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 $C$ be a complex $n \times n$ matrix with $\det C \neq 0$. Prove that there is a complex matrix $B$ such that $C = e^B$
Hint: use the Jordan form matrices for comlexas

share|cite|improve this question
up vote 4 down vote accepted

Suppose $A$ is the Jordan block of size $n$ for some non-zero eigenvalue $\lambda$. That is, $$ A = \pmatrix{ \lambda&1&&\\ &\lambda&1&\\ &&\ddots\\ &&&\lambda } $$ We compute $e^A = e^\lambda K$, where $K$ is the matrix given by $$ K=\pmatrix{ 1&1&\cdots&1\\ &1&\cdots&1\\ &&\ddots&\vdots\\ &&&1 } $$ Note that $K$ is similar to the Jordan block of size $n$ corresponding to $1$.

Let $M$ be the matrix given by $$ M = \pmatrix{ 1&e^{-\lambda}&\\ &1&e^{-\lambda}&\\ &&\ddots&\\ &&&1 } $$ Note that $M$ is also similar to the Jordan block of size $n$ corresponding to $1$. Thus, $M$ and $K$ are similar. Take $S$ such that $K = SMS^{-1}$. We have $$ e^A = e^\lambda K = e^{\lambda}SMS^{-1} = S(e^{\lambda}M)S^{-1}\\ S^{-1}e^A S = e^{S^{-1}AS} = e^\lambda M $$ That is, we have $$ e^{S^{-1}AS} = e^{\lambda}M = \pmatrix{ e^{\lambda}&1&\\ &e^{\lambda}&1&\\ &&\ddots&\\ &&&e^{\lambda} } $$ So, suppose you begin with a matrix of the form $$ C = \pmatrix{ e^{\lambda}&1&\\ &e^{\lambda}&1&\\ &&\ddots&\\ &&&e^{\lambda} } $$ We may take $B = S^{-1}AS$, with $S$ and $A$ as defined above.

Note that any matrix with a non-zero determinant has a Jordan form consisting of blocks of the above form.

share|cite|improve this answer
I do not understand why the K is that form – João Apr 12 '14 at 20:19
Compute $e^A$ and it should be clear – Omnomnomnom Apr 12 '14 at 20:23
Or do you mean why $K$ has that Jordan canonical form? Note that $K$ has only $1$ eigenvector. – Omnomnomnom Apr 12 '14 at 20:24
got it, thank you – João Apr 12 '14 at 20:29
I do not understand why you choose C only at the end, the C would not have to be a generic matrix? – Croos Apr 12 '14 at 20:57

The fact that $\det C \ne 0$ implies that all the eigenvalues of $C$ are nonzero; thus when $C$ is transformed to Jordan form via a similarity transformation by the appropriate nonsingular matrix $P$, $C \to P^{-1}CP = D$, we see that $D = \text{diag}(D_1, D_2, \ldots, D_l)$ is block diagonal consisting of Jordan blocks of the form $D_k = \lambda I_m + N_m$, where $\lambda \ne 0$ is an eigenvalue of $C$ and $N$ may be written

$N_m = [n_{ij}] \; \text{with} \; n_{i, i + 1} = 1; \; n_{ij} = 0, j \ne i + 1, 1 \le i, j \le m, \tag{1}$

and $I_m$ is the $m \times m$ identity matrix. For each such block $D_k$ we may find a matrix $E_k$ such that $e^{E_k} = D_k$ in the following manner: set

$F_k = -\sum_1^{m - 1} \dfrac{(-N_m)^p}{p \lambda^p}; \tag{2}$

$F_k$ is derived by considering the power series for $\ln(1 + s)$ for small $s$ and substituting $N_m / \lambda$ for $s$; since $N_m^p = 0$ for $p \ge m$ the resulting series becomes the mere polynomial in $N_m$ given by (2); thus it is well-defined. We put

$E_k = (\ln \lambda) I_m + F_k, \tag{3}$

and we find, since $I_m$ and $F_k$ commute, that is $I_m F_k = F_k I_m$, that

$e^{E_k} = e^{\ln \lambda} e^{F_k} = \lambda e^{F_k}; \tag{4}$


$e^{F_k} = I_m + \dfrac{1}{\lambda}N_m, \tag {5}$

which may be readily seen by noting that $e^{\ln (1 + s)} = 1 + s$, since the algebraic maneuvers required in evaluating the coefficients of the power series of $e^{\ln (1 + s)}$ are the same for $s$ and $N_m / \lambda$, given that $N_m^m = 0$. Thus

$e^{E_k} = \lambda I_m + N_m = D_k, \tag{6}$

which holds for any Jordan block of $D$; taking

$E = \text{diag}(E_1, E_2, \ldots, E_l) \tag{7}$

thus yields a matrix such that

$D = e^E; \tag{8}$


$C = PDP^{-1} = Pe^EP^{-1} = e^{PEP^{-1}}; \tag{9}$


Hope this helps. Cheerio,

and as always,

Fiat Lux!!!

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.