Consider the following theorem regarding convergence in distribution

THEOREM 1: Let $(X_n)_{n\geq1}$, $X$ be $\mathbb{R}^d$-valued random variables. Then $X_n$ converges to $X$ in distribution if and only if $\lim\limits_{n\to\infty}E\{f(X_n)\}=E\{f(X)\}$ for all continuous, bounded functions $f$ on $\mathbb{R}^d$.

Then consider the following theorem and its proof

THEOREM 2 Let $(X_n)_{n\geq1}$, $X$ be defined on a given fixed probability space $(\Omega\text{, }\mathcal{A}\text{, }\mathbb{P})$. If $X_n$ converges to $X$ in distribution, and if $X$ is a r.v. a.s. equal to a constant, then $X_n$ converges to $X$ in probability as well.

PROOF: Suppose that $X$ is a.s. equal to a constant $a$ (that is $\mathbb{P}(X=a)=1)$. The function $f(x)=\frac{|x-a|}{1+|x-a|}$ is bounded and continous. Therefore, $\lim\limits_{n\to\infty}E\Big\{\frac{|X_n-a|}{1+|X_n-a|}\Big\}=0$, hence $X_n$ converges to $a$ in probability by a result assuring that.

I have a doubt about the proof.
In particular, since $X_n$ converges to $X$ in distribution, given THEOREM 2, $\lim\limits_{n\to\infty}E\Big\{\frac{|X_n-a|}{1+|X_n-a|}\Big\}=E\Big\{\lim\limits_{n\to\infty}\Big\{\frac{|X_n-a|}{1+|X_n-a|}\Big\}\Big\}$. At this point, since in the proof it is directly stated that "$\lim\limits_{n\to\infty}E\Big\{\frac{|X_n-a|}{1+|X_n-a|}\Big\}=0$", I guess it holds that $\lim\limits_{n\to\infty}\Big\{\frac{|X_n-a|}{1+|X_n-a|}\Big\}=0$, but I cannot figure out why this is true.

So, the question is: how can I show that $$\lim\limits_{n\to\infty}\Bigg\{\frac{|X_n-a|}{1+|X_n-a|}\Bigg\}=0$$?


1 Answer 1


We have $X_n \stackrel{d}\to a$ (convergence in distribution). Thus $$\lim_n\mathbb{E}(f(X_n)) = \mathbb{E}(f(a))= f(a)$$ for every bounded continuous function $f$.

With your choice of $f$, the result immediately follows.

The following result/definition was used:

$$X_n \stackrel{d}\to X$$ $$\iff$$ $$\mathbb{P}_{X_n} \stackrel{w}\to \mathbb{P}_X$$ $$\iff$$ $$\int fd\mathbb{P}_{X_n }= \int f(X_n) d \mathbb{P}=\mathbb{E}(f(X_n))\to \mathbb{E}(f(X)) = \int f(X) d \mathbb{P} = \int f d\mathbb{P}_X$$ for all $f \in C_b$.

Your claim that $$\lim_n \frac{|X_n-a|}{1+|X_n-a|}=0$$ need not be true.

  • 1
    $\begingroup$ I made the edit. Let me know if it is clear! $\endgroup$ Jun 7, 2020 at 9:04
  • 1
    $\begingroup$ Can I ask what your definition of $X_n \to X$ in distribution is? Do you know that this means that $E(f(X_n)) \to E(f(X))$ for all continuous bounded functions? $\endgroup$ Jun 7, 2020 at 9:05
  • 1
    $\begingroup$ This is the formula of the image measure. Suppose we have a measurable map between measurable spaces $T:(\Omega, \mathcal{F}) \to (\Omega', \mathcal{F}')$, and $\mu$ is a measure on $\mathcal{F}$ (here $\mu = \mathbb{P}, X= T, \Omega' = \mathbb{R}$), then we have an image measure $\mu_T$ on $\mathcal{F}'$ defined by $\mu_T(A) := \mu(T^{-1}(A))$ and the following holds: $\int_\Omega f \circ T d \mu = \int_{\Omega'} f d \mu_T$ for a measurable function $f: \Omega' \to \mathbb{R}$, in the sense that one side exists if and only the other one exists and then they are equal. $\endgroup$ Jun 7, 2020 at 9:16
  • 1
    $\begingroup$ As always, glad to help! $\endgroup$ Jun 7, 2020 at 9:19
  • 1
    $\begingroup$ If I may be honest, some of your questions seem to be more real analysis questions than measure theory questions (for example the $a_n^2 \to 0 \implies a_n \to 0$ one), so maybe start with that. If you feel confident with that, then you might have a look at: Billingsley "Probability and measure" (develops probability and measure simultaneously), Axler's "Measure theory" (freely available on Axler's homepage, book that looks like it will be a quick read) and more advanced and my personal favorite Cohn's "Measure theory". The last two books do not really have a probability orientation though. $\endgroup$ Jun 7, 2020 at 9:28

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.