Let $a, b, c \geq 0$ and satisfy \[ a^2+b^2+c^2 +abc = 4 . \] Show that \[ 0 \le ab + bc + ca - abc \leq 2. \]
Problem
Source: USAMO 2001 #3
Tags: inequalities, symmetry, trigonometry, AMC, USA(J)MO, USAMO, quadratics
27.06.2004 23:13
cn2_71828182846 wrote: Can someone explain a solution to USAMO 2001 problem 3? I am fine on the left inequality, but I'm a little hazy on the right-hand one. Thanks for assistance. Let a, b, c, be :ge: 0 and satisfy a :^2: + b :^2: + c :^2: + abc = 4. Show that 0 :le: ab + bc + ca - abc :le: 2. I'm not so sure but I'll give it a shot. -abc = a :^2: + b :^2: + c :^2: - 4 0 :le: ab + bc + ca + a :^2: + b :^2: + c :^2: - 4 :le: 2 4 :le: ab + bc + ca + a :^2: + b :^2: + c :^2: :le: 6 4 :le: [(a+b):^2: + (b+c):^2: + (a+c):^2:]/2 :le: 6 8 :le: (a+b):^2: + (b+c):^2: + (a+c):^2: :le: 12 That's all I've gotten so far, I'll work on it later.
28.06.2004 00:06
I think I got the first part If one of a,b,c is 0, we are done. So assume a,b,c >0 By AM-HM(or cauchy): \frac{1}{a}+\frac{1}{b}+\frac{1}{c} \geq \frac{9}{a+b+c} By the equation, a, b, c :le: 2. So a+b+c can be at most 6. Therefore \frac{9}{a+b+c} \geq 1 and \frac{1}{a}+\frac{1}{b}+\frac{1}{c} \geq 1 multiply both side by abc gives us the first inequality.
28.06.2004 00:19
For the left side, just notice that not all of the variables can be >1, so WLOG assume a<=1. Then, bc(1-a)+a(b+c) >= 0 since each term is non-negative. It's the right side that is difficult. I have a solution, but it isn't entirely mine so I feel guilty posting it.
28.06.2004 00:36
All I have for the second inequality: a^2+b^2 \geq 2ab b^2+c^2 \geq 2bc a^2+c^2 \geq 2ac add all together and divide by 2 a^2+b^2+c^2 \geq ab+bc+ac so ab+bc+ac+abc \leq 4 ab+bc+ac-abc \leq 4-2abc I will try to do figure it out later.
28.06.2004 08:11
following off beta's work (i'm not quite sure if this works): The last inequality holds true if 2abc \ge 2 abc \ge 1 \sqrt[3]{abc} \ge 1 By GM-HM, we can substitute to get \displaystyle \frac{3}{\frac{1}{a}+\frac{1}{b}+\frac{1}{c}} \ge 1 3 \ge \frac{1}{a}+\frac{1}{b}+\frac{1}{c} 3abc \ge ab+bc+ca From cauchy or AM-GM, we find a^2+b^2+c^2 \ge ab+bc+ca So we substitute again 3abc \ge a^2+b^2+c^2 Sorta guessing, but I think it can be proved that as long as the condition is satisfied, this inequality holds true, but i dont know how to do that.
28.06.2004 15:40
paladin8 wrote: 3abc \ge ab+bc+ca From cauchy or AM-GM, we find a^2+b^2+c^2 \ge ab+bc+ca So we substitute again 3abc \ge a^2+b^2+c^2 That' not true, you can't substitute like that.
28.06.2004 20:48
Yes, it works. If this is proved: 3abc \ge a^2+b^2+c^2 and this is true: a^2+b^2+c^2 \ge ab+bc+ca then this is true: 3abc \ge ab+bc+ca which is what we wish to prove.
