Speeding up LTC Calculations with Machine Learning

You're trying to figure out how well a crystal can conduct heat. Traditionally, scientists use complex calculations called first-principles to predict this property, called lattice thermal conductivity (LTC). These calculations need a lot of computational power, especially when you want to explore many materials at once. To make things more efficient, researchers started using machine learning to help out. Instead of calculating everything from scratch, they trained machines to predict the forces between atoms. This way, they could speed up the process and save a ton of computing time. In a recent study, scientists integrated these machine learning potentials directly into their calculations. They designed a smart workflow that combined different software tools. This approach allowed them to calculate LTC for 103 different materials, including those with structures like wurtzite, zincblende, and rocksalt. The results were impressive. By using machine learning, they significantly cut down on the computing resources needed. This means they could explore more materials in less time, opening up new possibilities for finding better heat-conducting crystals.