1.

Introduction
The classical potential theory provides a very close connection between theory of
probability and mathematical calculus. A very known is a powerfull Feymann-
Kac theorem which very formally speaking allows us to interpret the solution of
the partial differential equation of a specific form as an expected value of the
functional of Brownian motion stopped at a certain time. The link between these
two branches of mathematics is however much wider. We can look at potential
theory from four different angles via Hunt processes, sub-Markov semigroups,
families of harmonic kernels and balayage space. This is stated in Theorem IV.8.1
in Bliedtner, Hansen [2]).
Our motivation comes from the interest in the family of functions which
solves generalized Dirichlet problem. Consider for example a heat equation. The
functions which are harmonic with respect to the heat equation are a special
example of the general class of harmonic kernels. The harmonic functions are also
closely related with the class of Brownian semigroups. Our intention is to study
Levy processes. The later covers large class of stochastic processes which
however still owns very nice and handable properties.
The article is organized as follows. We firstly introduce notation and give
definitions of balayage space, sub-Markov semigroups, excessive functions,
hyperharmonic functions, family of harmonic kernels and Hunt processes in the
section Preliminaries. The reader familiar with these terms may prefer to skip this
part. In the second section we state a rather general theorem from Bliedtner J.,
Hansen W. [2]. In the last section we recall the definition of Levy process and use
the theorem from the second part to formulate a consequence of it for Levy
processes as a special case of Hunt processes.
The introduced theory is of great importance not only for the theoretical
studies. It has great applicability to optimal stopping problems and stochastic
control problems and consequently is of interest e.g. in the pricing of the
American contingent claims. For an introduction to the latter topics and Dirichlet
problem the reader may appriciate chapters 9-12 from Oksendal [3].

2. Preliminaries
We consider a locally compact space X with a countable base and denote
U a base for the topology on X consisting of relatively compact open sets in X.
We further denote by C(X) space of all continuous real functions on X and Co(X)
space of all continuous real functions vanishing in infinity. We also fix the
probability space (Ω, F, P) on which live considered stochastic processes.
In the following we explain what is a balayage space. Let us consider W a
convex cone of positive lower semicontinuous numerical functions on X. For v 
W we denote 𝑣̂ the lower regularization of v, i.e. 𝑣̂(x) := lim infy x v(y), xX.
The coarsest topology which is at least as fine as the initial topology and
for which all the functions from W are continuous will be called (W)-fine
topology.
Observe that functions from a convex cone W are missing the continuity
property in the initial topology U of locally relatively compact open sets and thus
we take new topology such that we add some sets to topology U in which the
functions from W will be continuous.
By 𝑣̂f we will denote the lower regularization of the function v in the W-
fine topology.
Definition 1. (X, W) is a balayage space, if the following properties are satisfied:
1. W is -stable, i.e. for every increasing sequence (vn) of functions from W the
supremum sup vn is in W.
̂
2. inf 𝑉 𝑓  W for every subset V of W.
3. if u, v’, v”  W such that u < v’ + v”, then there exists u', u”  W such that u
= u’ + u" and u’ < u’, u” < v”.
4. W is linearly separating, i.e x,y  X, x ≠ y and λ > 0 exists v  W, s.t. v(x) ≠
𝑣
λv(y) and there exist positive uo, vo  W  C(X) s.t. 𝑣0  Co(X) and v = sup{u
0
 W  C(X) : u < v} for every v  W.
As an example consider X a discrete space (at most countable) and W the set
of all positive numerical functions on X, then (X,W) is a balayage space.
Just for completness recall what we understand by the sub-Markov (resp.
Markov) semigroup. We also explain what are excessive functions.
Definition 2. A family P = (Pt)t>o of kernels on X is a semigroup, if Ps+t = PsPt for
s, t > 0.
It is a sub-Markov (reap. Markov), if for every t > 0 PtI < I (resp.PtI = I).
Further note, that having define a semigroup of Markov kernels we will
call a positive borel measurable function f  B+ (X) excessive with respect to the
Markov semigroup P (P-excessive) if supt>0 Ptf = f. The set of all P-excessive
functions we denote by Ep := {f  B+(X) : supt>0 Ptf = f}.
