Let $f$ be a some real-valued function differentiable at least three times (I want to say twice, or maybe twice with a continuous second derivative, but I'll play it safe).

A saddle point of $f$ is some value $c$ in its domain, not a local extremum, which satisfies $f'(c) = 0$.

Proposition: $c$ is a saddle point of $f$ if and only if $f'(c) = 0$, $f''(c) = 0$ and $\exists \epsilon \in \mathbb{R_+}$ such that $\forall y \in (c - \epsilon, c)$ and $\forall x \in (c, c+ \epsilon), f''(x) \cdot f''(y)≤0$.

In other words, the second derivative "changes sign" at $c$. Intuitively, I am hypothesizing that a point is a saddle point iff it's a critical point and an inflection point.

Does this hold?

  • $\begingroup$ Consider the function $f(x) = \begin{cases} (x+1)^3, & x<-1 \\ 0, & -1\le x\le 1 \\ (x-1)^3, & x>1\end{cases}$. What type of point is $(0,0)$? $\endgroup$ – user137731 Jan 10 '16 at 5:21
  • $\begingroup$ (I deleted the previous comment). But could you elaborate, perhaps? $\endgroup$ – MathematicsStudent1122 Jan 10 '16 at 5:29
  • $\begingroup$ This function is at least twice differentiable (I think it's probably three times differentiable, but I'm too lazy to check now). The point $(0,0)$ is certainly not an extremum, thus by your definition it is a saddle point. But taking $\epsilon < 1$ we get $f''(x)\cdot f''(y)=0$. $\endgroup$ – user137731 Jan 10 '16 at 5:32
  • $\begingroup$ @Bye_World Was my definition of a "saddle point" careless? I never considered constant functions. $\endgroup$ – MathematicsStudent1122 Jan 10 '16 at 5:34
  • $\begingroup$ It's fine -- I believe that's the usual definition. But I just showed an example where you can have a saddle point that is not an inflection point. So your proposition is false. $\endgroup$ – user137731 Jan 10 '16 at 5:35

Consider the function $f(x):=x^n\sin{\displaystyle{1\over x}}$ $(x\ne0)$, $\ f(0):=0$, with $n\geq2$ to your liking.

Concerning you comment below: It's not clear what your exact definition of a saddle point is. In any case, if $f(0)=f'(0)=0$ then the following formula, obtained through partial integration, may be of interest to you: $$f(x)=\int_0^x(x-t)f''(t)\>dt\ .$$

  • $\begingroup$ I understand; the function's derivative vanishes at 0 but it's not an extremum. And the second derivative will change sign infinitely in any neighbourhood about 0. So this is a valid counterexample. $\endgroup$ – MathematicsStudent1122 Jan 10 '16 at 18:59
  • $\begingroup$ But, if I may ask another question, is there a counterexample for the biconditional in the other direction? Namely, you have something satisfying the condition but which is not a saddle point. $\endgroup$ – MathematicsStudent1122 Jan 10 '16 at 19:00

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.