Suppose I have two independent Lévy processes $(X_t)_t$ and $(Y_t)_t$, both not continuous.

Is anyone familiar and can refer me to a result (or a counterexample) which states that

${\displaystyle \sum_{0\leq s\leq t}}|\Delta X_{s}\cdot \Delta Y_{s}|=0 $ for all $t\in \mathbb{R}$ a.s?

A different yet equivalent formulation of this is

$\Delta X_{t}=0$ or $\Delta Y_{t}=0$ a.s. for all $t\in \mathbb{R}$

In words, every two independent Lévy processes have no simultaneous jumps a.s. I know it holds for independent Poisson processes and I'm wondering if it generalizes.


Let $(X_t)_{t \geq 0}$, $(Y_t)_{t \geq0}$ be independent Lévy processes with Lévy triplet $(\ell_1,q_1,\nu_1)$ and $(\ell_2,q_2,\nu_2)$, respectively. By the independence of the processes, we know that $(X_t,Y_t)_{t \geq 0}$ is a Lévy process and using Lévy-Khinchine's formula we find that its Lévy measure equals

$$\nu(dx,dy) := \nu_1(dx) \otimes \delta_{\{0\}}(dy)+ \delta_{\{0\}}(dx)\otimes \nu_2(dy)$$

where $\delta_{\{0\}}$ denotes the Dirac measure centered at $0$ (for a proof see e.g. this question). Roughly speaking, the jumps of the process $(X,Y)$ are concentrated on the coordinate axis. Indeed: Let

$$A(\varepsilon) := \mathbb{R}^2 \backslash \bigg(B[0,\varepsilon] \cup \{(x,y) \in \mathbb{R}^2; x = 0 \vee y=0\} \bigg)$$

for $\varepsilon>0$. Obviously, $A(\varepsilon)$ is open and satisfies $\nu(A(\varepsilon)) = 0$. Consequently, there exists $\Omega(\varepsilon)$, $\mathbb{P}(\Omega(\varepsilon))=1$, such that

$$\forall t \geq 0: \Delta (X_t,Y_t)(\omega) = (\Delta X_t(\omega),\Delta Y_t(\omega)) \notin A(\varepsilon)$$

for all $\omega \in \Omega(\varepsilon)$. Set $\Omega' := \bigcap_{k \in \mathbb{N}} \Omega(1/k)$. Then $\mathbb{P}(\Omega')=1$ and

$$\forall t \geq 0: (\Delta X_t(\omega),\Delta Y_t(\omega)) \notin \mathbb{R}^2 \backslash \{(x,y); x = 0 \vee y=0\}$$

for any $\omega \in \Omega'$. This shows that $X$ and $Y$ do not jump simultaneously a.s.


Let $X$, $Y$ be independent real-valued Lévy processes and $\varepsilon>0$. Without loss of generality set $t=1$. Since the processes have cadlag paths, we find

$$\begin{align*} & \quad \mathbb{P}(\exists s \in [0,1]: \Delta X_s > \varepsilon, \Delta Y_s>\varepsilon) \\ &\leq \mathbb{P} \bigg( \bigcup_{i=0}^{\infty} \bigcap_{j \geq i} \underbrace{\left\{ \exists k \in \{0,\ldots,j\}: |X_{(k+1)/j}-X_{k/j}| > \frac{\varepsilon}{2}, |Y_{(k+1)/j}-Y_{k/j}| > \frac{\varepsilon}{2} \right\}}_{=: A_j} \bigg) \\ &= \mathbb{P} \left( \liminf_{j \to \infty} A_j \right) \leq \liminf_{j \to \infty} \mathbb{P}(A_j) \end{align*}$$

Using the stationarity of the increments as well as the independence of the processes, we obtain

$$\begin{align*} \mathbb{P}(A_j) &\leq \sum_{k=0}^j \mathbb{P}\left(|X_{(k+1)/j}-X_{k/j}| > \frac{\varepsilon}{2} \right) \cdot \mathbb{P}\left(|Y_{(k+1)/j}-Y_{k/j}|> \frac{\varepsilon}{2} \right) \\ &= j \cdot \mathbb{P}\left(|X_{1/j}|> \frac{\varepsilon}{2} \right) \cdot \mathbb{P}\left(|Y_{1/j}|> \frac{\varepsilon}{2} \right) \end{align*}$$

Applying this lemma yields that

$$j \cdot \mathbb{P} \left(|X_j| > \frac{\varepsilon}{2} \right) \leq C, \qquad j \in \mathbb{N}$$

for some constant $C>0$. On the other hand, by the stochastic continuity, $\mathbb{P}(|Y_{1/j}|>\varepsilon/2) \to 0$ as $j \to \infty$. Consequently,

$$ \mathbb{P}(\exists s \in [0,1]: \Delta X_s > \varepsilon, \Delta Y_s>\varepsilon) \leq \liminf_{j \to \infty} \mathbb{P}(A_j) = 0$$

Since $\varepsilon>0$ is arbitrary, this finishes the proof.

Note that the jumps of a Lévy process $X$ cannot occur at fixed times. Indeed,

$$\begin{align*}\DeclareMathOperator \re {Re} \DeclareMathOperator \im {Im} \mathbb{P}(|\Delta X_t|>\varepsilon) &= \mathbb{P} \left( \bigcup_{j \geq 1} \bigcap_{k \geq j} |X_t-X_{t-1/k}| \geq \varepsilon \right) \\ &\leq \liminf_{k \to \infty} \mathbb{P}(|X_t-X_{t-1/k}| \geq \varepsilon) = 0 \end{align*}$$

by the stochastic continuity of $X$. Thus, $\Delta X_t = 0$ a.s. Obviously, this implies

$$\Delta X_t \cdot \Delta Y_t = 0$$

for any two Lévy processes $X$, $Y$.

