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

I need some help in finding a (as simple as possible) smooth function $f:\mathbb R \rightarrow \mathbb R$ which does NOT satisfy the following:

There exist a constant $C>0$, a compact $K\subset\mathbb R$ and $h_0>0$ such that for every $|h| \leq h_0$ and every $x\in\mathbb R\setminus K$

$|h|^{-1}|f(x+h) - f (x)| \leq C |f'(x)|$

EDIT: and there exists a $\tilde C>0$ such that $|f'(x)|>\tilde C$ for $x\in\mathbb R\setminus K$.

EDIT 2: my intuition is that such an $f$ may look like this: the first derivative stays always positive and oscillates (around g(x)=|x| for example), the oscillations becoming both faster and larger in amplitude when $x$C goes to infinity.

Many thanks.

share|cite|improve this question
It can't be oscillating if $f'$ is not allowed to have zeros. Also, do you want the "sufficiently small" uniformly in $x$? Maybe try to rephrase the whole question to make it a little clearer. – Lukas Geyer Oct 22 '12 at 16:42
How did ${\mathbb R}^n$ get in there? – Robert Israel Oct 22 '12 at 16:53
@LukasGeyer and Robert Israel. Thanks for your comments, I edited, hope it is clear now (yes the "sufficiently small" should be uniform in $x$). With oscillating I meant not the function itself, but higher derivatives. Anyway I cancelled the comment on the oscillations, maybe it is only confusing. – Hans Oct 22 '12 at 17:00
up vote 4 down vote accepted

Take $$ \begin{eqnarray*} f(x) &=& x + e^x + sin(e^x) \\ f'(x) &=& 1 + e^x + e^xcos(e^x) \end{eqnarray*} $$ Then $f'(x) \geq 1 + e^x - e^x \geq 1$

You have $f'(\log(\pi + 2\pi n)) = 1$ and $f'(\log(2\pi n)) = 1 + 2e^x$, i.e. arbitrarily close points with arbitrarily large differences of $f'$. This looks promising. Take now, for example, the differential quotient at $$ \begin{eqnarray} x &=& \log(\pi + 2\pi n) \\ h &=& \log(2\pi (n+1)) - \log(\pi + 2\pi n) \end{eqnarray} $$ which is $$ \begin{eqnarray} && \frac{h + e^{x+h} - e^{x} + sin(e^{x+h}) - sin(e^{x})}{h} \\ &=& \frac{h}{h} + \frac{2\pi(n+1) - (\pi + 2\pi n)}{h} + \frac{0 - 0}{h} \\ &=& 1+ \frac{\pi}{h} \end{eqnarray} $$

Now, since $h \to 0$ as $n \to \infty$, and since $f'(x) = 1$, the differential quotient at $x$ isn't bound by $f'$ globally.

share|cite|improve this answer
Tanks, this is what I was looking for! – Hans Oct 24 '12 at 15:46

Try e.g. $f(x) = \cos(x)$. All you need is that $f$ is not constant on any interval and $f'$ has arbitrarily large zeros.

EDIT: With the new condition, take $$f(x) = \int_0^x (1 + t^2 \cos^2(t^2))\ dt = x+\frac{x \sin \left( 2\,{x}^{2} \right)}{8}-\frac{\sqrt {\pi }}{16}{\rm FresnelS} \left( 2\,{\frac {x}{\sqrt {\pi }}} \right) +\frac{{x}^{3}}{6} $$ Note that for $x > 0$ $$\eqalign{\frac{f(x+h) -f(x)}{h} &= \frac{1}{h} \int_x^{x+h} (1 + t^2 \cos^2(t^2))\ dt\cr &> \frac{x^2}{h} \int_x^{x+h} \cos^2(t^2)\ dt \cr &= \frac{x^2}{h} \left(\frac{h}{2} + \frac{\sqrt{\pi}}{4} \left({\rm FresnelC}\left( \frac{2(x+h)}{\sqrt{\pi}}\right) - {\rm FresnelC}\left( \frac{2x}{\sqrt{\pi}}\right) \right)\right) \cr &= \frac{x^2}{2} + o(x^2)}$$ as $x \to \infty$ for any fixed $h$ (since ${\rm FresnelC}(t) \to 1/2$ as $t \to \infty$), while $f'(\sqrt{(n+1/2)\pi})=1$ for positive integers $n$.

share|cite|improve this answer
ops, I forgot to write that I want also the first derivative bounded away from zero..see edit. sorry. – Hans Oct 22 '12 at 15:21
I accept the other answer because it is bit simpler, but this is interesting anyway, thanks! – Hans Oct 24 '12 at 15:45

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.