Kvant Math Problem 1192
The polyhedron has all edges of equal length and every edge is tangent to a sphere.
Verified: no
Verdicts: SKIP + SKIP
Solve time: 8m00s
Source on kvant.digital
Problem
It is known that all edges of the polyhedron $M$ are equal to one another and are tangent to some sphere.
- Suppose that one of the faces of $M$ has an odd number of sides. Prove that there exists a sphere circumscribed about $M$.
- Under the conditions of part (a), must there necessarily exist a sphere inscribed in $M$?
- Suppose that all faces of $M$ have the same number of sides. Prove that there exists a sphere inscribed in $M$.
- Under the conditions of part (c), must there necessarily exist a circumscribed sphere?
V. A. Senderov
Exploration
The polyhedron has all edges of equal length and every edge is tangent to a sphere. Let this sphere have center $O$. Since each edge has the same length, say $a$, the point of tangency with an edge is its midpoint. Indeed, if a segment of fixed length $a$ is tangent to a sphere of radius $r$, then the tangency point is the foot of the perpendicular from $O$ to the line of the segment; among all points of the line it is the nearest to $O$, hence it must be the midpoint because the distances from $O$ to the endpoints are equal.
Consider a vertex $V$. Every edge through $V$ has length $a$ and is tangent to the sphere. If $T$ is the midpoint of such an edge, then in the right triangle $OVT$ we have
$$OV^2=OT^2+VT^2=r^2+\frac{a^2}{4}.$$
Hence $OV$ is the same for every vertex. This already suggests that all vertices lie on a sphere centered at $O$.
Part (a) asks for a circumscribed sphere under the additional assumption that one face has an odd number of sides. The computation above seems to show something stronger: every vertex lies on one sphere regardless of the parity condition. The odd face condition must therefore be related to the standard fact that from an edge-tangent polyhedron one cannot always deduce that the tangency point is the midpoint. Let us examine this more carefully.
Let a face have successive tangency lengths along its boundary
$$x_1,x_2,\dots,x_n,$$
where $x_i$ is the distance from vertex $V_i$ to the tangency point on side $V_iV_{i+1}$. Since all sides have length $a$,
$$x_i+x_{i+1}=a.$$
If $n$ is odd, solving these equations cyclically gives
$$x_1=x_2=\cdots=x_n=\frac a2.$$
Thus on an odd face every tangency point is the midpoint of its edge. For any vertex of that face,
$$OV^2=r^2+\frac{a^2}{4}.$$
Now every edge incident with such a vertex also has midpoint tangency, because the tangent lengths from one vertex to the sphere along all incident edges are equal. Therefore every edge of the polyhedron has midpoint tangency. Then the previous computation yields that all vertices are equidistant from $O$.
This seems to be the intended argument.
For part (b), does an inscribed sphere necessarily exist? We need a counterexample. The given sphere tangent to all edges need not be tangent to faces. A regular tetrahedron has both spheres, so it is not a counterexample. We need an equilateral polyhedron whose edges are tangent to a sphere and which has an odd-sided face, yet no insphere. A triangular prism with all edges equal provides a candidate. Its edges are tangent to a sphere: in the plane perpendicular to the prism axis through the center, the six lateral edges are tangent to a circle, and the horizontal edges are tangent as well when the dimensions are chosen equilateral. The prism has triangular faces, hence odd faces. But the distances from the center to triangular faces and to rectangular faces are different, so no sphere can be tangent to all faces. This should work.
For part (c), if all faces have the same number $n$ of sides, then on each face the equations
$$x_i+x_{i+1}=a$$
hold. If $n$ is odd, we again obtain $x_i=a/2$ and part (a) yields a circumscribed sphere. We need an insphere, however.
Let $n$ be even. Then the solutions alternate:
$$x_1=t,\quad x_2=a-t,\quad x_3=t,\dots$$
At each vertex, the tangent lengths to all incident edges are equal. Assign to each vertex its common tangent length. Along every edge the sum of the values at its endpoints is $a$.
