Let $A,B,C$ complex $n\times n$ Matrices with ones on the diagonal and entries of absolute value 1. Further, let $$A=(a_{ij})_{i,j=1}^n,\; B=(b_{ij})_{i,j=1}^n \mbox{ and } C=(a_{ij}b_{ij})_{i,j=1}^n.$$

Can we say something about the positive semi-definiteness of $A$ if we know that $B$ and $C$ have this property?

Any hint to a similar problem or result or counter-example would be greatly appreciated.

  • 1
    $\begingroup$ Are $A$, $B$, and $C$ related somehow? Why should positive semi-definiteness of $A$ depend at all on $B$ and $C$? $\endgroup$ – User8128 Jun 28 '17 at 17:43
  • $\begingroup$ @User8128 Thank you. I forgot something ... edited the question $\endgroup$ – Vincent.W. Jun 28 '17 at 17:46
  • $\begingroup$ Got it. Thanks makes more sense! $\endgroup$ – User8128 Jun 28 '17 at 17:46
  • $\begingroup$ Not sure about the star notation. Is $C$ equal to the Hadamard product of $A$ and $B$? $\endgroup$ – Paul Aljabar Jun 28 '17 at 17:51
  • $\begingroup$ @PaulAljabar Yes.. its just a componentwise product. $\endgroup$ – Vincent.W. Jun 28 '17 at 17:53

If by complex matrices you mean Hermitian matrices the answer is yes, for a silly reason and a non-silly reason. Suppose $B$ and $C$ are PSD with $1$s on the diagonal and all off diagonal entries of absolute value $1$. You ask if the elementwise quotient $A=(c_{i,j}/b_{i,j})$ is PSD. The silly reason is that the complex conjugate of a complex number of absolute value $1$ is its reciprocal. So the matrix $D=(\bar{b}_{i,j})$ is the matrix of reciprocals of the entries in $B$, and $A$ is the element-wise product of the two PSD matrices $A$ and $D$. Which it is also PSD, by the non-silly Schur product theorem: https://en.wikipedia.org/wiki/Schur_product_theorem.

  • $\begingroup$ Thank you! However, by complex matrix I mean a matrix with complex entries-no further regularity assumed. $\endgroup$ – Vincent.W. Jun 29 '17 at 6:54
  • $\begingroup$ I mean, what is a complex positive definite matrix for you? Do you require $\bar z' M z\ge 0$ for all complex vectors $z$, with $\bar z'$ denoting the conjugate transpose? If so, the argument I supplied answers your question in the affirmative. $\endgroup$ – kimchi lover Jun 29 '17 at 15:02
  • $\begingroup$ Yes, its just that you wrote that the answer is yes for hermitian matrices. But,if $B$ is psd and not hermitian then also its transpose conjugate $D$ will be psd. However then you dont have $B=D$ and hence $C\circ D\neq A$ ($\circ$ denotes hadamard) since $c_{i,j}*b^{-1}_{j,i}\neq a_{i,j}$. $\endgroup$ – Vincent.W. Jun 29 '17 at 15:44
  • $\begingroup$ Can you tell me precisely what you mean by PSD, and supply an example of a non-Hermitian PSD matrix? The [Wikipedia article]( en.wikipedia.org/wiki/Positive-definite_matrix) does not help me understand more clearly what you are thinking about. $\endgroup$ – kimchi lover Jun 29 '17 at 17:58
  • $\begingroup$ You are right. Psd implies hermitian. $\endgroup$ – Vincent.W. Jun 29 '17 at 18:41

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.