4
$\begingroup$

The following inequality comes up in connection with motion in a dipole field. One has to show that $$\int_\overline{\theta}^{\pi}\frac{d\theta}{\sqrt{1-\lambda \cos{\theta}}}\gt\pi$$ where $$\overline{\theta}= \begin{cases}0, & 0\lt\lambda \lt1\\ \arccos(\frac{1}{\lambda}), & \lambda \gt 1\end{cases}.$$ The case $0\lt\lambda\lt 1$ is easily verified. For then $$\int_{0}^{\pi}\frac{d\theta}{\sqrt{1-\lambda \cos{\theta}}}=\int_{0}^{\pi/2}\frac{d\theta}{\sqrt{1-\lambda \cos{\theta}}}+\int_{\pi/2}^{\pi}\frac{d\theta}{\sqrt{1-\lambda \cos{\theta}}}$$ $$\implies\int_{0}^{\pi}\frac{d\theta}{\sqrt{1-\lambda \cos{\theta}}}=\int_{0}^{\pi/2}\frac{d\theta}{\sqrt{1-\lambda \cos{\theta}}}+\int_{0}^{\pi/2}\frac{d\theta}{\sqrt{1-\lambda \sin{\theta}}}\gt2\int_{0}^{\pi/2}d\theta=\pi.$$ However I am finding it difficult to prove the second case. Any pointers would be helpful.

Thanks.

NOTE: The inequality does not hold for all values of $\lambda > 1$, as has been noted in the following responses.

$\endgroup$
  • 1
    $\begingroup$ So the question amounts to bounding $\cos(\theta)$ for $\theta \in [\arccos(1/\lambda),\pi]$ from below. On $[\arccos(1/\lambda),\pi/2]$ you have the convenient lower bound given by concavity: $\cos(\theta)$ is lies above the line between $(\arccos(1/\lambda),1/\lambda)$ and $(\pi/2,0)$. On $[\pi/2,\pi]$ you can use convexity to bound it from below by the tangent line, so in that region you have $\cos(\theta) \geq -(\theta-\pi/2)$. How good of a bound do these two give you? $\endgroup$ – Ian Nov 1 '15 at 21:03
  • 2
    $\begingroup$ Actually, are you sure this result is true? For instance WA tells me that $\int_{\pi/3}^\pi \frac{1}{\sqrt{1-2\cos(x)}} dx$ (the relevant integral for $\lambda=2$) is about $2.1$. $\endgroup$ – Ian Nov 1 '15 at 21:21
  • $\begingroup$ @Ian: Using concavity, as suggested on $[\arccos(\frac{1}{\lambda}), \pi / 2]$ , I obtained $\int_{\arccos(\frac{1}{\lambda})}^{\pi /2} \frac{d\theta}{\sqrt{1-\lambda \cos (\theta)}} \gt \pi - 2 \arccos (\frac{1}{\lambda})$. Still to work out in the range $[\pi/2, \pi]$. The inequality is necessarilly true, but the argument i have for this is physical. Basically, $\theta$ here is the angular position (with z-axis) of a particle moving in a dipole field. The angular position in such a field cannot have turning points. $\endgroup$ – vnd Nov 1 '15 at 21:27
  • 1
    $\begingroup$ Is it possible that it does not hold for $\lambda$ too large? Just playing around on Wolfram I find that the inequality appears to fail somewhere in between $\lambda=1.2$ and $\lambda=1.3$. $\endgroup$ – Ian Nov 1 '15 at 21:29
  • $\begingroup$ It should be not too difficult to integrate this exactly using elliptic integrals... maybe that would be an alternative... $\endgroup$ – tired Nov 1 '15 at 21:52
1
$\begingroup$

If I divide by $\lambda$ the inequality becomes: $$\int_{\underline{\theta}}^{\pi}\frac{d\theta}{\sqrt{\cos(\underline{\theta})-\cos(\theta)}}\geq \pi \sqrt{\lambda}$$