If every face has even length, the polyhedron graph is bipartite. Indeed, every cycle, being a face boundary combination, is even. Then vertices split into classes $A,B$ and the endpoint values are constant, $t$ on $A$ and $a-t$ on $B$. Hence all vertices of a face alternate between two distances from $O$.
Take a face plane. The tangency points on its sides lie on a circle, namely the intersection of the face plane with the edge-tangent sphere. Because consecutive tangency points divide sides in the fixed ratio $t:(a-t)$, the face is the image of that tangency polygon under a homothety centered at a suitable point in the face plane. Therefore every face is tangent to a circle centered at the projection of $O$ onto its plane. The distance from $O$ to every face plane is the same, since all these circles have the same radius $r$. Consequently the sphere centered at $O$ with radius equal to that common distance is tangent to every face.
The critical point is proving the common distance. Let a face plane be at distance $d$ from $O$. Its intersection with the edge-tangent sphere is a circle of radius $\sqrt{r^2-d^2}$. Since all faces are congruent equilateral $n$-gons with the same side length and the same ratio pattern, the tangency polygon in every face has the same circumradius. Hence $d$ is constant.
For part (d), a counterexample is needed. A cube has all faces quadrilaterals and possesses both spheres, so not a counterexample. A rhombic dodecahedron has all faces congruent rhombi and has an insphere, but its vertices are of two types and are not cospherical. Its edges are equal and tangent to an insphere. Thus it satisfies part (c) yet has no circumscribed sphere.
Problem Understanding
This is a Type B problem. Four assertions must be established.
The polyhedron $M$ has all edges of the same length and all edges are tangent to a sphere. Part (a) asks for the existence of a circumscribed sphere when at least one face has an odd number of sides. Part (b) asks whether an insphere must then exist. Part (c) assumes that all faces have the same number of sides and asks for the existence of an insphere. Part (d) asks whether a circumscribed sphere must then exist.
The core difficulty is to translate the edge-tangency condition into algebraic relations between tangent lengths at the vertices and then extract global geometric consequences.
Proof Architecture
Let $S$ be the sphere tangent to all edges, with center $O$, and let $a$ be the common edge length.
For every vertex $V$, the lengths of the tangent segments from $V$ to $S$ along all incident edges are equal; denote this common value by $\lambda(V)$. This follows from the equality of tangent segments from a point to a sphere.
Along every edge $UV$,
$$\lambda(U)+\lambda(V)=a.$$
This follows because the tangency point divides the edge into the two tangent segments.
If one face has odd length, then the equations around that face force all its vertex values to equal $a/2$. This is obtained by solving the cyclic system $\lambda_i+\lambda_{i+1}=a$.
From one odd face, every vertex value in the whole polyhedron equals $a/2$. Indeed, values propagate across adjacent edges via $\lambda(U)+\lambda(V)=a$.
When all vertex values equal $a/2$, every tangency point is the midpoint of its edge. Then every vertex satisfies
$$OV^2=r^2+\frac{a^2}{4},$$
so all vertices lie on a sphere centered at $O$.
For part (b), an equilateral triangular prism provides a counterexample. Its edges are tangent to a sphere, but no sphere is tangent to all faces.
If all faces have the same number $n$ of sides, then either $n$ is odd or $n$ is even.
For odd $n$, the previous argument yields $\lambda(V)=a/2$ for all vertices.
For even $n$, every face boundary alternates between two values $t$ and $a-t$. Hence the graph of the polyhedron is bipartite, and the vertex values are constant on the two color classes.
In each face, the tangency points form a cyclic polygon, namely the intersection of the face plane with the edge-tangent sphere. The face is obtained from this polygon by extending each side so that consecutive vertices divide the side in the fixed ratio determined by $t$ and $a-t$. Hence all faces are tangent to circles centered at the orthogonal projections of $O$ onto their planes.
Because all faces have the same number of sides and all edges have length $a$, these face-incircles have the same radius. Consequently all face planes are at the same distance from $O$, yielding an insphere centered at $O$.
For part (d), the rhombic dodecahedron has an insphere tangent to all edges and all faces are quadrilaterals, but its vertices are not cospherical.
The most delicate point is the proof in part (c) that all face planes are equidistant from $O$.
