Ipopt.jl is a Julia interface to the Ipopt nonlinear solver.

Default Installation: julia> Pkg.add("Ipopt")

This will install Ipopt.jl, as well as Ipopt itself. A binary will be downloaded by default on macOS or Windows, and Ipopt will be automatically built from source on Linux unless a pre-existing version is found on the LD_LIBRARY_PATH. If your platform is not supported, or if you prefer to compile your own version of Ipopt in order to use commercial sparse linear algebra libraries, use the instructions below.

Custom Installation:

Make sure you have the required packages before installing, e.g.,

sudo apt-get install build-essential gfortran pkg-config liblapack-dev libblas-dev

The script below was tested successfully for installing Ipopt. You may modify the configuration options, but be sure to install Ipopt into the correct prefix location.

wget https://www.coin-or.org/download/source/Ipopt/Ipopt-3.12.8.tgz
tar xvzf Ipopt-3.12.8.tgz
cd Ipopt-3.12.8/
# Blas and Lapack must be installed already. If not, run
# ThirdParty/Blas/get.Blas and ThirdParty/Lapack/get.Lapack.
# ASL is required even if you do not plan to use it.
cd ThirdParty/ASL/
./get.ASL
cd ..
cd Mumps
# Compiling Mumps requires gfortran.
./get.Mumps
cd ../..
# Update the prefix location! The following is correct only for Julia 0.6.
./configure --prefix=$HOME/.julia/v0.6/Ipopt/deps/usr
make
make test
make install

Now in Julia:

julia> Pkg.build("Ipopt")
INFO: Building Ipopt

julia> Pkg.test("Ipopt")
...
INFO: Ipopt tests passed

Ipopt implements the solver-independent MathProgBase interface, and so can be used within modeling software like JuMP. The solver object is called IpoptSolver. All options listed in the Ipopt documentation may be passed directly. For example, you can suppress output by saying IpoptSolver(print_level=0). If you wish to pass an option specifically for the restoration phase, instead of using the prefix resto., use the prefix resto_. For example IpoptSolver(resto_max_iter=0).

Full documentation for the Ipopt C wrapper is available here. Use of the nonlinear MathProgBase interface is recommended over the low-level C interface because it permits one to easily switch between solvers.