Module solver 

Concrete solver implementations.

Re-exports

pub use augmented_lagrangian_method::AugmentedLagrangianMethod;

pub use barrier_method::BarrierMethod;

pub use basin_hopping::AcceptanceTest;

pub use basin_hopping::BasinHopping;

pub use basin_hopping::Metropolis;

pub use basin_hopping::RandomDisplacement;

pub use basin_hopping::StepTaker;

pub use bfgs::Bfgs;

pub use bobyqa::Bobyqa;

pub use bounded_cma_es::BoundedCmaEs;

pub use bounded_cma_inject::BoundedCmaInject;

pub use bounded_levenberg_marquardt::BoundedLevenbergMarquardt;

pub use brent::Brent;

pub use brent_derivative::BrentDerivative;

pub use cma_es::CmaEs;

pub use cma_inject::ClosureInner;

pub use cma_inject::CmaInject;

pub use cma_inject::MemeticInner;

pub use cobyla::Cobyla;

pub use de::De;

pub use de::DeCrossover;

pub use de::DeMutation;

pub use de_inject::DeInject;

pub use direct::Direct;

pub use gauss_newton::GaussNewton;

pub use gbnm::Gbnm;

pub use global_best_pso::GlobalBestPso;

pub use global_best_pso::PsoBoundaryHandling;

pub use global_best_pso::PsoVelocityLimit;

pub use golden_section::GoldenSection;

pub use gradient_descent::GradientDescent;

pub use levenberg_marquardt::LevenbergMarquardt;

pub use levenberg_marquardt::LevenbergMarquardtQr;

pub use levenberg_marquardt::LmDamping;

pub use lincoa::Lincoa;

pub use ma_ls_ch::MaLsCh;

pub use ma_ls_ch_cma::MaLsChCma;

pub use ma_ls_ch_sw::MaLsChSw;

pub use mads::Bounded;

pub use mads::Constrained;

pub use mads::Mads;

pub use nelder_mead::NelderMead;

pub use nelder_mead::Projected;

pub use nelder_mead::Unbounded;

pub use newuoa::Newuoa;

pub use nonlinear_cg::CgUpdate;

pub use nonlinear_cg::NonlinearCg;

pub use projected_gradient_descent::ProjectedGradientDescent;

pub use random_search::RandomSearch;

pub use sgd::Sgd;

pub use simulated_annealing::Neighbor;

pub use simulated_annealing::Reannealing;

pub use simulated_annealing::SimulatedAnnealing;

pub use simulated_annealing::TemperatureSchedule;

pub use solis_wets::SolisWets;

pub use ssga::Ssga;

pub use trust_region::CauchyPoint;

pub use trust_region::Dogleg;

pub use trust_region::ExactHessian;

pub use trust_region::MatrixFree;

pub use trust_region::MoreSorensen;

pub use trust_region::Steihaug;

pub use trust_region::TrustRegion;

pub use trust_region_reflective::TrustRegionReflective;

pub use slsqp::Slsqp;

pub use slsqp::SlsqpFailure;

Modules

augmented_lagrangian_method

Augmented-Lagrangian method for linear equality constraints A x = b (penalty + multiplier updates over any unconstrained inner solver; tolerates infeasible starts). Augmented-Lagrangian method for linear equality constraints A x = b.

barrier_method

Two-phase log-barrier method for linear inequality constraints A x ≤ b; automatically finds a strictly feasible start when one exists. Log-barrier (sequential unconstrained minimization) method for linear inequality constraints A x ≤ b.

basin_hopping

Basin-hopping (Wales & Doye 1997): Metropolis Monte-Carlo walk over the Ẽ(x) = min{f(x)} transform, wrapping any WarmStart local solver with a pluggable step taker and acceptance test. Basin-hopping: a stochastic global optimizer that walks the transformed landscape Ẽ(x) = min{f(x)} (the cost of the local minimum reached from x) with a Metropolis Monte-Carlo loop.

bfgs

BFGS quasi-Newton solver (dense inverse Hessian; Vec, nalgebra, ndarray, and faer, with f32 or f64).

bobyqa

BOBYQA (Powell 2009): bound-constrained model-based derivative-free trust-region solver (shared Powell core + TRSBOX + ALTMOV + RESCUE). BOBYQA (Powell 2009): bound-constrained model-based derivative-free optimization.

bounded_cma_es

Box-constrained CMA-ES with adaptive quadratic boundary penalty (Hansen BoundPenalty, the default in pycma).

bounded_cma_inject

Memetic BoundedCmaEs with Hansen-2011 injection: sibling of cma_inject over the bounded outer. Inners: Nelder-Mead, Levenberg-Marquardt, L-BFGS-B.

bounded_levenberg_marquardt

Levenberg-Marquardt with Coleman-Li box scaling.

