 1. In particular, this implies that limn!y GBn ðx; yÞ
¼ GB ðx; yÞ for x; y A B and limn!y PBn ðx; yÞ ¼ PB ðx; yÞ for x A B, y A int B c .

3. Main results
First we define an unbounded cone with vertex at 0 ¼ ð0; 0; . . . ; 0Þ and
symmetric with respect to the d-th axis. This is a set V defined by
V ¼ fx : h
jðx1 ; x2 ; . . . ; xd1 Þj < xd g;
pffiffiffiffiffiffiffiffiffiffiffiffiffi
where h A ðy; yÞ. The aperture of V is the angle g ¼ arccosðh= 1 þ h 2 Þ.
For g ¼ p we define V by V ¼ Rd nfð0; 0; . . . ; 0; xd Þ : xd a 0g. Generally, an
unbounded cone with vertex at 0 is a set V 0 isomorphic to the cone V defined
above and satisfies rV 0 ¼ V 0 . Finally, an unbounded cone with vertex at
Q A Rd is a set V 0 þ Q.
For an unbounded cone V with vertex at Q we denote by 1 a point on
the axis of V such that j1  Qj ¼ 1. Since the process Xt is homogeneous
6 Krzysztof Michalik

and translationally invariant, we will often assume that Q ¼ 0 and 1 ¼

ð0; 0; . . . ; 0; 1Þ.
Let V be a circular cone with vertex at 0 and aperture g A ð0; p. Assume
that 1 ¼ ð0; 0; . . . ; 0; 1Þ. By [2, Theorem 3.2], there exists the so-called Martin
kernel with pole at infinity. This is a unique nonnegative function MV on Rd
such that MV ð1Þ ¼ 1, MV 1 0 on V c and MV is regular a-harmonic on every
open bounded subset of V . Moreover, MV is locally bounded on Rd and
homogeneous of degree b A ½0; aÞ, that is,
MV ðxÞ ¼ jxj b MV ðx=jxjÞ; x A V: ð10Þ
Furthermore, b ¼ bðV ; aÞ is a strictly decreasing function of g ([2, Lemma
3.3]). We will call b the characteristics of V .
We know the explicit formulas of MV in several cases. If g ¼ p=2, then V
a=2
is the half-space fðx1 ; x2 ; . . . ; xd Þ : xd > 0g. In this case MV ðxÞ ¼ xd , x A V
and this gives b ¼ a=2. If g ¼ p and (d > 2 or a a 1), then V c is polar
([25]). This implies that MV ðxÞ ¼ 1V ðxÞ and b ¼ 0 in this case. If g ¼ p,
d ¼ 2 and a > 1, then V c is not polar and due to [2, Theorem 3.4] we have
b > 0. Actually, it is proved in [14] that b ¼ ða  1Þ=2. We do not know the
formula of MV in this case.
Later on, we will prove that MV is nothing but a constant multiple of a
classical Martin kernel MV ðx; yÞ defined by (6) (see Theorem 3.13).
We now present further properties of MV . This function plays an im-
portant role in further analysis.
Lemma 3.1. For n A N let Vn and V be circular cones with vertices at 0,
apertures gn and g respectively, characteristics bn and b respectively and Martin
kernels with pole at infinity Mn and M respectively. Assume that gn , g A ð0; p
and 1 ¼ ð0; 0; . . . ; 0; 1Þ is on the axis of symmetry of V and every Vn . Then
Mn ðxÞ ! MðxÞ for all x A V if limn!y gn ¼ g. Furthermore, M is continuous
on Rd i¤ g < p or d ¼ 2, g ¼ p, a > 1.

Proof. Since gn ! g, we may assume that gn > g=2 for all n, without loss
of generality. Since g > 0, there is a constant e > 0 such that for every n,
Bð1; eÞ H Vn V Bð0; 3=2Þ. We will prove that there exists C > 0 such that for
every jxj a 1, MðxÞ a C, where C does not depend on V .
Since M is regular a-harmonic on Bð0; rÞ V V for r A ½3=2; 2, we get

MðxÞ ¼ E x MðXtV VBð0; rÞ Þ ¼ E x fMðXtV VBð0; rÞ Þ : XtV VBð0; rÞ A V g

¼ E x fMðXtBð0; rÞ Þ : XtV VBð0; rÞ A V g a E x MðXtBð0; rÞ Þ

ð
¼ Pr ðx; yÞMðyÞdy; ð11Þ
 Estimates in cones for stable processes 7
 a=2
r 2 jxj 2 1
where Pr ðx; yÞ ¼ Cd; a 1fj yj>rg
2
. Let Pe ðyÞ be the Poisson
j yj r 2 jx yj d
kernel for the ball Bð1; eÞ for the process starting from 1. We introduce a
regularized version of the Poisson kernel Pr ðx; yÞ Ðin a similar way as in
2
[12]. Fix a nonnegative j A Ccy ðð3=2; 2ÞÞ such that 3=2 jðrÞdr ¼ 1 and define
Ð2
P~ðx; yÞ ¼ 3=2 jðrÞPr ðx; yÞdr. Then by [12, Lemma 3.11] we have
P~ðx; yÞ a C=ð1 þ jyjÞ dþa ;

where C depends on d and a. Moreover, for jyj b 3=2,

ea C
Pe ð yÞ ¼ b
ðj y  1j  2
e 2 Þa=2 jy  1j d
jyj dþa
for some constant C ¼ Cðd; a; eÞ. Hence
Ð for jxj a 1 and jyj b 3=2 we obtain
P~ðx; yÞ a CPe ð yÞ. Therefore, since P~ðx; yÞdy ¼ 1, by (11) we get
ð ð
MðxÞ a P~ðx; yÞMð yÞdy a C Pe ðyÞMðyÞdy
j yj>3=2 j yj>3=2
ð
aC Pe ð yÞMð yÞdy ¼ CMð1Þ ¼ C:
B c ð1; eÞ

