Atom-HiFi

Atomistic High-Fidelity representative-set selection framework.

Atom-HiFi finds the smallest subset S of a structure library that achieves high Fidelity = L / R — S covers the library’s atomic-environment diversity (high Likeness L) without redundancy (low Redundancy R). It is agnostic to the downstream task: MLIP training-set curation, active-learning loops, chemical-motif analysis, and diversity-aware sampling from large databases.

This wiki is generated with Sphinx. The narrative pages reuse the project’s canonical Markdown sources (README.md, CONTEXT.md); the API reference is generated from the in-code docstrings. ADRs stay local-only and are not published here.

Guide

• Overview
• Atom-HiFi
  • What is Atom-HiFi?
  • Key concepts
  • Installation
  • Documentation
  • Quick start
  • Output files
  • Configuration reference
  • Advanced usage
  • Citation
• Configuration
  • Build a config for your system
  • Starter template
  • Loading and validation
  • Custom descriptor
• Frozen HDBSCAN deployment
  • What an accepted environment means
  • Density attachment rule
  • Selection groups versus deployment groups
  • Persistence and compatibility
  • Run the saved-Judge example
  • Integrating the later HDBSCAN lifetime arm
  • Sources and scope
• HDBSCAN deployment validation
  • Behavioral checks
  • Reference-only canary
  • Isolation and limits
• Command-line interface
  • Entry points
  • Workflow stages
• Experimental: temporal assessment (opt-in)
  • Quickstart
  • Configuration (YAML or JSON)
  • Identity: which candidates may share a front
  • What B and E are
  • Exact arithmetic and the optional budget
  • Fronts
  • Outputs
  • Reusing existing labels or a saved model (no campaign rerun)
  • The PIT-WATER example is an example
  • Limitations

Reference

• API reference
  • Runner (orchestration)
  • Configuration
  • Focus corpus helpers
  • Piecing (super-large structures)
  • CLI
  • Judge — cover + set-cover + microstate reference
  • HiFiWorkflow — L/R scan workflow
  • Clustering primitives and persistence
  • Clusterer backends
  • NSGA-II c-factor optimiser
  • QHO σ_thermal table
  • Force truth (ε_F)
  • RDF δr_ref ε_noise
  • Workflow persistence
  • Plotting
  • ED-SOAP descriptor
  • Experimental: temporal assessment

Indices

• Index

• Module Index

• Search Page