This paper presents an improved algorithm for adaptive random search for global optimization problems. The method is compared to others for the same problem class and reported to be equal to or better than the other candidates.

The authors have included algorithmic steps for implementation of the method. They report results from several standard test cases, as well as two parameter estimation problems that motivated the work. The implementation is in general use in their lab. The major advantage of this improvement to adaptive random search is the robustness of the method for even naive users.

The paper is clearly written and the references are complete. The test results are not as complete as they should be. Only one sequence of iterations is reported for each problem. In order to infer algorithm performance, the authors should have made several (many) sequences in the usual fashion for Monte Carlo methods. Their test case analysis is just as suspect as the results for one simulation of a queueing network. Treatment of the initial transient and the use of multiple random number seeds are critical elements of the analysis that have been ignored in this report.