Stochastic Processes and their Applications 120 (2010) 1680–1700

www.elsevier.com/locate/spa

Regularity of semigroups generated by Lévy type

operators via coupling
Jian Wang ⇤
School of Mathematics and Computer Science, Fujian Normal University, Fuzhou, 350007, PR China
Institut für Mathematische Stochastik, TU Dresden, Dresden, 01062, Germany
Received 9 November 2009; received in revised form 9 April 2010; accepted 20 April 2010
Available online 5 May 2010

Abstract

By adopting the coupling method, we obtain new verifiable sufficient conditions about the Cb (Rd )-
Feller continuity, the Lipschitz continuity and the strong Feller continuity of the semigroups associated
with Lévy type operators. These results easily apply to jump–diffusion processes and stochastic differential
equations driven by Lévy processes. Our results also yield the criterion for the e-property (namely the
characterization about the equi-continuity of semigroups acting on bounded Lipschitz functions) of Lévy
type operators, and show that both genuine Lévy processes and the Ornstein–Uhlenbeck type processes are
e-processes.
c 2010 Elsevier B.V. All rights reserved.

MSC: 60J25; 60J75; 60J35

Keywords: Lévy type operators; Cb (Rd )-Feller continuity; Lipschitz continuity; Strong Feller continuity; Coupling;
Feller processes; e-property

1. Introduction

Consider a d-dimensional Markov process {X t }t 0, whose infinitesimal generator acting on

smooth functions is given by
d d
1 X X
L f (x) = ai j (x)@i j f (x) + bi (x)@i f (x)
2 i, j=1 i=1
Z
+ f (x + z) f (x) r f (x) · z1{|z|1} ⌫(x, dz), (1.1)

⇤ Corresponding address: Institut für Mathematische Stochastik, TU Dresden, Dresden, 01062, Germany.
E-mail address: [email protected].

0304-4149/$ - see front matter c 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.spa.2010.04.007
 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700 1681

where (a(x), b(x), ⌫(x, dz)) are the Lévy characteristics, i.e. a(x) := (ai j (x))d⇥d is a nonnega-
tive definite matrix-valued function, b(x) := (bi (x))
R is a measurable function, and ⌫(x, dz) is a
nonnegative, -finite kernel on Rd \ {0} such that (1 ^ |z|2 )⌫(x, dz) < +1 for every x 2 Rd .
For all f 2 Bb (Rd ) (the set of bounded measurable functions), set
Pt f (x) = E x ( f (X t )), t 0, x 2 Rd .
The family of operator (Pt )t 0 forms a semigroup of bounded linear operators on Bb (Rd ).
In this paper, we consider various continuous properties of the semigroup (Pt )t 0 directly
through the operator L given by (1.1). One of the principle purposes is to provide easily verifiable
sufficient conditions under which
(1) X has the Cb (Rd )-Feller continuous property, i.e. for all t > 0, Pt maps Cb (Rd ) into itself;
(2) X fulfils the strong Feller continuity, i.e. for all t > 0, Pt maps Bb (Rd ) into Cb (Rd ).
The other concern of our paper is to show when the semigroup (Pt )t 0 is Lipschitz continuous,
i.e. for all t > 0, Pt maps Lipb (Rd ) into Lipb (Rd ), where Lipb (Rd ) denotes the set of bounded
Lipschitz continuous functions. Particularly, new characterization about the equi-continuity of
semigroups acting on Lipb (Rd ) (which we call e-property adopting from [15]) is presented.
For all continuous properties of the semigroups above, we find the coupling argument, which
has been developed in [6,30] for Markov jump processes and diffusions. However, due to the
appearing of Lévy measure ⌫(x, dz) in (1.1), the coupling operator and the coupling process
here should be constructed and applied more carefully (see Section 3.1 below). To the best of
knowledge, our results (Theorems 2.1 and 2.3) about the Cb (Rd )-Feller continuity and the strong
Feller continuity respectively for Lévy type operators did not appear before, and both of them
improve many existing assertions on related topics (cf. [4,26,28]); Theorem 2.2 is the first result
in the literature about the e-property of the semigroups associated with Lévy jumps. We also
note that, a wide variety of examples can be taken into consideration by applying our results; for
example, we can deal with jump–diffusion processes and stochastic differential equations driven
by Lévy processes.
Though derivation of the Feller continuous properties is interesting in its own right, this
was motivated by the ergodicity of stochastic differential equations with jumps and that of
Lévy type operators (see [16,18,19,31,32]). For example, one result of [24] states that, for a
Cb (Rd )-Feller continuous semigroup {Pt }t 0 , there are R r two mutually exclusive possibilities:
either an invariant probability exists, or limr !1 supµ 0 µPs (1C )ds = 0 for any compact set
C, where the supremum is taken over all initial distributions µ. Furthermore, suppose that the
semigroup {Pt }t 0 is strong Feller and an invariant probability measure µ exists. Then, the
process is recurrent and for all starting points x the transition probability Pt (x, ·) converges
to µ in the total variation norm as t ! 1 (e.g. see [25]). Therefore, our work supplies
previous research in these quoted papers. On the other hand, it is well known that the strong
Feller property and the irreducibility ensure the uniqueness of the invariant measure for (Pt )t 0 .
However, to deal with the ergodicity of Lévy type operators, the strong Feller continuity is a
rigorous restriction (cf. see Theorem 2.3). Recently, [15, Theorem 1] shows that there exists a
unique invariant probability measure for a Markov process, provided that its semigroups have the
e-property and the pointwise Doeblin condition (e.g. see [15, (1.1)] or (2.6) below) holds. We
stress that, according to Theorem 2.2, both genuine Lévy processes and the Ornstein–Uhlenbeck
type processes have e-property, and they surely are not necessarily strong Feller. We believe that
the e-property is more efficient to study the ergodicity of Lévy type operators.
1682 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700

Another starting point of this paper is due to the fact that the operator (1.1) formally is the
generator corresponding to a Feller process, i.e. the process whose associated semigroups defined
on the continuous functions vanishing at infinity enjoy the Feller property. Restricted on Cc1 (Rd )
(the set of smooth functions with compact support), the generator L given by (1.1) becomes a
pseudo-differential operator
Z
L f (x) = p(x, D) f (x) := (2⇡) d/2 eix·⇠ p(x, ⇠ ) fˆ(⇠ )d⇠, (1.2)
Rd
R
where fˆ(x) = (2⇡) d/2 f (⇠ )e ix⇠ d⇠ and the symbol function p : Rd ⇥ Rd ! C possesses
the Lévy–Khinchine representation
Z ⇣ ⌘
1
p(x, ⇠ ) = ⇠ · a(x)⇠ ib(x) · ⇠ + 1 eiz·⇠ + iz · ⇠ 1{|z|1} ⌫(x, dz).
2
Feller processes have been particularly investigated by many researchers by means of the theory
of pseudo-differential operators, and nowadays a lot of results are available (cf. see [10–13]
and the references therein). However, the general cases about the Cb -Feller, strong Feller and
Lipschitz continuities have not been studied here.
The remainder of this paper is organized as follows. In the next section we first set up
regularity assumptions on the Lévy type operator given by (1.1) and necessary notations, then
we describe the main results about the Cb (Rd )-Feller continuity, the Lipschitz continuity and the
strong Feller continuity. Section 3 is devoted to the construction of a suitable coupling process
and the proofs of our results. In the final section, we provide some comments on our results and
the coupling method of Lévy type operators for the sake of complement.

