A fractional LP optimum is tempting to round, but rounding is not a proof. It can leave the feasible region or land on a weaker integer point. This lesson shows both failures on the same asymmetric IP.

highlighted = computed this step

Rounded point

Rounding the LP point gives (1, 2), but x plus three y equals 7 and the limit is 6. Why: rounding can cross a constraint boundary.

x+3y=7x+{}3y=7
Rounding checkRounding and the true integer optimum are marked against the LP relaxation.Feasible region2x + y ≤ 4x + 3y ≤ 6(0, 0)(0, 2)(6/5, 8/5)(2, 0)roundedoptimal z=22/5

Nearby feasible point

The feasible point (1, 1) has z=3. Why: a nearby feasible integer point can still miss the best integer point.

z=3z=3
Rounding checkRounding and the true integer optimum are marked against the LP relaxation.Feasible region2x + y ≤ 4x + 3y ≤ 6(0, 0)(0, 2)(6/5, 8/5)(2, 0)roundedoptimal z=22/5

True integer optimum

The true integer optimum is (0, 2) with z=4. Why: branch-and-bound proves the best point instead of guessing by rounding.

zIP=4z_{IP}=4
Rounding checkRounding and the true integer optimum are marked against the LP relaxation.Feasible region2x + y ≤ 4x + 3y ≤ 6(0, 0)(0, 2)(6/5, 8/5)(2, 0)roundedoptimal z=22/5

Diagram note

The rounded marker is intentionally outside the feasible region; the green marker is the recomputed IP optimum. Pixel positions are rounded for layout; every number shown is exact.

rounding is not a certificate\text{rounding is not a certificate}
Rounding checkRounding and the true integer optimum are marked against the LP relaxation.Feasible region2x + y ≤ 4x + 3y ≤ 6(0, 0)(0, 2)(6/5, 8/5)(2, 0)roundedoptimal z=22/5