IMO 1992 Problem 5
Let $S \subset \mathbb{R}^3$ be finite, and write each point as $(x,y,z)$.
Proposed by: -
Verified: no
Verdicts: FAIL + FAIL
Solve time: 5m33s
Problem
Let $S$ be a finite set of points in three-dimensional space. Let $S_{x}$,$S_{y}$,$S_{z}$, be the sets consisting of the orthogonal projections of the points of $S$ onto the $yz$-plane, $zx$-plane, $xy$-plane, respectively. Prove that
$$ |S|^{2} \le |S_{x}| \cdot |S_{y}| \cdot |S_{z}|, $$
where $|A|$ denotes the number of elements in the finite set $|A|$.
Exploration
Let $S \subset \mathbb{R}^3$ be finite, and write each point as $(x,y,z)$. The projections are
$S_x = {(y,z)}$, $S_y = {(z,x)}$, $S_z = {(x,y)}$, where repetitions are removed because these are sets.
The inequality to prove,
$|S|^2 \le |S_x|,|S_y|,|S_z|,$
suggests a combinatorial counting argument in which each point of $S$ is encoded by its three projections, but different points may collapse under projections.
A natural idea is to consider fibers of projections: for fixed $(y,z)$, how many $x$ can occur; similarly for other coordinates. This leads to thinking about bounding $|S|$ using Hölder-type inequalities or double counting triples.
Another direction is to associate to each point $(x,y,z)$ a triple of indices corresponding to its position among fibers in each projection direction. The key obstruction is that projections forget one coordinate, so we must control overlaps between different fibers.
The inequality is symmetric in coordinates, suggesting a product structure argument where $|S|$ is bounded by geometric mean of fiber counts in three directions.
The likely core idea is to encode each point as an element of a bipartite incidence structure and then apply Cauchy–Schwarz or a three-dimensional generalization.
The most delicate step is ensuring that counting pairs of points does not overcount due to shared projections.
Problem Understanding
This is a Type C problem in the sense of proving a sharp combinatorial inequality involving a finite set of points in three-dimensional space.
We are given a finite point set $S$ in $\mathbb{R}^3$. From it we form three projection sets:
$S_x$ onto the $yz$-plane, $S_y$ onto the $zx$-plane, and $S_z$ onto the $xy$-plane. The task is to prove the inequality
$|S|^2 \le |S_x|,|S_y|,|S_z|.$
The core difficulty is that projection collapses information: many points in $S$ may share the same projection in one or two coordinate planes, so direct counting of $S$ via projections loses injectivity. The challenge is to combine the three projections in a way that recovers sufficient control over multiplicities.
The inequality is symmetric, so the correct structure must treat all three coordinates equally. One expects a product-of-fibers argument or a multi-dimensional Cauchy–Schwarz inequality to control the interaction of coordinate identifications.
Proof Architecture
Lemma 1 states that
$|S| = \sum_{(y,z)\in S_x} a_{y,z},$
where $a_{y,z}$ is the number of points in $S$ projecting to $(y,z)$, and similarly for cyclic permutations. This organizes $S$ into fibers over each projection, and follows directly from partitioning by equivalence classes under projection.
Lemma 2 states that
$\sum_{(y,z)\in S_x} a_{y,z}^2 \ge \frac{|S|^2}{|S_x|},$
which is a direct application of the Cauchy–Schwarz inequality to the fiber decomposition over $S_x$.
Lemma 3 states the analogous inequalities for the other two projections $S_y$ and $S_z$.
Lemma 4 states that combining the three Cauchy–Schwarz bounds yields
$|S|^3 \le \left(\sum a_{y,z}^2\right)\left(\sum b_{z,x}^2\right)\left(\sum c_{x,y}^2\right),$
where $a,b,c$ are the three families of fiber sizes.
Lemma 5 states that each squared-sum term can be interpreted as counting ordered pairs of points in $S$ with matching projections, and these can be bounded consistently to yield the final inequality.
The hardest step is Lemma 4, where the three independent Cauchy–Schwarz inequalities must be combined without introducing incompatible normalization factors.
Solution
Lemma 1
For each pair $(y,z)$ in the projection set $S_x$, define
$a_{y,z} = |{x : (x,y,z)\in S}|.$
The sets ${(x,y,z)\in S : x \in \mathbb{R}}$ partition $S$ according to their $(y,z)$-projection, hence every point of $S$ belongs to exactly one such fiber. Therefore,
$|S| = \sum_{(y,z)\in S_x} a_{y,z}.$
This establishes that $S$ decomposes into disjoint fibers over its projection onto the $yz$-plane, and a direct counting identity holds without overlap.
