diffsol-nl

Non-linear solver foundation for diffsol — the non-linear operator and solver traits used by the ODE/DAE solvers, together with a Newton implementation.

Most users should depend on the diffsol crate directly. Use diffsol-nl when you want to build on the non-linear solver layer itself.

Traits

• Operators: the NonLinearOp trait describing a function F(x), and NonLinearOpJacobian describing its Jacobian J(x). These are deliberately time-unaware; the richer, time-aware operator traits live in the diffsol crate.
• Solvers: the NonLinearSolver trait for solving F(x) = 0.
• Support types: Convergence, the LineSearch trait with NoLineSearch and BacktrackingLineSearch implementations, and the NlError error type.

Implementations

• NewtonNonlinearSolver: a Newton solver built on top of a diffsol-la LinearSolver.

Each linear-algebra backend is selected via a feature flag, forwarded to diffsol-la:

• nalgebra: nalgebra-backed containers and solvers (enabled by default).
• faer: faer-backed containers and solvers (enabled by default).
• cuda: in-built CUDA containers and solvers (disabled by default).
• suitesparse: the KLU sparse direct solver from SuiteSparse (disabled by default).

Links

• Main crate: https://crates.io/crates/diffsol
• Documentation: https://docs.rs/diffsol-nl
• Book: https://martinjrobins.github.io/diffsol/

License

Licensed under the MIT license. See the repository for details.