Molecular Dynamics Metrics beta

This task tests how well ML force fields reproduce structural, thermodynamic, and vibrational observables of ab-initio molecular dynamics (AIMD) at finite temperature.

Reference data: DynaMat. The combined molecular dynamics score (CMDS, higher is better) combines trajectory fidelity with rollout speed.

Leaderboard

Preliminary: the reference set and metrics are evolving, so rankings may change. Private-label energy and force errors are diagnostics, excluded from CMDS.

42 of 42 eligible models · Sorted by CMDS (descending) · Row numbers follow this filtered order.

#									↑
1	GRACE-2L-OAM-L	0.7	75.9	1.4	22.8	0.72	0.72	43.1	0.759	18.4k	1.97	26.4M	EFS_G	2025-09-09	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
2	GRACE-2L-OAM	0.8	89.3	1.4	23.3	0.80	0.80	50.3	0.751	14k	1.49	12.6M	EFS_G	2025-02-06	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
3	ORB v2 MPA	1.3	102.4	1.6	24.0	0.88	0.89	57.4	0.750	9.38k	1	25.2M	EFS_D	2024-10-11	10 Å	3.25M (32.1M) MPtrj+Alex
4	GRACE-1L-OAM	1.1	121.5	1.5	24.8	0.76	0.78	52.9	0.749	11.7k	1.24	3.45M	EFS_G	2025-02-06	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
5	GRACE-3L-OAM-L	0.5	54.2	1.3	22.0	0.80	0.81	43.2	0.739	27.2k	2.9	42.1M	EFS_G	2026-07-02	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
6	ORB v3	0.8	77.3	1.5	24.0	0.81	0.83	51.0	0.736	16.9k	1.8	25.5M	EFS_G	2025-04-05	6 Å	6.47M (133M) MPtrj+Alex+OMat24
7	TACE-OAM-L	0.5	60.1	1.4	22.2	0.74	0.75	42.5	0.728	33.6k	3.58	82.9M	EFS_G	2026-04-09	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
8	ORB v2 MPtrj	1.9	115.5	2.5	23.2	0.93	0.95	65.2	0.727	9.44k	1.01	25.2M	EFS_D	2024-10-14	10 Å	146k (1.58M) MPtrj
9	GRACE-2L-MPtrj	1.4	148.0	2.1	25.2	0.78	0.79	64.6	0.698	14.8k	1.58	15.3M	EFS_G	2024-11-21	6 Å	146k (1.58M) MPtrj
10	EquFlash*	0.6	70.2	1.3	22.2	0.82	0.83	46.6	0.678	65.8k	7.01	28.7M	EFS_G	2025-06-23	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
11	Nequip-OAM-L	0.7	74.9	1.5	22.4	0.82	0.82	49.1	0.670	65.1k	6.94	9.6M	EFS_G	2025-09-08	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
12	MACE-MPA-0	1.1	109.7	1.5	22.8	1.21	1.21	65.8	0.659	31.8k	3.39	9.06M	EFS_G	2024-12-09	6 Å	3.37M (12M) MPtrj+sAlex
13	MatterSim v1 5M	1.2	112.7	2.0	23.0	0.93	0.95	64.3	0.659	33.7k	3.59	4.55M	EFS_G	2024-12-16	5 Å	n/a MatterSim
14	HIENet	1.2	118.9	2.3	24.4	0.72	0.73	47.6	0.653	82.6k	8.81	7.51M	EFS_G	2025-07-01	5 Å	146k (1.58M) MPtrj
15	AlphaNet-v1-OAM*	1.3	142.3	2.2	28.1	0.97	0.97	61.1	0.653	33.8k	3.6	4.65M	EFS_G	2025-05-12	5 Å	6.6M (113M) MPtrj+OMat24+sAlex
16	EquFlashV2*	0.5	49.7	1.4	21.5	0.73	0.74	41.7	0.651	142k	15.09	44.9M	EFS_G	2026-06-11	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
17	SevenNet-l3i5*	1.2	123.3	2.1	24.8	0.81	0.82	56.1	0.651	54.4k	5.79	1.17M	EFS_G	2024-12-10	5 Å	146k (1.58M) MPtrj
18	DPA-3.1-3M-FT	0.7	68.9	1.3	22.4	0.75	0.76	50.1	0.647	93.1k	9.93	3.27M	EFS_G	2025-06-05	6 Å	n/a OpenLAM
