Paralelização da reconstrução de geometrias moleculares

Abstract

In this thesis we present an methodology to parallelize the Discrete Molecular Distance Geometry Problem (DMDGP) using Dataflow. The method consists in partitioning the molecule according to its intrinsic characteristic, the symmetry vertices. The idea is to break the molecule into parts according to the distribution of some very specific vertices, to solve each part in parallel and then to join the partial solutions found by using rotation matrices. We will also deal with the parallelization of Interval Discretizable Molecular Distance Geometry Problem (iDMDGP) using Dataflow. However, instead of breaking the molecule, we will partition the search space. We present some computational experiments performed and analyze the behavior of the proposed approaches as a function of the number of cores employed. The results were encouraging, gains were obtained (speedups) in the vast majority of the tests performed, in some of these reached speedups above 12, which demonstrates the effectiveness of the approaches employed.