Efficient Predictions of Electric Response

Predicting how materials react to electric fields is a central task for microelectronics and telecommunications. Researchers often struggle with these calculations because they demand significant computing resources. Now, a new machine-learning framework offers a faster path to these results.

Bradley Martin and his team at University College London introduced a model capable of calculating electric response quantities for inorganic crystals with high accuracy. The team built upon existing machine-learning models by incorporating external electric fields directly into the architecture. They trained the system on quantum-mechanical data, allowing the model to determine an electric-enthalpy functional. This functional provides direct access to key properties like electric polarization, polarizability, and Born effective charges.

These metrics are vital for understanding how atomic vibrations and lattice strain redistribute charge within a material. The model successfully predicted how the ferroelectric material barium titanate reacts to different histories of applied electric fields. It also replicated theoretical infrared and Raman spectra for quartz. The researchers suggest the framework applies to diverse systems ranging from electrolytes to materials found in the deep Earth.

This development marks a shift in how scientists approach the design of functional materials. By reducing the computational cost of these predictions, the framework clears a hurdle for researchers aiming to identify materials for future technologies. The findings were recently published in the inaugural issue of PRX Intelligence.