Predicted errors of a GAP model. The figure shows the
energetics
of the pathway leading from perfect diamond-type silicon (left) to
the formation of a 4-fold defect (right); the color of the curve corresponds
to the largest predicted atomic energy error in the simulation cell,
given by the Bayesian error estimate. For small distortions, the GAP
prediction is in practically quantitative agreement with a DFT reference
(show as thin black line); for larger distortions, roughly to the
right of the dashed vertical line, the prediction deviates from the
DFT result, and concomitantly the predicted error rises notably. Note
that DFT predicts the 4-fold defect as a local minimum (highlighted
by an arrow), whereas the GAP does not. Adapted from ref (139), where details, as well
as other example cases, may be found. Original figure published under
the CC BY 4.0 license (https://creativecommons.org/licenses/by/4.0/).