Suppose I have a random sequence $X_n$ of cadlag functions on $[0,1]$ that converge weakly to $X$. In general this is meant with respect to the Skorkhod metric but suppose here I have that $X$ is continuous and $X_n$ converge weakly to $X$ with respect to the uniform metric on the space of cadlag processes $[0,1]$. (Under this metric this is a complete metric space however it is not separable.) I need that $X_n$ is well defined with respect to the uniform topology but suppose I have this as well.

Now suppose $X_n$ are piecewise constant and consider $\tilde{X}_n$, the linearized version of $X_n$. By this I mean the function I get by drawing straight lines between the jump discontinuities of $X_n$. Is it true that $\tilde{X}_n$ converges in law to $X$ again with respect to the uniform metric on continuous functions?

I'm pretty sure that this is true, and I believe I have a proof but I'd like confirmation. I thought this would be written down somewhere. I tried Billingsley, Convergence of Probability Measures, especially Chapter 3, Section 18 (The Uniform Topology), but I couldn't find it. Thanks for your help.

  • $\begingroup$ The question is not really clear, so let's translate it to distributions. You consider functions $f:[0,1] \to \Bbb R$ that are cadlag, right? Let's say $\Omega$ is a space of all such functions with the uniform metric convergence. You have probability measures $\mu_n$ that converge weakly (w.r.t. topology of $\Omega$) to $\mu$. These $\mu_n$ are distributions of $X_n$. Now you suppose that $X_n$ is piecewise constand and $\tilde X_n$ is a linearized version of $X_n$. What does that mean. Also, your question now sounds like is it true that if I consider $\tilde X_n$, then $X_n \to X$ $\endgroup$ – Ilya Jul 23 '14 at 19:55
  • $\begingroup$ The linearised version of a piecewise function is the corresponding continuous function constructed by drawing straight lines between all the jump discontinuities. I fixed the tilde so that the question makes sense. Thank you. $\endgroup$ – Ben Jul 23 '14 at 23:07
  • 2
    $\begingroup$ So $\tilde X_n$ is not defined uniquely, you have same parameters of the linearization - correct? Depending on how you define it, perhaps the continuous mapping theorem could work. As far as I understand, $\|\tilde f - \tilde g\| \leq \|f - g\|$ where $f,g$ are piece-wise constant functions, $\tilde f,\tilde g$ are their linearizations done in a good way and $\|\cdot\|$ is the uniform sup norm. $\endgroup$ – Ilya Jul 24 '14 at 6:51
  • $\begingroup$ Nice. That's a lot cleaner than what I had. Thanks. $\endgroup$ – Ben Jul 24 '14 at 15:31

If I understand the question correctly, the linearization map $l$ have some parameters and can be chosen to satisfy some nice properties. For example, we can define it in such a way that $$ \|l(f) - l(g)\|\leq \|f-g\| $$ where $\|f\|:=\sup_{x\in [0,1]}|f(x)|$ denotes the uniform sup-norm. Hence, $l$ is a Lipschitz map and thus continuous. The result follows from the continuous mapping theorem.


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.