The exercise is to construct a sequence of continuous functions $f_n:\mathbb{R}\rightarrow \mathbb{R}, n\in \mathbb{N}$ , which is pointwise convergent to $f(x)=0 , x\in \mathbb{R}$ and not uniformly convergent on any interval $(a,b)$.

I posted this question once and I got the answer, but I had forgotten to mention that $f_n$ must be continuous. The example of sequence of discontinuous functions looks like this:

Let $\{r_n\}_{n\geq 1}$ be an enumeration of the rationals, and define $f_n(x)$ as follows: $$ f_n(x)=\begin{cases} 1,& x\in \{r_k\}_{k\geq n}\\ 0,& else \end{cases} $$

The person, who posted this, made the remark:

If you want to make the $\{f_n\}$ continuous, you can modify this example by using bump functions supported on $(r_n-2^{-n},r_n+2^{-n})$ in place of the spikes to $1$ at $r_n$.

I'm not realy sure what he meant. I was thinking of defining $g_k$ as a broken line or maybe more smooth function going through $(r_k-2^{-k},0), (r_k,1), (r_k+2^{-k},0)$ and zero on $\mathbb{R}\setminus [r_k-2^{-k},r_k+2^{-k}]$ and then defining $f_n(x)=\sup\{g_k(x):k\ge n\}$. But I don't know whether it converges to zero and whether $f_n$ are even continuous.

On the other hand I have found this topic A sequence of continuous functions on $[0,1]$ which converge pointwise a.e. but does not converge uniformly on any interval and there is only an example of a sequence converging to zero almost everywhere. That's why I begin to doubt that there exists a sequence converging to zero everywhere (that would mean there is a mistake in my book because the exercise says "construct such a sequence").

  • $\begingroup$ Are you familiar with convolution and mollifiers? $\endgroup$ – Silvia Ghinassi Feb 10 '16 at 16:41
  • $\begingroup$ @SilviaGhinassi: What is your idea, even if the OP is not familiar with these notions? (I am interested in the answer, too). $\endgroup$ – PhoemueX Feb 10 '16 at 18:35
  • $\begingroup$ @PhoemueX I didn't check if it really works, and I won't have access to a computer in the next few hours. I'll try to think more about it as soon as I can. Sorry about it. $\endgroup$ – Silvia Ghinassi Feb 10 '16 at 18:38
  • $\begingroup$ The remark you quote looks interesting, but I don't see how that could work. If we add the "spikes" , then the sum diverges. $\endgroup$ – leonbloy Feb 11 '16 at 18:43
  • 1
    $\begingroup$ @SilviaGhinassi with the current structure of rational numbers and irrational numbers the mollifier method can be hard to use, since the integral is zero. Perhaps we need a further discussion, not only to use the "rational-irrational" structure, but perhaps with some step functions. $\endgroup$ – Peter Feb 12 '16 at 17:37

According to the hint in "Problems in Mathematical Analysis", Biler and Witkowski, problem 4.112.

Define $g_n$ as the function which is $0$ on the intervals $(-\infty, 0]$ and $[2/n,\infty)$, $1$ at $1/n$, and linear otherwise. Then $g_n$ converges pointwise to $0$ but is not uniform in any interval containing $0$.

Now define $$f_n(x) = \sum_{k=0}^{\infty} 2^{-k} g_n(x - r_k).$$

Then the $f_n$ are continuous, converge pointwise to $0$, but do not converge uniformly on any open interval.

| cite | improve this answer | |
  • $\begingroup$ Could you help me with the proof? I managed to prove that $f_n$ are well defined and continuous and $(f_n(x))$ converges to zero for all $x$. But I have problems with showing that this sequence doesn't converge uniformly on any interval. $\endgroup$ – Kulisty Feb 13 '16 at 22:31
  • $\begingroup$ And just to make sure: Is $(r_n)$ an enumeration of the rationals? $\endgroup$ – Kulisty Feb 13 '16 at 23:10
  • 1
    $\begingroup$ Yes, $r_n$ is an enumeration of the rationals. For nonuniform convergence, does it make sense that the convergence of $g_n$ is not uniform on any interval containing $0$? (For any interval containing $0$ and any number $N$, there is an $n > N$ and $x$ in the interval such that $g_n(x)=1$.) Similarly, for the nonuniform convergence of $f_n$: given any interval, there is a rational $r_k$ inside the interval, so for any number $N$ there is an $n > N$ and $x$ in the interval such that $f_n(x) \geq 2^{-k}$. $\endgroup$ – Michael Harrison Feb 14 '16 at 4:08
  • 1
    $\begingroup$ Thank you very much, nice answer. I just came up with a different construction using Baire. I will award you the bounty if no other answers come up. $\endgroup$ – PhoemueX Feb 16 '16 at 9:38

There is indeed such a sequence, as one can show using a Baire category argument, using a suitable Banach space.

