[jiweiqi]

Merge the development line from Subsapce-Lab/Arrhenius.jl: native one-dimensional flame solvers verified against Cantera, plus refreshed documentation.

What is new

• Native 1-D flames: free, burner-stabilized (including fixed-temperature profiles), counterflow diffusion, premixed and twin premixed, impinging jet, catalytic and ionized flames, with adaptive refinement and restart.
• The default discretization for free and burner flames is the finite-difference scheme that reproduces Cantera's Flow1D equations; the conservative scheme remains available through the discretization keyword.
• Solver behaviour aligned with Cantera where it affects results: velocity interpolation at refinement midpoints, successful-step time-step budget and retry cadence, mass-action rules for negative trace concentrations, robust recovery from singular factorizations.
• Concise README; details moved into the Documenter site (flames, solver options, reactors, mechanisms, examples, validation methods). The documentation workflow now builds with Julia 1.12.

Verification

Two independent checks against Cantera on NH3/H2 flames (several mechanisms, all flame kinds, 1-5 atm), with Newton-converged references:

• Same discrete problem: native and Cantera solved on one shared grid agree within 1e-4 in every profile and within 1e-8 in the residuals evaluated at the same state.
• Solution accuracy on adaptive grids: within absolute tiers of the converged references, or within Cantera's own error plus its measured refinement-path variation.
  On a validation set of 26 conditions all are accepted; on a separate held-out set of 15 conditions 13 are accepted, and the two others have references that are not grid-converged.

Speed relative to Cantera varies by case (faster in many, slower in some); performance work continues.

🤖 Generated with Claude Code