3
$\begingroup$

Please see the mean value theorem stated here: MVT.

Note that the MVT stated there requires that the segment connecting $x$ and $y$ lies in $G$. So clearly, MVT works for any pair of points if $G$ is made a convex open set.

Here are my questions:

  1. What happens if $G$ is not convex, but connected? Looking at the proof of MVT, I think the MVT will still work as long as there is a path (a continuous map) from $x$ to $y$. And this is the case in a connected set. Am I right in thinking this?
  2. Suppose the answer to (1) is yes. (Of course, if I am wrong in (1), please ignore this question.)Assume that $G$ is open and connected and let $x,y \in G$. Then there is some path (not necessarily a line segment) connecting $x$ and $y$ in $G$. MVT guarantees that there is a $z$ on this path such that $$f(x) - f(y) = \nabla f(z) \cdot (x - y).$$ Can we say something about the distance between $z$ and the points $x$,$y$? If the path is a line segment, then certainly, $$|x-z|\leq |x-y| \quad \text{and} \quad |y-z| \leq |x-y|.$$ If the path is not a line segment, what happens?
$\endgroup$
  • $\begingroup$ If you look at the proof of MVT as sketched in Wikipedia (your source) it is important that the parametrization of the 'path' joining $x$ and $y$ is linear in the variable $t$. Otherwise (in wikipedia notation) $g'(c) \neq \nabla f((1-c)x + cy)$ there will be extraneous factors involved. If $G$ is not convex the best you can do is partition $G$ into finitely many convex sets $G_{i}$ such that $G_{i} \cap G_{i+1}$ is nonempty and $x \in G_{0}$ and $y \in G_{n}$. Then apply MVT to each partition separately. $\endgroup$ – DBS Jul 26 '13 at 5:25
  • $\begingroup$ But a path $g$ joining $x$ and $y$ (even if it's not linear) still satisfies: $g(0) = x, g(1) = y,$ and of course, $g$ is continuous since it's a path. $\endgroup$ – Alex Strife Jul 26 '13 at 6:15
  • $\begingroup$ Cosider a path $\gamma(t) = (1-t^{2})x + t^{2}.y, 0 \leq t \leq 1$ then $g^{'}(c^{2}) \neq \nabla f((1-c^{2})x + c^{2}y)$ it will be $\nabla f((1-c^{2})x + c^{2}y). < -2c, 2c>$. So it no longer has a nice form. $\endgroup$ – DBS Jul 26 '13 at 6:20
  • $\begingroup$ I see. I get it now. Thanks :) $\endgroup$ – Alex Strife Jul 26 '13 at 7:30
2
$\begingroup$

Question resolved in comments.

$\endgroup$

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.