2. Assumptions, basic notations and statements of results

We first recall the notion of martingale problem from [1, Section 6.7.3]. Let ⌦x be the
path space of all càdlàg functions ! from R+ to Rd for which !(0) = x and let Fx be the
-algebra generated by the cylinder sets. We say that a probability measure P x on (⌦x , Fx )
solves the martingale problem associated with L given by (1.1) if P x (X 0 = x) = 1 and for all
f 2 Cc1 (Rd ),
⇢ Z t
f (X t ) L f (X s )ds, Ft , t 0
0

is a local martingale under P x , where X t is the coordinated process, i.e. X (t, !) = !(t) for all
t 0, and Ft is the natural filtration (X s : s  t). The martingale problem is said to be well
posed if such a family of measures (P x )x2Rd exists and is unique.
Our assumptions in this paper are as follows:
R R
Assumption (H). (1) The coefficients ai j , bi , {|z|>1} ⌫(·, dz) and {|z|1} |z|2 ⌫(·, dz) are
continuous on Rd .
(2) The martingale problem for Lévy type operator L (1.1) is well posed.
(3) There exist a -finite measure space (U, U , M) and a function c(x, u) on Rd ⇥ U such that
for a.e. u 2 U , the function c(·, u) is continuous on Rd and
Z Z
1 (c(x, u))M(du) = 1 (z)⌫(x, dz), 2 B(Rd ). (2.1)
U {|z|1}
 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700 1683

Let us comment on these hypotheses. First, under condition (1), the Taylor formula shows
that (1.1) is meaningful for any f 2 Cb2 (Rd ). Second, general answers to the uniqueness
(i.e. well posed) of martingale problem are driven from the locally uniqueness and the non-
explosive property. For the former, we refer to [3,8,12] and the references therein. On the other
hand, there already have existed a number of results about non-explosiveness of the martingale
problem for Lévy type operators. Starting from the integral–differential representation (1.1) and
employing the Meyn and Tweedie’s stability theory, [32, Theorem 3.1] has presented criteria
for the conservativeness of the martingale problem for general Lévy type operators. When
the coefficients in the Lévy type operator (1.1) are bounded, [22, Theorem 5.2] proved that a
sufficient (also almost necessary) condition for non-explosiveness is that the symbol p(x, ⇠ )
satisfies p(x, 0) = 0 for each x 2 Rd . This result has been generalized to the unboundedness
case in [32, Theorem 2.1]. Third, the last assumption in (H) is our technique condition, which
is due to the construction of the coupling operator in Section 3.1. It is remarkable that under
the conditions that U is Lusin space and M is a -finite diffusive measure on (U, U ), the
construction of the coefficient c(x, u) in (2.1) has been given in [14]. The paper [29] has also
studied the case that the Lévy measure ⌫(x, dz) has a generalized polar decomposition. For
further details about Rcondition (3) in (H) we would refer to [27, Chapter 3]. Forth, we mention
that, in the case that {|z|1} |z|⌫(x, dz) < 1 for each x 2 Rd , the condition (3) can be removed
(see some more detailed discussion in Section 4.1). We also stress that there are two important
classes of processes automatically satisfy condition (3). One is the stochastic differential equation
driven by Lévy process

dX t = F(X t )dZ t ,

where Z t is an r -dimensional Lévy process and F : Rd ! Rd⌦r . See the recent monograph [1]
for fundamental facts of this process. Another is the jump–diffusion process
Z
dX t = b(X t )dt + (X t )dWt + c(X t , z)µ̃(dz, dt),

where Wt is an r1 -dimensional standard Wiener process, {µ(B, I ) : I ⇢ R+ , B ⇢ Rr2 \ {0}}

is a Poisson random measure with intensity measure ⌫(dz)dt, µ̃(dz, dt) is the compensated
measure of µ(dz, dt), i.e. µ̃(dz, dt) = µ(dz, dt) ⌫(dz)dt, and the coefficients b : Rd ! Rd ,
: Rd ! Rd⌦r1 , c : Rd ⇥ Rr2 ! Rd . A jump–diffusion process is one of useful models in
many applications.

Necessary notations. Throughout our work, we suppose that there exists 0 such that
ha(x)h, hi |h|2 for x and h 2 Rd . Denote by be the unique symmetric nonnegative
definite matrix-valued function such that 2 = a I. For simplicity, set (x) = 0 (x) for
x 2 Rd . For any x, y 2 Rd , define

1
↵ (x, y) = k (x) (y)k2 , B(x, y) = hb(x) b(y), x yi,
2 Z
k⌫(x, ·) ⌫(y, ·)kVar,>1 = |⌫(x, dz) ⌫(y, dz)|,
{|z|>1}
Z
1
ke
⌫(x, ·) ⌫(y, ·)kVar,1
e = |c(x, z) c(y, z)|2 M(dz).
2 U
1684 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700

We also introduce the following notation, which will be used frequently:


↵ (x, y) + B(x, y) + ke ⌫(x, ·) e⌫(y, ·)kVar,1
K (x, y) =
|x y|2 (1 + |x y|)
k⌫(x, ·) ⌫(y, ·)kVar,>1
+ .
|x y|

2.1. Cb (Rd )-Feller continuity

Theorem 2.1. The process X t is Cb (Rd )-Feller continuous, if for some 2 (0, 1), there exists a
constant C > 0 such that for any x, y 2 Rd with |x y|  ,
2
(K (x, y) C ) |x y|  . (2.2)
(1 + |x y|)2
Theorem 2.1 indicates that genuine Lévy processes (i.e. the coefficients in (1.1) are constants)
are Cb (Rd )-Feller continuous. As a direct consequence of Theorem 2.1, the proposition below
shows that the local Lipschitz continuity of the coefficients in (1.1) implies the Cb (Rd )-Feller
continuity of the process X t .

Proposition 2.1. Suppose that there exists > 0 such that for any x, y 2 Rd with |x y|  ,
the coefficients a, b and ⌫ in the Lévy type operator (1.1) satisfy
k (x) (y)k2 + 2hb(x) b(y), x yi + ke
⌫(x, ·) ⌫(y, ·)kVar,1
e
+ (1 + |x y|)|x y|k⌫(x, ·) ⌫(y, ·)kVar,>1  C |x y|2 . (2.3)
Then the process X t is Cb (Rd )-Feller continuous. Particularly, when for x, y 2 Rd with
|x y|  ,
k (x) (y)k2 + |b(x) b(y)|2 + ke
⌫(x, ·) ⌫(y, ·)kVar,1
e
+ |x y|k⌫(x, ·) ⌫(y, ·)kVar,>1  C 0 |x y|2 , (2.4)
the process X t has the Cb (Rd )-Feller continuous property.
Proposition 2.1 is new even in the context of diffusions, and it extends [28, Theorem 6.3.4]
and [4, Proposition 5.6.4].

2.2. Lipb (Rd ) continuity and e-property

Indeed, the argument of Theorem 2.1 also gives us the following interesting result.

Proposition 2.2. Under the assumptions in Theorem 2.1 (particularly under condition (2.3)), the
semigroups (Pt )t 0 of X t map Lipb (Rd ) into itself, where Lipb (Rd ) denotes the set of bounded
Lipschitz continuous functions. More precisely, for any f 2 Lipb (Rd ) and t 0,
⇢
(2 + 1 )eC t C( f )|x y|, |x y|  ;
|Pt f (x) Pt f (y)| 
2 1 k f k|x y|, |x y| > ,
where C( f ) is a constant depending on k f k and the Lipschitz constant L( f ).
The semigroup preserving Lipb (Rd ) is called a Lipschitz semigroup in [2]. Proposition 2.2
is an extension of results in [2, Section 6]. Next, based on Theorem 2.1 and Proposition 2.2,
 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700 1685

we consider the e-property of the semigroups, which is important in studying the existence of a
unique invariant probability about Lévy type operators. First, according to [15, Definition 2.1],
a transition semigroup (Pt )t 0 on Rd has e-property if for any f 2 Lipb (Rd ) the family of the
functions (Pt f )t 0 is equi-continuous at every x 2 Rd . That is, for any f 2 Lipb (Rd ), x 2 Rd
and " > 0, there exists := ( f, ") > 0 such that for any z 2 B(x, ) and t 0, it holds that
|Pt f (z) Pt f (x)|  ". The corresponding process X shall be called an e-process.

