Speaker
Description
Molecular crystals occupy an important part of material sciences and investigating them through methods of crystallography allows us to study the particular intermolecular interactions within them. Ionic liquids (ILs) represent one specific class of molecular crystal, effectively being salts. They are known e.g. for their exceptionally low vapor pressures, making them prospect candidates for solvents of the future. The exact packing within molecular crystals, including those of ILs, largely influences their behavior and prospect usage. Furthermore, many substances are able to form multiple different crystals (the so-called polymorphs) depending on the exact external conditions. The exact causes of the potential existence of multiple polymorphs or monomorphic behavior for a given substance is still largely unknown. If the phase behavior is known for a given compound, determining the crystal structure of said polymorphs often presents a big challenge. Additionally, the reported phase behavior may not be exhaustive enough, such as in the case of “late-appearing” polymorphs, which might resemble the existing crystal structures or exhibit completely different packing motifs. For some compounds, there is no evidence of a crystal phase being ever formed with little or no reason or explanation as to why.
Methods of in silico crystal structure prediction have been shown not only to reproduce existing knowledge about crystal structures but also provide ample directions where additional polymorphs can be located. They usually consist of two parts: structure generation and re-ranking. Many different methodologies of generating crystal structures exist, including evolutionary algorithms, (quasi-)random sampling, simulated annealing etc. The common denominator of these methods is that they have to be reasonably cheap to compute while providing qualitatively correct structures, due to the sheer size of the structure space that needs to be sampled. Afterwards, an additional step is performed to more accurately assess the polymorph ranking of the structures, usually considering lattice energies computed with a quantum-mechanical method. Even more precisely, a finite-temperature free energy correction can be employed. In essence, the entire scheme represents a sort of funnel, with the cheapest methods being applied to a large number of structures in order to select a smaller amount of prospect structures for the following stage, culminating in the most precise methods available being applied to a handful of structures. This is important as the differences between the free energy of experimental polymorphs, if they are known, are often rather small, below 10 kJ mol$^{-1}$ and usually around 2 kJ mol$^{-1}$, which far exceeds the precision of many cheap methods.
This work centers around crystal structure prediction for crystalline ionic liquids with the goal of developing a pipeline of methods that would reproduce known experimental crystal structures and potentially find new polymorphs or suggest crystal structures for non-crystallizing ILs, such as for the case of emIm (1-ethyl-3-methylimidazolium) EtSO$_4$, with the methodology being validated on the known crystal structure of emIm MeSO$_3$. The scheme currently consists of first sampling structures using quasi-random sampling combined with basin hopping, as implemented in the software package mol-CSPy and re-optimization using periodic DFT-D3.
