Skip to main content
. 2017 Apr 26;7:1192. doi: 10.1038/s41598-017-01251-z

Figure 1.

Figure 1

Histogram of the errors between a machine-learned analytical potential and the directly computed DFT energies for carbon in the diamond structure. The network used here, trained in PROPhet, contained 2 hidden layers each containing 35 nodes. The errors shown here are for 2000 structures that were not used in the fitting procedure.