Execution errors during model generation

 Top  Previous  Next

Execution errors may arise when GAMS is generating the model before passing it to the solver.  These again can be arithmetic errors or can also be model structure errors.  Such model structure errors arise because some equations are improperly set up i.e. being inherently infeasible or the wrong solver is being used.

Discovery of such execution errors is sometimes very straightforward but can at other times be fairly involved.  An error in the middle of a multi-dimensional equation block and/or in a multi-dimensional equation term within a block can be difficult to find.  The most practical way of finding such errors is to use the Limcol/Limrow option commands.

Consider the example executmd.gms

 

sets elems /s1*s25/

parameter data1(elems)   data to be exponentiated

          datadiv(elems) divisors

          datamult(elems)     x limits;

data1(elems)=1;

data1("s20")=-1;

datadiv(elems)=1;

datadiv("s21")=0;

datamult(elems)=1;

datamult("s22")=0;

positive variables x(elems) variables

variables  obj;

equations  objr        objective with bad exponentiation

           xlim(elems) constraints with bad divisor;

objr.. obj=e=sum(elems,data1(elems)**2.1*x(elems));

xlim(elems).. datamult(elems)/datadiv(elems)*x(elems)=e=1;

model executerr /all/

option limrow=30; option limcol=30;

solve executerr using lp maximizing obj;

 

In this example, we will have execution errors arise during model generation because of arithmetic problems.  Namely

•In the objective function for the "S20" element where we are exponentiating a negative constant to a real power (because of the assignment in line 9);
•In the XLIM constraint associated with element "S21" where we are dividing by zero (because of the assignment in line 11); and
•In the XLIM "S22" constraint where we set zero equal to one which results in an infeasible constraint (because of the assignment in line 15).

When we run GAMS, we see the execution error messages as follows:

 

--- Generating model EXECUTERR

--- EXECUTMD.GMS(16) 134 Kb

*** ExecError 10 at Line 16

    ILLEGAL ARGUMENTS IN ** OPERATION

--- EXECUTMD.GMS(17) 134 Kb 1 Errors

*** ExecError 0 at Line 17

    DIVISION BY ZERO

*** ExecError 28 at Line 17

    EQUATION INFEASIBLE DUE TO RHS VALUE

--- EXECUTMD.GMS(20) 134 Kb 3 Errors

*** SOLVE aborted

*** Status: Execution error(s)

 

The Limrow section of the LST file reveals

 

**** EXECUTION ERROR 10 AT LINE 16 .. ILLEGAL ARGUMENTS IN ** OPERATION

---- OBJR        =E=  objective with bad exponentiation

OBJR..  - X(s1) - X(s2) - X(s3) - X(s4) - X(s5) - X(s6) - X(s7) - X(s8)

      - X(s9) - X(s10) - X(s11) - X(s12) - X(s13) - X(s14) - X(s15) - X(s16)

      - X(s17) - X(s18) - X(s19) + UNDF*X(s20) - X(s21) - X(s22) - X(s23)

      - X(s24) - X(s25) + OBJ =E= UNDF ; (LHS = UNDF, INFES = UNDF ***)

 

**** EXECUTION ERROR 0 AT LINE 17 .. DIVISION BY ZERO

**** EXECUTION ERROR 28 AT LINE 17 .. EQUATION INFEASIBLE DUE TO RHS VALUE

 

**** INFEASIBLE EQUATIONS ...

---- XLIM        =E=  constraints with bad divisor

XLIM(s22)..  0 =E= 1 ; (LHS = 0, INFES = 1 ***)

 

This does not display the equation limrow listing for the divide by zero error.  To get it you have to fix the infeasibility and run again.  Then you get

 

XLIM(s21)..  UNDF*X(s21) =E= UNDF ; (LHS = UNDF, INFES = UNDF ***)

 

Regardless the Limrow display shows the exact elements within the model where the problems have arisen and one may then investigate.  Such an investigation may involve displays of the input data used within the equation calculations so one can investigate the numerical properties of specific elements associated with problems.  Often even more displays will be involved where one traces faulty input data back through the program investigating places where these data have been calculated to eventually see why these data have taken on the specific numerical values they have.