28.06.2004 21:01
ok, i see. I thought you proved 3abc \ge ab+bc+ca with these two steps, not realizing that you are working backwards a^2+b^2+c^2 \ge ab+bc+ca 3abc \ge a^2+b^2+c^2
28.06.2004 21:11
beta wrote: All I have for the second inequality: a^2+b^2 \geq 2ab b^2+c^2 \geq 2bc a^2+c^2 \geq 2ac add all together and divide by 2 a^2+b^2+c^2 \geq ab+bc+ac so ab+bc+ac+abc \leq 4 ab+bc+ac-abc \leq 4-2abc I will try to do figure it out later. I don't think my idea is going to lead anywhere. (It's too strong.) To prove this statement to be true, we need abc \ge 1. However, by applying AM-GM to a^2 + b^2 + c^2 + abc = 4, \displaystyle{{a^2 + b^2 + c^2 + abc}\over{4}} \ge \sqrt[4]{a^3b^3c^3} From which we obtain abc \le 1. Hmm, thinking on then... Paladin: your steps don't hold. Try a = 2, b = c = 0.
28.06.2004 22:12
Some more work One of a, b, c must be < or =1, one must be > or =1. Let a=1+x, let b=1-y. , where x, y \leq 1 so (1+x)(1-y)c \leq 1 (1+x)^2+(1-y)^2+c^2+(1+x)(1-y)(c)=4 since x^2+y^2+c^2-xyc \geq 0 because xy+yc+xc-xyc=xy+yc+xc(1-y) \geq 0 so we have (2+c)(1+x-y) \leq 4
29.06.2004 01:11
Quote: Paladin: your steps don't hold. Try a = 2, b = c = 0. I assumed a, b, c > 0 because it is trivial to prove if any are 0 as noted by beta earlier on.
29.06.2004 01:54
Let a = 1.999, b = c = 0.01. It still doesn't work.
29.06.2004 04:25
that doesnt satisfy the original condition, but i have also found a counterexample, (1.99, .1, .1)
29.06.2004 04:55
It doesn't?
29.06.2004 21:22
1.999 :^2: +.01 :^2: +.01 :^2: +1.999*.01*.01 = 3.9964009.
29.06.2004 21:34
Close enough. In any case, beta's assertion is an inequality that can't be proven. So we have to think about another way.
31.10.2005 15:45
To prove this problem. We have to prove that, $ab+bc+ca-abc \leq 2 =\frac{a^2+b^2+c^2+abc}{2}\\ \Longleftrightarrow 2(ab+bc+ca) \leq a^2+b^2+c^2+3abc$
01.11.2005 03:36
Well, let's see... By AM-GM, we have $a^2+b^2+c^2+abc\ge 4\sqrt[4]{(abc)^3}\Rightarrow 4\ge 4\sqrt[4]{(abc)^3}\Rightarrow 0\le abc\le 1$ (since $a,b,c\ge 0$). From there, we have $4\ge a^2+b^2+c^2\ge 3$. By AM-GM, $ab+ac+bc\ge 3\sqrt[3]{(abc)^2}$, so $ab+ac+bc\le 3$. All that's left to prove is that $ab+ac+bc\le 3abc$, but I'm not sure how to do that.
03.11.2005 01:43
By the symmetry of the given equation, without loss of generality, $0\leq c\leq b\leq a,$ we have $4=a^2+b^2+c^2+abc\leq a^2+a^2+a^2+a\cdot a\cdot a\cdot\Longleftrightarrow a^3+3a^2-4\geq 0$ $\Longleftrightarrow (a-1)(a+2)^2\geq 0\Longleftrightarrow a\geq 1$,similarly we have $b\geq 1, c\geq1.$ Thus $4-2(ab+bc+ca-abc)=(a^2+b^2+c^2+abc)-2(ab+bc+ca-abc)$ $=a^2+b^2+c^2-2(ab+bc+ca)+3abc$ $\geq 3abc-(ab+bc+ca)\ (\because a^2+b^2+c^2\geq ab+bc+ca)$ $=abc+abc+abc-ab-bc-ca=ab(c-1)+bc(a-1)+ca(b-1)\geq 0$ $\Longleftrightarrow ab+bc+ca-abc\leq 2.$ Q.E.D.
17.02.2015 17:42
Another way to solve the problem : We may take $a=\frac{2x}{\sqrt{(x+y)(x+z)}},b=\frac{2y}{\sqrt{(y+z)(y+x)}},c=\frac{2z}{\sqrt{(z+x)(z+y)}}$ where $x,y,z$ are positive real numbers.
