Show uniform convergence implies uniform convergence of squares/root

98 Views Asked by At

Let $f_n ,\,f\colon [0,1]\rightarrow [0,\infty)$ be given.

I wanted to show that if $f_{n}{\underset {n\ }{\rightrightarrows }}f,$ then $f^{1/2}_{n}{\underset {n\ }{\rightrightarrows }}f^{1/2}$

Using $|\sqrt{a}-\sqrt{b}|\le \sqrt{|a-b|}$ we see that $$\lim_{n\to \infty}\sup_{x\in[0,1]}\left | \sqrt{f_n(x)}-\sqrt{f(x)}\right | \le \lim_{n\to \infty}\sup_{x\in[0,1]}\sqrt{\left | f_n(x)-f(x)\right |}\le \lim_{n\to \infty}\sup_{x\in[0,1]}\left | f_n(x)-f(x)\right |=0$$

and the claim follows.

Is this correct?

I however can not find an example where

$f_{n}{\underset {n\ }{\rightrightarrows }}f,$ then $f^{2}_{n}{\underset {n\ }{\rightrightarrows }}f^{2}$

This cannot be true I think, can it?

1

There are 1 best solutions below

3
On BEST ANSWER

In the first part, the claim $f_{n}{\underset {n\ }{\rightrightarrows }}f \implies f_{n}^{1/2}{\underset {n\ }{\rightrightarrows }}f^{1/2} $ for $f_n,f :[0,1] \to [0,\infty)$ is true. Your proof starts out correctly, but your last inequality is not valid. You can complete the argument by stating that for all $\epsilon > 0$ there exists $N$ independent of $x$ such that when $n > N$ we have for all $x \in [0,1]$,

$$|f_n(x) - f(x)| < \epsilon^2, $$

and, thus,

$$\left|\sqrt{f_n(x)} - \sqrt{f(x)}\right| \leqslant \sqrt{|f_n(x) - f(x)|} < \epsilon.$$

Second Part

Clearly, there are examples where both $f_{n}{\underset {n\ }{\rightrightarrows }}f$ and $f^{2}_{n}{\underset {n\ }{\rightrightarrows }}f^{2}$ hold, either by virtue of your first proposition or the trivial example $f_n(x) = 1/n, \, f(x) = 0$.

However, the implication$f_{n}{\underset {n\ }{\rightrightarrows }}f\implies f^{2}_{n}{\underset {n\ }{\rightrightarrows }}f^{2}$ is not true.

For a counterexample, take $f_n,f:(0,1] \to \mathbb{R}$ where

$$f_n(x) = \log \frac{nx}{n+1}, \quad f(x) = \log x$$

We have $f_{n}{\underset {n\ }{\rightrightarrows }}f$, since

$$|f_n(x) - f(x)| = \log \frac{n+1}{n} \to 0$$

However, $$|f_n^2(x) - f^2(x)| = \left|\log \frac{nx}{n+1 } - \log x\right|\left|\log \frac{nx}{n+1 } + \log x\right| = \log\frac{n+1}{n} \left|\log \frac{nx^2}{n+1 }\right|,$$

and by choosing $x_n = 1/(n e^n)$ we have

$$|f_n^2(x_n) - f^2(x_n)| = \log\frac{n+1}{n} \log (n^2 + n) + 2n \log\frac{n+1}{n} $$

Since the RHS does not converge to $0$ as $n \to \infty$, it follows that $f_{n}^2{\underset {n\ }{\rightrightarrows }}f^2$ does not hold on $(0,1].$

This example is easily modified for the domain $[0,1]$ and range $[0,\infty)$ as $f_n(x) = -\log \frac{nx}{n+1} \mathbf{1}_{(0,1]}.$