19	Eqnorm MPtrj	1.3	125.0	2.6	25.7	0.82	0.83	56.6	0.639	61.2k	6.53	1.31M	EFS_G	2025-05-26	6 Å	146k (1.58M) MPtrj
20	EquiformerV3+DeNS-OAM	0.4	43.0	1.4	21.8	0.81	0.81	41.5	0.638	174k	18.54	30.3M	EFS_G	2026-04-07	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
21	PET-OAM-XL	0.5	43.3	1.3	21.7	0.72	0.74	46.1	0.634	149k	15.85	730M	EFS_G	2026-01-10	n/a	6.6M (113M) MPtrj+OMat24+sAlex
22	MACE-MP-0	1.7	165.7	3.3	31.2	0.80	0.82	66.4	0.633	29.4k	3.13	4.69M	EFS_G	2023-07-14	6 Å	146k (1.58M) MPtrj
23	TECE-OAM-RRA-1.0	0.4	42.4	1.4	21.6	0.80	0.80	43.1	0.630	188k	20.04	222M	EFS_G	2026-07-05	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
24	Nequip-MP-L	1.2	116.8	2.2	24.4	0.92	0.92	60.1	0.629	65.2k	6.95	9.6M	EFS_G	2025-09-08	6 Å	146k (1.58M) MPtrj
25	SevenNet-Omni-i12*	0.5	47.2	1.4	21.8	0.76	0.77	45.2	0.626	178k	19	54.9M	EFS_G	2026-01-12	6 Å	n/a COSMOSDataset
26	eSEN-30M-OAM	0.5	50.8	1.3	21.8	0.79	0.81	42.2	0.625	213k	22.73	30.2M	EFS_G	2025-03-17	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
27	Nequip-OAM-XL	0.6	62.9	1.4	22.2	0.88	0.88	50.0	0.623	141k	15.07	32.1M	EFS_G	2025-11-30	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
28	EquiformerV3+DeNS-MP	0.8	67.2	1.9	22.0	0.94	0.95	45.9	0.622	178k	18.99	30.3M	EFS_G	2026-04-07	6 Å	146k (1.58M) MPtrj
29	Nequix MP PFT	1.4	131.1	2.8	27.5	0.79	0.80	69.5	0.616	41.4k	4.42	708k	EFSH_G	2026-01-08	6 Å	154k (1.59M) MPtrj+MDR-MP PBE ω_q
30	eqV2 S DeNS	6.1	110.2	2.2	26.5	0.97	0.99	56.0	0.615	93.2k	9.93	31.2M	EFS_D	2024-10-18	12 Å	146k (1.58M) MPtrj
31	BAM-MP-core	1.3	124.0	2.1	26.9	0.90	0.90	61.2	0.607	80.2k	8.54	17M	EFS_G	2026-07-16	6 Å	146k (1.58M) MPtrj
32	Nequix MP	1.4	136.1	2.7	27.7	0.96	0.96	73.8	0.603	41.3k	4.4	708k	EFS_G	2025-08-17	6 Å	146k (1.58M) MPtrj
33	Allegro-OAM-L	0.7	84.8	1.4	23.2	0.78	0.80	47.1	0.602	228k	24.34	9.7M	EFS_G	2025-09-08	7 Å	6.6M (113M) MPtrj+OMat24+sAlex
34	DPA-4.0.1-Pro-MPtrj	0.6	59.4	1.4	21.6	0.77	0.79	47.4	0.598	264k	28.1	22.8M	EFS_G	2026-06-11	6 Å	146k (1.58M) MPtrj
35	eqV2 M	0.6	44.6	1.4	22.3	0.90	0.91	58.2	0.595	151k	16.12	86.6M	EFS_D	2024-10-18	12 Å	3.37M (102M) MPtrj+OMat24
36	M3GNet	2.6	222.0	6.4	36.5	1.84	1.85	80.1	0.593	19.5k	2.08	228k	EFS_G	2022-09-20	5 Å	62.8k (188k) MPF
37	eSEN-30M-MP	0.9	81.4	1.9	22.8	0.78	0.79	51.5	0.593	214k	22.82	30.1M	EFS_G	2025-03-17	6 Å	146k (1.58M) MPtrj
38	MatRIS-10M-MP	0.9	100.4	1.6	22.2	0.81	0.82	45.9	0.593	508k	54.2	10.4M	EFS_GM	2025-10-29	6 Å	146k (1.58M) MPtrj
39	MatRIS-10M-OAM	0.7	71.9	1.5	22.0	0.83	0.83	50.0	0.581	511k	54.52	10.4M	EFS_GM	2025-10-29	6 Å	6.6M (113M) MPtrj+OMat24+sAlex
40	Prophet-OAME-MBD	n/a	n/a	1.4	21.9	0.76	0.77	51.2	0.578	587k	62.56	62.3M	EFS_G	2026-09-13	7 Å	16.4M (324M) MPtrj+OMat24+sAlex+ELEMENTA
41	Allegro-MP-L	1.1	114.7	1.7	24.3	0.81	0.82	56.5	0.556	289k	30.77	18.7M	EFS_G	2025-09-08	6 Å	146k (1.58M) MPtrj
42	CHGNet	2.4	202.2	6.3	35.6	1.11	1.12	65.9	0.554	87k	9.28	413k	EFS_GM	2023-03-03	5 Å	146k (1.58M) MPtrj