Therefore, for every n and jxj a 1 we have

Mn ðxÞ a C: ð12Þ

This implies that for jxj > 1, by homogeneity of Mn we get

Mn ðxÞ ¼ jxj bn Mn ðx=jxjÞ a Cjxj bn : ð13Þ

We know that bn < a for every n and bn increases as gn decreases. Therefore,

if g > 0, we may assume by (12) and (13) that there is b 0 < a such that for
every n,
0
Mn ðxÞ a C maxðjxj b ; 1Þ: ð14Þ
It implies that Mn are uniformly bounded on every bounded subset B H V .
Also, Mn are equicontinuous on compact subsets F H V because of the
gradient estimates (see [15]). Therefore, by the Arzeli Theorem and diagonal
procedure we find a subsequence Mnk almost uniformly convergent to some
nonnegative function M ~ on V .
We prove now that M ~ is a-harmonic on V . Let x A V and r > 0 such
that B ¼ Bðx; rÞ H Bðx; rÞ H Vn . Then we have
ð
Mn ðxÞ ¼ Mn ð yÞPr ðx; yÞdy: ð15Þ
Bc
8 Krzysztof Michalik

If R > 0 is su‰ciently large, then for jyj > R we have

ra C
Pr ðx; yÞ ¼ a ;
2
ðj y  xj  r 2 Þa=2 jx  yj d
jyj dþa

where C ¼ Cðd; a; r; RÞ. Moreover, by (14), for every n we have Mn ðyÞ a

0
Cjyj b , jyj b 1, and Mn ð yÞ a C, jyj a R. Therefore, we get
( 0
Cjyj b da ; jyj > R;
Mn ð yÞPr ðx; yÞ a
CPr ðx; yÞ; jyj a R:

The right hand side of the above inequality is integrable over B c since b 0 < a
Ð
and Pr ðx; yÞdy ¼ 1. Hence, by dominated convergence, we get
ð ð
~ ðxÞ ¼ lim Mnk ðxÞ ¼
M lim Mnk ð yÞPr ðx; yÞdy ¼ ~ ðyÞPr ðx; yÞdy;
M
k!y B c k!y Bc

which proves that M ~ is a-harmonic on V . Obviously, M ~ 1 0 on V c and

M~ ð1Þ ¼ 1.
Now we need to prove that M ~ is regular a-harmonic on bounded subsets
of V . Assume first that g < p or d > 2 or a a 1. Let B ¼ Bð0; RÞ for some
R > 0. We show that the set G ¼ V V B satisfies the assumptions of Lemma
2.5. This is immediate if 0 < g < p, since G are bounded Lipschitz domains
in this case. If g ¼ p, then V ¼ Rd nfð0; 0; . . . ; 0; xd Þ : xd a 0g. Since d b 3
or a a 1, the set V c is polar ([25]) and M 1 1V is not continuous on Rd .
Let x A G. Since the process does not hit qB when leaving B, we see that
P x ðtG A qGÞ ¼ 0. Obviously, P x ðtG < yÞ ¼ 1 as G is bounded. Now take
open sets Bn H Bn H G such that Bn % G. Since M ~ is a-harmonic and locally
~ x ~
bounded on V , we get M ðxÞ ¼ E M ðXtBn Þ ! E x M ~ ðXtG Þ by Lemma 2.5.
Hence M ~ is regular a-harmonic on V V B.
Now let g ¼ p, d ¼ 2 and a > 1. This case requires di¤erent arguments
since V c is not polar. Let x A V . Fix z A qV and take R > 2 such that z A
Bð0; R=4Þ. First we prove that there exists a constant C ¼ Cðd; a; RÞ such that
for x A V V Bð0; R=4Þ,
~ ðxÞ a CMðxÞ:
M ð16Þ

To show it we define fn ðxÞ ¼ P x ðXtGn A B c Þ and f ðxÞ ¼ P x ðXtG A B c Þ where

Gn ¼ Vn V B, G ¼ V V B. As Gn % G, we may assume that x A Gn for all n
for a given x. Furthermore, Bð1; 1=2Þ H Gn . Next we observe that fn and Mn
are regular a-harmonic on Gn and vanish on Vnc V B. Similarly, f and M are
regular a-harmonic on G and vanish on V c V B. Therefore, by the Boundary
Harnack Principle we obtain
 Estimates in cones for stable processes 9

fn ð1Þ fn ðxÞ fn ð1Þ

c a aC ;
Mn ð1Þ Mn ðxÞ Mn ð1Þ
f ð1Þ f ðxÞ f ð1Þ
c a aC ; ð17Þ
Mð1Þ MðxÞ Mð1Þ

where c, C depend on d, a, R. Since fn ðxÞ a f ðxÞ and 1 b f ð1Þ b fn ð1Þ b

P1 ðXtBð1; 1=2Þ A B c Þ ¼ C > 0, (17) gives Mn ðxÞ a C 0 MðxÞ and letting n ! y we
obtain (16).
Next we show that M is continuous on Rd . We only need to show that

lim MðxÞ ¼ 0: ð18Þ

x!z

First assume that z ¼ 0. M is locally bounded and MðxÞ ¼ jxj b Mðx=jxjÞ

with b > 0 so (18) is immediate. Now let z 0 0. From (17) we get f ðxÞ a
CMðxÞ ! 0 as x ! 0. Let y ¼ x  z. Observe that G  z H V V Bðz; RÞ.
Hence

f ðxÞ ¼ P x ðXtG A B c Þ ¼ P y ðXtGz A B c ðz; RÞÞ

a P y ðXtV VBðz; RÞ A B c ðz; RÞÞ ¼ gðyÞ:

Take r > 0 such that Bð0; rÞ H Bðz; RÞ. Then g is regular a-harmonic on
V V Bð0; rÞ and vanishes on V c V Bð0; rÞ. Hence for y su‰ciently close to 0
we obtain gðyÞ a CMðyÞ by similar arguments as for (17). Therefore,
limy!0 gðyÞ ¼ 0, which implies limx!z f ðxÞ ¼ 0. Now (18) follows from (17).
Combining this with (16) we see that M ~ is continuous on Rd . But we know
that if a function is a-harmonic on a bounded set D and is continuous on D,
then is regular a-harmonic on D ([27]). Hence, as in the previous cases, M ~
is regular a-harmonic on every open bounded subset of V . Therefore, by the
uniqueness of M ([2, Theorem 3.2]), M ~ 1 M and limn!y Mn ¼ M on V .
Finally, continuity of M for g < p will follow from Lemma 3.3 stated below.
This completes the proof. r

The next theorem provides some information about b.

