10
$\begingroup$

Suppose we are given a function $f$ in terms of its power series \begin{align} f(x)= \sum_{n+0}^\infty a_n x^n. \end{align} We assume that we know all $a_n$'s and $f$ has infinite radius of convergense.

Can we say based on $a_n$ if the function is integrable or not? That is if \begin{align} \int_{\mathbb{R}} |f(x)| dx<\infty. \end{align}

For example, if $a_n \ge 0$ then we have that \begin{align} f(x) \ge a_0+a_1 x,\ x >0 , \end{align} and the function is not integrable.

The case that I am interested is when $a_n$'s have alternating sign. Specifically, can we determine if the following $f(x)$ is integrable \begin{align} f(x)= \sum_{k=0}^\infty \frac{\cos( \frac{\pi}{2}k)\ (k+1)^{\frac{k+1}{6}}}{k!} x^k. \end{align}

Here is the plot of $f(x)$ where the series was computed up to $N=600$. Blues is the $f(x)$ and red is $\frac{sin(\pi/2 x)}{\pi/2 x}$

where blue curve is $f(x)$ and red curve is $sinc(\pi/2 x)= \frac{\sin(\pi/2 x)}{\pi/2 x}$.

It seems that $f(x)$ has a tail that decays faster than $sinc(\pi/2 x)$.

Moreover, next we give a plot of $f(x) \cdot x$ and $f(x) \cdot x^2$ enter image description here

where blue curve is $f(x) \cdot x$ and red curve is $f(x) \cdot x^2$.

This seems to indicate that $f(x)$ decreases faster than $\frac{1}{x^2}$ which would imply that $f(x)$ is integrable.

Therefore, it would also be interesting to show that \begin{align} \lim_{x \to \infty} f(x)=0, \end{align} which at this moment I do not know how to do.

Thanks you.

**Edit: ** Please see a possible solution via Mellin transform.

$\endgroup$
  • 1
    $\begingroup$ If the power series converges in all of $\mathbb{R}$, the function is analytic, so it should automatically be integrable, unless I'm being really stupid, which I conceivably might be. $\endgroup$ – Daminark Dec 26 '16 at 4:26
  • 2
    $\begingroup$ If $a_n > 0$ for even $n$, $f(x) + f(-x) = 2 \sum_{n\ \text{even}} a_n x^n \ge a_2 x^2$ for $x > 0$, and $f(x)$ and $f(-x)$ can't both be integrable on $[0,\infty)$, and $f$ can't be integrable on $\mathbb R$. $\endgroup$ – Robert Israel Dec 26 '16 at 5:33
  • 2
    $\begingroup$ sinc is not integrable, although its improper integral converges. For integrability you'll want a decay better than the $1/|x|$ of sinc, maybe $1/|x|^p$ for some $p > 1$. The picture might suggest that this is the case, but of course it's not a proof. $\endgroup$ – Robert Israel Jan 3 '17 at 20:49
  • 1
    $\begingroup$ If that's true (which I'm not convinced of), then it is integrable. $\endgroup$ – Robert Israel Jan 3 '17 at 23:45
  • 1
    $\begingroup$ @boby Yeah sure, I don't have time at the moment, but I'll make it an answer and add some references and fill in a few details later. $\endgroup$ – user335907 Jan 4 '17 at 21:18
4
+50
$\begingroup$

To start consider mellin's inversion theorem.

Mellin's inversion theorem can be stated as, if

$$f(x) = \frac{1}{2\pi i}\int_{\sigma - i\infty}^{\sigma + i\infty}F(z)x^{-z}\,dz$$

Assuming everything is convergent, $0 < \sigma < 1$, $F$ holomorphic, then

$$F(z) = \int_0^\infty f(x)x^{z-1}\,dx$$

for $0 < \Re(z) < 1$

Ramanujan's master theorem gives us a really beautiful result based off this. If $z\Gamma(-z)g(z) \to 0$ as $z \to \infty$ when $|\arg(z)| < \pi/2$ and $g(z) = O(e^{\tau|\Im(z)|})$ for $\tau < \pi/2$, then

Ramanujan's Master theorem says

$$f(x) = \sum_{n=0}^\infty g(n)\frac{(-x)^n}{n!} = \frac{1}{2\pi i} \int_{\sigma - i \infty}^{\sigma+ i \infty} \Gamma(z)g(-z)x^{-z}\,dz$$

If you need more justification of this fact look at Ramanujan's master theorem, even wikipedia has something on this. There are a few ways of stating this, but the version I'm using is perfectly valid.

