1
$\begingroup$

Let $A\in\mathbb{R}^{n\times n}$ be a non-negative matrix and let $D\in\mathbb{R}^{n\times n}$ be a signature matrix, i.e. a diagonal matrix having either $+ 1$ or $-1$ elements on its diagonal. Notice that it holds $D=D^{-1}$. Consider the matrix $$ B:=AD. $$ Observe that $B$ is equal to $A$ except for the fact that some of its columns may have opposite sign.

The problem. I'm interested in some estimates of the 2-norm of the matrix exponential of $B$. In particular, I'm wondering whether it's possible to derive upper bounds on $\|\exp (B)\|_2$ of the form $$ \|\exp (B)\|_2\leq f(A,D)\|\exp(A)\|_2, $$ for some function $f$ of $A$ and/or $D$. Any suggestions?

Note 1. Using the subadditivity and submultiplicativity of matrix 2-norm, we can obtain the bound $$ \|\exp (B)\|_2\leq \exp(\|A\|_2). $$ However, this bound is quite loose and not very informative.

Note 2. It can be assumed that $A$ is symmetric, if that helps in deriving a bound.

$\endgroup$
  • $\begingroup$ I don't think you'll be able to do any better than note 1. Since $A$ is symmetric, you can express that 2-norm in terms of the eigenvalues of $A$. $\endgroup$ – Omnomnomnom Feb 13 '17 at 15:35
  • $\begingroup$ Is your 2-norm the operator norm (the "spectral norm") or the Frobenius norm? $\endgroup$ – Omnomnomnom Feb 13 '17 at 15:36
  • $\begingroup$ @Omnomnomnom: The operator norm. However, if it's possible to derive a bound for the Frobenius norm, I'm interested in it as well. $\endgroup$ – Ludwig Feb 13 '17 at 15:40
  • $\begingroup$ I think if you want any kind of improvement to your inequality, you'll have to say something about the eigenvectors of $A$. $\endgroup$ – Omnomnomnom Feb 13 '17 at 15:43
  • $\begingroup$ Alternatively, if we partition the matrix $B$ in the form $$ \begin{bmatrix}A_{11} &-A_{12} \\ A_{21} &-A_{22} \end{bmatrix},$$ is it possible to derive a bound on $\|\exp(B)\|_2$ using the norm of the exp of the diagonal blocks? $\endgroup$ – Ludwig Feb 13 '17 at 15:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.