Same-Workload DHFR Stretch Status

Date: 2026-07-15

Engine pair: mlx_atomistic vs openmm-reference. Pair ids: dhfr-implicit, dhfr-explicit-pme.

These are the real-system DHFR stretch rows for the same-workload comparison plan. dhfr-implicit runs as a one-step MLX/OpenMM Reference smoke comparison. dhfr-explicit-pme now runs under the explicit uniform_neutralizing_plasma convention and passes fixed-coordinate OpenMM energy/complete-force parity, but it still has no throughput ratio because the available runtime rows do not share a matching manifest.

Row Status

Pair id	Intended system	Physics target	Metric	Status	Raw output
dhfr-implicit	DHFR implicit solvent, ~2.5k atoms	implicit GBSA/OBC target	ns/day	comparable one-step smoke row; ratio in summary	MLX: results/same-workload-openmm-comparison/mlx-dhfr-implicit.json; OpenMM Reference: results/same-workload-openmm-comparison/openmm-dhfr-implicit.json; summary: results/same-workload-openmm-comparison/summary.json
dhfr-explicit-pme	Charged AMBER20 JAC explicit PME, 23,558 atoms	explicit solvent PME with uniform neutralizing plasma	energy/complete-force parity; one-step MLX ns/day diagnostic	parity passed; MLX runtime passed; no throughput ratio	results/scalable-charged-pme-runtime/jac-1x/charged_pme_parity_report.json; results/scalable-charged-pme-runtime/jac-1x/runtime-smoke.json

OpenMM Reference

The reference side is already measured in openmm-opencl-dhfr.md:

OpenMM system	Size	Result
DHFR Implicit GBSA	~2.5k atoms	1762.1 ns/day
DHFR Explicit PME	23.6k atoms	752.5 ns/day

The committed OpenMM OpenCL numbers are reference context, not the same thing as the one-step OpenMM Reference shape-check rows used by the same-workload comparison helper.

MLX Status

mlx_atomistic can resolve the local DHFR inputs. The two rows now differ:

• dhfr-implicit: runnable. The prep path derives a DHFR GBSA/OBC artifact from OpenMM amber99sb.xml and amber99_obc.xml, including gbsa_radius, gbsa_scale, bonded terms, torsions, constraints, and nonbonded exceptions. The MLX runtime currently reports a one-step 0.004 ps smoke row.
• dhfr-explicit-pme: runnable with the explicit uniform_neutralizing_plasma policy. The 23,558-atom fixed-coordinate report passed with 0.00013610 kJ/mol/atom total-energy error, 1.05591e-5 relative energy error, 0.07247 kJ/mol/nm force RMS, and 1.18571 kJ/mol/nm maximum force error. The one-step MLX NVT smoke reported 0.047433 ns/day, one PME plan build, lazy topology, mlx_cell_blocks, and no materialized dense pair cache. These values come from the two JAC 1x raw files listed in the row table.

The comparison helper computes a ratio only when operation, atom count, step count, timing metric, and PME configuration match. The charged explicit-PME energy/force comparison meets that contract and passes. The one-step MLX NVT timing does not have a matching OpenMM runtime manifest, so the performance ratio remains suppressed.

Reproducer Boundary

MLX row reproducers:

UV_CACHE_DIR=/tmp/mlx-atomistic-uv-cache uv run python -m mlx_atomistic.benchmarks.dhfr --case dhfr-implicit --steps 1 --json > results/same-workload-openmm-comparison/mlx-dhfr-implicit.json
UV_CACHE_DIR=/tmp/mlx-atomistic-uv-cache uv run python -m mlx_atomistic.benchmarks.dhfr --case dhfr-explicit-pme --steps 1 --amber-topology results/inputs/Amber20_Benchmark_Suite/PME/Topologies/JAC.prmtop --amber-coordinates results/inputs/Amber20_Benchmark_Suite/PME/Coordinates/JAC.inpcrd --json > results/scalable-charged-pme-runtime/jac-1x/runtime-smoke.json

OpenMM same-workload shape-check and charged parity reproducers:

UV_CACHE_DIR=/tmp/mlx-atomistic-uv-cache uv run python scripts/benchmark_openmm_dhfr.py --case dhfr-implicit --platform Reference --steps 1 --json > results/same-workload-openmm-comparison/openmm-dhfr-implicit.json
UV_CACHE_DIR=/tmp/mlx-atomistic-uv-cache uv run --with openmm python scripts/run_charged_pme_parity.py --mlx-prepared results/dhfr-artifacts/dhfr-explicit-pme --amber-prmtop results/inputs/Amber20_Benchmark_Suite/PME/Topologies/JAC.prmtop --amber-coordinates results/inputs/Amber20_Benchmark_Suite/PME/Coordinates/JAC.inpcrd --replicas 1,1,1 --platform OpenCL --out results/scalable-charged-pme-runtime/jac-1x

OpenMM OpenCL context is preserved as internal reference evidence; it is not a package workflow or a required PyPI validation step.

The OpenMM Reference commands above are one-step same-workload shape checks. They are not the OpenCL performance numbers in the context table.