1

Let $0<u<\sin x<x<\frac{\pi}{2}$, and $$ f(x)=\frac{x}{u}\left(\frac{1}{(\sin x) - u}-\frac{1}{(\sin x) + u}\right)-\frac{1}{x-u}-\frac{1}{x+u}. $$ Show that $f(x)$ as a function of $x$ with fixed $u$ is convex.

We can also write $$ f(x)=\frac{2x}{(\sin^2x)-u^2}-\frac{2x}{x^2-u^2}. $$

I have calculated $$ f''(x) = \frac{2x\cos^2x}{u}\left(\frac{1}{((\sin x) -u)^3}-\frac{1}{((\sin x) + u)^3}\right) + \frac{x(\sin x)-2\cos x}{u}\left(\frac{1}{((\sin x) -u)^2}-\frac{1}{((\sin x) + u)^2}\right) -\frac{2}{(x-u)^3}-\frac{2}{(x+u)^3}. $$

Plots of the two-variable function support the conclusion.

I have been able to show $$ \lim_{u\to 0^+}f''(x;u) = \frac{12x\cos^2x}{\sin^4x} - \frac{8\cos x}{\sin^3x} + \frac{4x}{\sin^2x} - \frac{4}{x^3}>0. $$

Hui Zhang
  • 606
  • are the signs of the 2 last fractions ok, I would expect one - and one + so as to keep same symmetry than for the sine ones ? second question, convexity is regarding to $x$, to $u$, both ? – zwim Jun 27 '24 at 09:05
  • @zwim yes, they are both - for 1/(x-u) and 1/(x+u). The convexity is about $x$ for fixed $u$. Thanks for your attention! – Hui Zhang Jun 27 '24 at 09:06
  • Graphically https://www.desmos.com/calculator/6zjklzwshc it is concave with $h''<0$, but anyway of constant sign. Yet no clue how to prove this, the $\frac xu$ factor seems to play a major role to make the function go to $0$ in $0$, but it's harder to see the reason it makes everything concave. – zwim Jun 27 '24 at 09:55
  • I think multiple parameters should be separated with $,$ not $;$. – Lucenaposition Jun 27 '24 at 10:46
  • 1
    @zwim It is a nice plot. Some confusion from my writing is that the $\sin x-u$ should be $(\sin x)-u$ and as such I have revised the question. – Hui Zhang Jun 28 '24 at 06:12

1 Answers1

1

Some thoughts.

We rephrase the problem as follows.

Problem. Let $u \in (0, 1)$ be given. Let $$f(x) := \frac{x}{u}\left(\frac{1}{\sin x - u}-\frac{1}{\sin x + u}\right)-\frac{1}{x-u}-\frac{1}{x+u}$$ for $x\in (\arcsin{u}, \pi/2)$. Prove that $f$ is convex on $(\arcsin{u}, \pi/2)$.

Denote $c := \cos x, s := \sin x$. We have \begin{align*} f''(x) &= \frac{2xc^2}{u}\left(\frac{1}{(s-u)^3} - \frac{1}{(s+u)^3}\right)\\ &\quad + \frac{xs - 2c}{u}\left(\frac{1}{(s-u)^2} - \frac{1}{(s+u)^2}\right) - \frac{2}{(x-u)^3} - \frac{2}{(x+u)^3}. \end{align*}

It is easy to prove that $c \le 1 - \frac12 sx$ for all $x\in [0, \pi/2]$. We have \begin{align*} f''(x) &\ge \frac{2xc^2}{u}\left(\frac{1}{(s-u)^3} - \frac{1}{(s+u)^3}\right)\\ &\qquad + \frac{xs - 2\cdot (1 - \frac12 sx) }{u}\left(\frac{1}{(s-u)^2} - \frac{1}{(s+u)^2}\right)\\ &\qquad - \frac{2}{(x-u)^3} - \frac{2}{(x+u)^3}.\tag{1} \end{align*} From (1), to prove that $f''(x) \ge 0$, after clearing the denominators, it suffices to prove that $$q_4 v^4 + q_3 v^3 + q_2 v^2 + q_1v + q_0 \ge 0 \tag{2}$$ where $v := u^2$ and \begin{align*} q_4 &:= 3\,{s}^{2}x-2\,s+2\,x , \\ q_3 &:= {s}^{4}x-9\,{s}^{2}{x}^{3}+2\,{s}^{3}-12\,{s}^{2}x+6\,s{x}^{2}+4\,{x}^ {3}, \\ q_2 &:= -3\,{s}^{4}{x}^{3}+9\,{s}^{2}{x}^{5}+9\,{s}^{4}x-6\,{s}^{3}{x}^{2}+6\, {s}^{2}{x}^{3}-6\,s{x}^{4}-3\,{x}^{5},\\ q_1 &:= 3\,{s}^{4}{x}^{5}-3\,{s}^{2}{x}^{7}-3\,{s}^{6}x+3\,{s}^{4}{x}^{3}+6\,{ s}^{3}{x}^{4}-9\,{s}^{2}{x}^{5}+2\,s{x}^{6}+{x}^{7}, \\ q_0 &:= -{s}^{4}{x}^{7}-{s}^{6}{x}^{3}-2\,{s}^{3}{x}^{6}+3\,{s}^{2}{x}^{7}. \end{align*}

(2) is true for all $x\in [0, \pi/2]$ and $v \in [0, 1]$ which is verified (in several seconds) by Mathematica - a Computer Algebra System (CAS). A human verifiable proof is required.


