$${\phi}_{t}:f\mapsto {e}^{rt}{P}_{t}f\vee g$$
for
$t\ge 0$
. If
$f\le h$
,
${P}_{t}f\le {P}_{t}h=h$
for all
$t\ge 0$
, since the operators
${P}_{t}$
are positive and linear, and
$h$
was assumed to be harmonic. Thus, since
$g\le h$
and
$r>0$
, one must have
${\phi}_{t}f\le h$
for all
$f\le h$
and
$t\ge 0$
. Moreover, the operators
${P}_{t}$
preserve subharmonicity and the maximum of two subharmonic functions is subharmonic again, therefore
${\phi}_{t}f$
must be subharmonic for all subharmonic
$f$
. Finally, since
${P}_{t}$
is monotone,
${\phi}_{t}$
has to be monotone for all
$t\ge 0$
Summarising this, we obtain
Lemma 8.
Using the notation introduced previously,
${\phi}_{t}:Q\to Q$
and whenever
${f}_{0}\le {f}_{1}$
,
${\phi}_{t}{f}_{0}\le {\phi}_{t}{f}_{1}$
for all
$t\ge 0$
.
As a consequence, we derive from the two Lemmas 7 and 8 the following:
Corollary 3.
If we define
${\mathcal{K}}_{t}:=\mathcal{\mathcal{R}}\circ {\phi}_{t}$
(adopting the notation of the previous paragraph), we have
${\mathcal{K}}_{t}:Q\to Q$
and whenever
${f}_{0}\le {f}_{1}$
,
${\mathcal{K}}_{t}{f}_{0}\le {\mathcal{K}}_{t}{f}_{1}$
.
This already suffices to prove the following
Theorem 2.
Let
$t\ge 0$
. Then for all
$n\in {\mathbb{N}}_{0}$
,
$$\begin{array}{c}{{\mathcal{K}}_{t}}^{n+1}(g\vee 0)\ge {{\mathcal{K}}_{t}}^{n}(g\vee 0).\end{array}$$ 
(2)

Furthermore,
$$q:={sup}_{n\in {\mathbb{N}}_{0}}{{\mathcal{K}}_{t}}^{n}(g\vee 0)$$
(which a priori is only defined as a function with range in
$\mathbb{R}\cup \{+\infty \}$
) is an element of
$Q$
and indeed is the least nonnegative fixed point of
${\mathcal{K}}_{t}$
.

Proof.

1.
The proof of equation ( 2 ) is a straightforward induction on
$n$
where we have to use the monotonicity of
${\mathcal{K}}_{t}$
in the induction step.

2.
Since
${\mathcal{K}}_{t}$
maps
$Q$
to itself, the whole sequence
${\left({{\mathcal{K}}_{t}}^{n}(g\vee 0)\right)}_{n\in {\mathbb{N}}_{0}}$
is bounded by
$h$
. This entails
$q\le h$
as well. Applying Beppo Levi’s Theorem on swapping
$sup$
and
$\int \cdot d\mu $
– for bounded monotonely increasing sequences of measurable nonnegative functions and an arbitrary measure
$\mu $
– to the measures
${P}_{t}(\cdot ,x)$
,
$x\in {\mathbb{R}}^{d}$
and the sequence
${\left({{\mathcal{K}}_{t}}^{n}(g\vee 0)\right)}_{n\in {\mathbb{N}}_{0}}$
, we can exploit the subharmonicity of the functions
${{\mathcal{K}}_{t}}^{n}(g\vee 0)$
,
$n\in {\mathbb{N}}_{0}$
, to deduce
$$\begin{array}{ccc}\forall x\in {\mathbb{R}}^{d}{P}_{t}q\left(x\right)& =& {sup}_{n\in {\mathbb{N}}_{0}}{P}_{t}\left({{\mathcal{K}}_{t}}^{n}(g\vee 0)\right)\left(x\right)\end{array}$$  
$$\begin{array}{ccc}& \ge & {sup}_{n\in {\mathbb{N}}_{0}}{{\mathcal{K}}_{t}}^{n}(g\vee 0)\left(x\right)=q\left(x\right),\end{array}$$  
which is the subharmonocity of
$q$
. As we have already seen,
$q\le h$
, so
$q\in Q$
.

