#3544 closed defect (fixed)
Integration failed due to ordering of the states equations
| Reported by: | Lennart Ochel | Owned by: | Willi Braun |
|---|---|---|---|
| Priority: | blocker | Milestone: | 1.9.4 |
| Component: | Run-time | Version: | |
| Keywords: | Cc: | Per Östlund |
Description
Some recent commits (Nov 4) lead to a failure in the following model: PNlib.Examples.ConTest.Conflict
The error message is:
DASKR-- AT T (=R1) AND STEPSIZE H (=R2) THE
In above, R1 = 4.7579600408158E-03 R2 = 4.5386523588368E-09
DASKR-- NONLINEAR SOLVER FAILED TO CONVERGE
DASKR-- REPEATEDLY OR WITH ABS(H)=HMIN
stdout | warning | The corrector could not converge.
stdout | warning | can't continue. time = 0.004758
stdout | info | model terminate | Integrator failed. | Simulation terminated at time 0.00475796
Change History (11)
comment:1 by , 9 years ago
| Cc: | added |
|---|
comment:2 by , 9 years ago
I had a quick look at the model, but as expected the only change is that some variables and equations/algorithms have changed place. I don't think the order should matter in this case, so probably there's some issue in the back-end/run-time.
comment:3 by , 9 years ago
You are right, it seems that the problem is causes by the post optimization module detectJacobianSparsePattern, since the simulation works just fine if that module is turned off.
comment:4 by , 9 years ago
| Component: | Run-time → Backend |
|---|---|
| Owner: | changed from to |
| Status: | new → assigned |
comment:5 by , 9 years ago
| Status: | assigned → accepted |
|---|
comment:6 by , 9 years ago
| Summary: | New error in PNlib → Integration failed due to Jacobian sparsity information |
|---|
comment:7 by , 9 years ago
| Component: | Backend → Run-time |
|---|---|
| Keywords: | PNlib removed |
| Summary: | Integration failed due to Jacobian sparsity information → Integration failed due to ordering of the states equations |
It not an issue of the detectJacobianSparsePattern it just permutes the ordering of the states and that causes an issue. This makes also sense since originally this issue just pop up by changing the ordering.
The real problem then is actually caused by bad selection of starting values for the underlying non-linear system which is needed to solve while jacobian calculation. Since currently we select the start values for the non-linear system based on the last solving or based on ring-buffer without taking into account the context of the evaluation.
Actually I have already a fix for this issue since it is connected to #3111, but unfortunately it breaks currently other tests in our test suite.
comment:9 by , 9 years ago
| Resolution: | → fixed |
|---|---|
| Status: | accepted → closed |
This started to fail with this commit bf9ac8.