Theorem 2.2. If for some 2 (0, 1), there exists a constant C 0 such that for any x, y 2 Rd
with |x y|  ,
2
(K (x, y) + C ) |x y|  , (2.5)
(1 + |x y|)2
then the process X t is an e-process; moreover, for any f 2 Lipb (Rd ) and t 0,
⇢
(2 + 1 )e C t C( f )|x y|, |x y|  ;
|Pt f (x) Pt f (y)| 
2 1 k f k|x y|, |x y| > ,
where C( f ) is a constant depending on k f k and the Lipschitz constant L( f ).
As mentioned above, Theorem 2.2 implies that both genuine Lévy processes and the
Ornstein–Uhlenbeck type processes are e-processes. Particularly, combining Theorem 2.2 with
[15, Theorem 1] yields that

Corollary 2.1. Suppose that Assumption (H) and condition (2.5) hold. If there exists a 2 Rd
such that for every > 0 and x 2 Rd ,
Z
1 T
lim Pt (x, B(a, ))dt > 0, (2.6)
T !1 T 0
then the process X t admits a unique invariant probability measure.

2.3. Strong Feller continuity

In this part, we will see that a coupling argument, more straightforward and of a probabilistic
nature, also yields efficient conditions about the strong Feller continuous property.

Theorem 2.3. Assume that for a fixed 2 (0, 1), there exists C > 0 such that for any x, y 2 Rd
with |x y|  ,
2
K (x, y)|x y| + C  . (2.7)
(1 + |x y|)2
Then for any f 2 Bb (Rd ) and t > 0, the function x 7! Pt f (x) is Lipschitz continuous;
moreover,
⇣ ⌘
|Pt f (x) Pt f (y)|  2 (1 + ) (C t) 1 + 1 k f k|x y|.
Particularly, under (2.7) the process X t is strong Feller continuous.
Suppose that the diffusion matrix a(x) is strictly definite (i.e. > 0) and the coefficients
satisfy the local Lipschitz continuous conditions (2.4). Then, (2.7) holds and the process X t
is strong Feller continuous. So, Theorem 2.3 shows that genuine Lévy processes with non-
degenerate Gaussian part are strong Feller continuous. Theorem 2.3 also improves the strong
1686 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700

Feller conditions pointed out in [26, Remark in Page 232 and Remark in Page 237] and
[17, Theorem 2.1]. In addition, we emphasis that, Theorem 2.2 above has discussed the equi-
continuous property of (Pt f )t 0 by using condition (2.5), which is exactly between conditions
(2.2) and (2.7).
To end up this section, we recall the following result and then state the difference between the
conclusions in Theorems 2.1 and 2.3.
Lemma 2.1 ([21, Lemma 2.2]). Let (Pt )t 0 be a Markov semigroups on Bb (Rd ), let c > 0 and
t > 0 be fixed. Then the following conditions are equivalent.
(1) For any 2 Bb (Rd ) and any x, y 2 Rd , |Pt (x) Pt (y)|  ck k|x y|.
(2) For any 2 Cb2 (Rd ) and any x, y 2 Rd , |Pt (x) Pt (y)|  ck k|x y|.
Clearly, both conditions in Lemma 2.1 are also equivalent to
(3) For any 2 Lipb (Rd ) and any x, y 2 Rd , |Pt (x) Pt (y)|  ck k|x y|.
According to Lemma 2.1 and the remark above, we provide an equivalent conclusion of
Theorem 2.3.
Proposition 2.3. Under condition (2.7), for any f 2 Lipb (Rd ) and t 0,
⇣ ⌘
|Pt f (x) Pt f (y)|  2 (1 + ) (C t) 1 + 1 k f k|x y|.

Both Propositions 2.2 and 2.3 ensure the Lipschitz continuous property of the family of
(Pt f )t 0 for any f 2 Lipb (Rd ). The difference between the consequence of conditions (2.2)
and (2.7) only dues to the estimates about the Lipschitz constants (L(Pt f ))t 0 .

3. Coupling argument and Feller continuity

3.1. Construction of coupling process

L on R2d is called a coupling operator of the Lévy type operator L, if

Recall that an operator e
e
L satisfies the marginal condition:
(e
L f )(x1 , x2 ) = (L f )(x1 ), f 2 Cb2 (Rd ), x1 , x2 2 Rd ,
where on the left hand side of the above equality f is regarded as a function in Cb2 (R2d ). Lévy
type operator L defined by (1.1) consists of the diffusion (i.e. differential) part L 1 and the pure
jump (i.e. integral) part L 2 . The coupling methods for diffusion operators and Markov jump
processes have been well studied in [5–7,30]. These facts combining with the linear property of
e
L motivate us to split the construction of the coupling operator e L into the following two parts.
First, for the diffusion part L 1 , we use the coupling by reflection-marching e
L 1 (cf. see [5,20]). Set
C(x, y) = + (x) (y)⇤ for x, y 2 Rd .
Then, the corresponding diffusion matrix of e
L 1 is given by
✓ ◆
a(x) C(x, y)
ã(x, y) := , (3.1)
C(x, y)⇤ a(y)

and the drift term is b̃(x, y) := (b(x), b(y)). Second, the coupling operator e
L 2 of the pure Lévy
type jump part L 2 is established in the sense of Marching Coupling (cf. see [7, Example 2.11]).
Here, we use condition (3) in Assumption (H) to deal with small jumps. For any f 2 Cb2 (R2d ),
define
 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700 1687
Z
e
L 2 f (x, y) = ( f (x + z, y) f (x, y)) (⌫(x, dz) ⌫(y, dz))+
{|z|>1}
Z
+ ( f (x, y + z) f (x, y)) (⌫(y, dz) ⌫(x, dz))+
{|z|>1}
Z
+ ( f (x + z, y + z) f (x, y)) (⌫(y, dz) ^ ⌫(x, dz))
{|z|>1}
Z
+ f (x + c(x, z), y + (y, z)) f (x, y)
U
@x f (x, y) · c(x, z) @ y f (x, y) · c(y, z) M(dz)
= : L 2,>1 f (x, y) + e
e L 2,M f (x, y).
Then, it is easy to check that e
L 2 is a coupling of L 2 . So, we obtain that e
L := e L1 + eL 2 is a
coupling of the Lévy type operator L.
The process corresponding to the reflection–marching coupling operator e L 1 of diffusion
part has been constructed in [20, Section 3.1], which is equivalent to saying (e.g. [9, Chapter
IV; Proposition 2.1 and Theorem 2.3]) that there are a stochastic basis (⌦ e, F e, F
et , P)
e and a
2d-dimensional Brownian motion W̃t such that the following process X̃ t takes values in R̂2d (the
standard one point compactification of R2d ) and is generated by e
L 1 before the explosive time:
d X̃ t = ˜ ( X̃ t )dW̃t + b̃( X̃ t )dt,
where for all x 2 R2d , ˜ (x) ˜ (x)⇤ = ã(x) and ã(x) is given by (3.1). Without loss of generality,
we assume that on (⌦ e, Fe, F et , P),
e there also exists a Poisson point process L̃ t , independent of
W̃t , with Poisson measure M on (U, U ). Now, we consider a 2d-dimensional solution Z̃ t of the
following stochastic differential equation
Z
d Z̃ t = ˜ ( Z̃ t )dW̃t + b̃( Z̃ t )dt + c̃( Z̃ t , z)µ̃(dz, dt), (3.2)
U

