[Ipopt] Question about largest Jacobian

[Stefan Vigerske]

Hi,

On 06/08/2015 07:51 PM, Panos Lambrianides wrote:
> Hi everyone,
>
> I have a question about the largest size of the Jacobian for the
> constraints.
>
> While setting up the problem the variable nnz_jac_g is used to set up the
> number of non-zero constraint for the Jacobian.  This is of type *Index*,
> which is itself an *int* in C++.  The upper limit for int is 32767 (2^15-1).

I believe that int's are 32bit, so you get an upper limit of around 
2147483647.

> Is there a reason why this is an int (versus a long for example) and more
> importantly is there a way to overcome this limitation?  I would be happy
> to tweak the code if someone would point out the right places to do so.

There are some feature requests on this:
https://projects.coin-or.org/Ipopt/ticket/251
https://projects.coin-or.org/Ipopt/ticket/259

The main thing to do is to extend a number of interfaces to 3rd-party 
software (linear algebra and linear solvers) to convert Ipopt's 
Index-arrays to their int-arrays, when necessary (or convince them to 
use also a long for indices).

Stefan