Expose dual variables of constraints

[bchasnov]

We want to expose the dual variables associated with the constraint of an optimization problem.

For example,

// initialize parameters and variables
prob pr;
par a(pr, ...);
par b(pr, ...);
var x(pr, ...);

// setup and solve problem
minimize(...);
cons &c = (a*x <= b)
pr.solve();

// output
x.col(0); // solution to x
cons.dual_value(...); // dual variable associated with constraint a*x <= b

The last line is wanted but unimplemented.

Places to take a look

• csocp/solver_lib.h, structslvOut_t.y and slvOut_t.s. These are the slack variables that are computed by the csocp module.
• cprs/source/prob.cpp, function prob::unparse(slvOut_t). This is where we want to iterate through m_cons_list and populate the data in there using pSO->y and pSO->x.
• cprs/header/cons.h. We should add a public get variable to access the data set in the previous set. the cons::m_nSlacks variable shows how many slack variables are associated with this one constraint.
• search for slack in the project to find references of where slack variables are being set. For example, line 727 of cprs/source/prob.cpp states that there is one slack variable per inequality.

Things to worry about

• indexing
• it looks like we want to subtract pSO->y - pSO->s for some dual variables.

[bchasnov]

initial tests committed to dualvalues branch on cprs