where for all x, y 2 Rd

and z 2 U , c̃((x, y), z) = (c(x, z), c(y, z)) and µ̃ is the compensated
Poisson measure corresponding to the point process RL̃ t . According to condition (1) in (H) and
the transformation (2.1), the coefficients ˜ , b̃ and U c̃(·, z)2 M(dz) are continuous on R2d ;
moreover, R by the dominated convergence theorem and condition (3) in (H), for every x 2 R2d ,
2
lim y!x U (c̃(y, z) c̃(x, z)) M(dz) = 0. Thus, applying the Euler approximation method
(e.g. [23, Theorem 175] and [9, Chapter IV; Theorems 2.2 and 2.3]), we could obtain that there
exists a weak solution to the Eq. (3.2) on R̂2d . By the Itô formula of general Lévy type stochastic
integrals ([1, Theorem 4.4.7]), this process is a solution to the martingale problem for the operator
e
L1 + e L 2,M . Furthermore, for any x, y 2 Rd and A 2 B(R2d ), define
Z
⌫2,>1 ((x, y), A) :=
e (⌫(x, dz) ⌫(y, dz))+
{u2 R d :(u,0)2A with |u|>1 }
Z
+ (⌫(y, dz) ⌫(z, dz))+
{u2R :(0,u)2A with |u|>1}
d
Z
+ (⌫(y, dz) ^ ⌫(x, dz)) .
{u2Rd :(u,u)2A with |u|>1}
R
Then, thanks to the fact that {|z|>1} ⌫(·, dz) < 1, for any x, y 2 Rd , e ⌫2,>1 ((x, y), ·) is
2d
a finite measure on R , and so the operator e L 2,>1 can be viewed as a generator of some
1688 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700

compound Poisson process corresponding to e ⌫2,>1 . So, by applying the standard interlacing
procedure (e.g. [1, Theorem 6.2.9, p. 311]) to Eq. (3.2), on (⌦e, Fe, F e there is a process Z̃ t0
et , P),
such that its generator is the coupling operator L̃. Moreover, the Itô formula yields that for each
f 2 Cb2 (R2d ),
⇢ Z t^⇣
0
f ( Z̃ t^⇣ ) L f ( Z̃ s0 )ds, t
e 0
0

is an F et local martingale, where ⇣ is the explosion time of the process Z̃ 0 = ( X̃ 0 , Ỹt ), i.e.
t t
⇣ = limn!1 n and n = inf{t > 0 : | X̃ t0 | + |Ỹt | n}. The marginal processes X̃ t0 and Ỹt
are associated with the operator L starting from x and y, respectively. According to assumption
(2) in (H), the processes X̃ t0 and Ỹt are non-explosive, so one has ⇣ = 1. Thus, the coupling
operator e L generates a non-explosive process Z̃ t0 = ( X̃ t0 , Ỹt ). Let T be the coupling time of Z̃ t0 ,
i.e. T = inf{t 0 : X̃ t0 = Ỹt }. This is an F
et -stopping time; define a new process X̃ t = X̃ 0 if
t
t  T ; X̃ t = Ỹt if t T . Following Page 251 in [20, Section 3.1], for any f 2 Cc1 (Rd ),
Z t ✓ Z t^T ◆
f (X t0 ) L f (X s0 )ds = 0
f (X t^T ) L f (X s0 )ds
0 0
✓ Z t ◆
+ f (X t0 ) 0
f (X t^T ) 0
L f (X s )ds
t^T
✓ Z t^T ◆
= f (X t^T ) L f (X s )ds
0
✓ Z t ◆
+ f (Yt ) f (Yt^T ) L f (Ys )ds .
t^T

The equality above combining with the Doob stopping theorem yields that the process X̃ t0
is also associated with the Lévy type operator L given by (1.1). Thanks to condition (2) in
Assumption (H) again and [5, Lemma 2.1], we know that Z t := ( X̃ t , Ỹt ) is a coupling process of
X t . Recall that, X̃ t = Ỹt a.s. for any t T , and the generator of Z t before the coupling time T
is just the coupling operator e L constructed before.
To obtain some basic properties about the coupling operator e L, we will adopt some more
notations introduced in Section 2. First, we introduce an auxiliary function sequence {h n }n 1
such that for any n 1, h n 2 C 2 (R), 0  h n (x)  |x| for |x|  1/n and h n (x) = |x| for
|x| 1/n (cf. see the proof of [31, Theorem 1.4]).

Proposition 3.1. For any x, y 2 Rd with |x y| 1/n, we have

h i
e 2
L f n (|x y|)  |x y|(1+|x y|)2
+ K (x, y) f (|x y|), (3.3)

where f (r ) = r (1 + r ) 1 and f n (r ) = f (h n (r )).

Proof. According to the structure of the coupling operator e

L, we split the proof into three parts.
(1) Note that for |x y| 1/n, f n (|x y|) = f (|x y|). Using the straightforward formula
[5, (2.8)],
 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700 1689

e
L 1 f n (|x y|) = e
L 1 f (|x y|)
trA(x, y) A(x, y) + B(x, y) 0
= A(x, y) f 00 (|x y|) + f (|x y|),
|x y|
where
A(x, y) = a(x) + a(y) 2C(x, y),
A(x, y) = hx y, A(x, y)(x y)i/|x y|2 .
Note that, (e.g. see Page 256 in the proof of [20, Theorem 3.4]),
A(x, y) 4
and
1
trA(x, y) A(x, y) = k (x) (y)k2 = ↵ (x, y).
2
Since f 0 (r ) = (1 + r ) 2
> 0 and f 00 (r ) = 2(1 + r ) 3 < 0, it follows that

k (x) (y)k2 + 2B(x, y)
e
L 1 f n (|x y|)  2 f 00 (|x y|) + f 0 (|x y|)
2|x y|

2 ↵ (x, y) + B(x, y)
= + f (|x y|). (3.4)
|x y|(1 + |x y|)2 |x y|2 (1 + |x y|)
(2) For any x, y and z 2 Rd ,
|x + z y| |x y| 1
<1= + .
1 + |x + z y| 1 + |x y| 1 + |x y|
Thus, for |x y| 1/n, by the definition of f n ,
Z
( f n (|x + z y|) f n (|x y|)) (⌫(x, dz) ⌫(y, dz))+
{|z|>1}
Z
 ( f (|x + z y|) f (|x y|)) (⌫(x, dz) ⌫(y, dz))+
{|z|>1}
Z ✓ ◆
|x + z y| |x y|
= (⌫(x, dz) ⌫(y, dz))+
{|z|>1} 1 + |x + z y| 1 + |x y|
R R
{|z|>1} (⌫(x, dz) ⌫(y, dz))+ {|z|>1} (⌫(x, dz) ⌫(y, dz))+
 = f (|x y|).
1 + |x y| |x y|
Similarly,
Z
( f n (|x y z|) f n (|x y|)) (⌫(y, dz) ⌫(x, dz))+
{|z|>1}
R
{|z|>1} (⌫(y, dz) ⌫(x, dz))+
 f (|x y|).
|x y|
Furthermore,
Z
( f n (|x + z y z|) f n (|x y|)) (⌫(y, dz) ^ ⌫(x, dz))
{|z|>1}
Z
 ( f (|x + z y z|) f (|x y|)) (⌫(y, dz) ^ ⌫(x, dz)) = 0.
{|z|>1}
1690 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700

