petsc-3.8.4 2018-03-24

SNESLINESEARCHL2

Secant search in the L2 norm of the function or the objective function, if it is provided with SNESSetObjective(). Attempts to solve min_lambda f(x + lambda y) using the secant method with the initial bracketing of lambda between [0,damping]. Differences of f() are used to approximate the first and second derivative of f() with respect to lambda, f'() and f''(). The secant method is run for maxit iterations.

When an objective function is provided f(w) is the objective function otherwise f(w) = ||F(w)||^2. x is the current step and y is the search direction.

This has no checks on whether the secant method is actually converging.

Options Database Keys

	-snes_linesearch_max_it <maxit>	- maximum number of iterations, default is 1
	-snes_linesearch_maxstep <length>	- the algorithm insures that a step length is never longer than this value
	-snes_linesearch_damping <damping>	- initial step is scaled back by this factor, default is 1.0
	-snes_linesearch_minlambda <minlambda>	- minimum allowable lambda

Developer Notes: A better name for this method might be SNESLINESEARCHSECANT, L2 is not descriptive

Keywords

SNES, nonlinear, line search, norm, secant

SNESLINESEARCHL2

Options Database Keys

Keywords

See Also