[Ipopt-tickets] [Ipopt] #237: bug with "hessian_constant" (C++ library)

[Ipopt]

#237: bug with "hessian_constant" (C++ library)
-------------------+------------------------
Reporter:  namzug  |      Owner:  ipopt-team
    Type:  defect  |     Status:  new
Priority:  normal  |  Component:  Ipopt
 Version:  3.11    |   Severity:  major
Keywords:          |
-------------------+------------------------
 When the option "hessian_constant" is left as "no", a quadratic model
 solves as expected.  However, when the option is set to "yes", models
 often take longer to converge or do not converge at all.  The derivative
 checker also does not indicate any errors.

 It should be noted that in the provided example "Cpp_example", if the
 option is set to "yes", on my system, the number of iterations required to
 solve the example increases from 6 to 10.

 Attached is a small, self-contained C++ source file that implements a
 small quadratic model (shown at the top in AMPL form... the binary
 variables are just continuous [0,1] variables in the C++ code).  The
 model, with the option set to "yes", stops after 1071 iterations due to
 the "acceptable level" heuristic.  When the option is set to "no", it
 solves in 27 iterations.  There are lines in the code that output some
 diagnostic Hessian information (using "cout") for convenience.  One could
 comment out the line setting "hessian_constant" to yes and get much faster
 convergence despite the constant 2nd derivatives being returned.

 Thank you.

--
Ticket URL: <https://projects.coin-or.org/Ipopt/ticket/237>
Ipopt <http://projects.coin-or.org/Ipopt>
Interior-point optimizer for nonlinear programs.