Add Andrews squeezing mechanism benchmark with MM, DAE, and MTK formu…#1485
Conversation
|
It doesn't look like this one is working? |
|
@ChrisRackauckas I’ve cleaned up the Andrews benchmark and fixed the WPD section so it now reflects the only working solver path (direct RADAU5). All code runs locally and plots generate correctly. |
|
Update ODEInterfaceDiffEq.jl so that you can just use the standard work-precision diagram directly. You just need to give the algorithm type the arguments required to forward them here. |
|
@ChrisRackauckas I’ve updated the benchmark to use the standard WorkPrecisionSet interface. Changes made: Patched ODEInterfaceDiffEq to forward M1/M2, DIMOFIND*, and mass matrix options Replaced direct ODEInterface.radau5 calls with standard solve Restored proper work-precision diagrams Adjusted tolerance range for index-3 stability All code runs locally and plots now generate correctly. Wrapper PR: SciML/ODEInterfaceDiffEq.jl#90 |
singhharsh1708
force-pushed
the
andrews-mechanism-benchmark
branch
from
004ee93 to
458a7b4
Compare
|
I’ve updated the benchmark based on your suggestions. |
…lations and WPD analysis
…U5 manual work-precision
The radau5() wrapper with M1/M2/DIMOFIND kwargs is not yet released to the Julia registry (PR merged to master but not tagged). CI installs from registry and gets MethodError on the wrapper constructor. Reverted Reference Solution and WPD sections to use direct OI.radau5() API (same approach that passed CI in build #4040): - make_radau5_opts() helper for ODEInterface options - Direct OI.radau5() calls with output_fcn for trajectory - Manual WPD loop with median timing instead of WorkPrecisionSet - Added Statistics for median() - Added Convergence Table section
singhharsh1708
force-pushed
the
andrews-mechanism-benchmark
branch
from
9d9ccfd to
888e25e
Compare
Replace manual WPD loops with standard WorkPrecisionSet from DiffEqDevTools. Use radau5() DiffEq wrapper with DIMOFIND kwargs instead of direct OI.radau5 API. Remove 13-component rhs_andrews (only needed for direct OI API). Key changes: - Reference solution: radau5 wrapper at tol=3e-8 with dt=1e-6 (INITIALSS) - WPD: WorkPrecisionSet with tolerances [3e-5..1e-7], dt=1e-6 - Removed manual WPD for-loop, convergence table - Removed Statistics import and OI alias - Aligns with SciMLBenchmarks repo style (cf. Car Axis benchmark)
|
This benchmark now uses the Reference solution computed with abstol=reltol=3e-8 and dt=1e-6. All tests run locally and the WPD generates successfully. |
|
@ChrisRackauckas there are few errors can you check if they are fine or need changes? |
2 participants
Adds Andrews' squeezing mechanism benchmark including mass-matrix, residual DAE, and MTK index-reduced formulations. Includes solver behavior analysis and work-precision diagrams.