Let $C_{b}\left(\mathbb{R}\right)$ denote the space of bounded, continuous functions and let $$ V:=\left\{ \left(f_{n}\right)_{n\in\mathbb{N}}\in\left(C_{b}\left(\mathbb{R}\right)\right)^{\mathbb{N}}\,\middle|\,\left\Vert f\right\Vert _{V}:=\sup_{n\in\mathbb{N}}\left\Vert f_{n}\right\Vert _{\infty}<\infty\right\} . $$ It is not hard to see that this is a Banach space, equipped with $\left\Vert \cdot\right\Vert _{V}$. Furthermore, define $$ V_{0}:=\left\{ \left(f_{n}\right)_{n\in\mathbb{N}}\in V\,\middle|\,\forall x\in\mathbb{R}:\quad f_{n}\left(x\right)\xrightarrow[n\to\infty]{}0\right\} . $$ Then $V_{0}$ is complete, since it is a closed subspace of $V$. To see this, let $\left(g^{\left(m\right)}\right)_{m\in\mathbb{N}}=\left(\left(g_{n}^{\left(m\right)}\right)_{n\in\mathbb{N}}\right)_{m\in\mathbb{N}}$ be a sequence in $V_{0}$ with $g^{\left(m\right)}\xrightarrow[m\to\infty]{}g=\left(g_{n}\right)_{n\in\mathbb{N}}\in V$. We want to show $g\in V_{0}$. To this end, let $x\in\mathbb{R}$ and $\varepsilon>0$ be arbitrary. Then, there is some $m_{0}\in\mathbb{N}$ with $\left\Vert g^{\left(m\right)}-g\right\Vert _{V}<\frac{\varepsilon}{2}$ for all $m\geq m_{0}$. Because of $g^{\left(m_{0}\right)}\in V_{0}$, we have $g_{n}^{\left(m_{0}\right)}\left(x\right)\xrightarrow[n\to\infty]{}0$, so that there is $n_{0}\in\mathbb{N}$ with $\left|g_{n}^{\left(m_{0}\right)}\left(x\right)\right|<\frac{\varepsilon}{2}$ for all $n\geq n_{0}$. But for $n\geq n_{0}$, this implies \begin{align*} \left|g_{n}\left(x\right)\right| & \leq\left|g_{n}\left(x\right)-g_{n}^{\left(m_{0}\right)}\left(x\right)\right|+\left|g_{n}^{\left(m_{0}\right)}\left(x\right)\right|\\ & \leq\left\Vert g-g^{\left(m_{0}\right)}\right\Vert _{V}+\frac{\varepsilon}{2}\\ & <\frac{\varepsilon}{2}+\frac{\varepsilon}{2}=\varepsilon. \end{align*} Hence, $g_{n}\left(x\right)\xrightarrow[n\to\infty]{}0$. Since $x\in\mathbb{R}$ was arbitrary, we have shown $g\in V_{0}$, so that $V_{0}\leq V$ is closed and hence complete.

Now, let $\emptyset\neq I=\left(a,b\right)\subset\mathbb{R}$ be an arbitrary open, bounded, nonempty interval and set $$ V_{0}^{\left(I\right)}:=\left\{ \left(f_{n}\right)_{n\in\mathbb{N}}\in V_{0}\,\middle|\,\left(f_{n}\right)_{n\in\mathbb{N}}\text{ does {\color{red}not} converge uniformly on }I\right\} . $$ I claim that $V_{0}^{\left(I\right)}\subset V_{0}$ is open and dense. Indeed,

