4
$\begingroup$

Let $\mathbb {\overline B}^n\subset \mathbb R^n$ be a closed unit ball in and let $\mathbb B^k\subset \mathbb R^k$ be a open unit ball.

Suppose $F$ is a smooth function on $\mathbb {\overline B}^n\times \mathbb B^k$ that has the following properties.

1) $F(x,y)$ tends to $-\infty$ when $|y|\to 1$ for $(x,y)\in \mathbb {\overline B}^n\times \mathbb B^k$.

2) For any $y_0\in \mathbb B^k$ the strict minimum of $F$ on the ball $\mathbb {\overline B}^n\times y_0$ is attained in the interior of the ball. (Here by strict I mean, that the minimum of $F$ on the boundary of $\mathbb {\overline B}^n\times y_0$ is larger than the minimum of $F$ over the whole ball.)

Question. Is it true that $F$ has a critical point $(x_0,y_0)$ in $\mathbb {\overline B}^n\times \mathbb B^k$ such that $F$ restricted on $\mathbb {\overline B}^n\times y_0$ has a minimum at $(x_0,y_0)$? If yes, I would like to know if there are some natural generalizations of this statement. If no, could one impose some additional conditions that guarantee a positive answer?

$\endgroup$
2
  • $\begingroup$ When you say "strict minimum" does this imply that the minimum is attained at a single, unique point? $\endgroup$ Sep 13, 2016 at 13:42
  • $\begingroup$ Jaap, thanks for your question. I only mean that the values of F on the boundary of the ball are larger than the minimum of F in the ball. I'll will adjust the question accordingly $\endgroup$
    – aglearner
    Sep 13, 2016 at 14:09

1 Answer 1

5
+300
$\begingroup$

I think the following is a counter example to your first question. Let $k = n = 1$, so we consider $(x,y) \in [-1,1] \times (-1,1)$. Construct $F$ such that for $y_0 \le 0$ the minimum of $F(\,\cdot\,,y_0)$ is attained at $x = -1/2$ with value $F(-1/2,y_0) = \frac{-1}{y_0+1}$, and for $y_0 \ge 0$ construct a minimum at $x = 1/2$ with $F(1/2,y_0) = \frac{1}{y_0-1}$, and extend $F$ smoothly around this and growing along $x$.

I think this contradicts your requested result, since the set of minima for fixed $y_0$ is precisely those given above, while those are not critical points since $\frac{\partial F}{\partial y} \neq 0$ there.

One way of making sure that you have a critical point, is if you have a differentiable path of minima $x(y_0)$ for $F(\,\cdot\,,y_0)$. Then you can restrict $F$ to that path and apply a version of the mean value theorem to conclude that $\frac{\partial F}{\partial y}(x(y_0),y_0) = 0$ for some $y_0$.

$\endgroup$
2
  • $\begingroup$ Thanks Jaap! Indeed this looks like a counterexample. Do you think you can elaborate this example further so that $F$ does not have any critical point in $[-1,1]\times (-1,1)$? $\endgroup$
    – aglearner
    Sep 13, 2016 at 21:10
  • $\begingroup$ My feeling is that there always exists a critical point, because you have directions along $x$ where $f$ increases and along $y$ where it decreases. This creates some kind of saddle configuration, that I think should always contain a critical point. Maybe Morse theory or some adaptation of the (finite-dimensional) mountain pass lemma may help you here, but I don't know these topics well. $\endgroup$ Sep 14, 2016 at 19:39

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.