Grant: Discovery Grants Program - Individual

The speed and power of modern computers is truly impressive. Given the regular announcements of more and more powerful super-computers, one might believe any reasonable problem can be solved, by brute force, in reasonable time. Surprisingly this is not the case. There are certain problems that seem intractable by nature. Classic examples of such problems include scheduling, vehicle routing, and facility location problems. All of these problems maybe viewed as assigning resources (time slots, routes, locations) to some consumer (exam writing students, vehicles, factories). The goal is to assign the resources in the most efficient way possible, subject to constraints restricting the assignments. In general we call these Constraint Satisfaction Problems or CSPs. The challenge in computing the most efficient assignment is the sheer number of possibilities. For each resource there can be many ways to assign it to a consumer. Moreover, each choice one makes may affect what other choices can be made in the future. In this way we are forced to consider all the possible combinations of assignments of resources to consumers. For a fast computer to exhaustively search through all the possible combinations looking for the optimum assignment could take thousands of years. This is an example of "combinatorial explosion"; the sheer number of combinations defeats brute force solutions. More efficient algorithms, if they exist, must be employed. The primary aim of my program is to understand the nature of these challenging combinatorial problems. It turns out that some of these problems contain (mathematical) structure. This structure can be exploited to produce efficient algorithms, i.e. we do not need to consider all possible resource assignments, but rather we can zoom-in on the optimal solution. On the other hand, some problems seem to lack sufficient structure to admit an efficient solution. At the highest level the goal of my program is to identify those CSPs which admit efficient algorithms (polynomial time solvable) and those which are intractable (NP-complete). The long term goal is to classify families of related problems by giving a structural description of the efficient versus hard dichotomy. Many efficient algorithms are based on mathematical structure known as "good characterizations". A good characterization guides one through the combinatorial explosion to an optimal solution for the CSP or to a proof that the CSP has no solution, i.e. there are too many constraints, at which point the search can stop. In order to use this mathematical structure, we often need to develop new mathematics. The develop of this theory can be used to solve particular CSPs, but more importantly it gives insight into the fundamental dichotomy question above. My research program focuses on combinatorial problems from graph homomorphisms and graph packings and coverings.