CMDS weights

ΔvDOS ΔADF Speed ΔP

Other weights adjust proportionally. Total: 100%.

CMDS vs t_wall

This defaults to a cost-vs-fidelity Pareto: each model's total rollout wall time against its CMDS, with marker size showing model parameters and color the training-set size. Use the axis/color/size selectors to compare any pair of metrics: the RDF, ADF and vDOS errors range from 0% (perfect match with the AIMD reference) to 100% (as different from the reference as an ideal gas / non-overlapping distributions).

Marker size

Params Params · 42 models

Log Scale Color Size

Test set: DynaMat v1.0

17 systems · 22 elements · 293–1500 K

Data by Gawkowski et al.

Ab-initio NVT trajectories span metals, alloys, perovskites, molecular crystals, and transition-metal dichalcogenides. They test structural, thermodynamic, and vibrational observables at finite temperature.

Public trajectories omit energies and forces. These withheld labels are used only for maintainer-computed diagnostics, outside CMDS.

Trajectories (HDF5) Reference paper

H 6

Li 2

C 6

N 1

Mg 2

Al 3

S 1

Ti 1

Co 1

Cu 3

Zn 2

Se 1

Br 1

Zr 1

Mo 1

Ag 1

Sn 3

I 1

Cs 1

Pt 1

Au 2

Pb 1

Each element is counted once per system.

System	Composition	Atoms	Temperature (K)	Reference duration (ps)
CsSnI3	CsSnI3	135	500	14.76
MAPbBr3	H6PbCBr3N	96	300	30.00
TiSe2	TiSe2	108	400	14.71
anthracene	H5C7	48	293	8.00
Ag	Ag	108	600	24.88
Au	Au	192	1.5k	20.16
CuAu	CuAu	128	500	1.25
CuZrAl	Zr23Al4Cu23	150	1.5k	2.72
Cu	Cu	108	1k	36.01
LiMgAlZnSn	LiMgAlZnSn	500	600	3.01
LiMgAlZnSn	LiMgAlZnSn	500	900	2.18
MoS2	MoS2	96	300	20.00
Pt3Co	CoPt3	108	300	20.93
naphthalene	H4C5	144	295	3.78
pentacene	H7C11	72	295	7.79
picene	H7C11	72	295	7.08
tetracene	H2C3	60	295	8.00

Methodology

Simulation protocol, reference data, and metric definitions

Each model runs NVT simulations from the same initial structures and thermodynamic conditions as the reference first-principles trajectories. The resulting trajectories are compared via radial distribution functions (RDF), angular distribution functions (ADF), pressure distributions from the stress tensor trace, and the vibrational density of states (vDOS) obtained from the velocity autocorrelation function. Energy-fluctuation and force RMSEs are shown as maintainer-computed private-label diagnostics when available, but they are excluded from CMDS.

The DynaMat reference set spans pure metals, alloys, high-entropy alloys, transition-metal dichalcogenides, perovskites, and molecular crystals at finite temperature. Contributions to expand its coverage are welcome. The public reference data intentionally omits energies and forces. Use the shared runner, models/run_md.py, to evaluate your model and submit its metrics via PR.

For details on the MD modeling task, the DynaMat reference set and the CMDS metric, refer to arXiv:2607.03433.