Suppose $\mu,\mu_n$ are Borel probability measures on $\mathbb{R}$ with $\mu_n$ converging weakly to $\mu$. I am asked to find some probability space $(\Omega,\mathcal{F},\mathbb{P})$ and random variables $X,X_n$ such that $X$ has law $\mu$, $X_n$ has law $\mu_n$ and $X_n \to X$ almost surely as $n \to \infty$.

So far I tried to let $(\Omega,\mathcal{F},\mathbb{P}) = ((0,1),\mathcal{B}_{0,1},\text{Lebesgue})$ and defined $X_n$ as for $\omega \in (0,1)$ we let $X_n(\omega) = \inf\{x \in \mathbb{R}: \omega \in \mu_n((-\infty,x])\}$. Then $\mathbb{P}(X_n \in (-\infty,x]) = \mathbb{P}(X_n^{-1}((-\infty,x])) = \mu_n((-\infty,x])$ so $X_n$ has law $\mu_n$ and similarly for $X$.

However I am not able to prove that $X_n \to X$ almost surely. I have tried using contradiction, if $X_n \not\to X$ almost surely then we have that $|X_n-X| \geq 0$ and this does not converge to 0 almost surely. So the integral of this does not converge to $0$ as $n \to \infty$. However at this point I get stuck. I can't see where I should bring in the fact that $\mu_n \to \mu$ weakly.

How should I proceed?

  • $\begingroup$ This is a special case of Skhorohod Representation Theorem. You can search Wikipedia for this theorem. $\endgroup$ – Kavi Rama Murthy Sep 22 '18 at 12:42

Let $\mu_n$, $\mu$ be probability measures on $\mathbb{R}$ such that $\mu_n$ converges weakly to $\mu$. The associated cumulative distribution functions $$F_n(x) := \mu_n((-\infty,x]) \qquad F(x) := \mu((-\infty,x])$$ then satisfy $$F_n(x) \to F(x) \tag{1}$$ for any continuity point $x$ of $F$.

Lemma Denote by $$\begin{align*} F_n^{-1}(t) &:= \inf\{x \in \mathbb{R}; F_n(x) \geq t\} \\ F^{-1}(t) &:= \inf\{x \in \mathbb{R}; F(x) \geq t\} \end{align*}$$ the generalized inverse functions of $F_n$ and $F$, respectively. If $t \in (0,1)$ is a continuity point of $F^{-1}$, then $$\lim_{n \to \infty} F_n^{-1}(t) = F^{-1}(t). \tag{2}$$

Once we have proved the result, we find immediately that the sequence of random variables defined in your question $$X_n(t) = F_n^{-1}(t) \qquad X(t) = F^{-1}(t)$$ satisifies $X_n(t) \to X(t)$ for any continuity point $t$ of $F^{-1}$; as $F^{-1}$ has at most countably many discontinuies, this shows $X_n \to X$ almost surely.

Proof of the lemma: Let $t \in (0,1)$ be a continuity point of $F^{-1}$. Since $F$ has at most countably many discontinuities, we can find for any $\epsilon>0$ a continuity point $x$ of $F$ such that $$F^{-1}(t)- \epsilon< x < F^{-1}(t).$$ As $$x < F^{-1}(t) \implies F(x) < t$$ and $\lim_n F_n(x) = F(x)$, we have $F_n(x)<t$ for large $n \in \mathbb{N}$, and the definition of the inverse yields $x \leq F_n^{-1}(t)$. Consequently, $$F^{-1}(t) - \epsilon < x \leq F_n^{-1}(t)$$ implying $$F^{-1}(t) \leq \liminf_{n \to \infty} F_n^{-1}(t).$$ It remains to show that $$\limsup_{n \to \infty} F_n^{-1}(t) \leq F^{-1}(t). \tag{3}$$ To this end, we note that for any $u>t$ and $\epsilon>0$ we can find a continuity point $y$ of $F$ such that $$F^{-1}(u) < y < F^{-1}(u)+\epsilon.$$ This implies $$F(y) \geq u > t.$$ As $y$ is a continuity point of $F$, this means that $F_n(y) \geq t$ for large $n \in \mathbb{N}$. Hence, $y \geq F_n^{-1}(t)$, and so $$F^{-1}(u)+\epsilon > y \geq F^{-1}_n(t)$$ for large $n$. Thus, $$\limsup_{n \to \infty} F_n^{-1}(t) \leq F^{-1}(u).$$ Letting $u \downarrow t$ we find from the (right)continuity of $F^{-1}$ at $t$ that $(3)$ holds.

Reference: Theorem 8.3.2 in S.I. Resnick: A probability path, Birkhäuser 2013.


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.