2
$\begingroup$

Let $A,B$ be two sets. We say the hyperplane $\langle a,x\rangle =c$ separates $A,B$ if $A\subset H^-$ and $B\subset H^+$, that is $$ x\in A \implies \langle a,x\rangle \leq c\\ x\in B \implies \langle a,x \rangle \geq c $$ I am confused about the fact that both of these inequalities allow equality. If both of them are equal than can't the hyperplanes be touching at one point? This doesn't make sense to me unless we allow the half spaces to be closed (Which I guess we must).

I am wondering then, why the separating hyperplane theorem requires that $A\cap B =\emptyset$ though? Is it just that the separating hyperplane theorem only applies when the intersection is empty, but there can be a separating hyperplane when the sets touch at one point? (my interactions with separating hyperplanes up to this point was only through this theorem, and I did not/still don't fully understand it, so until now I had assumed that the sets must have an empty intersection, which seems to perhaps not be true in order to have a separating hyperplane, but which obviously must be true to apply the separating hyperplane theorem)

Also, in the case where there is equality, then the supporting hyperplane theorem will always be applicable as well, won't it (on both sets given they are convex), since the point the hyperplane passes through seems like it must be on the boundary of each of the sets.

Assume convex, non-empty sets.

Thanks

$\endgroup$
  • 1
    $\begingroup$ The inequalities should be strict. $\endgroup$ – Lee Mosher Jan 16 '16 at 2:38
  • 2
    $\begingroup$ @LeeMosher Some authors distinguish "separates" and "strictly separates" which makes this post confusing for me. $\endgroup$ – Cameron Williams Jan 16 '16 at 2:40
  • 1
    $\begingroup$ The question is: if you have to circles touching at the origin, the tangent line to both is a hyperplane. Does it separate them? $\endgroup$ – Martin Argerami Jan 16 '16 at 2:45
  • $\begingroup$ @MartinArgerami: One can be sticky and say that it only separates their interiors. =) $\endgroup$ – user21820 Jan 16 '16 at 6:48
  • 1
    $\begingroup$ I am used to seeing "separates" and "strictly separates". The separating hyperplane theorem I guess requires "strictly separates". However, I am wondering if two sets are not "strictly separated", whether there is still a hyperplane that "separates" them. From the definition of (not strictly) separates, I think the answer is yes? $\endgroup$ – majmun Jan 16 '16 at 16:48
1
$\begingroup$

The answer is yes (or perhaps it depends on your definition). With the definition I'm using, the two sets can be tangent at a point -- like two circles that both touching the the origin -- and still say a hyperplane separates them (but not strictly separates).

$\endgroup$

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.