1
$\begingroup$

My question is: is there a standard name for a structure like the following?

For positive integers $n$, $k < n$ define a "$k$-set-free test for $n$" as a set $C$ of subsets of the integers $\{0, \dots, n-1\}$ such that: for every $i \in \{0, \dots, n-1\}$ and every $S \subset \{0, \dots, n-1\} \setminus \{i\}$ with $|S| \le k$ there exists $T \in C$ such that $i \in T$ and $S \cap T = \emptyset$.

Is this a standard combinatorial structure, or a simple transform of a standard combinatorial structure? Does anybody have a convenient reference to this sort of problem?

Follow up notes and questions.

Of interest is: bounding $|C|$.

Our application is: we have an unknown set $B \subset \{0, \cdots, n-1\}$ with $|B| \le k$ of "bad" integers and a test procedure that says a set $T$ is spoiled iff $T \cap B \neq \emptyset$. Find a small set of test-sets to identify $B$ using the test procedure.

Trivially the set $C = \{ \{ i \} | i \in \{0, \dots, n-1\} \}$ is an $n-1$-set-free test for $n$ of size $n$. However, it is easy to show for small $k$ there are $C$ with $|C|$ small (for example $|C| = 2 k \lceil 1 + (k+1) \log_2(n)\rceil $ will do, link).

The problem is a standard error-correcting coding problem if test was working over a field (i.e. counting intersection sizes instead of checking non-emptiness).

Is there an $O(k \log(n))$ sized solution? Or can one establish a lower bound that is larger than that?

$\endgroup$

1 Answer 1

2
$\begingroup$

This is an instance of the set cover problem. I used integer linear programming to obtain the minimum $|C|$ for $1\le k < n \le 12$. Do any of these results surprise you?

\begin{matrix} n\backslash k &1 &2 &3 &4 &5 &6 &7 &8 &9 &10 &11 \\ \hline 2 &2 \\ 3 &3 &3 \\ 4 &4 &4 &4 \\ 5 &4 &5 &5 &5 \\ 6 &4 &6 &6 &6 &6 \\ 7 &5 &7 &7 &7 &7 &7 \\ 8 &5 &8 &8 &8 &8 &8 &8 \\ 9 &5 &9 &9 &9 &9 &9 &9 &9 \\ 10 &5 &9 &10 &10 &10 &10 &10 &10 &10 \\ 11 &6 &9 &11 &11 &11 &11 &11 &11 &11 &11 \\ 12 &6 &9 &12 &12 &12 &12 &12 &12 &12 &12 &12 \\ \end{matrix}

Also, this type of problem is called group testing.

$\endgroup$
1
  • $\begingroup$ The references are great and exactly what I needed. The table results are not surprising, as the method isn't going to work until n is much larger than k. Thanks! $\endgroup$
    – John Mount
    Commented Apr 14, 2020 at 14:41

You must log in to answer this question.

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