Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Let $e_i$ denote a column-vector of length $n$ whose entries are all zero except for the $i$-th entry that is 1. Now consider the set of $n\times n$ matrices given by $$\mathcal{M}_n=\left\lbrace \left(e_i-e_j\right)\left(e_i-e_j\right)^\mathrm{T}\mid 1\leq j<i\leq n\right\rbrace\cup\left\lbrace \left(e_i+e_j\right)\left(e_i+e_j\right)^\mathrm{T}\mid 1\leq j<i\leq n\right\rbrace.$$

My question is that can we obtain all $n\times n$ real symmetric positive-semidefinite matrices as conic combinations of matrices in $\mathcal{M}_n$?

share|improve this question

1 Answer 1

up vote 1 down vote accepted

No. This is not the case for any $n$. For $n=1$, $\mathcal M_n$ is empty, so the cone consists only of the zero matrix, whereas there are non-zero positive-definite matrices. For $n=2$, the diagonal elements of a matrix in the cone are always equal, which need not be the case for a positive-semidefinite matrix. For $n\gt2$, the matrix with all entries $1$ is positive-semidefinite, but it's not in the cone, since for a matrix in the cone the sum of diagonal elements cannot be less than the sum of off-diagonal elements.

share|improve this answer
    
I made a mistake when I wrote the second set in $\mathcal{M}_n$. For the second set $i=j$ is permitted. Would that make any difference? –  S.B. Sep 27 '12 at 1:59
1  
@S.B.: It certainly makes a difference for $n=1$, where you now have the multiple of the identity matrix that you need. I'm not sure for $n=2$, but for $n\gt2$ it doesn't make a difference, since this new matrix can only be used to increase, not to decrease the diagonal elements. –  joriki Sep 27 '12 at 2:02
    
I see. Thanks for your simple counter example. Is it possible at all to express the real symmetric PSD matrices as a conic combinations of a finite set of matrices? –  S.B. Sep 27 '12 at 2:11
    
@S.B.: I don't think so. For $n=2$, the matrix $$ \pmatrix{a&b\\b&c} $$ is positive-semidefinite iff $a\ge0$, $c\ge0$ and $ac\ge b^2$. I don't think a curved region like that can be the conical hull of a finite set of points, which I would expect to be bounded by hyperplanes. –  joriki Sep 27 '12 at 2:25

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.