First version of the goldilocks app

🚀 Goldilocks v0.1.0 — Initial Research Release

Goldilocks is an application for automatic generation of k-point meshes for Quantum ESPRESSO SCF single-point total-energy calculations, using machine-learning models trained on a dataset of converged meshes for 3D inorganic compounds.

This first public release focuses on reliable, uncertainty-aware k-point selection, helping users avoid both under-converged and unnecessarily expensive calculations. We showed that using Goldilocks values for DFT SCF calculations is beneficial compared to standard High-throughput approaches of fixing parameters at one value for the whole dataset.

Key features

1. Automatic k-point mesh prediction for periodic inorganic materials (for QE, SSSP1.3 PBESol library)
2. Two models are available for inference: ALIGNN and RF (best models for this task based on preliminary research)
3. Uncertainty-aware recommendations via conformalised quantile regression
4. Streamlit web interface allowing download structures from several databases, modify them, and generate QE input file.
5. MCP-server, which exposes tools for uncertainty-aware k-points spacing prediction and k-mesh generation https://github.com/stfc/goldilocks-mcp