Theorem 3.2. Let Vn and V be as in Lemma 3.1. Let g A ð0; p. Then
(i) If limn!y gn ¼ g, then limn!y bn ¼ b.
( ii ) If limn!y gn ¼ 0, then limn!y bn ¼ a.
(iii) If g ¼ p then b ¼ ða  1Þ=2 in case d ¼ 2 and a > 1, otherwise b ¼ 0.
(iv) If g ¼ p=2, then b ¼ a=2.

Proof. From (10) we have Mn ðð1=2Þ
1Þ ¼ ð1=2Þ bn and Mðð1=2Þ
1Þ ¼

ð1=2Þ b . To prove (i) we use Lemma 3.1. We have Mn ðð1=2Þ
1Þ ! Mðð1=2Þ

10 Krzysztof Michalik

1Þ and it gives limn!y bn ¼ b. It remains to prove (ii) as (iii) and (iv) were
discussed before. By [23, Theorem 4.6], for x A Vn V Bð0; 2Þ we have
0
cdxa=2 ðVn Þjxj a=2e a GVn ðx; 8
1Þ a Cdxa=2 ðVn Þjxj a=2e ; ð19Þ

where c, C depend on d, a, n and 0 < e 0 < e satisfy e; e 0 ! 0 if gn ! 0 [23,

Lemma 3.7]. Since Mn and GVn ð
; 8
1Þ are regular a-harmonic on Vn V Bð0; 6Þ
and vanish on Vnc V Bð0; 6Þ, we obtain by the Boundary Harnack Principle for
all n and x ¼ r
1, r a 1
Mn ðr
1Þ GVn ðr
1; 8
1Þ
¼ :
Mn ð1Þ GVn ð1; 8
1Þ
0
Combining this with (19) we see that cr ae a Mn ðr
1Þ a Cr ae . From (10)
we have Mn ðr
1Þ ¼ r bn Mn ð1Þ ¼ r bn and this implies that a  e a bn a a  e 0 .
Since e; e 0 ! 0 in case n ! y, we get bn ! a. This completes the proof.
r
The rest of the section is devoted to study the behaviour of the Green
function, Poisson kernels and Martin kernels for cones of apertures less than
p. For any cone V with its vertex at 0 and aperture g < p we define its inner
smooth set BV . This will be a fixed C 1; 1 domain such that
ðBð0; 3=2ÞnBð0; 1=16ÞÞ V V H BV H ðBð0; 2ÞnBð0; 1=32ÞÞ V V :
We notice that BV can be chosen to be dependent only on d and g. Next we
define a bounded (‘smooth’) cone of length 2 and vertex at 0 by
V2 ¼ BV U ðV V Bð0; 1=16ÞÞ:
Similarly we define a bounded (‘smooth’) cone of length R > 0 and vertex at 0
by VR ¼ ðR=2ÞV2 . Finally, a bounded (‘smooth’) cone of length R and vertex
at Q is a set of the form Q þ VR .
From now on V and VR will denote, respectively, unbounded and bounded
cones with vertices at a fixed Q. We also set RV ¼ ð1=8Þ sin g a 1=8. This
implies that V V Bðz; 4RV Þ H BV if z A qV V qBðQ; 3=4Þ. Furthermore, let x 0 ¼
ð1=16Þ
1 ¼ ð0; 0; . . . ; 0; 1=16Þ.
The next lemma provides sharp estimates for MV .
Lemma 3.3. There exist constants c ¼ cðd; a; gÞ, C ¼ Cðd; a; gÞ such that
for every x A V we have

cjx  Qj ba=2 dxa=2 ðV Þ a MV ðxÞ a Cjx  Qj ba=2 dxa=2 ðV Þ: ð20Þ

Proof. We may assume that Q ¼ 0 and 1 lies on the axis of V . Let
z A qV V qBð0; 1Þ. Since MV is homogeneous, we obtain
MV ðxÞ ¼ MV ðx=jxjÞjxj b : ð21Þ
 Estimates in cones for stable processes 11

By definition of BV and RV we have x 0 A BV and jz  x 0 j > 4RV . So the

functions MV ð
Þ and GBV ð
; x 0 Þ are regular a-harmonic on V V Bðz; 4RV Þ
and vanish on V c V Bðz; 4RV Þ. Therefore, if x=jxj A Bðz; RV Þ, then by the
Boundary Harnack Principle we obtain

GBV ðx=jxj; x 0 Þ MV ðx=jxjÞ GBV ðx=jxj; x 0 Þ

c 0
a aC ; ð22Þ
GBV ðAz; RV ; x Þ MV ðAz; RV Þ GBV ðAz; RV ; x 0 Þ

where c, C depend only on d, a, g, and the definition of the points Az; RV refers
to the set BV . Since dAz; RV ðV Þ ¼ dAz; RV ðBV Þ b kRV , all the points Az; RV are in a
compact subset of V . Since MV ð
Þ is continuous, positive on V and GBV ð
; x 0 Þ
is continuous, positive on BV nfx 0 g, we get

c a GBV ðAz; RV ; x 0 Þ a C;

c a MV ðAz; RV Þ a C; ð23Þ

and, again, c, C depend only on d, a, g. By the definition of BV we have

dx=jxj ðBV Þ ¼ dx=jxj ðV Þ if x=jxj A Bðz; RV Þ. Therefore, from Theorem 2.2 we
have
a=2 a=2
a=2
dx=jxj ðBV Þdx 0 ðBV Þ
c 0 dx=jxj ðV Þ a c a GBV ðx=jxj; x 0 Þ
jx=jxj  x 0 j d
a=2 a=2
dx=jxj ðBV Þdx 0 ; BV a=2
aC a C 0 dx=jxj ðV Þ ð24Þ
jx=jxj  x 0 j d

for x such that x=jxj A Bðz; RV Þ and for any z A qV V qBð0; 1Þ. Now (24)
extends easily to all x=jxj A V V qBð0; 1Þ since dx ðV Þ and GBV ðx; x 0 Þ are positive
and continuous functions of x on V V qBð0; 1Þ. Since dx=jxj ðV Þ ¼ dx ðV Þ=jxj, we
obtain (20) by combining (21), (22), (23) and (24). The proof is completed.
r
A bounded cone is a Lipschitz domain with a localization radius R 0 and
a Lipschitz constant l b h. Recall that fVR ðxÞ ¼ minðGVR ðx; x0 Þ; Cd; a R 0ad Þ.
We know that if D is a Lipschitz domain with its localization radius r0 , then rD
is a Lipschitz domain with its localization radius rr0 . This implies that fVR has
the same scaling property as GVR .
With this remark we are ready to present estimates for fVR .

