Several Jacobian comparison tests related to multi-particle dynamic reactions continue to fail and appear to be a follow-up issue to #476.
The failures occur across both GRM and GRM2D configurations.
Failing Tests:
- GRM (Jacobian vs AD):
- Bulk
- Particle
- Modified particle
- Bulk + particle
- Bulk + modified particle
GRM2D (Time derivative Jacobian vs FD):
- Bulk
- Particle
- Modified particle
- Bulk + particle
- Bulk + modified particle
Observed Behavior:
GRM:
- Non-deterministic behavior on Windows (tests pass only when run in isolation).
- On macOS, the last Jacobian entry is non-zero although it should be zero.
- In some runs, multiple Jacobian entries are misaligned (this did not appear in the initial executions).
GRM2D:
- Two analytical Jacobian entries consistently disagree with the FD reference.
Summary:
The issue indicates platform-dependent numerical inconsistencies and potential instability in the analytical Jacobian implementations for multi-particle dynamic reactions. Further investigation into the reaction Jacobian assembly logic is required.