10

Prove or disprove:

if $f(x)\ge 0,\forall x\in [-\pi,\pi]$,show that $$\left(\int_{-\pi}^{\pi}f(x)\sin{x}dx\right)^2+\left(\int_{-\pi}^{\pi}f(x)\cos{x}dx\right)^2\le\dfrac{\pi}{2}\int_{-\pi}^{+\pi}f^2(x)dx$$

I can prove this if $2\pi$ takes the place of $\dfrac{\pi}{2}$

because use Cauchy-schwarz inequality we have $$\left(\int_{-\pi}^{\pi}f(x)\sin{x}dx\right)^2\le\int_{-\pi}^{\pi}\sin^2{x}dx\int_{-\pi}^{\pi}f^2(x)dx$$

$$\left(\int_{-\pi}^{\pi}f(x)\cos{x}dx\right)^2\le\int_{-\pi}^{\pi}\cos^2{x}dx\int_{-\pi}^{\pi}f^2(x)dx$$ add this two inequality, we have $$\begin{align*}\left(\int_{-\pi}^{\pi}f(x)\sin{x}dx\right)^2+\left(\int_{-\pi}^{\pi}f(x)\cos{x}dx\right)^2 &\le \int_{-\pi}^{\pi}f^2(x)dx\int_{-\pi}^{\pi}(\sin^2{x}+\cos^2{x})dx\\ &=2\pi\int_{-\pi}^{+\pi}f^2(x)dx\end{align*}$$

see this Discrete form of inequality:Prove this inequality with Cauchy-Schwarz inequality

So far, I haven't found any counterexamples,such $f(x)=1,\sin{x}+1$ it such this inequality

math110
  • 94,932
  • 17
  • 148
  • 519

3 Answers3

5

Note that \begin{align*} \int_{-\pi}^\pi f(x) \sin x \,dx &= \int_0^{\pi/2} (f(x) - f(x - \pi) + f(\pi - x) - f(-x)) \sin x \,dx \\ &= \int_0^{\pi/2} (g(x) + h(x)) \sin x \,dx \end{align*} and similarly \begin{align*} \int_{-\pi}^\pi f(x) \cos x \,dx &= \int_0^{\pi/2} (f(x) - f(x - \pi) - f(\pi - x) + f(-x)) \cos x \,dx \\ &= \int_0^{\pi/2} (g(x) - h(x)) \cos x \,dx \end{align*} where we define $g, h : [0, \pi/2] \to \mathbb{R}$ by $g(x) = f(x) - f(x - \pi)$ and $h(x) = f(\pi - x) - f(-x)$.

Then by Cauchy-Schwarz, \begin{align*} \left(\int_0^{\pi/2} (g(x) + h(x)) \sin x \,dx\right)^2 &\leq \int_0^{\pi/2} \sin^2 x \,dx \int_0^{\pi/2} (g(x) + h(x))^2 \,dx \\ &= \frac{\pi}{4} \int_0^{\pi/2} (g(x) + h(x))^2 \,dx \end{align*} and \begin{align*} \left(\int_0^{\pi/2} (g(x) - h(x)) \cos x \,dx\right)^2 &\leq \int_0^{\pi/2} \cos^2 x \,dx \int_0^{\pi/2} (g(x) - h(x))^2 \,dx \\ &= \frac{\pi}{4} \int_0^{\pi/2} (g(x) - h(x))^2 \,dx \end{align*} hence \begin{align*} \left(\int_{-\pi}^\pi f(x) \sin x \,dx\right)^2 &+ \left(\int_{-\pi}^\pi f(x) \cos x \,dx\right)^2 \\ &\leq \frac{\pi}{4} \int_0^{\pi/2} (g(x) + h(x))^2 \,dx + \frac{\pi}{4} \int_0^{\pi/2} (g(x) - h(x))^2 \,dx\\ &= \frac{\pi}{2} \int_0^{\pi/2} (g(x))^2 + (h(x))^2 \,dx \\ &\leq \frac{\pi}{2} \int_0^{\pi/2} (f(x))^2 + (f(x - \pi))^2 + (f(\pi - x))^2 + (f(-x))^2 \,dx \\ &= \frac{\pi}{2} \int_{-\pi}^\pi (f(x))^2 \,dx \end{align*} as desired, where the last inequality holds because $f$ is nonnegative.

user125932
  • 8,871
  • Nice,+1 so the Discrete inequality can you prove it?Thanks https://math.stackexchange.com/questions/3486966/prove-this-inequality-with-cauchy-schwarz-inequality?noredirect=1&lq=1 – math110 Jan 07 '20 at 04:13
  • You should mention that you need the non-negativity of $f$ in the very last inequality. Nice Answer! – daw Jan 07 '20 at 07:41
  • @functionsug I'm sorry -- I haven't been able to prove it. my answer in the continuous case above does not really seem to translate to the discrete case; translating the other answer here to the discrete case gives the first comment on the other question by WimC, but it's not clear to me how to proceed. If the arguments to $\cos$ and $\sin$ were $2\pi k/n$ instead of $k$, I think the inequality would not be so hard, but the fact that they are integers, not rational multiples of $\pi$, creates a distribution of points mod $2\pi$ which is hard to work with. – user125932 Jan 09 '20 at 07:52
3

