Let $$ M = \begin{pmatrix}A& B\\C& D\end{pmatrix}. $$ Is there a general formula for $e^M$ in terms of the sub-matrices $A$, $B$, $C$ and $D$? If not, can anything useful be said about the properties of $e^M$, given properties of its component blocks?

I've tried a few different approaches but haven't got anywhere, mostly because the formula $e^{A + B}=e^Ae^B$ only works if $A$ and $B$ commute.

If it helps, I'm most interested in the case where $M$ is a symmetryic real matrix, and an 'infinitesimal stochastic' matrix in particular. (An infinitesimal stochastic matrix is a symmetric real matrix whose column sums are all zero, and which has no negative off-diagonal elements.) If there are other special sub-classes of matrices for which something useful can be said here, I would be interested to know that.

  • $\begingroup$ the case where $M$ is a real symmetric matrix, or where the sub-matrices are? $\endgroup$ – Thoth Oct 23 '15 at 5:33
  • $\begingroup$ Since any matrix $M$ can be represented in block form as indicated, it is unclear why any special expression for the exponential of $M$ ought to be expected. $\endgroup$ – hardmath Oct 23 '15 at 5:39
  • $\begingroup$ @Thoth the case where $M$ is - I've clarified in the question $\endgroup$ – Nathaniel Oct 23 '15 at 7:01
  • $\begingroup$ @hardmath that's a good point. (I was coming from the point of view of "it would be really useful to me if there was a formula for this," not "I have some reason to expect there would be a formula for this.") I've edited the question to ask whether there's anything else that can be said about $e^M$, given properties of its component blocks. $\endgroup$ – Nathaniel Oct 23 '15 at 7:02
  • 1
    $\begingroup$ I doubt you can say much about $A,B,C,D$, but if $M$ is symmetric then it has an eigen-decomp as $M=P^TDP$ and so I think you have $e^M=P^Te^DP$. I don't know if that helps any. $D$ contains the (possibly absolute value or squared can't remember) eigenvalues of $M$, but unless the top right or bottom left block is the zero block, there is no relationship known between the eigenvalues of the blocks and the eigenvalues of $M$. $\endgroup$ – Thoth Oct 23 '15 at 8:35

If $M$ is a squared symmetric real matrix then \begin{equation} M=\begin{pmatrix} A & B \\ B^\top & C\end{pmatrix}. \end{equation} If $B=0$, then \begin{equation} \exp(M)=\begin{pmatrix} \exp(A) & 0 \\ 0 & \exp(C)\end{pmatrix}. \end{equation}

Moreover, since $M$ is symmetric positive semidefinite then it is diagonalizable, i.e. $M=V D V^\top$ with $D$ diagonal matrix and $V$ orthogonal matrix where each column is an eigenvector of $M$. If you work a bit with the relation \begin{equation}\begin{pmatrix} D_1 & 0 \\ 0 & D_2\end{pmatrix}= D=V^\top M V = \begin{pmatrix} V_1^\top & V_3^\top \\ V_2^\top & V_4^\top\end{pmatrix}\begin{pmatrix} A & B \\ B^\top & C\end{pmatrix}\begin{pmatrix} V_1 & V_2 \\ V_3 & V_4\end{pmatrix}.\end{equation} you might find the block diagonal matrices $D_1$ and $D_2$ and block matrices $V_1$, $V_2$, $V_3$ and $V_4$. Hence you can apply $\exp(M)=V\exp(D)V^\top$.

| cite | improve this answer | |

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.