Date of Degree
James L. Cox
In this dissertation, we present some enhancement methods for backtracking-search in solving multiple permutation problems. Some well-known NP-complete multiple permutation problems are Quasigroup Completion Problem and Sudoku. Multiple permutation problems have been getting a lot of attention in the literature in the recent years due to having a highly structured nature and being a challenging combinatorial search problem. Furthermore, it has been shown that many real-world problems in scheduling and experimental design take the form of multiple permutation problems. Therefore, it has been suggested that they can be used as a benchmark problem to test various enhancement methods for solving constraint satisfaction problems. Then it is hoped that the insight gained from studying them can be applied to other hard structured as well as unstructured problems. Our supplementary and novel enhancement methods for backtracking-search in solving these multiple permutation problems can be summarized as follows:
We came up with a novel way to encode multiple permutation problems and then we designed and developed an arc-consistency algorithm that is tailored towards this modeling. We implemented five versions of this arc-consistency algorithm where the last version eliminates almost all of the possible propagation redundancy. Then we introduced the novel notion of interlinking dynamic variable ordering with dynamic value ordering, where the dynamic value ordering is also used as a second tie-breaker for the dynamic variable ordering. We also proposed the concept of integrating dynamic variable ordering and dynamic value ordering into an arc-consistency algorithm by using greedy counting assertions. We developed the concept of enforcing local-consistency between variables from different redundant models of the problem. Finally, we introduced an embarrassingly parallel task distribution process at the beginning of the search.
We theoretically proved that the limited form of the Hall's theorem is enforced by our modeling of the multiple permutation problems. We showed with our empirical results that the ``fail-first" principle is confirmed in terms of minimizing the total number of explored nodes, but is refuted in terms of minimizing the depth of the search tree when finding a single solution, which correlates with previously published results. We further showed that the performance (total number instances solved at the phase transition point within a given time limit) of a given search heuristic is closely related to the underlying pruning algorithm that is being employed to maintain some level of local-consistency during backtracking-search. We also extended the previously established hypothesis, which stated that the second peak of hardness for NP-complete problems is algorithm dependent, to second peak of hardness for NP-complete problems is also search-heuristic dependent. Then we showed with our empirical results that several of our enhancement methods on backtracking-search perform better than the constraint solvers MAC-LAH and Minion as well as the SAT solvers Satz and MiniSat for previously tested instances of multiple permutation problems on these solvers.
Pay, Tayfun, "Some Enhancement Methods For Backtracking-Search In Solving Multiple Permutation Problems" (2015). CUNY Academic Works.
This work is embargoed and will be available for download on Sunday, May 27, 2018
Graduate Center users:
To read this work, log in to your GC ILL account and place a thesis request.
See the GC’s lending policies to learn more.