(169c) Tackling the Inverse Design Problem in Quantum Chemistry

Modern quantum chemistry facilitates unprecedented predictions of molecular and materials properties in complex systems. However, there is a disconnect between the fundamental building blocks of quantum mechanics and synthetic chemistry, which becomes evident when contrasting the languages of indistinguishable electrons used in quantum mechanics and distinguishable functional groups used in synthetic chemistry. While there are well-established quantum mechanical approaches to translate a molecular structure into a predicted molecular energy or property, the inverse mapping from desirable property into optimal molecular structure is not obvious. As a result, optimizing molecular structure with respect to a given observable such as an energy (e.g., a reaction barrier in catalysis) or property (e.g., the polarizability of a dielectric material) typically requires an extensive approach that includes either guessing candidate structures and manually checking their suitability empirically or time-consuming training of empirical models to predict suitable new candidates. This work takes initial steps in formulating an inverse mapping approach that determines, entirely from first principles, a molecular structure optimized for a desired observable. To this end, we introduce an approach that partitions the response of a molecular wave function to an external perturbation into contributions from molecular fragments. Our approach, which we term linear response for molecular interactions, or LR(MI), allows us to distinguish additive and cooperative molecular contributions to a target property and thus identifies optimal molecular fragments for bottom-up property design that can be directly interpreted by synthetic chemists. As a result, our work goes beyond the traditional structure-to-property paradigm and provides a direct route to property-to-structure prediction. Therefore, this inverse mapping approach informs the rational design of molecules with desirable properties directly from first principles, thus having the potential to be truly predictive.