9
$\begingroup$

Is it true that the smallest root $t$ of the polynomial

$$ 20 t^3 - 30 t^2 + (12 - 4 \cos^2 \alpha - 4 \cos^2 \beta - 4 \cos^2 \gamma) t + \cos^2 \alpha + \cos^2 \beta + \cos^2 \gamma - 2 \cos \alpha \cos \beta \cos \gamma - 1 $$

is always no greater than $\frac{1}{5}$ for all $\alpha, \beta, \gamma \in [0,\pi]$ with $\alpha + \beta \geq \gamma$, $\alpha + \gamma \geq \beta$, $\beta + \gamma \geq \alpha$, and $\alpha + \beta + \gamma \leq 2\pi$?

Thanks a lot for any helpful answer.

$\endgroup$
10
  • $\begingroup$ I have performed numerous computations but have not found any counterexample. $\endgroup$
    – Venus
    Commented Sep 15 at 5:29
  • 2
    $\begingroup$ This looks like a characteristic polynomial of a certain $3\times 3$ matrix, is not it arised this way? $\endgroup$ Commented Sep 15 at 5:56
  • $\begingroup$ @FedorPetrov It may not be completely that, but do you have any insights on finding the answer regarding the algebraic or analytic property of polynomials? Thanks. $\endgroup$
    – Venus
    Commented Sep 15 at 6:58
  • 2
    $\begingroup$ I don't understand the close votes. The problem does not seem quite trivial, and it could possibly have arisen in some research. $\endgroup$ Commented Sep 15 at 13:42
  • 11
    $\begingroup$ @IosifPinelis I think it's an excellent question and have certainly not downvoted/closevoted, but I suspect the folks who are are hoping for more context; it's clear that this polynomial comes from somewhere rather than appearing like Venus from the waves, and it would be helpful/interesting/etc. to know its origins. $\endgroup$ Commented Sep 15 at 19:37

3 Answers 3

12
$\begingroup$

This proof is similar to River Li's, but it requires no machine calculation. Also, we will show more, namely that the original polynomial has three real roots, the smallest of which is at most $$u:=\frac{5-\sqrt{15}}{10}=0.11270166\dotsc$$ This estimate is sharp, because in case of $\alpha=\beta=\gamma=\pi/2$, the polynomial is $$20t^3-30t^2+12t-1=20\left(t-\frac{1}{2}\right)\left(t-\frac{5-\sqrt{15}}{10}\right)\left(t-\frac{5+\sqrt{15}}{10}\right).$$ We turn to the proof. With the notation $$Q:=\frac{\cos^2\alpha+\cos^2\beta+\cos^2\gamma}{3} \qquad\text{and}\qquad R:=\cos\alpha\cos\beta\cos\gamma,$$ the polynomial in the original post equals $$p(t):=20t^3-30t^2+12(1-Q)t+(3Q-2R-1).$$ The coefficients of this cubic polynomial are $$a:=20,\qquad b:=-30,\qquad c:=12(1-Q),\qquad d:=3Q-2R-1,$$ hence its discriminant equals \begin{align*} \Delta\ := \ & 18 a b c d - 4 b^3 d + b^2 c^2 - 4 a c^3 - 27 a^2 d^2 \\ \ = \ & 2160 (1 + 12 Q + 3 Q^2 + 64 Q^3 - 60 Q R - 20 R^2). \end{align*} Here $2 QR\leq Q^2+R^2$, and also $R^2\leq Q^3$ by the AM-GM inequality, whence \begin{align*} \Delta/2160 \ & \geq \ 1 + 12 Q + 3 Q^2 + 64 Q^3 - 30 Q^2 - 50 R^2 \\ & \geq \ 1 + 12 Q - 27 Q^2 + 14 Q^3 \\ & = \ (Q-1)^2(1+14Q) \\ & \geq \ 0. \end{align*} So $p(t)$ has three real roots (counted with multiplicity), and the same is true of the shifted polynomial $$p(u+t)=20t^3-6\sqrt{15}t^2+6(1-2Q)t+\left(6\sqrt{\frac{3}{5}}Q-3Q-2R\right).$$ It remains to show that this new polynomial $p(u+t)$ has a nonpositive root $t$. If this is not the case, then $p(u+t)$ has three positive roots (counted with multiplicity), hence in particular $$1-2Q>0>6\sqrt{\frac{3}{5}}Q-3Q-2R.$$ However, this is a contradiction, because it would yield that $$Q^3\geq R^2>\left(3\sqrt{\frac{3}{5}}-\frac{3}{2}\right)^2 Q^2\geq\frac{1}{2}Q^2\geq Q^3.$$

