Development of methods for structure prediction and characterising the lattice energy surfaces of organic molecular crystals

This work focuses on the sampling and classification of lattice energy landscapes of organic molecular crystal systems. For the prediction of crystal structures, we combine quasi-random structure generation with Monte Carlo global optimization, combining the low-discrepancy sampling provided by quasi-random sequences with the efficiency of global optimization at locating low energy structures. Simulated annealing and basin hopping are both implemented as part of this work as a means of global optimization. The former is constrained by program efficiency, and conflicts between the global and local searching. Through tests on a set of single-component molecular crystals and co-crystals, the latter combined method of quasi-random searching and basin hopping (QR-BH) is demonstrated, providing more efficient location of low energy structures than pure quasi-random sampling, and maintaining the efficient location of higher energy structures that are essential for identifying important polymorphs. Looking beyond local minima, the transition states between structures hold significant information about the energy landscape. Connectivity is sampled by threshold algorithms in both single-component crystals and co-crystals. Transitions between experimental structures from different space groups are explored with P1 unit cells and supercells. Structures are classified into groups by the energy barrier dividing experimental structures, which can be regarded as a clustering procedure. Geometrical descriptors are thus introduced to reproduce the basins of connectivity from threshold algorithms. Two descriptors focusing on local atomic environments, atom-centered symmetry functions and the smooth overlap of atomic positions, are combined with a dimensionality reduction and clustering method. The chosen descriptors and algorithms are found unable to reproduce the pockets determined by the threshold algorithm. We explore possible reasons for the limitations of this method and suggest future avenues of investigation