Therefore, for any |x y| 1/n,

Z
( f n (|x + z y|) f n (|x y|)) (⌫(x, dz) ⌫(y, dz))+
{|z|>1}
Z
+ ( f n (|x y z|) f n (|x y|)) (⌫(y, dz) ⌫(x, dz))+
{|z|>1}
Z
+ ( f n (|x + z y z|) f n (|x y|)) (⌫(y, dz) ^ ⌫(x, dz))
{|z|>1}
k⌫(x, ·) ⌫(y, ·)kVar,>1
 ⇥ f (|x y|). (3.5)
|x y|

(3) We find that, for any |x y| 1/n,

Z
( f n (|x y + c(x, z) c(y, z)|) f n (|x y|)
U
@x f n (|x y|) · c(x, z) @ y f n (|x y|) · c(y, z) M(dz)
Z
 ( f (|x y + c(x, z) c(y, z)|) f (|x y|)
U
@x f (|x y|) · (c(x, z) c(y, z))) M(dz)
Z 
|x y + c(x, z) c(y, z)| |x y|
=
U (1 + |x y + c(x, z) c(y, z)|)(1 + |x y|)
h(x y)/|x y|, c(x, z) c(y, z)i
M(dz)
(1 + |x y|)2
Z
= I (x, y, c(x, z) c(y, z))M(dz),
U

where

|x + z y| |x y| h(x y)/|x y|, zi

I (x, y, z) := . (3.6)
(1 + |x + z y|)(1 + |x y|) (1 + |x y|)2

For any x, y and z 2 Rd ,

|x + z y| |x y| |x + z y| |x y|
I (x, y, z) =
(1 + |x + z y|)(1 + |x y|) (1 + |x y|)2
|x + z y| |x y| h(x y)/|x y|, zi
+
(1 + |x y|)2
✓ ◆
|x + z y| |x y| 1 1
=
1 + |x y| 1 + |x + z y| 1 + |x y|
|x + z y| |x y| h(x y)/|x y|, zi
+
(1 + |x y|)2
|x + z y| |x y| h(x y)/|x y|, zi
 .
(1 + |x y|)2
 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700 1691

Using the inequality that 2hx y, zi = |x y + z|2 |x y|2 |z|2 , we get

|x + z y| |x y| h(x y)/|x y|, zi
1 ⇣ ⌘
= 2|x y||x + z y| 2|x y|2 |x y + z|2 + |x y|2 + |z|2
2|x y|

1 ⇣ ⌘ |z|2
= |z|2 (|x y| |x y + z|)2  , (3.7)
2|x y| 2|x y|
and so
|z|2
I (x, y, z)  .
2|x y|(1 + |x y|)2
Thus, for |x y| 1/n,
Z
f n (|x y + c(x, z) c(y, z)|) f n (|x y|)
U

@x f n (|x y|) · c(x, z) @ y f n (|x y|) · c(y, z) M(dz)

R
U |c(x, z) c(y, z)|2 M(dz) ke
⌫(x, ·) e ⌫(y, ·)kVar,1
 2
= 2
f (|x y|). (3.8)
2|x y|(1 + |x y|) 2|x y| (1 + |x y|)
The required assertion follows from (3.4), (3.5) and (3.8). ⇤

In many applications, the Lévy measure ⌫(x, dz) in (1.1) may satisfy some integrability
conditions, which give us further conclusions about the coupling operator L̃. The result below is
one of the examples.

Proposition 3.2. Assume that there exist a constant 1 > > 0 such that for each x 2 Rd ,
Z
|z| ⌫(x, dz) < 1.
{|z|>1}

Then, for 0 < ↵  ,

e
L f n (|x y|)  ↵|x y|↵ 2 [2(↵ 1) + G(x, y)]
Z
+ |z|↵ |⌫(y, dz) ⌫(z, dz)| , |x y| 1/n,
{|z|>1}

where f (r ) = r ↵ , f n (r ) = f (h n (r )) and
G(x, y) = ↵ (x, y) + B(x, y) + ke
⌫(x, ·) ⌫(y, ·)kVar,1 .
e

Proof. We follow the proof of Proposition 3.1 with some modifications. By the argument of
(3.4), for |x y| 1/n, we get
e 2
L 1 f n (|x y|) = e
L 1 f (|x y|) = ↵ [2(↵ 1) + ↵ (x, y) + B(x, y)] |x y|↵ .
According to the definition of f n and the element inequality that (a + b)↵  a ↵ + b↵ for a,
b > 0,
1692 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700
Z
( f n (|x + z y|) f n (|x y|)) (⌫(x, dz) ⌫(y, dz))+
{|z|>1} Z
+ ( f n (|y + z x|) f n (|x y|)) (⌫(y, dz) ⌫(x, dz))+
Z {|z|>1}
 ( f (|x + z y|) f (|x y|)) (⌫(x, dz) ⌫(y, dz))+
Z
{|z|>1}
+ ( f (|y + z x|) f (|x y|)) (⌫(y, dz) ⌫(x, dz))+
Z {|z|>1}
 ( f (|x y| + |z|) f (|x y|)) (⌫(x, dz) ⌫(y, dz))+
Z
{|z|>1}
+ ( f (|y x| + |z|) f (|x y|)) (⌫(y, dz) ⌫(x, dz))+
Z {|z|>1}
 |z|↵ |⌫(y, dz) ⌫(z, dz)|.
{|z|>1}
On the other hand, by using the mean value theorem,
Z
( f n (|x y + c(x, z) c(y, z)|) f n (|x y|)
U
Z @x f n (|x y|) · c(x, z) @ y f n (|x y|) · c(y, z) M(dz)
 ( f (|x y + c(x, z) c(y, z)|) f (|x y|)
U
Z @x f (|x y|) · (c(x, z) c(y, z))) M(dz)
⇥ 0
 f (|x y|) (|x y + c(x, z) c(y, z)| |x y|)
U ⇤
@x f (|x y|) · (c(x, z) c(y, z)) M(dz)
Z 
= ↵|x y|↵ 1 |x y + c(x, z) c(y, z)| |x y|
U
(x y) · (c(x, z) c(y, z))
M(dz)
|x y|
↵
 |x y|↵ 2 ke ⌫(x, ·) e
⌫(y, ·)kVar,1 ,
2
where in the last inequality we also have used (3.7). The required conclusion follows from all the
estimates above. ⇤
3.2. Proofs of main results
At the beginning, we are ready to prove Theorem 2.1 by using the coupling method and
Proposition 3.1.
Proof of Theorem 2.1. The Cb (Rd )-Feller continuity is equivalent to saying that for any t 0,
Pt (x, ·) converges weakly to Pt (y, ·) as x ! y. So, in order to obtain the Cb (Rd )-Feller
continuity of Pt , thanks to [6, Theorem 5.6], it suffices to verify that
W⇢ (Pt (y, ·), Pt (x, ·)) ! 0 as y ! x.
Here,
|x y|
⇢(x, y) = f (|x y|) = ,
1 + |x y|
and W⇢ is the Wasserstein metric. i.e. for two probability measures P1 and P2 ,
Z
W⇢ (P1 , P2 ) = inf ⇢(x, y)P(dx, dy),
P
 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700 1693

where P varies over all coupling probability measures with marginals P1 and P2 . In the
following, we use the coupling operator e L and the process Z t = ( X̃ t , Ỹt ) constructed in
Section 3.1. Denote by P (x,y) and E (x,y) the distribution and the expectation of Z t starting from
(x, y), respectively. Therefore, what we need is to prove

E (x,y) f (| X̃ t Ỹt |) ! 0 as y ! x. (3.9)

First, note that condition (2.2) is equivalent to

2
K (x, y) C . (3.10)
|x y|(1 + |x y|)2
Then, according to (3.10) and (3.3),
e
L f n (|x y|)  C f (|x y|) = C f n (|x y|)
d
for any x, y 2 R with1/n  |x y|  . (3.11)
To prove (3.9), we use the standard martingale method. For 0 < < 1 and n, N > 1, set
n o
S = inf t 0 : | X̃ t Ỹt | > ,
n o
Tn = inf t 0 : | X̃ t Ỹt | < 1/n ,
n o
R = inf t 0 : | X̃ t | + |Ỹt | > N ,
T ,n,N = S ^ Tn ^ N.

