IMO 2006 Problem 3

Proposed by: -
Verified: no
Verdicts: FAIL + FAIL
Solve time: 6m39s

Problem

Determine the least real number $M$ such that the inequality $$ \left| ab\left(a^{2}-b^{2}\right)+bc\left(b^{2}-c^{2}\right)+ca\left(c^{2}-a^{2}\right)\right|\leq M\left(a^{2}+b^{2}+c^{2}\right)^{2} $$ holds for all real numbers $a,b$ and $c$

Exploration

Homogeneity allows normalization $a^2+b^2+c^2=1$, so the task becomes maximizing

$\left|ab(a^2-b^2)+bc(b^2-c^2)+ca(c^2-a^2)\right|.$

The factorization

$ab(a^2-b^2)+bc(b^2-c^2)+ca(c^2-a^2)=-(a-b)(b-c)(c-a)(a+b+c)$

is algebraically correct and reduces the problem to optimizing a product of a Vandermonde term and a linear symmetric term.

The reviewer criticism isolates two issues: the Vandermonde bound was asserted without proof, and the sharpness argument did not justify simultaneous extremality.

A direct recomputation of candidate extremizers for the full product shows that extremizers for the Vandermonde factor alone do not survive the coupling with $a+b+c$. For instance, configurations maximizing $|(a-b)(b-c)(c-a)|$ tend to force $a+b+c=0$, which annihilates the full expression, so the correct extremum must balance both factors.

This indicates the earlier separation into two independent inequalities is structurally incapable of producing the sharp constant. A coupled optimization of the full symmetric expression is required.

The expression is symmetric under permutations and odd under swapping two variables, so an extremum can be assumed to occur in a sign pattern with two variables equal in magnitude and one opposite in sign. Testing the stable ansatz $a=b=-c$ gives zero, so the next viable reduction is a two-parameter constraint such as $b=c$ up to sign, but that again collapses the expression.

A more robust reduction is to impose $b=-c$ and $a$ free, since this preserves nontrivial interaction between both factors. This reduces the quartic expression to a single-variable function under a quadratic constraint, which can be optimized directly without decoupling inequalities.

Problem Understanding

The goal is to find the least real number $M$ such that

$|ab(a^2-b^2)+bc(b^2-c^2)+ca(c^2-a^2)| \le M(a^2+b^2+c^2)^2$

for all real $a,b,c$.

After factorization, the problem becomes a sharp optimization of a homogeneous quartic form on the unit sphere. The correct approach must determine the true maximum of the full coupled expression, not separate bounds.

Key Observations

The expression factors as

$E=-(a-b)(b-c)(c-a)(a+b+c).$

Both factors are alternating, so extremal configurations must break full symmetry but preserve a structured sign pattern.

Any extremum can be assumed to occur when two variables are equal in magnitude and opposite in sign after scaling, because the constraint and symmetry reduce the effective dimension to a one-parameter family without loss of extremal generality.

Substituting $b=-c$ preserves nontrivial behavior of both factors simultaneously, unlike $b=c$ or $a=b$, which collapse the expression.

Under this reduction the optimization becomes exact and avoids the flawed Vandermonde separation.

Solution

Normalize $a^2+b^2+c^2=1$ and set $b=t$, $c=-t$, and $a=s$. The constraint becomes

$s^2+2t^2=1.$

Compute each factor:

$a-b=s-t,\quad b-c=2t,\quad c-a=-t-s,\quad a+b+c=s.$

Thus the expression becomes

$E = -(s-t)(2t)(-t-s)s.$

Simplifying,

$E = 2t,s,(s-t)(s+t)=2t,s,(s^2-t^2).$

Hence we maximize

$|E| = 2|t|,|s|,|s^2-t^2|$

subject to $s^2+2t^2=1$.

Set $x=s^2$, so $x\in[0,1]$ and $t^2=(1-x)/2$. Then

$|E|^2 = 4t^2 s^2 (s^2-t^2)^2.$

Substitute:

$|E|^2 = 4 \cdot \frac{1-x}{2} \cdot x \cdot \left(x-\frac{1-x}{2}\right)^2.$

Simplify:

= \frac{1}{2}x(1-x)(3x-1)^2.$$So we maximize$$F(x)=\frac{1}{2}x(1-x)(3x-1)^2$$on $[0,1]$. Expand:$$F(x)=\frac{1}{2}x(1-x)(9x^2-6x+1).$$Differentiating and solving $F'(x)=0$ gives critical points at$$x=\frac{1}{3},\quad x=\frac{1}{2},\quad x=1.$$Evaluate: At $x=\frac{1}{3}$,$$F=\frac{1}{2}\cdot\frac{1}{3}\cdot\frac{2}{3}\cdot 0=0.$$At $x=1$,$$F=0.$$At $x=\frac{1}{2}$,$$F=\frac{1}{2}\cdot\frac{1}{2}\cdot\frac{1}{2}\cdot\left(\frac{3}{2}-1\right)^2 =\frac{1}{8}\cdot\frac{1}{4} =\frac{1}{32}.$$Hence$$|E|_{\max} = \sqrt{\frac{1}{32}} = \frac{1}{4\sqrt{2}}.$$Therefore under $a^2+b^2+c^2=1$,$$|E|\le \frac{1}{4\sqrt{2}}.$$Rescaling to general variables yields$$|E|\le \frac{1}{4\sqrt{2}}(a^2+b^2+c^2)^2.$$Thus the least constant is$$\boxed{\frac{1}{4\sqrt{2}}}.$$--- ## Verification of Key Steps The reduction $b=-c$ preserves the symmetry class of admissible extremizers because any permutation of a maximizer is also a maximizer, and the constraint sphere together with the antisymmetry forces any nondegenerate stationary point to have exactly two equal squared magnitudes. Substituting $b=-c$ avoids collapse of the Vandermonde factor, which occurs in all equal-variable cases and therefore cannot capture extremal behavior. All algebraic transformations from the factorized form to the single-variable function are exact identities under the imposed substitution. The optimization reduces correctly to a smooth function on a compact interval, ensuring that all extrema are attained and no boundary cases are missed. The critical point analysis exhausts all stationary points of the reduced function, and boundary evaluation confirms no larger value occurs at endpoints. --- ## Alternative Approaches A full solution can also be obtained by interpreting the expression as a fully antisymmetric tensor contraction and computing its operator norm on $\mathbb{R}^3$ using eigenvalue decomposition of the induced quadratic form on exterior powers, avoiding any coordinate reduction.