Skip to content

Releases: thematteroftime/agentic-md-for-dummies

Release list

v0.2.0 - Three reproductions, AI/human dual path

v0.2.0 - Three reproductions, AI/human dual path Latest
Latest

Choose a tag to compare

View all tags
@thematteroftime thematteroftime released this 10 May 03:39
v0.2.0
947b8cb

agentic-md-for-dummies v0.2.0

Three reproductions, two integrators, two extension flows, two paths
into the framework — one for AI agents and one for human developers.

Highlights:

  • Third reproduction: Pedersen, Schroder, Dyre PRL 120, 165501 (2018) Kob-Andersen
    binary LJ. Structural g_AA / g_AB peaks match the analytic sigma_pq * 2^(1/6)
    targets within ~5%; thermostat fidelity within 1% of T_target across three
    state points on the rho=1.2 isochore.
  • Layer 1 directory-per-class refactor: forces/ and integrators/ packages
    with single-source forwarding station at tools/registry.py and regression
    tests that catch local/global registry drift.
  • New BAOAB Wiener-noise integrator (integrators/baoab_langevin.py) that
    is fluctuation-dissipation-balanced and reduces algebraically to Velocity
    Verlet at nu=0. Implemented entirely by an autonomous sub-agent walking
    the new 9-step integrator extension flow.
  • Repository renamed from md-for-dummies to agentic-md-for-dummies, with
    the rationale spelled out in the README's new "Two ways to use this repo"
    section that documents both the AI-agent prompt patterns and the
    human-developer hand-coded path through every extension surface.
  • Eight skill regression tests covering the contracts that hold the
    architecture together. Sixty pytest passing in total.

Cross-session handoff document at
D:/code/pythonProjects/phySimu/forSimu/memory/v0.2.0-handoff.md.

Assets 2
Loading

v0.1.0 — initial public release

v0.1.0 — initial public release

Choose a tag to compare

View all tags
@thematteroftime thematteroftime released this 09 May 05:36
v0.1.0
77194f4

md-for-dummies v0.1.0 — initial public release

A small but complete molecular-dynamics framework for reproducing physics papers, driven by an AI skill that turns a paper into a runnable experiment config.

What is in this release

  • 4-layer architecture (config → adapter → platform → infrastructure) with hard contracts at every boundary
  • AI skill (paper-to-experiment) that walks a paper into a validated config in 7 steps
  • Two end-to-end reference reproductions:
    • PRX 2015 (non-reciprocal Hertzian): slope_A = 0.6617 vs paper 2/3 (0.74% error)
    • PRL 2008 (anisotropic Yukawa): chain phase with g_∥/g_⊥ = 5.33×, ⟨L⟩ = 5.15
  • Class-name dispatch — add a new analyzer / visualizer / aggregator with one file plus one registry line
  • Layered validation — schema, physics rules, manifest §3.2, registry. Bad configs fail before any GPU is touched.
  • 8 reproduction figures under docs/images/
  • Meta-skill (creator) scaffold for generating paper-to-experiment skills tailored to other simulation frameworks

Tested on

Python 3.10, Taichi 1.7.4 + CUDA, RTX 5060 Laptop (8 GB VRAM).

Getting started

git clone https://github.com/thematteroftime/md-for-dummies
cd md-for-dummies
pip install -r requirements.txt
python scripts/validate_config.py configs/examples/plan_e_damping.json --strict

See README.md for the full quickstart and AI-skill workflow, ARCHITECTURE.md for the design spec, CONTRIBUTING.md for the contribution guide.

Loading