Product Engineering
The window can be empty — and knowing that beats an optimum that does not exist.
— units
Feasible band satisfying both specifications; negative means none exists
Both responses are modelled as linear in one factor, which is a first-order approximation of a genuine response surface — real optima are usually curved and a linear fit extrapolated beyond the data that produced it is not a prediction. Fit the coefficients from a designed experiment across the range you intend to run, and treat the window as provisional until confirmed by a trial at the midpoint. A window narrower than the process can actually hold is the same as no window at all: compare the width against the standard deviation of the factor in production before believing it. A single midpoint also ignores which margin is more expensive to lose.
Response Surface Operating Window Optimizer — free, with the formula and a worked example, at Textile School.