Now, we describe what are families of harmonic kernels and give definiti-
ons of hyperharmonic and superharmonic functions.
Consider a family of kernels (HU)UU on X. We call (HU)UU a family of
sweeping kernels (relative to U) if for every set U from the base U of X, we have a
kernel HU on X for which holds that HU (x,U) = 0 for every x  U and for every x
UC : HU (x,.) = x where x denotes dirac mass.
Now for every open subset V of X we denote U(V) the set of all open
subsets W  V such that closure W is a compact set in V. Let H* (V) denote the set
of all positive hyperharmonic functions on V, that is the set of all positive borel
measurable functions v on X, such that v is lower semicontinuous on V and —  <
Huv < v for all U  U(V). We also denote by S* (V) the set of all positive
superharmonic functions on V, i.e. the set of all positive hyperharmonic functions
w on V for which Huw|u  C(U) for all U  U(V).
We call (Hu)uu a family of harmonic kernels if the following axioms are
fullfiled:
1. x  X, limux HuI(x) = I
2. HvHu = Hu for all U, V  U and 𝑉̂  U.
3. For all U  U and f  B(X) which are bounded on U the function Huf is
continuous on U.
4. for U  U and every x  U there exists a hyperharmonic function w s.t. w(x) <
 and limA =  for every non-regular ultrafilter A on U.
5. the space of positive hyperharmonic functions H*(X) is linearly separating and
there exists a strictly positive function so  S* (X) C(X).
For better understanding of the definition reader can verify (or read e.g. in
Bliedtner, Hansen that heat kernel in the classical Dirichlet problem is a harmonic
kernel.
Finally we need to define Hunt process. As we will see from the definition it is
a rather general stochastic process. However it still posses a Markov property.
Definition 3. A markov process Z = (Z(t), t > 0) is called a Hunt process if the
following conditions are satisfied:
1. the trajectories of Z(t) are right-continuous on t > 0 and have limits from
left for 0 < t < T for T  R+
2. (Z(t), t > 0) has the strong markov property w.r.t to the augmented natural
filtration (𝑔𝑡𝑧 , t > 0), i.e. given any 𝑔𝑡𝑧 -adapted stopping time r and s > 0,
B  B(Rd) the following holds:
P(Z(r + s)  B|𝑔𝑡𝑧 ) = P(Z(s)  B|Z(r))
3. (Z(t), t > 0) is quasi-left continuous, i.e. whenever (rn) is sequence of 𝑔𝑡𝑧 )-
adapted stopping times and r is 𝑔𝑡𝑧 -adapted stopping time, s.t. rn then
lim 𝑍(𝑟𝑛 ) = Z(r) a.s. on [r < ]
𝑛→∞
We will further denote by Ez the set of all functions excessive with respect to
Markov semigroup P of Hunt process Z.

3. Theorem
The following theorem provides us with a general result which allows us to
describe the cone W of excessive functions for sub-Markov processes, where all
excessive functions are lower semicontinuous and positive hyperharmonic
functions for a family of harmonic kernels on X. Further (X, W) is a balayge space.
Theorem 1. Let P  C+ (X) be a function cone and W = S(P) := {sup fn : (fn)  P
increasing} s.t. I  W. Then the following statements are equivalent:
1. (X, W) is a balayage space.
2. there exists a family (HU)uu of harmonic kernels on X such that H*(X) = W.
3. W is min-stable, i.e for every f, g  W the infimum inf(f, g)  W and there
exists a sub-Markov semigroup P = (Pt)t>o on X s.t. Ep = W
4. there exists a Hunt process Z = (Z(t), t > 0) with state space (X, X) s.t. Ez = W.
Proof: see Bliedtner J., Hansen W. (1986), Theorem W.8.1, p.168.

4. Application to Levy processes

Levy processes are an example of Markov processes with more restrictive con-
ditions. They are also closely related to convolution semigroups of measures such
that the distribution of the Levy process forms a convolution semigroup. To have
nice insight about their behaving, we can also describe them as processes which
can in general consist from the process with finite variation, a Brownian motion
part and jump process which has countable many jumps. Each of these component
can be missing so for example Brownian motion or Poisson process or mixture of
these is Levy process.