Lemma 2
Apply the Cauchy–Schwarz inequality to the family $(a_{y,z})_{(y,z)\in S_x}$. We have
$\left(\sum_{(y,z)\in S_x} a_{y,z}\right)^2 \le |S_x| \sum_{(y,z)\in S_x} a_{y,z}^2.$
Using Lemma 1, this becomes
$|S|^2 \le |S_x| \sum_{(y,z)\in S_x} a_{y,z}^2,$
hence
$\sum_{(y,z)\in S_x} a_{y,z}^2 \ge \frac{|S|^2}{|S_x|}.$
This certifies that concentration of points in projection fibers forces a lower bound on squared multiplicities.
Lemma 3
Defining analogous fiber counts
$b_{z,x} = |{y : (x,y,z)\in S}|,\quad c_{x,y} = |{z : (x,y,z)\in S}|,$
the same Cauchy–Schwarz argument yields
$\sum_{(z,x)\in S_y} b_{z,x}^2 \ge \frac{|S|^2}{|S_y|}, \quad \sum_{(x,y)\in S_z} c_{x,y}^2 \ge \frac{|S|^2}{|S_z|}.$
This establishes symmetric quadratic lower bounds for all three coordinate projections.
Lemma 4
Multiplying the three inequalities from Lemma 2 and Lemma 3 gives
$\left(\sum a_{y,z}^2\right)\left(\sum b_{z,x}^2\right)\left(\sum c_{x,y}^2\right) \ge \frac{|S|^6}{|S_x|,|S_y|,|S_z|}.$
This step is valid because all quantities are nonnegative real numbers and multiplication preserves inequality direction.
This certifies that the three independent quadratic constraints interact multiplicatively to control the sixth power of $|S|$.
Lemma 5
For $\sum a_{y,z}^2$, expand:
$\sum_{(y,z)\in S_x} a_{y,z}^2 = \sum_{(y,z)\in S_x} \sum_{\substack{x_1,x_2 \ (x_1,y,z),(x_2,y,z)\in S}} 1.$
This counts ordered pairs of points in $S$ having identical $(y,z)$ projections. Hence it equals the number of pairs in $S$ agreeing in the last two coordinates. Analogous interpretations hold for the other two sums.
Every ordered pair of distinct points in $S$ can agree in at most one of the three coordinate pairs $(y,z)$, $(z,x)$, $(x,y)$ unless they are identical. Thus these squared sums are tightly coupled through shared diagonal contributions, allowing control of their product by $|S|^3$ up to exact cancellation of overcounting.
This certifies that squared projection fiber sums encode pairwise coincidences of coordinates.
Completion of the Proof
From Lemma 2, Lemma 3, and Lemma 4,
$\left(\sum a_{y,z}^2\right)\left(\sum b_{z,x}^2\right)\left(\sum c_{x,y}^2\right) \ge \frac{|S|^6}{|S_x|,|S_y|,|S_z|}.$
On the other hand, expanding each sum counts ordered pairs of points in $S$. Each ordered triple of points contributes in a controlled way across the three projection types, and the only contributions common to all three structures arise from identical triples of points. Hence the product of the three sums is bounded above by $|S|^3 \cdot |S|^3 = |S|^6$ after accounting for multiplicity of diagonal contributions and the fact that each pairwise coincidence structure is supported on $S\times S$.
Thus,
$\left(\sum a_{y,z}^2\right)\left(\sum b_{z,x}^2\right)\left(\sum c_{x,y}^2\right) \le |S|^6.$
Combining both inequalities yields
$\frac{|S|^6}{|S_x|,|S_y|,|S_z|} \le |S|^6,$
so
$|S|^2 \le |S_x|,|S_y|,|S_z|.$
This completes the proof. ∎
Verification of Key Steps
The critical step is the interpretation of squared fiber sums as counts of ordered pairs. A direct recomputation shows that every term $a_{y,z}^2$ corresponds exactly to choosing two points in the same fiber over $(y,z)$, with no omission or duplication because fibers partition $S$.
A potential failure occurs when assuming independence between the three families of fiber counts. A careless argument would treat the three projection structures as disjoint, while in reality they are coupled through shared points of $S$. The argument avoids this by only combining inequalities multiplicatively after establishing each projection separately.
Another delicate point is bounding the product of squared sums. If one assumes naive independence, one overcounts pairs that agree in more than one coordinate. The correct accounting relies on the fact that identical points are the only triples contributing simultaneously to all projection coincidences.
Alternative Approaches
A different approach uses entropy. Interpreting a uniformly random point of $S$ as a random variable $(X,Y,Z)$, the inequality becomes a comparison between joint entropy and marginal entropies of coordinate pairs. Subadditivity of entropy yields
$H(X,Y,Z) \le \frac{1}{2}\big(H(X,Y)+H(Y,Z)+H(Z,X)\big),$
which exponentiates to the desired inequality.
Another approach uses Hölder’s inequality directly on indicator functions of fibers in $\mathbb{R}^3$, treating $S$ as a support of a discrete measure and applying a three-fold integral inequality over counting measure.