2D LJ atomic-wall channel flow. 384particles:288mobile,96fixedwalls.

bash run.sh          fresh full 16case Slurm calculation and analysis
bash analyze.sh work analyze your existing completed work,without new jobs
bash analyze.sh      analyze provided actual source outputs,without new jobs
bash plot.sh reference redraw provided actual source data

Needs both P100 and V10016 partition permissions. Each of two MD allocations
uses1GPU,2allocatedCPUcores,1MPI and1OMP;8cases runserially perGPU.
All GPU MD uses normal Slurm;dynamic TC membership and V10032 excluded.
Postprocessing is single threaded. No SSH to GPU nodes.
Inspect work/SUBMISSIONS.json and squeue after terminal interruption.
Never resubmit into an existing work directory;use a fresh extraction.

2D hex lattice density0.8,16x12latticeperiods,periodicx/nonperiodicy.
Fixed atomic walls:3rows eachside. Pair lj/cut2.5 shifted,epsilon_ff=ww=1,
epsilon_fw=.25/1,sigma=1,mass1. Fluid only NVE plus Langevin Ty control,
compute temp/partial0 1 0,extra/dof0 explicitly;no x streaming damping.
5000warmupsteps+20000drivensteps,dt.002,bodyforce.02/.04,seeds58001/2.
Profiles saved every500steps,41frames;last10occupancyweighted means.
Roughwallreference planes are innermostfixedatomcenters. Quadratic fits and
signed apparent extrapolation lengths are finite short-trajectory summaries,
not converged slip or viscosity. Negative/large values are retained.
Initial2D order is not a proof of an equilibriumliquidphase.

Provided raw flow.profile has native count/value fields joined by the original
custom format. The parser recovers a unique canonical printf split using an
independently binned vx,then checks thecount separately. Fresh calculations
use a leading separator in the format;physics and tolerances are unchanged.
CSV tables have explicit columns. No MD is repeated merely to reformat data.

LAMMPS30Sep2026 GCC13.3/CUDA12.8/OpenMPI5.0.10/Kokkos5.2.1.
NumPy/Matplotlib in existing ASE3.29.0 environment. All units are LJ reduced;
2D channel is not mapped to a particular real fluid or physical viscosity.

The short-tail positive-flow criterion remains in NUMERICAL-VALIDATION.json.
A negative tail mean is not silently accepted as steadypositiveflow. Physical
identity failures stop analysis; a short-tail flow failure retains passed=false,
prints a request for longer statistics,and still exports signed plots/tables.
An existing completed work directory is analyzed without duplicate MD submission.
Incomplete work is rejected;inspect original job IDs rather than resubmitting.