For any x, y 2 Rd with |x y|  , choose n 0 > 1 such that |x y| 1/n 0 . Then, for n n0,
by (3.11), we have
⇣ ⌘
E (x,y) f n | X̃ t^T ,n,N Ỹt^T ,n,N |
Z t^T ,n,N ⇣ ⌘
= f n (|x y|) + E (x,y) L f n | X̃ s Ỹs | ds
0
Z t^T ,n,N ⇣ ⌘
(x,y)
 f (|x y|) + C E f n | X̃ s Ỹs | ds
0
Z t ⇣ ⌘
= f (|x y|) + C E (x,y) f n | X̃ s^T ,n,N Ỹs^T ,n,N | ds.
0

Applying the Gronwall inequality,

⇣ ⌘
E (x,y) f n | X̃ t^T ,n,N Ỹt^T ,N |  eC t f (|x y|).

Since the coupling process Z t is non-explosive, N ! 1 as N ! 1. Letting N " 1, we

arrive at
⇣ ⌘
E (x,y) f n | X̃ t^Tn ^S Ỹt^Tn ^S |  eC t f (|x y|).

By the dominated convergence theorem and the definitions of f n , one further has
⇣ ⌘
E (x,y) f | X̃ t^T ^S Ỹt^T ^S |  eC t f (|x y|).
1694 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700

Thus,
⇣ ⇣ ⌘ ⌘
1
(1 + ) P (x,y) (S < t ^ T )  E (x,y) f | X̃ t^T ^S Ỹt^T ^S | 1{S <t^T }
⇣ ⌘
 E (x,y) f | X̃ t^T ^S Ỹt^T ^S |
 eC t f (|x y|).
That is,
1
P (x,y) (S < t ^ T )  (1 + )eC t f (|x y|).
Now, for any t > 0, X̃ t = Ỹt a.s. if t T . So,
E (x,y) f (| X̃ t Ỹt |) = E (x,y) f (| X̃ t^T Ỹt^T |)
⇣ ⌘
= E (x,y) f (| X̃ t^T Ỹt^T |)1{S <t^T }
⇣ ⌘
+ E (x,y) f (| X̃ t^T Ỹt^T |)1{S t^T }
⇣ ⌘
 P (x,y) (S < t ^ T ) + E (x,y) f | X̃ t^T ^S Ỹt^T ^S |
1
 (1 + 2 )eC t f (|x y|),
which yields the desired assertion (3.9) by the fact that limr !0 f (r ) = 0. The proof is complete.
⇤
Remark 3.1. The application of coupling method to deal with the Cb (Rd )-Feller property has
already appeared in previous works (cf. see [6, Section 13] and [33, Section 4]). Here, we point
out the difference between [33, Section 4] and our result. As shown by the argument above,
one key step to use coupling method requires properly choosing the test function sequence
( f n (|x y|))n 1 in Proposition 3.1 and the proof of Theorem 2.1. Here, we use f (r ) = r/(1 +r )
instead of the usual function f (r ) = r 2 in [33, Section 4]. This point completely depends on
the
R characteristic of Lévy measure ⌫(x, dz) in general Lévy type operator L. For example, if
|z| 2 ⌫(x, dz) = 1 for some x 2 Rd , then the Euclidean metric g(x, y) := |x y|2 is
{|z| 1}
not well defined for the coupling operator e L. That means that the test function f (r ) = r 2 does
not work in our setting. Additionally, a little lengthy calculations and an approximating function
sequence (h n )n 1 are also needed to argue the Cb (Rd )-Feller continuity condition (2.2).
Next, we consider the Lipschitz continuity.
Proof of Proposition 2.2. The proof of Theorem 2.1 indeed gives us that for |x y|  ,
1
W⇢ (Pt (x, ·), Pt (y, ·))  (1 + 2 )eC t ⇢(x, y).
Recall that (cf. [6, Theorem 5.10])
⇢Z Z
W⇢ (P1 , P2 ) = sup f d P1 f d P2 : f 2 Lipb,⇢ (Rd ), L⇢ ( f )  1 ,

where Lipb,⇢ (Rd ) denotes the set of Lipschitz continuous (with respect to the metric ⇢) functions,
and L⇢ ( f ) denotes the Lipschitz constant of f 2 Lipb,⇢ (Rd ). Thus, for any f 2 Lipb,⇢ (Rd ) and
|x y|  ,
1 1
|Pt f (x) Pt f (y)|  (1 + 2 )eC t L⇢ ( f )⇢(x, y)  (1 + 2 )eC t L⇢ ( f )|x y|.
 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700 1695

Note that, if |x y| , then

1
|Pt f (x) Pt f (y)|  2 k f k|x y|.
Therefore, for any t 0,
1
|Pt f (x) Pt f (y)|  2(1 + )eC t C( f )|x y|,
where C( f ) is a constant only depending on k f k and L⇢ ( f ). This is, Pt f is Lipschitz continuous
with respect to the usual Euclidean metric. The proof is complete by the fact that Lipb (Rd ) =
Lipb,⇢ (Rd ). ⇤

Theorem 2.2 mainly follows from the arguments of Theorem 2.1 and Proposition 2.2, and the
details are omitted here. Finally, we follow the idea of [20, Theorem 3.4] to study the strong
Feller continuity.

Proof of Theorem 2.3. Set C ⇤ = C (1 + ) 1. Condition (2.7) gives us that for any x, y 2 Rd
with |x y|  ,
K (x, y)|x y| 2
 C ⇤. (3.12)
1 + |x y| (1 + |x y|)3

We still use the coupling operator L̃ constructed in Section 3.1 and the test function sequence
( f n )n 1 given in Proposition 3.1. According to (3.12), for any x, y 2 Rd with 1/n  |x y|  ,
it holds that
L̃ f n (|x y|)  C ⇤. (3.13)

Motivated by the proof of Theorem 2.1, we define

Tn := inf{t 0 : | X̃ t Ỹt | < 1/n};
S := inf{t 0 : | X̃ t Ỹt | > };
N := inf{t 0 : | X̃ t | + |Ỹt | N }.
Given x 6= y with |x y|  , choose n 0 1 such that |x y| > 1/n 0 . Then, for any n n0,
by (3.13),
f ( )P (x,y) (Tn ^ N > S ) = f n ( )P (x,y) (Tn ^ N >S)
(x,y)
 E f n (| X̃ Tn ^S ^ N ỸTn ^S ^ N |)
✓Z Tn ^S ^ N
◆
(x,y)
= f n (|x y|) + E L̃ f n (| X̃ s Ỹs |)ds
0
 f (|x y|) C ⇤ E (x,y) (Tn ^ S ^ N ).

