Modernizing `quadprog` for convex quadratic programming

13

Quick update for those interested: I’ve wrapped up the first iteration on the original code. It exposes a couple of new derived-types and interfaces for solving (strictly) convex quadratic programs. Here is a simple example

program test_program
    use QuadProg
    implicit none

    integer, parameter :: n = 3
    ! Size of the problem.
    type(qp_problem) :: problem
    type(OptimizeResult) :: result
    real(dp) :: P(n, n), q(n), C(n, n), d(n)

    ! Qyadratic cost.
    P = 0.0_dp; forall (i=1:n) P(i, i) = 1.0_dp
    q = [0.0_dp, 5.0_dp, 0.0_dp]

    ! Inequality constraints
    C(1, :) = [-4, -3, 0]
    C(2, :) = [2, 1, 0]
    C(3, :) = [0, -2, 1]
    d = [-8, 2, 0]

    problem = qp_problem(P, q, C=C, d=d)
    result = solve(problem)

    print *, "Success ?", result%success
    print *, "Solution :", result%x
    print *, "Lagrange multipliers :", result%y
    print *, "Objective function :", result%obj
end program

You can also find a rudimentary model predictive control (MPC) example using a double integrator with bounded actuation signal in the example folder.

3 Likes