Re: checking large scale linear programming solution with

*To*: mathgroup at smc.vnet.net*Subject*: [mg113397] Re: checking large scale linear programming solution with*From*: Daniel Lichtblau <danl at wolfram.com>*Date*: Wed, 27 Oct 2010 05:15:05 -0400 (EDT)

Mala Jozsef wrote: > Dear Group, > NMinimize sometimes gives a[39,3]->1.0000000000000178` as a solution > instead of 1 when dealing with large scale linear programming task. Moreover, > Mathematica sometimes gives a "solution" when there is no feasible one. > Here is an example: > In[207]:= NMinimize[{1,x+y>=0,x+y<=-1,Element[x,Integers]},{x,y}] > Out[207]= {1.,{x->-1,y->0.}} > I would like to check a solution with a given accuracy, say, some > digits. Is there any "almost built in" or easy solution? > Thanks in advance. > Jozsef NMinimize will attempt to enforce constraints but will not necessarily succeed. Minimize will enforce them but requires either a purely integer or real program, not mixed. In principle one could recast using Linearprogramming, as that allows for mixed problems. Unfortunately you then hit a bug wherein it might also fail to report infeasibility. http://forums.wolfram.com/mathgroup/archive/2009/Jan/msg00296.html This has been fixed for an upcoming release. Daniel Lichtblau Wolfram Research