Mesostructures and phase transition of imidazolium-based ionic liquid systems for training deep neural network algorithm

X-ray scattering data of ionic liquid (IL)/water mixtures were obtained and analyzed as training data sets for developing deep neural network (DNN) based computation architecture that can predict non-equilibrium, dynamical phenomena such as chemical reactions, self-assembly, and ionization. Specialized DNN based architectures are accurate and computationally efficient alternatives to the computationally expensive quantum mechanical simulations. But despite their broad applicability in chemical and materials discovery, they cannot describe non-equilibrium processes such as the long-range transfer of electron charge and finite electron temperature effects. ILs are known to form various hierarchical, mesoscale ordered structures when mixed with water, similar to the phase behavior of lyotropic liquid crystals. The hierarchical self-assembly of the ionic liquid systems is a complex phenomenon not easily predictable by computation. Thus, the IL/water mixtures is an excellent platform to produce large data set to train the DNN algorithm. Model systems of ILs, consisting of a cation having linear hydrocarbon tail and simple anions such as thiocyanate or nitrate, were mixed with varying amount of water to form mesoscale ordered structures and subsequently measured by X-ray scattering. The data was quantitatively analyzed by crystallographic methods and radial distribution function calculations. Potential implementation of the obtained data to DNN training will be discussed.