1
$\begingroup$

Consider the power series

$F(z) = \sum\limits_{n=1}^{\infty} a_n z^n$

my question is how should the $a_n$ decay for the radius of convergence of the series to be greater than 1.

  • 5
    You want to have $\limsup_{n \to \infty} \sqrt[n]{|a_n|} \lt 1$. See [Wikipedia](http://en.wikipedia.org/wiki/Radius_of_convergence#Finding_the_radius_of_convergence). Could you be more specific on what sort of decay property you are looking for?2011-06-10
  • 0
    @Theo Buehler: does faster than $\frac{1}{n^p}$ for any $p \in \mathbb{N}$ sufficient ?2011-06-10
  • 0
    @Rajesh: but is then for p=1 the series not divergent ( harmonic series)?2011-06-10
  • 0
    @Gottfried Helms : I mean faster than the case of any $p \in \mathbb{N}$, i do not mean $p = 1$, i mean for any arbitrarily large $p$.2011-06-10
  • 0
    Are you looking at the Fourier coefficients of a periodic real analytic function? Or what exactly do you mean by "faster than $\frac{1}{n^p}$"?2011-06-10
  • 0
    @Theo Buehler: yes2011-06-10
  • 0
    @Theo Buehler: Does the condition stated by you is met by the condition stated by me (I apologize if this question is trivial) ?2011-06-10
  • 1
    Why the -1 vote on this question?2011-06-10
  • 0
    @Rajesh: The answer to your last question to @Theo is No. Consider $a_n=\exp(-(\log n)^2)$, then $a_n\ll n^{-p}$ for every $p$ since $(\log n)^2\gg p\log n$, but $\limsup\sqrt[n]{a_n}=1$ since $(\log n)^2\ll n$ hence your condition implies only a radius at least $1$. // Relatedly, I fail to see how you can consider Eric's post as answering your question, as opposed to Theo's very first comment above. Oh well.2011-06-14
  • 0
    @Didier Piau : I only took Eric's answer as an hint. At the time I asked the question to Theo as a comment Eric's answer wasn't yet posted. Thank you for the comment.2011-06-14

2 Answers 2

2

As Theo said it:

The radius of convergence is greater than $1$ if and only if $\limsup\sqrt[n]{|a_n|}<1$.

In turn, this condition is met if and only if there exists a finite $c$ and a real number $r<1$ such that $|a_n|\le cr^n$ for every $n$.

The condition that $|a_n|\le cn^{-p}$ is not sufficient, for any positive $p$, it only ensures that the radius of convergence is at least $1$ but not that it is greater than $1$. Even the condition that $n^p|a_n|\to0$ for every given $p$ is not enough. Consider for example the sequence $a_n=\exp(-(\log n)^2)$, then $\sqrt[n]{|a_n|}\to1$ because $(\log n)^2\ll n$ hence the radius of convergence of the associated series is $1$ although $n^p|a_n|\to0$ for every given $p$ because $p\log n\ll(\log n)^2$.

In fact, powers of $n$ are on a too fine grained scale for radiuses of convergence: to wit, for every sequence $(a_n)$, the series $\displaystyle\sum a_nz^n$ and $\displaystyle\sum a_nn^pz^n$ have the same radius of convergence, for every real number $p$.

Likewise, for every finite positive $R$, the radius of convergence is greater than $R$ if and only if there exists a finite $c$ and a real number $r<1/R$ such that $|a_n|\le cr^n$ for every $n$.

2

Hint:

What happens if $a_n=\frac{1}{2^n}$? Consider the series $$\sum_{n=1}^\infty \frac{z^n}{2^n}.$$ Does it converge if $z=1$? What about $z=1.5$? And $z=2$?

Also, here is another series: Let $a_n=\frac{1}{n!}$, so we are looking at $\sum_{n=1}^\infty \frac{z^n}{n!}$. It turns out that this series will converge for every $z\in\mathbb{C}$. Try to use the ratio test to prove why.

Hope that helps,