Note that $\Delta X_t \cdot \Delta Y_t = 0$ a.s., $t \geq 0$, does in general not imply

$$\sum_{s \leq t} \Delta X_s \cdot \Delta Y_s = 0,$$

see Did's comment below.

  • $\begingroup$ Careful, you just proved the result when X=Y (a case where it does not hold). $\endgroup$ – Did Oct 7 '13 at 17:33
  • $\begingroup$ @Did I don't see your point. Obviously, $\Delta X_t \cdot \Delta X_t = 0$ a.s for fixed $t$. I did not claim that $\Delta X_t \cdot \Delta Y_t = 0$ implies $\sum_{s \leq t} \Delta X_s \cdot \Delta Y_s$ for any two Lévy processes $X$ and $Y$. $\endgroup$ – saz Oct 7 '13 at 17:45
  • $\begingroup$ Sorry if I caused any confusion in the formulation of my question. I probably should have put it $\bigtriangleup X_{t}(\omega)=0$ or $\bigtriangleup Y_{t}(\omega)=0$ for all t a.s. This is like the difference between the definition of two processes being a version of one another and being indistinguishable. I suspected it would lead to confusion but I saw it in Jean Bertoin paper on subordinators and I thought it might help. $\endgroup$ – Ofer Oct 11 '13 at 8:05
  • $\begingroup$ My point is that you state (correctly) that "Note that (1) does in general not imply (2)" but that you only show (1) while the question is about (2). Your add-on makes clear the difference between (1) and (2) (thanks for that)--but now I wonder why you do not answer (2)... $\endgroup$ – Did Oct 13 '13 at 11:58
  • $\begingroup$ @Did It sounded to me as ofer was aware of a proof of the equivalence $(1) \Leftrightarrow (2)$. Later it became clear that this is not the case. Unfortunately, I recognized belatedly that my own proof doesn't work out either. So if you (or someone else) answer the question, I'll remove my answer. $\endgroup$ – saz Oct 13 '13 at 13:41

by Fubini's theorem we know that $\sum_{0\leq s\leq \infty}|\triangle X_{s}\triangle Y_{s}|$ is measurable. We can write it this way $\sum_{i=1}^{\infty}|\triangle X_{s}\triangle Y_{s}|1_{\{i-1\leq s\leq i\}}$. Take expectation, and by independence,the MCT and the continuity in probability of the Levy processes we have

$E\Bigl(\sum_{i=1}^{\infty}|\triangle X_{s}\triangle Y_{s}|1_{\{i-1\leq s\leq i\}}\Bigr)=\sum_{i=1}^{\infty}E\Bigl(|\triangle X_{s}\triangle Y_{s}|1_{\{i-1\leq s\leq i\}}\Bigr)=\sum_{i=1}^{\infty}E\Bigl(|\triangle X_{s}|\Bigr)E\Bigl(|\triangle Y_{s}|\Bigr)1_{\{i-1\leq s\leq i\}}=0$

so $\sum_{0\leq s\leq \infty}|\triangle X_{s}\triangle Y_{s}|=0$ a.s. Since the trajectories of the summand are increasing it implies for all $t>0$.

  • $\begingroup$ Probably, I've missed your point, but: Why should $\sum_{0 \leq s \leq t} |\Delta X_s \cdot \Delta Y_s|$ equal $\sum_{i=1}^{\infty} |\Delta X_s \cdot \Delta Y_s| 1_{(i-1) \wedge t \leq s \leq i \wedge t}$? In my opinion, the latter equals $|\Delta X_s \cdot \Delta Y_a| \cdot 1_{[0,t]}(s)$. $\endgroup$ – saz Oct 28 '13 at 16:39
  • $\begingroup$ You are right. I fixed it. Thanks. $\endgroup$ – Ofer Nov 4 '13 at 8:00
  • $\begingroup$ That's not what I meant. Fix $s \in [i-1,i]$ (note that you do not sum over $s$, so it is fixed!), then $$\sum_{i=1}^{\infty} |\Delta X_s \cdot \Delta Y_s| \cdot 1_{i-1 \leq s \leq i} = |\Delta X_s \cdot \Delta Y_s|$$, i.e. $\sum_{i=1}^{\infty} |\Delta X_s \cdot \Delta Y_s|=0$ holds trivially since $\Delta X_s = \Delta Y_s = 0$ a.s. In particular, your calculation does not imply $$\sum_{0 \leq s \leq \infty} |\Delta X_s \cdot \Delta Y_s|=0$$ $\endgroup$ – saz Nov 4 '13 at 15:55
  • $\begingroup$ Moreover, your "proof" (if it was correct) would also work for $X=Y$: Obviously, we have $$\mathbb{E}(|\Delta X_s \cdot \Delta Y_s|) = \mathbb{E}(|\Delta X_s|^2) = 0$$ and therefore $$\sum_{i=1}^{\infty} |\Delta X_s \cdot \Delta Y_s| 1_{i-1 \leq s \leq i} = 0$$ And as @did already mentioned, the statement does not hold for $X=Y$. $\endgroup$ – saz Nov 4 '13 at 15:58
  • $\begingroup$ you are right. it's wrong. $\endgroup$ – Ofer Nov 4 '13 at 17:20

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.