06.03.2016 20:31
For the first inequality, observe that by AM-GM it is enough to show that $ab+bc+ca \ge 3 \cdot (abc)^{\frac{2}{3}} \ge abc$. That is $27 \ge abc$, which is clear. For the second inequality, write $a=2 \cos A, b=2 \cos B, c=2 \cos C$ where $A+B+C=\pi$, thanks to the hypothesis and the identity $\cos^2 A+\cos^2 B+\cos^2 C+2 \cos A \cos B \cos C=1$. Rewrite the desired as \[ \sum_{cyc} 2 \cos A \cos B + \sum_{cyc} \cos^2 A - 1 = \sum_{cyc} (\cos A + \cos B)^2 - 1 \le 2 \]Subtracting $\cos^2 A + \cos^2 B + \cos^2 C$ from both sides, this becomes \[ (\cos A + \cos B + \cos C)^2 \le \sin^2 A + \sin^2 B + \sin^2 C \]Now we will prove two trig identities. Lemma 1: $\sum_{cyc} \cos A \sin B \sin C = \frac{1}{2} (\sin^2 A + \sin^2 B + \sin^2 C)$ where $A+B+C = \pi$. Proof: \begin{align*} \cos A \sin B \sin C &= \frac{1}{2} \cos A (\cos (B-C) + \cos A) \\ &= \frac{1}{2} \cos^2 A + \frac{1}{4} (\cos(A+B-C)+\cos(A+C-B)) \\ &= \frac{1}{2} \cos^2 A - \frac{1}{4} (\cos 2B + \cos 2C) \\ &= \frac{1}{2} \cos^2 A + \frac{1}{2} \sin^2 B + \frac{1}{2} \sin^2 C - \frac{1}{2} \\ &= \frac{1}{2} (\sin^2 B + \sin^2 C - \sin^2 A) \end{align*}Sum cyclically. Lemma 2: $\sin 2A + \sin 2B + \sin 2C = 4\sin A \sin B \sin C$ where $A+B+C = \pi$. Proof: \begin{align*} \sin 2A + \sin 2B + \sin 2C &= 2 \sin (A+B) \cos (A-B) + \sin 2C \\ &= 2 \sin C (\cos (A-B) + \cos C) \\ &= 2 \sin C (\cos(A-B) - \cos (A+B)) \\ &= 4 \sin A \sin B \sin C \end{align*}As a simple consequence, we have that \begin{align*} \sum_{cyc} \frac{\cos A}{\sin B \sin C} &= \frac{\sin A \cos A + \sin B \cos B + \sin C \cos C}{\sin A \sin B \sin C} \\ &= \frac{\sin 2A + \sin 2B + \sin 2C}{2 \sin A \sin B \sin C} \\ &= 2 \end{align*} Now we apply Cauchy to finish the proof, combined with Lemmas 1 and 2: \[ \sin^2 A + \sin^2 B + \sin^2 C = (\sum_{cyc} \cos A \sin B \sin C)(\sum_{cyc} \frac{\cos A}{\sin B \sin C}) \ge (\cos A + \cos B + \cos C)^2 \]We are done.
30.09.2017 17:10
v_Enhance wrote: Case 3: Now suppose that $a=b \neq c$. That means $\lambda = \frac{1}{2-c}$ Since we have that \[ (a-b)(\left[ 2\lambda - 1 \right] - \lambda c) = 0. \]The result quoted seems to point to the fact that we need $(\left[ 2\lambda - 1 \right] - \lambda c)$ to equal zero, which confuses me. Isn't $(a-b)(\left[ 2\lambda - 1 \right] - \lambda c)$ already zero because $a=b$? Also, how does this work: v_Enhance wrote: Now \[ 2a+2c = a+b+2c = \lambda (2c+ab) = \frac{1}{2-c} (2c+a^2) \]which implies \[ 4a+4c-2a^2-2ac = 2c+a^2 \] Should it be $\lambda=\frac{1}{2-a}$? A final more general question about process, what was the motivation for letting having the condition for $U$ be $a^2+b^2+c^2<1000$? Why not let it be say $999$ or $1001$?. Is there anything specific about $1000$, or is it just a matter of choosing a number that is sufficiently large?
