Let $f(x)$ be a real-valued function with $c$ in its domain. Given
1. $f(c)=0, $
2. $f$ is not identically zero in any interval about $c$, and
3. $f$ is differentiable at $c$.

a. If $f(x)$ changes sign at $c$, then $f’(x)$ does not change sign at $c$.
b. If $f(x)$ does not change sign at $c$, then $f’(x)$ changes sign at $c$.

How would I go about proving this claim?

The main difficulty I have with this question is that I cannot express "changing sign" in a way involving derivatives that makes this result amenable to a proof I can come up with.

The one precise meaning of "changing sign" I thought of is:
$f(x)$ changes sign at $c$ if there exists an open set $S$ containing $c$ such that for all $y\in S$ and $y<c$, $f(y)$ is positive (negative) and for all $y\in S$ and $y>c$, $f(y)$ is negative (positive).

It's also clear that: If $f(x)$ changes sign at $c$, $f(c)=0$, but the converse is not true.

Suggestions would be appreciated.

  • $\begingroup$ Hi all - thanks for the suggestion, I have some thoughts I have on this question so far. Suggestions to proceed, or a counterexample, would be appreciated! $\endgroup$
    – FreshAir
    Oct 31, 2015 at 1:13

1 Answer 1


Here's a counterexample to (b) according to your definition.

$$f(x)=x^3sin(1/x), x \neq 0$$ $$f(0)=0$$

According to your definition, neither f nor f' changes sign at $x=0$.

  • $\begingroup$ Your example clearly proves that $f$ differentiable at $c$ is not nearly good enough. And by replacing $x^3$ with $x^n$, it's also easy to show that $n$ times differentiable at $c$ isn't good enough either. That only leaves $C^\infty$ and analytic to check. I suspect all counter examples have bad oscillation like this. Can we prove that? $\endgroup$
    – Zach Stone
    Oct 31, 2015 at 2:44
  • $\begingroup$ @ZachStone, to eliminate the infinite oscillation case, you need to require that on each side of c there exists an open interval where f does not change sign. $\endgroup$
    – Paul
    Oct 31, 2015 at 2:52
  • $\begingroup$ For $C^{\infty}$ couldn't you do something like $e^{-1/x^2}\sin(1/x)$? I'm pretty sure that this is true for analytic though. $\endgroup$
    – Alex Zorn
    Oct 31, 2015 at 2:54
  • 1
    $\begingroup$ @Paul In the case where it has constant sign on either side of $c$, continuity plus mean value theorem combine to give a simple proof of (b). Even better, it doesn't require $f$ to even be continuously differentiable. $\endgroup$
    – Zach Stone
    Oct 31, 2015 at 2:56

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.