4
$\begingroup$

Consider the complete graph on n vertices. How many ways are there to form this graph by unioning together triangles?

The triangles are distinct but may overlap.

$\endgroup$
1
  • $\begingroup$ A little less than 2^{n choose 3}. A large subset that has to fail to cover must omit at least n-2 triangles which share an edge, so the number of triangle sets which fail to cover is much smaller. Gerhard "Did You Need Something Exact?" Paseman, 2018.04.04. $\endgroup$ Commented Apr 5, 2018 at 2:42

2 Answers 2

13
$\begingroup$

For $3\le n\le 12$:
1, 5, 388, 477965, 19199206747, 48058624241034238, 14791854612528152343049939, 112039538006858119010793653395340973, 42077073837090084518028123082446810913910363417140, 1580746220785841888200914891451910938106730059214980821153579470917.

The next value is computable but expensive. I added it to OEIS as A302394.

Let $\mathcal{G}_n$ be the set of unlabelled graphs of order $n$. For $G\in\mathcal{G}_n$, let $a(G)$ be the order of its automorphism group, $e(G)$ be the number of edges in its complement, and $t(G)$ be its number of triangles. Then the number of sets of triangles which cover every vertex-pair is $$ \sum_{G\in\mathcal{G}_n} \frac{n!}{a(G)} (-1)^{e(G)} 2^{t(G)}.$$

The formula is a standard inclusion-exclusion. $n!/a(G)$ is the number of labelled graphs isomorphic to $G$. An event is a vertex-pair that is not covered. $2^{t(n)}$ is the number of sets of triangles that don't cover any edge in the complement of $G$.

If $t_r=t_r(n)$ is the contribution to the above sum for $e(G)=r$, apart from the $(-1)^{e(G)}$, then the summation $t_0-t_1+t_2-\cdots$ has the Bonferroni property: its partial sums alternate above and below the full sum. I calculate: $$\eqalign{ t_0 &= 2^{\binom n3} \\ t_1 &= t_0 \, (n-1)\,n\,2^{1-n} \\ t_2 &= t_0 \, (n-1)(n-2)(n+5)\,n \,2^{1-2n} \\ t_3 &= t_0 \, \tfrac13\,(n-1)(n-2)(n^3+12n^2+39n-220)\,n\,2^{2-3n}, }$$ which gives very narrow bounds already.

$\endgroup$
7
  • $\begingroup$ Thanks for independently counting. Setting alpha to (n choose 2)/2^{n-2}, it looks like the next largest term is roughly alpha^2 of the total, suggesting that a good approximation for n bigger than 10 would use a geometric series or (1/(1+alpha)) of the total number of subsets of triangles. Does this intuition seem useful, or are the lower order terms much larger than the tail of the geometric series? Gerhard "Is Wondering About Fat Tails" Paseman, 2018.04.05. $\endgroup$ Commented Apr 6, 2018 at 4:42
  • 1
    $\begingroup$ @GerhardPaseman It's a bit more irregular than that. I'll add more. $\endgroup$ Commented Apr 6, 2018 at 8:19
  • 2
    $\begingroup$ This sequence deserves to be added to OEIS, I think. $\endgroup$ Commented Apr 6, 2018 at 12:49
  • 1
    $\begingroup$ Thank you for mentioning the terms. Do you mean to write $t_0 + (-1)^1 t_1 + (-1)^2 t_2 + \cdots$ ? Gerhard "Or Is Something Else Alternating?" Paseman, 2018.04.06. $\endgroup$ Commented Apr 6, 2018 at 16:18
  • 1
    $\begingroup$ Good. I understand there is a polynomial of degree 2 in n as part of term 1, and one of degree 6 for term 3. Shouldn't the polynomial part of term 2 be degree 4? Gerhard "Haven't Checked it Myself Yet" Paseman, 2018.04.06. $\endgroup$ Commented Apr 6, 2018 at 23:34
3
$\begingroup$

Leaving the cases of $n \lt 3$ to others, I note there is exactly one such subset of triangles when $n=3$ and 5 subsets (whose edge union is the set of all edges of $K_n$) when $n=4$. I am guessing the name of the game is to enumerate such distinct subsets (effectively meaning the edges are labelled).

For $n=5$, the game becomes a challenge. $K_5$ contains 10 triangles, and of the collections of triangles at most four in number, one finds 10 subsets which cover. This is because such covers require three triangles to share an edge. Every subset of 8 or more triangles cover, while only 10 subsets of 7 triangles do not cover. For each of these 10 subsets of 7 triangles, there are 7 subsets of six which also do not cover and 21 subsets of 5, of which six of these 21 are involved in double counting, giving 10 + 70 + 150 + 6*10/2 = 260 subsets of five to seven triangles which do not cover. Combine this with the 1+ 10 + 45 +120 + (210-10) small subsets which do not cover gives 636 non covering subsets of triangles, meaning 388 subsets do cover for $n=5$.

For larger n, the fraction of all subsets that do not cover is less than (n choose 2)/2^(n-2), which is less than 1 for $ n \gt 5 $, and decreases almost exponentially. Thus for n at least ten, greater than 80% of all the subsets of triangles cover the graph, so 2^(n choose 3) is within a small multiplicative factor of the correct number of covering subsets. Assuming montonicity of the growth of this ratio for n greater than 3, at least 5/16 of all subsets of triangles cover the graph.

Gerhard "That's A Lot Of Subsets" Paseman, 2018.04.05.

$\endgroup$
3
  • $\begingroup$ The sequence 1, 5, 388 is currently unknown to the Online Encyclopedia of Integer Sequences. $\endgroup$ Commented Apr 6, 2018 at 0:40
  • $\begingroup$ Maybe I should check my work. Gerhard "Or Let Someone Else Check" Paseman, 2018.04.05. $\endgroup$ Commented Apr 6, 2018 at 1:32
  • $\begingroup$ 1,5,388 are correct, no need to check. $\endgroup$ Commented Apr 6, 2018 at 3:43

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .