We will use a geometric interpretation to informally show that solving (4c) and (4d) not only meets the necessary conditions, but also guarantees that the confidence interval width is minimized. Consider the example problem presented in Figure 1 where T follows a binomial distribution with n = 20 and to = 4, and CL = .95. We can define the feasible region in the (θ_H,θ_L) plane using constraints (3b)-(3d).