13

Earlier, a very interesting proof of an inequality has been proposed at MSE: How prove this inequality $\tan{(\sin{x})}>\sin{(\tan{x})}$

Here the question is: How to prove that $$\sin(\tanh x) \ge \tanh(\sin x), ~~ \text{for}~~ x \in [0,\pi/2]$$ Interestingly, the first three terms of the Mclaurin series are identical for both the functions.

Boris Valderrama
  • 768
  • 6
  • 19
Z Ahmed
  • 43,235
  • 1
    Let $f(x)=\sin(\tanh x)-\tanh(\sin x)$. Isn't there some numerically-based way to prove that $f(x)\geq 0$ on $[0,\pi/2]$? I mean, it's clearly true when we plug it into wolfram. So how does wolfram know? – Ben W Mar 30 '20 at 07:23
  • 1
    It appears that the inequality is actually true for all positive $x$. But that might be harder to prove. – 2'5 9'2 Mar 30 '20 at 07:39
  • @alex jordan Yeah, a proof is welcome! – Z Ahmed Mar 30 '20 at 08:04
  • @Ben W yes, but an (analytic) proof by hand is required. – Z Ahmed Mar 30 '20 at 08:06
  • @alex.jordan Due to the fact that $\tanh x\uparrow 1$ as $x\to\infty$ we and $\pi/2\approx .92$ have $\sin(\tanh x)\geq\sin .91\approx .79$ when $x\geq\pi/2$. On the other hand $\tanh(\sin x)\leq\tanh(1)\approx .76$. – Ben W Mar 30 '20 at 10:52
  • @DrZafarAhmedDSc Sure, I understand, but numerical methods can be supported by analytic proofs in some cases. For instance, we can numerically estimate $f$ on $[0,\pi/2]$. Isn't there some theorem that bounds the estimate? And mightn't that bound be sufficient for the inequality $f(x)\geq 0$? – Ben W Mar 30 '20 at 10:55

2 Answers2

3

Here is a proof:

Fact 1: It holds that, for all $x \ge 0$, $$\tanh x \ge \frac{10x^3 + 105x}{x^4 + 45x^2 + 105}.$$ (The proof is given at the end.)

Fact 2: It holds that, for all $u \in [0, 1]$, $$\tanh u \le \frac{u^3 + 15u}{6u^2 + 15}.$$ (The proof is given at the end.)

Using Fact 1 and $\tanh x \le 1$, we have $$\sin (\tanh x) \ge \sin \left(\frac{10x^3 + 105x}{x^4 + 45x^2 + 105}\right). \tag{1}$$

Using Fact 2 and $\sin x \in [0, 1]$, we have $$\tanh(\sin x) \le \frac{\sin^3 x + 15\sin x}{6\sin^2 x + 15}. \tag{2}$$

Using (1) and (2), it suffices to prove that $$\sin \left(\frac{10x^3 + 105x}{x^4 + 45x^2 + 105}\right) \ge \frac{\sin^3 x + 15\sin x}{6\sin^2 x + 15}. \tag{3}$$

Let $$A := \frac{10x^3 + 105x}{x^4 + 45x^2 + 105}, \quad B := x - \frac{x^3}{3!} + \frac{x^5}{5!}.$$

Using $\sin u \ge u - \frac{u^3}{3!} + \frac{u^5}{5!} - \frac{u^7}{7!}$ for all $u\ge 0$, we have $$\sin \left(\frac{10x^3 + 105x}{x^4 + 45x^2 + 105}\right) \ge A - \frac{A^3}{3!} + \frac{A^5}{5!} - \frac{A^7}{7!}. \tag{4}$$

Using $\sin u \le u - \frac{u^3}{3!} + \frac{u^5}{5!}$ for all $u \ge 0$, noting that $u\mapsto \frac{u^3 + 15u}{6u^2 + 15}$ is strictly increasing on $u\ge 0$, we have $$\frac{\sin^3 x + 15\sin x}{6\sin^2 x + 15} \le \frac{B^3 + 15B}{6B^2 + 15}. \tag{5}$$

