specular.optimization

Composable direction rules, step size rules, and an iteration wrapper for scalar or vector optimization. SPEG combines the normalized negative specular gradient with a step size rule.

Function or class	Description
specular_gradient	Run SPEG with a selected step size rule.
minimize	Combine a direction and a step size rule in one solver.
make_direction	Create a SPEG or custom direction callable.
make_step_size	Create a schedule or a bounded line search.
OptimizationResult	Inspect the accepted solution, history, and stopping reason.
LineSearchError	Identify a search that could not accept a step.

For worked examples, step size formulas, and the distinction between classical and specular search derivatives, see the optimization user guide.

The optimizer uses the selected calculation backend for numerical specular differentiation, with an eager Python loop and NumPy state.