Proof Verification of a Linear Algebra problem

73 Views Asked by At

Consider a symmetric and positive semidefinite matrix $A \in \mathbb{R}^{n \times n}$ Assume additionally that there exists no vector $x \in \mathbb{R}^n \setminus \{ 0 \}$ which satisfies $x^T A y = 0$ for all $y \in \mathbb{R}^n$. Prove that $A$ is positive definite.


My attempt:

Since $A$ is psd, for any $\bf z\in\mathbb{R}^n\setminus\{0\}$ we have $\bf z^{T}A\bf z\ge0$. Now according to the question $\not\exists\bf x\in\mathbb R^n\setminus\{0\}$ such that $\bf x^{T}A\bf y=0$ for all $\bf y\in\mathbb R^n$. In particular, if we take $\bf y=\bf z$, then $\bf z^{T}A\bf z\ne0$, but since $A$ is psd $\bf z^{T}A\bf z\ge0$. This implies $\bf z^{T}A\bf z >0$. Thus $A$ is positive definite.

Is this proof correct? If not where is the flaw?

Thank you.

1

There are 1 best solutions below

0
On

As stated in the comments, your proof does not work. However, the result is true. Indeed, if $A$ was merely positive semidefinite and not positive definite, then there would have to be some nonzero vector $x$ such that $x^t A x = 0$. Now by the spectral theorem, there is some orthogonal matrix $B$ such that $A = BDB^t$ for $D$ diagonal. Now let $y = B^{-1}x = B^t x \neq 0$. $x^t A x = 0$ implies that $(By)^t (BDB^t) (By) = 0$. This left hand side is equal to $y^t B^tBDB^tBy = y^t D y$. Let $y^t = (y_1, \dots, y_n)$. Then $0 = y^t D y = \sum D_{ii} y_i^2$. As $A$ is positive semi-definite $D$ is as well, so all the $D_{ii} \geq 0$. Indeed, if $D_{ii} < 0$ then $e_i^t D e_i < 0$, a contradiction. Thus, each $D_{ii} y_i^2 \geq 0$ and their sum is zero, so each $D_{ii} y_i^2 = 0$. Now, $y \neq 0$ so some $y_j \neq 0$. Thus, that corresponding $D_{jj} = 0$. In other words, $e_j$ is an eigenvector of $D$ of eigenvalue $0$. $ABe_j = B D e_j = 0$, so $0 \neq Be_j$ is an eigenvector of $A$ of eigenvalue $0$. Now, let $z \in \mathbb R^n$ and we shall compute $(B e_j)^t A z$. Indeed, as $A$ is symmetric this is $(Be_j)^t A^t z = (ABe_j)^t z = 0$. But $z$ was chosen arbitrarily and as $B$ is orthogonal, $B e_j \neq 0$. We assumed that no such nonzero vector would exist, so $A$ must have been positive definite.