Shadow prices are not extrapolation rules. This lesson pushes the RHS beyond the allowable range, shows the old basis becoming infeasible, and recomputes the LP.

highlighted = computed this step

Too far

Increase the first RHS by 5. Motivation: this deliberately moves past the allowable range endpoint.

Δb1=5\Delta b_1=5
Outside the allowable rangeThe old basis candidate has a negative coordinate, so the LP is recomputed.Feasible region2x + y ≤ 9x + 2y ≤ 4(0, 0)(0, 2)(4, 0)optimal z=4

Old basis candidate

The old basis candidate is (14/3, -1/3). Why: its negative coordinate violates the nonnegative bound, so the old certificate cannot be reused.

old basis candidate is infeasible\text{old basis candidate is infeasible}
Outside the allowable rangeThe old basis candidate has a negative coordinate, so the LP is recomputed.Feasible region2x + y ≤ 9x + 2y ≤ 4(0, 0)(0, 2)(4, 0)optimal z=4

Recomputed optimum

The stale prediction is 13/3, but the recomputed optimum is (4, 0) with value 4. Interpretation: outside the range, recompute beats extrapolation.

znew=4z_{\text{new}}=4
Outside the allowable rangeThe old basis candidate has a negative coordinate, so the LP is recomputed.Feasible region2x + y ≤ 9x + 2y ≤ 4(0, 0)(0, 2)(4, 0)optimal z=4

Certificate note

RHS sensitivity is local: the shadow-price prediction is exact while the same basis remains primal feasible and dual feasible. Outside that range, the LP must be recomputed. Pixel positions are rounded for layout; every number shown is exact.

range failure forces a new certificate\text{range failure forces a new certificate}
Outside the allowable rangeThe old basis candidate has a negative coordinate, so the LP is recomputed.Feasible region2x + y ≤ 9x + 2y ≤ 4(0, 0)(0, 2)(4, 0)optimal z=4