This package implements a wrapper to allow derivative-based nonlinear solvers to function as conic solvers for problems involving linear, (rotated) second-order, and exponential cones. For example:
# min -2y -1z
# st x == 1,
# x >= norm(y, z)
using MathProgBase, ConicNonlinearBridge, Ipopt
solver = ConicNLPWrapper(nlp_solver=IpoptSolver())
m = MathProgBase.ConicModel(solver)
MathProgBase.loadproblem!(m, [0, -2, -1], [1 0 0], [1], [(:Zero, 1)], [(:SOC, 1:3)])
MathProgBase.optimize!(m)
MathProgBase.status(m) # :Optimal
MathProgBase.getsolution(m) # [1.0, 0.894427, 0.447214]
MathProgBase.getobjval(m) # -2.236067
MathProgBase.freemodel!(m)
You may replace IpoptSolver above with any NLP solver (e.g. Knitro) accessible through MathProgBase, and you may pass valid options to the NLP solver directly (e.g. IpoptSolver(print_level=1).
This wrapper is experimental. If you are experiencing convergence troubles with existing conic solvers, this wrapper may be helpful. In general, however, specialized conic solvers are more reliable than derivative-based nonlinear solvers, especially for detection of infeasibility and unboundedness. If you find this wrapper useful, please let us know.