Under what conditions is the sum of two positive semidefinite matrices positive definite?

123 Views Asked by At

My question is different than this question because the latter asks whether it is possible, but I am asking under what general condition(s) it is possible.


Question: Under what condition is the sum of two (symmetric) positive semidefinite matrices positive definite? Let $A$ and $B$ be two $n \times n$ positive semi-definite matrices, i.e., $x^{\top}Ax \geq 0$ and $y^{\top}By \geq 0$ for all $\mathbb{R}^n \ni x\neq 0 \neq y \in \mathbb{R}^n$. What condition on $A$ and $B$ guarantees $x^{\top}(A+B)x > 0$ for all $0 \neq x \in \mathbb{R}^n$?


My thoughts

If there is $0 \neq x \in \mathbb{R}^n$ such that $x^{\top}(A+B)x > 0$, this means that $x \in \text{Null}(A)$ and $x \in \text{Null}(B)$, i.e., or $\text{Null}(A) \cap \text{Null}(B) \neq \emptyset$. However, this condition is not practical. For example, let $A=vv^{\top}$ and $B=ww^{\top}$ for $0 \neq v \in \mathbb{R}^n$ and $0 \neq w \in \mathbb{R}^n$. How can I find a condition on $v, w$? I know in this case, the range of $A$ and $B$ is span{$v$} and span{$w$} but we want the entire space be covered.