Using (3), (4) and (5), it suffices to prove that $$A - \frac{A^3}{3!} + \frac{A^5}{5!} - \frac{A^7}{7!} \ge \frac{B^3 + 15B}{6B^2 + 15}$$ or $$\frac{x^7 h(x)}{( {x}^{4}+45\,{x}^{2}+105) ^{7} ( {x}^{10}-40{x}^{ 8}+640{x}^{6}-4800{x}^{4}+14400{x}^{2}+36000) }\ge 0 \tag{6}$$ where \begin{align*} h(x) &:= -7\,{x}^{36}-1785\,{x}^{34}-181440\,{x}^{32}-8778175\,{x}^{30}- 152703075\,{x}^{28}\\ &\qquad +3298464750\,{x}^{26}+165337839125\,{x}^{24}+ 826367130625\,{x}^{22}\\ &\qquad -45036632734375\,{x}^{20}-480887233668750\,{x}^{ 18}+2020497175546875\,{x}^{16}\\ &\qquad +5494328646890625\,{x}^{14}+ 335334922673625000\,{x}^{12}\\ &\qquad +6005165669281406250\,{x}^{10} + 43232877600909609375\,{x}^{8}\\ &\qquad +170152296290845312500\,{x}^{6} + 390099811698253125000\,{x}^{4}\\ &\qquad +490539329059500000000\,{x}^{2}+ 262396086816937500000. \end{align*}

We can prove that $h(x) \ge 0$ on $[0, \pi/2]$. Thus, (6) is true.

We are done.

$\phantom{2}$


Proof of Fact 1:

Using $\tanh x = 1 - \frac{2}{\mathrm{e}^{2x} + 1}$, it suffices to prove that $$f(x) := 2x - \ln \frac{x^4 + 10x^3 + 45x^2 + 105x + 105}{x^4 - 10x^3 + 45x^2 - 105x + 105} \ge 0.$$

We have $$f'(x) = \frac{2x^8}{x^8 - 10x^6 + 135x^4 - 1575x^2 + 11025} \ge 0.$$ Also, $f(0) = 0$. Thus, we have $f(x) \ge 0$ for all $x \ge 0$.

We are done.


Proof of Fact 2:

Using $\tanh u = 1 - \frac{2}{\mathrm{e}^{2u} + 1}$, it suffices to prove that $$g(u) := \ln\frac{u^3 + 6u^2 + 15u + 15}{-u^3 + 6u^2 - 15u + 15} - 2u \ge 0.$$

We have $$g'(u) = \frac{2u^6}{-u^6 + 6u^4 - 45u^2 + 225} \ge 0.$$ Also, $g(0) = 0$. Thus, we have $g(u) \ge 0$ on $[0, 1]$.

We are done.

River Li
  • 37,323
0

This is an attach from the right side of the interval, so only "half" of an argument. An attack from the left side would complete an argument. (Taylor's approximation might work, but it appears that you would need to establish an upper bound on the 8th derivative of $\sin(\tanh(x))-\tanh(\sin(x))$, which is not pretty.)

If $x$ is in $(0,\pi/2)$ and greater than $\tanh^{-1}(\arcsin(\tanh(\sin(\pi/2))))=\tanh^{-1}(\arcsin(\tanh(1)))$, then $\sin(\tanh(x))>\tanh(1)$, but $\tanh(\sin(x))$ is always at most $\tanh(1)$. So the claim is true on $\left(\tanh^{-1}(\arcsin(\tanh(1))),\frac{\pi}4\right)$.

Let $a_1=\tanh^{-1}(\arcsin(\tanh(\sin(\pi/2))))\approx1.316$.

Let $a_2=\tanh^{-1}(\arcsin(\tanh(\sin(a_1))))\approx1.237$.

Consider $x$ in $(a_2,a_1]$. Now $\sin(\tanh(x))>\tanh(\sin(a_1))$, but $\tanh(\sin(x))$ is at most $\tanh(\sin(a_1))$. So the claim is true on $\left(a_2,\frac{\pi}4\right)$.

In this manner, if we recursively define $a_n=\tanh^{-1}(\arcsin(\tanh(\sin(a_{n-1}))))$, we can show that the claim is true on $\left(a_n,\frac{\pi}4\right)$. And in fact, that it is true on $\left(\lim\limits_{n\to\infty}a_n,\frac{\pi}4\right)$.

So what is $L=\lim\limits_{n\to\infty}a_n$? If it is $0$, we are done. Well, we'd have to solve $\tanh^{-1}(\arcsin(\tanh(\sin(L))))=L$, equivalent to $\tanh(\sin(L))=\sin(\tanh(L))$. But if we were able to establish that the only solution that equation is $0$, then we'd have an easier proof in the first place (since our functions are continuous and it's easy to show that for very small $x$, $\sin(\tanh(x))>\tanh(\sin(x))$.

But numerically, $a_{12}<1$. (So says my computer. Tedious application of some rigorously bounded approximations of these functions could establish that.) So the claim holds at least on $[1,\pi/2]$. It would remain to show that the claim is true for $x$ on the left, that is, $x$ in $[0,1)$.

2'5 9'2
  • 54,717
  • The Cauchy argument principle together with a numerical computation means $f(z)$ has 7 zeros less poles on the complex unit disk. One of them is $z=0$. If you can find the other six, you'll be done. – Ben W Mar 31 '20 at 15:58