Abstract

Designer solvents such as ionic liquids or deep eutectic solvents (DESs) have combinatorial design spaces and functionalities, and these complex structure-property-reactivity relationships can now in principle be studied explicitly thanks to machine learning interatomic potentials (MLIPs). Thus, MLI Ps promise to accelerate rational design of functional, tunable, reactive liquids. I will discuss how we use MLI Ps to generate such design principles for a promising class of green solvents, focusing on how we develop a generalizable workflow from first principles. We begin with the benchmarking of DFT , followed by use of active learning through the FLARE framework. We also employ classical force fields for configurational sampling and find that multiple training iterations of the Allegro MLIP architecture are important for MLIP stability, thermodynamics, and kinetics. The final MLIP is used to approximate the transition state pathway for a fully solvated SN2 reaction. We find that the strong hydrogen bond network pins chlorine in place near reaction sites, resulting in uphill reactivity. In the design of future solvents, understanding how hydrogen bonding modifies the potential energy landscape seems to be an under-explored frontier. In summary, our general approach can be used to efficiently study a variety of neotoric solvents undergoing reactivity.

Paper: Julia H .Yang, Amanda Whai Shin Ooi, Zachary A.H. Goodwin, Yu Xie, Jingxuan Ding, Stefano Falletta, Ah-Hyung Alissa Park, Boris Kozinsky. Room-temperature decomposition of the ethaline deep eutectic solvent, The Journal of Physical Chemistry Letters, in press DOI : 10.1021/acs.jpclett.4c03645