4
$\begingroup$

Let $f(x)=\sum_{n\geq 1} c_n\cdot x^n$ be a function given by a power series. Further there is some $\alpha >1$ such that for all $n$, $c_n = \Theta(1/n^{\alpha})$. What can one say about the asymptotics (in terms of n) of the coefficients of the Taylor expansion (around $0$) of the inverse of f assuming it exists? For instance, is the $n$-th coefficient less than $1/\beta^n$ for some $\beta>0$?

I couldn't get much mileage out of trying to analyze the Lagrange-Burrman formula or Bell polynomials.

$\endgroup$
2
  • 2
    $\begingroup$ To talk about inverse function at $0$ you need first of all that $f(0)=0$, that is $c_0=0$. Then the rate of coefficients of the inverse function is regulated by the radius of the largest disk centered at $0$ in which this inverse function exists. And this has nothing to do with the rate of $c_n$. Even when $c_n$ is a finite sequence, the rate of the coefficients of the inverse can be anything. $\endgroup$ – Alexandre Eremenko Nov 1 '17 at 13:24
  • $\begingroup$ Great - thanks for the information. I had a more specific function in mind, and the radius of the disk you mentioned gave me the bound I desire. $\endgroup$ – user116726 Nov 1 '17 at 18:18
4
$\begingroup$

You can use the Faa-di-Bruno formula, like it is done on page 7 of here or the proof of claim (b) on page 21 of here.

$\endgroup$
1
  • $\begingroup$ Thanks - I think I'm able to understand the gist of what's going on. The reference is a little notation heavy given my background, so I just wanted to confirm the following: What is the dependence of beta on alpha? (any bound at all is ok, I don't need to know the optimal dependence). $\endgroup$ – user116726 Nov 1 '17 at 10:07

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.