Module Sundials_NonlinearSolver.Custom

Create a nonlinear solver from a set of callback functions.

The callbacks should indicate failure by raising an exception (preferably one of the exceptions in this package). Raising Sundials.RecoverableFailure indicates a generic recoverable failure.

The expected operations are:

• init: initializes the nonlinear solver.

• setup: sets up the nonlinear solver with an initial iteration value.

• set_lsetup_fn: receive a linear solver setup callback.

• set_lsolve_fn: receive a linear solver callback.

• set_convtest_fn: receive a convergence test callback.

• set_max_iters: sets the maximum number of iterations.

• set_info_file: sets a logfile for informational messages.

• set_print_level: sets the level of verbosity for informational messages (0 = none).

• get_num_iters: returns the number of iterations in the most recent solve.

• get_cur_iter: returns the iteration index of the current solve. This function is required when using a convergence test provided by Sundials or one of the spils linear solvers.

• get_num_conv_fails: return the number of convergence failures in the most recent solve.

• nls_type: the type of problem solved.

• solve: the call solve y0 y w tol callLSetup mem should solve the nonlinear system $F(y) = 0$ or $G(y) = y$ , given the initial iterate y0, which must not be modified, the solution error-weight vector w used for computing weighted error norms, the requested solution tolerance in the weighted root-mean-squared norm tol, a flag callLSetup indicating whether the integrator recommends calling the setup function, and a memory value to be passed to the system function.

• set_sys_fn: receive the system callback.

Note that the setup and solve functions are passed the payload data directly, whereas the lsolvefn and sysfns require the data to be wrapped in an nvector. This asymmetry is awkward but, unfortunately, unavoidable given the implementation of nvectors and the different constraints for C-to-OCaml calls and OCaml-to-C calls.

Create a nonlinear solver from a set of callback functions for sensitivity problems that pass arrays of nvectors. As for the Sundials_NonlinearSolver.Custom.make function except that the callbacks receive arrays of values.

Writing custom nonlinear solvers for use with some forward sensitivity methods requires the "internal" senswrapper type. Any attempt to use Sundials_NonlinearSolver.Senswrapper.ts outside of the call to setup or solve that provides them will result in an Sundials_NonlinearSolver.IncorrectUse exception. They must only be used to extract the underlying data with Sundials_NonlinearSolver.Senswrapper.data or as arguments for lsolve_fn, convtest_fn, or sys_fn. There are no restrictions on the arrays extracted with Sundials_NonlinearSolver.Senswrapper.data.