Tessero\Optimize\NonlinSolve

broyden1(callable $F, array $xin, ?int $maxiter = null): array

scipy.optimize.broyden1 ("good" Broyden, low-rank inverse-Jacobian update).

broyden2(callable $F, array $xin, ?int $maxiter = null): array

scipy.optimize.broyden2 ("bad" Broyden).

anderson(callable $F, array $xin, ?int $maxiter = null): array

scipy.optimize.anderson (extended Anderson mixing).

diagbroyden(callable $F, array $xin, ?int $maxiter = null): array

scipy.optimize.diagbroyden (diagonal Broyden approximation).

linearmixing(callable $F, array $xin, ?int $maxiter = null): array

scipy.optimize.linearmixing: root via a constant scalar (-alpha) Jacobian approximation.

excitingmixing(callable $F, array $xin, ?int $maxiter = null): array

scipy.optimize.excitingmixing: root via a tuned diagonal Jacobian approximation.

newtonKrylov(callable $F, array $xin, ?int $maxiter = null): array

scipy.optimize.newton_krylov: Jacobian-free (matrix-free) inexact-Newton with a restarted-GMRES inner solve ({@see KrylovJacobian}). The Jacobian's action on a vector is a finite difference of the residual, so no analytic/explicit Jacobian is needed. Converges to the same root scipy's newton_krylov reaches (unique-root systems match to ~1e-6 relative; the inner forcing tolerance follows the inexact-Newton schedule, so the iterate path is numerically but not bit-for-bit identical to scipy's gmres). Real-valued systems only.

nonlinSolve(callable $F, array $xin, string $jacobian, ?int $maxiter = null): array

nonlin_solve driver (line_search='armijo', the scipy default).

n2(array $v): float

2-norm (scipy.linalg.norm default). @param list<float> $v

safeNorm(array $v): float

_safe_norm: inf if any non-finite. @param list<float> $v

dot(array $a, array $b): float

vdot. @param list<float> $a @param list<float> $b

solveLin(array $A, array $b): array

Gaussian elimination with partial pivoting (scipy.linalg.solve stand-in for the small MxM systems inside the Jacobian approximations).