How can the following inequation be proven?
$$a^2 + b^2 + c^2 \ge ab + bc + ca$$
How can the following inequation be proven?
$$a^2 + b^2 + c^2 \ge ab + bc + ca$$
This is a specific form of Cauchy-Schwarz inequality.
Let $x = (a, b, c)$ and $y = (b, c, a)$ as vectors.
The inequality is $ | \left< x,y \right>| \le \|x\|\|y\|. $ with standard inner product definition. One neat trick to prove this is using an auxilary parameter $t,$ and expanding $$ \| x+ty \|^2 = \left< x+ty,x+ty \right> = \|x\|^2 + 2 \left< x,y \right>t +\|y\|^2t^2.$$ We know, this being a square, is non-negative. Therefore, the discriminant of the polynomial in $t$ is less or equal to zero. Which is $\left< x,y \right>^2 - (\|x\|\|y\|)^2 \le 0.$ Substituting the values for $x$ and $y$ will do the job.
$$\sum_{cyc}(a^2-ab)=\frac{1}{2}\sum_{cyc}(a^2-2ab+b^2)=\frac{1}{2}\sum_{cyc}(a-b)^2\geq0$$
$$q := a^2 + b^2 + c^2 - ab - bc - ca = \frac12 \begin{bmatrix} a\\ b \\ c\end{bmatrix}^\top \underbrace{\begin{bmatrix} 2 & -1 & -1\\ -1 & 2 & -1 \\ -1 & -1 & 2\end{bmatrix}}_{=: {\rm L}} \begin{bmatrix} a\\ b \\ c\end{bmatrix}$$
where matrix $\rm L$ is the Laplacian of the cycle graph with $3$ vertices, whose (signed) incidence matrix is
$${\rm C} = \begin{bmatrix} -1 & \color{blue}{1} & 0\\ 0 & -1 & \color{blue}{1} \\ \color{blue}{1} & 0 & -1\end{bmatrix}$$
Since $\rm L = C^\top C$, we obtain the following sum of squares (SOS) decomposition
$$2q = (\color{blue}{b} - a)^2 + (\color{blue}{c} - b)^2 + (\color{blue}{a} - c)^2 \geq 0$$
which is the SOS decomposition proposed by Mark Bennet. Since matrix $\rm L$ is rank-$2$, a terser SOS decomposition with only $2$ terms can easily be found — say, via the Cholesky decomposition.
Using Macaulay2,
Macaulay2, version 1.16
with packages: ConwayPolynomials, Elimination, IntegralClosure, InverseSystems, LLLBases, MinimalPrimes, PrimaryDecomposition, ReesAlgebra, TangentCone, Truncations
i1 : needsPackage( "SumsOfSquares" );
--loading configuration for package "NumericalAlgebraicGeometry" from file /Users/rodrigo/Library/Application Support/Macaulay2/init-NumericalAlgebraicGeometry.m2
--loading configuration for package "Bertini" from file /Users/rodrigo/Library/Application Support/Macaulay2/init-Bertini.m2
--warning: symbol "Verbosity" in MinimalPrimes.Dictionary is shadowed by a symbol in SemidefiniteProgramming.Dictionary
-- use the synonym MinimalPrimes$Verbosity
i2 : R = QQ[a,b,c];
i3 : q = a^2 + b^2 + c^2 - ab - ac - b*c
2 2 2
o3 = a - ab + b - ac - b*c + c
o3 : R
i4 : sosPoly solveSOS q
1 1 2 3 2
o4 = (1)(a - -b - -c) + (-)(b - c)
2 2 4
o4 : SOSPoly
i5 : tex o4
o5 = $\texttt{SOSPoly}\left{\texttt{coefficients},\Rightarrow,\left{1,,\frac{3}{4}\right},,\texttt{generators},\Rightarrow,\left{a-\frac{1}{2},b-\frac{1}{2},c,,b-c\right
},,\texttt{ring},\Rightarrow,R\right}$
In $\TeX$,
$$\texttt{SOSPoly}\left\{\texttt{coefficients}\,\Rightarrow\,\left\{1,\,\frac{3}{4}\right\},\,\texttt{generators}\,\Rightarrow\,\left\{a-\frac{1}{2}\,b-\frac{1}{2}\,c,\,b-c\right\},\,\texttt{ring}\,\Rightarrow\,R\right\}$$
From Cauchy-Schwarz
$ab+bc+ac=\sqrt{a^2}\sqrt{b^2}+\sqrt{b^2}\sqrt{c^2}+\sqrt{a^2}\sqrt{c^2} \leq \sqrt{a^2+b^2+c^2}\sqrt{a^2+b^2+c^2}$
Moving on;
$ab+bc+ac \leq a^2+b^2+c^2$
Done!
If $\ c> a ,a^2+c^2 \gt 2ac $, because $\ (a-c)^2 \gt 0$ If $\ c>b>a ,c^2+b^2/2+a^2/2 \gt ac+bc $ and $\ b^2/2+a^2/2 \gt ab $, sum of them $\ a^2+b^2+c^2 \gt ab+bc+ac $ If $\ c=b \gt a $ or $\ a=b=c $, it can be solved with same logic.
$a^2+b^2+c^2-ab-bc-ca\\ =(a, b, c)\begin{pmatrix}1&-1/2&-1/2\\-1/2& 1&-1/2\\-1/2&-1/2&1\end{pmatrix}\left(\begin{array}{c}a\\b\\c\end{array}\right)$
$=(a, b, c)A\begin{pmatrix}a\\b\\c\end{pmatrix}$
It is sufficient to prove $A$ is a positive semi-definite.
