39

Let $a$, $b$ and $c$ be positive numbers. Prove that: $$\frac{a^3}{13a^2+5b^2}+\frac{b^3}{13b^2+5c^2}+\frac{c^3}{13c^2+5a^2}\geq\frac{a+b+c}{18}$$

This inequality is strengthening of the following Vasile Cirtoaje's one, which he created in 2005.

Let $a$, $b$ and $c$ be positive numbers. Prove that: $$\frac{a^3}{2a^2+b^2}+\frac{b^3}{2b^2+c^2}+\frac{c^3}{2c^2+a^2}\geq\frac{a+b+c}{3}.$$

My proof of this inequality you can see here: https://artofproblemsolving.com/community/c6h22937p427220

But this way does not help for the starting inequality.

A big problem we have around the point $(a,b,c)=(0.785, 1.25, 1.861)$ because the difference between the LHS and the RHS in this point is $0.0000158...$.

I tried also to use Cauchy-Schwarz, but without success.

Also, I think the BW (see here https://math.stackexchange.com/tags/buffalo-way/info I tryed!) does not help.

  • 2
    Is the function $f : (a,b,c) \rightarrow \sum\limits_{cyc} \frac{6a^3}{13a^2+5b^2} $ concave? Also, C-S = "Cauchy-Schwarz inequality"? Avoid this kind of unconventional abbreviation, please. – Vincent May 08 '16 at 18:17
  • See also http://math.stackexchange.com/questions/1775572/olympiad-inequality-sum-cyc-fracx48x35y3-geqslant-fracxyz13 – David Quinn May 08 '16 at 18:42
  • @David Quinn my inequality is much harder. BW does not help here. – Michael Rozenberg Jan 30 '17 at 14:08
  • Since this inequality lacks convexity and is not stable under slight changes to its coefficients $(5,13)$, I don't expect there to be a "nice" proof.

    A computer algebra system reduces this problem to a system of polynomial equations and then solves this system by computing a Gröbner basis and numerically approximating the roots of polynomials with very high degrees.

    What kind of proof are you looking for?

