Table 2.

Prediction errors on H₂ and H₂O

Molecule	M	ML-KS				ML-HK
		ΔE		ΔR o	Δθ 0	ΔE		$\Delta E_{\mathrm{D}}^{\mathrm{ML}}$		ΔR o	Δθ 0
		MAE	Max			MAE	Max	MAE	Max
H2	5	1.3	4.3	2.2	—	0.70	2.9	0.18	0.54	1.1	—
	7	0.37	1.4	0.23	—	0.17	0.73	0.054	0.16	0.19	—
	10	0.080	0.41	0.23	—	0.019	0.11	0.017	0.086	0.073	—
H2O	5	1.4	5.0	2.1	2.2	1.1	4.9	0.056	0.17	2.3	3.8
	10	0.27	0.93	0.63	1.9	0.12	0.39	0.099	0.59	0.12	0.38
	15	0.12	0.47	0.19	0.41	0.043	0.25	0.029	0.14	0.064	0.23
	20	0.015	0.064	0.043	0.16	0.0091	0.060	0.011	0.058	0.024	0.066

Errors are shown for increasing numbers of training points M for the ML-KS and ML-HK approaches. In addition, the estimated density-driven contribution to the error for the ML-HK approach (Eq. (9)) is given. Energies are given in kcal/mol, bond-lengths in pm, and angles in degrees