We know that, If we start with a basis of a finite dimensional vector space $V$ over $\mathbb{R}$ by using lexicographic ordering with respect to that basis, we have a total ordering $\lt$ on $V$ satisfying,

i) $v \in V, v_1 \lt v_2 \implies v_1 + v \lt v_2 + v$

ii) $ \alpha \in \mathbb{R}, v_1 \lt v_2 \implies $

a) $\alpha v_1 \lt \alpha v_2$ if $\alpha \gt 0$

b) $\alpha v_1 \gt \alpha v_2$ if $\alpha \lt 0 $

I just wondering, given a such a total ordering on $V$ , can we find a basis of $V$ with respect to which the lexicographic order will be the same as the one we start with?

is there any similar theory of infinite dimensional vector spaces?

Thanks in Advance

  • $\begingroup$ I would conjecture that for sucn an order, there is a subspace $U$ of codimension one (a hyperplane) so that all vectors to one side of $U$ are $>0$, and all vectors on the other side are $<0$. If so, you just pick the first vector of your basis on the positive side of $U$, then proceed inductively by doing the same thing in $U$. $\endgroup$ – Harald Hanche-Olsen May 10 '14 at 10:54
  • $\begingroup$ @HaraldHanche-Olsen cant we find that $U$ explicitly or prove its existence?. Then the problem is over as you remarked. Is it really a conjecture? $\endgroup$ – GA316 May 10 '14 at 10:58
  • $\begingroup$ See the answer by @Hans below. Yes, it was a conjecture in the sense that I was pretty sure that is how it goes, but I didn't have the time to cobble together a proof – so I could have been wrong, but as it turns out, I wasn't. $\endgroup$ – Harald Hanche-Olsen May 10 '14 at 15:57

Let $<$ be such a total ordering on $V$ (which is, as in the question, finite dimensional). Then the set $C=\{v \in V: v\geq 0\}$ clearly is a convex cone. Since $C \cap (-C) =\{0\}$ and since $V=(-C) \cup C$ there is a $v_0 \in V$ which does not lie in the closure of $C$. Thus there is a nonzero linear form $l$ on $V$ such that $l \geq 0$ on $C$ (see below in "edit"). Now the zero set of $l$ is a hyperplane as Harald thought of:

Let $l(v)<0$ then $v \not\in C$ thus $v<0$.

Let $l(v)>0$ and assume $v \leq 0$ then $-v \in C$ which contradicts $l(-v)<0$.

edit: This follows from biduality of convex cones, see http://arxiv.org/abs/1006.4894 (section 2.1): Assume that $C^*=\{0\}$ (notation as in the paper), then we have $V=\{0\}^*=(C^*)^*=\textrm{cl}(C)$, but we have $V \neq \textrm{cl}(C)$ ($\textrm{cl}(C)$ is the closure of $C$).

I expect that you can find the proof of this biduality theorem (which uses the separation theorem that I mentioned in the first place) in a standard book about convexity, for example in Barvinok's "A First Course in Convexity".

  • $\begingroup$ what is separation theorem for convex cones? any reference. In google search I couldnt find much. thanks. $\endgroup$ – GA316 May 10 '14 at 12:04
  • $\begingroup$ please see the edit... $\endgroup$ – Hans May 10 '14 at 14:07
  • $\begingroup$ Unfortunately there extsts a counterexample to the hyperplane assumption. $\endgroup$ – Keinstein May 11 '14 at 20:34
  • $\begingroup$ but not in the case of $V$ finite dimensional, which is part of the assumptions in the original question. but perhaps I should mention it in my answer again... $\endgroup$ – Hans May 12 '14 at 9:19
  • 1
    $\begingroup$ Ok. Then you should mention that this is well known as Hahn's theorem in the finite case. There are many introductory books about lattice-ordered should contain it. $\endgroup$ – Keinstein May 13 '14 at 16:18

In case of an infinite dimensional vector space there exists a basis that represents the order.

Nevertheless the separating “Hyperplane” may be the whole vector space itself. For example take $\mathbb{R}^{\mathbb N}$ with the natural orders on $\mathbb R$ and $\mathbb N$. In this vector space every vector is infinitesimal with respect to some other vector. If there were a hyperplane that separates the positive cone and the negative cone, then there would have been a vector that is not infinetesimal with respect to any other vectors.

This topic is discussed in this article.


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.