[sundials-users] Forward Sensitivity Analysis with SensErrCon

2 views

Skip to first unread message

Klauss, Daniel

unread,

Feb 5, 2026, 11:45:47 AMFeb 5

to SUNDIAL...@listserv.llnl.gov

Hello there,

I am currently working on CADET (https://github.com/cadet/cadet-core), a numerical simulator for bioprocesses. We use SUNDIALS IDAS as the time integrator for our model systems. Until recently, we were limited to SUNDIALS v3.2.1, and I am now in the process of upgrading to a more recent version (currently testing v7.5.0, commit c8dabcea90ca8bf195474da120e4f3dd39aa711f).

For solving the linear systems in IDAS, we provide our own custom linear solver. To allow our solver to use the Newton convergence factor (0.33) for its own tolerance decisions, I call
IDASetEpsLin(mem, 1) and IDASetLSNormFactor(mem, 1). With this setup, I have been able to reproduce all standard simulations in our test framework.

However, one issue remains with our benchmark cases that use forward sensitivity analysis. We use the staggered approach. With SensErrCon disabled, I can reproduce the same solutions as with v3.2.1. In contrast, when enabling sensitivity error control (which we would like to use), IDAS in v7 does not increase the time step properly and exceeds any reasonable maxsteps, whereas it worked with v3.2.1.

In the test case I am using for debugging, I set the sensitivity tolerances via IDASetSensSTolerances as follows:

abstol = 1e-12
reltol = reltolS = 1e-10
abstolS = 1e-5

with an initial step size of initstep = 1e-10.

Increasing the initial step size leads to instability at t = 0. While v3.2.1 increases the step size within the first few steps to approximately 1e-3, v7.5.0 remains around 1e-6, eventually reaches maxsteps, terminating with ” At t = 1.16756665347275, mxstep steps taken before reaching tout.”.

If helpful, I can provide a minimal reproducible example; the relevant code is available under https://github.com/cadet/CADET-Core/blob/feature/update-sundials-7.1.1/src/libcadet/SimulatorImpl.cpp

As a side note, I noticed a small inconsistency in the documentation: the default value of sensmaxnonliniters is stated as 3 in the setter function documentation, while the general description of optional inputs lists the default as 4, see https://sundials.readthedocs.io/en/latest/idas/Usage/FSA.html#optional-inputs-for-forward-sensitivity-analysis.

Thank you very much for your time and any insights you might have.

Best regards

Daniel Klauß

Master Student/Matse

Institut für Bio- und Geowissenschaften 1

Telefon: 02461 619066

E-Mail: d.kl...@fz-juelich.de

Internet: https://www.fz-juelich.de/ibg/ibg-1

---------------------------------------------------------------------------------------------
---------------------------------------------------------------------------------------------
Forschungszentrum Jülich GmbH
52425 Jülich
Sitz der Gesellschaft: Jülich
Eingetragen im Handelsregister des Amtsgerichts Düren Nr. HR B 3498
Vorsitzender des Aufsichtsrats: MinDir Stefan Müller
Geschäftsführung: Prof. Dr. Astrid Lambrecht (Vorsitzende),
Dr. Stephanie Bauer (stellvertretende Vorsitzende),
Prof. Dr. Ir. Pieter Jansens, Prof. Dr. Laurens Kuipers
---------------------------------------------------------------------------------------------
---------------------------------------------------------------------------------------------

To unsubscribe from the SUNDIALS-USERS list: write to: mailto:SUNDIALS-USERS-...@LISTSERV.LLNL.GOV

Reply all

Reply to author

Forward

0 new messages