[Ipopt] IPOPT Binary Status and linear solver choices.

[Ruhollah Tavakoli]

Jesse,

it seems that you have lot of questions which are actually beoyund scope of
this mailing list so i will feedback you in a PM, however, regarding to
compliing with mingw or cygwin and using in native windows application,
there are some ways, a while ago i work around it and just say that it's
possible but a bit tricky, i'm so busy and can not specifically guide you.
but if you google you found lot of materials. But as starting point you can
see this web:

http://www.symscape.com/openfoam-on-windows

in which peoples compile OpenFoam for native windows application (just know
that compiling OpenFoam is in deed very difficult and tedious to in windows
platform even by cygwin, so you can imagin their skill). I had already
contact with them, they are kind and feedback you.


Cheers

RT

PS: what is structure of your problem, you may use more simpler,
compact (and probably efficient) nonlinear programming solver. Based on your
speech (difficulty of derivative computation, noise, ...) i guess you do not
have nonlinear constraints, probably just have box ones. I think power of
IPOPT is more related to excellent treatment of nonlinear constraints. Also
what is size of your problem? for small problems SQP type methods may works
better, look at software by Prof. Klaus Schittkowski:
http://www.math.uni-bayreuth.de/~kschittkowski/software.htm, his codes are
available upon request ...












On Fri, Feb 20, 2009 at 5:55 PM, Jesse Perla <jesseperla at gmail.com> wrote:

> Thanks Ruhollah, that is very helpful:
>
>> First of all, if you are insisted to use windows, why do not use cygwin or
>> mingw, ipopt works well under these shells, and also there are some trick to
>> use precompiled libraries them into MSVC projects (just notice that using
>> cygwin has performance penalty, in my case ~ 50% slower). Also notice that
>> under windows you do not have chance of using Pardiso, as there is
>> neither any cygwin/mingw version nor native windows complers like
>> MSVC/intel. It seems that tis developper do not attend to it (i personally
>> send couple of request for it, but without at least any response ;) )
>>
>
> 0) Since you opened the can of worms, I am going to ask a whole bunch of
> somewhat unrelated - and likely ignorant - questions!
>
> 1) In my mind, I am separating the compilation of libraries from the user
> of the libraries. If I can link to the compiled libraries on windows from
> the intel C++ compiler, then I don't really care how it is built. I would
> prefer not to have to use the cygwin.dll, but that is a small price to pay
> (well... maybe 50% overhead is a large price if it is uniformly that bad) 2)
> My knowledge of this stuff is a little shaky, but I have been staying away
> from cygwin or MinGW because of the use the GNU compiler. My understanding
> is that the binary static libraries generated by cygwin are not compatible
> with linking into Intel/MSVC applications on windows. Note that I am using
> Intel for both the optimizing compiler and the tuned MKL. If I am mistaken
> and I can link cygwin generated .a files into normal intel C++ project, then
> please tell me how? This would really make my life easier in many many
> ways...
>
> 3) The other problem I have with cygwin, etc. is that I can't figure out
> how you control which runtime library it links to. I need to make "debug
> builds" of the libraries that link to the debug runtime libraries (building
> /MD and /MDd on windows compilers). I don't necessarily need the libraries I
> am building to have debug info themselves, but without this I will get
> symbol clashes when linking the library in my own debug build.   4) I was
> under the impression that I could get MSYS to work with the intel compiler
> and generate proper .lib files for windows, but I am not sure exactly how to
> do this.  This might have been another strategy.
>
> 5) Intel MKL, which comes with the Intel C++ and Fortran Compiler, has a
> binary of Paradiso for all of its platforms.  It also has a tuned BLAS,
> BLACS, LAPACK, ScaLAPACK, etc. that I want to use on all platforms.  It is
> on the clusters I am interested in and even free for academic use on Linux
> desktop.  My understanding is that Paradiso may be a little behind the
> versions, but I assume it will eventually work with IPOPT.
>
>>
>> 2  your question is ambigious for me, but take a look at this page (i may
>> helps), "sintef":
>> http://www.sintef.no/Home/Information-and-Communication-Technology-ICT/Applied-Mathematics/Research-Areas/Geometry/Software/
>>
>> 3  regarding to qudrature i do not understand your enquiry well, but to my
>> understanding, "Deal II" library is an OOP FEM library with arbitrary order
>> quadrature on logically rectangular (hypercube) grid (support qudtree/octree
>> refinement), Also FEniCS project (
>> http://www.fenics.org/wiki/FEniCS_Project) provide an automatic
>> variational compiler and of course support qudrature integration (support
>> only *affine* mappings). Both of these not easy to use with native win, but
>> easy with cygwin.
>>
> The links are great, thanks.  Since I am not working with PDEs or in the
> geometry world, I haven't been looking in the right places.  For quadrature,
> we end up with a lot of integrals over stochastic noise in economics (which
> would be in the objective function I use with IPOPT), so I was thinking
> about a library that does multi-dimensional versions of
> http://en.wikipedia.org/wiki/Gauss–Hermite_quadrature <http://en.wikipedia.org/wiki/Gauss%E2%80%93Hermite_quadrature>
>  etc.
>
>  Thanks again everyone.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://list.coin-or.org/pipermail/ipopt/attachments/20090220/5dd33f14/attachment.html