2
$\begingroup$

Set-up and question. Let $\mathcal{X}$ be a complete separable metric space which is not locally-compact. Let $V: \mathcal{X} \to [0; +\infty]$ be a function and $(X_t)_{t\geq 0}$ a Markov process in $\mathcal{X}$ such that $P\{ V(X _t) < \infty \text{ for all } t \geq 0 \} = 1$. Let $g: \mathcal{X} \to \mathbb{R}$ be such that $$ V(X _t) - \int ^t _0 g(X _s) ds \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \quad \quad \quad \quad \quad \quad \quad \quad (1) $$ is a martingale, where $$ g(x) \leq c - V(x), \quad \text{ for all } x \in \{z \in \mathcal{X}: V(z) < \infty \} \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ (2) $$ for some $c >0$. Is it possible to deduce that $$ \limsup\limits _{t \to \infty} \mathbb{E} V(X _t) < \infty? \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ (3) $$

My thoughts. If $V$ was 'norm-like' in the sense that $\{x \in \mathcal {X} : V (x) \leq r \}$ was compact for $r >0$ and $\mathcal{X}$ was locally-compact, we would get ergodicity from (1) and (2) and thus (3) would hold. I think that without $V$ being norm-like and local compactness of $\mathcal{X}$ ergodicity cannot follow from (1) and (2), however a weaker property like (3) should hold. I have an idea how to prove (3) if additionally $(X_t)_{t\geq 0}$ is assumed to be continuous a.s. (finding an ergodic birth-death process which would stochastically dominate $(V(X_t))_{t\geq 0}$), however without continuity assumption I am currently out of ideas. I did not find any similar result in the literature, but any related reference would be kindly appreciated.

Similar posts. The set-up is similar to this, however the assumptions and the questions are different.

Edit Edited the condition on $g$. The initial version was not well formulated and resulted in a triviality as was pointed out by Iosif Pinelis.

$\endgroup$
1
$\begingroup$

The conditions $g\ge0$, (2), and $0\le V(X_t)<\infty$ almost surely (a.s.) imply $V(X_t)\le c$ a.s. (Indeed, on the event $\{V(X_t)<\infty\ \forall t\ge0\}$ -- which is of probability $1$, we have $0\le g(X_t)\le c-V(X_t)$ and hence $0\le c-V(X_t)$ and $V(X_t)\le c$.) So, $$\limsup_{t\to\infty}EV(X_t)\le c<\infty$$ if $c<\infty$.

Added in response to the edit of the OP: Even when the condition $g\ge0$ is removed, the conclusion (3) holds. Indeed, letting $h(t):=Eg(X_t)$ and $v(t):=EV(X_t)$, and using the martingale condition, we have \begin{equation} v(t)-\int_0^t h(s)\,ds=E\Big(V(X_t) - \int^t_0 g(X_s)\, ds\Big)=EV(X_0), \end{equation} whence $v'=h$. Also, by (2), \begin{equation} h(t)=Eg(X_t)\le c-EV(X_t)=c-v(t), \end{equation} whence $v'(t)+v(t)\le c$, or, for $u(t):=v(t)e^t$,
\begin{equation} u'(t)\le ce^t \end{equation} and hence $u(t)\le u(0)+ce^t$, that is, \begin{equation} EV(X_t)=v(t)\le u(0)e^{-t}+c, \end{equation} so that \begin{equation} \limsup_{t\to\infty}EV(X_t)\le c<\infty, \end{equation} as desired.

$\endgroup$
  • $\begingroup$ Thank you for an elegant argument. I deleted my previous comments since I think they don't contribute to the post. $\endgroup$ – Sinusx Oct 25 '18 at 15:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.