I am interested in the properties of symmetric hollow matrices (diagonal full of zeros) for which all off-diagonal terms are strictly positive. An example for $n=3$ would be

$$ \left[\begin{array}{ccc} 0 & A_{21} & A_{31}\\ A_{21} & 0 & A_{32}\\ A_{31} & A_{32} & 0 \end{array}\right] $$ where $A_{21}$, $A_{31}$ and $A_{32}$ are all strictly positive. Let's call the set of these matrices $B$.

Now consider the subspace $S:\sum_i x_i=0$. I want to show that matrices in $B$ are negative definite on $S$. I.e. for all $x\in S$ and for a matrix $A\in B$ we have $x'Ax\leq -d \left\Vert x\right\Vert $ for some $d>0$. This seems to be the case for $n=2$ and $n=3$ but I would like a general proof. If the result does not hold, could we show that matrices in $B$ are negative semi definite on $S$ instead?

  • $\begingroup$ Hi user_lambda, I'm also looking for references for hollow matrices that are neg-def on subspace S. Have you found any good sources? Thanks! $\endgroup$ – jim h Jan 21 '19 at 1:31
  • $\begingroup$ Not really. If you find anything please let me know! $\endgroup$ – user_lambda Jan 22 '19 at 1:03
  • 1
    $\begingroup$ I don't know much, but you can try Donoghue 1974 Monotone matrix functions and analytic continuation, Micchelli 1986 Interpolation of scattered data: distance matrices and conditionally positive definite functions $\endgroup$ – jim h Jan 23 '19 at 21:01

Not true. For example, with $n=3$ take $$x = \pmatrix{1\cr 1\cr-2\cr}$$ with $A_{21}$ very large. $A_{21} > 2 A_{31} + 2 A_{32}$ will do.

| cite | improve this answer | |

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.