Consider a vector of length $m$($m\geq 3$): $a=[a_1,...,a_m]^T$, where $a_i\neq a_j, \forall 1\leq i<j\leq m$. And consider the matrix $A\in R^{m\times m}$, whose entries are $[A]_{ij}=(a_i−a_j)^2$. My simulation suggests that $A$ seems to be positive semi-definite, and I wonder if there is a proof or disproof.

BTW, one verifiable property of $A$ is that its rank is 3. Moreover, for the positive eigenvalues of $A$: $\lambda_1\geq\lambda_2\geq\lambda_3$,we have $\lambda_1=\lambda_2+\lambda_3$, which also requires a proof.


We have \begin{align} x^TAx & = \sum_{i,j=1}^n x_ix_j(a_i-a_j)^2 = \sum_{i=1}^nx_ia_i^2 \sum_{j=1}^nx_j + \sum_{j=1}^nx_ja_j^2 \sum_{i=1}^nx_i -2\left(\sum_{i=1}^na_ix_i \right)^2\\ & = 2\left(\sum_{i=1}^nx_ia_i^2 \cdot \sum_{j=1}^nx_j - \left(\sum_{i=1}^na_ix_i\right)^2\right) \end{align} Now using Cauchy–Schwarz inequality, for the vectors $\vec{p} = \begin{bmatrix} \sqrt{x_1}a_1 & \sqrt{x_2}a_2 & \cdots & \sqrt{x_n}a_n\end{bmatrix}$ and $\vec{q} = \begin{bmatrix} \overline{\sqrt{x_1}} & \overline{\sqrt{x_2}} & \cdots & \overline{\sqrt{x_n}}\end{bmatrix}$ we obtain that $x^TAx \geq 0$.

  • $\begingroup$ Thank you for your answer! I didn't realize to use Cauchy-Schwarz inequality for complex vectors when $x_i<0$ before your answer. And happy holiday! $\endgroup$ – SoftSail Dec 24 '15 at 10:11
  • $\begingroup$ And how about the eigenvalue result, i.e. $\lambda_1 = \lambda_2+\lambda_3$? Thanks! $\endgroup$ – SoftSail Dec 24 '15 at 10:15
  • $\begingroup$ I realize a problem that $<p,p>$ is not $\sum_{i=1}^n x_i a_i^2$ but $\sum_{i=1}^n| x_i a_i^2|$ instead. $\endgroup$ – SoftSail Dec 27 '15 at 10:05

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.