3.
If we employ Beppo Levi’s Theorem again, we can show that
${\mathcal{K}}_{t}$
and
${sup}_{n\in {\mathbb{N}}_{0}}$
commute for bounded monotonely increasing sequences of functions.
Thereby
$${\mathcal{K}}_{t}q={sup}_{n\in {\mathbb{N}}_{0}}{\mathcal{K}}_{t}{{\mathcal{K}}_{t}}^{n}(g\vee 0)={sup}_{n\in \mathbb{N}}{{\mathcal{K}}_{t}}^{n}(g\vee 0)=q.$$

4.
Any nonnegative fixed point
$p$
of
${\mathcal{K}}_{t}$
must be greater or equal
$g\vee 0$
.
Therefore by the monotonicity of
$sup$
and
${\mathcal{K}}_{t}$
,
$${sup}_{n\in {\mathbb{N}}_{0}}{{\mathcal{K}}_{t}}^{n}p\ge {sup}_{n\in {\mathbb{N}}_{0}}{{\mathcal{K}}_{t}}^{n}(g\vee 0)=q.$$
Example 2 (Bermudan put option with equidistant exercise times in
$t\cdot {\mathbb{N}}_{0}$
on the weighted arithmetic average of a basket in a special BlackScholes model).
Let
${\beta}_{1},...,{\beta}_{d}\in [0,1]$
be a convex combination and
${\left({P}_{t}\right)}_{t\ge 0}={\left({\nu}_{t/2,t}\right)}_{t\ge 0}$
. Then one has
$$L=\frac{1}{2}\Delta \left(\frac{1}{2},\cdots ,\frac{1}{2}\right)\cdot \nabla $$
(cf e g Revuz and Yor’s exposition [
5]
), and both
$$g:x\mapsto K{\sum}_{i=1}^{d}{\beta}_{i}exp\left({x}_{i}\right)$$
and
$h:=K$
(where
$K\ge 0$
is the option’s strike price) are harmonic.
Moreover, obviously
$g\le h$
. This allows us to deduce that whatever closed
${F}_{\sigma}$
set
$B\subset {\mathbb{R}}^{d}$
we choose (for the sake of computational efficiency one could think of a triangulisation for instance), the (perpetual) Bermudan option pricing algorithm that iteratively applies
${\mathcal{K}}_{t}$
(where
$t>0$
is the exercise mesh size of the option) to the payoff function
$g\vee 0$
on the
$log$
price space, will monotonely increase and will have a limit which is the smallest nonnegative fixed point of
${\mathcal{K}}_{t}$
.
Acknowledgements. This work originates from research conducted by the author for his doctoral thesis at the University of Oxford. The author is highly indebted to Professor Terry Lyons for numerous helpful discussions. Furthermore, he gratefully acknowledges funding from the German Academic Exchange Service (Doktorandenstipendium des Deutschen Akademischen Austauschdienstes).
References

K Ito, H P McKean Jr, Diffusion processes and their sample paths, Grundlehren der mathematischen Wissenschaften 125, Springer, Berlin 1974.

T J Lyons, personal communication.

PA Meyer, Probabilites et potentiel, Actualites scientifiqes et industrielles 1318, Hermann, Paris 1966.

S C Port, C J Stone, Brownian motion and classical potential theory, Academic Press, New York 1978.

D Revuz, M Yor, Continuous martingales and Brownian motion, 3rd ed, Grundlehren der mathematischen Wissenschaften 293, Springer, Berlin 1999.

D Stroock, Probability theory – an analytic view, Cambridge University Press, Cambridge 1993.