Statistical Methods
Check Integer Decisions Without Rounding a Relaxed Solution
Define the task
Whole sessions, kits or other indivisible units require integer decision variables. Mark which variables must be integers and which may vary continuously. Bounds alone do not impose integrality. Keep the objective, resource constraints and variable units with the model specification. A continuous relaxation can give an informative bound while still failing to produce an implementable allocation.
Carry out the check
Use an integer-capable solver and inspect termination status, feasible incumbent, objective bound and optimality gap. A time-limited feasible solution is not automatically proven optimal. Recompute all constraints after interpreting numerical tolerances; never round an arbitrary relaxed vector and assume feasibility or optimality survives. Modeling choices remain the owner's responsibility: an optimizer follows the encoded objective and cannot validate omitted practical constraints.
Worked hypothetical example
Hypothetical example: maximize 3x+2y subject to 2x+2y≤3 with nonnegative integer x,y. The continuous optimum is x=1.5,y=0, worth 4.5. Rounding x upward gives 2 and consumes 4 units, violating capacity 3. Enumerating the feasible integer points (0,0), (1,0), (0,1) gives values 0,3,2; the integer optimum is (1,0). This small example can be checked manually, unlike large instances where a solver certificate and explicit stopping criteria matter.
Method checked 22 November 2026 against Primary method documentation: milp.