Skip to main content
. 2020 Apr 15;11:1854. doi: 10.1038/s41467-020-15619-9

Fig. 6. Regression modelling of thermal conductivity at GBs of MgO.

Fig. 6

a, b Example of how predictor (input) variables Nm were generated from GB structures for multiple linear regression. a LDFs in the vicinity of high-angle Σ5(310)/[001] and low-angle Σ327(1719¯2)/[111] STGBs, and the Gaussian function G(x) centred on the GB plane used in calculating Nm. A log scale is used for LDF values to make it easier to distinguish differences within LAE groups. b Number of atoms per unit area in each LAE group, Nm, using hierarchical clustering results for the two GBs. c Parity plot of calculated against predicted GB thermal conductivities. Error bars indicate standard deviations in thermal conductivity calculated using perturbations of different magnitudes. d Ridge regression coefficients for Nm of each LAE group. The higher the LAE group number, the larger the LDF values in the group.