[Rd] [SoC09-Info] An IPopt interface for R

[Douglas Bates]

On Sun, Feb 22, 2009 at 3:05 PM, Douglas Bates <bates at stat.wisc.edu> wrote:
> There have been several messages on R-devel mentioning the interior
> point optimization software Ipopt, https://projects.coin-op/Ipopt/.
> This C++ library is released under a license called the Common Public
> License.
>
> I have two questions that readers of R-devel may be able to answer.
>
> 1) Would creating an Ipopt interface for R be duplicating existing
> efforts?  That is, has someone already done so or is already working
> on this project.
>
> 2) Could a package incorporating Ipopt be released under some version
> of the GPL or LGPL?
>
> Assuming that the answers to those questions are no and yes, I would
> be willing to mentor a student in a Google Summer of Code project to
> create such a package.  However, I don't have a whole lot of time to
> do so and would not object at all if another potential mentor, or
> perhaps a co-mentor, were to volunteer.  Failing that I will create an
> application, including a programming exercise.

Having looked in more detail at the Ipopt sources and installation
instructions I think there will be a problem.  While Ipopt itself is
covered by the Common Public License it requires a sparse indefinite
solver such as MUMPS, Pardiso or one of the Harwell Subroutine
Libraries, http://www.coin-or.org/Ipopt/documentation/node6.html .  As
far as I can see, none of these are covered by a license that allows
redistribution so they can't be included in a package on CRAN or, I
think, in an SoC project.

It's too bad.  I am encouraged by projects like COIN, www.coin-or.org,
and Ipopt that do recognize the importance of open source software
covered by a suitable license but so much of the numerical analysis
community still misses the boat on that one.

By the way, the URL in the quoted message was garbled.  It should have been

https://projects.coin-or.org/Ipopt