mcp06.gms : Test level/marginal correctness and conventions for MCP

Description

This model illustrates and tests the convention for returning
equation/variable levels/marginals for an MCP.

Assume we have a complementarity problem defined by
model m /f.x/.  The values to expect in GAMS are:
  x.l: solution computed by the solver
  f.l: computed as for all GAMS models (plug and chug)
  x.m: f.l - f.up/f.lo (whichever is not infinity)
       what if both are infinity??
  f.m: x.l

AFAIK, what is illustrated here has been the accepted behavior,
consistently implemented, for quite some time.

Contributor: Steve Dirkse, June 2006


Small Model of Type : MCP


Category : GAMS Test library


Main file : mcp06.gms

$title 'Test level/marginal correctness and conventions for MCP'  (MCP06,SEQ=340)

$onText

This model illustrates and tests the convention for returning
equation/variable levels/marginals for an MCP.

Assume we have a complementarity problem defined by
model m /f.x/.  The values to expect in GAMS are:
  x.l: solution computed by the solver
  f.l: computed as for all GAMS models (plug and chug)
  x.m: f.l - f.up/f.lo (whichever is not infinity)
       what if both are infinity??
  f.m: x.l

AFAIK, what is illustrated here has been the accepted behavior,
consistently implemented, for quite some time.

Contributor: Steve Dirkse, June 2006
$offText

$if not set TESTTOL $set TESTTOL 1e-6
scalars
  tol / %TESTTOL% /,
  failed / 0 /;

$escape =
$echo if{%=1, display '%=1 failed', '%=2'; failed=1}; > gtest.gms
$escape %

set J / j1 * j4 /;

parameters
   erhs(J)   /  j1      -1.0
                j2      1.0
                j3      -1.0
                j4      1.0 /
   ex_lo(J)   / j3      1.0 /
   ex_up(J)   / j4      -1.0 /
   ex_l(J)    / j1      -1.0
                j2      1.0
                j3      1.0
                j4      -1.0 /
   ex_m(J)    / j1      0
                j2      0
                j3      2.0
                j4      -2.0 /
   ;

equations
  ef(J);

variable
  ex(J);

ef(J).. ex(J) =e= erhs(J);

model m / ef.ex /;

ex.lo(J) $= ex_lo(J);
ex.up(J) $= ex_up(J);

solve m using mcp;

$ batInclude gtest "( m.solvestat <> %solveStat.normalCompletion% or (m.modelstat > %modelStat.locallyOptimal% and m.modelstat <> %modelStat.feasibleSolution%))" "wrong status codes"
$ batInclude gtest "(smax(J, abs(ex.l(J)-ex_l(J))) > tol)"    "bad ex.l"
$ batInclude gtest "(smax(J, abs(ex.m(J)-ex_m(J))) > tol)"    "bad ex.m"
abort$failed 'test failed';