1. PF Insights is off to a great start! Fresh and interesting articles on all things science and math. Here: PF Insights

Convex set : characteristic cone

  1. Hello :)
    I have been giving a mathematical problem. But I find difficulties solving this. Therefore, I will be very grateful if anybody might wanted to help?
    The problem is

    "Let K be a compact convex set in R^n and C a closed convex cone in R^n. Show that
    ccone (K + C) = C."

    - Julie.
     
  2. jcsd
  3. micromass

    micromass 18,725
    Staff Emeritus
    Science Advisor
    Education Advisor

    Hi wjulie and welcome to PF! :smile:

    I'm not familiar with the terminology "characteristic cone", is it perhaps the same thing as a recession cone? ( http://planetmath.org/encyclopedia/DirectionOfAConvexSet.html )

    It is obvious that [itex]C\subseteq ccone(K+C)[/itex]. Assume that this inclusion was strict, then there would be a direction d which is not in C. This d has a >0 distance from C. Thus the multiples of d grow further away from C. That is, the distance from d to C becomes arbitrarily big. But we still have that d is in ccone(K+C). Can you find a contradiction with that?
     
  4. Thanks for your reply :) and yes, a characteristic cone is the same as a recession cone.
    T S.jpg
    Then I must show that T = S

    I can show that if a belongs to S, then a must belong to T as well.
    Let a=y+d belong to C.
    if y belongs to C, and because C [itex]\subseteq[/itex] C+K, then y must belong to C+K
    Therefore a=y+d must belong to C+K.

    But how about the other way? I'm finding it quite difficult.
     
  5. micromass

    micromass 18,725
    Staff Emeritus
    Science Advisor
    Education Advisor

    I think you made a mistake in your picture since T and S are exactly the same there. :biggrin:

    But I see what you mean. Let's prove this in steps. Let's begin with this: let d be a direction not in C. Can you prove that the distance between x+rd and C becomes arbitrarily large as r becomes large?

    I.e. can you show that [itex]d(x+rd,C)\rightarrow +\infty[/itex] as [itex]r\rightarrow +\infty[/itex]?
     
  6. hmm i can't quite see the trick. But K has no direction because it is compact?
     
  7. micromass

    micromass 18,725
    Staff Emeritus
    Science Advisor
    Education Advisor

    Do you see intuitively why it must be true?
    Consider for example the cone [itex]C=\{(x,0)~\vert~x\in \mathbb{R}\}[/itex] in [itex]\mathbb{R}^2[/itex]. Take something not in C, for example (1,1). Do you see that multiples of (1,1) are getting further away from C? That is, if [itex]r\rightarrow +\infty[/itex], then the distance between (r,r) becomes arbitrarily large.

    The general case is quite the same...
     
  8. i can see the intuitive behind it now. But when i have shown that this distance grow larger, what's next? Where are we heading?
     
  9. micromass

    micromass 18,725
    Staff Emeritus
    Science Advisor
    Education Advisor

    Well, x+rd is getting further away from C. But if d is in ccone(K+C), it must hold that x+rd is in K+C. And thus we must be able to express x+rd=k+c. But as the distance between x+rd and c becomes large, then k must become large. Thus K must be unbounded.
     
  10. Aha! I see. I got it now. Thank you, you have saved my day :)
     
  11. "It is obvious that C⊆ccone(K+C)"

    why is this obvious, please explain ?

    /Olga
     
  12. micromass

    micromass 18,725
    Staff Emeritus
    Science Advisor
    Education Advisor

    See post #3.
     
  13. Well, that isn't a useable argument, in my opinion though...

    Isn't that just at proof of y+d belonging to the set (K+C), and not the characteristic cone(K+C)?
     
  14. micromass

    micromass 18,725
    Staff Emeritus
    Science Advisor
    Education Advisor

    Well, to see that

    [tex]C\subseteq ccone(C+K)[/tex]

    Take d in C, then for all x in C, we have that x+rd is in C. In particular rd is in C.
    Now, take c+k in C+K, then c+k+rd=k+(c+rd) is in C+K (since C is convex). Thus for every x in C+K, we have that c+rd is in C+K
     
Know someone interested in this topic? Share a link to this question via email, Google+, Twitter, or Facebook

Have something to add?