5

Suppose $f\in C^0(\mathbb R)$ is a continuous function on real numbers. If for any $\varepsilon>0$, there exists $\delta>0$ such that $|f(x+\delta)-f(x)|<\varepsilon$ for all $x\in\mathbb R$. Can we deduce that $f$ is unformly continuous on $\mathbb R$?

Some clues: For a particular kind of functions called Bohr almost periodic functions, this is true, cf 2.3.5 of this survey or reference there. I believe this is more or less standard. However Bohr almost periodic functions are all bounded and the condition of my question is quite weaker. Being unformly continuous means any sufficiently small translation $f_\delta$ by $\delta$ approximates the original function $f$ in the uniform norm--$\|f_\delta-f\|_\infty\to0$ as $\delta\to0$. But I can't get such $\delta$ in my question. And neither could I make a counterexample.

1 Answers1

3

We cannot deduce that $f$ is uniformly continuous

Preparations

Choose a conditionally convergent series $\sum_{k=0}^{\infty} a_k$, e.g.

$$\sum_{k=0}^{\infty} a_k = 1 - \frac{1}{2} + \frac{1}{3} - \frac{1}{4} + \frac{1}{5} - \frac{1}{6} + \frac{1}{7} - \frac{1}{8} + \ldots$$

Given $n \in \mathbb{Z}$, let $b_n$ denote the sum of all $a_k$ such that the ($2$-adic) binary expansion of $n$ has a $1$ on position $k$. In symbols:

$$b_n = \sum_{k=0}^{\infty} c_k \cdot a_k, \qquad \text{where in } \mathbb{Z}_2:\qquad n = \sum_{k=0}^{\infty} c_k \cdot 2^k, c_k \in \{ 0, 1 \}.$$

Note that for negative $n$ ultimately $c_k = 1$, so the series defining $b_n$ converges to a real number.

Claim. $b_{2^k+n} - b_n \xrightarrow{k \to \infty} 0$ uniformly with respect to $n$.

Proof. The binary expansions of $n$ and $n+2^k$ differ in the following way, where $k \leqslant m \leqslant \infty$:

$$\begin{align*} n &= *** \ \overset{\substack{m \\ \downarrow \\ \phantom{0}}}{0}11 \ldots \overset{\substack{k \\ \downarrow \\ \phantom{0}}}{1} \ *\mathrel{*}\overset{\substack{0 \\ \downarrow \\ \phantom{0}}}{*} \\ 2^k+n &= *** \ \underset{\substack{\phantom{0} \\ \uparrow \\ m}}{1}00 \ldots \underset{\substack{\phantom{0} \\ \uparrow \\ k}}{0} \ *\mathrel{*}\underset{\substack{\phantom{0} \\ \uparrow \\ 0}}{*} \end{align*}$$

It follows that if $m < \infty$,

$$\left| b_{2^k+n} - b_n \right| \leqslant |a_m| + \left| \sum_{i=k}^{m-1} a_i \right|$$

and the right hand side approaches zero because of the Cauchy condition.

If $m = \infty$, the reasoning is similar. $\square$

Counterexample

Define $f(x) = b_n \cdot \sin^2( \pi x )$ on each interval $[n, n+1]$ with $n \in \mathbb{Z}$.

  • Clearly $f \equiv 0$ on the integers where the piecewise definitions meet, so $f$ is continuous.

  • For any $k \in \mathbb{N}$, $n \in \mathbb{Z}$ and $x \in [n, n+1]$ we have $$|f(x+2^k) - f(x)| = |b_{n+2^k} \sin^2 \left( \pi \big( x + 2^k \big) \right) - b_{n} \sin^2( \pi x )| = |b_{n+2^k} - b_n| \cdot \sin^2( \pi x ).$$ It now follows from the Claim that $f$ satisfies the assumption: given $\varepsilon > 0$, take $\delta = 2^k$ where $k$ is so large that $|b_{n+2^k} - b_n| < \varepsilon$ for all $n \in \mathbb{Z}$.

  • $f$ is not uniformly continuous because $(b_n)$ is unbounded.

Thus the implication does not hold.

Adayah
  • 13,487