Is there a way to specify objective and constraint gradients when doing nonlinear optimization in the problem-based framework?

In the solver-based framework, user-applied gradients are specified as additional output arguments from the objective and constraint functions. However, in the problem-based framwork, the objective and constraints are specified as symbolic expressions, not function handles. So, the concept of additional outputs does not seem applicable. Is there a way to to supply analytical gradients and Hessians in problem-based implementations?

Answer 1

[Alan Weiss]

The problem-based approach calculates and uses gradients automatically for supported functions. However, to include gradients of unsupported functions or to include a Hessian, see Supply Derivatives in Problem-Based Workflow.

Alan Weiss

MATLAB mathematical toolbox documentation

[Matt J]

Thanks, Alan. But based on your link, I guess the answer (for non-automatic differentiation) is no? The procedure there appears to be just a recipe for converting the problem to the solver-based framework, where of course arbitrary derivative functions can be supplied.

[Alan Weiss]

You understand correctly, Matt.

Alan Weiss

MATLAB mathematical toolbox documentation