Fig. 3

Benchmarking descriptor against ab initio methods. a Comparing the SISSO-learned descriptor to QHA for 131 compounds not in the experimental dataset used to train or test the descriptor. b Distribution of residuals shown in a. c Comparing QHA to experiment for 37 compounds which appear in both FactSage and PhononDB. d Distribution of residuals shown in c. e Comparing the SISSO-learned descriptor to experiment of these same 37 compounds shown in c. f Distribution of residuals shown in e. The annotation within each figure is provided in the Fig. 2 caption