Lemma 3.4. There exist constants c ¼ cðd; a; gÞ, C ¼ Cðd; a; gÞ such that
for every x; y A VR we have

cR adb dxa=2 ðVR Þjx  Qj ba=2 a fVR ðxÞ a CR adb dxa=2 ðVR Þjx  Qj ba=2 :
12 Krzysztof Michalik

Proof. The steps taken are analogous to those in the proof of Lemma
3.3. We may assume that Q ¼ 0. Let R ¼ 2. We know that fV2 ðxÞ ¼
GV2 ðx; x0 Þ for x su‰ciently close to qV2 . Let z A qV2 and x A Bðz; RV Þ. If
jzj a 1, then applying the Boundary Harnack Principle to the functions
GV2 ð
; x0 Þ, MV and using Lemma 3.3, we obtain dxa=2 ðV Þjxj ba=2 a GV2 ðx; x0 Þ a
Cdxa=2 ðV Þjxj ba=2 . Since ð1=2Þdx ðV Þ a dx ðV2 Þ a dx ðV Þ, this leads to the desired
result. If jzj b 1, then we have c a jxj ba=2 a C. In this case we apply the
Boundary Harnack Principle to the functions GV2 ð
; x0 Þ and GBV ð
; x0 Þ. We
obtain cGBV ðx; x0 Þ a GV2 ðx; x0 Þ a CGBV ðx; x0 Þ. Now Theorem 2.2 completes
the proof for fV2 . The scaling property of fVR gives the estimates for all VR .
r
We also need the following technical lemma.

Lemma 3.5. Let Q ¼ 0. Assume that x; y A V2 . Then there exist con-

stants c ¼ cðd; a; gÞ, C ¼ Cðd; a; gÞ such that for A ¼ Aðx; yÞ we have

c maxðjxj; jyjÞ a jAj a C maxðjxj; j yjÞ: ð25Þ

Proof. We note that for V2 , R 0 can be chosen to be dependent only on d
and g. Recall that r ¼ maxðdx ðV2 Þ; dy ðV2 Þ; jx  yjÞ. This gives
r a maxðjxj; jyj; jxj þ jyjÞ a 2 maxðjxj; jyjÞ: ð26Þ
Assume first that r a R 0 =32. Then, by definition, A A Bðx; 3rÞ V Bðy; 3rÞ
and kr a dA ðV2 Þ a r. This, combined with (26), leads to jAj a jxj þ 3r a
7 maxðjxj; jyjÞ. Furthermore, jAj b jxj  3r, jAj b jyj  3r. This implies that

jAj b maxðjxj; jyjÞ  3r ð27Þ

and

jAj b dA ðV2 Þ b kr: ð28Þ

If r a maxðjxj; j yjÞ=ðk þ 3Þ, then from (27), jAj b maxðjxj; j yjÞk=ðk þ 3Þ. If
r b maxðjxj; jyjÞ=ðk þ 3Þ, then jAj b maxðjxj; jyjÞk=ðk þ 3Þ from (28). Hence
we proved that 7 maxðjxj; jyjÞ b jAj b maxðjxj; jyjÞk=ðk þ 3Þ.
If r > R 0 =32, then we have A ¼ x1 and by (26), 2 b maxðjxj; jyjÞ b
R 0 =64. Hence (25) holds immediately. The proof is completed. r

Now, with the use of Theorem 2.3 and the estimates for fVR we obtain
sharp estimates for the Green functions of bounded cones. This theorem
improves [23, Theorem 4.7].

Theorem 3.6. Let R > 0. There exist constants c ¼ cðd; a; gÞ, C ¼

Cðd; a; gÞ such that for every x; y A VR we have
 Estimates in cones for stable processes 13

  !
Ad; a dxa=2 ðVR Þdya=2 ðVR Þ minðjx  Qj; jy  QjÞ ba=2
c min ;
jx  yj da jx  yj d maxðjx  Qj; jy  QjÞ

a GVR ðx; yÞ
  !
Ad; a dxa=2 ðVR Þdya=2 ðVR Þ minðjx  Qj; jy  QjÞ ba=2
a min ;C :
jx  yj da jx  yj d maxðjx  Qj; jy  QjÞ

Proof. Let R ¼ 2 and Q ¼ 0. We note that in this case the choice of R 0

depends only on d and g. Assume that x; y A V2 and A ¼ Aðx; yÞ is as in
Theorem 2.3. From Lemma 3.4 we have

cdxa=2 ðV2 Þjxj ba=2 a fV2 ðxÞ a Cdxa=2 ðV2 Þjxj ba=2 ;

cdya=2 ðV2 Þj yj ba=2 a fV2 ð yÞ a Cdya=2 ðV2 Þj yj ba=2 ; ð29Þ

a=2 a=2
cdA ðV2 ÞjAj ba=2 a fV2 ðAÞ a CdA ðV2 ÞjAj ba=2 :

Recall that r ¼ maxðdx ðV2 Þ; dy ðV2 Þ; jx  yjÞ. We consider two cases.

