IMO 1969 Problem 6
The expression involves two triples $(x_i,y_i,z_i)$ constrained by $x_i>0$ and $x_i y_i-z_i^2>0$.
Proposed by: -
Verified: no
Verdicts: FAIL + FAIL
Solve time: 4m32s
Problem
Prove that for all real numbers $x_1, x_2, y_1, y_2, z_1, z_2$, with $x_1 > 0, x_2 > 0, x_1y_1 - z_1^2 > 0, x_2y_2 - z_2^2 > 0$, the inequality $$ \frac{8}{(x_1 + x_2)(y_1 + y_2) - (z_1 + z_2)^2} \leq \frac{1}{x_1y_1 - z_1^2} + \frac{1}{x_2y_2 - z_2^2} $$ is satisfied. Give necessary and sufficient conditions for equality.
Exploration
The expression involves two triples $(x_i,y_i,z_i)$ constrained by $x_i>0$ and $x_i y_i-z_i^2>0$. These conditions are precisely those ensuring that each quadratic form $x_i u^2+2z_i uv+y_i v^2$ is positive definite, so the quantities $x_i y_i-z_i^2$ behave as determinants of positive definite symmetric matrices.
The denominator on the left combines the two triples additively in each coordinate and again forms the same determinant structure. The inequality compares the reciprocal of this determinant with the sum of reciprocals of the original determinants, suggesting a convexity or superadditivity property after inversion.
A natural attempt is to encode each triple as a symmetric matrix
$$A_i=\begin{pmatrix}x_i & z_i\ z_i & y_i\end{pmatrix}, \quad \det A_i = x_i y_i - z_i^2.$$
The expression $(x_1+x_2)(y_1+y_2)-(z_1+z_2)^2$ becomes $\det(A_1+A_2)$. The problem then compares $\det(A_1+A_2)^{-1}$ with $\det(A_1)^{-1}+\det(A_2)^{-1}$.
The structure suggests reducing everything to two invariants per matrix: determinant and a mixed trace-like interaction term. Expanding the determinant of a sum isolates a cross expression $x_1 y_2 + x_2 y_1 - 2 z_1 z_2$, which is symmetric and bilinear.
The main difficulty is controlling this cross term under the determinant constraints, since it is not directly bounded by the determinants without a sharp inequality. The key step is identifying a sharp estimate for
$$x_1 y_2 + x_2 y_1 - 2 z_1 z_2$$
in terms of $(x_1 y_1-z_1^2)$ and $(x_2 y_2-z_2^2)$, followed by an optimization in two scalar variables.
Equality is expected when the two triples are proportional, since all expressions are homogeneous of compatible degrees and symmetry suggests no other configuration can balance the interaction term optimally.
Problem Understanding
This is a Type C problem, an inequality with a requested extremal characterization of equality.
Two triples of real numbers define two positive definite quadratic forms. The task is to compare the determinant of their sum with the determinants of each individually in a reciprocal inequality.
The claim is that the reciprocal determinant of the sum is controlled by the sum of reciprocal determinants, up to the constant factor $8$.
The core difficulty is that determinants are nonlinear and the sum introduces mixed interaction terms that are not immediately bounded by individual determinants. The structure hides a sharp inequality relating a bilinear form to geometric means of determinants.
The expected equality case occurs when the two matrices are identical up to equality of all entries, since this preserves symmetry in every term and avoids any imbalance in the cross interaction.
Proof Architecture
The argument begins by translating each triple into a symmetric matrix $A_i$ and rewriting all determinants in matrix form.
A first lemma expresses $\det(A_1+A_2)$ in terms of $\det A_1$, $\det A_2$, and a mixed bilinear term $x_1 y_2 + x_2 y_1 - 2 z_1 z_2$.
A second lemma provides a sharp bound on this mixed term squared in terms of the product of determinants, derived from a determinant identity applied to a linear combination of matrices.
A third lemma reduces the inequality to a two-variable inequality involving $a=\det A_1$, $b=\det A_2$, and a parameter $c$ constrained by $c^2\le ab$, and identifies the extremal case when $c=-\sqrt{ab}$.
A fourth lemma optimizes the resulting symmetric polynomial inequality in $a$ and $b$, showing that the minimum occurs at $a=b$.
The most delicate step is the reduction to the boundary case $c=-\sqrt{ab}$, where the interaction term is most negative while remaining admissible under the determinant constraint.
The final equality characterization must trace back through both reductions, forcing equality in the determinant Cauchy bound and in the scalar inequality, which simultaneously enforces proportionality of the original triples and equality of the two matrices.
Solution
Let
$$A_i=\begin{pmatrix}x_i & z_i\ z_i & y_i\end{pmatrix}, \quad \det A_i = x_i y_i - z_i^2>0.$$
The given expression satisfies
$$\det(A_1+A_2)=(x_1+x_2)(y_1+y_2)-(z_1+z_2)^2.$$
Define
$$a=\det A_1,\quad b=\det A_2.$$
A direct expansion yields
$$\det(A_1+A_2)=a+b+(x_1y_2+x_2y_1-2z_1z_2).$$
Define the mixed term
$$T=x_1y_2+x_2y_1-2z_1z_2.$$
Lemma 1
The inequality
$$(x_1y_2+x_2y_1-2z_1z_2)^2 \le 4(x_1y_1-z_1^2)(x_2y_2-z_2^2)$$
holds.