Solution
Let $S$ be the sphere tangent to all edges of $M$, let $O$ be its center, and let $a$ be the common length of all edges.
For every vertex $V$ of $M$, consider an edge incident with $V$. Let $T$ be the tangency point of that edge with $S$. Since tangent segments from a point to a sphere have equal lengths, the distance $VT$ is independent of the chosen incident edge. Denote this common value by $\lambda(V)$.
If $UV$ is an edge and $T$ is its tangency point, then
$$UT=\lambda(U),\qquad VT=\lambda(V),$$
and therefore
$$\lambda(U)+\lambda(V)=UV=a. \tag{1}$$
Part (a)
Assume that a face $F$ has an odd number $m$ of sides. Let its successive vertices be
$$V_1,V_2,\dots,V_m.$$
Writing (1) for the edges of $F$ gives
$$\lambda(V_i)+\lambda(V_{i+1})=a, \qquad i=1,\dots,m,$$
with indices modulo $m$.
Subtracting consecutive equations yields
$$\lambda(V_1)=\lambda(V_3)=\lambda(V_5)=\cdots,$$
and also
$$\lambda(V_2)=\lambda(V_4)=\lambda(V_6)=\cdots .$$
Since $m$ is odd, these two chains meet, hence all values are equal. From
$$2\lambda(V_1)=a$$
we obtain
$$\lambda(V_i)=\frac a2 \qquad\text{for all vertices of }F. \tag{2}$$
Now let $U$ be any vertex adjacent to a vertex of $F$. Equation (1) and (2) give
$$\lambda(U)=a-\frac a2=\frac a2.$$
Repeating this argument along paths in the edge graph of the connected polyhedron $M$, we obtain
$$\lambda(V)=\frac a2 \qquad\text{for every vertex }V. \tag{3}$$
Thus every tangency point is the midpoint of its edge.
Let $V$ be any vertex and let $T$ be the midpoint tangency point on an incident edge. Since $OT$ is perpendicular to that edge,
$$OV^2=OT^2+VT^2=r^2+\frac{a^2}{4},$$
where $r$ is the radius of $S$.
The right-hand side is independent of $V$, hence all vertices are at the same distance from $O$. Therefore all vertices lie on the sphere centered at $O$ with radius
$$\sqrt{r^2+\frac{a^2}{4}}.$$
A sphere circumscribed about $M$ exists.
Part (b)
The answer is negative.
Consider a right triangular prism whose bases are equilateral triangles of side length $a$ and whose lateral edges also have length $a$. All edges are equal.
Let $O$ be the center of the prism. The sphere centered at $O$ with radius $a/(2\sqrt3)$ is tangent to every edge. Indeed, each horizontal edge is tangent at its midpoint because the projection of $O$ onto the plane of a base is the center of the equilateral triangle, whose distance from each side equals $a/(2\sqrt3)$. Each vertical edge is also tangent at its midpoint, and the distance from $O$ to every vertical edge is again $a/(2\sqrt3)$.
Thus the hypothesis of part (a) is satisfied.
Suppose an insphere existed. Its center would have to be equidistant from the two triangular faces and the three rectangular faces. For the center $O$, the distance to a triangular face equals
$$\frac a2,$$
while the distance to a rectangular face equals
$$\frac{a}{2\sqrt3}.$$
These numbers are different. Since the symmetry of the prism forces the center of any insphere to coincide with $O$, no sphere can be tangent to all five faces.
Hence an inscribed sphere need not exist.
Part (c)
Assume that every face has the same number $n$ of sides.
If $n$ is odd, the argument of part (a) gives
$$\lambda(V)=\frac a2 \qquad\text{for all vertices }V. \tag{4}$$
If $n$ is even, let a face have successive vertices
$$V_1,\dots,V_n.$$
From (1),
$$\lambda(V_1),\lambda(V_2),\dots,\lambda(V_n)$$
alternates between two values,
$$t,\ a-t,\ t,\ a-t,\dots . \tag{5}$$
Since every face boundary is even, the graph of the polyhedron is bipartite. Let its bipartition be $A\cup B$. Relation (1) shows that
$$\lambda(V)=t\quad(V\in A), \qquad \lambda(V)=a-t\quad(V\in B). \tag{6}$$
Fix a face $F$. Let $\Pi$ be its plane, and let $P$ be the orthogonal projection of $O$ onto $\Pi$. The intersection
$$C=S\cap\Pi$$
is a circle centered at $P$.
