5
$\begingroup$

Suppose that ${n\choose k}, {n-1\choose k-1}, \ldots, {n-k+1\choose 1}$ are all even. (This happens for example if $k=2^\alpha-1$ and $n=2k$.) In this case, can we select ${n\choose k}/2$ sets of size $k$ from an $n$ element set such that any $i<k$ elements are contained in exactly ${n-i\choose k-i}/2$ of the selected sets?

This would be somekind of parity for the sets. If true, I am also interested in the natural generalization for $p\ne 2$ to define some $\!\!\mod p$ of a set.

The only nontrivial example that I know is for $k=3$ and $n=6$ which I have just learned from Douglas Zare who might have heard it from Robin Chapman, see his comment here.

$\endgroup$
2
  • $\begingroup$ There's a slightly larger example using $p=5$, $n=8$ and $k=4$, where each element is in $7$ sets, each pair is in $3$, and each triple is in $1$. It basically designs itself: take any set of size $4$ (e.g. $1234$), the complement of that (so $5678$), and then two sets for each pair in the first set ($12xx$, $12xx$), filled in using disjoint parts of the complement (like $1256$, $1278$) so that all three partitions of the complement are used for each element in the first set. $\endgroup$ Apr 22, 2014 at 22:25
  • $\begingroup$ I might've made it seem more complicated than it is. $\endgroup$ Apr 22, 2014 at 22:26

2 Answers 2

4
$\begingroup$

I wish I had a real answer for you!

You are essentially interested in a tough conjecture of Hartmann, known as the "halving conjecture", which is promoted heavily by Reza Khosrovshahi. Actually, the conjecture is more general than what you are asking, since it concerns large sets of $t$-designs with arbitrary block size $k$, and I believe you have specialized to $t=k-1$. In other words, the more general conjecture asks if your condition holds for $i \le t$ when only the first $t+1$ of your binomial coefficients are even.

Here is the closest thing I know to an open-access survey: http://math.ipm.ac.ir/combin/publications/files/2009/PA0900076.pdf

See Section 6 in particular.

The halving conjecture is known to be true for $2$-designs. I haven't read the proof, but it is apparently nontrivial. In any case, this means your question has an affirmative answer for $k=3$ and all admissible $n$. For general $t$, quite a few partial results are known, such as product, puncturing, and complementing constructions. There are also important connections with null designs and trades.

I hope some of this helps for your particular application(s).

$\endgroup$
5
$\begingroup$

Here are some solutions for the largest possible set of conditions, $n=2k$.

Suppose $p$ is a prime that divides ${2k\choose k}, {2k-1\choose k-1}, \ldots, {k+1\choose 1}$. Then the selection conditions in your question (that is, $b={2k\choose k}/p$ blocks, with each element in $r={2k-1\choose k-1}/p$ of them, and each pair in ${2k-2\choose k-2}/p$, etc.) would create a $t$-$(2k,k,\frac{k+1}{p})$ design with $t=k-1$.

With $k=2$ we want a partition of a $4$ element set into two sets of size $2$, which certainly exists. The example you had in mind is $k=3$, and the one in my comment is $k=4$, so those exist.

When $k=5$, there are no primes dividing everything up to ${2k \choose k}$.

When $k=6$, we are looking for a $5-(12,6,1)$ design, which is uniquely the small Witt design.

When $k=7$, Donald Kreher and Stanislaw Radziszowski constructed a $6-(14,7,4)$ design in J. Combin. Theory (vol 43.2, Nov 1986) http://dx.doi.org/10.1016/0097-3165(86)90064-6

Finally, for $k\geq8$, Donald Kreher names the $7-(16,8,3)$ as the smallest unsettled $k-1$-design with $\lambda \neq 1$ in "Simple t-designs with large t: A survey" (which is an inaccesible preprint on his current homepage, but can be found here https://www.math.ucdavis.edu/~deloera/MISC/BIBLIOTECA/trunk/Moura/large_tdesign_survey.pdf), and notes that the only other known $k-1$-designs for $k \geq 8$ have $\lambda = k!^{2k-1}$, which is not the case here.

This still leaves open the cases where $n > 2k$, for example, the trivial design with $p=3$, $n=7$, $k=2$, or the $3-(13,4,2)$ design with $p=5$, which exists.

$\endgroup$
2
  • $\begingroup$ Thank you for your detailed answer! Then summarizing, it is possible for every value of $k$, $n$ and $p$ to have such a set system that I ask for, right? $\endgroup$
    – domotorp
    Apr 23, 2014 at 10:33
  • $\begingroup$ Provided $p$ divides all of those binomials, there's no obvious obstruction. So they seem possible, but they have not been constructed or shown to exist. And by the same criteria, a projective plane of order 10 seems possible, and isn't. Your best bet for finding a counterexample is the case $k=3$, and your best bet for proving they all exist is time travel. $\endgroup$ Apr 23, 2014 at 15:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.