[Ipopt] understanding the filter method

Ipopt User ipoptuser at gmail.com
Fri Aug 26 16:09:29 EDT 2016

I'm using ipopt in l-bfgs mode with line search enabled. In the output, I'm
puzzled by iteration 51:

  50  1.3352705e+01 1.11e+00 4.96e+08   1.0 2.84e+00    -  7.92e-02
3.33e-03h  1
  51  2.7220570e+01 2.11e+00 5.90e+08   1.0 1.87e+00    -  4.34e-03
3.38e-01f  1

I think the 'f' means that the filter method was used for this iteration,
and that an improvement in the objective value was made. However, the
objective value becomes considerably worse (and so does the infeasibility).
Why is this step taken (instead of a line search)?

The filter method is described in the preprint ipopt paper (
http://www.optimization-online.org/DB_FILE/2004/03/836.pdf ). I think (19)
does not hold, but (20) does, right? Probably because some variables move
away from their bounds, right?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20160826/84df3b41/attachment.html>

More information about the Ipopt mailing list