Case 1. Let jx  yj b ð1=10Þ maxðdx ðV2 Þ; dy ðV2 ÞÞ. We have jx  yj a
r a 10jx  yj. If r a R 0 =32, then by definition, kr a dA ðV2 Þ a r. If r >
R 0 =32, then dA ðV2 Þ ¼ dx1 ðV2 Þ and 4 b jx  yj b R 0 =320. This implies that
cjx  yj a dA ðV2 Þ a Cjx  yj: ð30Þ

Now, combining (25), (29), (30) and (8) we obtain

 
dxa=2 ðV2 Þdya=2 ðV2 Þ minðjxj; jyjÞ ba=2
GV2 ðx; yÞ b c
jx  yj d maxðjxj; j yjÞ
  !
Ad; adxa=2 ðV2 Þdya=2 ðV2 Þ minðjxj; jyjÞ ba=2
b min ;c ;
jx  yj da jx  yj d maxðjxj; j yjÞ
and
 
dxa=2 ðV2 Þdya=2 ðV2 Þ minðjxj; j yjÞ ba=2
GV2 ðx; yÞ a C :
jx  yj d maxðjxj; jyjÞ

Since GV2 ðx; yÞ a Ad; a jx  yj ad by definition, this completes the proof in
Case 1.
Case 2. Let jx  yj a ð1=10Þ maxðdx ðV2 Þ; dy ðV2 ÞÞ. We may assume that
dx ðV2 Þ b dy ðV2 Þ. This means that jx  yj a ð1=10Þdx ðV2 Þ. Let S A qV2 be a
point for which dy ðV2 Þ ¼ j y  Sj. Then we obtain dx ðV2 Þ a jx  Sj a jx  yj þ
jy  Sj a ð1=10Þdx ðV2 Þ þ dy ðV2 Þ. This implies that
dy ðV2 Þ a dx ðV2 Þ a ð10=9Þdy ðV2 Þ; ð31Þ
14 Krzysztof Michalik

which leads to jx  yj a ð1=9Þ minðdx ðV2 Þ; dy ðV2 ÞÞ. Hence, by [22, Lemma 11],
ad
cjx  yj a GV2 ðx; yÞ a Ad; a jx  yj ad ;
where c depends on d, a. Hence, immediately,
  !
dxa=2 ðV2 Þdya=2 ðV2 Þ minðjxj; jyjÞ ba=2
Ad; a
GV2 ðx; yÞ b c min ; :
jx  yj da jx  yj d maxðjxj; jyjÞ

Furthermore, using (31) we have jxj a jx  yj þ jyj a ð1=10Þdx ðV2 Þ þ j yj a

ð1=9Þdy ðV2 Þ þ j yj a ð10=9Þj yj and, similarly, j yj a jx  yj þ jxj a ð1=10Þdx ðV2 Þ
þ jxj a ð11=10Þjxj. This implies that

ð9=10Þjxj a jyj a ð11=10Þjxj:

This means that
 ba=2
minðjxj; j yjÞ
ca a C:
maxðjxj; jyjÞ

As we have jx  yj a a ðð1=9Þ minðdx ðV2 Þ; dy ðV2 ÞÞÞa a ð1=9Þa dxa=2 ðV2 Þdya=2 ðV2 Þ,
 
Ad; a dxa=2 ðV2 Þdya=2 ðV2 Þ minðjxj; jyjÞ ba=2
GV2 ðx; yÞ a a C :
jx  yj da jx  yj d maxðjxj; jyjÞ

This completes the proof for GV2 . The scaling property (4) extends the
estimates to all GVR . r

In an analogous way we can obtain estimates for PVR as it is stated in the

next theorem. We present the proof for the convenience of the reader. The
estimates for PVR ðx; yÞ when dy ðVR Þ a R 0 =32 are the same as in [14, Theorem
3.2] and they have been proved independently and simultaneously.

Theorem 3.7. Let R > 0. Then there exist constants c ¼ cðd; a; gÞ, C ¼
Cðd; a; gÞ such that for x A VR , y A int VRc we have
 ba=2
dxa=2 ðVR Þ minðjx  Qj; jy  QjÞ
c a PVR ðx; yÞ
dya=2 ðVR Þjx  yj d maxðjx  Qj; jy  QjÞ
 ba=2
dxa=2 ðVR Þ minðjx  Qj; jy  QjÞ
aC
dya=2 ðVR Þjx  yj d maxðjx  Qj; jy  QjÞ

if dy ðVR Þ a R 0 =32, and

dxa=2 ðVR Þjx  Qj ba=2 R ab dxa=2 ðVR Þjx  Qj ba=2 R ab
c a PVR ðx; yÞ a C
jyj dþa jyj dþa
if dy ðVR Þ > R 0 =32.
 Estimates in cones for stable processes 15

Proof. Let R ¼ 2 and Q ¼ 0. We note that in this case the choice of R 0

depends only on d and g. Assume first that x A V2 , y A int VRc and dy ðV2 Þ a
R 0 =32: Let y 0 and A ¼ Aðx; y 0 Þ be as in Theorem 2.4. From Lemma 3.4 we
have

cdxa=2 ðV2 Þjxj ba=2 a fV2 ðxÞ a Cdxa=2 ðV2 Þjxj ba=2 ;
a=2 a=2
cdy 0 ðV2 Þj y 0 j ba=2 a fV2 ðy 0 Þ a Cdy 0 ðV2 Þj y 0 j ba=2 ; ð32Þ
a=2 a=2
cdA ðV2 ÞjAj ba=2 a fV2 ðAÞ a CdA ðV2 ÞjAj ba=2 :

First we prove that

cdya=2 ðV2 Þj yj ba=2 a fV2 ðy 0 Þ a Cdya=2 ðV2 Þj yj ba=2 ; ð33Þ

where c, C depend only on d, a, g. We observe that, by definition we

have kdy ðV2 Þ a dy 0 ðV2 Þ a dy ðV2 Þ. Next, we see that if dy ðV2 Þ ¼ jy  Sj, then
jy  y 0 j a j y  Sj þ jy 0  Sj a 2dy ðV2 Þ. This implies that jyj a j y  y 0 j þ jy 0 j
a 2dy ðV2 Þ þ jy 0 j a ð2=kÞdy 0 ðV2 Þ þ j y 0 j a ð3=kÞj y 0 j and similarly jy 0 j a jy  y 0 j þ
jyj a 2dy ðV2 Þ þ j yj a 3jyj. This gives

ðk=3Þj yj a jy 0 j a 3jyj: ð34Þ

Hence, (33) follows from (32).

