[Coin-discuss] license issues

[Andreas Waechter]

Hi John,

> Further to my earlier mail, I am wondering about IPOPT again. As I
> understand it you can't usefully emply IPOPT without one of Harwell,
> Pardiso or WSMP (just from my reading of the documentation, haven't
> drilled down any further than that at this stage). And apparently you
> also need AMPL Solver Library, although I'm not sure if that is required
> for compilation of the C++ interface, or only for building the AMPL
> plugin solver.
>
> If that is the case then it would seem that it's impossible to
> distribute binary versions of IPOPT anyway, right?
>
> If one were to attempt to use the IPOPT solver as part of a larger
> software system such as ASCEND (think of ASCEND as a GPL-licensed
> equivalent of AMPL), then it seems that a binary ASCEND distribution
> incorporating IPOPT is out of the question even without the issues of
> the CPL vs GPL.
>
> Does this sound right to you?

Given the information you have, this sounds right :)

However, an interface to the linear solver MUMPS has been contributed by 
some Ipopt users (mainly Damien Hocking).  A preliminary version of the 
interface is already in the current Ipopt distribution (without 
documentation), and I feel quite bad that I still haven't updated it and 
tested it more thoroughly (but we are all quite busy, right? ;).

MUMPS is public domain code, implemented in a sparse linear solver.  You 
can get some information here:

http://graal.ens-lyon.fr/MUMPS/

With this, you would be able to generate a completely free binary version 
of Ipopt.  But it might not be so robust and fast, since that depends 
heavily on the quality of the linear solver used (I don't know how well 
MUMPS works in the Ipopt context).

Note, the AMPL solver library is also free, you can get it from netlib - 
but the AMPL interpreter itself is not free.

I hope this clarifies this a bit.

Thanks in any case for your interest!

Andreas