lj-10000-argon

10,000-atom Lennard-Jones argon

A small example that exercises the full simulation pipeline at 10⁴ particles for 100 timesteps.

Layout

• argon.in.toml — simulation config (SI units; 1 fs timestep; 100 steps at 100 K)
• argon.in.xyz — 10,000 argon atoms on a 20 × 20 × 25 simple-cubic lattice with 4.0 Å spacing, centred at the origin in an 8 × 8 × 10 nm box.
• argon.in.analysis — example post-processing input declaring an Ar-Ar radial distribution function (200 bins, r_max = 3.5 nm).
• generate_init.py — regenerates argon.in.xyz deterministically.

Run

From this directory:

cargo run --release -- run argon.in.toml

Or with the debug binary already built:

../../target/debug/heddlemd run argon.in.toml

A run produces three files in this directory:

• argon.out.run.xyz — 11 trajectory frames (steps 0, 10, …, 100), extended-XYZ
• argon.out.run.log — 21 CSV rows of step, time, kinetic energy, temperature
• argon.out.run.timings — per-stage timing summary (kernels and host I/O)

Analyze

After heddlemd run has written argon.out.run.xyz, post-process the trajectory with:

cargo run --release -- analyze argon.in.analysis

The analyze run writes one CSV per declared analysis. For the bundled argon.in.analysis that is argon.out.ar-ar.csv (200 rows of r, g_r, count). See the Analysis chapter for the full file-format reference.

Notes

• The [[pair_interactions]] table uses standard LJ-argon parameters (σ = 3.4 Å, ε ≈ 120 k_B). Initial velocities are generated from a Maxwell-Boltzmann distribution, the centre-of-mass drift is removed, and the result is rescaled so the realised temperature is exactly 100 K (RNG seed = 1; deterministic across runs on the same GPU).
• The runner uses the O(N²) pair-force kernel; the pair buffer for this example occupies ~1.2 GB of GPU memory.