Recall that r ¼ maxðdx ðV2 Þ; dy 0 ðV2 Þ; jx  y 0 jÞ. We show that

ðk=ðk þ 2ÞÞjx  yj a r a 3jx  yj: ð35Þ

First, notice that maxðdx ðV2 Þ; dy ðV2 ÞÞ a jx  yj for all x A V2 , y B V2 . Hence

jx  y 0 j a jx  yj þ jy  y 0 j a jx  yj þ 2dy ðV2 Þ a 3jx  yj and dy 0 ðV2 Þ a dy ðV2 Þ
a jx  yj. This gives the upper bound in (35). Next, jx  yj a jx  y 0 j þ
jy  y 0 j a jx  y 0 j þ 2dy ðV2 Þ a r þ ð2=kÞdy 0 ðV2 Þ a ð1 þ 2=kÞr. Hence, (35) is
proved.
From (35) by the same arguments as for (30) we obtain

cjx  yj a dA ðV2 Þ a Cjx  yj: ð36Þ

Finally, 1 a 1 þ dy ðV2 Þ a 1 þ R 0 =32. Now using (9) and combining this
with (25) and (32)–(36), we obtain the desired estimates for R ¼ 2.
Now let dy ðV2 Þ > R 0 =32. Then y 0 ¼ x1 and r b dx1 ðV2 Þ b R 0 =4 so
A ¼ x1 , too. Next, R 0 =32 a dy ðV2 Þ a jyj a j y  Sj þ jSj a 2 þ dy ðV2 Þ a
ð1 þ 64=R 0 Þdy ðV2 Þ and dy ðV2 Þ a jx  yj a jxj þ jyj a 2 þ jyj a ð1 þ 64=R 0 Þj yj.
Combining this with (32) and (9), we complete the proof for R ¼ 2. The
scaling property (2) for the Poisson kernels completes the proof for every
R > 0. r
16 Krzysztof Michalik

As a simple application we estimate the probability that the process

starting ‘near’ the vertex of a cone VR exits the cone ‘far’ from the vertex. We
will use this result in further analysis.
Lemma 3.8. Let R > 1 and x A VR V BðQ; R=2Þ. Then there are constants
c ¼ cðd; a; gÞ, C ¼ Cðd; a; gÞ such that

cdxa=2 ðVR Þjx  Qj ba=2 =R b a P x ðXtVR A B c ðQ; 3R=4ÞÞ

a Cdxa=2 ðVR Þjx  Qj ba=2 =R b :

Proof. Take Q ¼ 0. From the scaling property (2) we see that
P x ðXtVR A B c ð0; 3R=4ÞÞ ¼ P x=R ðXtV1 A B c ð0; 3=4ÞÞ
ð
¼ PV1 ðx=R; yÞdy: ð37Þ
j yjb3=4

For jx=Rj < 1=2 and jyj b 3=4 we have ð1=3Þj yj a jx=R  yj a ð5=3Þj yj.
Since R > 1, ð1=2Þdx ðVR Þ=R a dx=R ðV1 Þ a 2dx ðVR Þ=R. Let x 0 A V1=2 . Com-
bining this with Theorem 3.7, we obtain
PV1 ðx=R; yÞ a=2 ba=2 C2 a=2 dxa=2 ðVR Þjxj ba=2
a Cdx=R ðV 1 Þðjxj=RÞ a :
PV1 ðx 0 ; yÞ Rb
Hence, by (37),
0 dxa=2 ðVR Þjxj ba=2
P x ðXtVR A B c ð0; 3R=4ÞÞ a CP x ðXtV1 A B c ð0; 3=4ÞÞ :
Rb
The proof is completed. r
Now we consider the case of unbounded cones. We have the following
Lemma 3.9. For all x A V we have P x ðtV < yÞ ¼ 1 and P x ðXtV A qV Þ
¼ 0.
Proof. Since b > 0, from [2, Theorem 4.1] we have E x tVt < y in case
t A ð0; b=aÞ. This implies that P x ðtV < yÞ ¼ 1.
Now take x A V V BðQ; R=2Þ. Let R > 1. Since the Lebesgue measure of
qV is 0 and P x ðXtVR A qVR Þ ¼ 0, we obtain P x ðXtVR A qV Þ ¼ 0. Hence we have

P x ðXtV A qV Þ ¼ P x ðXtV A qV ; XtVR A V c Þ þ P x ðXtV A qV ; XtVR A V nVR Þ

¼ P x ðXtVR A qV Þ þ P x ðXtV A qV ; XtVR A V nVR Þ

a P x ðXtVR A B c ðQ; 3R=4ÞÞ

and, by Lemma 3.8, the last term tends to 0 as R ! y. This completes the
proof. r
 Estimates in cones for stable processes 17

With this lemma we are able to describe the behaviour of the Green
function and the Poisson kernel of VR when R ! y. As an immediate
consequence of Lemma 2.5 we obtain

Lemma 3.10. For x; y A Rd we have limR!y GVR ðx; yÞ ¼ GV ðx; yÞ and

limR!y PVR ðx; yÞ ¼ PV ðx; yÞ.

