19
$\begingroup$

Exercise 1: Let $\mu_n$, $\mu$ be probability measures on $\left(\mathbb{R}, \mathcal{B}\left(\mathbb{R}\right)\right)$ with distribution functions $F_n$, $F$. Show: If $\left(\mu_n\right)$ converges weakly to $\mu$ and $F$ is continuous, then $\left(F_n\right)$ converges uniformly on $\mathbb{R}$ to $F$.

This is a problem that I am totally stuck at. I know the fact that $F_n$ converges pointwise to $F$ in this question. Also, I looked through Google and found out that I have to show first that $F_{n}(t_{n})$ converges to $F(t)$ if $t_{n}$ converges to $t$. But, no matter how I tried, I keep failing to prove the fact. Also, I have no idea how to use the fact to get the uniform convergence. So I'm just stymied.

Could anyone please help me with this?

Add : I managed to show that $F_n$ converges uniformly to $F$ on any compact intervals. However, the generalization to the whole real line is still not solved...Could anyone at least help me with the generalization to the real line?

$\endgroup$
3
  • $\begingroup$ What's your definition of weak convergence? $F_n \to F$ pointwise for all contuinty points of $F$? $\endgroup$
    – saz
    May 17, 2015 at 7:44
  • $\begingroup$ Yes, that is correct. $\endgroup$
    – Keith
    May 17, 2015 at 7:46
  • 4
    $\begingroup$ Could you (for the sake of other interested readers) add the idea how to prove the uniform convergence on compact intervals to your question? $\endgroup$
    – saz
    May 17, 2015 at 9:27

2 Answers 2

20
$\begingroup$

Fix $\epsilon>0$. Since $F$ is a distribution function, there exists $R >0$ such that $F(r) \leq \epsilon$ for all $r \leq -R$ and $F(r) \geq 1-\epsilon$ for all $r \geq R$. As $F_n \to F$ pointwise, we can choose $N \in \mathbb{N}$ such that

$$|F_n(-R) - F(-R)| \leq \epsilon \qquad \text{and} \qquad |F_n(R)-F(R)| \leq \epsilon$$

for all $n \geq N$. Hence, by the monotonicity of $F_n$ and $F$,

$$\begin{align*} |F_n(r)-F(r)| \leq |F_n(r)|+|F(r)| &\leq F_n(-R)+F(-R) \\ &= (F_n(-R)-F(-R)) + 2 F(-R) \\ &\leq 3\epsilon \tag{1} \end{align*}$$

for all $r \leq -R$. Similarly, it follows from

$$1 \geq F_n(r) \geq F_n(R) = (F_n(R)-F(R))+F(R) \geq 1-2\epsilon, \qquad r \geq R,$$

that

$$|F_n(r)-F(r)| \leq |F_n(r)-(1-\epsilon))|+ |F(r)-(1-\epsilon)| \leq 2 \epsilon \tag{2}$$

for all $r \geq R$. Combining $(1)$ and $(2)$ yields

$$\sup_{r \in [-R,R]^c} |F_n(r)-F(r)| \leq 3 \epsilon$$

for all $n \geq N$. Since you have already shown that $F_n$ converges to $F$ uniformly on compact intervals, there exists $N' \in \mathbb{N}$ such that

$$\sup_{r \in [-R,R]} |F_n(r)-F(r)| \leq \epsilon$$

for all $n \geq N'$. Setting $\tilde{N} := \max\{N,N'\}$, we get

$$\sup_{r \in \mathbb{R}} |F_n(r)-F(r)| \leq 3 \epsilon \qquad \text{for all $n \geq N$}.$$

$\endgroup$
3
  • $\begingroup$ Can you also provide the proof for the convergence in compact intervals? $\endgroup$ Mar 31, 2016 at 0:34
  • $\begingroup$ @Susan See this question: math.stackexchange.com/q/467976/36150 $\endgroup$
    – saz
    Mar 31, 2016 at 16:27
  • $\begingroup$ By $r\in[-R,R]^{c}$ do you want to say that the convergence is out of that interval? And how do you pass from that to sup $r\in[-R,R]$ ?@saz Thank you and sorry if this question is so basic. $\endgroup$
    – sopach96
    Oct 31, 2021 at 16:41
5
$\begingroup$

Here's another proof for the compact case which I find a bit more intuitive. We are trying to show that $F_n \rightarrow F$ uniformly on an arbitrary closed interval $[a,b]$.

Fix $\epsilon > 0$. Let $d = F(b)-F(a)$, and take $k$ large enough so that $\frac{d}{k} \leq \frac{\epsilon}{5}$. By the continuity of $F$, we can apply the intermediate value theorem to show there exist real numbers $a := x_0 < x_1 < ... < x_k := b$ such that $F(x_i) = F(a) + i\frac{d}{k}$ for each $i \in \{0,1,...,k\}$.

Since $F_n \rightarrow F$ pointwise, for each $i$ in $\{0,1,...,k\}$ there exists $N_i$ such that $|F_n(x_i)-F(x_i)| \leq \frac{\epsilon}{5}$ for all $n \geq N_i$. Taking $N = \max(N_0,N_1,...,N_k)$, we conclude that $|F_n(x_i)-F(x_i)| \leq \frac{\epsilon}{5}$ for all $n \geq N$, $i \in \{0,1,...,k\}$. This gives $F_n(x_{i+1})-F_n(x_i) \leq |F_n(x_{i+1})-F(x_{i+1})| + |F(x_{i+1})-F(x_i)| + |F(x_i)-F_n(x_i)| \leq \frac{\epsilon}{5} + \frac{d}{k} + \frac{\epsilon}{5} \leq \frac{3\epsilon}{5}$

For any $x \in [a,b)$, we have $x_i \leq x < x_{i+1}$ for some $i \in \{0,1,...,k\}$. Then for all $n \geq N$, we have $$ \begin{align} |F_n(x)-F(x)| & \leq |F_n(x)-F_n(x_i)| + |F_n(x_i) - F(x_i)| + |F(x_i)-F(x)| \\\\ & \leq |F_n(x_{i+1})-F_n(x_i)| + |F_n(x_i)-F(x_i)| + |F(x_{i+1})-F(x)| \\\\ & \leq \frac{3\epsilon}{5} + \frac{\epsilon}{5} + \frac{d}{k} \\\\ & \leq \epsilon \\\\ \end{align} $$ where the second inequality uses monotonicity of $F$ and $F_n$.

Of course, $|F_n(b)-F(b)| \leq \frac{\epsilon}{5} \leq \epsilon$ since $b = x_k$.

Thus, we've shown $|F_n(x)-F(x)| \leq \frac{5\epsilon}{3}$ for all $x \in [a,b]$, concluding the proof.

$\endgroup$

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.