Designing a physics experiment from a single sentence¶
Automation / physics verification: LAMMPS ideal-gas-law check · System: Polaris · Status: Completed
The ask¶
"run a LAMMPS simulation to show PV=NRT"
What happened¶
Given only an outcome to demonstrate — not a simulation spec — Trinity designed the entire experiment itself: chose simplified physical units where the check becomes especially clean, picked five different gas densities spanning from nearly-ideal to noticeably non-ideal behavior, wrote the full simulation setup (equilibration, then a production run with pressure averaging), and wrote the analysis script to compute the compressibility factor (a standard measure of how far a real gas deviates from ideal-gas behavior) at each density.
It also had to self-correct two job-submission failures caused by a scheduling-API detail — a missing initialization step meant standard software-loading commands weren't available on the compute node until the right startup file was explicitly sourced.
Results¶
- Compressibility factor of 0.995 at the most dilute density tested — within 0.5% of the ideal-gas value of exactly 1.0, as expected.
- Correctly reproduced the expected real-gas deviation at higher density (compressibility factor 0.934), consistent with the attractive intermolecular forces built into the simulation model — the right physical trend, not just a number close to 1.
- All 5 simulations completed in about 1 minute of total wall-clock time on 4 GPUs.