0

I have formulated a NLP problem where after solving I get following output. 

Iter Phase Ninf Infeasibility RGmax NSB Step InItr MX OK
0 0 2.0000000000E+01 (Input point)

Pre-triangular equations: 15
Post-triangular equations: 36

1 0 0.0000000000E+00 (After pre-processing)
2 0 0.0000000000E+00 (After scaling)

** Feasible solution. Value of objective = 62.6329262813

Iter Phase Ninf Objective RGmax NSB Step InItr MX OK
4 3 6.2632926281E+01 0.0E+00 0

** Optimal solution. There are no superbasic variables.

--- Restarting execution
--- _gams_py_gjo0.gms(143) 2 Mb
--- Reading solution for model SP_loc_dep_0
--- Executing after solve: elapsed 0:00:06.836
--- _gams_py_gjo0.gms(141) 3 Mb
*** Status: Normal completion
--- Job _gams_py_gjo0.gms Stop 12/03/17 14:37:32 elapsed 0:00:06.838

I have couple of questions.

Q1-It seems like the process had a normal completion. But my problem is a convex problem, so I should have a global optimum. However, the solver outputs Local Optimum.

Q2-In my equations I see following. Where (1) is in terms of exponential. Does this mean my constraints are infeasible? I do not get any infeasibility error in my output.

---- P1eqS =E= 


P1eqS(1,1,1).. (1)*alphaS(1,1,1) + P1S(1,1,1) =E= 1 ; (LHS = 0, INFES = 1 ****)
P1eqS(1,2,1).. P1S(1,2,1) =E= 1 ; (LHS = 0, INFES = 1 ****)
P1eqS(2,1,1).. (1)*alphaS(2,1,1) + P1S(2,1,1) =E= 1 ; (LHS = 0, INFES = 1 ****)
deepAgrawal
  • 673
  • 1
  • 7
  • 25
  • I have no clue about Gams, but: (1) This question seems broad and i think it's hard to help given this sparse info (2) There are convex-optimization problems with non-convex formulations. So a problem being convex does not necessarily lead to a convex-optimization problem where the solver finds the global-optimum! – sascha Dec 03 '17 at 22:06

1 Answers1

2

(1) You solved this NLP problem with Conopt. Conopt is a local solver, so it can only guarantee to find local solutions. Conopt does not know if the model is convex or not. (Rigorously checking whether a model is convex is not completely trivial).

(2) The fragment of the listing file

P1eqS(1,1,1).. (1)*alphaS(1,1,1) + P1S(1,1,1) =E= 1 ; (LHS = 0, INFES = 1 ****)
P1eqS(1,2,1).. P1S(1,2,1) =E= 1 ; (LHS = 0, INFES = 1 ****)
P1eqS(2,1,1).. (1)*alphaS(2,1,1) + P1S(2,1,1) =E= 1 ; (LHS = 0, INFES = 1 ****)

is the equation listing. The messages about infeasibilities are related to the initial point (i.e., before the solver is called). Note that the CONOPT log confirms that the initial point is infeasible.

The solution part in the listing is further down.

Erwin Kalvelagen
  • 15,677
  • 2
  • 14
  • 39