Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Rudin asked:

For $f\in L^{\infty}(\mathbb{R}^{1})$, define $f_{t}(x)=f(x-t)$. Assume that $$\lim_{t\rightarrow 0}|f_{t}-f|_{\infty}=0$$Prove that under these conditions there is a uniformly continuous function $g$ such that $g=f$ almost everywhere.

The suggested strategy is to find a sequence of functions $h_{n}\rightarrow \delta$, and use $g_{n}=h_{n}*f$. It is easy to show that $g_{n}\rightarrow f$ almost everywhere(for example, a proof by contradiction or by dominated convergence theorem). However to bridge from $g_{n}$ to $g$ I encountered a difficulty; the classical Arzelà–Ascoli only works in compact spaces, and $g_{n}$ is not compact supported in general. Therefore it is not clear if $g_{n}\rightarrow g$ uniformly even if we only looking for a subsequence of the original sequence.

share|cite|improve this question
What's $\delta$ here? Anyway, would it not be possible to select a sequence $h_n$ of compactly supported continuous functions with the desired limit? The convolution would preserve the nice properties you're looking for. – anonymous Jan 13 '13 at 17:35
$R$ is not compact. – Bombyx mori Jan 13 '13 at 17:39
Then why not just select a family $h_n$ of $C^{\infty}$ bump functions which form an approximate identity? Then $g_n$ is certainly a continuous function of compact support, since the convolution of two functions is at least as nice as the nicest. – anonymous Jan 13 '13 at 17:39
$h_{n}$ can be as nice as possible, and $g_{n}$ is uniformly continuous since $|g_{n}(x+t)-g_{n}(x)|\le |f_{t}-f|_{\infty}$. But $g_{n}(x)=\int_{A}f(x+y)h(y)dy$ is not necessarily compact supported. It is well defined for any $x$. – Bombyx mori Jan 13 '13 at 17:41
Is there a more direct proof? The condition above is 'essentially' uniform continuity. – copper.hat Jan 13 '13 at 22:15
up vote 1 down vote accepted

Here is another way: (I thought I had an elementary proof, but I had to rely on a not entirely elementary result.)

Let $\phi(t) = \text{esssup}_x |f(x+t)-f(x)|$. We have $\lim_{t\to 0} \phi(t) = 0$.

Since $f \in L^\infty(\mathbb{R})$, we have $f \in L^1(K)$ for any bounded set $K$. Let $f_n(x) = \frac{n}{2} \int_{-\frac{1}{n}}^\frac{1}{n} f(x+t) dt $ (and note that $f_n(x) = \frac{1}{2} \int_{-1}^1f(x+\frac{t}{n}) dt$). Since $f$ is essentially bounded, it follows that $f_n$ is continuous, and hence $f_n \in C([-M,M])$ for any $M>0$.

Suppose $n,m \geq N$, then we have \begin{eqnarray} |f_n(x)-f_m(x)| &\leq& \frac{1}{2} \int_{-1}^1 |f(x+\frac{t}{n})- f(x+\frac{t}{m})|dt \\ &\leq & \frac{1}{2} \int_{-1}^1 |f(x+\frac{t}{n})-f(x)|+|f(x)- f(x+\frac{t}{m})|dt \\ &\leq& \sup_{|t| \leq \frac{1}{N}}\phi(t) \end{eqnarray} Note that by assumption $\lim_{N \to \infty} \sup_{|t| \leq \frac{1}{N}}\phi(t) = 0$. Choose $M>0$, then the above shows that $f_n$ is Cauchy in $C([-M,M])$, and hence converges to some $\hat{f} \in C([-M,M])$. It is clear that if $M'>M$, then is is clear that the limit coincides on $[-M,M]$, hence this procedure defines a continuous function $\hat{f}$ on all of $\mathbb{R}$.

Now for the not entirely elementary part: Since $f \in L^1(K)$ for any bounded $K$, it follows from the Lebesgue differentiation theorem that almost every $x$ is a Lebesgue point, ie, $\lim_{n \to \infty} f_n(x) = f(x)$ a.e. Since $\lim_{n\to \infty} f_n(x) = \hat{f}(x)$ everywhere, and $\hat{f}$ is continuous, we have the desired result.

share|cite|improve this answer
Unfortunately, my approach is not that different after all. If one lets $h_n = \frac{n}{2} 1_{[-\frac{1}{n},\frac{1}{n}]}$, then $f_n = f * h_n$. – copper.hat Jan 14 '13 at 6:50
Indeed this is what Rudin hinted explicitly in his book. – Bombyx mori Jan 14 '13 at 10:01

You do not need Arzelà–Ascoli here; the fact that $(g_n)$ converge uniformly can be shown directly. The key point is that $h_n$ should have small support when $n$ is large. This makes sure that the convolution integral involves only $f_t$ for small $t$.
$$ \sup |g_n-g_m| = \sup_x \left|\int h_n(t)f_t(x)\,dt- \int h_m(t)f_t(x)\,dt\right| \\ \le \epsilon + \sup_x \left|\int h_n(t)f(x)\,dt- \int h_m(t)f(x)\,dt\right| \\ = \epsilon + \sup_x \left|f(x)\cdot 1 - f(x)\cdot 1 \right| = \epsilon $$ Here the second step is based on $\|f_t-f\|_\infty\le \epsilon$.

share|cite|improve this answer
Thanks! I should ask - is there a nice way to show $g_{n}\rightarrow f$ a.e? My proof passing from convergence almost everywhere to convergence in measure (because $h_{n}$ as compact support) and then by contradiction feels over-complicated. The dominated convergence theorem gives pointwise convergence, so I feel I might have made a mistake. – Bombyx mori Jan 13 '13 at 22:28
PavelPM, how did you become 5PM? – Bombyx mori Jan 14 '13 at 7:42

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.