By this result and Theorem 3.7 we easily obtain the following estimates for GV
and PV .
Theorem 3.11. There exist constants c ¼ cðd; a; gÞ, C ¼ Cðd; a; gÞ such
that
( i ) for x; y A V we have
  !
Ad; a dxa=2 ðV Þdya=2 ðV Þ minðjx  Qj; jy  QjÞ ba=2
c min ;
jx  yj da jx  yj d maxðjx  Qj; jy  QjÞ

a GV ðx; yÞ
  !
Ad; a dxa=2 ðV Þdya=2 ðV Þ minðjx  Qj; jy  QjÞ ba=2
a min ;C ;
jx  yj da jx  yj d maxðjx  Qj; j y  QjÞ

(ii) for x A V , y A int V c we have

 ba=2
dxa=2 ðV Þ minðjx  Qj; jy  QjÞ
c a PV ðx; yÞ
dya=2 ðV Þjx  yj d maxðjx  Qj; j y  QjÞ
 
dxa=2 ðV Þ minðjx  Qj; j y  QjÞ ba=2
aC :
dya=2 ðV Þjx  yj d maxðjx  Qj; jy  QjÞ

Remark 1. If V ¼ fð0; 0; . . . ; 0; xd Þ : xd > 0g, we know the explicit for-

a=2
mula of PV . This is equal to PV ðx; yÞ ¼ Cd; a xd j yja=2 jx  yjd (see [9,
Example 2]). This confirms the estimates from Theorem 3.11 as b ¼ a=2 in
this case. Note that this formula can be found by choosing a sequence of balls
Bn % V , using the explicit formulas of the Poisson kernels for Bn and applying
Lemma 2.5.

Now let us focus on Martin kernels of cones. Since a bounded cone is a

bounded Lipschitz domain, its Martin kernel may be defined as in (6). Hence,
as an immediate consequence of Theorem 3.6 we obtain the following esti-
mates.
Theorem 3.12. Let R > 0. Every singular a-harmonic function on VR
admits the Martin representation in the sense of (5) with the Martin kernel MVR
18 Krzysztof Michalik

defined as in (6). There are constants c ¼ cðd; a; gÞ, C ¼ Cðd; a; gÞ such that for
x A VR , z A qVR we have
 
jx0  zj d dxa=2 ðVR Þ minðjx  Qj; jz  QjÞ
maxðjx0  Qj; jz  QjÞ ba=2
c
jx  zj d dxa=2 ðVR Þ maxðjx  Qj; jz  QjÞ
minðjx0  Qj; jz  QjÞ
0

a MVR ðx; zÞ
 
jx0  zj d dxa=2 ðVR Þ minðjx  Qj; jz  QjÞ
maxðjx0  Qj; jz  QjÞ ba=2
aC
jx  zj d dxa=2 ðVR Þ maxðjx  Qj; jz  QjÞ
minðjx0  Qj; jz  QjÞ
0

if z 0 Q and

jx0  Qj dþba=2 dxa=2 ðVR Þ jx0  Qj dþba=2 dxa=2 ðVR Þ

c a MVR ðx; QÞ a C :
jx  Qj dþba=2 dxa=2
0
ðVR Þ jx  Qj dþba=2 dxa=2
0
ðVR Þ
The case of unbounded cones is more complicated. However, the results are
similar, with a natural extension fyg A qV .
Theorem 3.13. Every singular a-harmonic function on V admits Martin the
representation in the sense of (5) with the Martin kernel MV defined as in (6).
There are constants c ¼ cðd; a; gÞ, C ¼ Cðd; a; gÞ such that for x A V we have
 
jx0  zj d dxa=2 ðV Þ minðjx  Qj; jz  QjÞ
maxðjx0  Qj; jz  QjÞ ba=2
c
jx  zj d dxa=2
0
ðV Þ maxðjx  Qj; jz  QjÞ
minðjx0  Qj; jz  QjÞ

a MV ðx; zÞ
 
jx0  zj d dxa=2 ðV Þ minðjx  Qj; jz  QjÞ
maxðjx0  Qj; jz  QjÞ ba=2
aC
jx  zj d dxa=2 ðV Þ maxðjx  Qj; jz  QjÞ
minðjx0  Qj; jz  QjÞ
0

if z A qV nfQ U yg and
jx0  Qj dþba=2 dxa=2 ðV Þ jx0  Qj dþba=2 dxa=2 ðV Þ
c a MV ðx; QÞ a C ;
jx  Qj dþba=2 dxa=2
0
ðV Þ jx  Qj dþba=2 dxa=2
0
ðV Þ
jx  Qj ba=2 dxa=2 ðV Þ jx  Qj ba=2 dxa=2 ðV Þ
c a MV ðx; yÞ a C :
jx0  Qj ba=2 dxa=2
0
ðV Þ jx0  Qj ba=2 dxa=2
0
ðV Þ
Furthermore, we have
MV ðx; yÞ ¼ MV ðxÞ
MV ð1; yÞ:
Proof. The existence of the Martin representation is not immediate since
V is unbounded. To prove it, we introduce the Kelvin transform T (with pole
at 0) as
Tx ¼ x=jxj 2 ; x 0 0: ð38Þ
 Estimates in cones for stable processes 19

We have T 1 ¼ T and jTxj ¼ 1=jxj. Next we introduce the Kelvin transform

of a function as
TuðxÞ ¼ jxj ad uðTxÞ: ð39Þ

See [17] for further details.

We may assume that Q ¼ ð0; 0; . . . ; 0; 1Þ and 1 ¼ ð0; 0; . . . ; 0; 2Þ. V is a
circular domain and its profile consists of two half-lines starting from Q. By
symmetry, the image of V under T, TV is also circular and its profile is the
image of the profile of V . It is easy to check that the image of a line under T
is a circle provided the line does not pass through 0. Hence the profile of TV
consists of two arcs of the same radius and these arcs have two common points
Q and 0. It implies that TV is a bounded Lipchitz domain so, according to
[9], there exists the Martin representation on TV with classical Martin kernels
MTV ðx 0 ; z 0 Þ, x 0 A TV , z 0 A qTV . Then, by [9, Lemma 8], the classes of singular
a-harmonic functions on V and TV are isomorphic and there exists Martin
representation on V with its kernel function TMTV ð
; z 0 Þ. Since 0 B TV and
T 1 ¼ T, we have from [17, Proposition 2.6]