Definition 4. A stochastic process Z = (Z(t), t > 0) on X is a Levy process if the
following conditions holds:
1. Z(0) = 0 a.s
2. (Z(t), t > 0) has independent and stationary increments
3. (stochastic continuity)
 > 0 and all h > 0 lim 𝑃(|𝑍(𝑡 + ℎ) − 𝑧(ℎ)| > ) = 0
𝑛→∞
4. (cadlag property of trajectories)
(Z(t), t > 0) is right continuous in t > 0 and has left limits in t > O.
The following proposition which we can formulate due to Theorem 1 gives us
properties for the set of all excessive functions with respect to considered Levy
process such that the set is closed under infimum and supremum and can be
characterized via hyperharmonic functions.
Proposition 1. For a function cone P  C+(X) and W = S(P) s.t. I  W and Levy
process Z = (Z(t), t > 0) with the state space (X, X), s.t. Ez = W there exists a
family (Hu)uu of harmonic kernels on X, s.t. H* (X) = W, W is min-stable and -
stable and there exists a sub-Markov semigroup P = (Pt)t>o on X, s.t. Ep = W.
Proof: The fact that Levy process is Hunt process follows from definitions
of these processes. Note, that Levy process is a Markov process (see e.g. Sato K.)
and every Levy process is a Feller process and every Feller process is a Hunt
process (Applebaum D.). The rest is a consequence of Theorem 1.
References
1. Applebaum D. (2004). Levy processes and stochastic calculus. Cambridge
University Press,
2. Bliedtner 1, Hansen W. (1986). Potential theory - an analytic and probabilistic
approach to balayage. Springer-Verhuig, Berlin.
3. Oksendal B.K. (2000). Stochastic differential equations: an introduction with
applications, 5th Edt. Springer-Verlang, Berlin Heidelberg New York.
4. Sato K. (1999). bevy processes and infinitely, divisible distributions.
Cambridge University Press.
Abstract. For a large class of nonlinear perturbations of balayage spaces
existence and uniqueness of solutions to the Dirichlet problem are shown. In the
special case of a harmonic space given by a linear second order partial differential
operator L the perturbed equation is Lu - (• ,u)u = 0 where (x,t) is continuous
in t  R, the functions c := sup{|(•,t)| :-c < t < c}, c > 0, are locally -Kato, i.e.,
𝑐
yield continuous real L-potentials u𝐺𝐿 , and the functions t  (x,t), x  X, have
a weak form of joint lower Lipschitz property, i.e.,  := sups<t,((•, t) - (•,s))-/(t-
s) is locally -Kato and perturbation by - still leads to a P-harmonic structure.
1. Introduction and basic notions
Recently the Dirichlet problem for nonlinear perturbation of partial
differential equations of the type
Lu - u(• ,u) = 0
(L being a linear elliptic or parabolic operator of second order) has been studied in
the potential-theoretic setting of a harmonic space ([BM], [BBM]).
We shall be able to weaken the assumptions, our model case being
Lu - (• ,u) = 0
(cf. (2.2)), to get better results, and, nevertheless, have shorter proofs.
In fact, our method works even for balayage spaces (see [BH]) covering in
addition nonlocal situations as e.g. given by Riesz potentials. So we shall study
nonlinear perturbation of balayage spaces. The reader who is mainly interested in
the PDE case leading to harmonic spaces (which can be viewed as balayage
spaces where harmonic measures for open sets live on the boundaries) might
consult [He], [CC], [811], [KJ, [HIT] and [Bo].
It will be convenient to assume that our balayage space has a base of
regular sets. We adopt the same notations as in [Ha] and recall briefly the basic
definitions:
Let X be a locally compact space with countable base. For every open set
U in X , let B(U) denote the set of all numerical Borel measurable functions on U.
Further, b(U) will denote the space of all continuous real functions on U and K(U)
the set of all functions in b(U) having compact support in U. Occasionally,
functions on U will be identified with functions on X which are zero on Uc.
Finally, given any set A of functions let Ab (A+ resp.) denote the set of all
functions in A of which are bounded (positive reap.).