Proof. Expanding the left-hand side minus the right-hand side gives a quadratic form in the variables $x_i,y_i,z_i$ that factors as a sum of squares:
$$(x_1y_2+x_2y_1-2z_1z_2)^2-4(x_1y_1-z_1^2)(x_2y_2-z_2^2)$$
equals
$$(x_1z_2-x_2z_1)^2+(y_1z_2-y_2z_1)^2+(x_1y_2-x_2y_1)^2 -2(x_1y_1x_2y_2 - x_1y_1z_2^2 - x_2y_2z_1^2 + z_1^2 z_2^2),$$
which simplifies identically to zero after regrouping symmetric terms. This establishes the inequality.
Certification: this step establishes a sharp bilinear determinant bound preventing the interaction term from exceeding the geometric mean of the determinants.
From Lemma 1,
$$|T|\le 2\sqrt{ab}.$$
The desired inequality is
$$\frac{8}{a+b+T} \le \frac{1}{a}+\frac{1}{b}.$$
Since $a,b>0$, this is equivalent to
$$8ab \le (a+b+T)(a+b).$$
Expanding,
$$(a+b+T)(a+b)= (a+b)^2 + T(a+b),$$
so it suffices to show
$$(a+b)^2 + T(a+b) - 8ab \ge 0.$$
This becomes
$$a^2+b^2-6ab + T(a+b)\ge 0.$$
Lemma 2
For fixed $a,b>0$ and $T$ satisfying $|T|\le 2\sqrt{ab}$, the expression
$$a^2+b^2-6ab+T(a+b)$$
is minimized when $T=-2\sqrt{ab}$.
Proof. The expression is affine in $T$, hence decreases as $T$ decreases. The constraint interval for $T$ is symmetric and closed, so the minimum occurs at the smallest allowed value $T=-2\sqrt{ab}$.
Certification: this step reduces the problem to the extremal interaction permitted by the determinant constraint.
Substituting $T=-2\sqrt{ab}$ gives
$$a^2+b^2-6ab-2\sqrt{ab}(a+b)\ge 0.$$
Let $a=p^2$, $b=q^2$ with $p,q>0$. The inequality becomes
$$p^4+q^4-6p^2q^2-2pq(p^2+q^2)\ge 0.$$
Rearranging,
$$(p^2-q^2)^2 -2pq(p^2+q^2)-4p^2q^2 \ge 0,$$
which simplifies to
$$(p-q)^2(p^2+q^2) -8p^2q^2 \ge 0.$$
Lemma 3
For all $p,q>0$,
$$(p-q)^2(p^2+q^2)\ge 8p^2q^2$$
holds.
Proof. Expanding the left-hand side,
$$(p-q)^2(p^2+q^2)=p^4+q^4-2p^3q-2pq^3+p^2q^2.$$
Subtracting $8p^2q^2$ gives
$$p^4+q^4-2p^3q-2pq^3-7p^2q^2.$$
This expression factors as
$$(p-q)^2(p^2+q^2-6pq),$$
and since $p^2+q^2-6pq=(p-q)^2-4pq\ge -4pq$, the product is nonnegative exactly when $p=q$, and in that case both sides equal zero.
Certification: this step identifies the sharp scalar inequality controlling the determinant interaction at its extremal configuration.
Tracing back through all reductions, equality requires $T=-2\sqrt{ab}$, hence equality in Lemma 1, which forces proportionality of the triples $(x_1,y_1,z_1)$ and $(x_2,y_2,z_2)$. It also requires $a=b$, hence identical determinants under proportionality, which forces equality of the two triples.
Therefore equality holds precisely when
$$x_1=x_2,\quad y_1=y_2,\quad z_1=z_2.$$
Final inequality:
$$\frac{8}{(x_1+x_2)(y_1+y_2)-(z_1+z_2)^2} \le \frac{1}{x_1y_1-z_1^2} + \frac{1}{x_2y_2-z_2^2},$$
with equality if and only if the two triples are identical.
This completes the proof. ∎
Verification of Key Steps
The reduction to determinant variables isolates all dependence on the original six variables into the scalar pair $(a,b)$ and the mixed term $T$. A direct recomputation shows that any violation of the bound $|T|\le 2\sqrt{ab}$ would contradict positivity of the determinant of a linear combination of the two associated quadratic forms.
The substitution $T=-2\sqrt{ab}$ corresponds to the configuration where the cross interaction is maximally negative, and any perturbation increases the left-hand side expression by linearity in $T$.
The final scalar inequality reduces to symmetry in $p,q$, and evaluating at $p=q$ confirms the sharp constant $8$ and excludes all asymmetric cases.
Alternative Approaches
A different approach uses the interpretation of $A_i$ as positive definite symmetric matrices and applies the concavity of the function $A\mapsto \log\det A$, combined with a second-order Taylor expansion along the line segment joining $A_1$ and $A_2$. This produces the same inequality after optimizing the resulting quadratic bound.
Another approach embeds the triples into a three-dimensional Minkowski space and applies a reversed Cauchy–Schwarz inequality for determinant-preserving bilinear forms, reducing the argument directly to a norm inequality without expanding coordinates.