| --- |
| license: cc-by-nc-4.0 |
| tags: |
| - Materials |
| size_categories: |
| - 10M<n<100M |
| --- |
| # ELEMENTA Dataset |
|
|
| ELEMENTA is a large-scale, low-prior atomistic dataset labelled with density |
| functional theory (DFT). This first public release contains: |
|
|
| - nearly **39 million frames** from the core **ELEMENTA** dataset; |
| - the complete **ELEMENTA Spin** dataset, containing **4.7 million frames**. |
|
|
| ## Chemical-space coverage |
|
|
| The core release covers: |
|
|
| - all unary compositions across the periodic table; |
| - all binary compositions; |
| - ternary compositions whose reduced-formula stoichiometric coefficients sum |
| to no more than four. |
|
|
| The release is defined by explicit compositional and structural boundaries |
| rather than random subsampling. It therefore preserves complete and continuous |
| coverage within the released chemical domain, although it is a subset of the |
| full 210-million-frame ELEMENTA corpus. |
|
|
| ELEMENTA Spin complements the core dataset with large-scale magnetic |
| initializations and DFT-resolved magnetic energy landscapes. |
|
|
| ## Files |
|
|
| | File | Contents | |
| | --- | --- | |
| | `ELEMENTA_open.extxyz.tar.zst` | Core ELEMENTA public release | |
| | `ELEMENTA_spin.extxyz.tar.zst` | Complete ELEMENTA Spin release | |
|
|
| Both archives contain data in extended XYZ (`extxyz`) format and are compressed |
| with Zstandard. |
|
|
| To extract an archive: |
|
|
| ```bash |
| tar --use-compress-program=unzstd -xf ELEMENTA_open.extxyz.tar.zst |
| tar --use-compress-program=unzstd -xf ELEMENTA_spin.extxyz.tar.zst |
| ``` |
|
|
| ## Intended uses |
|
|
| The datasets provide a common benchmark domain for developing and evaluating |
| atomistic foundation models. Example applications include: |
|
|
| - energy and force prediction; |
| - out-of-composition generalization; |
| - polymorph recovery and ground-state identification; |
| - exact local-minima ranking; |
| - crystal structure generation; |
| - magnetic-state prediction; |
| - systematic analysis of model failures across compositions and structural |
| environments. |
|
|
| ## Citation and License |
|
|
| If you use this dataset, please cite: |
|
|
| > **[ELEMENTA: Toward Unbiased Scaling of Materials Data from First Principles](https://www.kairosmaterials.com/papers/ELEMENTA.pdf)** |
|
|
| Please refer to the accompanying release documentation for detailed data |
| fields, DFT protocols, quality-control procedures, licensing terms, and usage |
| instructions. |
|
|
|
|
