Code Overview

The code is organized into several packages that handle various aspects.

+solvers

This package contains interfaces to several solvers that can be used to minimize \phi_{\sigma}. Each interface must implement 2 methods (other than a constructor):

  • [fletcher, info] = solve()

    This calls the subsolver on the penalty function, and returns the fletcher_solver object, and an info struct:

    • info.sol.x: primal solution
    • info.sol.y: dual solution
    • info.sol.f: final objective value

  • [...] = post_iteration(...)

    This function should get called at the end of every iteration of the subsolver. It must at some point call fletcher_solver.post_iteration().

Currently supported solvers:

  • BCFLASH (equality constraints only)
  • TRPCG (equality constraints and explicit linear constraints)
  • KNITRO
  • IPOPT (not recommended)
  • SNOPT (not recommended)

+least_squares

This package contains interfaces for solving linear systems of the form

\[\begin{split}\begin{bmatrix} I & A \\ A^T & -\delta^2 I \end{bmatrix} \begin{bmatrix} p \\ q \end{bmatrix} = \begin{bmatrix} u \\ v \end{bmatrix}.\end{split}\]

Each interface must implement 2 methods (other than a constructor):

  • self = preprocess_local(self, A, Aprod, Atprod, Q, options)
  • [q,p] = lsq_local(u, v)

Currently supported linear system solvers:

  • Semi-normal equations (lssne)

    • Requires that Jacobian is explicitly available

  • LDL (lsldl)

    • Requires that Jacobian is explicitly available

  • LNLQ (lslnlq)

  • MINRES (lsminres) (not recommended)

+utils and +helpers

Contains various utilities and helper functions used throughout.