brent

Brent’s method (1D root or minimum bracketing).

brent_derivative

Brent’s method using first derivatives (“dbrent”, 1D minimization).

cma_es

Hansen 2016 (µ/µ_W, λ)-CMA-ES with negative weights.

cma_inject

Memetic CMA-ES with Hansen-2011 injection. Inners: Nelder-Mead, Levenberg-Marquardt. For L-BFGS-B inner with consistent bound handling, see bounded_cma_inject.

cobyla

COBYLA (Powell 1994): nonlinearly-constrained derivative-free solver (simplex linear models + L-infinity merit + trstlp). COBYLA (Powell 1994): Constrained Optimization BY Linear Approximations.

de

Configurable differential evolution: Storn-Price 1997 global optimizer on a feasible box, fully backend-generic.

de_inject

Memetic De with per-generation top-k local refinement: the DE-flavored sibling of cma_inject. Inners: Nelder-Mead, Levenberg-Marquardt, L-BFGS-B.

direct

Original DIRECT: deterministic global optimization over finite box bounds.

gauss_newton

Pure Gauss-Newton solver for nonlinear least squares.

gbnm

Globalized Bounded Nelder-Mead with probabilistic restarts.

global_best_pso

Synchronous global-best particle swarm optimization. Synchronous global-best particle swarm optimization.

golden_section

Golden-section search (1D minimization on a bracketed interval).

gradient_descent

Steepest-descent solver with a pluggable line search and optional heavy-ball momentum.

lbfgs

L-BFGS family: unconstrained Lbfgs<V, F, Unbounded> (two-loop recursion) and box-constrained Lbfgs<V, F, Bounded> (faithful port of Nocedal’s L-BFGS-B v3.0). Lbfgsb is a type alias for Lbfgs<V, F, Bounded>. Limited-memory BFGS, both modes.

least_squares_diagnostics

Opt-in native model and trial observations for least-squares solvers.

levenberg_marquardt

Levenberg-Marquardt solver for nonlinear least squares with scaled trust-region damping by default and an optional Nielsen 1999 update.

lincoa

LINCOA (Powell 2015): linearly-constrained model-based derivative-free trust-region solver (shared Powell core + TRSTEP projected-CG + GETACT). LINCOA (Powell 2015): linearly-constrained model-based derivative-free optimization.

ma_ls_ch

Generic MA-LSCh memetic framework: SSGA + local-search chains, generic over the resumable LS operator. Generic MA-LSCh: memetic algorithm with local-search chains, generic over the resumable LS operator.

ma_ls_ch_cma

MA-LSCh-CMA: memetic algorithm with LS chains (inner: CMA-ES). MA-LSCh-CMA: the CMA-ES-chain configuration of the generic MaLsCh memetic solver (Molina et al. 2010 §4.4).

ma_ls_ch_sw

MA-SW-Chains: memetic algorithm with LS chains (inner: Solis-Wets). MA-SW-Chains: the Solis-Wets-chain configuration of the generic MaLsCh memetic solver (Molina et al., CEC 2010).

mads

MADS (Audet & Dennis 2006): mesh adaptive direct search (deterministic OrthoMADS instance) for nonsmooth or non-continuous objectives. MADS (Audet & Dennis 2006): mesh adaptive direct search, the deterministic OrthoMADS instance (Abramson, Audet, Dennis & Le Digabel 2009).

nelder_mead

Nelder-Mead derivative-free simplex solver.

newuoa

NEWUOA (Powell 2006): model-based derivative-free trust-region solver (quadratic surrogate + least-Frobenius-norm update). NEWUOA (Powell 2006): model-based derivative-free optimization.

nonlinear_cg

Nonlinear conjugate gradient with selectable updates and a pluggable line search.

projected_gradient_descent

Projected gradient descent for box-constrained problems.

random_search

Elitist (1+λ) random search over a feasible box.

sgd

Mini-batch stochastic gradient descent with constant learning rate and optional Polyak heavy-ball momentum.

simulated_annealing

Generic simulated annealing with classical Metropolis acceptance and explicit cooling schedules. Generic simulated annealing with classical Metropolis acceptance.

slsqp

Sequential least-squares programming. Dense sequential least-squares programming (Kraft’s Han–Powell method).

solis_wets

Solis-Wets (1981) adaptive random local search: biased normal mutations with success/failure step-size control.

ssga

Steady-state real-coded GA with BLX-α + NAM + BGA + replace-worst.

trust_region

Trust-region Newton solver (Nocedal & Wright Algorithm 4.1) with pluggable subproblem strategies (Steihaug-CG, dogleg, Moré–Sorensen, Cauchy point). Trust-region (Newton) minimization.

trust_region_reflective

Dense trust-region-reflective least squares with rank-aware subproblems. Dense trust-region-reflective nonlinear least squares.