It follows from that the minor determinant of $A$ is one of $1,\dfrac{3}{4},0$
for $a,b,c>=0$
we know $(a-b-c)^2>=0$
i.e $a^2 + b^2 + c^2 - 2 (ab + bc + ca) \ge 0$
i.e $a^2 + b^2 + c^2 \ge ab + bc + ca$
This inequality can be solved by simple algebra
we have the equation $$a^2+b^2+c^2 >= ab+ac+cb$$ multiply and divide 2 on both sides $$\frac{2}{2}(a^2+b^2+c^2) >= \frac{2}{2}(ab+ac+cb)$$ putting the left side equation on the right side $$\frac{2a^2+2b^2+2c^2 - 2(ab+ac+cb)}{2} >= 0$$ then factorize and multiply by 2 on both the sides $$\frac{2a^2+2b^2+2c^2 - 2(ab+ac+cb)}{2} >= 0$$ $$\frac{(a-b)^2+(b-c)^2+(c-a)^2}{2} >= 0$$ $$\frac{(a-b)^2+(b-c)^2+(c-a)^2}{2} >= 0$$ $$2*\frac{(a-b)^2+(b-c)^2+(c-a)^2}{2} >= 2*0$$ $$(a-b)^2+(b-c)^2+(c-a)^2 >= 0$$ So the above equation is zero when $a=b=c$
So the above inequality is proved
Another way to approach this question would be to use the AM-GM Inequality. By AM-GM Inequality,
$ \begin{split} \frac{a^3+b^3+c^3}{3} \geq \sqrt[3]{a^3b^3c^3} & \implies a^3+b^3+c^3 \geq 3abc \\ & \implies a^3+b^3+c^3 - 3abc \geq0 \\ & \implies (a+b+c)(a^2+b^2+c^2-ab-bc-ca) \geq 0 \\ & \implies a^2+b^2+c^2 \geq ab+bc+ca \end{split} $
Obtained $$\sum a^{2}- \sum ab= \left ( c+ a- 2b \right )^{2}+ 3\left ( a- b \right )\left ( b- c \right )$$ by assuming $b:=\mathsf{med}\left ( a, b, c \right ).$
Obtained $$\sum a^{2}- \sum ab= \sum a\left ( a- b \right )\geq 0$$ by assuming $a\geq b\geq 0\geq c.$
Seems like various answers are being posted, so I’m adding one more.
This can be solved with Rearrangement Inequality, as mentioned in the OP’s comment.
Theorem 1. Greedy Algorithm
Greedy algorithm is a short-sighted algorithm to gain the best benefit. For example, we have three boxes with \$$20$, \$$50$, \$$100$ bills in each. We can take $3$, $4$, $5$ bills in three different boxes. To make the best benefit, you’ll take $5$ bills from \$$100$ box, $4$ from \$$50$, $3$ from \$$20$.
Of course, this algorithm sometimes(or often) have a failure, as the counterexample given below.
\begin{align} &7 \begin{cases} &20 \begin{cases} &10 \\ \\\ &11 \end{cases}\\ \\ & 19 \begin{cases} &14 \\ \\ &100\end{cases} \end{cases} \end{align}
To make the highest sum, you’ll gotta choose $7-19-100$. But the greedy algorithm chooses $7-20-11$, fails to achieve the goal. This is why the algorithm is called short-sighted.
Theorem 2. Rearrangement Inequality
Anyway, based on the algorithm above, we can make some successful inequality called Rearrangement Inequality.
The inequality is written as below:
$$a_1\le a_2\le \cdots\le a_n, b_1\le b_2\le \cdots\le b_n. \\ \sum_{i=1}^{n}a_ib_{n+1-i} \le \sum_{i=1}^{n}a_id_i \le \sum_{i=1}^{n} a_ib_i. \\ (d_i: \text{Rearrangement of sequence } b_i.)$$
I love proving this with induction.
Suppose that there exists a specific rearrangement of $b_i$(not equal to $b_i$) defined as $d_i$ which has the maximum of the sum of $a_id_i$.
Now, there exists a pair of distinct numbers $(k, l)$ which satisfies $k<l, d_k>d_l$.
So define new rearrangement $d’_i$ as:
$$d’_i=\begin{cases} & d_i & (i\neq k, i\neq l) \\ & d_k & (i=l) \\ & d_l & (i=k) \end{cases}$$
Then we have:
$$\sum_{i=1}^{n} a_id_i-\sum_{i=1}^{n} a_id’_i = a_kd_k+a_ld_l-a_kd_l-a_ld_k=(a_k-a_l)(d_k-d_l)<0. \\ \therefore \sum_{i=1}^{n} a_id_i < \sum_{i=1}^{n} a_id’_i.$$
So it’s a contradiction by the definition of sequence $d_i$.
ISW we can prove the minimum value.
Conclusion
This post seems to be needlessly long, but I just wanted to elaborate on the comment.
To conclude, without loss of generality let $a\leq b\leq c$.
Now, define $a_1=b_1=a, a_2=b_2=b, a_3=b_3=c$.
Applying the Rearrangement Inequality, we get:
$$a_1b_3+a_2b_2+a_3b_1 \leq a_1b_2+a_2b_3+a_3b_1 \leq a_1b_1+a_2b_2+a_3b_3$$
, which can be rewritten as:
$$ac+bb+ca\leq ab+bc+ca\leq aa+bb+cc$$
. So we have $ab+bc+ca\leq a^2+b^2+c^2$, and we are finally done!