I think so. We can assume that $f$ is $2\pi$-periodic over $\mathbb{R}$. The sum of integrals on the left-hand side is presented in the form $\left|\displaystyle\int\limits_{-\pi}^{\pi}f(x)e^{ix}\,dx\right|^2=|\rho e^{i\alpha}|^2=\rho^2$. But due to $2\pi$-periodicity, $$\rho=\rho e^{i\alpha}e^{-i\alpha}=\displaystyle\int\limits_{-\pi}^{\pi}f(x)e^{i(x-\alpha)}\,dx=\displaystyle\int\limits_{-\pi}^{\pi}f(t+\alpha)e^{it}\,dt.$$ Since $f$ is real-valued, $\rho=\displaystyle\int\limits_{-\pi}^{\pi}f(t+\alpha)\cos t\,dt$. Since $f\geq0$ we have $\rho\leq\displaystyle\int\limits_{-\frac{\pi}{2}}^{\frac{\pi}{2}}f(t+\alpha)\cos t\,dt$. Therefore, $$\rho^2\leq\displaystyle\int\limits_{-\frac{\pi}{2}}^{\frac{\pi}{2}}f^2(t+\alpha)\,dt\displaystyle\int\limits_{-\frac{\pi}{2}}^{\frac{\pi}{2}}\cos^2 t\,dt\leq\frac{\pi}{2}\displaystyle\int\limits_{-\pi}^{\pi}f^2(t+\alpha)\,dt=\frac{\pi}{2}\displaystyle\int\limits_{-\pi}^{\pi}f^2(\tau)\,d\tau.$$

peter a g
  • 5,423
thing
  • 1,758
  • 1
    I hope you don't mind the reformatting + massaging of English. It was perhaps more than you, the author, might consider to be polite - if so, I apologize. – peter a g Jan 07 '20 at 14:26
  • You might add, however, for emphasis, that $\rho\ge 0$ (as was obviously implied). – peter a g Jan 07 '20 at 14:37
  • In my opinion, this is obvious, since it is a module of a complex number. – thing Jan 07 '20 at 14:39
  • I agree - but you square an inequality $$ \rho \le \cdots$$ to obtain $$ \rho^2 \le (\cdots)^2,$$ which is valid because it involves positive numbers - hence the suggestion for 'emphasis' - but, as you say, it is 'obvious' – peter a g Jan 07 '20 at 14:41
-1

This is a consequence of parseval identity. Let $\mathcal{H}$ a Hilbert space (dont worry all you need to know is that you have an inner product and the concept of orthogonal vectors, i.e. when $\langle f, g\rangle = 0$ and that the norm is given by $\langle f, f \rangle = \lVert f \rVert^2$) and let $\{e_n\}$ be a numerable orthonormal set then if $f \in \mathcal{H}$ $$ \sum_n \lvert \langle e_n, f\rangle \rvert^2 \leq \lVert f \rVert^2. $$ with equality if $\{e_n\}$ is a basis. In this case $$ \lVert f \rVert^2 = \int_{-\pi}^{\pi} \lvert f(x) \rvert^2 \,dx $$ and $$ e_1(x) = A\cos(x), \quad e_2(x) = A\sin(x) $$ where $A$ is chosen so they are normal.

Parseval identity is deduced from the fact that for orthonormal (it also works with orthogonal, but the last equality is not true) vectors $u_n$ one has $$ \lVert \sum a_n u_n \rVert^2 = \sum \lVert a_n u_n \rVert^2 = \sum \lvert a_n \rvert^2 $$ and you can write $$ f = \langle e_1,f \rangle e_1 + \langle e_2, f\rangle e_2 + (f - \langle e_1,f \rangle e_1 + \langle e_2, f\rangle e_2) $$ where the 3 parts of the sum are orthogonal and you have that the norm of $f$ is bigger than the norm of the first 2 terms.

EDIT: it seems the constants are off. It does not coincide with the one in parseval's.

  • Parseval doesn't gives $\dfrac{\pi}{2}$ ,I think it only gives $2 \pi$. – N. S. Jan 07 '20 at 03:19
  • Note that this approach cannot work since you never used the positivity of $f$. And the inequality is definitely not true for all $f$. – N. S. Jan 07 '20 at 03:20
  • Well I am not 100% about the constant. but if I remember correctly it does indeed work as $\int cos(x)^2 = \pi /2$, you could also modify the norm by multiplying it by a constant. About positivity, I am sure you only need for $f$ to be real valued. Could you say why it doesnt work in general? Note that this requires $f$ to be square integrable. – k76u4vkweek547v7 Jan 07 '20 at 03:31
  • Ok, I get it, the constant IS the issue. The fact that $f$ is positive means that $\langle f, sin \rangle = 0$, just do the change of variables $x \mapsto -x$. – k76u4vkweek547v7 Jan 07 '20 at 03:34
  • The posted inequality falls for $f(x)=\sin(x)$. If your proof would show it, it would also work for this function. – N. S. Jan 07 '20 at 03:35
  • 1
    And no, the fact that $f$ is positive does NOT mean that $<f ,\sin(x)>=0$. – N. S. Jan 07 '20 at 03:37
  • 1
    you are right, in my mind confused $f > 0$ with $f$ being odd/even. – k76u4vkweek547v7 Jan 07 '20 at 03:42