An interface for solving an optimization problem with TAO where the problem itself is defined by a separate Problem class. More...

#include <TaoUtil.hh>

Public Member Functions
	TaoBasicSolver (MPI_Comm comm, const std::string &tao_type, Problem &prob)
	Construct a solver to solve `prob` using TAO algorithm `tao_type`.

virtual	~TaoBasicSolver ()

virtual std::shared_ptr< TerminationReason >	solve ()
	Solve the minimization problem.

virtual void	setMaximumIterations (int max_it)

virtual Problem &	problem ()

Protected Attributes
MPI_Comm	m_comm

petsc::Tao	m_tao

Problem &	m_problem

Detailed Description

template<class Problem>
class pism::taoutil::TaoBasicSolver< Problem >

An interface for solving an optimization problem with TAO where the problem itself is defined by a separate Problem class.

The primary interface to a TAO optimization problem is mediated by a PETSc-style TaoSolver object. The PISM TaoBasicSolver C++ class wraps a TaoSolver and some of its initialization boilierplate, and allows a separate class to define the function to be minimized.

To use a TaoBasicSolver you create a Problem class that defines the objective function and initial guess, as well any auxilliary callbacks desired. The Problem class must define a

void Problem::connect(TaoSolver solver);

method which gives the Problem an opportunity to register its methods as callbacks to the solver, perhaps taking advantage of the various TaoFooCallback classes provided in TaoUtil.hh to facilitate this. For example, a problem class MyProblem that did nothing more than register a combined objective/gradient callback could define

void MyProblem::connect(TaoSolver tao) {
typedef TaoObjGradCallback<Problem,&MyProblem::evaluateObjectiveAndGradient> ObjGradCallback; 
ObjGradCallback::connect(tao,*this);
}

In addition to the connect method, a Problem must define

TerminationReason::Ptr MyProblem::formInitialGuess(Vec *v)

which allows the problem to set the initial guess for optimization. If the minimization is successful, the solution will be found in the same vector that was returned by this method.

Assuming a MyProblem called problem has been constructed, solution of the minimization is done using, for example, the TAO algorithm tao_cg:

TaoBasicSolver<MyProblem> solver(com,"tao_cg",problem);
 
TerminationReason::Ptr reason = solver.solve();
 
if (reason->succeeded()) {
printf("Success: %s\n",reason->description().c_str());
} else {
printf("Failure: %s\n",reason->description().c_str());
}

Definition at line 92 of file TaoUtil.hh.

The documentation for this class was generated from the following file:

src/inverse/TaoUtil.hh

Public Member Functions

Protected Attributes

Detailed Description