Comparing LP Relaxations of Equivalent Formulations

Given two equivalent MIP formulations, compare the LP-relaxation feasible regions and optimal values to determine which formulation is stronger and gives a tighter bound on the integer optimum.

Step 1 of 157%

Tutorial

Comparing LP Feasible Regions

Two formulations of a mixed-integer program (MIP) are equivalent when they produce the same set of integer-feasible points. Their LP relaxations -- obtained by dropping integrality -- need not coincide.

Let PAP_A and PBP_B denote the LP feasible regions of two equivalent formulations AA and BB. Formulation AA is at least as strong as BB when

PAPB.P_A \subseteq P_B.

If in addition PAPBP_A \neq P_B, we say AA is strictly stronger than BB.

To prove PAPBP_A \subseteq P_B, show every constraint of BB is implied by the constraints of AA. To prove strictness, exhibit a fractional point in PBPAP_B \setminus P_A.

For example, take x[0,3]x \in [0,3], y{0,1}y \in \{0,1\}, with x=0x = 0 required when y=0y = 0. Compare

(A): x3y,(B): x5y.\text{(A): } x \le 3y, \qquad \text{(B): } x \le 5y.

If x3yx \le 3y holds, then x3y5yx \le 3y \le 5y (since y0y \ge 0), so PAPBP_A \subseteq P_B. The point (x,y)=(3,0.7)(x,y) = (3, 0.7) satisfies 35(0.7)=3.53 \le 5(0.7) = 3.5 but fails 33(0.7)=2.13 \le 3(0.7) = 2.1, so (3,0.7)PBPA(3, 0.7) \in P_B \setminus P_A. Therefore AA is strictly stronger than BB.

navigate · Enter open · Esc close · ⌘K/Ctrl K toggle