Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $A$ be a positive semidefinite matrix of rank $1$. Let $B$ be a general Hermitian matrix.

Under what conditions on $B$ (probably in terms of $A$) is $A-B$ positive semidefinite?

I was thinking that it may be along the lines of the generalized eigenvalue problem but can't quite see how.

share|cite|improve this question
up vote 0 down vote accepted

For your $A-B$ to be positive semidefinite,

  • one sufficient condition is $B$ being negative semidefinite;
  • one necessary condition is $\lambda_1^\downarrow(B)\le\lambda_1^\downarrow(A)$ and $\lambda_2^\downarrow(B),\,\lambda_3^\downarrow(B),\ldots,\, \lambda_n^\downarrow(B)\le0$, where $\lambda_k^\downarrow(\cdot)$ means the $k$-th largest eigenvalue; this is a consequence of Weyl's inequality;
  • if you are looking for a necessary and sufficient condition in terms of the eigenvalues of $A$ and $B$, I'm afraid there isn't one.
share|cite|improve this answer
Thanks user1551. As a follow up, is it possible to prove that a necessary and sufficient condition in terms of the eigenvalues doesn't exist? Are there any further constraints on the form of $A$ and $B$ that would engineer such a condition? – Stan Dec 11 '12 at 10:06
@Stan This is no such necessary and sufficient condition because the eigenvalues of $A$ and $B$ do not characterize the eigenvalues of $A-B$. For instance, Let $A=\mathrm{diag}(1,0)$, $B=\mathrm{diag}(1/2,0)$ and $B'=\mathrm{diag}(0,1/2)$. Then $B$ and $B'$ have identical spectrum, both the pairs $A,B$ and $A',B'$ satisfy the necessary condition as specified in my answer, yet $A-B=0$ is positive semidefinite but $A-B'=\mathrm{diag}(1,-1)$ is not. – user1551 Dec 12 '12 at 1:00
This example shows that you not only need to look at the eigenvalues, but also need to look at how the eigenvectors of $A$ and $B$ align. But then if you want to find a necessary and sufficient condition in terms of both eigenvalues and eigenvectors, things would become more complicated and it may be simpler to directly determine whether $A-B$ itself is positive semidefinite. – user1551 Dec 12 '12 at 1:04
Ok, that makes sense. In the overall problem I'm considering $B$ is an $nxn$ matrix and I have $n$ such $A$ ($A_i$). I am trying to find the most general $B$ such that $A_i - B$ is positive semidefinite for all $i$. As they are rank $1$ projectors I can write $A_i=v_i v_i^T$ so I guess I'd want to be able to write $B$ in terms of these $v_i$ and their inner products such that it satisfies the constraints. Does it sound plausible to find a necessary and sufficient condition on the form of $B$ along those lines? – Stan Dec 12 '12 at 9:30
If you only know that each $A_i$ has rank 1, but don't know the exact $v_i$, I think not much can be said about $B$. If you do know what each $v_i$ looks like, then it may be possible to characterize $B$, but there is also a possibility that $B$ does not exist. – user1551 Dec 12 '12 at 10:00

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.