Parameterized Complexity of Incomplete Connected Fair Division.

\textit{Fair division} of resources among competing agents is a fundamental problem in computational social choice and economic game theory. It has been intensively studied on various kinds of items (\textit{divisible} and \textit{indivisible}) and under various notions of \textit{fairness}. We focus on Connected Fair Division (\CFDO), the variant of fair division on graphs, where the \textit{resources} are modeled as an \textit{item graph}. Here, each agent has to be assigned a connected subgraph of the item graph, and each item has to be assigned to some agent. We introduce a generalization of \CFDO, termed Incomplete Connected Fair Division (\CFD), where exactly $p$ vertices of the item graph should be assigned to the agents. This might be useful, in particular when the allocations are intended to be ``economical'' as well as fair. We consider four well-known notions of fairness: \PROP, \EF, \EFO, \EFX. First, we prove that \EF-\CFD, \EFO-\CFD, and \EFX-\CFD are W[1]-hard parameterized by $p$ plus the number of agents, even for graphs having constant \textit{vertex cover number} ($\mathsf{vcn}$). In contrast, we present a randomized \FPT algorithm for \PROP-\CFD parameterized only by $p$. Additionally, we prove both positive and negative results concerning the kernelization complexity of \CFD under all four fairness notions, parameterized by $p$, $\mathsf{vcn}$, and the total number of different valuations in the item graph ($\mathsf{val}$).