Materials Discovery With Quantum-Enhanced Machine Learning Algorithms
Authors:
Ignacio F. Graña,
Savvas Varsamopoulos,
Tatsuhito Ando,
Hiroyuki Maeshima,
Nobuyuki N. Matsuzawa
Abstract:
Materials discovery is a computationally intensive process that requires exploring vast chemical spaces to identify promising candidates with desirable properties. In this work, we propose using quantum-enhanced machine learning algorithms following the extremal learning framework to predict novel heteroacene structures with low hole reorganization energy $λ$, a key property for organic semiconduc…
▽ More
Materials discovery is a computationally intensive process that requires exploring vast chemical spaces to identify promising candidates with desirable properties. In this work, we propose using quantum-enhanced machine learning algorithms following the extremal learning framework to predict novel heteroacene structures with low hole reorganization energy $λ$, a key property for organic semiconductors. We leverage chemical data generated in a previous large-scale virtual screening to construct three initial training datasets containing 54, 99 and 119 molecules encoded using $N=7,16$ and 22 bits, respectively. Furthermore, a sequential learning process is employed to augment the initial training data with compounds predicted by the algorithms through iterative retraining. Both algorithms are able to successfully extrapolate to heteroacene structures with lower $λ$ than in the initial dataset, demonstrating good generalization capabilities even when the amount of initial data is limited. We observe an improvement in the quality of the predicted compounds as the number of encoding bits $N$ increases, which offers an exciting prospect for applying the algorithms to richer chemical spaces that require larger values of $N$ and hence, in perspective, larger quantum circuits to deploy the proposed quantum-enhanced protocols.
△ Less
Submitted 12 March, 2025;
originally announced March 2025.