$\endgroup$
4
  • 1
    $\begingroup$ Nice. It is a solution without calculus. $\endgroup$
    – River Li
    Commented Sep 16 at 5:38
  • $\begingroup$ @RiverLi Thank you. I have now sharpened the conclusion slightly. $\endgroup$
    – GH from MO
    Commented Sep 16 at 6:29
  • 1
    $\begingroup$ Nice, without complicated computation, also the condition is redundant. $\endgroup$
    – River Li
    Commented Sep 16 at 7:03
  • 1
    $\begingroup$ @RiverLi Thanks again. I further simplified the proof, now the original polynomial is only shifted once. $\endgroup$
    – GH from MO
    Commented Sep 16 at 8:44
7
$\begingroup$

Solution.

With the substitution $t = \frac15 + s$, the equation is written as $$f(s) := 20s^3 - 18s^2 + qs - r = 0, \tag{1}$$ where \begin{align*} q &:= \frac{12}{5} - 4 \cos^2 \alpha - 4 \cos^2 \beta - 4 \cos^2 \gamma, \\ r &:= - \frac{9}{25} - \frac15\cos^2 \alpha - \frac15 \cos^2 \beta - \frac15 \cos^2 \gamma + 2 \cos \alpha \cos \beta \cos \gamma. \end{align*}

First, let us prove that (1) has three real roots. It suffices to prove that discriminant of $f$ is non-negative, or equivalently $$g(\cos \alpha, \cos \beta, \cos \gamma) \ge 0,$$ where \begin{align*} g(a, b, c) &:= 64\,{a}^{6}+192\,{a}^{4}{b}^{2}+192\,{a}^{4}{c}^{2}+192\,{a}^{2}{b}^{4 }\\ &\qquad -156\,{a}^{2}{b}^{2}{c}^{2}+192\,{a}^{2}{c}^{4} +64\,{b}^{6}+192\,{b}^ {4}{c}^{2}\\ &\qquad +192\,{b}^{2}{c}^{4} +64\,{c}^{6}-540\,{a}^{3}bc -540\,a{b}^{3 }c\\ &\qquad -540\,ab{c}^{3}+9\,{a}^{4} +18\,{a}^{2}{b}^{2}+18\,{a}^{2}{c}^{2}+9\, {b}^{4}\\ &\qquad +18\,{b}^{2}{c}^{2}+9\,{c}^{4}+108\,{a}^{2}+108\,{b}^{2}+108\,{ c}^{2}+27. \end{align*} We can prove that $g(a, b, c) \ge 0$ for all real numbers $a, b, c$ (this is verified by Mathematica). The desired result follows.

Second, we prove that (1) does not have three positive real roots. Assume, for the sake of contradiction, that (1) has three positive real roots. Then we have $q > 0, r > 0$. However, we can prove that $\frac54 r + \frac{3}{16}q \le 0$ that is $$- \cos^2 \alpha - \cos^2 \beta - \cos^2 \gamma + \frac52 \cos \alpha \cos \beta \cos \gamma \le 0. \tag{2}$$ (2) is true since $$\cos^2 \alpha + \cos^2 \beta + \cos^2 \gamma \ge 3\sqrt[3]{\cos^2 \alpha \cos^2 \beta \cos^2 \gamma}$$ $$\ge 3|\cos \alpha \cos \beta \cos \gamma| \ge \frac52 \cos \alpha \cos \beta \cos \gamma.$$ Contradiction. Thus, the claim is proved.

Thus, (1) has at least one real root $s \le 0$.

We are done.

