For a set $S\subseteq\mathbb R^m$ we denote with $pos(S)$ the set \begin{equation} \{\alpha_1x_1+\cdots+\alpha_nx_n:\alpha_i\geq 0,x_i\in S, n\in\mathbb N\}. \end{equation} How to prove that $conv(S)=aff(S)\cap pos(S)$ if $0\notin aff(S)$, where $aff(S)$ is afine hull, and $conv(S)$ is convex hull of $S$?

  • $\begingroup$ You should try showing $\mathrm{conv}(S) \subseteq \mathrm{aff}(S) \cap \mathrm{pos}(S)$ and $\mathrm{aff}(S) \cap \mathrm{pos}(S) \subseteq \mathrm{conv}(S)$ . Start with an arbitrary element of the LHS side and show it satisfies the conditions of the RHS set, then $\mathrm{LHS} \subseteq \mathrm{RHS}$. $\endgroup$ – Rammus May 12 at 17:38
  • $\begingroup$ Well, first inclusion is obvious. $\endgroup$ – Madara Uchiha May 12 at 17:52

Let $C = \operatorname{co} S$, $A= \operatorname{aff} S$ and $P= \operatorname{pos} S$. It is clear that $C \subset A$, $C \subset P$ so $C \subset A \cap P$.

If it is assumed that $0 \notin A$ then we have $A \cap P \subset C$.

To see this, suppose $x \in A \cap P$ and let $x = \sum_k \alpha_k x_n$ with $\alpha_k \ge 0, x_k \in S$. If $\sum_l \alpha_k =1$ we are finished, since $x \in C$, otherwise note that $x'= {1 \over \sum_k \alpha_k} \sum_k \alpha_k x_k \in A$ and the line through $x,x'$ passes through the origin and so $0 \in A$ a contradiction. Hence $0 \in C$.

If the condition $0 \notin A$ is removed then the other inclusion is not true. Take $S= \{ e_1,e_2,e_2+e_2 \} \subset \mathbb{R}^2$. Then $A=\mathbb{R}^2, P = \{x | x \ge 0 \}$ and so $0 \in A \cap P$ but clearly $0 \notin C$ (for example, $\phi(x)= x_1+x_2 \ge 1$ for $x \in C$) so the other inclusion is false in general.

| cite | improve this answer | |
  • $\begingroup$ What if $0\notin aff(S)$? Then, geometric intuition says it should be true, but how to prove it. $\endgroup$ – Madara Uchiha May 12 at 22:46
  • $\begingroup$ Are you changing your question? $\endgroup$ – copper.hat May 12 at 22:51
  • $\begingroup$ Yes, should i edit it? $\endgroup$ – Madara Uchiha May 12 at 22:52
  • $\begingroup$ Your call. I will add an answer in to your question in any event. $\endgroup$ – copper.hat May 12 at 22:57
  • $\begingroup$ Ok, i have edited it. $\endgroup$ – Madara Uchiha May 12 at 22:59

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.