1
$\begingroup$

This question is a related question see this post Vague convergence VS Laplace transform convergence. But now we assume that \begin{equation} \int_0^\infty e^{-sx}\mu_n(dx)\to \int_0^\infty e^{-sx}\mu(dx), \end{equation}

for all $s\in (0, \infty)$, where $\mu_n$ is a finite measure. Please check this book for vague convergence. Can we get what from the above Laplace transform convergence? From the above post, my understanding is that vague convergence is weaker than weakly convergence (please correct me, if I am wrong). Is it possible to get vague convergence and why or why not?

$\endgroup$
1
  • $\begingroup$ This (with $s > 0$ and with $s \geq 0$) are simple consequences of above mentioned Kallenberg (1997), Lemma 4.4 (approximation). Of course vague convergence is weaker than weak convergence, since for the first fewer test functions are used. $\endgroup$ Commented May 2, 2023 at 12:04

1 Answer 1

1
$\begingroup$

$\newcommand{\thh}{\theta}$In general, you cannot get the vague convergence here. E.g., suppose that $\mu=0$ and $\mu_n(dx)=x^2\,1(n<x<n+1)\,dx$. Then for each real $s>0$ \begin{equation*} 0\le\int_0^\infty e^{-sx}\mu_n(dx)=\int_n^{n+1} e^{-sx}x^2\,dx \le e^{-sn}(n+1)^2\to0 \end{equation*} (as $n\to\infty$), so that \begin{equation*} \int_0^\infty e^{-sx}\mu_n(dx)\to\int_0^\infty e^{-sx}\mu(dx). \end{equation*} However, letting $f(x):=\frac1{1+x}$ for real $x\ge0$, we have $f\in C_0([0,\infty))$ and \begin{equation*} \int_0^\infty f(x)\mu_n(dx)=\int_n^{n+1} f(x)x^2\,dx \ge \frac1{1+n+1}n^2\to\infty, \end{equation*} so that \begin{equation*} \int_0^\infty f(x)\mu_n(dx)\to\infty\ne0=\int_0^\infty f(x)\mu(dx). \end{equation*}


However, assuming that the measure $\mu$ is finite, you can get the following: \begin{equation*} \int_0^\infty f(x)\mu_n(dx)\to\int_0^\infty f(x)\mu(dx) \tag{$*$}\label{1} \end{equation*} for all $f\in C_c([0,\infty))$, that is, for all continuous functions $f$ on $[0,\infty)$ with a compact support.

Indeed, take any such $f$, so that $f=0$ on $[a,\infty)$ for some real $a>0$. Consider first the case when $f\ge0$.

Take any real $h>0$ and let \begin{equation*} \nu_{n,h}(dx):=e^{-hx}\mu_n(dx),\quad \nu_h(dx):=e^{-hx}\mu(dx). \end{equation*} Then for any real $s\ge0$ \begin{equation*} \int_0^\infty e^{-sx}\nu_{n,h}(dx)=\int_0^\infty e^{-(s+h)x}\mu_n(dx) \to\int_0^\infty e^{-(s+h)x}\mu(dx)=\int_0^\infty e^{-sx}\nu_h(dx). \end{equation*} So, by the previous answer, $\nu_{n,h}\to\nu_h$ weakly. So, \begin{equation*} \int_0^\infty f(x)e^{-hx}\mu_n(dx)=\int_0^\infty f(x)\nu_{n,h}(dx) \to\int_0^\infty f(x)\nu_h(dx)=\int_0^\infty f(x)e^{-hx}\mu(dx) \end{equation*} for each real $h>0$. Next, \begin{equation*} \int_0^\infty f(x)\mu_n(dx)\ge\int_0^\infty f(x)e^{-hx}\mu_n(dx) =\int_0^a f(x)e^{-hx}\mu_n(dx) \\ \ge e^{-ha}\int_0^a f(x)\mu_n(dx) =e^{-ha}\int_0^\infty f(x)\mu_n(dx), \end{equation*} so that \begin{equation*} \int_0^\infty f(x)\mu_n(dx)=e^{\thh_{n,h,f}ha}\int_0^\infty f(x)e^{-hx}\mu_n(dx) \end{equation*} for some $\thh_{n,h,f}\in[0,1]$ depending on $n,h,f$. Similarly, \begin{equation*} \int_0^\infty f(x)\mu(dx)=e^{\thh_{h,f}ha}\int_0^\infty f(x)e^{-hx}\mu(dx) \end{equation*} for some $\thh_{h,f}\in[0,1]$ depending on $h,f$. So, for each real $h>0$, \begin{equation*} \begin{aligned} \limsup_n\int_0^\infty f(x)\mu_n(dx)&\le e^{ha}\limsup_n\int_0^\infty f(x)e^{-hx}\mu_n(dx) \\ & =e^{ha}\int_0^\infty f(x)e^{-hx}\mu(dx) \\ & \le e^{ha}\int_0^\infty f(x)\mu(dx) \\ & \underset{h\downarrow0}\longrightarrow \int_0^\infty f(x)\mu(dx) \end{aligned} \end{equation*} and \begin{equation*} \begin{aligned} \liminf_n\int_0^\infty f(x)\mu_n(dx)&\ge\liminf_n\int_0^\infty f(x)e^{-hx}\mu_n(dx) \\ &=\int_0^\infty f(x)e^{-hx}\mu(dx) \\ &\ge e^{-ha}\int_0^\infty f(x)\mu(dx) \\ & \underset{h\downarrow0}\longrightarrow \int_0^\infty f(x)\mu(dx). \end{aligned} \end{equation*} So, $\limsup_n\int_0^\infty f(x)\mu_n(dx)=\liminf_n\int_0^\infty f(x)\mu_n(dx)=\int_0^\infty f(x)\mu(dx)$ and hence \eqref{1} holds for any nonnegative $f\in C_c([0,\infty))$.

Finally, writing $f=f_+-f_-$, where $f_\pm:=\max(0,\pm f)$, we get \eqref{1} for all $f\in C_c([0,\infty))$. $\quad\Box$

$\endgroup$
7
  • $\begingroup$ $f$ doesn't have compact support. $\endgroup$ Commented May 2, 2023 at 13:03
  • $\begingroup$ @DieterKadelka : No, it does not, and it does not have to, according to the definition of the vague convergence, which involves $C_0$, the space of continuous functions small (but not necessarily $0$) outside some compact sets. $\endgroup$ Commented May 2, 2023 at 13:24
  • $\begingroup$ Kallenberg explicitely demands test functions with compact support. $\endgroup$ Commented May 2, 2023 at 13:27
  • 1
    $\begingroup$ @DieterKadelka : Yes, he does, but I think most other people do not. Anyhow, now we have an answer for both versions of the vague convergence. $\endgroup$ Commented May 2, 2023 at 14:03
  • 1
    $\begingroup$ @DieterKadelka : Wikipedia gives $C_0$ as the principal version, and about the $C_c$ version it says "It is also not uncommon". But again, this is just a choice of terminology. $\endgroup$ Commented May 2, 2023 at 15:03

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .