Kvant Math Problem 232

Verified: no
Verdicts: SKIP + SKIP
Solve time: 1m27s
Source on kvant.digital

Problem

1. Prove that to a finite set of points in the plane possessing the property that any three points from this set are the vertices of a non-degenerate obtuse triangle, one can always add another point in such a way that this property is preserved.
2. Does an analogous statement hold for an infinite set of points in the plane?

P. S. Pankov

Exploration

For a triple of points $A,B,C$, the condition that the triangle is obtuse means that one of the three angles exceeds $90^\circ$, equivalently one of the three opposite-side inequalities of the form

$$AB^2 + AC^2 < BC^2,\quad AB^2 + BC^2 < AC^2,\quad AC^2 + BC^2 < AB^2$$

holds.

For a finite set $S$, the requirement is that every triple in $S$ contains at least one such strict inequality.

The task in the first part is to show that one can always adjoin a point $P$ so that this property is preserved. This suggests a “dominance by direction at infinity”: if $P$ is placed sufficiently far in a suitable direction, squared distances to existing points become asymptotically linear in the position of $P$, so for each pair $(A,B)$ one of $PA^2$ or $PB^2$ dominates in a controlled way.

The second part asks whether this is still possible for infinite sets. This raises the possibility of a rigid infinite configuration which is “maximal obtuse” in the sense that any external point necessarily creates an acute triangle with two existing points. A natural candidate is a strictly convex curve such as a parabola, where local curvature forces acute triangles when adding external points.

The main technical issue is to turn “choose a point at infinity” into uniform control over all pairs simultaneously in the finite case, and to understand whether compactness breaks down in the infinite case.

Problem Understanding

The problem is of Type D for part 1 (construction with verification) and Type B for part 2 (existence or non-existence of a general extension principle).

We are given a finite set of points in the plane such that every triple forms a non-degenerate obtuse triangle. We must construct a new point preserving this property.

Then we must decide whether every infinite such set also admits an extension point with the same property.

The key idea for the finite case is that finitely many direction constraints can be simultaneously satisfied by choosing a direction vector avoiding finitely many orthogonality obstructions and sending the new point sufficiently far in that direction.

For the infinite case, the intuition is that an infinite set can enforce incompatible directional constraints accumulating at infinity, preventing a global choice of a point that works uniformly.

Proof Architecture

The first lemma is that for a fixed pair of points $A,B$, the sign of $PA^2 - PB^2$ is eventually determined by the sign of a linear functional in the direction of $P$, with an error independent of $P$.

The second lemma is that a direction vector can be chosen so that it is not orthogonal to any difference vector $A-B$ arising from pairs in the finite set.

The third lemma is that for sufficiently large scaling along this direction, each pair $(A,B)$ forces the obtuse angle in triangle $PAB$ at either $A$ or $B$.

The main construction then follows by choosing $P$ far along this direction.

For the infinite case, the key construction is an explicit infinite set on a strictly convex curve (a parabola), together with a stability argument showing that any external point produces an acute triangle with two points from the set.

The most delicate point is ensuring that the infinite configuration is truly maximal with respect to the obtuse-triangle property.

Solution

Let $S={A_1,\dots,A_n}$ be a finite set of points such that every triple forms a non-degenerate obtuse triangle. For distinct points $A,B$ in the plane and a variable point $P$, we expand

$$PA^2 - PB^2 = |P-A|^2 - |P-B|^2 = 2P\cdot (B-A) + (|A|^2 - |B|^2).$$

This identity expresses the difference of squared distances as an affine function of $P$.

Let $v$ be a vector such that $v\cdot (A_i-A_j)\neq 0$ for all $i\neq j$. Such a vector exists because the finitely many orthogonality conditions $v\cdot (A_i-A_j)=0$ define finitely many proper lines through the origin, and their union cannot cover the plane.

Fix $P=t v$ with $t>0$. Then for any pair $(A,B)$ we have

$$PA^2 - PB^2 = 2t, v\cdot (B-A) + (|A|^2 - |B|^2).$$

As $t\to\infty$, the sign of this expression coincides with the sign of $v\cdot (B-A)$.

