1
$\begingroup$

Let $(X,\mu)$ be a measure space with $\mu(X)=1$. Let $f$ be a measurable function such that $0<||f||_{\infty}< \infty$. Let $a_n=\int|f|^nd\mu$

Show that:

(1) $\lim_{n\to\infty}a_{n+1}/a_n=||f||_\infty$

(2) $\lim_{n\to\infty}a_n^{1/n}=||f||_\infty$

I have no idea how to proceed.

$\endgroup$
  • 2
    $\begingroup$ For the second one : you can see here. $\endgroup$ – Watson Feb 15 '16 at 15:34
  • $\begingroup$ For the first one, I am not even sure the question is defined.What if $a_n=0$, say for instance for $f=\mathbb{1}_{\mathbb{Q}}$? $\endgroup$ – Clement C. Feb 15 '16 at 15:48
  • $\begingroup$ @ClementC. $\| \mathbf{1}_\mathbb{Q} \|_\infty = 0$, assuming that $\| \cdot \|_\infty$ is the essential supremum. $\endgroup$ – Nigel Overmars Feb 15 '16 at 15:53
  • $\begingroup$ @NigelOvermars granted, but this "assuming that" is not really part of the question. $\endgroup$ – Clement C. Feb 15 '16 at 15:54
  • $\begingroup$ @Clement: In a measure theory context, $\|f\|_\infty $ is nearly always the essential supremum of $|f|$. $\endgroup$ – PhoemueX Feb 15 '16 at 15:59
1
$\begingroup$

By renormalizing, we can assume $\left\Vert f\right\Vert _{\infty}=1$. Set $a_{n}:=\int\left|f\right|^{n}\,{\rm d}\mu$ and $b_{n}:=\frac{a_{n+1}}{a_{n}}$ for $n\in\mathbb{N}$. Then $\left\Vert f\right\Vert _{\infty}$ implies $a_{n+1}\leq a_{n}$ and thus $b_{n}\leq1$ for all $n\in\mathbb{N}$.

Now, the Cauchy-Schwarz inequality implies that $$ a_{n+1}=\int\left|f\right|^{n+1}\,{\rm d}\mu=\int\left|f\right|^{\frac{n}{2}}\cdot\left|f\right|^{\frac{n+2}{2}}\,{\rm d}\mu\leq\sqrt{\int\left|f\right|^{n}\,{\rm d}\mu}\cdot\sqrt{\int\left|f\right|^{n+2}\,{\rm d}\mu}, $$ i.e. $a_{n+1}^{2}\leq a_{n}\cdot a_{n+2}$. For the quotient $b_{n}=\frac{a_{n+1}}{a_{n}}$, this implies $$ b_{n}=\frac{a_{n+1}}{a_{n}}\leq\frac{a_{n+2}}{a_{n+1}}=b_{n+1}, $$ so that $\left(b_{n}\right)_{n\in\mathbb{N}}$ is nondecreasing and bounded above by $1$. Thus, $b:=\lim_{n\to\infty}b_{n}=\sup_{n\in\mathbb{N}}b_{n}\in\left(0,1\right]$ exists.

Now, the ratio test easily implies that the power series $$ \sum_{n=1}^{\infty}a_{n}x^{n} $$ has radius of convergence $R=\frac{1}{b}$.

But as noted in the comments, the question Limit of $L^p$ norm shows that we have $$ \sqrt[n]{a_{n}}\xrightarrow[n\to\infty]{}\left\Vert f\right\Vert _{\infty}=1. $$ Using the root test, we see that the power series from above also has the radius of convergence $R=\frac{1}{\lim_{n\to\infty}\sqrt[n]{a_{n}}}=1$. All in all, we conclude $\lim_{n\to\infty}b_{n}=R=1=\left\Vert f\right\Vert _{\infty}$, as desired.

EDIT: What I am using here is in more detail as follows: From above, we know that the limit $b := \lim_{n \to \infty} \frac{a_{n+1}}{a_n}$ exists.

Now, if $|x| < 1/b$, we have $$ \bigg|\frac{a_{n+1}x^{n+1}}{a_n x^n}\bigg| = \frac{a_{n+1}}{a_n} \cdot |x| \to b |x| <1, $$ so that the series $\sum_n a_n x^n$ converges (absolutely).

Conversely, if $|x|> 1/b$, a similar argument shows that the series diverges. Hence, the radius of convergence is given by $1/b$.

Essentially the same holds for the root test.

$\endgroup$
  • $\begingroup$ The part of proof that the limit $a_{n+1}/a_{n}$ is incorrect using root and ration tests on power series. at the boundary of both tests, the result should be inconclusive. $\endgroup$ – runaround Feb 19 '16 at 19:14
  • $\begingroup$ @runaround: Yes, at the boundary, it is inconclusive. Nevertheless, the radius of convergence is given by $\lim_n |a_{n}/a_{n+1}|$ if that limit exists. Likewise for the root test. See e.g. here: en.wikipedia.org/wiki/… $\endgroup$ – PhoemueX Feb 19 '16 at 19:27
  • $\begingroup$ @PhiemueX, if the power serie has the convergence radius 1, it does not give the value of the convergence radius from the ratio test is also 1. $\endgroup$ – runaround Feb 19 '16 at 19:50
  • $\begingroup$ @runaround: Please see my edit. $\endgroup$ – PhoemueX Feb 19 '16 at 20:48

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.