    – cafaxo Apr 13 '17 at 14:23
  • @cafaxo In my life I proved many inequalities by hand, which very hard to prove even with computer. I am ready to show an example. Hence, I know, that I ask. – Michael Rozenberg Apr 13 '17 at 14:34
  • Hello Michael Rozenberg , I can reduce your problem in a problem with one variable . For it I use the software SageMath .All the proof need basic Tools (Gröbner basis is useless ) . I will post it in an another comment . The only matter is that the proof is ugly ...Have a good day. – max8128 Apr 18 '17 at 15:59
  • @max8128 Thank you for your interest. I believe that there is a nice proof for this inequality. – Michael Rozenberg Apr 18 '17 at 17:35
  • @MichaelRozenberg Hi! I'm curious, I ask respectfully, interested in what are you doing. Since I never encountered problems such as this in my math education. What's the deal with inequalities? How did you came up with that problem, is it a conjecture of yours? Does this came up of a sub-problem while trying to solve another problem? You seem to be an expert in inequalities from your profile. – Santropedro Apr 19 '17 at 17:22
  • 1
    @Santropedro There is the following known inequality $\sum\limits_{cyc}\frac{a^3}{a^2+b^2}\geq\frac{a+b+c}{2}$. I think, it's just interesting, for which maximal value of $k$ the following inequality is still true. $\sum\limits_{cyc}\frac{a^3}{ka^2+b^2}\geq\frac{a+b+c}{k+1}$. – Michael Rozenberg Apr 19 '17 at 17:28
  • 1
    @Santropedro The value $,k=2.6,$ in the OP is pretty sharp. I found that $k=2.603279$ is already beyond: Then Michael's given $(0.785, 1.25, 1.861)$ violates the ineq.. – Hanno May 11 '17 at 18:24
  • 1
    Further result obtained via random-based simulation: $(0.785, 1.2535, 1.873)$ doesn't satisfy the inequality if $,k=2.603262,$. Below that $k$-value I didn't detect a violation. – Hanno May 11 '17 at 18:50
  • @Hanno I don't understand why do you reply that to me, maybe you are directing to others? I don't understand what you are trying to convey me. – Santropedro May 12 '17 at 02:37
  • @Santropedro My numerical output refers to Michael R.'s last comment which is replying to yours. So I thought this might be interesting for you as well, hence the alert. (And Michael R. as OP is alerted anyway.) Beyond the existence of a 'nice proof' (for which math.SE is longing) the quest for the maximal $k$-value such that the ineq still holds true $\forall, a,b,c>0,$ is a quite interesting follow-up. – Hanno May 12 '17 at 05:57
  • @Hanno Ah ok thanks. – Santropedro May 12 '17 at 16:20
  • @Michael Rozenberg . Your inequality is an application of the Levinson's Inequality. – max8128 May 16 '17 at 10:52
  • @max8128 I don't think so. – Michael Rozenberg May 16 '17 at 11:18
  • @Michael Rozenberg what do you think about this? – max8128 May 24 '17 at 09:48
  • 1
    @max8128 I don't like it. I am sorry. – Michael Rozenberg May 24 '17 at 10:19
  • The general problem is here : https://artofproblemsolving.com/community/c6h20513p815736 – knm Jun 13 '17 at 18:56
  • @knm I knew about existing of this topic. Now we need to prove it! – Michael Rozenberg Jun 13 '17 at 18:58
  • @Vincent I drew the graph of $f(a,b,1)$ with Mathcad and it is not concave. – Alex Ravsky Aug 26 '17 at 13:02
  • @knm Indeed, it is always better to cite one's sources, if only for context. – Did Aug 26 '17 at 13:22
  • Dear @Did I created the starting inequality by myself, when solved the following problem https://artofproblemsolving.com/community/c6h22937p427220 Easy to understand that it was before the Ji Chen's generalization. By the way, the Ji Chen's problem is open, but my problem we'll can solve, I hope. – Michael Rozenberg Aug 28 '17 at 07:48
  • @MichaelRozenberg Do you believe me if I say we can use Jensen's inequality on this problem ? – Miss and Mister cassoulet char Dec 07 '19 at 13:27
  • I don't believe that it's possible – Michael Rozenberg Dec 07 '19 at 13:36
  • @MichaelRozenberg If you don' believe that we cannot use convexity and Karamata's inequality see here.It's not a complete proof but it's the first time I see Karamata's inequality and Buffalo's way together .Teach this to your pupils =) – Miss and Mister cassoulet char Mar 11 '20 at 09:19
  • Do you want to avoid using Lagrange multipliers? It wouldn't be so bad with just three variables. – Maxim Gilula Mar 15 '20 at 10:53
  • 1
    @Maxim Gilula No. If it goes to solution by hand, why no. – Michael Rozenberg Mar 15 '20 at 10:55
  • @MichaelRozenberg I tried for a few hours, but the best I could do by hand was reduce this problem to finding roots of a fourth degree polynomial. So unless you know the quartic formula, this idea might be useless :( – Maxim Gilula Mar 15 '20 at 14:53
  • Can this form of the problem be solved using determinants? – Cyriac Antony Mar 17 '20 at 15:55

8 Answers8

10

@Maxim Gilula asked a good question: What is the point of presenting a solution via computer?

Let me address it from two aspects.

First, from my experience (the problems from MSE, AoPS, academic research etc.), the inequalities with elegant solutions (e.g. by hand) are usually those designed e.g. for contest. Except for the designed inequalities, the inequalities require usually the help of computer. Many inequalities in MSE, AoPS are open for many years without any solution by hand.

Second, although also with the help of computer, the solutions are different.

1) Some solutions e.g. Buffalo Way (BW) or Sum of Squares (SOS) provide step-by-step rigorous complete analytical solutions with detailed explanation. Moreover, these solutions are often not very long. Usually one or several A4 pages are enough to write down the step-by-step rigorous complete analytical solutions with detailed explanation.

2) In contrast, some solutions does not provided step-by-step solutions. For example, the step-by-step solutions are in Mathematica's innards, and for some cases, Mathematica run several hours to output 'true' which means that the step-by-step solutions may be very, very long, probably 10,000 A4 pages are required to write down the step-by-step solutions. Sometimes the method of Lagrange Multiplier works well, however, sometimes it results in very complicated equations which requires Mathematica etc.

3) Often, although the thinking process is very complicated and requires the help of computer, the solution is elementary and easy to check even by hand. For example, recently, I used computer to solve an inequality in AoPS to get the identity $a^3 + b^3 + c^3 - a^2b - b^2c - c^2a = \frac{(a^2+b^2-2c^2)^2 + 3(a^2-b^2)^2 + \sum_{\mathrm{cyc}} 4(a+b)(c+a)(a-b)^2}{8(a+b+c)}$. In contrast, Mathematica only outputs 'true'. Which solution is better?

