[Ipopt] Computing the jacobians and hessians for IPOPT in single precision

[Daniel Arteaga]

Hi all,

We are using automatic differentiation with Tensorflow to compute the
derivatives (jacobians and hessians) for IPOPT. Tensorflow can only compute
such derivatives in single precision.

We were finding some issues with the stability of the minimization, and we
were wondering if using single precision derivatives could be the root
cause.

Do you think it is acceptable to compute derivatives with single precision?
Function evaluations are still computed with double precision.

Thank you very much,

Daniel
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20180726/86923437/attachment-0001.html>