  1. $\left(V_{0}^{\left(I\right)}\right)^{c}\subset V_{0}$ is closed, because for a sequence $\left(g^{\left(m\right)}\right)_{m\in\mathbb{N}}=\left(\left(g_{n}^{\left(m\right)}\right)_{n\in\mathbb{N}}\right)_{m\in\mathbb{N}}$ in $\left(V_{0}^{\left(I\right)}\right)^{c}$ with $g^{\left(m\right)}\xrightarrow[m\to\infty]{}g=\left(g_{n}\right)_{n\in\mathbb{N}}\in V_{0}$, and $\varepsilon>0$, the following holds: There is some $m_{0}\in\mathbb{N}$ with $\left\Vert g-g^{\left(m\right)}\right\Vert _{V}<\frac{\varepsilon}{2}$ for all $m\geq m_{0}$. Furthermore, since $g^{\left(m_{0}\right)}\in\left(V_{0}^{\left(I\right)}\right)^{c}$, there is $n_{0}\in\mathbb{N}$ with $$ \frac{\varepsilon}{2}>\left\Vert g_{n}^{\left(m_{0}\right)}\right\Vert _{\sup,I}:=\sup_{x\in I}\left|g_{n}^{\left(m_{0}\right)}\left(x\right)\right| $$ for all $n\geq n_{0}$. But this implies \begin{align*} \left\Vert g_{n}\right\Vert _{\sup,I} & \leq\left\Vert g_{n}-g_{n}^{\left(m_{0}\right)}\right\Vert _{\sup,I}+\left\Vert g_{n}^{\left(m_{0}\right)}\right\Vert _{\sup,I}\\ & \leq\left\Vert g-g^{\left(m_{0}\right)}\right\Vert _{V}+\left\Vert g_{n}^{\left(m_{0}\right)}\right\Vert _{\sup,I}\\ & <\varepsilon \end{align*} for all $n\geq n_{0}$ and thus $g=\left(g_{n}\right)_{n\in\mathbb{N}}\in\left(V_{0}^{\left(I\right)}\right)^{c}$.

  2. To see that $V_{0}^{\left(I\right)}\subset V_{0}$ is dense, let $g=\left(g_{n}\right)_{n\in\mathbb{N}}\in V_{0}$ be arbitrary. We want to show $g\in\overline{V_{0}^{\left(I\right)}}$. If $g\in V_{0}^{\left(I\right)}$, this is clear. Hence, we can assume $g\notin V_{0}^{\left(I\right)}$, i.e. $g_{n}\to0$ uniformly on $I$. Now, let $\varepsilon>0$ be arbitrary. We want to find some $h=\left(h_{n}\right)_{n\in\mathbb{N}}\in V_{0}^{\left(I\right)}$with $\left\Vert g-h\right\Vert _{V}\leq\varepsilon$. To this end, let $f_{n}\equiv0$ for $n\leq\frac{2}{b-a}$, i.e. for $a+\frac{2}{n}\geq b$. For $n>\frac{2}{b-a}$, define $$ f_{n}\left(x\right):=\begin{cases} 0, & \text{if }x\leq a,\\ \varepsilon\cdot n\left(x-a\right), & \text{if }a\leq x\leq a+\frac{1}{n},\\ \varepsilon\cdot\left(1-n\cdot\left(x-a-\frac{1}{n}\right)\right), & \text{if }a+\frac{1}{n}\leq x\leq a+\frac{2}{n},\\ 0, & \text{if }x\geq a+\frac{2}{n}. \end{cases} $$ It is clear that $f:=\left(f_{n}\right)_{n\in\mathbb{N}}\in V_{0}$ (pointwise convergence is clear for $x\leq a$. But for $x>a$, we have $x\geq a+\frac{1}{n}$ for all $n\geq n_{x}$ and hence $f_{n}\left(x\right)=0$) with $\left\Vert f\right\Vert _{V}=\varepsilon$. Furthermore, we have $\left\Vert f_{n}\right\Vert _{\sup,I}=\varepsilon$ for all $n>\frac{2}{b-a}$ and hence $\left(f_{n}\right)_{n\in\mathbb{N}}$ does not converge uniformly (to $0$) on $I$. Thus, neither does $h:=\left(h_{n}\right)_{n\in\mathbb{N}}:=\left(f_{n}+g_{n}\right)_{n\in\mathbb{N}}$, because otherwise, $f_{n}=h_{n}+\left(-g_{n}\right)\to0$ uniformly on $I$, by uniform convergence of $\left(g_{n}\right)_{n\in\mathbb{N}}$.

Altogether, $h\in V_{0}^{\left(I\right)}$ with $\left\Vert g-h\right\Vert _{V}=\left\Vert f\right\Vert _{V}=\varepsilon$, so that $V_{0}^{\left(I\right)}\subset V_{0}$ is dense.

Finally, let $$ J:=\left\{ \left(a,b\right)\,\middle|\, a,b\in\mathbb{Q}\text{ and }a<b\right\} . $$ Then $J$ is countable, so that Baire's theorem shows that $$ M:=\bigcap_{I\in J}V_{0}^{\left(I\right)}\subset V_{0} $$ is dense. In particular, $M\neq\emptyset$. But any $f=\left(f_{n}\right)_{n\in\mathbb{N}}\in M$ satisfies $f_{n}\to0$ pointwise, but $\left(f_{n}\right)_{n\in\mathbb{N}}$ does not converge uniformly (to $0$) on any interval from $J$. Since every (nontrivial) interval contains an interval from $J$, we are done.

| cite | improve this answer | |

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.