1
$\begingroup$

Find the best constants $c_1$ and $c_2$ such that: $$\frac{c_1x}{1+x^2}\leq \arctan(x)\leq \frac{c_2x}{1+x}\qquad \forall x\geq0$$

What I've tried so far: It's easy to determine the constant $c_1$ just by studying $f(x)=\frac{(1+x^2)\arctan(x)}{x}$.

By analyzing this function we can see that it has a lower bound which is $1$, so the best constant is clearly $c_1=1$. But what about the second inequality? I tried to do the same by studying $g(x)=\frac{(1+x)\arctan(x)}{x}$ but I really have difficulties in finding the maximum of the function (which I know, by computation, exists).

Any hint/help is really appreciated.

$\endgroup$
2
  • 1
    $\begingroup$ NoTe : taht ....left hand side is wrong for $-1<x<0$ !!! $\endgroup$
    – Khosrotash
    Aug 22 '17 at 9:54
  • $\begingroup$ Oh, sorry, I forgot to say for all $x\geq 0$. I'll edit. $\endgroup$ Aug 22 '17 at 9:56
3
$\begingroup$

For any $c\in(1,2)$, the function $f_c(x)=\frac{c x}{1+x}-\arctan(x)$ attains its maximum value at $x=\frac{1+\sqrt{2c-c^2}}{c-1}$. Let $d=\frac{1+\sqrt{2c-c^2}}{c-1}$: we have $d\in(1,+\infty)$ and $c=\frac{(1+d)^2}{1+d^2}$, so the problem boils down to finding a zero of $$g(d)=\frac{(1+d)^2}{1+d^2}\cdot\frac{d}{1+d}-\arctan(d)=\frac{d(1+d)}{1+d^2}-\arctan(d) $$ on $(1,+\infty)$, or, by setting $d=\tan\theta$, a zero of $$h(\theta) = \sin\theta\,(\sin\theta+\cos\theta)-\theta $$ on the interval $\theta\in\left(\frac{\pi}{4},\frac{\pi}{2}\right)$. If we apply a step of Newton's method to $\frac{h(\theta)}{\theta}$ with starting point $\theta_0=\frac{\pi}{2}$, we get that the only zero of $h(\theta)$ is close to $\frac{2\pi}{2+\pi}$, hence:

$$ c_2 \approx \color{blue}{1+\sin\left(\tfrac{4\pi}{2+\pi}\right)}\approx\tfrac{5}{3}. $$

The above substitutions show that the problem of finding $c_2$ is essentially equivalent to solving Kepler's equation, and there is a large amount of literature on such topic. See, for instance, Colwell - Solving Kepler's Equations Over three Centuries.

$\endgroup$
1
$\begingroup$

You are not going to find the exact value of $c_2$ I think. You should study the function $f_a(x)=\frac{ax}{1+x}-\arctan x$. Since $$\lim_{x\to\infty}\frac{ax}{1+x}-\arctan x=a-\frac\pi2$$ you want $a\ge\frac\pi2$. Computing the derivative of $f_a$ you get that $$\frac{d}{dx}\left( \frac{ax}{1+x}-\arctan x\right) =\frac{1}{\left( x^{2}+1\right) \left( x+1\right) ^{2}}\left( (a-1)x^{2}-2x+a-1\right)\ge 0 $$ for $(a-1)x^{2}-2x+a-1\geq0$. This equation has two distinct solutions for $0<a<2$, which are $$ \frac{1\pm\sqrt{1-(a-1)^{2}}}{a-1}. $$ Hence, the function increases for $0<x<\frac{1-\sqrt{1-(a-1)^{2}}}{a-1}$ and decreases for $x>\frac{1+\sqrt{1-(a-1)^{2}}}{a-1}$. This shows that at $\frac{1+\sqrt{1-(a-1)^{2}}}{a-1}$ you have a global minimum. You want the function $f_a$ to be always positive, so you are interested in the value of $f_a$ at $\frac{1+\sqrt{1-(a-1)^{2}}}{a-1}$. Define $$g(a):=f_a\left(\frac{1+\sqrt{1-(a-1)^{2}}}{a-1}\right)=\frac{a+a\sqrt{1-(a-1)^{2}}}{a+\sqrt{1-(a-1)^{2}}}-\arctan\frac {1+\sqrt{1-(a-1)^{2}}}{a-1}% $$ For $a=\frac\pi2$ you have $g(a)=1.1959- 1. 5708<0$ while for $a=2$ you have $g(2)=1-\arctan 1>0$. Since $g$ is continuous, there exists a first value of $a$ after $\frac\pi2$ such that $g(a)=0$. This gives you the best constant.

$\endgroup$
1
$\begingroup$

We certainly need $c:=c_2\geq{\pi\over2}$. The function $$f(x):={cx\over 1+x}-\arctan x\qquad(x\geq0)$$ has derivative $$f'(x)={(c-1)x^2-2x+(c-1)\over(1+x)^2(1+x^2)}\ ,$$ hence increases for small $x\geq0$ to a local maximum, then decreases to a local minimum, and from then on increases indefinitely, see the following figure. The minimum is at $x_*=\bigl(1+\sqrt{2c-c^2}\bigr))/(c-1)$. We now have to find the smallest $c$ for which $f(x_*)\geq0$. Due to the $\arctan$ this leads to a transcendental equation, so that we have to settle with a practical numerical value. It so happens that $c={5\over3}$ leads to $x_*=(3+\sqrt{5})/2$, and above all to $f(x_*)\approx 0.0000788309>0$.

enter image description here

$\endgroup$

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.