01.10.2017 03:50
$\lambda = \frac{1}{2-a}$ was meant in both places. Thanks, I'll fix it. There's nothing special about $1000$ of course: as you say any large number suffices. The point is to capture the constraint set in some compact set, for which the boundary ($a^2+b^2+c^2=1000$ in this case) is a non-issue.
31.05.2019 00:48
If $a,b,c\ge0$ such that $a^2+b^2+c^2+abc=4$, then $$a+b+c\le\sqrt{8+abc}\qquad$$
05.08.2019 11:53
v_Enhance wrote: OK let's practice bashing. The left-hand side of the inequality is trivial; just note that $\min \left\{ a,b,c \right\} \le 1$. Hence, we focus on the right side. We use Lagrange Multipliers. Define \[ U = \left\{ (a,b,c) \mid a,b,c > 0 \text{ and } a^2+b^2+c^2 < 1000 \right\}. \]This is an intersection of open sets, so it is open. Its closure is \[ \overline U = \left\{ (a,b,c) \mid a,b,c \ge 0 \text{ and } a^2+b^2+c^2 \le 1000 \right\}. \]Hence the constraint set \[ \overline S = \left\{ \mathbf x \in \overline U : g(\mathbf x) = 4 \right\} \]is compact, where $g(a,b,c) = a^2+b^2+c^2+abc$. Excellent. Define \[ f(a,b,c) = a^2+b^2+c^2+ab+bc+ca. \]It's equivalent to show that $f \le 6$ subject to $g$. Over $\overline S$, it must achieve a global maximum. Now we consider two cases. If $\mathbf x$ lies on the boundary, that means one of the components is zero (since $a^2+b^2+c^2=1000$ is clearly impossible). WLOG $c=0$, then we wish to show $a^2+b^2+ab \le 6$ for $a^2+b^2=4$, which is trivial. Now for the interior $U$, we may use the method of Lagrange Multipliers. Consider a local maximum $\mathbf x \in U$. Compute \[ \nabla f = \left<2a+b+c, 2b+c+a, 2c+a+b \right> \]and \[ \nabla g = \left<2a+bc, 2b+ca, 2c+ab\right>. \]Of course, $\nabla g \neq \mathbf 0$ everywhere, so introducing our multiplier yields \[ \left<2a+b+c,a+2b+c,a+b+2c\right> = \lambda \left<2a+bc,2b+ca,2c+ab\right>. \]Note that $\lambda \neq 0$ since $a,b,c > 0$. Subtracting $2a+b+c = \lambda(2a+bc)$ from $a+2b+c = \lambda(2b+ca)$ implies that \[ (a-b)(\left[ 2\lambda - 1 \right] - \lambda c) = 0. \]We can derive similar equations for the others. Hence, we have three cases. Case 1: If $a=b=c$, then $a=b=c=1$, and this satisfies $f(1,1,1) \le 6$. Case 2: If $a$, $b$, $c$ are pairwise distinct, then we derive $a = b = c = 2 - \lambda^{-1}$, contradiction. Case 3: Now suppose that $a=b \neq c$. That means $\lambda = \frac{1}{2-a}$ (of course our conditions force $c < 2$). Now \[ 2a+2c = a+b+2c = \lambda (2c+ab) = \frac{1}{2-a} (2c+a^2) \]which implies \[ 4a+4c-2a^2-2ac = 2c+a^2 \]meaning (with the additional note that $a \neq 1$) we have \[ c = \frac{3a^2-4a}{2-2a}. \]Note that at this point, $c > 0$ forces $1 < a < \frac 43$. The constraint $a^2+b^2+c^2+abc=4 \iff c^2 + + a^2c + (2a^2-4) = 0$ now gives \[ \left( 3a^2-4a \right)^2 + a^2 \left( 3a^2-4a \right)\left( 2-2a \right) + \left( 2a^2-4 \right)\left( 2-2a \right)^2 = 0. \]Before expanding this, it is prudent to see if it has any rational roots. A quick inspection finds