NonlinearSolve.jl

NonlinearSolveFirstOrder NonlinearSolveFirstOrder-v2.1.0

Diff since NonlinearSolveFirstOrder-v2.0.0

Merged pull requests:

• CompatHelper: bump compat for NonlinearSolveFirstOrder to 2, (keep existing compat) (#822) (@github-actions[bot])
• CompatHelper: bump compat for NonlinearSolveFirstOrder to 2 for package docs, (keep existing compat) (#823) (@github-actions[bot])
• CompatHelper: bump compat for NonlinearSolveFirstOrder in [extras] to 2 for package SCCNonlinearSolve, (keep existing compat) (#824) (@github-actions[bot])
• Fix Jacobian diagonal calculation in LM when using StaticArrays (#825) (@JamesWrigley)
• fix modAB typos (#826) (@oscardssmith)
• switch default bracketing solver to ModAB (#828) (@oscardssmith)
• Fix Runic CI workflow and formatting (#830) (@ChrisRackauckas-Claude)
• Refactor ModAB constructor and update method name (#831) (@Just-A-Random-Geek)
• Add ModAB method to bracketing nonlinear solve documentation (#833) (@Just-A-Random-Geek)
• fix: remove closure in vector SCC branch, well-type sols vector (#834) (@ChrisRackauckas-Claude)
• feat: AutoSpecialize norecompile infrastructure for NonlinearSolveBase (#838) (@ChrisRackauckas-Claude)
• Fix ForwardDiff through bracketing solvers when Duals are in tspan (#840) (@ChrisRackauckas-Claude)
• Handle mixed case in tspan-Dual ForwardDiff extension (#841) (@ChrisRackauckas-Claude)
• Handle null u0 in SimpleNonlinearSolve for NonlinearLeastSquaresProblem (#843) (@ChrisRackauckas-Claude)
• Fix tspan-Dual ForwardDiff ext: secant step instead of foreign Dual tag (#844) (@ChrisRackauckas-Claude)
• Handle MixedDuplicated in Enzyme custom rules for solve_up (#845) (@ChrisRackauckas-Claude)
• Fix support for bounds when using init() (#848) (@JamesWrigley)
• Revise recommended methods for bracketing solvers (#849) (@Just-A-Random-Geek)
• Update convergence rate in ModAB method description (#850) (@Just-A-Random-Geek)
• Improved ModAB with adaptive k factor and threshold for bisection fallback (#852) (@Just-A-Random-Geek)
• Update symmetry factor and deviation factor calculations (#853) (@Just-A-Random-Geek)
• Fix SimpleNonlinearSolve GPU compat without unnecessary @generated (#854) (@ChrisRackauckas-Claude)
• Update LinearSolve compat to 3.48 (#856) (@JamesWrigley)
• Apply runic formatting to fix CI (#857) (@JamesWrigley)
• Add precompile workload for Dual and SubArray broadcast operations (#858) (@ChrisRackauckas-Claude)
• Make ModAB iteration always shrink the bracket even if there's numerical noise (#861) (@DaniGlez)
• fix warning messages for reverse diff backends (#863) (@hexaeder)
• Bump julia-actions/cache from 2 to 3 (#864) (@dependabot[bot])
• Remove @test_broken for LevenbergMarquardt Zygote+SVector (#866) (@AdityaPandeyCN)
• Replace Buildkite CI with GitHub Actions self-hosted runners (#867) (@ChrisRackauckas-Claude)
• Fix CUDA V100 compatibility with LocalPreferences.toml (#868) (@ChrisRackauckas-Claude)
• Fix type inference for polyalgorithm solve!() (#870) (@ChrisRackauckas-Claude)
• Avoid MArrays for out-of-place problems with SciMLJacobianOperators (#871) (@dingraha)
• Remove V100 CUDA compatibility workarounds (#873) (@ChrisRackauckas-Claude)
• Fix Enzyme reverse rule NamedTuple broadcasting with MTKParameters (#879) (@ChrisRackauckas-Claude)
• remove-redundant-var (#881) (@oscardssmith)
• Fix incorrect termination mode names and maxiters default in docs (#882) (@ChrisRackauckas-Claude)
• Simplify _scc_solve for Enzyme compatibility (#884) (@ChrisRackauckas-Claude)
• Fix Enzyme rule to handle DuplicatedNoNeed and other annotation types (#885) (@ChrisRackauckas-Claude)
• Bump codecov/codecov-action from 5 to 6 (#886) (@dependabot[bot])
• Bump minor versions for NonlinearSolve and NonlinearSolveBase (#888) (@ChrisRackauckas-Claude)
• Accept any parameter type in AutoSpecialize wrapping (#891) (@ChrisRackauckas-Claude)

Closed issues:

• InternalAPI.reinit! for polyalg is missing p and u0 (#504)
• Using LevenbergMarquardt for Linear Least Squares Problem Fails after 3.15.1 (#530)
• Solution of overdetermined systems fails with StaticArrays (#560)
• NonlinearSolve.jl is robust to NLsolve.jl Jacobian column norm underflow bug (#835)
• Migration guide: NLsolve.nlsolve with manual Jacobian and autoscale for kite power system (#836)
• ModAB Anderson-Bjork modification produces NaN with ForwardDiff.Dual numbers (#839)
• SimpleNewtonRaphson cannot solve NonlinearLeastSquaresProblem with empty u0 (#842)
• Docs refer to nonexistent AbsSafeBestTerminationMode termination condition (#846)
• nextfloat(::Int64) in BracketingNonlinearSolve ensemble solution (#847)
• ModAB iteration can fail to shrink the interval due to floating-point error (#860)
• Enzyme MixedReturnException with default PolyAlgorithm on Julia 1.10 (#869)
• You can remove this redundant assignment in ModAB algorithm: x0 = x3 (#880)