A feasible sequential quadratic programming strategy with iterated second-order corrections

This package provides a solver for parametric nonconvex programs of the form

dependencies

feasible_sqp uses the sparse implementation of the QP solver qpOASES which depends on MA57. Unfortunately, this complicates installation a bit. You need to obtain a valid license for HSL and its source code (see this page). After obtaining the source code for HSL, you'll need to download the METIS package (version 4.3 is required) here, extract the code and place it inside the root folder of coinhsl. Proceed then with the installation instructions of HSL, i.e., run ./configure, make and make install.

installation

• clone the repo:

git clone https://github.com/zanellia/feasible_sqp.git

• init submodules (qpOASES, CasADi):

git submodule update --init

• install the Python package:

cd feasible_sqp
pip install -e .

• build and install the dependencies:

import feasible_sqp
feasible_sqp.install_dependencies(hsl_lib_path=<...>)

usage

from feasible_sqp import *

hsl_lib_path =     <...> # e.g., '/usr/local/lib/libcoinhsl.so'

install_dependencies(hsl_lib_path=hsl_lib_path)

# number of primal variables
nv = 2

# create solver
solver = feasible_sqp(nv)

# get primal variables
y = solver.y

# define cost
f = 1.0/2.0*ca.dot(y-10,y-10)

# define constraints
g = ca.vertcat(ca.sin(y[1]) - y[0])

# define bounds
lby = -np.ones((nv,1))
uby = np.ones((nv,1))

# generate solver
solver.generate_solver(f, f, g, lby = lby, uby = uby)

# solve NLP
solver.solve()

notice: a call to solver.generate_solver() generates a script called set_LD_LIBRARY_PATH.sh which can be sourced in order to update the LD_LIBRARY_PATH environment variable such that the shared libraries needed by the solver can be located.