[Ipopt] regularization for projected spmd Hessians

[Martin Neuenhofen]

Dear all,

I have a problem where the Hessian is computed as HL = P' * D * P, where D
is diagonal non-negative. P is size m x n with m>=n and has column rank n.
Since P could be augmented by columns to something invertable Sylvesters
theorem applies and HL should be positive semi-definite as well. However, I
observe that Ipopt with solvers ma27 as well as ma97 is taking considerable
large regularization parameters. Does anybody know why it does so? I would
have hoped that no regularization is taken.

Kind regards,
Martin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20170419/f4cc8868/attachment.html>