4
$\begingroup$

Consider the sequence of functions $\{f_n \}$ on $\Bbb R$ defined by $$f_n(x) = n \log \left (1 + \frac {x^2} {n} \right ),\ \ x \in \Bbb R$$ for all $n \geq 1.$ Show that the sequence of functions $\{f_n \}$ is not uniformly convergent on $\Bbb R.$

My attempt $:$ First I observe that the sequence of functions $\{f_n \}$ converges pointwise to the everywhere continuous function $f$ defined by $$f(x) = x^2,\ \ x \in \Bbb R.$$

If the sequence of functions $\{f_n \}$ converges uniformly to the continuous limit function $f$ on whole of $\Bbb R$ then in particular the sequence of functions $\{f_n \}$ converges uniformly to the continuous limit function $f$ on $[0,1].$

But then we have $$\lim\limits_{n \rightarrow \infty} \int_{0}^{1} f_n(x)\ dx = \int_{0}^{1} f(x)\ dx.\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ (1)$$

Now what I found is that $$\int_{0}^{1} f_n(x)\ dx = n \log \left (1 + \frac 1 n \right ) - 2n + 2n^{\frac 3 2} \arctan \left (\frac {1} {\sqrt n} \right ).$$ So we have $$\lim\limits_{n \rightarrow \infty} \int_{0}^{1} f_n(x)\ dx = - \infty.$$ where as $$\int_{0}^{1} f(x)\ dx = \frac 1 3.$$ which is a contradiction to $(1).$ Hence our assumption is false. So the sequence of functions $\{f_n \}$ is not uniformly convergent on $\Bbb R,$ as required.

EDIT $:$ I found the wrong limit. By L'Hospital the limit $$\lim\limits_{n \rightarrow \infty} n \left ( \log \left (1 + \frac 1 n \right ) - 2 \right ) = 1.$$

Now what will be $$\lim\limits_{n \rightarrow \infty} 2n^{\frac 3 2} \arctan \left (\frac {1} {\sqrt n} \right )\ ?$$

By L'Hospital I got infinity. Isn't it the case? But then the overall limit becomes $+\infty.$

$\endgroup$
  • $\begingroup$ Oh! God. Online calculator shows that the limit is $\frac 1 3.$ $\endgroup$ – math maniac. Jan 16 at 15:12
  • 2
    $\begingroup$ Clearly, your calculations must be wrong because the integrand is positive everywhere, so the integral couldn't be negative. $\endgroup$ – Fimpellizieri Jan 16 at 15:21
  • $\begingroup$ @Fimpellizieri I think the limit is $+\infty.$ Am I still doing any mistake? $\endgroup$ – math maniac. Jan 16 at 15:28
  • $\begingroup$ As it turns out, $f_n$ does converge uniformly to its limit on $[0,1]$. So, your line of reasoning is bound to fail. $\endgroup$ – Ben Grossmann Jan 16 at 15:34
3
$\begingroup$

Here's an approach: note that $$ f_{n}(x) - f(x) = n \log \left (1 + \frac {x^2} {n} \right ) - \log(e^{x^2}) = \log \left( \frac{(1 + x^2/n)^n}{e^{x^2}} \right). $$ Now, we note that for any fixed $n$, we have $$ \begin{align} \lim_{x \to \infty} \frac{(1 + x^2/n)^n}{e^{x^2}} &= \left(\lim_{x \to \infty} \frac{1 + x^2/n}{e^{x^2/n}}\right)^n = 0 \end{align} $$ (since the numerator is a polynomial) from which we may conclude that (again, for any fixed $n$) the set $\{|f_n(x) - f(x)|: x \in \Bbb R\}$ is unbounded. Thus, if we fix (for instance) $\epsilon = 1$, we can see that for any $n$ there exists an $x$ such that $|f_n(x) - f(x)| \geq \epsilon$, which is to say that $\|f_n - f\|_{\sup} \geq \epsilon$.

So, the sequence is not uniformly convergent.


Regarding your limit: with the substitution $m = \sqrt{1/n}$, we have $$ \begin{align} \lim_{m \to 0^+} 2m^{-3} \arctan \left (m \right ) &= 2\lim_{m \to 0^+}\frac{\arctan(m)}{m^3} = 2\lim_{m \to 0^+}\frac{\frac{1}{m^2 + 1}}{3m^2} = \frac 23 \lim_{m \to 0^+} \frac{1}{m^2(m^2 + 1)} = +\infty \end{align} $$

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Very nice! +1 ${}$ $\endgroup$ – Fimpellizieri Jan 16 at 15:38
  • $\begingroup$ @F Thanks! And good catch. $\endgroup$ – Ben Grossmann Jan 16 at 15:38
  • $\begingroup$ Can you help me in the computation of the above limit? If the given sequence of functions was uniformly convergent on $[0,1]$ then the limit should be equal to $\frac 1 3.$ $\endgroup$ – math maniac. Jan 16 at 15:45
  • $\begingroup$ See my latest edit. We could do the second limit, for instance, with two iterations of L'Hospital $\endgroup$ – Ben Grossmann Jan 16 at 15:50
  • $\begingroup$ I misunderstood, I thought you meant the limit from my question. I'm not particularly interested in ironing that limit of integrals out. Regardless of uniform convergence, the limit has to be $\frac 13$ by the monotone convergence theorem. $\endgroup$ – Ben Grossmann Jan 16 at 15:51
2
$\begingroup$

@mathmaniac.: (too big for a comment) using Taylor series, you rewrite $$n\log\Big(1+\frac{1}{n}\Big)-2n+2n\sqrt n\arctan\Big(\frac{1}{\sqrt{n}}\Big)$$ as $$n\Big(\frac{1}{n}+o\Big(\frac{1}{n}\Big)\Big)-2n+2n\sqrt{n}\Big(\frac{1}{\sqrt n}-\frac{1}{3(\sqrt n)^3}+o\Big(\frac{1}{(\sqrt n)^3}\Big)\Big)=1-\frac{2}{3}+o(1)\underset{n\to\infty}{\longrightarrow}\frac{1}{3},$$

so perhaps your mistake was to only use degree $1$ for the development of $\arctan$.

| cite | improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ Great answer. Thanks @Ballon. +1 $\endgroup$ – math maniac. Jan 16 at 16:45
  • $\begingroup$ @mathmaniac.: You're welcome! $\endgroup$ – Balloon Jan 16 at 16:46

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.