$\endgroup$
8
  • 2
    $\begingroup$ It is confusing to write $f(a,b,c)$ when you already use the notation $f(s)$. $\endgroup$
    – GH from MO
    Commented Sep 15 at 16:35
  • 1
    $\begingroup$ @GHfromMO Thanks. I will edit. $\endgroup$
    – River Li
    Commented Sep 15 at 23:35
  • $\begingroup$ Thank you for your answer. Are verification and proof the same? Since you said that you "can prove that $g(a, b, c) \ge 0$ for all real numbers $a, b, c$," could you please show the proof? Thanks. $\endgroup$
    – Venus
    Commented Sep 16 at 1:10
  • $\begingroup$ @Venus I think proof is referred to as a human verifiable proof of the inequalities (sometimes we use computer to motivate the proof, but the proof itself is verifiable by hand usually), while verification is to use computer to verify the inequalities without showing the steps. $\endgroup$
    – River Li
    Commented Sep 16 at 1:20
  • 1
    $\begingroup$ @RiverLi I added a simple proof that (1) has three real roots. See my response below. $\endgroup$
    – GH from MO
    Commented Sep 16 at 5:32
7
$\begingroup$

Write $a,b,c$ instead of $\alpha,\beta,\gamma$. Let $p(t)$ be the polynomial in question. Let $$u:=\cos^2a+\cos^2b+\cos^2c,\quad v:=\cos a\,\cos b\,\cos c,$$ $$t_*:=\frac{1}{30} \left(15-\sqrt{15} \sqrt{4 u+3}\right).$$ Then $$0\le u\le3,\quad v\le(u/3)^{3/2},$$ $$\begin{aligned}p(t_*)&=\frac{(4 u+3)^{3/2}}{3 \sqrt{15}}-u-2 v \\ &\ge\frac{(4 u+3)^{3/2}}{3 \sqrt{15}}-u-2 (u/3)^{3/2} \\ &=:g(u)\ge g(3)=0 \end{aligned}$$ (the latter inequality holds because $u=3$ is the only critical point of $g$ and $g(0)>0$),
$$t_*>1/5\implies u<3/5,$$ $$\begin{aligned}p(1/5)&=\frac{1}{25} (5 u-50 v+9) \\ &\ge\frac{1}{25} (5 u-50(u/3)^{3/2}+9) \\ &=:h(u)\ge\min(h(0),h(3/5)) \\ &>0 \text{ if }0\le u<3/5 \end{aligned}$$ (the 2nd inequality in the latter display holds because $h$ is concave). So, $p(\min(t_*,1/5))\ge0$.

Also, $p(-\infty+)=-\infty<0$.

Thus, $p(t)=0$ for some real $t\le1/5$. $\quad\Box$


Note that no conditions on $a,b,c$ (that is, on $\alpha,\beta,\gamma$) were needed or used here.

$\endgroup$
10
  • 2
    $\begingroup$ "Claims (1) and (2) can be verified algorithmically" -- technically, the initial claim can also be verified algorithmically (by writing all the conditions in terms of sines and cosines and using Tarski--Seidenberg). I think what you implicitly mean is that they "can be verified algorithmically before the heat death of the Universe". $\endgroup$ Commented Sep 15 at 14:49
  • $\begingroup$ @AlekseiKulikov : You are quite right. I have further simplified the expressions for $t_*$ and $p(t_*)$ to reduce the total Mathematica execution time to under 0.5 sec. $\endgroup$ Commented Sep 15 at 15:59
  • $\begingroup$ The proof is now much simpler and without using Mathematica. $\endgroup$ Commented Sep 15 at 18:34
  • $\begingroup$ The new argument looks extremely condensed, even the logic is hard to follow, it would benefit from a few english words I think. Anyway, if I understood it right, in the middle you claim that $g(u) \ge g(3)$ for $0 \le u \le 3$ -- why is that true? (you also claim that $h(u) \ge \max(h(0), h(3/5))$ which is just straight up false, but I assume you've meant $\min$ there and not $\max$, which I managed to convince myself in) $\endgroup$ Commented Sep 15 at 18:57
  • 1
    $\begingroup$ @AlekseiKulikov : Such a quest, with such a rather simple but apparently unremarkable function $g$, would look rather strange to me. $\endgroup$ Commented Sep 15 at 21:33

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .