GROMACS essential dynamics constraints and flooding examples

28 actual CPU double-precision MD conditions,3atom flexible teaching monomer.
24conditions: two seeds, monitor, five ED constraint algorithms with directional/zero controls, Gaussian/inverted Gaussian/constant-force flooding.
4additional conditions: Gaussian/inverted Gaussian with2.5kJ/mol strength, alongside weak0.02kJ/mol cases.
Each trajectory:200steps x0.0002ps =0.04ps,201native TRR frames.
This is a short operational model, not converged protein PCA, equilibrium thermodynamics or a free-energy estimate.

1 sha256sum -c MANIFEST.sha256
2 bash view.sh : rebuild reference CSV and4figures, no new simulation.
   Uses MDAnalysis2.9.0, NumPy and Matplotlib; ANALYSIS_PYTHON can point to another suitable Python.
3 module load gromacs/2026.4-cpu-double
   python3 make_basis.py : regenerate nonmassweighted PCA in basis-new/ from the included actual unbiased40fs reference trajectory.
4 bash prepare.sh : prepare28new case folders using frozen PCA vectors.
5 sbatch job-p100.slurm or sbatch job-v100.slurm : sequentially compute all28cases.
   These templates use available GPU partitions and reserve their required matching GPU, but this double-precision binary computes on one CPU core; no GPU acceleration is claimed.
6 After completion, analyze fresh results in a new directory:
   /public/toolkit/miniconda3/envs/qvasp-2.26/bin/python analyze.py --cases cases --output analysis-new
   /public/toolkit/miniconda3/envs/qvasp-2.26/bin/python plot_geometry.py --cases cases --output analysis-new/figures

Nonmassweighted PCA and ED conventions are deliberately consistent. Do not substitute massweighted Hessian vectors into these ED constraints. The existing Hessian/normal-mode/monitoring tutorial uses a different convention.
The basis comes from201actual native MD coordinates, with explicit physical atom masses in the TPR. No masses are guessed. The source trajectory is too short to define a statistically converged protein basis.
PCA vector numbers are TRR step labels, not their eigenvalue/time field.
For ED constraints/monitoring, output row n describes postupdate coordinates stored in trajectory frame n+1;200pairs are directly comparable. The lastEDrow has no saved postupdate frame.
Flooding output row n corresponds to trajectory frame n. The global Gaussian bias energy is repeated for each EV in native output; do not add these repeated columns.
Printed native ED q,V,F use6significantdigits. Analysis propagates exact rounding intervals in energy/force checks, while independently checking coordinates/projections and physical geometry.
Constant-force flooding reports zero Vfl yet applies its requested nonzero force; zero in that column does not mean no bias.
radfix uses explicit external minimum as origin and maxedsteps0, avoiding radius-origin resets.
All28reference full TRR/TPR/EDR/EDI files and scientific md.log are retained. No shared program, module or service is modified.