that $a=2$ is such a root (precisely, $16-32+16=0$). Now, we can expand and try to factor: \begin{align*} 0 &= -6a^5 + 31a^4 - 48a^3 + 8a^2 + 32a - 16 \\ &= (a-2)(-6a^4 + 19a^3 - 10a^2 - 12a + 8) \\ &= (a-2)^2 \left( -6a^3+7a^2+4a-4 \right) \\ \intertext{Seeing a cubic, we desperately try every rational root and miraculously discover} &= (a-2)^2 (2-3a)(2a^2-a-2). \end{align*}The only root $a$ in the interval $\left( 1,\tfrac43 \right)$ is $a = \frac 14 \left( 1 + \sqrt{17} \right)$. You can guess what comes next -- write \[ c = \frac{a(3a-4)}{2-2a} = \frac{1}{8} \left( 7 - \sqrt{17} \right) \]and \[ f(a,b,c) = 3a^2+2ac+c^2 = \frac{1}{32} \left( 121 + 17\sqrt{17} \right) . \]This is the last critical point, so we're done once we check this is less than $6$. This follows from the inequality $17^3 < (6 \cdot 32 - 121)^2$; in fact, we actually have \[ \frac{1}{32} \left( 121 + 17\sqrt{17} \right) \approx 5.97165. \]This completes the solution. In case 3 , can anyone tell how c<2 is concluded?
06.08.2019 19:35
Anyone ??
06.08.2019 19:38
$c\ge2 \Longrightarrow c^2 \ge 4 \Longrightarrow a^2 + b^2 + c^2 + abc >4,$ contradiction.
18.05.2020 17:31
cn2_71828182846 wrote: Let $a, b, c \geq 0$ and satisfy \[ a^2+b^2+c^2 +abc = 4 . \]Show that \[ 0 \le ab + bc + ca - abc \leq 2. \] 2017 Preparation for juniors, Romania: Let $a, b, c > 0$ and satisfy \[ a^2+b^2+c^2 = 3. \]Show that \[ ab + bc + ca - abc \leq 2. \]
19.05.2020 14:02
Let $a \leq 1$ and $b,c \geq 1$. Now, obviously $(b-1)(c-1) \geq 0$ and: $ab+bc+ca-abc=a(b+c)+bc(1-a) \geq 0$ We will solve quadratic equation $a^2+b^2+c^2+abc-4=0$ for $a$, so: $a=\frac{\sqrt{(4-b^2)(4-c^2)}-bc}{2}$. And we will use this in second equation: $ab+bc+ca-abc=-abc+ab+ac-a+a+bc=-a(b-1)(c-1)+a+bc \leq a+bc=\frac{bc+\sqrt{(4-b^2)(4-c^2)}}{2}$ By Cauchy-Schwarz inequality we see that: $\frac{bc+\sqrt{(4-b^2)(4-c^2)}}{2} \leq \frac{\sqrt{(4-b^2+b^2)(4-c^2+c^2)}}{2}=\frac{\sqrt{16}}{2}=2$ So, $0 \leq ab+bc+ca-abc \leq 2$
07.02.2021 19:02
It is well-known that we can substitute $a=2\cos A,b=2\cos B,c=2\cos C$ for a triangle $ABC$. Observe that \[ab+bc+ca\ge 3\sqrt[3]{a^2b^2c^2}\ge abc\]because $abc\le 27$, which shows the first part of the inequality. For the latter part, rewrite it as showing \[3\ge \sum_{\mbox{cyc}}(\cos A+\cos B)^2.\]Note since triangle $ABC$ is not obtuse, we can let $a^2=y+z,b^2=z+x,c^2=x+y$ for some nonnegative $x,y,z$, so the problem amounts to showing \[3\ge \sum_{\mbox{cyc}}\left(\frac{b^2+c^2-a^2}{2bc}+\frac{a^2+c^2-b^2}{2ac}\right)^2 = 2\sum_{\mbox{cyc}}\frac{x^2}{(x+y)(x+z)}+2\sum_{\mbox{cyc}}\frac{yz}{(y+z)\sqrt{(x+z)(x+y)}}.\]By rearranging, the problem amounts to checking \[6xyz+3\sum_{\mbox{sym}}x^2y\ge 2\sum_{\mbox{cyc}}x^2(y+z)+2\sum_{\mbox{cyc}}yz\sqrt{(x+z)(x+y)}.\]By cancelling, it is equivalent to show \[6xyz+\sum_{\mbox{sym}}x^2y\ge 2\sum_{\mbox{cyc}}yz\sqrt{(x+z)(x+y)}.\]But by AM-GM, the latter expression is at most \[\sum_{\mbox{cyc}}(2xyz+y^2z+z^2y),\]so done.