EDIT

Remark: Actually, the Buffalo Way works though the solution is ugly. I do not put it here. With computer, here is an SOS (Sum of Squares) solution:

WLOG, assume that $c = \min(a, b, c)$. We have \begin{align*} &\frac{a^3}{13a^2+5b^2} + \frac{b^3}{13b^2+5c^2} + \frac{c^3}{13c^2+5a^2} - \frac{a+b+c}{18}\\ =\ & \frac{1}{18(13a^2+5b^2)(13b^2+5c^2)(13c^2+5a^2)}\\ &\quad \cdot \frac{1}{2223}\Big(cz_1^TA_1 z_1 + c(a-c)(b-c)z_2^TA_2z_2 + (b-c)z_3^TA_3z_3 + (a-c)z_4^TA_4z_4\Big) \end{align*} (Remark: $A_1, A_2, A_3, A_4$ are all positive definite. So the inequality is true.) where $$z_1 = \left(\begin{array}{c} {\left(a - c\right)}^2\, \left(b - c\right)\\ \left(a - c\right)\, {\left(b - c\right)}^2\\ c\, {\left(a - c\right)}^2\\ c\, \left(a - c\right)\, \left(b - c\right)\\ c\, {\left(b - c\right)}^2\\ c^2\, \left(b - c\right)\\ c^2\, \left(a - c\right) \end{array}\right), \quad z_2 = \left(\begin{array}{c} {\left(a - c\right)}^2\\ \left(a - c\right)\, \left(b - c\right)\\ {\left(b - c\right)}^2\\ c\, \left(a - c\right)\\ c^2\\ c\, \left(b - c\right) \end{array}\right), $$ $$z_3 = \left(\begin{array}{c} {\left(a - c\right)}^2\, \left(b - c\right)\\ \left(a - c\right)\, {\left(b - c\right)}^2\\ c\, {\left(a - c\right)}^2\\ c\, \left(a - c\right)\, \left(b - c\right)\\ c\, {\left(b - c\right)}^2\\ c^2\, \left(b - c\right)\\ c^2\, \left(a - c\right) \end{array}\right) , \quad z_4 = \left(\begin{array}{c} {\left(a - c\right)}^2\, \left(b - c\right)\\ \left(a - c\right)\, {\left(b - c\right)}^2\\ c\, {\left(a - c\right)}^2\\ c\, \left(a - c\right)\, \left(b - c\right)\\ c\, {\left(b - c\right)}^2\\ c^2\, \left(b - c\right)\\ c^2\, \left(a - c\right) \end{array}\right),$$ and $$A_1 = \left(\begin{array}{ccccccc} 453245 & -100035 & 111397 & 113607 & -146718 & 24687 & 6498\\ -100035 & 166231 & -19773 & 24453 & -82004 & -11286 & 31356\\ 111397 & -19773 & 444600 & 86526 & -144261 & 13585 & -6669\\ 113607 & 24453 & 86526 & 760500 & -344736 & -126711 & 149188\\ -146718 & -82004 & -144261 & -344736 & 297882 & -26676 & -53352\\ 24687 & -11286 & 13585 & -126711 & -26676 & 160056 & -111150\\ 6498 & 31356 & -6669 & 149188 & -53352 & -111150 & 160056 \end{array}\right), $$ $$A_2 = \left(\begin{array}{cccccc} 281580 & 33098 & -64467 & 40261 & 4275 & -24219\\ 33098 & 329004 & -124982 & 13572 & 20241 & -51129\\ -64467 & -124982 & 106704 & -51129 & -6840 & -30875\\ 40261 & 13572 & -51129 & 326610 & -8645 & -79794\\ 4275 & 20241 & -6840 & -8645 & 62244 & -46683\\ -24219 & -51129 & -30875 & -79794 & -46683 & 153216 \end{array}\right) ,$$ $$A_3 = \left(\begin{array}{ccccccc} 258609 & -100035 & 102011 & 75582 & -56069 & -57798 & 126945\\ -100035 & 55575 & -37791 & -29393 & 2223 & 26923 & -26676\\ 102011 & -37791 & 589342 & 224757 & -195624 & -155376 & 313272\\ 75582 & -29393 & 224757 & 693576 & -279851 & -320283 & 535977\\ -56069 & 2223 & -195624 & -279851 & 200070 & 91143 & -306945\\ -57798 & 26923 & -155376 & -320283 & 91143 & 293436 & -435708\\ 126945 & -26676 & 313272 & 535977 & -306945 & -435708 & 1016158 \end{array}\right) ,$$ $$A_4 = \left(\begin{array}{ccccccc} 144495 & -57057 & 3705 & 24206 & -22230 & 2457 & 26923\\ -57057 & 55575 & 4199 & -15561 & -22724 & 13338 & -1989\\ 3705 & 4199 & 200070 & 18031 & -46449 & -9063 & 17784\\ 24206 & -15561 & 18031 & 351468 & -140049 & -6435 & -13509\\ -22230 & -22724 & -46449 & -140049 & 189202 & -117990 & -46449\\ 2457 & 13338 & -9063 & -6435 & -117990 & 177840 & -48659\\ 26923 & -1989 & 17784 & -13509 & -46449 & -48659 & 253422 \end{array}\right). $$

