Demonstration only. Corollary is a non-commercial research demo — no warranty, no support, not for commercial use or clinical decisions. Read the disclaimer

Corollary Labs
All datasets

Materials Science · LeMaterial

LeMat-Traj-subset

Curated molecular-dynamics trajectory subset for prototyping and small-scale interatomic potential training.

Original source

What LeMat-Traj-subset is

LeMat-Traj-subset is a curated data collection published by LeMaterial for researchers in materials science, chemistry, physics, and engineering. It provides a manageable slice of the broader LeMat-Traj molecular-dynamics trajectory collection, structured to make trajectory data accessible without the storage and processing demands of the full corpus.

The dataset addresses the computational bottlenecks associated with handling massive simulation collections. By offering a smaller, representative subset, it allows scientists to prototype workflows and train small-scale machine-learned interatomic potentials (MLIPs) using standard data analysis tools.

How it was made

The dataset is curated as a smaller subset extracted from the primary LeMat-Traj molecular-dynamics trajectory corpus. It is distributed in Parquet format, providing tabular and textual representations compatible with standard data tools including Polars, Dask, and the Hugging Face datasets ecosystem.

Key facts

Publisher
LeMaterial
Dataset size
Between 10 million and 100 million entries
File format
Parquet
Data modalities
Tabular, text
Primary fields
Materials science, chemistry, physics, engineering
Source corpus
LeMat-Traj

What you can use it for

Machine-learned interatomic potential prototyping
Test novel potential architectures and feature sets quickly on molecular-dynamics trajectories before scaling up to full trajectory archives.
Small-scale MLIP training
Train baseline interatomic potential models using a bounded set of trajectory configurations.
Tabular trajectory exploration
Inspect and analyse atomic trajectories in materials science and chemistry pipelines using tools like Polars and Dask.

Explore the data

No licence declared

The authors have not published a licence, which means no permission to redistribute the data has been granted to anyone. The rows are not served here; read them on the original dataset page.

How the columns are distributed

Over 5,772,852 rows of pbe/train

cartesian_site_positions

list
min
1
median
11
mean
14.72
max
392

chemical_formula_anonymous

string_text
min
1
median
5
mean
5.16
max
18

chemical_formula_descriptive

string_text
min
1
median
10
mean
10.76
max
30

chemical_formula_reduced

string_text
min
1
median
6
mean
6.45
max
22

cross_compatibility

bool
  • True100%

dimension_types

list
min
3
median
3
mean
3
max
3

elements

list
min
1
median
3
mean
3.23
max
8

elements_ratios

list
min
1
median
3
mean
3.23
max
8

energy

float
min
-2749.01
median
-52.20
mean
-77.70
max
2.15

energy_corrected

float
min
-2897.40
median
-53.18
mean
-80.35
max
2.15

2.0% empty

forces

list
min
1
median
11
mean
14.72
max
392

functional

string_label
  • pbe100%

id

string_text
min
12
median
19
mean
18.83
max
20

immutable_id

string_text
min
4
median
12
mean
11.77
max
12

last_modified

datetime

lattice_vectors

list
min
3
median
3
mean
3
max
3

nelements

int
min
1
median
3
mean
3.23
max
8

nperiodic_dimensions

int
min
3
median
3
mean
3
max
3

nsites

int
min
1
median
11
mean
14.72
max
392

relaxation_number

int
min
0
median
0
mean
0.35
max
2

relaxation_step

int
min
0
median
25
mean
57.15
max
731

space_group_it_number

int
min
1
median
14
mean
55.30
max
230

species

string_text
min
138
median
420
mean
451.19
max
1,116

species_at_sites

list
min
1
median
11
mean
14.72
max
392

stress_tensor

list
min
3
median
3
mean
3
max
3

Strengths and limits

Good at

  • Offers a manageable volume of trajectory data tailored for rapid prototyping and low-overhead training.
  • Packaged in columnar Parquet format for fast querying and compatibility with scalable data engines.
  • Spans multiple disciplinary domains across chemistry, materials science, and physics.

Watch out for

  • Contains only a subset of the full LeMat-Traj repository, making it unsuitable for applications requiring exhaustive trajectory sampling.
  • Constrained to molecular-dynamics trajectory data and tabular representations, which may not fit non-trajectory materials science problems.

Questions

Can I look inside LeMat-Traj-subset without downloading it?

Not here: it declares no licence, so no permission to redistribute its rows has been granted. Its description and statistics are still shown.

What licence is LeMat-Traj-subset under?

Its authors have not declared a licence, which means no permission to reuse or redistribute it has been granted.

What is the relationship between LeMat-Traj-subset and LeMat-Traj?

LeMat-Traj-subset is a curated, smaller slice extracted directly from the comprehensive LeMat-Traj molecular-dynamics trajectory collection.

How large is the subset?

The dataset contains between 10 million and 100 million records.

What primary tasks is this dataset designed for?

It is intended for prototyping molecular-modelling pipelines and training small-scale machine-learned interatomic potentials.

Written from LeMat-Traj-subset's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.

Related datasets