I am developing an optimization model using pyomo with python (python 3 - i manage packages with anaconda on windows). I need to use a non-linear solver, ipopt. It seems that the default linear solver used by ipopt (mumps or ma27) is relatively slow and not threadable, i wish to use the pardiso solver for the linear part of ipopt. There seems to be two versions of it: the MKL-intel pardiso version and the "independent" version - I am fine with any of them. - but i do not manage to use pardiso.
I have ipopt working, i choose it for my pyomo problem by:
solver = SolverFactory('ipopt')
And i choose 'pardiso' as the linear solver using the ipopt options
solver.options['linear_solver']='pardiso'
I get the following error:
Exception message: Selected linear solver Pardiso not available.
Tried to obtain Pardiso from shared library "libpardiso.dll", but the following error occured:
Windows error while loading dynamic library libpardiso.dll, error = 126.
Which I guess means the solver is not installed. If i choose 'mumps' or do not specify anything, everything works.
However it says that the pardiso solver comes within the mkl library (which I have), now ideally i would like to know which packages i have to install from anaconda to have the pardiso solver working ?
This post stipulates that in order to work, specific Anaconda libraries folders have to be added to the system path, which i did
sys.path.append('C:\Programs\pkgs\mkl-2019.3-203\Library\bin')
sys.path.append('C:\Programs\pkgs\intel-openmp-2019.3-203\Library\bin')
Nothing changes
Alternatively I downloaded the pardiso model from https://www.pardiso-project.org/ which gave me 3 files libpardiso600-WIN-X86-64.dll .exp and .lib with which i don't know what to do or where to place (no explanation on the solver website)
thanks for your help,
