Fix zero sensitivity for conservation laws with parameter-dependent initial values

[dweindl]

A boundary-condition or constant=true SBML species with no counteracting rate rule gets dx/dt=0 and is eliminated into a conservation law. dtcldp (the sensitivity of a conservation law's total abundance w.r.t. free parameters) was unconditionally hardcoded to zero whenever a conservation law spanned only one species, on the assumption that a single-species total abundance can't depend on a parameter. This breaks as soon as the species' initial value does depend on a parameter (e.g. via an InitialAssignment) -- the chain rule for any other state whose rate law references the eliminated species (via its tcl symbol) then silently drops that dependency.

Whether a conservation law's total abundance depends on a free parameter can't be decided from the model's own symbolic initial conditions alone: preequilibration (steady-state Newton solve) and PEtab condition-table state reinitialization compute/override initial states and sensitivities at runtime, entirely bypassing the model's compiled fx0/fsx0. So a species' initial value can depend on a parameter even where the model's own static formula shows no such dependence -- for single- and multi-species conservation laws alike.

dtcldp is therefore now always emitted as a symbol, never folded to a literal zero. The correct numeric value is always computed at runtime via fdtotal_cldp/fdtotal_cldx_rdata.

Verified on SBML semantic test suite case 00783 (parameter-dependent conservation law, single species): sensitivities now match finite differences, where they were previously hardcoded to zero. Also verified against PEtab test suite cases 0010/0017 (multi-species conservation law under partial preequilibration and reinitialization), which an earlier version of this fix -- one that tried to statically detect the parameter-independent case instead of always emitting a symbol -- broke.

Fixes #3249

🤖 Generated with AI assistance.

Co-Authored-By: Claude Sonnet 5 noreply@anthropic.com

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 78.52%. Comparing base (3d695c5) to head (074f978).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3259      +/-   ##
==========================================
- Coverage   78.52%   78.52%   -0.01%     
==========================================
  Files         318      318              
  Lines       22199    22197       -2     
  Branches     1490     1491       +1     
==========================================
- Hits        17432    17430       -2     
  Misses       4759     4759              
  Partials        8        8              

Flag	Coverage Δ
cpp	72.30% <100.00%> (+<0.01%)	⬆️
cpp_python	35.23% <100.00%> (-0.02%)	⬇️
petab	48.43% <100.00%> (-0.01%)	⬇️
petab_sciml	16.45% <33.33%> (+<0.01%)	⬆️
petab_sciml_benchmarks	15.00% <33.33%> (+<0.01%)	⬆️
python	70.74% <100.00%> (-0.02%)	⬇️
sbmlsuite-jax	31.74% <33.33%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
python/sdist/amici/_symbolic/de_model.py	91.79% <100.00%> (-0.02%)	⬇️
python/sdist/amici/adapters/fiddy.py	88.18% <ø> (-2.37%)	⬇️

... and 4 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.