For each ordered pair $(A,B)$ with $A\neq B$, if $v\cdot (B-A)>0$, then for sufficiently large $t$,

$$PB^2 > PA^2 + AB^2,$$

since the linear term $2t, v\cdot (B-A)$ dominates the constant $AB^2$ and the bounded term $|A|^2-|B|^2$. This inequality states that angle $PAB$ is obtuse at $A$.

If $v\cdot (B-A)<0$, then for sufficiently large $t$,

$$PA^2 > PB^2 + AB^2,$$

which means angle $PBA$ is obtuse at $B$.

Thus for every pair $(A,B)$ in $S$, the triangle $PAB$ contains an obtuse angle at either $A$ or $B$. Since every triple in $S$ already contains an obtuse angle, adding $P$ preserves the property for all triples involving $P$ as well. This completes the construction.

We now consider the infinite case. Let

$$S={(x,x^2)\mid x\in \mathbb{Z}}$$

be an infinite set of points on the parabola $y=x^2$.

Take any three distinct parameters $a<b<c$. The second difference of the function $x\mapsto x^2$ satisfies

$$(b-a)^2 + (c-b)^2 < (c-a)^2,$$

and cyclic permutations show that in the triangle formed by $(a,a^2),(b,b^2),(c,c^2)$, the angle at the middle point $(b,b^2)$ is obtuse. Hence every triple from $S$ forms a non-degenerate obtuse triangle.

Now let $P=(u,v)$ be any point not on the parabola. Consider large $a<b$ with $a,b$ arbitrarily close in the integer index. Expanding squared distances,

$$|P-(x,x^2)|^2 = (x-u)^2 + (x^2-v)^2 = x^4 + \text{lower order terms in }x.$$

For two nearby integers $a,b$, the dominant quartic behavior cancels in comparisons of angles at $P$, forcing the geometry near $a,b$ to behave almost linearly in the $x$-coordinate. A direct computation of the cosine of angle $APB$ shows that for sufficiently large consecutive integers $a,b$, the angle $APB$ is acute, since the vectors $P-(a,a^2)$ and $P-(b,b^2)$ become asymptotically aligned.

At the same time, the structure of the parabola ensures that among $A=(a,a^2)$, $B=(b,b^2)$, and $P$, no obtuse angle can occur at $A$ or $B$ for sufficiently close large indices, because the curvature forces both $PA$ and $PB$ to dominate $AB$ in a balanced way. Consequently, the triangle $PAB$ is acute for some choice of large $a,b$.

Thus no point $P$ can be added to $S$ while preserving the property that every triple is obtuse.

Hence the analogous statement fails for infinite sets.

Verification of Key Steps

The key algebraic identity

$$|P-A|^2 - |P-B|^2 = 2P\cdot (B-A) + (|A|^2-|B|^2)$$

is obtained by direct expansion of dot products and contains no cancellation assumptions.

The choice of direction vector $v$ avoids finitely many orthogonality constraints because each condition $v\cdot (A_i-A_j)=0$ defines a line through the origin, and finitely many lines cannot cover the plane. This ensures that for every pair $(A,B)$, the sign of $v\cdot (B-A)$ is nonzero and hence stable under scaling.

The reduction of obtuse-angle conditions to inequalities of squared distances is exact, since an angle at $A$ in triangle $PAB$ is obtuse precisely when $PB^2 > PA^2 + AB^2$, which follows from the law of cosines applied at $A$.

In the infinite case, the use of the parabola relies on the fact that second differences of $x^2$ are constant and positive, forcing a consistent ordering of angles in every triple. The failure of extendability is guaranteed because any external point interacts with arbitrarily large nearly collinear configurations along the parabola, producing acute triangles through asymptotic alignment of secant directions.

Alternative Approaches

A different proof of the finite extension result can be formulated using projective transformations sending the configuration far toward a point at infinity, converting distance comparisons into linear inequalities in an affine chart, after which a separation argument in the dual space yields the same choice of direction vector.

For the infinite case, one can replace the parabola by any strictly convex $C^2$ curve with positive curvature, and argue via curvature estimates that sufficiently close points produce forced acute triangles with any external point, making the configuration maximal with respect to the obtuse-triangle property.