dc.description.abstract |
The Stable Marriage Problem is a classic problem of matching under preferences.
It involves two sets of agents, classically refered to as men and women each having a
preference list over the other set. A matching is a disjoint collection of man woman
pairs. We define a blocking edge with respect to a matching to be a (man,woman) pair,
say (m,w), such that both m and w prefer each other over their matched partners in
the matching. A Stable Matching is a matching with no blocking pairs. In the 1960s
Gale and Shapley gave an algorithm, now known as the Gale-Shapley algorithm, that
solves the stable marriage problem in polynomial time.
Various extentions of the Stable Marriage problem have been studied. The simplest
and most natural extension is to incomplete lists, called a the Stable Marriage
problem with incomplete lists, or SMI. As the name suggests an instance of this
problem would involve the preference lists to be incomplete. The GS algorithm can
be easily extended to this problem and again it has been shown that a stable matching
always exists and can be found in polynomial time. If indifference is introduced
in the problem instance, that is the preference lists may involve ties and are incomplete
as well, the problem is referred to as The Stable Marriage problem with Ties
and Incomplete Lists or SMTI in short. With ties there are three different notions of
stablity - super stability, strong stability and weak stability. We are only interested in
weakly stable matchings. A matching in an SMTI instance is weakly stable if there is
no pair (m,w) such that m and w both strictly prefer each other over their respective
matched partners.
It has been shown that arbitrarily breaking the ties in an SMTI problem instance I
we get a corresponding SMI instance I� such that a stable matching in I� is a weakly
stable matching in I. Depending on how we break the ties the size of the stable
matching obtained may vary. The problem of finding a maximum sized or a minimum
sized weakly stable matching is shown to be NP-hard. When the underlying graph
is not a bipartite graph, the problem is commonly known as the Stable Roommates
problem. We can define versions of SMI and SMTI for the Stable Roommates setting
as well and we call them SRI and SRTI respectively. Ronn showed that in an SRTI
instance the problem of deciding whether a weakly stable matching exists or not is
NP-complete.
In this dissertation we study the problem of finding a maximum (minimum) sized
weakly stable matching in SMTI and SRTI instances in the realm of parameterized
xi
xii
complexity. We show that these problems are fixed parameter tractable by giving
polynomial kernels. Also when the underlying graph is planar we show that we can
improve the kernel further to a linear kernel thus giving us better algorithms.
Another extension of the Stable Marriage problem gives us the well known Hospital/
Residents Problem (HR Problem). This problem is a many-one stable matching
problem. An instance of Hospital/Residents Problem is again a bipartite graph,
H ∪ R, where H is the set of hospitals and R the set of residents. Each hospital has
a strict preference over a subset of residents and each resident has a strict preference
over a subset of hospitals. With every hospital we have an associated number called
its upper quota. Our goal is to assign residents to hospitals such that for every hospital
not more than its upper quota residents are assigned to it, and the assignment
is ‘stable’. Again a stable assignment is an assignment with no blocking pairs, but
the definition of a blocking pair is slightly different than the one defined for a stable
marraige instance. The GS algorithm can be used in this setting as well to find a
stable matching and again a stable matching always exists.
When we associate a lower quota with every hospital and add an additional constraint
to the problem of assigning at least lower capacity number of residents to each
hospital. Now we want to find such a feasible matching which is as stable as possible.
A stable matching found by GS may not be feasible and by the Rural Hospital Theorem
which states that all stable matchings match the same set of vertices, we are
assured that no stable matching would be feasible. So we look for a feasible matching
as ‘stable’ as possible. In other words we look for a matching with either fewest number
of blocking edges or fewest number of blocking residents. Both these problems
of finding a feasible matching minimizing the number of blocking edges or blocking
residents are shown to be NP-hard. We study the Hospital/Residents problem with
lower quotas in the realm of parameterized complexity. We give an XP Algorithm
with parameter blocking residents. For other parameters sum of lower quotas and
colleges with positive lower quota we show that the problem is W[1] hard. |
en_US |