Does there exist a definition for matrix exponentiation? If we have, say, an integer, one can define $A^B$ as follows:

$$\prod_{n = 1}^B A$$

We can define exponentials of fractions as a power of a radical, and we even have the following definition of the exponential:

$$e^z = \sum_{n = 0}^\infty \frac{z^n}{n!}$$

which comes from a taylor series for the function $\exp(z)$. Now, a problem seems to arise when we attempt to calculate $\exp(A)$, where $A$ is an $n$ x $n$ (square) matrix. We cannot define it as multiplication a "matrix number of times" as this makes no sense. The only reasonable definition that could work is the latter definition (the infinite series):

$$e^A = 1 + A + \frac{AA}{2!} + \frac{AAA}{3!} + \frac{AAAA}{4!} + \cdots$$

where we can define matrix exponentiation to the power of an integer, which is all that is required here. We know that $e^x$ will converge absolutely for all complex numbers, but do we know that this is true for matrices? Can this "matrix sum" diverge, and are there ways to test divergence/convergence when a matrix is applied? Or is this concept of "matrix divergence" not well defined?


  • $\begingroup$ This exponentiation appears in the Lie theory. If $ A $ is a lie algebra, then the exponentiation provides the local structure of the corresponding Lie group. In the case of Lie theory, at least, we can say that this exponentiation converges. $\endgroup$ – user122283 Jan 20 '14 at 15:45
  • $\begingroup$ Isn't a matrix a higher dimensional complex number? If so, then couldn't we apply the same divergence or convergence rules to this exponentiation also? $\endgroup$ – user122283 Jan 20 '14 at 15:46
  • $\begingroup$ I'd suggest editing the title, as you're only really asking about $e^A$, rather than $x^A$ for a given $x\in\Bbb R$. $\endgroup$ – Jonathan Y. Jan 20 '14 at 15:59

Yes, the exponential of a matrix can be defined by that Taylor series, and it is a very useful thing. See e.g. Wikipedia The proof that it converges is not difficult, using any sub-multiplicative matrix norm. If $\| \cdot \|$ is such a norm, then any power series $\sum_n c_n A^n$ converges whenever the real series $\sum_n |c_n| \|A\|^n$ converges.


My favourite definition - as I teach quite often ODEs - is using differential equations.

Let us first note that $x(t)=\mathrm{e}^{at}$ is the unique solution of the initial value problem: $$ x'=ax, \quad x(0)=1. $$ Generalise that a little bit: Let $A$ be a square matrix ($A\in\mathbb R^{n\times n}$). Define as $\mathrm{e}^{tA}$ the unique, and globally defined, solution of $$ X'=AX, \quad X(0)=I, $$ where $X$ is an $n\times n$ matrix, and $I$ the identity matrix. Through this definition a lot of properties of the exponential of a matrix, (which in order to prove require messy calculations) are proved very elegantly if we use the above definition.


For diagonalizable $A$, you can write $A = P \Lambda P^{-1}$, where the column vectors of $P$ are eigenvectors and $\Lambda$ is diagonal matrix with eigenvalue entries.

Then $$e^{A} = e^{P \Lambda P^{-1}} = PIP^{-1} + P \Lambda P^{-1} + \frac{1}{2!}(P \Lambda P^{-1})^2 + \ldots = Pe^{\Lambda}P^{-1}$$

And raising $e$ to a diagonal matrix is easy—just replace all the diagonal entries with those raised by $e$.


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.