River Li
  • 37,323
  • If you are allowed to use a computer, there are 100 different solutions. What is the point of presenting a solution via computer? Might as well copy and paste wolfram alpha "minimize f(x,y,z)." This doesn't deserve 100 bounty -- no answer here does. Lagrange may be less work and can be actually done by hand if you somehow memorized how to solve quartic equations and doesn't involve a computer, or once you're at that step use a computer. – Maxim Gilula Mar 18 '20 at 20:17
  • @MaximGilula Thank you for comments. Let me reply in the beginning of the solution because the comments are too long. – River Li Mar 19 '20 at 01:03
  • Thanks for your useful remarks! – Maxim Gilula Mar 19 '20 at 01:17
  • @MaximGilula Welcome to discuss. This inequality is waiting for elegant solutions (e.g. by hand) for many years. – River Li Mar 19 '20 at 01:23
  • I have found this link https://mathworld.wolfram.com/WitchofAgnesi.html . Where we recognize (the cartesian equation) an element of the LHS.So do you think there exists a geometric proof of this fact ?Thanks. – Miss and Mister cassoulet char Apr 20 '20 at 09:49
  • @The.old.boy Interesting. You found that it has some geometric meaning. – River Li Apr 20 '20 at 10:28
  • @RiverLi How did you find those matrices? – didgogns Aug 04 '22 at 05:16
  • 1
    @didgogns We used sostools+Matlab to find them. – River Li Aug 04 '22 at 05:26
3

A bit algebra shows that the inequality is equivalent to $$ 5 \left(5 a^5 \left(13 b^2+5 c^2\right)+13 a^4 \left(5 b^3-13 b^2 c-5 b c^2-5 c^3\right)+a^3 \left(-65 b^4+144 b^2 c^2+65 c^4\right)+a^2 \left(25 b^5-65 b^4 c+144 b^3 c^2+144 b^2 c^3-169 b c^4+65 c^5\right)-13 a \left(13 b^4 c^2+5 b^2 c^4\right)+5 b^2 c^2 \left(13 b^3+13 b^2 c-13 b c^2+5 c^3\right)\right) \ge0 $$ The left hand side is a polynomial, so this can be solved by Cylindrical Algebra Decomposition -- https://en.wikipedia.org/wiki/Cylindrical_algebraic_decomposition

The following code in Mathematica does the job.

ex1 = a^3/(13 a^2 + 5 b^2) + b^3/(13 b^2 + 5 c^2) + c^3/(5 a^2 + 13 c^2) >= 1/18 (a + b + c);
ex2 = ex1[[1]] - ex1[[2]] // Together // Numerator // Simplify;
ex3 = ForAll[{a, b, c}, And @@ {a >= 0, b >= 0, c >= 0}, ex2 >= 0];
CylindricalDecomposition[ex3, {}]

Note this may take a few minutes to run.

