advances.sciencemag.org/cgi/content/full/5/5/eaaw2210/DC1

Supplementary Materials for

A unified picture of the covalent bond within quantum-accurate force fields:

From organic molecules to metallic complexes’ reactivity

Alessandro Lunghi* and Stefano Sanvito*

*Corresponding author. Email: lunghia@tcd.ie (A.L.); sanvitos@tcd.ie (S.S.)

Published 31 May 2019, Sci. Adv. 5, eaaw2210 (2019)

DOI: 10.1126/sciadv.aaw2210

This PDF file includes:

Fig. S1. FF training curve for methane. Fig. S2. FF training curve for benzene. Fig. S3. FF training curve for the distorted octahedral [MnCl6]

3– complex. Fig. S4. FF training curve for the trigonal bipyramidal [MnCl5]

2– complex. Fig. S5. FF training curve for the square pyramidal [VOCl4]

2– complex. Fig. S6. FF training curve for the square planar [NiCl4]

2– complex. Fig. S7. FF training curve for the tetrahedral [ZnCl4]

2– complex. Fig. S8. Alanine molecular dynamics blind test.

SUPPLEMENTARY NOTE 1. TRAINING CURVES

Ff training curve for methane. The deviation between the DFT and the FF total energy of Me is plotted asfunction of the number of configurations included in the training set. The validation and test sets are always composed bythe same 200 configurations.

Ff training curve for Benzene. The deviation between the DFT and the FF total energy of Benzene is plotted asfunction of the number of configurations included in the training set. The validation and test sets are always composed bythe same 200 configurations.

Fig. S1.

Fig. S2.

Ff training curve for the distorted octahedral [MnCl6]3− complex. The deviation between the DFT and the

FF total energy of [MnCl6]3− is plotted as function of the number of configurations included in the training set. The validationand test sets are always composed by the same 200 configurations.

F training curve for the trigonal bipyramidal [MnCl5]2− complex. The deviation between the DFT and the

FF total energy of [MnCl5]2− is plotted as function of the number of configurations included in the training set. The validationand test sets are always composed by the same 200 configurations.

f

Fig. S3.

Fig. S4.

Ff training curve for the square pyramidal [VOCl4]2− complex. The deviation between the DFT and the FF

total energy of [VOCl4]2− is plotted as function of the number of configurations included in the training set. The validation andtest sets are always composed by the same 200 configurations.

Ff training curve for the square planar [NiCl4]2− complex. The deviation between the DFT and the FF

total energy of [NiCl4]2− is plotted as function of the number of configurations included in the training set. The validation andtest sets are always composed by the same 200 configurations.

Ff training curve for the tetrahedral [ZnCl4]2− complex. The deviation between the DFT and the FF total

energy of [ZnCl4]2− is plotted as function of the number of configurations included in the training set. The validation and testsets are always composed by the same 200 configurations.

Fig. S5.

Fig. S6.

Fig. S7.

Alanine molecular dynamics blind test. The DFT total energy is plotted against that computed by FF forboth SNAP (green dots) and GAFF (purple dots) for 450 configurations obtained by molecular dynamics. One can notice thatSNAP clearly outperforms GAFF, in particular for large distortions (large energies), where GAFF fails to describe anharmoniccontributions to the energy.

Fig. S8.