Now $$\int_{\pi/2}^{\pi}\frac{d\theta}{\sqrt{\cos(\underline{\theta})-\cos(\theta)}}\leq \int_{\pi/2}^{\pi}\frac{d\theta}{\sqrt{-\cos(\theta)}}d\theta=\int_0^{\pi/2}\frac{d\theta}{\sqrt{\sin(\theta)}}<+\infty$$ and as if $\theta \in [ \underline{\theta}, \pi/2]$, we have $\cos(\theta)-\cos(\underline{\theta})=(\theta-\underline{\theta})(-\sin(c))$ for some $c\in [ \underline{\theta}, \pi/2]$, we get $|\cos(\theta)-\cos(\underline{\theta})|\geq(\theta-\underline{\theta})\sin(\underline{\theta}) $ Hence $$\int_{\underline{\theta}}^{\pi/2}\frac{d\theta}{\sqrt{\cos(\underline{\theta})-\cos(\theta)}}\leq\frac{1}{\sqrt{\sin(\underline{\theta})}} \int_{\underline{\theta}}^{\pi/2}\frac{d\theta}{\sqrt{\theta-\underline{\theta}}}= 2\frac{\sqrt{\pi/2-\underline{\theta}}}{\sqrt{\sin(\underline{\theta})}} $$

Now as $\displaystyle \sin(\underline{\theta})=\sqrt{1-\frac{1}{\lambda^2}}$, we have $\displaystyle 2\frac{\sqrt{\pi/2-\underline{\theta}}}{\sqrt{\sin(\underline{\theta})}}$ bounded as $\lambda\to +\infty$, so if I am not wrong, there is a problem.

$\endgroup$
  • $\begingroup$ The sine will actually be the square root of $1-1/\lambda^2$. But that does not impede your point. In fact, you've actually proven that the integral goes to zero, because $\arccos(1/\lambda)$ goes to $\pi/2$ as $\lambda \to \infty$. $\endgroup$ – Ian Nov 1 '15 at 22:09
  • $\begingroup$ Oh yes thanks, I have edited. Have you seen a problem in my computations ? $\endgroup$ – Kelenner Nov 1 '15 at 22:10
  • $\begingroup$ Sorry, I should say that you proved that the integral up to $\pi/2$ goes to zero. But I do think you have proven that the result does not hold for all $\lambda$: if it did, the rescaled integral would need to blow up, but it does not. In fact I think the integral is something like $O(\lambda^{-1/2})$. $\endgroup$ – Ian Nov 1 '15 at 22:13
  • $\begingroup$ Ok, thanks. Perhaps the OP can modify the hypothesis. $\endgroup$ – Kelenner Nov 1 '15 at 22:14
1
$\begingroup$

Your inequality is wrong in general (indicated below), and your proof for $\lambda<1$ has a small mistake. Let us start with the latter.

The problem with your proof

You divide the integral in the intervals $(0,\pi/2)$ and $(\pi/2,\pi)$ and that is fine. But since $\cos \theta<0$ in the interval $(\pi/2,\pi)$ it is impossible that you get as an integrand $1/\sqrt{1-\lambda\sin \theta}$ after your change of variables. What you get if you just shift $\theta\mapsto \theta-\pi/2$ is $1/\sqrt{1+\lambda\sin\theta}$. But this is less than $1$ on the interval. Instead, let $\theta\mapsto \pi-\theta$. Then, you will end up with $$ \int_0^{\pi/2}\frac{1}{\sqrt{1-\lambda\cos\theta}}+\frac{1}{\sqrt{1+\lambda\cos\theta}}\,d\theta. $$ Then, according to the inequality $a+b\geq 2\sqrt{ab}$, you find that your integrand is bounded below by $$ \frac{2}{\sqrt[4]{1-\lambda^2\cos^2\theta}} $$ which is greater than $2$ on the interval. Integrating you get a lower bound of $\pi$.

Proof that the inequality is not true for all $\lambda>1$

Here, as suggested by @Ian, it is sufficient to consider $\lambda=2$, and hence show that the integral $$ \int_{\pi/3}^\pi\frac{1}{\sqrt{1-2\cos\theta}}\,d\theta<\pi. $$ In the interval $(\pi/3,\pi)$ it holds that $$ 1-2\cos\theta>\frac{9}{2\pi}(\theta-\pi/3). $$ Thus $$ \int_{\pi/3}^\pi\frac{1}{\sqrt{1-2\cos\theta}}\,d\theta< \int_{\pi/3}^\pi\frac{1}{\sqrt{\frac{9}{2\pi}(\theta-\pi/3)}}\,d\theta=\frac{4\pi}{3\sqrt{3}}<\pi. $$

$\endgroup$

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.