the large eigenvalues. He can then re-arrange the equations to either

eliminate these terms from the Gauss-Seidel partition, or make them

outputs.

 The tear variable values used for the example problems were

computed using the method suggested by Naphtali and Sandholm (1971).

This method requires initial guesses for temperature and L/V ratios on

all stages. In addition, all K-values are assumed to be independent of

composition. The T and L/V values used were those used by Naphtali and

Sandholm in their examples. This results in the same starting point for

the solution procedures and makes a comparison more meaningful.

 The estimated variable values used in computing the eigenvalues

were order of magnitude guesses based on engineering intuition. The

more accurate the intuition, the better the eigenvalue estimates, and

the better the solution procedure.

 The algorithms successfully generated a Gauss-Seidel solution

procedure for the first problem, which was indeed solved by Gauss-

Seidel. It also generated a Gauss-Seidel solution procedure for the

second problem. Further analysis indicated that Gauss-Seidel would not

solve that problem. (The Gauss-Seidel solution procedure was tried for

the second problem and found to diverge even when ch correct values

were guessed for the tear variables.) A modified solution procedure,

which exhibited better convergence characteristics, was proposed and

found to be convergent.