1

Let $H_n$ be a $(n+1)\times (n+1)$ real symmetric matrix, and let $D_0,D_1,\dots, D_n$ be the leading principal minors of $H_n$.

What I know is:

  1. If $H_n$ is positive definite (resp. positive semi definite), then $D_n> 0$ (resp. $D_n\geq 0$).
  2. If $D_k>0$ for all $0\leq k\leq n$, then $H_n$ is positive definite (by Sylvester's criterion).

What I want to know is, assuming that $H_n$ is positive semi-definite,

$\quad$ Q1. If $D_n>0$, then $H_n$ is positive definite.

$\quad$ Q2. If $H_n$ is not positive definite, then $D_n=0$.

For Q1: I believe it's done by induction over $n$. For $n=0$: If $D_0>0$, then $H_0$ is positive definite, by second point. For $n=1$: If $D_1>0$, how do you know that $D_0\neq 0$, so that we can use second point again?

For Q2: We know that $H_n$ is positive semi-definite by assumption, so $D_n\geq 0$ by first point. But, since $H_n$ is not positive semi-definite, we can't have $D_n>0$, so $D_n=0$. Is that it?

James2020
  • 573
  • The answer to both questions is yes. In other words, a positive semidefinite matrix is positive definite if and only if it is invertible (has non-zero determinant). – Ben Grossmann Aug 19 '20 at 18:58
  • @BenGrossmann Wow, that's an useful info you gave me there. Could you please inform me where you got this from? I'd love to know the proof of "if"-part. – James2020 Aug 19 '20 at 19:56
  • It's a pretty standard fact, normally taken as a consequence of the following: a symmetric matrix is positive definite if and only if its eigenvalues are real and positive semidefinite if and only if its eigenvalues are non-negative. From there, the determinant of a matrix is the product of its eigenvalues – Ben Grossmann Aug 19 '20 at 20:13
  • For a more direct proof, it suffices to note that for a (symmetric) positive semidefinite matrix $H$, we have $x^THx = 0 \iff Hx = 0$. In my post here, I prove this in a few different ways. From there, note that a matrix has zero determinant if and only if its nullspace (AKA kernel) is non-trivial) – Ben Grossmann Aug 19 '20 at 20:16
  • It is settled, thank you. If you can write your two first comments in an answer, I'll accept it. – James2020 Aug 21 '20 at 14:27

1 Answers1

1

A positive semidefinite matrix is positive definite if and only if it is invertible (has non-zero determinant).

This is normally taken as a consequence of the following: a symmetric matrix is positive definite if and only if its eigenvalues are real and positive semidefinite if and only if its eigenvalues are non-negative. From there, we note that the determinant of a matrix is the product of its eigenvalues.

For a more direct proof, it suffices to note that for a (symmetric) positive semidefinite matrix $H$, we have $x^THx = 0 \iff Hx = 0$. In my post here, I prove this in a few different ways. From there, note that a matrix has zero determinant if and only if its nullspace (AKA kernel) is non-trivial.

Ben Grossmann
  • 225,327