[Ipopt] Restoration phase messing up convergence

Sun Apr 1 14:09:40 EDT 2018

Hi all,

On multiple occasions I have seen ipopt initiating the restoration phase when this does not seem necessary, since the problem seems to be converging (the primal infeasibility is decreasing). Moreover, after the restoration phase has completed, the solution is worse than before. This effectively causes my optimizations not to converge.

See e.g. below where the objective, primal and dual infeasibilities are smaller before restoration (iteration 310) than after (iteration 330). Is there

Can someone explain why the restoration phase is initiated and why it is allowed to terminate with a worse result than the initial result? I thought its purpose is to reduce the primal infeasibility, which it does not seem to do?

Thank you!

Filip

iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
 300  4.9413587e+01 1.39e+01 9.19e+01  -6.3 7.62e+04    -  6.08e-02 5.90e-03h  4
 301  4.9434016e+01 1.38e+01 9.31e+01  -6.3 7.61e+04    -  6.43e-02 5.32e-03h  6
 302  4.9439824e+01 1.38e+01 9.28e+01  -6.3 5.42e+04    -  6.23e-03 1.42e-03h  2
 303  4.9445688e+01 1.37e+01 9.18e+01  -6.3 5.45e+04    -  1.46e-02 1.42e-03h  1
 304  4.9462371e+01 1.37e+01 9.24e+01  -6.3 5.53e+04    -  1.56e-03 3.98e-03h  2
 305  4.9481166e+01 1.36e+01 8.90e+01  -6.3 5.93e+04    -  4.32e-02 4.48e-03h  1
 306  4.9502736e+01 1.36e+01 8.87e+01  -6.3 6.90e+04    -  5.57e-04 5.44e-03h  3
 307  4.9532790e+01 1.34e+01 1.14e+02  -6.3 6.87e+04    -  9.69e-03 7.62e-03h  3
 308  4.9555527e+01 1.34e+01 1.14e+02  -6.3 6.97e+04    -  1.72e-02 5.87e-03h  1
 309  4.9616944e+01 1.32e+01 1.10e+02  -6.3 1.01e+05    -  2.95e-04 1.50e-02h  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
 310  4.9911875e+01 1.22e+01 1.15e+02  -6.3 7.60e+04    -  3.04e-04 7.45e-02h  1
 311r 4.9911875e+01 1.22e+01 9.99e+02   4.0 0.00e+00    -  0.00e+00 3.72e-07R  3
 312r 4.9893340e+01 1.22e+01 9.99e+02   4.0 5.87e+11    -  4.52e-09 2.90e-10f  1
 313r 5.1041288e+01 2.67e+02 9.50e+03   3.3 8.75e+09    -  6.55e-04 6.75e-06f  1
 314r 6.5746661e+01 5.21e+01 9.49e+03   3.3 3.62e+07    -  1.25e-03 1.56e-03f  1
 315r 8.9289704e+01 2.21e+02 9.47e+03   3.3 1.73e+07    -  2.66e-03 2.88e-03f  1
 316r 1.4193131e+02 4.56e+02 9.40e+03   3.3 4.48e+06    -  8.24e-03 8.52e-03f  1
 317r 1.7319660e+02 4.04e+03 9.34e+03   3.3 1.68e+06    -  6.72e-03 7.90e-03f  1
 318r 2.5758435e+02 4.35e+02 9.15e+03   3.3 1.00e+06    -  2.15e-02 2.50e-02f  1
 319r 3.4323045e+02 5.29e+03 8.84e+03   3.3 4.99e+05    -  3.40e-02 3.47e-02f  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
 320r 5.6263873e+02 3.56e+02 7.79e+03   3.3 2.06e+05    -  1.25e-01 9.60e-02f  1
 321r 9.1823497e+02 3.70e+04 6.55e+04   3.3 4.41e+04    -  1.24e-01 2.75e-01f  1
 322r 1.2039772e+03 1.41e+02 3.93e+03   3.3 5.18e+04    -  8.78e-01 3.74e-01f  1
 323r 1.5456710e+03 1.16e+05 6.62e+03   3.3 8.43e+03    -  7.80e-01 1.00e+00f  1
 324r 1.5256602e+03 2.62e+01 1.25e+04   3.3 4.72e+03    -  1.00e+00 1.00e+00f  1
 325r 4.4331694e+02 4.21e+04 1.51e+04   2.6 2.20e+03    -  8.58e-01 8.99e-01f  1
 326r 6.7541139e+02 4.89e+04 4.87e+03   2.6 5.59e+03    -  6.66e-01 1.00e+00f  1
 327r 9.6890867e+02 8.97e+03 1.53e+03   2.6 1.57e+04    -  1.00e+00 1.00e+00f  1
 328r 9.7454742e+02 1.97e+01 3.14e+03   2.6 6.34e+03    -  1.00e+00 1.00e+00f  1
 329r 9.5111247e+02 4.17e+01 8.34e+02   2.6 8.46e+02    -  1.00e+00 1.00e+00h  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
 330  9.4599206e+02 4.41e+01 1.51e+02  -6.3 6.78e+06    -  2.80e-03 3.04e-03h  1

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20180401/3889869e/attachment.html>