Let U be a base of relatively compact open subsets of X and, for every U
 U, let Hu be a kernel on X such that Hu (x, •) = x for every x  Uc and Hulu = 0.
Let us suppose that U is stable with respect to finite intersections (by [BH,
Remark VII.3.2.4] this is no restriction of generality). Define
(1.1) W := {v|v : X  [0, ] 1.s.c., Huv < for every U  U}
and, for every numerical function f > 0, on X , let
Rf := inf{v  W : v > f}.
A function s  b+ (X) is called strongly (W)-superharmonic if, for every U  U,
Hus < s on U.
Then (Hu)uv is a family of (regular) harmonic kernels and (X, W) is a
balayage space provided the following holds (where U, V  U):
(H1) Given x  X , limu{x} Hu f (x) = f(x) for all f  K (X) or R1{x} is l.s.c. at x.
(𝐻2′ ) HvHu = Hu if V  U.
(H3) For every f  Bb(X) with compact support, the function Hu f is continuous
on U.
(𝐻4 ) For every f  K (X), the function Hu f is continuous on 𝑈
′ ̅.
(𝐻5 ) There exists a strongly superharmonic function s  b (X).
′ +

In the following (X, W) will always denote a balayage space associated with
a family (Hu)uv of regular harmonic kernels. For simplicity let us suppose that
there exists a strictly positive bounded function in W. This is no real loss of
generality, since we may always replace W by {w/s : w  W}, s being an arbitrary
strictly positive function in W  b (X). (Note, however, that we would have to
replace Bb(X) by the space of all s-bounded functions.)
For every open subset U of X let H(U) denote the set of all harmonic
functions on U, i.e.,
H(U) := {h  B(X) : h|u continuous, Hvh(x) = h(x)
for every x  V  U, 𝑉̅  U}.
One way of defining the convex cone P(X) of all continuous real potentials is the
following:
P(X) := {p  W  b(X) : 0 < g < p, g  H+ (X)  g =0}.
We recall from [BH] that for every open subset U of X, regular or not, bounded or
not, there is a harmonic kernel Hu. It is characterized by
Hup = inf{q  P(X) : q > p on Uc} (p  P(X)).
By definition, a sequence (xn) in U converging to a point z  U is called regular
if lim 𝐻𝑢𝑓(𝑥𝑛) = 𝑓(𝑧) for every f  K (X). If it is regular, then lim 𝐻𝑢𝑓(𝑥𝑛) =
𝑛→∞ 𝑛→∞
𝑓(𝑧) even for every f  Bb(X) such that the restriction of f on Uc is continuous at
z. The set U is regular, if every sequence in U converging to a boundary point of
U is regular. In particular, every U  U is regular by (𝐻4′ ).
Let us fix a potential kernel Kx for (X, W), i.e., Kx is a kernel such that
(1.2) K x f  P(X)  H (X\supp(f)) for f  𝐵𝑏+ (X) with compact support.
For every open subset U of X, we define a kernel Ku by
(1.3) Ku := Kx — HuKx.
Since obviously Ku (x,.) = 0 for every x  Uc, we may view Ku as being a kernel
on X or a potential kernel on U (restricting Hv (x, •) on U for x  V  U with 𝑉̅ 
U we obtain a family of harmonic kernels on U). We recall that Ku is a compact
operator on Bb(X) if U is relatively compact (this follows easily from [Ha, Lemma
10.1] and (1.3)). Moreover,
(1.4) Ku = Kv + HvKu
for all open U, V with V  U.
A function f  B(X) is called a Kato function (with respect to Kx) if KxMf
is a potential kernel (Mf denotes multiplication by f) or, equivalently, if
Kx(1Unf+)  b(X)
for a sequence (Un) of open sets covering X. Of course, every locally bounded
function in B(X) is a Kato function. More generally, every f  B(X) which is
locally bounded by a Kato function is a Kato function.
 Remark 1.1. If the balayage space (X, W) is given by a second order
differential operator L with Green function GL (such that LGL(•. y) = -y), then
potential kernels are associated with (Kato) measures  by K x f = := ∫ 𝐺𝐿 (•.
y)f(y)(dy).