3
$\begingroup$

I have a reference request which I hope some reader here can help me with.

I have encountered a set that has all the properties that one would expect from a polyhedral set (in the sense of finite dimensional convex analysis - an intersection of finitely many half-spaces), however in my case the number of intersecting half-spaces could be infinite. I am interested in things such as extreme points, extreme rays etc. However, the catch is that the set is itself in the space of functions. I am therefore looking for a principled generalization of the concept related to polyhedra to infinite-dimensional spaces. Could someone help me out with this?

$\endgroup$
6
  • 2
    $\begingroup$ If you have a finite number of half-spaces, then presumably you have a finite number of linear functions. Then you want to consider the span of these linear functions. This would give you a map from your infinite-dimensional space to a finite-dimensional space and all of the polytope business comes from the latter. So I think that you can separate the infinite-dimensional issues and the polyhedral issues. $\endgroup$ Feb 6, 2014 at 19:41
  • $\begingroup$ You might look at Paolo d'Alessandro, "Generalizing polyhedra to infinite dimension," 2011. PDF download link. $\endgroup$ Feb 6, 2014 at 19:42
  • $\begingroup$ @LevBorisov: The map you describe is in fact a projecion onto a finite-dimensional space in a direction parallel to each of the defining half-spaces. Therefore the original, infinite-dimensional polytope is the Cartesian product of a finite-dimensional one with an infinite-dimensional space. You should write your comment in the "answer" box. $\endgroup$ Feb 6, 2014 at 20:35
  • $\begingroup$ In my case, I could have infinitely many half-spaces intersecting. Can something be said about this case? (I edited my question with this clarification) $\endgroup$
    – Ankur
    Feb 7, 2014 at 2:59
  • 1
    $\begingroup$ In the case of an countably infinite number of half-spaces, it should also be mentioned whether the set of normal directions is discrete or dense on the unit sphere. $\endgroup$ Feb 7, 2014 at 11:59

3 Answers 3

5
$\begingroup$

Intersecting countably many half-spaces, you can get a general closed convex set (in a separable Banach space, say). For extreme points and other interesting things, I like:

Robert R. Phelps, Lectures on Choquet's Theorem, second edition, Lecture Notes in Mathematics, Vol. 1767. Springer- Verlag, 2001.

$\endgroup$
3
$\begingroup$

The theorem of Hahn Banach states that EVERY closed, convex subset of a normed space, say, is the intersection of an infinite family of closed half planes so that you are essentially talking about the geometry of such sets. A suitable introductory text would be the popular "Geometric Functional Analysis" by R.B. Holmes.

$\endgroup$
0
$\begingroup$

For a very good reference see Eckland and Turnbull, Infinite-Dimensional Optimization and Convexity.

$\endgroup$

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.