1

Your inequality is equivalent to : $$\sum_{cyc}\frac{a}{13}\sin(\arctan(\sqrt{\frac{13}{5}}\frac{a}{b}))^2\geq \frac{a+b+c}{18}$$ Each side is divided by $b$, We get: $$\frac{a}{13b} \sin(\arctan(\sqrt{\frac{13}{5}}\frac{a}{b}))^2+\frac{1}{13}\sin(\arctan(\sqrt{\frac{13}{5}}\frac{b}{c}))^2+\frac{c}{13b}\sin(\arctan(\sqrt{\frac{13}{5}}\frac{c}{a}))^2\geq \frac{1+\frac{a}{b}+\frac{c}{b}}{18}$$ Now we put $\sqrt{\frac{13}{5}}\frac{a}{b}=x$, $\sqrt{\frac{13}{5}}\frac{b}{c}=y$, $\sqrt{\frac{13}{5}}\frac{c}{a}=z$, your inequality is equivalent to: $$\sqrt{\dfrac{5}{13}}\frac{1}{13}\dfrac{(x)^3}{(x^2+1)}+\dfrac{1}{13}\dfrac{(y)^2}{(y^2+1)}+\sqrt{\dfrac{13}{5}}\dfrac{1}{13(y)}\dfrac{(z)^2}{(z^2+1)}$$$$\geq \dfrac{1+\sqrt{\dfrac{5}{13}}x+\sqrt{\dfrac{13}{5}}\dfrac{1}{y}}{18}$$ with the condition $xyz=(\sqrt{\frac{13}{5}})^3$.

We study the following function: $$f(x)=\sqrt{\dfrac{5}{13}}\frac{1}{13}\dfrac{(x)^3}{(x^2+1)}+\dfrac{1}{13}\dfrac{(y)^2}{(y^2+1)}+\sqrt{\dfrac{13}{5}}\dfrac{1}{13(y)}\dfrac{(z)^2}{(z^2+1)}-\dfrac{1+\sqrt{\dfrac{5}{13}}x+\sqrt{\dfrac{13}{5}}\dfrac{1}{y}}{18}$$ This function is easily differentiable and the minimum is for $x=\sqrt{\frac{3\sqrt{77}}{17}-\frac{26}{17}}=\alpha$. So with the condition $xyz=(\sqrt{\frac{13}{5}})^3$ becomes $yz=\frac{(\sqrt{\frac{13}{5}})^3}{\alpha}=\beta$. So we have this inequality just with $y$: $$\sqrt{\dfrac{5}{13}}\frac{1}{13}\dfrac{(\alpha)^3}{(\alpha^2+1)}+\dfrac{1}{13}\dfrac{(y)^2}{(y^2+1)}+\sqrt{\dfrac{13}{5}}\dfrac{1}{13(y)}\dfrac{(\frac{\beta}{y})^2}{((\frac{\beta}{y})^2+1)}\geq\dfrac{1+\sqrt{\dfrac{5}{13}}\alpha+\sqrt{\dfrac{13}{5}}\dfrac{1}{y}}{18}$$ which is easily analyzable. Done!

Ѕᴀᴀᴅ
  • 34,263
max8128
  • 652
  • Actually, what he did is to transform the problem to $F(x, y, z) = g(x) + h(y, z) \ge 0$ with constraints $xyz = (\sqrt{\frac{13}{5}})^3$ and $x, y, z > 0$. Here, $g(x) = \sqrt{\tfrac{5}{13}}\frac{1}{13}\tfrac{(x)^3}{(x^2+1)}
    • \tfrac{1}{18}\sqrt{\tfrac{5}{13}}, x$.
    – River Li Mar 18 '20 at 10:43
  • Yes, it seems "$x$" is separated. Then he minimized $g(x)$ at $x = \alpha = \sqrt{\frac{3\sqrt{77}}{17} - \frac{26}{17}}$ (this is not true). With $x = \alpha$, then he wanted to prove $g(\alpha ) + h(y, z) \ge 0$ with the constraints $yz = (\sqrt{\frac{13}{5}})^3/\alpha = \beta$ and $y, z > 0$. Finally, with $z = \frac{\beta}{y}$, he wanted to prove $g(\alpha) + h(y, \frac{\beta}{y}) \ge 0$ with the constraint $y > 0$. However, the mistake arises when he minimized $g(x)$. – River Li Mar 18 '20 at 10:43
  • @RiverLi I was saying 2 upvotes is 2 too many... Wrong or not. And now there are 4! – Maxim Gilula Mar 18 '20 at 20:02
  • @MaximGilula Yes, I just agree with you. – River Li Mar 19 '20 at 01:05
  • @RiverLi I see! – Maxim Gilula Mar 19 '20 at 01:15