I'm trying to find a relation between derivative and uniform continuity on $\mathbb{R}$. First, I found this property:

  1. If $f$ is differentiable on $\mathbb{R}$ and $\displaystyle \lim_{x\rightarrow +\infty} f'(x)=+\infty$, then $f$ is not uniformly continuous on $\mathbb{R}$.

Proving this is not so difficult. Indeed, if $f$ is uniformly continous on $\mathbb{R}$, then for every $\epsilon$, there exists $\delta$ such that for all $x,y$ satisfying $|x-y|<\delta$ then $|f(x)-f(y)|<\epsilon$. Since $f'\rightarrow +\infty$, there exists $X$ such that for all $x>X$, $f'(x)>\frac{2\epsilon}{\delta}$. Choose an arbitrary $x>X$ and consider Lagrange theorem: $$\left|f\left(x+\frac{\delta}{2} \right)-f(x)\right|=f'(c_x)\cdot \frac{\delta}{2}>\epsilon$$

since $c_x>x>X$. This is a contradiction.

Now I want to make the condition a little bit stricter.

  1. If $f$ is differentiable on $\mathbb{R}^+$ and $f'$ is not bounded above on $\mathbb{R}^+$, then $f$ is not uniformly continuous on $\mathbb{R}^+$.

($\mathbb{R}^+$ is just for convinience)

Since $f'$ is not bounded, there exists a sequence $(x_n)$ such that $x_n\rightarrow =\infty$ and $f'(x_n)\rightarrow +\infty$. Then there exists $N$ sucht that $f'(x_n)>\frac{2\epsilon}{\delta}$ for all $n>N$. Since $f$ is continuous, for all $n>N$, there exists a neighborhood $(x_n-\delta_n,x_n+\delta_n)$ such that $f'(x)>\frac{2\epsilon}{\delta}$ for all $x\in (x_n-\delta_n,x_n+\delta_n)$. If $\sup\delta_n>\frac{\delta}{2}$, there exists $k>N$ such that $\delta_k>\frac{\delta}{2}$. Applying Lagrange theorem for arbitrary $x,y\in(x_k-\delta_k,x_k+\delta_k)$ such that $|x-y|=\delta$: $$|f(x)-f(y)|=f'(c_{xy})|x-y|>2\epsilon$$

I'm stuck with the case $\sup \delta_n\le \frac{\delta}{2}$. Cound I continue, or the statement is just false itself?

  • $\begingroup$ why cant the same arguements be repeated in 2 as in 1 $\endgroup$ – Learnmore Jan 14 '15 at 4:49
  • 1
    $\begingroup$ Since in the first statement, $f'\rightarrow+\infty$, while in the second one it is just unbounded. Notice that "unbounded" in not the same as "tending to infinity". Following the proof, if $\sup\delta_n\le \frac{\delta}{2}$, then we can not choose $x,y$ such that $|x-y|=\delta$. $\endgroup$ – Tien Kha Pham Jan 14 '15 at 4:54

Take $$f(x)=\begin{cases}x^2\cos(1/x^2)&\text{ for }x\neq0\\0&\text{ for }x=0\end{cases}$$ Then $$f'(x)=\begin{cases}2x\cos(1/x^2)+2\frac{1}{x}\sin(1/x^2)&\text{ for }x\neq0\\0&\text{ for }x=0\end{cases}$$

Observe that $f(x)$ is continuous in the compact $[-1,1]$. Therefore $f(x)$ is uniformly continuous in $[-1,1]$. But we can see that $f'(x)$ is not bounded in $[-1,1]$.

This is an example where the problem is at the origin and we only define it in $[-1,1]$. To get a function on $\mathbb{R}^+$ you first ignore the part for $x<0$ and continue $f(x)$ to $[1,+\infty)$ by smoothly continuing the graph such that it bends into a constant function for $x$ large.

If we want the problem to be at $+\infty$ we can take

$$g(x)=\frac{1}{x}\cos(x^3)\text{ for }x\geq 1$$

which by a translation later you can assume defined in $\mathbb{R}^+$. Then


The derivative is unbounded. Let us prove $g(x)$ is uniformly continuous in $[1,+\infty)$.

Given $\epsilon>0$.

Choose $M>0$ such that if $x>M$ then $|g(x)|<\epsilon/2$. In particular if $x,y>M$ then $$|f(x)-f(y)|\leq |f(x)|+|f(y)|<\epsilon$$

On the compact $[1,M]$ the function $g$ is continuous and therefore uniformly continuous. Therefore there is $M/2>\delta_1>0$ such that if $x,y\in[1,M]$ with $|x-y|<\delta_1$ then $|f(x)-f(y)|<\epsilon$.

Finally, on the compact set $C:=[M-\delta_1,M+\delta_1]$ the function $g(x)$ is continuous and therefore there is a $\delta_2$ such that if $x,y\in C$ with $|x-y|<\delta_2$ then $|f(x)-f(y)|<\epsilon$.

Finally, take $\min(\delta_1,\delta_2)>\delta>0$, then if $x,y\geq1$ are such that $|x-y|<\delta$ then $x$ and $y$ lay both in one the the sets $[M,+\infty)$, $[1,M]$ or $C$. In each case $|f(x)-f(y)|<\epsilon$ and therefore $g(x)$ is uniformly continuous on $[1,+\infty)$.

Therefore $h(x)=g(x+1)$ does the job on $\mathbb{R}^+$.


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.