jxj ad jyj ad GTV ðTx; TyÞ ¼ GV ðx; yÞ; x; y A V : ð40Þ

Let Tx0 be the reference point for the kernel MTV ð
;
Þ. From (40) we
obtain

GV ðx; yÞ jxj ad GTV ðTx; TyÞ

¼ : ð41Þ
GV ðx0 ; yÞ jx0 j ad GTV ðTx0 ; TyÞ

If y ! z, then Ty ! Tz with a natural extension T0 ¼ y, Ty ¼ 0. Hence,

letting y ! z in (41) and using (6), we obtain

jxj ad MTV ðTx; TzÞ

MV ðx; zÞ ¼ :
jx0 j ad

By (39) this implies that MV ðx; zÞ ¼ jx0 j da TMTV ðx; TzÞ. It means that
classical Martin kernels MV ð
;
Þ exist and the kernels TMTV ð
; z 0 Þ are their
constant multiples. Hence there exists the classical Martin representation on V
with the kernel functions MV ð
;
Þ. The estimates of MV ð
;
Þ are an imme-
diate consequence of (6) and Theorem 3.6.
It remains to prove the last part of the theorem. We see that the function
f ðxÞ ¼ MV ðx; yÞ=MV ð1; yÞ is nonnegative singular a-harmonic in V and
f ð1Þ ¼ 1. Moreover, the estimates of MV ð
; yÞ imply that f is locally
bounded. Hence, by [10, Lemma 17], f is regular a-harmonic on every open
bounded subset of V . From the uniqueness of MV ð
Þ ([2, Theorem 3.2]) we
obtain f 1 MV ð
Þ, which completes the proof. r
20 Krzysztof Michalik

Acknowledgements

The author is grateful to dr hab. K. Bogdan and dr hab. M. Ryznar

for introducing him into the subject and for many helpful discussions on the
paper.

References

[ 1 ] V. S. Azarin, Generalization of a theorem of Hayman’s on a subharmonic function in an

n-dimensional cone (in Russian), Mat. Sb. (N.S.) 66(108) (1965), 248–264.
[ 2 ] R. Bañuelos and K. Bogdan, Symmetric stable processes in cones, Potential Anal. 21
(2004), no. 3, 263–288.
[ 3 ] R. Bañuelos and R. G. Smits, Brownian motion in cones, Probab. Theory Related Fields
108(3) (1997), 299–319.
[ 4 ] R. F. Bass, Probablistic Techniques in Analysis, Springer-Verlag, New York, 1995.
[ 5 ] R. F. Bass and K. Burdzy, A probabilistic proof of the boundary Harnack principle, In
Seminar on Stochastic Processes, 1989 (San Diego, CA, 1989), pages 1–16, Birkhäuser Boston,
1990, Boston, MA.
[ 6 ] R. F. Bass and M. Cranston, Exit times for symmetric stable processes in R n , Ann.
Probab. 11(3) (1983), 578–588.
[ 7 ] R. M. Blumenthal and R. K. Getoor, Markov Processes and Potential Theory, Academic
Press, New York, 1968.
[ 8 ] R. M. Blumenthal, R. K. Getoor and D. B. Ray, On the distribution of first hits for the
symmetric stable processes, Trans. Amer. Math. Soc., 99 (1961), 540–554.
[ 9 ] K. Bogdan, Representation of a-harmonic functions in Lipschitz domains, Hiroshima
Math. J. 29 (1999), 227–243.
[10] K. Bogdan, The boundary Harnack principle for the fractional Laplacian, Studia Math.
123(1) (1997), 43–80.
[11] K. Bogdan, Sharp estimates for the Green function in Lipschitz domains, J. Math. Anal.
Appl. 243 (2000), 326–337.
[12] K. Bogdan and T. Byczkowski, Potential theory for the a-stable Schrödinger operator on
bounded Lipschitz domains, Studia Math. 133(1) (1999), 53–92.
[13] K. Bogdan and T. Byczkowski, Probabilistic proof of boundary Harnack principle for a-
harmonic functions, Potential Anal. 11 (1999), 135–156.
[14] K. Bogdan and T. Jakubowski, Problème de Dirichlet pour les fonctions a-harmoniques sur
les domaines coniques, Annales Mathematiques Blaise Pascal 12 (2005), 297–308.
[15] K. Bogdan, T. Kulczycki and A. Nowak, Gradient estimates for harmonic and q-harmonic
functions of symmetric stable processes, Illinois J. Math. 46 (2002), 541–556.
[16] K. Bogdan, A. Stós, P. Sztonyk, Potential theory for Lévy stable processes, Bull. of The
Polish Academy of Sciences, Mathematics, vol. 50, no. 3, (2002), 361–392.
[17] K. Bogdan and T. Żak, On Kelvin transformation, J. Theor. Probab. (2004), to appear.
[18] D. L. Burkholder, Exit times of Brownian motion, harmonic majorization and Hardy
spaces, Adv. in Math. 26 (1977), 182–205.
[19] Z.-Q. Chen and R. Song, Estimates on Green functions and Poisson kernels for symmetric
stable processes, Math. Ann. 312 (1998), 465–501.
[20] Z.-Q. Chen and R. Song, Martin boundary and integral representation for harmonic
functions of symmetric stable processes, J. Funct. Anal. 159 (1998), 267–294.
 Estimates in cones for stable processes 21

[21] R. D. DeBlassie, The first exit time of a two-dimensional symmetric stable process from a
wedge, Ann. Probab. 18(3) (1990), 1034–1070.
[22] T. Jakubowski, The estimates for the Green function in Lipschitz domains for the symmetric
stable processes, Probability and Mathematical Statistics, vol. 22, Fasc. 2 (2002), 419–441.
[23] T. Kulczycki, Exit time and Green function of cone for symmetric stable processes,
Probability and Mathematical Statistics, vol. 19, Fasc. 2 (1999), 337–374.
[24] T. Kulczycki, Properties of Green function of symmetric stable processes, Probability and
Mathematical Statistics, vol. 17, Fasc. 2 (1997), 339–364.
[25] N. S. Landkof, Foundations of Modern Potential Theory, Springer, New York 1972.
[26] P. J. Méndez-Hernández, Exit times from cones in R n of symmetric stable processes,
Illinois J. Math. 46 (2002), 155–163.
[27] R. Song, J.-M. Wu, Boundary Harnack Principle for symmetric stable processes, J. Funct.
Anal. 168 (1999), 403–427.

Krzysztof Michalik
Institute of Mathematics
Wrocław University of Technology
Wyb. Wyspiańskiego 27
50-370 Wrocław, Poland
email: [email protected]