What is great is that

$$\int_0^\infty |f(x)| x^{-\sigma}\,dx < \infty$$

Which is another consequence of Mellin's inversion theorem.

Now your function $F(z) = \cos(\pi/2 z)(z+1)^{(z+1)/6}$ satisfies these bounds. It is rather trivial to show this, if need be I'll edit this in. By the above

$$f(x) = \sum_{n=0}^\infty \cos(\pi/2n)(n+1)^{(n+1)/6}\frac{(-x)^n}{n!}$$

satisfies for $0 < \Re(z) < 1$

$$\int_0^\infty f(x)x^{z-1}\,dx = \Gamma(z)F(z)$$

Therefore off the start, your function $f$ is in a weighted $L^{1}$ space (with weight $x^{-\sigma}$ (this is also part of Mellin's inversion theorem). All the odd powers disappear, meaning the function is symmetric about zero. Therefore

$$\int_{-\infty}^\infty |f(x)|x^{-\sigma}\,dx < \infty$$

for all $ 0 < \sigma < 1$. I'll leave it to you to show the subtleties required in making it $L^1$. If you can't, I'll show it .

$\endgroup$
  • 1
    $\begingroup$ I just wanted to break up your argument into logical steps. Let me know if the following is correct. The steps are: 1) Identify that $g(z)=\cos(\pi/2 z) (z+1)^{(z+1)/6}$ has the property that $z \Gamma(-z) g(z) \to 0$ as $z \to \infty$. 2) By Ramanujan's master theorem this implies that the function $f(z)= \sum_{n=0}^\infty g(n) \frac{(-x)^n}{n!}$ can be expressed as $f(z)= \frac{1}{2\pi i} \int_{\sigma-i \infty}^{\sigma+i \infty} \Gamma(z) g(-z) x^{-z} dz$. 3) This is the step I am confused about. How do we now know that $\int_{-\infty}^\infty |f(x)| x^{-\sigma} dx<\infty$??? $\endgroup$ – Boby Jan 5 '17 at 18:54
  • $\begingroup$ @Boby Am I being stupid or is this just because $f(z)$ is the Mellin transform of $\Gamma(z)g(-z)$, so $\int_0^\infty f(x)x^{z-1}\,dx = \Gamma(z)F(z)$? $\endgroup$ – Anon Jan 6 '17 at 2:10
  • $\begingroup$ @Anon I am not sure. I guess the claims is that if $f(z)$ is Mellin transform of $\Gamma(z)g(-z)$ then $f(z)$ satisfies $\int_{0}^\infty |f(z)| x^{-\sigma} dx<\infty$. But I don't know where is this comming from? $\endgroup$ – Boby Jan 6 '17 at 3:47
  • $\begingroup$ @Boby I think it's from Mellin's inversion theorem which he mentioned. I think that will give you his second-last formula (by applying the inversion theorem to his third formula which has the form of an inverse Mellin transform), which (with what he says at the beginning) I think gives the last formula. $\endgroup$ – Anon Jan 6 '17 at 4:18
  • $\begingroup$ Yeah, it's all because of Mellin's inversion theorem. $\int_0^\infty f(x)x^{z-1}\,dx = \Gamma(z)F(z)$, which isn't trvial to prove. If I showed the whole thing it would take like 5 pages. Just look up Mellin's inversion theorem, and Ramanujan's master theorem. Both of these results should clear up your confusion. $\endgroup$ – user335907 Jan 6 '17 at 18:32

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.