SciML/ModelingToolkit.jl v11.42.0

on GitHub

ModelingToolkit v11.42.0

Diff since v11.41.0

Added

• A System can now carry its own time span, and every time-dependent problem constructor falls back to it when the caller omits one:

  @mtkcompile sys = System([D(x) ~ -x], t; tspan = (0.0, 5.0))
  prob = ODEProblem(sys, [])              # tspan == (0.0, 5.0)
  prob = ODEProblem(sys, [], (0.0, 2.0))  # an explicit time span still wins

  This covers ODEProblem, DAEProblem, DDEProblem, SDEProblem, SDDEProblem, BVProblem, DiscreteProblem and ImplicitDiscreteProblem. Omitting the time span for a system that does not carry one throws MissingTspanError, which names the system and both ways to fix it, rather than silently building a problem with a nothing time span.

  JumpProblem deliberately keeps its required time span: a two-argument method with an untyped operating point is ambiguous with JumpProcesses.JumpProblem(prob, jumps...).

  This adds no new accessor. get_tspan and has_tspan already existed and were already public; only the field behind them was missing. The field itself and the constructor defaults shipped in ModelingToolkitBase 1.69.0 (#5085, originally #5027 by @carldlund); this release raises the ModelingToolkitBase floor to 1.69 so that a ModelingToolkit user is guaranteed to have them.

Fixed

• mtkcompile no longer drops a system's time span when the system has Brownian variables (#5091). That path rebuilds the System field by field rather than through @set!, so it silently lost the field while keeping tstops. It is only reachable with ModelingToolkit itself loaded, so the ModelingToolkitBase test suite could not catch it; the regression test now runs from the root InterfaceII group.

Changed

• ModelingToolkitBase compat floor 1.64 → 1.69, required because src/systems/systems.jl now passes tspan to the System constructor.

Downstream packages that want to rely on a system carrying its own time span should depend on ModelingToolkit 11.42.0 rather than 11.41: an 11.41 user can resolve ModelingToolkitBase 1.68 and not have the field at all.

🤖 Posted by an AI agent (Claude Code, model claude-opus-5) on behalf of Chris Rackauckas.

https://claude.ai/code/session_01H5RuLDtEazwTKPkNbPeZVu

Merged pull requests:

• Document interactive integrator workflow (#5055) (@ChrisRackauckas-Claude)
• Fix JET typo-mode findings in ModelingToolkitBase (#5065) (@ChrisRackauckas-Claude)
• fix: preserve structured parameters for vector remake (#5070) (@ChrisRackauckas-Claude)
• Fix scalarized array initial conditions (#5071) (@ChrisRackauckas-Claude)
• Require CasADi ≥ 1.3.1 for the CasADi dynamic-optimization extension (#5074) (@ChrisRackauckas-Claude)
• Run the Optimization test group on a dedicated GitHub-hosted runner (#5075) (@ChrisRackauckas-Claude)
• [bot] Solve implicit affects at the ODE solver's tolerance (#5080) (@ChrisRackauckas)
• Document observed access and plotting from a live integrator (#5081) (@ChrisRackauckas-Claude)
• Add analytical parameter jacobian code generation (#5083) (@AJ0070)
• Release ModelingToolkitBase 1.68.3 (#5084) (@ChrisRackauckas)
• Default every time-dependent problem constructor to the system's tspan (#5085) (@ChrisRackauckas-Claude)
• Release ModelingToolkitBase 1.69.0 (#5086) (@ChrisRackauckas)
• Recommend the inputs keyword for interactive external signals (#5087) (@ChrisRackauckas-Claude)
• Document unbound_inputs as the fallback for discovering an input (#5090) (@ChrisRackauckas-Claude)
• Keep the system tspan through SDE structural simplification (#5091) (@ChrisRackauckas-Claude)

Closed issues:

• Symbolic tspan (#847)