Skip to content

MD simulation algorithm research backlog

Status: Active
Vertical: md
Registry: benchmarks/competitive/algo_registry.json (family md, ids 101–120)
Implement loop: docs/ecosystem/sim-algorithm-backlog.md on cursor/sim-algo-plan-loop

todos:

  • id: md-r0-sota-survey content: "LAMMPS/GROMACS/OpenMM — integrators, neighbor lists, cutoffs; map to algo_registry 101–120" status: completed study_only: true

  • id: md-r1-stability-matrix content: "CFL / neighbor-skin / size scaling for md_lennard_jones; tier-0 stability row proposal" status: completed study_only: true

  • id: md-r2-neighbor-list-gap content: "sim-md-research: pending plan todo md-r2-neighbor-list-gap — gap orchestrator" status: pending handoff_implement: sim-p1-md-neighbor-cell

  • id: md-r3-oracle-plan content: "sim-md-research: pending plan todo md-r3-oracle-plan — gap orchestrator" status: pending study_only: true

Agent instructions

  • One todo per loop iteration (sim-algo-research-plan-loop.py).
  • Agent: numerics_researcher (or autoresearch when novel: true on todo).
  • Gates: SIM_RESEARCH_VERTICAL=md ./scripts/sim-algo-research-gates.sh.
  • Deliverable: docs/numerics/studies/YYYY-MM-DD-<todo-id>.md with grade matrix (see sim-algo-research-grading.md).
  • Survey/stability todos: study_only: true — gates require study file, not full bench timing.
  • Push branch cursor/sim-md-research-loop every iteration.