The following is of Philip Protter at page 26 of the book Stochastic integration and Differential equations that I have not been able to proved yet.

Let $X$ a Levy process, and $\Lambda$ a borel set in $\mathbb{R}$ away from $0$ (that is, $0 \notin \bar{\Lambda}$), then $\nu(\Lambda) = E[N_{1}^{\Lambda}] < \infty$. Where $N_{t}^{\Lambda} = \sum_{0 < s \leq t} 1_{\Lambda}(\Delta X_{s})$. Protter says that it follows from Theorem 34, but for me it means I have to prove first that the jumping times have independent and stationary increments something that I am not sure.

My attempt I assume that the jumping times $T_{\Lambda}^{1} := \lbrace t >0 : \Delta X_t \in \Lambda \rbrace, \cdots, T_{\Lambda}^{n}:= \lbrace t > T_{\Lambda}^{n-1}: \Delta X_{t} \in \Lambda \rbrace,...$ have independent and stationary increments. Therefore, taking into account that $ T_{\Lambda}^{n} = T_{\Lambda}^{1} + \sum_{k=1}^{n} T_{\Lambda}^{k} - T_{\Lambda}^{k-1} $ we get

\begin{align} \nu(\Lambda) &= E[N_{1}^{\Lambda}] \\ &=\sum_{n=1}^{\infty}P[T_{\Lambda}^{n} \leq 1] \\ &\leq \sum_{n=1}^{\infty}P[T_{\Lambda}^{1} \leq 1, ..., T_{\Lambda}^{n} - T_{\Lambda}^{n-1} \leq 1 ] \\ &\leq \sum_{n=1}^{\infty}(P[T_{\Lambda}^{1} \leq 1])^{n} < \infty \end{align} and the last part depends on the fact $P[T_{\Lambda}^{1} \leq 1] < 1$ something that I am not quite sure. Apparently, this is not at all an intuitive fact.

Any hint will be welcome.


Since the Lévy process $(X_t)_{t \geq 0}$ has right-continuous paths with probability $1$, we can choose $t>0$ such that $\mathbb{P}(T_{\Lambda}^1 \leq t) < 1$. Following the reasoning in your question we find

$$t \nu(\Lambda) = \mathbb{E}(N_t^{\Lambda}) \leq \sum_{n \geq 1} (\mathbb{P}(T_{\Lambda}^1 \leq t))^n < \infty$$

and so $\nu(\Lambda)<\infty$.

  • $\begingroup$ Thanks @saz for the insights. $\endgroup$ – Ivan Apr 2 '18 at 23:48

It seems that the times $(T_{\Lambda}^{n})$ does not have independent and stationary increments for any Borel set $\Lambda$ away from zero. Fortunately, by the same arguments made by @saz the proof can be done.

Proof of the problem: Because $\Lambda$ is away from zero, there exists a $C > 0$ such that $\Lambda \subset \Gamma_{C}$, where $\Gamma_{C}:= (-\infty, -C] \cup [C, \infty) $. If we denote

\begin{align} T_{\Gamma_{C}}^{1} := \inf\lbrace t: X_{t} \in \Gamma_{C}\rbrace \cdots T_{\Gamma_{C}}^{n} := \inf\lbrace t > T_{\Gamma_{C}}^{n-1}: X_{t} \in \Gamma_{C} \rbrace \cdots , \end{align}

these are stopping times that are stationary and have independent increments as @saz proved in one of these related questions. Therefore, we get

\begin{align} P(T_{\Gamma_{C}}^{n} < t) \leq \dfrac{E[e^{-T_{\Gamma_{C}}^{n}}]}{e^{-t}} = \dfrac{(E[e^{-T_{\Gamma_{C}}^{1}}])^{n}}{e^{-t}} \leq e^{t} \epsilon^{n} \end{align} for some $\epsilon $, $0 \leq \epsilon < 1$.

Now, using the equality \begin{align} N_{t}^{\Lambda} = \sum_{0 < s \leq t}1_{\Lambda}(\Delta X_{s}) = \sum_{n=1}^{\infty} 1_{ \lbrace T_{\Lambda}^{n} \leq t \rbrace} \end{align} we have \begin{align} N_{t}^{\Lambda} = \sum_{n=1}^{\infty} 1_{ \lbrace T_{\Lambda}^{n} \leq t \rbrace} = \sum_{0 < s \leq t}1_{\Lambda}(\Delta X_{s}) \leq \sum_{0 < s \leq t}1_{\Gamma_{C}}(\Delta X_{s}) = \sum_{n=1}^{\infty} 1_{ \lbrace T_{\Gamma_{C}}^{n} \leq t \rbrace} \end{align} As a result, \begin{align} \nu(\Lambda) &= E[N_{1}^{\Lambda}] \\ &=\sum_{n=1}^{\infty}P[T_{\Lambda}^{n} \leq 1] \\ &\leq \sum_{n=1}^{\infty}P[T_{\Gamma_{C}}^{n} \leq 1 ] \\ &\leq \sum_{n=1}^{\infty}(P[T_{\Gamma_{C}}^{1} \leq 1 ])^{n} \\ &\leq e\sum_{n=1}^{\infty} \epsilon^{n} < \infty \end{align}


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.