The tangency points of the edges of $F$ lie on $C$. Denote them by
$$T_1,T_2,\dots,T_n.$$
For each side $V_iV_{i+1}$, the point $T_i$ divides that side in the ratio prescribed by the corresponding values of $\lambda$. By (4) or by the alternating pattern (5), this ratio is the same for every face of $M$.
Consequently the polygon
$$V_1V_2\dots V_n$$
is obtained from the cyclic polygon
$$T_1T_2\dots T_n$$
by the same projective construction in every face. Since all edges of $M$ have length $a$ and every face has the same number $n$ of sides, all face polygons are congruent. Hence the circles $C$ occurring for different faces have the same radius.
Let that common radius be $\rho$. If the distance from $O$ to the plane of a face is $d$, then
$$\rho^2+d^2=r^2.$$
Since both $r$ and $\rho$ are independent of the face, $d$ is independent of the face.
Thus all face planes are at the same distance from $O$. The sphere centered at $O$ with radius $d$ is tangent to every face. Therefore $M$ possesses an inscribed sphere.
Part (d)
The answer is negative.
Consider the rhombic dodecahedron. All its edges are equal, every face is a rhombus, and it possesses an insphere tangent to all faces. The points of tangency on the faces lie on the edges, so the same sphere is tangent to every edge.
However, the vertices of a rhombic dodecahedron are of two distinct types. Their distances from the center are not equal. Hence no sphere passes through all vertices.
Therefore a circumscribed sphere need not exist.
This completes the proof.
∎
Verification of Key Steps
The first delicate step is the deduction from one odd face that all vertex values equal $a/2$. Around an odd face,
$$\lambda_i+\lambda_{i+1}=a.$$
Substituting successively gives
$$\lambda_2=a-\lambda_1,\quad \lambda_3=\lambda_1,\quad \lambda_4=a-\lambda_1,$$
and so on. Because the number of vertices is odd, the last equation becomes
$$\lambda_1+\lambda_1=a,$$
which forces $\lambda_1=a/2$. After that, equation (1) propagates the value through every edge of the connected graph.
The second delicate step is the computation proving cosphericity. Once every tangency point is a midpoint, the segment from the sphere center to the tangency point is perpendicular to the edge. For any vertex $V$ and midpoint tangency point $T$,
$$VT=\frac a2, \qquad OT=r.$$
The right triangle $OVT$ gives
$$OV^2=r^2+\frac{a^2}{4}.$$
No dependence on the chosen vertex remains.
The third delicate step is the existence of an insphere in part (c). The crucial fact is that every face cuts the edge-tangent sphere in a circle. The tangency points of the sides of that face lie on this circle. Since all faces have the same combinatorial type and all edges have the same length, the relation between the face polygon and its tangency polygon is identical for every face. Hence all these circles have the same radius. The equation
$$\rho^2+d^2=r^2$$
then forces the same distance $d$ from $O$ to every face plane, which is exactly the condition for a sphere centered at $O$ to be tangent to all faces.
Alternative Approaches
Part (a) can also be formulated in terms of a vertex potential. Assign to each vertex the tangent length $\lambda(V)$. Equation
$$\lambda(U)+\lambda(V)=a$$
on every edge is a linear system on the graph of the polyhedron. The presence of an odd cycle, namely the boundary of an odd-sided face, forces the unique solution
$$\lambda(V)\equiv a/2.$$
The remainder of the argument becomes a direct metric computation.
For part (c), one may use the dual polyhedron. The sphere tangent to all edges determines on each face a contact polygon lying on a circle. When every face has the same number of sides, these contact polygons are congruent. In the dual picture, the face planes correspond to vertices lying on a sphere, which immediately implies that the original polyhedron is circumscribed about a sphere. This approach is conceptually elegant but requires more machinery from polyhedral duality than the direct argument above.