MathOverflow is a question and answer site for professional mathematicians. 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'm a bit embarrassed to admit that:

a) This is a rather unmotivated question.

b) I can't remember whether or not I've asked this before, but searching doesn't seem to turn anything up so ...

Consider some "shape" function $\phi: \mathbf{R} \to \mathbf{R}$. Then given some function $f: \mathbf{R} \to \mathbf{R}$, one can ask whether the "difference quotient",

$\lim_{y\to x} \frac{f(y)-f(x)}{\phi(y-x)}$,

exists at various points $x$. Letting $\phi(x) = x$ corresponds to taking normal derivatives, and intuitively when the limit exists this means that near $x$, the function $f$ "looks like" $\phi$ does near 0.

However, if the ratio $\phi(x)/x$ is not bounded above or away from 0 as $x\to 0$ (I'm mostly thinking of the case when it is neither, so that $\phi$ is "wildly oscillating" in some sense), then anywhere the above limit exists and is nonzero, the function $f$ is necessarily non-differentiable.

My question: If $\phi$ is some wildly oscillating function as described above (pick your favorite), can there be an $f$ for which this limit exists everywhere?

(Edit: I suppose I really want $\phi$ and $f$ to be continuous functions.)

share|cite|improve this question
Do you want the limit to be nonzero almost everywhere? – Douglas Zare Mar 15 '10 at 22:50
How about taking $\phi$ to be a discontinuous additive function and $f=\phi$? – Jonas Meyer Mar 15 '10 at 22:51
@Douglas Zare Not necessarily. Is there an easy (non-constant) example if it isn't required? @Jonas Meyer Good point. I guess I really want $\phi$ to be continuous, and $f$ to be a continuous nowhere-differentiable function. I really should have put the adjective continuous in a lot of places. – Mike Hall Mar 16 '10 at 0:32
@Mike, Any $C^{0,\alpha}$-function $f$ and $\phi(x)=|x|^\beta$ for $0<\beta<\alpha<1$ will do. – Anton Petrunin Mar 16 '10 at 1:12
But Anton, I don't think that meets the "wildly oscillating" criterion. – Jonas Meyer Mar 16 '10 at 1:18
up vote 5 down vote accepted

Assume WLOG that $\phi(x)>0$ when $x>0$. Since the limit described exists for all $x$ in the source of $f$. We get for any $x$ the bound:

$f(x+\delta)-f(x) \leq C\phi(\delta)$

for $0 < \delta < \delta_0$ for some $C,\delta_0>0$ which may depend on $x$.

diving by $\delta$ we get by the assumptions on $\phi$ that

$\underline{\lim}_{\delta \to 0} ( \frac{f(x+\delta) - f(x)}{\delta}) \leq 0$

This is one the four derivatives of $f$, and proposition 2 chapter 5 in Real Analysis by H.L. Royden states that if $f$ is continuous then it is (non-strictly) decreasing. Similar for increasing. So $f$ is constant.

share|cite|improve this answer
“Assume WLOG that $\phi(x)>0$ when $x>0$.” Why? M Hall specifically wants to consider the ‘wildly oscillatory’ case. (I am supposing that $x \mapsto x\sin(1/x)$ is wildly oscillatory, anyway.) – L Spice May 9 '11 at 2:13
Wildly oscilating refers to $\phi(x)/x$ being both close to zero and unbounded for very small $x$. The assumption WLOG refers to the fact that the limit would not exist if $\phi$ were not strictly positive or negative in a small $(0,\epsilon)$ (assuming continuity). – Thomas Kragh May 27 '11 at 15:12

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.