About the verifying:

We use Mathematica to verify a stronger inequality: $$p_4 v^4 + p_3 v^3 + p_2 v^2 + p_1v + p_0 \ge 0$$ where $v \in [0, 1]$ and $x \in [0, 2]$ and $0 \le w \le 1$ and $$x - x^3/6 \le w \le x - x^3/6 + \frac{x^5}{120},$$ and \begin{align*} p_4 &:= 3\,{w}^{2}x-2\,w+2\,x , \\ p_3 &:= {w}^{4}x-9\,{w}^{2}{x}^{3}+2\,{w}^{3}-12\,{w}^{2}x+6\,w{x}^{2}+4\,{x}^ {3}, \\ p_2 &:= -3\,{w}^{4}{x}^{3}+9\,{w}^{2}{x}^{5}+9\,{w}^{4}x-6\,{w}^{3}{x}^{2}+6\, {w}^{2}{x}^{3}-6\,w{x}^{4}-3\,{x}^{5},\\ p_1 &:= 3\,{w}^{4}{x}^{5}-3\,{w}^{2}{x}^{7}-3\,{w}^{6}x+3\,{w}^{4}{x}^{3}+6\,{ w}^{3}{x}^{4}-9\,{w}^{2}{x}^{5}+2\,w{x}^{6}+{x}^{7}, \\ p_0 &:= -{w}^{4}{x}^{7}-{w}^{6}{x}^{3}-2\,{w}^{3}{x}^{6}+3\,{w}^{2}{x}^{7}. \end{align*} (Note: We use $x - x^3/6 \le w \le x - x^3/6 + \frac{x^5}{120}$ to avoid dealing with $\sin x$.)

f[x_, w_, v_] := (3w^2x - 2w + 2x)* v^4 + (w^4x - 9w^2x^3 + 2w^3 - 12w^2x + 6wx^2 + 4x^3) v^3 + (-3w^4x^3 + 9w^2x^5 + 9w^4x - 6w^3x^2 + 6w^2x^3 - 6wx^4 - 3x^5) v^2 + (3w^4x^5 - 3w^2x^7 - 3w^6x + 3w^4x^3 + 6w^3x^4 - 9w^2x^5 + 2wx^6 + x^7)v - w^4x^7 - w^6x^3 - 2w^3x^6 + 3w^2*x^7

pos1 := v >= 0 && v <= 1 && w >= x - x^3/6 && w <= x - x^3/6 + x^5/120 && x >= 0 && x <= 2 && w <= 1

Resolve[ForAll[{x, w, v}, pos1, f[x, w, v] >= 0], Reals]

River Li
  • 49,125
  • How is (2) verified by Mathematica? Can you show me the code? Thanks! – Hui Zhang Jun 29 '24 at 02:30
  • @HuiZhang Have you used the Resolve in Mathematica? – River Li Jun 29 '24 at 02:49
  • I do not know but I can learn. I mean: is that a real proof or just eyeball check? – Hui Zhang Jun 29 '24 at 02:50
  • @HuiZhang It is a verification that the inequality is true. It is not a human verifiable proof. See my answer in this question (in 2019, I used mathematica to verify the inequality, this year I gave a human verifiable proof): https://math.stackexchange.com/questions/2016364/inequality-a-b-c-d-subset0-1-rightarrow-sum-limits-cyca4a2b28-pr/3446031#3446031 – River Li Jun 29 '24 at 02:55
  • Thanks! That sounds very reliable. I am going to try it. Thanks again! – Hui Zhang Jun 29 '24 at 02:58
  • @HuiZhang When you are familiar with Resolve command (You can try some simple examples), we can discuss my answer here. For this question, we can not deal with $\sin x$ directly in Resolve. We use a trick. – River Li Jun 29 '24 at 03:00
  • It seems Resolve can not resolve much the inequality. I tried to use Resolve directly on $f''(x)>0$, $0<u<1$, $asin u< x<pi/2$ and got no answer. – Hui Zhang Jun 29 '24 at 06:29
  • Yes. Resolve[Exists[x,x*Sin[x]/2>1-Cos[x] && 0< x && x<Pi/2],Reals] worked. – Hui Zhang Jun 29 '24 at 06:35
  • I did not clear the denominators. – Hui Zhang Jun 29 '24 at 06:42
  • Thanks a lot! I found it worked! So finally the condition of $w^2>v$ i.e. $\sin x>u$ is not used?! – Hui Zhang Jun 29 '24 at 08:42
  • @HuiZhang We clear the denominators by multiplying something like $(\sin x - u)^3$ to get the last inequality which is true for all $u\in [0, 1]$ and $0\le x \le \pi/2$. By the way, to obtain a proof by hand, perhaps we can use bounds in the original expression of $f''(x)$. – River Li Jun 29 '24 at 09:15