MathOverflow is a question and answer site for professional mathematicians. 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 $A(t)$ be a $n\times n$-matrix-valued continuous (plus possibly other niceness conditions; see below) curve, with the matrix entries being complex in general. If I am not mistaken, $A(t)$ generates a minimal Lie algebra $\mathfrak{g}$ of matrices, in the sense that $\mathfrak{g}$ is the intersection of all $n\times n$ matrix Lie algebras containing $A(t)$ for all $t$.

Now consider the equation $$\frac{d U(t,t_0)}{d t}=A(t) U(t,t_0)$$ with initial condition $$U(t_0,t_0)=\mathbf{1}_n$$ where $\mathbf{1}_n$ is the $n \times n$ unit matrix.

I am interested in the matrix logarithms of $U(t,t_0)$ for arbitrarily large $t$. For $t$ close to $t_0$ at least one of them lies in $\mathfrak{g}$ because the condition $ \int_{t_0}^t \|A(s)\| ds < \pi$ is satisfied -as long as $A(t)$ is nice- and thus the Magnus series for the logarithm converges. For large $t$ I have no reason to expect that the Magnus series will continue to converge; however $U(t,t_0)$ has at least one logarithm, because the matrix exponential is surjective when considering matrices with complex-valued entries. My question is this:

Where do the matrix logarithms of $U(t,t_0)$ lie? Is it possible that they all lie outside $\mathfrak{g}$?

I would also appreciate it if anyone knows any good references dealing with these equations. Thank you!

EDIT: Robert Bryant gave an example of a $U$ whose logarithm lies outside $\mathfrak{g}$. However one may extend $\mathfrak{g}$ by adding the unit matrix, so that that logarithm is now included. Would this $\mathfrak{g'}$ include at least one of the logarithms of $U(t,t_0)$ in general?

share|cite|improve this question
I don't understand your notation $U(t,t_0)$, and I'm sure you don't mean to have $U(t,t_0) = \mathbf{1}_n$, which doesn't make sense. Don't you just mean $U'(t) = A(t)U(t)$ with initial condition $U(t_0)=\mathbf{1}_n$? – Robert Bryant Apr 9 '12 at 1:06
Sorry, I was being sloppy. I meant $U(t_0,t_0)=\mathbf{1}_n$. Fixing... – AlexArvanitakis Apr 9 '12 at 1:14
Still, what is the point of having two arguments for $U$? – Robert Bryant Apr 9 '12 at 1:15
The point of using $U(t,t_0)$ is that the matrices $U(t_1,t_2)$ possess the semigroup property $U(t,t_0)=U(t,t_1)U(t_1,t_0)$. – AlexArvanitakis Apr 9 '12 at 1:19
But, if you take my $1$-variable solution, your $2$-variable solution is expressed as $U(t,t_0) = U(t)U(t_0)^{-1}$, so there's no need for a $2$-variable expression. – Robert Bryant Apr 9 '12 at 1:45
up vote 8 down vote accepted

It can't always be done. Take $n=2$, and start with a smooth curve $V:[0,1]\to SL(2,\mathbb{C})$ such that $V(0) = \mathbb{1}_2$ and $$ V(1) = \pmatrix{-1&1\\\\ 0&-1}. $$

Set $A(t) = V'(t)V(t)^{-1}$. Then $A:[0,1]\to{\frak{sl}}(2,\mathbb{C})$, and it satisfies $V'(t) = A(t)V(t)$ with $V(0) = \mathbb{1}_2$, but $V(1)$ is not the square of any element of $SL(2,\mathbb{C})$ and hence is not the exponential of any element of ${\frak{g}} = {\frak{sl}}(2,\mathbb{C})$.

Added after the question was edited: Giving yourself just the extra 'room' of adding the center of ${\frak{gl}}(n,\mathbb{C})$ to your algebra $\frak{g}$ won't do it. Here's a counterexample: Consider $SL(2,\mathbb{C})$ as a subgroup of $GL(3,\mathbb{C})$ in the obvious way, and let $W:[0,1]\to GL(3,\mathbb{C})$ be defined by $$ W(t) = \pmatrix{V(t)&0\\\\ 0&1}, $$ where $V$ is as above. Set $A(t) = W'(t)W(t)^{-1}$. Then $A:[0,1]\to {\frak{g}} = {\frak{sl}}(2,\mathbb{C})\subset {\frak{gl}}(3,\mathbb{C})$, and it satisfies $W'(t) = A(t)W(t)$ with $W(0) = \mathbb{1}_3$. But it is easy to see that $W(1)$ is not the exponential of any element of ${\frak{g}'} = \mathbb{C}\cdot\mathbb{1}_3\oplus {\frak{sl}}(2,\mathbb{C}) = \mathbb{C}\cdot\mathbb{1}_3\oplus {\frak{g}}$.

share|cite|improve this answer
That is correct, of course. (One of) the logarithms of V(1) as given above is $i \pi\mathbf{1}_2$−[0 1;0 0] (That last one is the $2\times 2$ upper triangular nilpotent matrix with -1 on the superdiagonal). If I were to enlarge $\mathfrak{g}$ to $\mathfrak{g'}$ by including the unit matrix, would I then get a logarithm in $\mathfrak{g'}$? (I will edit the question to include this tomorrow) – AlexArvanitakis Apr 9 '12 at 2:20
That's what I was looking for. Thank you very much for your time! – AlexArvanitakis Apr 9 '12 at 22:16

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.