We also found that TNPACK with the standard MC did not work for large-scale molecular minimization problems such as BPTI.
Comparison of TNPACK with Other Two CHARMM Minimizers and LM-BFGS for BPTI (1704 Variables)
(min.) TNPACK 1.14 x [10.sup.-6] 227 5.7 LM-BFGS 6.3 x [10.sup.-5] 4622 12.61 ABNR 8.9 x [10.sup.-6] 8330 25.17 CONJ 9.9 x [10.sup.-6] 32661 97.80 Table VI compares the performance of TNPACK for BPTI minimizations with two other CHARMM minimizers, ABNR (an adopted-basis Newton-Raphson method) and CONJ (a nonlinear conjugate gradient method), as well as LM-BFGS with u = 5 stored updates [Liu and Nocedal 1989].
Performance of TNPACK for BPTI Using Criterion 1 (C1) versus Criterion 2 (C2) in the Line Search
(min.) C1 176 (1387) 537 6.99 C2 85 (1029) 250 4.35 To show that a significant improvement can be observed when Criterion 2 (C2) of the line search is used, we consider TNPACK for BPTI potential function minimization.