WP11 mixed large-scale evidence¶
WP11 records performance and reproducibility evidence for the already bounded
mixed linear_static workflow. It does not promote a numerical capability or
create a general scalability claim.
Mandatory model¶
The measured model contains 9,091 repeated conforming chains. Each chain has
one TET4, one WEDGE6 and one HEX8, with shared triangular TET4/WEDGE6 and
quadrilateral WEDGE6/HEX8 interfaces. The aggregate has 100,001 nodes,
300,003 translational DOFs and 27,273 elements: 9,091 of each family. Each
component is fully constrained on its x=0 nodes and receives a unit total Z
load on two free top nodes. Material is homogeneous isotropic linear elasticity
with E=210e9, nu=0.3, rho=7800, in SI units.
This repeated-chain construction is deliberately explicit scale evidence. It is not an arbitrary connected industrial mesh and does not support a universal large-model performance or accuracy claim.
Results¶
The small two-chain baseline passed before the large run. Both mandatory replays
passed with the same physics digest. The CG/Jacobi SciPy solve used 12
iterations and produced a relative residual of 2.484933369237869e-9.
Force and moment closure errors were respectively 2.0730893516296557e-12
and 4.358628815121694e-12; the linear energy identity error was 0.0.
The reduced stiffness had 1,309,104 nonzeros.
Measured phase timings for replays 1/2 were:
| Phase | Replay 1 | Replay 2 |
|---|---|---|
| Assembly | 46.1126 s | 46.0056 s |
| Linear solve | 0.0537 s | 0.0516 s |
| Total | 62.4159 s | 62.5280 s |
Peak RSS after each replay was 754,372,608 bytes. The environment pack records
Windows 10, AMD64, Python/NumPy/SciPy package metadata, backend, preconditioner,
MPI rank count and the exact input/replay digests.
The optional 999,999-DOF target and 2,999,997-DOF stretch target were not run; both remain non-blockers and are not represented as measured results.
Machine-readable records:
Bounded claim¶
The permitted claim is: “Recorded large-scale mixed TET4/WEDGE6/HEX8 evidence for this specific linear-static route and environment.” No modal, dynamic, nonlinear, PYRAMID5, HEX8-SRI, nonconforming-interface, accuracy, production or universal performance claim is made. No 0.2.7 evidence was changed.