6
$\begingroup$

If $f : [0,\infty) \to [0,\infty)$ is an integrable function, then what is the best possible constant $k$, for which the following ineqality holds:

$$\int_0^{\infty}f(x)dx \leq k\left(\int_0^{\infty}\sqrt{x}f(x)dx\right)^{1/2}\cdot\left(\int_0^{\infty}f^{2}(x)dx\right)^{1/4}$$

For example, $k=2$ is a bound, since, $$\int_0^{\infty}f(x)dx = \int_0^{y}f(x)dx + \int_y^{\infty}f(x)dx < \sqrt{y}\left(\int_0^{\infty}f^2(x)dx\right)^{1/2} + \frac{1}{\sqrt{y}}\int_0^{\infty}\sqrt{x}f(x)dx$$

and setting, $ y = \dfrac{\displaystyle \int_0^{\infty}\sqrt{x}f(x)dx}{\left(\displaystyle\int_0^{\infty} f^2(x)dx\right)^{1/2}}$, establishes the inequality for the case $k=2$, but it is not strict.

How to improve the bound $k$ and also determine the function where equality holds for the best constant $k$ ?

Thank you.

$\endgroup$
  • 2
    $\begingroup$ One can get, rather trivially, that $k \ge \sqrt{3/2}$; but I strongly doubt that this is sharp. $\endgroup$ – user61527 Jul 13 '14 at 22:13
  • $\begingroup$ @T.Bongers: did you get that with a test function? $\endgroup$ – robjohn Jul 14 '14 at 13:26
  • $\begingroup$ @robjohn No, I just integrated $\chi_{[0,1]}$. $\endgroup$ – user61527 Jul 14 '14 at 17:44
  • $\begingroup$ @T.Bongers: so $\chi_{[0,1]}$ was your test function :-) $\endgroup$ – robjohn Jul 14 '14 at 18:12
  • $\begingroup$ @robjohn But test functions are smooth! :) $\endgroup$ – user61527 Jul 14 '14 at 23:19
6
$\begingroup$

Suppose $$ \int_0^\infty\sqrt{x}f(x)\,\mathrm{d}x=A\tag{1} $$ and $$ \int_0^\infty f^2(x)\,\mathrm{d}x=B\tag{2} $$ Then, letting $g(x)^2=A^{1/2}B^{-3/4}f(AB^{-1/2}x)$, we get $$ \int_0^\infty\sqrt{x}g(x)^2\,\mathrm{d}x=\int_0^\infty g(x)^4\,\mathrm{d}x=1\tag{3} $$ and $$ \int_0^\infty f(x)\,\mathrm{d}x=A^{1/2}B^{1/4}\int_0^\infty g(x)^2\,\mathrm{d}x\tag{4} $$ Thus, maximizing $\int_0^\infty g(x)^2\,\mathrm{d}x$ given $(3)$ yields the best constant for the given inequlity.


The variation of $(4)$ is stationary when $$ \int_0^\infty g(x)\,\delta g(x)\,\mathrm{d}x=0\tag{5} $$ Equations $(3)$ imply $$ \int_0^\infty\sqrt{x}g(x)\,\delta g(x)\,\mathrm{d}x=\int_0^\infty g(x)^3\,\delta g(x)\,\mathrm{d}x=0\tag{6} $$ To insure $(5)$ holds whenever $(6)$ does, we must have that $g(x)^2=0$ or $g(x)^2=a-b\sqrt{x}$.

Thus, consider $g(x)^2=a-b\sqrt{x}$ on $\left[0,\frac{a^2}{b^2}\right]$ and $g(x)^2=0$ for $x\gt\frac{a^2}{b^2}$. $$ \begin{align} \int_0^{a^2/b^2}\sqrt{x}g(x)^2\,\mathrm{d}x &=\int_0^{a^2/b^2}\left(a\sqrt{x}-bx\right)\,\mathrm{d}x\\ &=\frac{a^4}{6b^3}\tag{7} \end{align} $$ $$ \begin{align} \int_0^{a^2/b^2}g(x)^4\,\mathrm{d}x &=\int_0^{a^2/b^2}\left(a^2-2ab\sqrt{x}+b^2x\right)\,\mathrm{d}x\\ &=\frac{a^4}{6b^2}\tag{8} \end{align} $$ $$ \begin{align} \int_0^{a^2/b^2}g(x)^2\,\mathrm{d}x &=\int_0^{a^2/b^2}\left(a-b\sqrt{x}\right)\,\mathrm{d}x\\ &=\frac{a^3}{3b^2}\tag{9} \end{align} $$


Solving $(7)$ and $(8)$ given $(3)$ yields $$ (a,b)=\left(6^{1/4},1\right)\tag{10} $$ Plugging $(10)$ into $(9)$ yields an optimal constant of $$ \left(\frac83\right)^{1/4}\doteq1.277886208492545\tag{11} $$ attained by the function $$ \left(6^{1/4}-\sqrt{x}\right)\left[0\le x\le6^{1/2}\right]\tag{12} $$ where $\left[\dots\vphantom{6^{1/2}}\right]$ are Iverson brackets.


The Slick Answer

Suppose $f(x)\ge0$ and $$ \int_0^\infty\sqrt{x}f(x)\,\mathrm{d}x=\int_0^\infty f(x)^2\,\mathrm{d}x=1\tag{13} $$ Then letting $u(x)=\left(6^{1/4}-\sqrt{x}\right)\left[0\le x\le6^{1/2}\right]$ $$ \begin{align} 0 &\le\int_0^\infty(f(x)-u(x))^2\,\mathrm{d}x\\ &=\int_0^\infty\left(f(x)^2-2f(x)u(x)+u(x)^2\right)\,\mathrm{d}x\tag{14} \end{align} $$ Applying $(13)$ to $(14)$, dividing by $2$, and rearranging, we get $$ \begin{align} 1 &\ge\int_0^\infty f(x)u(x)\,\mathrm{d}x\\ &\ge\int_0^\infty f(x)\left(6^{1/4}-\sqrt{x}\right)\,\mathrm{d}x\\ &=6^{1/4}\int_0^\infty f(x)\,\mathrm{d}x-1\tag{15} \end{align} $$ Therefore, $$ \int_0^\infty f(x)\,\mathrm{d}x\le\left(\frac83\right)^{1/4}\tag{16} $$

$\endgroup$
  • $\begingroup$ Oh !!! This is fantastic !!!! :D $\endgroup$ – r9m Jul 15 '14 at 0:21

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.