7

Is the following conjecture true: For $n$ non-overlapping circles (or disks) in a triangle, the maximum number of tangent points is $3n$.

The tangent points include points where two circles are tangent, and points where a circle is tangent to a side of the triangle.

For example, here are $7$ circles and $21$ tangent points.

enter image description here

My thoughts

The conjecture is not true for other polygons: for example, a single circle inscribed in a square has four tangent points.

Could this be related to the fact that a circle can be surrounded by at most six other circles of the same size?

I am not very familiar with graph theory, but I suspect that may be used here.

Context

I was playing with circles in triangles, and I started to wonder about he maximum number of tangent points.

Dan
  • 35,053

2 Answers2

9

I'm not entirely sure but I think the following works: Given a triangle and $n$ circles as above. Let us define a graph $G$ as follows.

Take $n+3$ vertices, identifying each circle with a vertex plus three additional vertices $u_1,u_2,u_3$ corresponding to the three sides of the triangle.

Now we add an edge between any two vertices, when the corresponding circles share a tangent point. Also add edges between $u_i$ and vertices corresponding to circles that are tangent to the corresponding triangle side.

One can check that $G$ defined in this way is planar. A famous result of planar graph theory is that a planar graph on $m\ge 3$ vertices can have at most $3m-6$ edges.

Therefore $G$ has at most $3(n+3)-6 -3 = 3n$ edges (because the edges $u_1u_2, u_1u_3, u_2u_3$ could be added to $G$ and it would remain planar). The edges in our graph correspond exactly to the tangent points. So we have $3n$ at most.

EDIT:

Here's a picture of how $G$ would look like for the $7$ circles and the triangle as in the question. The nodes $1-7$ correspond to the circles in the picture. triangle graph

Immanuel
  • 576
3

@Immanuel's answer answered the question perfectly. Here, I will slightly generalize that answer and show that the upper bound can be attained.


Consider $n$ non-overlapping circles inside a $k$-gon. As in @Immanuel's answer, we consider a vertex for each of the $n$ circles and each edge of the $k$-gon, so there are $n+k$ vertices. We add an edge between two vertices if the corresponding objects touch. This again yields a planar graph $G$, so the total number of edges is at most $3(n+k)-6$.

In fact, we can add $2k-3$ more edges to $G$ and still get another planar graph $G’$. These edges are added between the $k$ vertices associated with the edges of the $k$-gon: $k$ of these form a $k$-gon, and the remaining $k-3$ form diagonals on the outside. Therefore, the bound $3(n+k)-6$ applies to $G’$. This then gives the bound $$ 3(n+k)-6-(2k-3) = 3n+k-3, $$ on the maximum number of edges of $G$.


In fact, this bound can be achieved very easily. Start with a regular $k$-gon with a circle inscribed in it. This satisfies the bound for $n=1$. Pick a vertex of the $k$-gon, and consider the two edges that are incident to it. Add a new circle that touches the previous circle and these two edges. This adds $3$ tangent points for each new circle.

Pranay
  • 5,390
  • I understand why you subtract $2k-3$, but the wording confused me at first: you wrote "we can add $2k-3$ more edges...", then you subtract $(2k-3)$. – Dan Jan 05 '25 at 21:41
  • 2
    @Dan. I can see why it’s confusing. I added because the graph constructed in the previous paragraph doesn’t have these edges. My intention was to say that we can add these edges while still keeping the graph planar and satisfying the bound $3(n+k)-6$. And then we subtract these edges to go back to the graph that we are interested in. – Pranay Jan 05 '25 at 21:44
  • @Dan I edited my answer to clarify what I meant. I’m sure you already understand what I meant, but I hope the wording is clearer now. – Pranay Jan 05 '25 at 21:47
  • 1
    Perhaps still a little confusing, but maybe it's just me. I would say it like this: Planar graph $G$ has at most $3(n+k)-6$ edges, but $2k-3$ edges do not correspond to tangent points; rather, they correspond to... so we subtract them from $3(n+k)-6$." Anyway, thanks for generalizing, and showing that the upper bound can be attained. – Dan Jan 05 '25 at 21:53
  • 3
    Also a really nice result. When something is $3n$ for $k = 3$ my natural guess would be that it is $4n$ for $k = 4$, not $3n + 1$. Very nice to have my intuition corrected here! – Vincent Jan 06 '25 at 22:34