m as having the largest activity coefficient rather than activity. But this alternative cannot be used with dilute solutes having high activity coefficients or vapor trial phases.In the region of a plait point, the phase-splitting algorithm predicts two similar trial phases and four iterations of the minimization algorithm are often not sufficient to reduce Gibbs free energy. Occasionally 6 or 7 iterations are necessary.Similar tests of the phase-splitting algorithm should be successful for phase diagrams with at least one partially miscible pair. For a design study, it is important to identify regions where the algorithm fails, and either modify or avoid using the algorithm in these regions. Systems must contain at least one partially miscible pair for the phase-splitting algorithm to apply.The decision to accept the first trial phases that satisfy AG:FtiQ)' > 10-5G(k), rather than examine all combinations of source and trial phases, saves much computer time. But, it gives rise to the possibility that the same two trial phases are accepted repeatedly, with one rejected after q iterations of the minimization algorithm, preventing other trial phases from being considered. As an example, for a VLlLz system at equilibrium, with initial guesses for two liquids L1L2, it might be possible for the sequence t l L $ 3 LIL& + LILz to be repeated, preventing examination of the trial phases L2V. This is highly unlikely, as the composition of phases LILz would be closer to a constrained minimum in Gibbs free energy prior to the next split and the criteria, AGtltZ > 10-5G(k', would not likely be satisfied.