02.05.2021 05:47
For the lower bound, write $0 \le ab + bc + ca - abc$ as \[0\le -a((b-1)(c-1)-1)+bc.\]This is clearly true since $(b-1)(c-1)-1$ must be nonpositive. Now we deal with the upper bound. WLOG, we can let $a-1$ and $b-1$ have the same sign. Then $(a-1)(b-1)=ab-b-a+1 \ge 0 \implies ab+1 \ge a+b$. Multiplying by $c$ and subtracting $abc$, we get $c \ge ac+bc-abc$. Adding $ab, $ \[c+ab \ge ac+bc+ab-abc.\]This means to prove the upper bound it suffices to prove $c+ab \le 2$. Assume FTSOC that $c+ab>2$. Then, \begin{align*} 4 &=a^2+b^2+c(c+ab) \\ &\ge 2ab+c(c+ab) \\ &> 2ab+2c \end{align*} which implies that $2>ab+c,$ contradiction. Hence, $c+ab \le 2$ and we are done.
08.07.2021 17:56
My solution: We know that $min\{a,b,c\}\leq1$ so wlog any two of $a,b,c$ can be $\geq1$ or $\leq1$ So let $a\leq 1$ and $b,c\geq 1$ so we have \[ab+bc+ca-abc=a(b+c)+bc(1-a)\geq 0\]So lower bound is proved.So now we need to prove upper bound so \[a^2+b^2+c^2+abc=4\]and we know that $b^2+c^2\geq 2bc$ by AM-GM so now put this up\[a^2+2bc\leq 4\implies bc(2+a)\leq 4-a^2=(2+a)(2-a)\\\implies bc\leq 2-a\]now lets see what we have until now, $\bullet$ $(1-b)(1-c)\geq 0$ for both the cases $\bullet$ $bc\leq 2-a$ So now begin the sol lets see what we have \[ab+bc+ca-abc\leq ab+2-a+ca-abc=2-a(1-b-c+bc)=2-a(1-b)(1-c)\leq 2\]So we are done now !
23.04.2022 12:28
Part 1: $0\leq ab+bc+ca.$ Notice $$\min (a,b,c) \leq 1 \implies ab+bc+ca \geq bc \geq abc.$$Equality when $a=b=0, c=2.$ Q.E.D. Part 2: $ab+bc+ca-abc\leq 2.$ WLOG, $a\leq 1, c\geq 1,$ then $$abc+b \geq ab+bc \implies b(a-1)(c-1)\geq 0.$$We have to prove $2\geq ac+b.$ Notice: $$a^2+c^2+b(ac+b)=4 \implies 2ac+b(ac+b) \leq 4 \implies (b+2)(ac+b-2) \leq 0 \implies ac+b \leq 2.$$Equality when $a=b=c=1.$ Q.E.D.
20.09.2024 20:25
Set $a=2\cos A,b=2\cos B,c=2\cos C$ for $\triangle ABC$ acute. Then the stronger lower bound $ab+bc+ca-3abc\ge 0$ follows from Erdos-Mordell from the circumcenter of $ABC$ to its tangential triangle. If $\triangle ABC$ has side lengths $x,y,z$ we can write $a,b,c$ in terms of $x,y,z$ by Law of Cosines. We can check that \[abc+2-ab-bc-ca=\sum_{\text{cyc}}bc\cdot\frac{(y-z)^2}{2yz}\]which is nonnegative so we are done.