How do I show that formal logarithm is the inverse of the formal exponential?

1.6k Views Asked by At

Let $A$ be a unital commutative and associative $\mathbb{Q}$-algebra.

Define $exp(f):=\sum_{n=0}^\infty \frac{f^n}{n!}$ for each $f\in XA[[X]]$.

Define $log(f):=\sum_{n=1}^\infty \frac{(-1)^{n+1}}{n} (f-1)^n$ for each $f\in 1+XA[[X]]$.

Hence, we have two maps $exp:XA[[X]]\rightarrow 1+XA[[X]]$ and $log:1+XA[[X]]\rightarrow XA[[X]]$.

I am trying to prove that $log$ map is the inverse of the $exp$ map.

The first thing I tried is to directly show $log\circ exp=id$ and $exp\circ log = id$ by checking if the identities hold for every element $f$, but this does not work well since this way involves too many calculations. For example, $$[X^n]exp(log(f))=[X^n]\sum_{k=0}^n log(f)^k/k! = \sum_{k=0}^n \frac{1}{k!} [X^n]log(f)^k= \sum_{k=0}^n \frac{1}{k!} [X^n](\sum_{l=1}^k \frac{(-1)^{l+1}}{l} (f-1)^l)^k$$.

Thus, we have to show that $$\sum_{k=0}^n \frac{1}{k!} [X^n](\sum_{l=1}^k \frac{(-1)^{l+1}}{l} (f-1)^l)^k=[X^n]f$$.

But this calculation is really a nightmare.. Is there a clever way to show this? If not, how do I wisely calculate to show the above identity?

Thank you in advance.

2

There are 2 best solutions below

0
On

Here is one possible argument. One way to define the exponential $\exp(x)$ as a formal power series is that it is the unique formal power series $f(x)$ (over any commutative $\mathbb{Q}$-algebra) satisfying $f(0) = 1$ and

$$f'(x) = f(x).$$

Repeatedly differentiating this identity easily gives $[x^n] \exp(x) = \frac{1}{n!}$ as expected. Similarly, one way to define the logarithm $\log (1 + x)$ as a formal power series is that it is the unique formal power series $g(x)$ satisfying $g(0) = 0$ and

$$g'(x) = \frac{1}{1 + x} = \sum_{n=0}^{\infty} (-1)^n x^n.$$

So what can we say about the composite $\exp \log (1 + x)$? Well, by the formal chain rule, we have

$$\frac{d}{dx} f(g(x)) = f'(g(x)) g'(x) = f(g(x)) \frac{1}{1 + x}.$$

So $f(g(x))$ is a solution $h(x)$ to the differential equation $h'(x) = h(x) \frac{1}{1 + x}$ with initial condition $h(0) = f(g(0)) = 1$. We clearly have $h(x) = 1 + x$ is a solution, and we can appeal to a formal version of the Picard-Lindelof theorem to assert that formal solutions to ODEs exist and are unique, so we conclude that

$$\exp \log (1 + x) = 1 + x.$$

Similarly, what can we say about $\log \exp x = \log ((\exp x - 1) + 1)$? Well, again by the formal chain rule, we have

$$\frac{d}{dx} g(f(x) - 1) = g'(f(x) - 1) f'(x) = \frac{f'(x)}{f(x)} = 1.$$

So $g(f(x) - 1)$ is a solution to the ODE $\frac{d}{dx} h(x) = 1$ with initial condition $h(0) = g(f(0) - 1) = 0$. Here it's a bit simpler to see that we must have $h(x) = x$, so

$$\log \exp(x) = x.$$

0
On

Another way might be (feeling too lazy right now to work out all the details) to use the functional equations:

For ln, $f(xy) = f(x)+f(y), f'(1) = 1$.

For exp, $g(x+y) = g(x)g(y), g'(0) = 1$.

From these, can get $f'(x) =\dfrac1{x} $ and $g'(x) = g(x)$.