I have a set of real variables $\{a_1,a_2,a_3,...,a_n\}$.
Let $u=\sqrt{a_1^2+a_2^2+...+a_n^2}$. I am trying to prove that the following matrix is positive definite: $A\in \mathbb{R}^{n\times n}$,
$$A_{ij}=-a_i a_j$$ $$A_{ii}=u+u^2-a_i^2$$ where $i,j=1,2,...,n$, and $i\neq j$.
I can't find a method to prove it. I tried proving directly using $x^TAx$, but I didn't get anywhere. Calculating eigenvalues seem to be getting too complicated too fast. Same with determinants.
What I understood so far, is that matrix is symmetric, and that all values in its main diagonal are non-negative. All values on the main diagonal are positive if there is at least one $a_i$ that is nonzero, for some $i$.


The matrix isn't necessarily positive definite, since it's the zero matrix if all $a_i$ are zero. However, it's positive semidefinite, and is positive definite if at least one $a_i$ is non-zero.

Let $a$ denote the vector $(a_1,\dotsc,a_n)$. The projection matrix $aa^\top$ has one eigenvalue $a^\top a=u^2$ and $n-1$ eigenvalues $0$. Your matrix is $(u+u^2)I-aa^\top$, so it has one eigenvalue $u$ and $n-1$ eigenvalues $u^2+u$, all of which are non-negative, and all of which are positive if and only if all $a_i$ are non-zero.

  • $\begingroup$ "..,so it has one eigenvalue $u$ and $n-1$ eigenvalues $u^2+u$". How did u calculate this? $\endgroup$ – Sunny88 Feb 25 '13 at 15:41
  • $\begingroup$ @Sunny88: Adding $\lambda I$ to $-aa^\top$ shifts all its eigenvalues by $\lambda$. The one eigenvalue $-u^2$ is shifted to $(u+u^2)-u^2=u$, and the $n-1$ eigenvalues $0$ are shifted to $(u+u^2)+0=u+u^2$. $\endgroup$ – joriki Feb 25 '13 at 15:52
  • $\begingroup$ I think that in your answer "if and only if all $a_i$ are non-zero" should be changed to "if and only if not all $a_i$ are zero". $\endgroup$ – Sunny88 Dec 4 '13 at 23:13

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.