---
license: cc0-1.0
task_categories:
  - other
tags:
  - chemistry
  - crystallography
  - xrd
  - materials-science
pretty_name: COD XRD Patterns
size_categories:
  - 100K<n<1M
---

# Crystallography Open Database — Synthetic XRD Patterns

436,196 synthetic powder X-ray diffraction (XRD) patterns generated from every CIF file in the [Crystallography Open Database](https://www.crystallography.net/cod/) (COD).

## What's in the dataset

Each pattern is a simulated powder diffractogram computed from a published crystal structure. The dataset covers the full chemical and structural diversity of COD: metals, minerals, organics, MOFs, and everything in between. No measured (experimental) patterns are included — these are purely computational.

## Files and subsets

| File | Patterns | Purpose |
|---|---|---|
| `COD_xrd_patterns_1000.pt` | 1,000 | Quick prototyping |
| `COD_xrd_patterns_10000.pt` | 10,000 | Development / hyperparameter tuning |
| `COD_xrd_patterns_50000.pt` | 50,000 | Development / hyperparameter tuning |
| `COD_xrd_patterns_100000.pt` | 100,000 | Development / hyperparameter tuning |
| `COD_xrd_patterns.pt` | 436,196 | Full dataset |

The smaller subsets are random samples drawn from the full set.

## Data format

Each `.pt` file is a Python dictionary saved with `torch.save()`. Load it with:

```python
import torch

data = torch.load("COD_xrd_patterns.pt")

patterns = data["patterns"]    # torch.Tensor, shape (N, 4500)
filenames = data["filenames"]  # list[str], length N
```

- **`patterns`**: A float tensor where each row is a 4,500-point XRD intensity vector.
- **`filenames`**: The corresponding COD CIF filenames (e.g. `"4326570.cif"`).

## Generation parameters

| Parameter | Value |
|---|---|
| Software | pymatgen `XRDCalculator` |
| Radiation | Cu K-alpha (lambda = 1.54184 angstrom) |
| 2-theta range | 0 to 90 degrees |
| Step size | 0.02 degrees |
| Points per pattern | 4,500 |
| Normalisation | Min-max scaled to [0, 1] |

Patterns were generated by reading each CIF file, computing the theoretical diffraction peak positions and intensities, and interpolating onto the uniform 2-theta grid. Regions with no diffraction peaks are zero-padded.

## Finding the original CIF files

Every filename maps directly to a COD entry. Strip the `.cif` extension to get the COD ID, then download the structure file:

```
Filename:  4326570.cif
COD ID:    4326570
URL:       https://www.crystallography.net/cod/4326570.cif
```

## License

CC0 1.0 Universal — same as the Crystallography Open Database itself.

## Citation

If you use this dataset, please cite:

**COD:**
Grazulis, S. et al. "Crystallography Open Database — an open-access collection of crystal structures." *Journal of Applied Crystallography*, 42(4), 726–729, 2009.

**pymatgen:**
Ong, S. P. et al. "Python Materials Genomics (pymatgen): A robust, open-source python library for materials analysis." *Computational Materials Science*, 68, 314–319, 2013.