0
$\begingroup$

If I randomly select subsets $K$ and $R$ of set $N$, what is the expected value of $|K \cap R|$ i.e. the number of elements in both $K$ and $R$?

EDIT: Clarifications: $N$ is finite and of known size, and $K$ and $R$ are uniformly distributed over subsets of given sizes $k$ and $r$. In other words, I'm looking for a function $P(k, r, n)$ that returns $|K \cap R|$.

I initially tried modelling this as a computation. If I select an element from $R$ and try to add it to $K$, the probability that said element is already in $K$ is $\frac{k}{n}$ ($k = |K|$; $n = |N|$; $r = |R|$). So the expected number of elements in $K$ after I add an element from $R$ should be $\frac{k}{n} k + \frac{n - k}{n} \left ( k + 1 \right )$. Following this train of logic gives us the recurrence relation

$P(k, r, n) = \frac{k}{n} P(k, r - 1, n) + \frac{n - k}{n} P(k + 1, r - 1, n)$

with the obvious boundary conditions $\forall k, n: P(k, 0, n) = 1$ and $\forall n, r: P(n, r, n) = 0$. (Finding $P(k, r, n)$ will obviously give us $|K \cup R|$, not $|K \cap R|$, but then $|K \cap R| = |K| + |R| - |K \cup R|$.) But I have no idea how to turn this recurrence relation into a closed-form, and I'm not even fully confident that it's correct in the first place. I would just write a C program to calculate $P(k, r, n)$ for me, but I'm dealing with values of $k$ and $r$ that are large enough (up to $2^{16}$) that I would blow my stack space.

  • 0
    Just to clarify: $N$ is a finite set, and you're picking $K$ and $R$ using the uniform meausre on $\mathcal{P}(N)$ (so all subsets of $N$ are equally likely)? – 2017-02-13
  • 0
    Also, I'm a massive idiot and just rubber-ducked myself into a viable Monte Carlo algorithm that'll work in O(r) time and O(1) space per iteration. Still curious to see if analytic solutions exist. – 2017-02-13

2 Answers 2

3

We have $$ |K \cap R| = \sum_{x \in N}1_{x \in K \cap R} = \sum_{x \in N}1_{x\in K}1_{x\in R} $$ and taking expected values and using independence, $$ E|K \cap R| = \sum_{x \in N}P(x \in K) P(x \in R). $$ So symmetry demands that $P(x \in K), P(x \in R)$ don't depend on $x$, only $k$ and $r$. The question remains, how many total subsets of $N$ have size $k$, and of those how many contain some fixed $x$?. By definition, $\binom{n}{k} = \frac{n!}{k!(n-k)!}$ is how many ways there are to choose a subset of size $k$ from a set of size $n=|N|$. Choosing a subset of size $k$ containing $x$ is the same as choosing a subset of size $k-1$ from $N\setminus\{x\}$, so there are $\binom{n-1}{k-1}$ ways of doing this. Thus $$ E|K \cap R| = |N| \frac{\binom{n-1}{k-1}}{\binom{n}{k}}\frac{\binom{n-1}{r-1}}{\binom{n}{r}} $$ and I believe you are capable of simplifying from here.

  • 0
    Your answer has led me to realize a missing piece of information from the original question. Thank you for the accidental correction; I will update the question momentarily. – 2017-02-13
  • 0
    See updated answer – 2017-02-13
1

When you need to calculate the expectation of the cardinality of some set, its usually convinient to express this value as the sum of indicator functions. Let $X_n, n \in N$ be a random variable that is an indicator of the event $n \in K \cap R$. Note that $|K \cap R| = \sum_{n \in N} X_n$. Now, using the linearity of expectation and uniform distribution of subsets $$\mathbb{E}|K \cap R| = \sum_{n \in N} \mathbb{E}X_n = \sum_{n \in N}\mathbb{P}(n \in K \cap R) = |N| \cdot \mathbb{P}(n \in K \cap R).$$ To calculate the last probability, note that to choose a pair of subsets of $N$ with $n \in K \cap R$ is the same as fixing $n$ and choosing a pair of subsets of $N \setminus \{n\}$. So $\mathbb{P}(n \in K \cap R) = \frac{2^{2(|N| - 1)}}{2^{2|N|}} = \frac{1}{4}.$ Finally, we get $\mathbb{E}|K \cap R| = \frac{|N|}{4}$.