Title: MDS Variable Generation and Secure Summation with User Selection

Abstract: A collection of $K$ random variables are called $(K,n)$-MDS if any $n$ of the $K$ variables are independent and determine all remaining variables. In the MDS variable generation problem, $K$ users wish to generate variables that are $(K,n)$-MDS using a randomness variable owned by each user. We show that to generate $1$ bit of $(K,n)$-MDS variables for each $n \in \{1,2,\cdots, K\}$, the minimum size of the randomness variable at each user is $1 + 1/2 + \cdots + 1/K$ bits.
An intimately related problem is secure summation with user selection, where a server may select an arbitrary subset of $K$ users and securely compute the sum of the inputs of the selected users. We show that to compute $1$ bit of an arbitrarily chosen sum securely, the minimum size of the key held by each user is $1 + 1/2 + \cdots + 1/(K-1)$ bits, whose achievability uses the generation of $(K,n)$-MDS variables for $n \in \{1,2,\cdots,K-1\}$.