[Ipopt] complex step in derivative checker for matlab interface?

[Stefan Vigerske]

Hi,

[1] No.

[2] The derivative checker will tell when it sees a nonzero value in its 
Jacobian approximation that hasn't been specified inn the users sparsity 
pattern. See the discussion on the 'v' entry at 
https://www.coin-or.org/Ipopt/documentation/node30.html

Stefan

On 02/01/2017 05:56 PM, Martin Neuenhofen wrote:
> Hi all,
>
> can one overload the derivative checker with a complex step for the matlab
> interface? [1]
>
> I am just observing that whereas my complex step validated derivatives
> indicate accuracy up to the 15th digit the ipopt derivatives checker
> indicates that my gradients would have relative errors of 1e2 (so more than
> completely wrong).
> In case that not has anyone an idea what can be done. Can one for instance
> check if Ipopt got the sparsity pattern and non-zero alignment right? [2]
>
> Bug report: Apart from the above I observed that the provided Jacobian
> sparsitiy pattern has an influence on the derivative check. Giving a
> sparsitiy pattern that indicates full dense-ness and filling up the
> Jacobian by adding 1e-200 to each zero entry the derivative checker checks
> all numbers whereas otherwise it seems to check exclusively the non-zeros
> (since each second zero was thrown as an error when using pseudo-dense
> pattern whereas with actual sparse pattern it did not even find one of
> them).
>
> Kind regards
> Martin
>
>
>
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/ipopt
>


-- 
http://www.gams.com/~stefan