Os-2024-06-05

../../_images/mlp_dist40.png

The current structure dataset comprises 12974 structures. Procedures to generate structures and estimate MLPs are found in A. Seko, J. Appl. Phys. 133, 011101 (2023).

Predictions using Pareto optimal MLPs

../../_images/eqm_properties40.png

These properties are calculated for MLP equilibrium structures obtained by performing local structure optimizations from the DFT equilibrium structures. These DFT equilibrium structures are obtained by optimizing prototype structures that are included in ICSD. As a result, the structure type of the converged structure may sometimes differ from the one shown in the legend.

The other properties predicted using each Pareto optimal MLP are available from column Predictions in the following table.

Os-2024-06-05 shows large prediction errors. Distributed MLPs should be carefully used. MLPs are often accurate for reasonable structures, but it is sometimes inaccurate for unrealistic structures.

Pareto optimals (on convex hull)

Name

Time

RMSE

Predictions

Files

polymlp-00064

0.158 / 0.016

15.037 / 0.2419

polymlp-00093

0.195 / 0.020

13.279 / 0.2307

polymlp-00096

0.234 / 0.022

12.541 / 0.2132

polymlp-00065

0.412 / 0.033

9.525 / 0.1862

polymlp-00068

0.452 / 0.034

8.908 / 0.1778

polymlp-00069

0.798 / 0.044

6.788 / 0.1593

polymlp-00070

0.922 / 0.050

6.049 / 0.1594

predictions

polymlp.lammps polymlp.in

polymlp-00071

1.281 / 0.067

4.969 / 0.1459

predictions

polymlp.lammps polymlp.in

polymlp-00100

1.713 / 0.092

4.532 / 0.1403

predictions

polymlp.lammps polymlp.in

polymlp-00187

2.592 / 0.123

3.924 / 0.1405

predictions

polymlp.lammps polymlp.in

polymlp-00284

9.954 / 0.682

3.030 / 0.1219

predictions

polymlp.lammps polymlp.in

Units:

  • Time: [ms] (1core/36cores)

  • RMSE: [meV/atom]/[eV/ang.]

Column “Time” shows the time required to compute the energy and forces for 1 MD step and 1 atom, which is estimated from 10 runs for a large structure using a workstation with Intel(R) Xeon(R) CPU E5-2695 v4 @ 2.10GHz. Note that these MLPs should be carefully used for extreme structures. The MLPs often return meaningless values for them.

  • All Pareto optimal MLPs are available here.