2
$\begingroup$

Denote by $D(0,T)$ the space of right continuous functions with left limits defined on $[0,T]$. Let $\mathbb P$ be a probability measure on $D(0,T)$. Define

$$cont(\mathbb P):=\Big\{t\in [0,T]:~ \mathbb P\big(\big\{f\in D(0,T):~ f(t-)=f(t)\big\}\big) = 1\Big\}.$$

Is $cont(\mathbb P)$ dense in $[0,T]$?

I strongly believe the answer is yes, but cannot find a reference.

$\endgroup$
2
  • $\begingroup$ Doesn’t this follow from Fubini’s theorem? A càdlàg function has at most countable many discontinuities, right? $\endgroup$ Commented Sep 12, 2020 at 15:56
  • $\begingroup$ Yes. If we denote by $X_t$ the coordinate projection, i.e. X_t(f):=f(t), then we can deduce $\{t\in [0,T]: \mathbb P\circ X_{t-}^{-1} = \mathbb P\circ X_{t}^{-1}\}$ is dense. But the above result is stronger. Could you please detail a bit more? $\endgroup$
    – user128095
    Commented Sep 12, 2020 at 16:00

1 Answer 1

1
$\begingroup$

Let $f\in D(0,T)$. Let $\mathsf{Disc}_n(f)=\{t\colon |f(t^+)-f(t^-)|\ge \frac 1n\}$. I claim this set is discrete. If not, there is a sequence of distinct points $(t_k)\in\mathsf{Disc}_n(f)$ converging to some $t_0$. There is then a subsequence converging to $t_0$ consisting entirely of points on the left or entirely of points on the right. This contradicts the left limit or right continuity property at $t_0$. Hence $\mathsf{Disc}_n(f)$ is discrete, and therefore finite. Hence $\mathsf{Disc}(f)=\bigcup_n\mathsf{Disc}_n(f)$ is countable and so of measure 0.

Now consider $$ S=\{(f,t)\colon f\in D(0,T),\, t\in\mathsf{Disc}(f)\}. $$ By the above calculation, for each $f\in D(0,T)$, $\text{Leb}(\mathsf{Disc}(f))=0$. Hence $\mathbb P\times\text{Leb}(S)=0$. So by Fubini's theorem, $$ \mathbb P\times\text{Leb}(S)=\int_0^T \mathbb P(\{f\in D(0,T)\colon t\in\mathsf{Disc}(f)\})\,dt=0. $$ Hence for $\text{Leb}$-a.e. $t$, $$ \mathbb P(\{f\in D(0,T)\colon t\in\mathsf{Disc}(f)\})=0.$$

$\endgroup$
2
  • $\begingroup$ For this to be complete, one needs to show that the set $S$ is measurable with respect to the product $\sigma$-field. I guess that follows from the statements that $(f,t) \mapsto f(t)$ and $(f,t) \mapsto f(t-)$ are jointly measurable. $\endgroup$ Commented Sep 12, 2020 at 17:45
  • $\begingroup$ Thanks for the nice proof, as well as the complement of @NateEldredge $\endgroup$
    – user128095
    Commented Sep 12, 2020 at 18:59

You must log in to answer this question.