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

I am interested in computing the following integral of a matrix exponential.

\begin{equation} \int_0^t e^{A(t-t')} e^{A^T (t-t')} dt' \end{equation}

The only assumption is that $A_{n\times n}$ is real. This is simple (albeit cumbersome) to compute, given a particular $A$. I was wondering, however, if there were any other steps I could take to progress the problem further a little bit further.

For instance (assuming I did not make any mistakes), if $A$ is normal then $A = U \Lambda U^H$, where $\Lambda=\text{diag}(\lambda1,\lambda2,\ldots,\lambda_n)$ contains the eigenvalues and $U$ is unitary. Then

\begin{equation} \int_0^t e^{A(t-t')} e^{A^T (t-t')} dt' = U \left[\int_0^t e^{2\Lambda(t-t')} dt'\right]U^H = U \left[(2\Lambda)^{-1} (e^{2\Lambda t} - I)\right]U^H \end{equation}

However, in general $A$ is not normal.

I appreciate the support in advance.

share|cite|improve this question
A trivial remark: we can do the substitution $s=t-t'$. The trace of the integral looks like a inner product. – Davide Giraudo Jul 19 '12 at 9:46

When the sum of any two complex eigenvalues of $A$ -- counting multiplicity -- is nonzero, some simplifications are possible.

(I) This is similar to solving the Lyapunov equation. Let $f(s)=e^{-sA}e^{-sA^T},\ C=\int_0^t f(s)ds$ and $B = \int_0^t e^{(t-s)A}e^{(t-s)A^T}ds = e^{tA} C e^{tA^T}$. Then \begin{align*} f'(s) &= - Af(s) - f(s)A^T,\\ f(t)-f(0) &= - AC - CA^T,\\ e^{-tA}e^{-tA^T} - I &= - AC - CA^T,\\ I - e^{tA}e^{tA^T} &= -AB-BA^T,\\ AB+BA^T &= e^{tA}e^{tA^T} - I,\\ (I\otimes A+A\otimes I)\operatorname{vec}(B)&=\operatorname{vec}(e^{tA}e^{tA^T} - I).\tag{1} \end{align*} Since no two eigenvalues of $A$ sum up to zero, $I\otimes A+A\otimes I$ is invertible. Therefore the unique solution of $(1)$ corresponds to our desired $B$.

(II) When $A$ is also $\mathbb{C}$-diagonalisable. Let $A=P\Lambda P^{-1}$ be a (not necessarily orthogonal) diagonalisation, where $\Lambda=\operatorname{diag}(\lambda_1,\ldots,\lambda_n)$ and $P$ may be nonreal. Then \begin{align*} f(s)&=Pe^{-s\Lambda}(P^{-1}P^{-T})e^{-s\Lambda}P^T\\ &=P\left((e^{-s(\lambda_i+\lambda_j)})\circ(P^{-1}P^{-T})\right)P^T\\ \end{align*} where "$\circ$" denotes a Hadamard product. Therefore $$ B=e^{tA}P\left(\left(\frac{1-e^{-t(\lambda_i+\lambda_j)}}{\lambda_i+\lambda_j}\right)\circ(P^{-1}P^{-T})\right)P^Te^{tA^T}. $$

share|cite|improve this answer

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.