solve.hpp File Reference

Problem-level solve(): the uniform solve(problem, algorithm) -> result verb over ODEProblem and LinearProblem. Repeated or warm-started linear solves use init(problem, algorithm) + solve(cache). The low-level cg()/gmres()/... kernels and ODE steppers remain the in-place primitives underneath. Stochastic sampling (MCMC) lives in solve/sample.hpp as sample(model, sampler). More...

#include "core/concepts.hpp"
#include "linalg/matrix_properties.hpp"
#include "linalg/solvers/cg.hpp"
#include "linalg/solvers/gmres.hpp"
#include "linalg/solvers/minres.hpp"
#include "linalg/solvers/pcg.hpp"
#include "linalg/solvers/solver_result.hpp"
#include "ode/ode.hpp"
#include "solve/algorithms.hpp"
#include "solve/problems.hpp"
#include <utility>

Go to the source code of this file.

Classes
struct	num::LinearSolution
	Result of a linear solve: the solution vector plus convergence stats. More...
struct	num::LinearCache< Op, Alg >
	Reusable linear-solve cache (CommonSolve init/solve!): a view of the problem plus the algorithm and the current iterate. Re-solving warm-starts from cache.u. More...

Namespaces
namespace	num
namespace	num::detail

Functions
template<IsODEProblem P>
ODEResult	num::solve (const P &prob, const RK45 &alg, ObserverFn obs=nullptr)
template<IsODEProblem P>
ODEResult	num::solve (const P &prob, const RK4 &alg, ObserverFn obs=nullptr)
template<IsODEProblem P>
ODEResult	num::solve (const P &prob, const Euler &alg, ObserverFn obs=nullptr)
SolverResult	num::detail::run (const linalg::SPDMatrix< Matrix > &A, const Vector &b, Vector &u, const CG &a)
template<class Op > requires SPDLinearOperator<Op, Vector, Vector>
SolverResult	num::detail::run (const Op &A, const Vector &b, Vector &u, const CG &a)
SolverResult	num::detail::run (const Matrix &A, const Vector &b, Vector &u, const GMRES &a)
SolverResult	num::detail::run (const SparseMatrix &A, const Vector &b, Vector &u, const GMRES &a)
template<class Op > requires LinearOperator<Op, Vector, Vector>
SolverResult	num::detail::run (const Op &A, const Vector &b, Vector &u, const GMRES &a)
template<class Op > requires SymmetricLinearOperator<Op, Vector, Vector>
SolverResult	num::detail::run (const Op &A, const Vector &b, Vector &u, const MINRES &a)
template<class Op , class M > requires SPDLinearOperator<Op, Vector, Vector> && Preconditioner<M>
SolverResult	num::detail::run (const Op &A, const Vector &b, Vector &u, const PCG< M > &a)
template<class Op , class Alg >
LinearCache< Op, Alg >	num::init (const LinearProblem< Op > &prob, const Alg &alg)
	Build a solve cache; the iterate starts at zero.
template<class Op , class Alg >
LinearCache< Op, Alg >	num::init (const LinearProblem< Op > &prob, const Alg &alg, Vector u0)
	Build a solve cache seeded with an initial guess u0 (warm start).
template<class Op , class Alg >
LinearSolution	num::solve (LinearCache< Op, Alg > &cache)
	Re-solve from a cache, warm-starting from its current iterate; cache.u is updated in place (the C++ spelling of CommonSolve solve!).
template<class Op , class Alg >
LinearSolution	num::solve (const LinearProblem< Op > &prob, const Alg &alg)
	One-shot linear solve: solve(problem, algorithm) == solve(init(...)).

Detailed Description

Problem-level solve(): the uniform solve(problem, algorithm) -> result verb over ODEProblem and LinearProblem. Repeated or warm-started linear solves use init(problem, algorithm) + solve(cache). The low-level cg()/gmres()/... kernels and ODE steppers remain the in-place primitives underneath. Stochastic sampling (MCMC) lives in solve/sample.hpp as sample(model, sampler).

Definition in file solve.hpp.