That is,
f ( )P (x,y) (Tn ^ N > S ) + C ⇤ E (x,y) (Tn ^ S ^ N)  f (|x y|).

Denote the coupling time by T = inf{t 0 : X̃ t = Ỹt }. Then, Tn ! T as n ! 1. Since the

coupling process is non-explosive, it is also valid that N ! 1 as N ! 1. So,
f ( )P (x,y) (T > S ) + C ⇤ E (x,y) (T ^ S )  f (|x y|) for any x 6= y with |x y|  .
1696 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700

Then, for any t > 0 and x 6= y with |x y|  ,

(x,y) (x,y)
P (T > t) = P (T > t, S > t) + P (x,y) (T > t, S  t)
 P (x,y) (T ^ S > t) + P (x,y) (T > S )
E (x,y) (T ^ S )
 + P (x,y) (T > S )
✓ t ◆
1 1
 + f (|x y|).
C ⇤t f( )
Therefore, for any h 2 Bb (Rd ), t > 0 and x 6= y with |x y|  , by the coupling property,
(x,y)
|Pt h(x) Pt h(y)|  E |h( X̃ t ) h(Ỹt )|  2khkP (x,y) (T > t)
✓ ◆
1 1
2 + khk f (|x y|), (3.14)
C ⇤t f( )
which gives us the strong Feller continuity of the semigroups {Pt }t>0 . For any |x y| > ,
1
|Pt h(x) Pt h(y)|  2 khk|x y|.
Combining all the inequalities yields the first required assertion. The strong Feller property of Pt
immediately follows from it, and the proof is complete. ⇤

4. Comments and complements

In this section, we give some necessary remarks about our results and coupling methods for
Lévy type operators. Several conclusions included here are deduced from the approach we have
adopted in previous sections. We only present proofs of some non-standard facts for reader’s
convenience.
4.1. Situation that without using condition (3) in Assumption (H)

In this part, assume that

Z
|z|⌫(x, dz) < 1, for each x 2 Rd . (4.1)
{|z|1}

Then, we present two ways to conclude that Theorems 2.1–2.3 still hold without using condition
(3) in Assumption (H).
First, under (4.1), the operator L given by (1.1) becomes
d d Z
1 X X
L f (x) = ai j (x)@i j f (x) + bi⇤ (x)@i f (x) + ( f (x + z) f (x)) ⌫(x, dz),
2 i, j=1 i=1

where
Z
b⇤ (x) = b(x) + z⌫(x, dz).
{|z|1}

Thus, one could remove condition (3) in Assumption (H) and follow the same lines of the
arguments in Section 3 to obtain the Cb (Rd )-Feller property, the Lipschitz continuity and the
strong Feller property of Lévy type operators.
Another way to avoid condition (3) under condition (4.1) is modifying the coupling operator
e
L 2 of pure Lévy jump part into the following more natural form
 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700 1697
Z
L ⇤2 f (x, y) =
e ( f (x + z, y) f (x, y)) (⌫(x, dz) ⌫(y, dz))+
{|z|>1}
Z
+ ( f (x, y + z) f (x, y)) (⌫(y, dz) ⌫(x, dz))+
Z{|z|>1}
+ ( f (x + z, y + z) f (x, y)) (⌫(y, dz) ^ ⌫(x, dz))
Z{|z|>1}
+ ( f (x + z, y) f (x, y) @x f (x, y) · z) (⌫(x, dz) ⌫(y, dz))+
{|z|1}
Z
+ f (x, y + z) f (x, y) @ y f (x, y) · z (⌫(y, dz) ⌫(x, dz))+
Z{|z|1}
+ ( f (x + z, y + z) f (x, y) @x f (x, y) · z
{|z|1}
@ y f (x, y) · z (⌫(x, dz) ^ ⌫(y, dz)) .
Then, thanks to the proposition below, one should obtain that Theorems 2.1–2.3 still hold by
replacing ke
⌫(x, ·) e⌫(y, ·)kVar,1 with
|x ⌫ ⇤ (x, ·)
y|ke ⌫ ⇤ (y, ·)kVar,1 ,
e
where
Z
⌫ ⇤ (x, ·)
ke ⌫ ⇤ (y, ·)kVar,1 =
e |z||⌫(x, dz) ⌫(y, dz)|.
{|z|1}

Proposition 4.1. Let f and f n be functions given in Proposition 3.1 Then, for any |x y| 1/n,
we have
L ⇤ f n (|x y|) := (e
e L1 + eL ⇤2 ) f n (|x y|)

2
 + K 1 (x, y) f (|x y|), (4.2)
|x y|(1 + |x y|)2
where 
↵ (x, y) + B(x, y) + |x y|ke⌫ ⇤ (x, ·) e
⌫ ⇤ (y, ·)kVar,1
K 1 (x, y) =
|x y|2 (1 + |x y|)
k⌫(x, ·) ⌫(y, ·)kVar,>1
+ .
|x y|

Sketch of the Proof of Proposition 4.1. It is sufficient to change Part (3) in the proof of
Proposition 3.1. For any |x y| 1/n, by the definitions of e L ⇤2 and f n ,
Z
( f n (|x + z y|) f n (|x y|) @x f n (|x y|) · z) (⌫(x, dz) ⌫(y, dz))+
{|z|1}
Z 
|x + z y| |x y|

{|z|1} (1 + |x + z y|)(1 + |x y|)
h(x y)/|x y|, zi
(⌫(x, dz) ⌫(y, dz))+
(1 + |x y|)2
Z
= I (x, y, z) (⌫(x, dz) ⌫(y, dz))+ ,
{|z|1}
1698 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700

where I (x, y, z) is given by (3.6). Since

|x + z y| |x y| h(x y)/|x y|, zi
1 ⇣ ⌘
= |z|2 (|x y| |x y + z|)2
2|x y|
1
= (|z| + |x y + z| |x y|) (|z| + |x y| |x y + z|)
2|x y|
1
 (|z| + |z| + |x y| |x y|) (|x y + z|
2|x y|
+ |x y| + |x y| |x y + z|)
= 2|z|,
we arrive at
2|z|
I (x, y, z)  .
(1 + |x y|)2
Thus, for any |x y| 1/n,
Z
( f n (|x + z y|) f n (|x y|) @x f n (|x y|) · z) (⌫(x, dz) ⌫(y, dz))+
{|z|1}
Z
2
 |z| (⌫(x, dz) ⌫(y, dz))+
(1 + |x y|)2 {|z|1}
R
2 {|z|1} |z| (⌫(x, dz) ⌫(y, dz))+
 f (|x y|). (4.3)
|x y|(1 + |x y|)
Then, the desired assertion follows from (4.3) and the proof of Proposition 3.1. ⇤

Remark 4.1. One shall want to adopt the coupling operator e L ⇤ := e

L1 + eL ⇤2 above to deal with
Feller continuity of general Lévy type operators. Following the argument of Proposition 3.1,
one could get that Theorems 2.1–2.3 still hold by replacing ke ⌫(x, ·) e ⌫(y, ·)kVar,1 with
⌫ ⇤⇤ (x, ·) e
ke ⌫ ⇤,⇤ (y, ·)kVar,1 , while
Z
ke ⇤⇤
⌫ (x, ·) e ⇤⇤
⌫ (y, ·)kVar,1 = |z|2 |⌫(x, dz) ⌫(y, dz)|.
{|z|1}
However, according to the transformation (2.1), it holds that
ke
⌫(x, ·) ⌫ ⇤⇤ (x, ·)
⌫(y, ·)kVar,1  ke
e ⌫ ⇤⇤ (y, ·)kVar,1 .
e
When the coefficients in Lévy type operators are locally Lipschitz continuous, we find that as
|x y| ! 0 the term
⌫ ⇤⇤ (x, ·)
ke ⌫ ⇤⇤ (y, ·)kVar,1 ⇡ |x
e y|;
while
ke
⌫(x, ·) ⌫(y, ·)kVar,1 ⇡ |x
e y|2 ,
which is what we need in the definition of K (x, y). Therefore, the transformation (2.1) is
necessary and reasonable in dealing with general Lévy type operators. We also have explained the
reason why we use condition (3) in Assumption (H) in the construction of the coupling operator
in Section 3.
 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700 1699

