ml-cluster-expansion

What problem does it solve?

This Skill helps you build a Cluster Expansion (CE) model for disordered, lattice-based materials so you can efficiently predict energies and explore configuration space with Monte Carlo.

Core Features & Use Cases

• Agent-driven CE build loop: prepares disordered inputs, generates initial ordered structures, relaxes/lables via MCP tools, trains the CE, and optionally iterates with active learning.
• Monte Carlo sampling from a trained CE: runs lattice Monte Carlo to sample finite-temperature configurations and supports extraction of candidate structures for continued training.
• Flexible fitting workflows: trains CE from relaxation datasets and supports optional direct feature-matrix fitting for advanced regularization (e.g., sparse group lasso).

Quick Start

Use the ml-cluster-expansion skill to train a cluster expansion for your disordered alloy by supplying a primordial CIF, generating an initial sampling set, relaxing structures with an MLIP MCP tool, then training with mcp_smol_train_cluster_expansion to produce ce_project/cluster_expansion.json.