9
$\begingroup$

This question is more of a graph problem. Suppose $A=\{1,2,..,n\}$, and we want to construct the sets $B(i)$,$1\leq i\leq n$, such that $i \not\in B(i)$ and the following constraints on the set $B(i)$ are satisfied:

1) If $j \in B(i)$ then $ i \notin B(j)$,

2) For all $i \in B(j)$, $B(j) \setminus \{i\} \not \subset B(i)$,

3) For all $i \notin B(k)$, $B(i) \not \subset {\{k\} \cup B(k)}$

Can we construct such subsets for all values of $n$? I can prove that you can construct such subsets for $n=4,5,6,..10$, but I can not generalize the proof for all values of $n$. I believe that such subsets cannot be constructed in general.

$\endgroup$
11
  • $\begingroup$ If you think of the "dual" problem, then your question might be a relative of this mathoverflow.net/questions/268798/… $\endgroup$ Commented May 24, 2017 at 16:20
  • 2
    $\begingroup$ A random tournament works for large $n$. $\endgroup$ Commented May 24, 2017 at 16:21
  • $\begingroup$ @Fedor: Could you elaborate on it or give me some link for reference. $\endgroup$ Commented May 24, 2017 at 17:37
  • $\begingroup$ The dual problem of this seems to be messy. If you look at it from the graph theory perspective I am claiming that when the graph has only directed edge from i to j then it cannot have an edge from j to i. Also after constructing such subsets for all 1,2,...,n ; this condition would never get violated. This seems to be difficult considering the subset constraints that are present. Please let me know if you get better insights on it. If possible can you tell me what you think is a dual problem for this. $\endgroup$ Commented May 24, 2017 at 17:48
  • 1
    $\begingroup$ This question reminds me a lot of a 2003 Romanian Math Olympiad problem given for the 9th grade (I added the grade just in case one wants to look for the exact formulation). The conditions seemed a bit more mild in that problem. They only asked for $i \not\in B(i)$ plus condition 1) plus disjointness of the $B(i)$s. Then answer was that such configurations exist only for $n \geq 7$. For every $n \geq 7$ they gave explicit examples. $\endgroup$
    – Alex
    Commented May 29, 2017 at 10:39

2 Answers 2

6
$\begingroup$

Consider a random tournament on $n$ vertices: for any two distinct vertices $i$, $j$ either $i\in B(j)$ or $j\in B(i)$ with probability $1/2$, all $\binom{n}2$ pairs $(i,j)$ are independent. Then the probability that for given vertices $i,k$ we have $B(i)\subset k\cup B(k)$ does not exceed $(3/4)^{n-2}$, since for any vertex $x\ne i,k$ the probability that $x\in B(i)\setminus B(k)$ equals $1/4$ and these events are independent. Using union bound, we see that the probability that some pair violates your condition does not exceed $n(n-1)(3/4)^{n-2}<1$ for large $n$.

$\endgroup$
3
  • $\begingroup$ But the thing is i $\in$ B(i) is not allowed. That is the additional constraint. In graph terms you cannot have self loops. $\endgroup$ Commented May 25, 2017 at 8:01
  • 1
    $\begingroup$ of course, no loops, this is a tournament $\endgroup$ Commented May 25, 2017 at 9:15
  • $\begingroup$ @მამუკაჯიბლაძე of course, thank you $\endgroup$ Commented May 25, 2017 at 14:11
1
$\begingroup$

Encouraged by the proof in the answer by Fedor Petrov, I started random search and found such subsets for each $n\leqslant200$ starting from $n=7$, which in addition satisfy the condition imposed by Fedor: for all $i\ne j$, either $i\in B(j)$ or $j\in B(i)$. Here is one for $n=7$: $$ \begin{aligned} B(1)&=\{2,3,5\},\\ B(2)&=\{3,4,6\},\\ B(3)&=\{4,5,7\},\\ B(4)&=\{5,6,1\},\\ B(5)&=\{6,7,2\},\\ B(6)&=\{7,1,3\},\\ B(7)&=\{1,2,4\}. \end{aligned} $$ As noticed in the comment by Gerhard "Nothing Escapes Him" Paseman, this one can be realized by a bijection between points and lines of the Fano plane.

Later

Here is, in fact, a (one of many) simple inductive construction of a tournament $B_n$ on vertices $\{1,...,n\}$ with needed properties for each $n\geqslant7$.

For $n=7$ just take $B_7(k)=B(k)$, $k=1,...,7$, with $B$ the one above.

Then for $n>7$ let $B_n(n)=B_7(7)=\{1,2,4\}$ for $n$ odd and $B_n(n)=B_7^{-1}(7)=\{3,5,6\}$ for $n$ even, and $$ B_n(k)= \begin{cases} B_{n-1}(k)\cup\{n\},&k\notin B_n(n),\\ B_{n-1}(k),&k\in B_n(n) \end{cases} $$ for $k=1,...,n-1$. That's it.

Still later

The proof seems to be not entirely trivial, so I decided to supply at least a sketch of it.

Let us denote for simplicity $i\in B(j)$ by $j\to i$. Note that for tournaments (i. e. when exactly one of $i\to j$ or $j\to i$ holds iff $i\ne j$), the requirements are equivalent to the existence, for all $i\to j$, of a $k$ with $i\to k\to j$.

Now by induction, if $B_{n-1}$ satisfies this, we have to check the above for $n\to i$ and for $i\to n$.

Now if $n\to i$ holds (i. e. $i\in B_n(n)$), then, noting that each $B_n(n)$ is in fact a 3-cycle (either $1\to2\to4\to1$ or $3\to5\to6\to3$), there always is a $j\in B_n(n)$ (i. e. $n\to j$) with $j\to i$.

And if $i\to n$ holds then $i\notin B_n(n)$, and, again by induction (see below), there is a $j\notin B_n(n)$ (hence $j\to n$) with $i\to j$.

To complete the induction, it remains to show that ($n\notin B_{n+1}(n+1)$ and) there is a $j\notin B_{n+1}(n+1)$ with $n\to j$. This is clear since $B_n(n)$ and $B_{n+1}(n+1)$ do not intersect, so we can take any $j\in B_n(n)$.

$\endgroup$
6
  • 1
    $\begingroup$ This looks familiar. Gerhard "Can You Say Fano Plane?" Paseman, 2017.05.25. $\endgroup$ Commented May 25, 2017 at 18:53
  • $\begingroup$ Thanks for the acknowledgement. I am thinking about such a bijection for other finite projective planes. If you take the (Hasse diagram of the) inclusion lattice, you may be able to find an arrangement and labelling that reflects the poster's property 1). Gerhard "Feel Free To Solve It" Paseman, 2017.05.25. $\endgroup$ Commented May 25, 2017 at 19:23
  • $\begingroup$ @GerhardPaseman The ones I found have the $B(i)$ of size about half of $n$ on average, and $B(i)\cap B(j)$ of size about $1/4$ of $n$, so these look more like half-dimensional subspaces rather than lines... $\endgroup$ Commented May 25, 2017 at 19:30
  • $\begingroup$ For the projective plane parameter n=5, there should be 31 points and 31 lines each with 6 points. Does a labelling show up for that? Gerhard "Hope To Catch This One" Paseman, 2017.05.25. $\endgroup$ Commented May 25, 2017 at 19:40
  • $\begingroup$ @GerhardPaseman With my random ones, sizes of $B(1)$, ..., $B(31)$ range somewhere between $10$ and $20$, and their pairwise intersections roughly between $3$ and $15$ $\endgroup$ Commented May 25, 2017 at 19:58

You must log in to answer this question.

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