5

Let $f$ be a continuously differentiable function on $[1, +\infty]$.

Question: Is it true that $\int_1^{+\infty} (f')^2dx < +\infty$ implies $\int_1^{+\infty} (\frac f x)^2 dx< +\infty$?

This is the integral version of the question in this post: If a positive series converge in square sum, will its average series converge in square sum?.

I am interested in this question because it seems rather simple and concrete. I think there should be some techniques to deal with such kind of problems. However, I am not able to prove or give a counterexample of the claim.

What I have got: (1) The claim holds when $f$ is a power function. (2) The claim fails if 'square integrability' of $f'$ and $\frac f x$ is replaced by 'absolute integrability', since $f$ can be chosen to be a constant $1$.

Alex Ravsky
  • 90,434
  • 1
    This seems to be related to Hardy inequalities ($\leftarrow$ google keyword), but I cannot find this result – Calvin Khor Oct 25 '20 at 09:05
  • @CalvinKhor Thank you so much. This is exactly Hardy inequality. Would you mind posting the comment as an answer? – Zhang Yuhan Oct 25 '20 at 10:46
  • 2
    No, not exactly, it is not starting from $1$, and the derivative might not be positive (which seems required in dimension $1$) – LL 3.14 Oct 25 '20 at 10:51
  • 1
    This may also be a good link but I don't have the time to go through the proofs atm – Calvin Khor Oct 25 '20 at 11:03
  • @LL3.14 I see it now. It is not mentioned in the link in my previous comment that the derivative needs to be positive. – Zhang Yuhan Oct 25 '20 at 11:14
  • In the link you mentionned, the derivative is called $f$ (what you call $f'$ here). Of course it has to be positive (if not, $f^p$ might be a complex number). However, similar proof as the usual Hardy inequality gives a similar answer (see my answer). Remark that the $f(1)$ is added, so it is different from Hardy's inequality. – LL 3.14 Oct 25 '20 at 11:20

1 Answers1

5

We have an inequality similar to Hardy's inequality: $$ \boxed{\left\|\frac{f(x)}{x}\right\|_{L^2[1,\infty)} ≤ f(1) + 2 \left\|f'\right\|_{L^2[1,\infty)}} $$

Proof: First remark that $$ ∫_1^\infty \left|\frac{f(x)}{x}\right|^2\mathrm d x = \left\|\frac{1}{x} \left(f(1)+ \int_1^x f'\right)\right\|_{L^2[1,\infty)}^2 \\ ≤ \left(\left\|\frac{f(1)}{x}\right\|_{L^2[1,\infty)} + \left\|\frac{1}{x} \int_1^x f'\right\|_{L^2[1,\infty)}\right)^2 $$ and the first integral is easily bounded since $$ \left\|\frac{f(1)}{x}\right\|_{L^2[1,\infty)} = f(1) \left(\int_1^\infty x^{-2}\,\mathrm d x\right)^{1/2} = f(1) $$ To bound the second integral, we can do the same strategy as for the classical Hardy's inequality and use first a change of variable $t= sx$ to get $$ \left\|\frac{1}{x} \int_1^x f'(t)\,\mathrm d t\right\|_{L^2[1,\infty)} = \left\|\int_{1/x}^1 f'(sx)\,\mathrm d s\right\|_{L^2[1,\infty)} \\ = \left\|\int_0^1 \mathbf{1}_{\{s>1/x\}} f'(sx)\,\mathrm d s\right\|_{L^2_x[1,\infty)} \\ ≤ \int_0^1 \left\| \mathbf{1}_{\{sx>1\}} f'(sx)\right\|_{L^2_x[1,\infty)} \,\mathrm d s $$ and then a second change of variable giving $$ \int_0^1 \left\| \mathbf{1}_{\{sx>1\}} f'(sx)\right\|_{L^2_x[1,\infty)} \,\mathrm d s = \int_0^1 \left(\int_1^\infty |\mathbf{1}_{\{sx>1\}} f'(sx)|^2\,\mathrm d x\right)^{1/2} \,\mathrm d s \\ = \int_0^1 \left(\int_{s}^\infty |\mathbf{1}_{\{y>1\}} f'(y)|^2\,\,\mathrm d y\right)^{1/2} s^{-1/2}\,\mathrm d s \\ = \int_0^1 \left(\int_{1}^\infty |f'(y)|^2\,\,\mathrm d y\right)^{1/2} s^{-1/2}\,\mathrm d s = 2 \|f'\|_{L^2[1,\infty)} $$ Therefore $$ ∫_1^\infty \left|\frac{f(x)}{x}\right|^2\mathrm d x ≤ \left(f(1) + 2 \left\|f'\right\|_{L^2[1,\infty)}\right)^2 $$

LL 3.14
  • 12,457
  • Of course not, the primitive of $s^{-1/2}$ is $2s^{1/2}$ ... the global criterion is that $s^a$ is integrable near $0$ iff $a>-1$. – LL 3.14 Oct 25 '20 at 11:30
  • I added a step to clarify. Since there is the indicator function and $s<1$, the integral just starts from $1$ in any cases. – LL 3.14 Oct 25 '20 at 11:35
  • It seems $f'$ being positive is not essential even if the integral in the question is from $0$ to $+\infty$ as long as $f(0) = 0$. Is it true that even if $f'$ is signed, Hardy Inequality still holds as long as $p$ is chosen that $(f')^p$ is not a complex number? – Zhang Yuhan Oct 25 '20 at 11:54
  • It might just be true by taking $|f'|^p$ I suppose. – LL 3.14 Oct 25 '20 at 19:59