4.2. Role of drift term in strong Feller continuity

The sufficient condition about the strong Feller property in Theorem 2.3 heavily depends on
the diffusion matrix. As mentioned before, there are some relationships between strong Feller
continuity and ergodic theory. It has been shown in [31,32] that the drift term b(x) plays an
important role in the ergodicity of Lévy type operators. So, one interesting question is what
effect of the drift term b(x) in studying the strong Feller property of Lévy type operators. The
coupling method also gives us part of its answer. To further illustrate the power of this technique,
we take a specific example.

Example 4.1. Consider the operator

d
X Z
L f (x) = bi (x)@i f (x) + f (x + z) f (x) r f (x) · z1{|z|1} ⌫(dz),
i=1
R
where b(x) is a continuous function and ⌫ is a Lévy measure such that {|z|>1} |z| d⌫(dz) < 1
for some > 0. Assume that the martingale problem for L is well posed, and denote its solution
by (X t , P x ). If there exist 0 < ↵ < and > 0 such that for any x, y 2 Rd with |x y|  ,
hb(x) b(y), x yi  |x y|2 ↵
,
then the process X t is strong Feller continuous.
The assertion of Example 4.1 mainly follows from Proposition 3.2 and the argument of
Theorem 2.3. Example 4.1 shows that, some more singular requirements are needed on drift
term b(x) to ensure the strong Feller continuity. That fact (more or less) points out that why
until now there remain unsatisfying answers about the strong Feller property of the processes
corresponding to pure jump Lévy type operators.

Acknowledgements

The author is very grateful to the generous support of Alexander–Humboldt Foundation, and
would like to thank Professor René L. Schilling for providing him an excellent environment to
work in TU Dresden. He also thanks an anonymous referee’s comments on the first version of
this paper.

References

[1] D. Applebaum, Lévy Processes and Stochastic Calculus, Cambridge Univ. Press, Cambridge, 2004.
[2] R.F. Bass, Markov processes with Lipschitz semigroups, Trans. Amer. Math. Soc. 267 (1981) 307–320.
[3] R.F. Bass, Stochastic differential equations with jumps, Probab. Surv. 1 (2004) 1–19.
[4] K. Bichteler, Stochastic Integration with Jumps, Cambridge Univ. Press, Cambridge, 2002.
[5] M.F. Chen, S.F. Li, Coupling methods for multidimensional diffusion processes, Ann. Probab. 17 (1989) 151–177.
[6] M.F. Chen, From Markov Chains to Non-Equilibrium Particle Systems, 2nd, World Scientific, Singapore, 2004.
[7] M.F. Chen, Eigenvalue, Inequalities and Ergodicity, Spring, Singapore, 2005.
[8] W. Hoh, Pseudo-differential Operators Generating Markov Processes, Habilitationsschrift, Universität Bielefeld,
1998.
[9] N. Ikeda, S. Watanabe, Stochastic Differential Equations and Diffuison Processes, North-Holland, Amsterdam,
1981.
[10] N. Jacob, Pseudo-differential Operators and Markov Processes, vol.1, Imperial College Press, London, 2001.
[11] N. Jacob, Pseudo-differential Operators and Markov Processes, vol.2, Imperial College Press, London, 2002.
[12] N. Jacob, Pseudo-differential Operators and Markov Processes, vol.3, Imperial College Press, London, 2005.
1700 J. Wang / Stochastic Processes and their Applications 120 (2010) 1680–1700

[13] N. Jacob, R.L. Schilling, Lévy-type processes and pseudo-differential operators, in: O.E. Barndorff-Nielsen, et al.
(Eds.), Lévy Processes: Theory and Applications, Birkhäuser, Boston, 2001, pp. 139–168.
[14] N.EI. Karoui, J.P. Lepeltier, Representation des processus ponctuels multivariés à l’aide dun processus d’e Poisson,
Z. Wahrscheinlichkeitstheor. Verwandte Geb. 39 (1977) 111–133.
[15] T. Komorowski, S. Peszat, T. Szerek, On ergodicity of some Markov processes, Ann. Probab. (2009) (in press)
http://arxiv.org/abs/0810.4609.
[16] A. Kulik, Exponential ergodicity of the solutions to SDE’s with a jump noise, Stochastic Process. Appl. 119 (2009)
602–632.
[17] Y. Kwon, C. Lee, Strong Feller property and irreducibility of diffuisons with jumps, Stoch. Stoch. Rep. 67 (1999)
147–157.
[18] H. Masuda, Ergodicity and exponential -mixing bound for multidimensional diffusions with jump, Stochastic
Process. Appl. 117 (2007) 36–56; Stochastic Process. Appl. 119 (2007) 676–678. (erratum).
[19] H. Masuda, On stability of diffusions with compound-Poisson jumps, Bull. Inform. Cybernet. 40 (2008) 61–74.
[20] E. Priola, F.Y. Wang, Gradient estimates for diffusion semigroups with singular coefficients, J. Funct. Anal. 236
(2006) 244–264.
[21] S. Peszat, J. Zabczyk, Strong Feller property and irreducibility for diffusion on Hilbert spaces, Ann. Probab. 23
(1995) 157–173.
[22] R.L. Schilling, Conservativeness and extensions of Feller semigroups, Positivity 2 (1998) 239–256.
[23] R. Situ, Theory of Stochastic Differential Equations with Jumps and Applications, Springer, New York, 2005.
[24] L. Stettner, On the existence and uniquness of invariant measure for continuous time Markov processes, Technical
Report, Lefschetz Center for Dynamical Systems Preprint, 86-18, Brown University, Providence, RI, 1986.
[25] L. Stettner, Remarks on ergodic conditions for Markov processes on Polish spaces, Bull. Pol. Acad. Sci. Math. 42
(1994) 103–114.
[26] D. Stroock, Diffusion processes associated with Lévy generators, Z. Wahrscheinlichkeitstheor. Verwandte Geb. 32
(1975) 209–244.
[27] D.W. Stroock, Markov Processes from K. Itô’s Prespective, Princeton Univ. Press, Pronceton and Oxford, 2003.
[28] D.W. Stroock, S.R.S. Varadhan, Multidimensional Diffusion Processes, Springer, New York, 1979.
[29] M. Tsuchiya, Lévy measure with generalized polar decomposition and the associated SDE with jumps, Stoch.
Stoch. Rep. 38 (1992) 95–117.
[30] F.Y. Wang, Functional Inequalities, Markov Semigroups and Spectral Theory, Science Press, Beijing, 2004.
[31] J. Wang, Criteria for ergodicity of Lévy type operators in dimension one, Stochastic Process. Appl. 118 (2008)
1909–1928.
[32] J. Wang, Stability of Markov processes generated by Lévy type operators, preprint, TU Dresden, 2009.
[33] F.B. Xi, On the stability of jump-diffusions with Markovian switching, J. Math. Anal. Appl. 341 (2008) 588–600.