UNSPECIFIED (1998) An efficient Monte Carlo approach to optimization. JOURNAL OF CHEMICAL INFORMATION AND COMPUTER SCIENCES, 38 (5). pp. 791-797. ISSN 0095-2338

Full text not available from this repository.

Abstract

A highly efficient Monte Carlo algorithm for global optimization has been developed which accepts beneficial moves, rejects all detrimental ones, picks a new step size at random from a guided range, and samples a new region of the response surface using a randomly generated directional search technique. The step size guide is the modified Heaviside function: r = r(0) 0 < m less than or equal to nn), r = r/2 + 1/r (nn < m less than or equal to 3nn),and r = {r/2 + 1/r} sin(2r(0)r) (m > 3nn), where m is the number of frustrated steps and nn a decision parameter. This approach quickly steps through local; optima. Two strategies have been evolved leading to solutions for both the highly intractable modified n-dimensional shekel function and COSn, the "cosine function". The strategies are efficient: starting from the point (10,10), an average of only 187 steps (successful and otherwise) taking a total of 0.22 shekels of time were required to find the minimum of the COS2 function to three significant figures with a success rate of 987/1000. In the case of COS4 the current calculations started from {x(i) = 10} and found the global minimum to three significant figures in 864 steps and 1.80 shekels with a success rate of 999/1000. COS100 has also been solved. In almost all comparisons with currently accepted methods, considerable efficiencies are achieved, and, more importantly, global optimization could be implemented in all test cases.

Item Type:	Journal Article
Subjects:	Q Science > QA Mathematics > QA76 Electronic computers. Computer science. Computer software Q Science > QD Chemistry
Journal or Publication Title:	JOURNAL OF CHEMICAL INFORMATION AND COMPUTER SCIENCES
Publisher:	AMER CHEMICAL SOC
ISSN:	0095-2338
Date:	September 1998
Volume:	38
Number:	5
Number of Pages:	7
Page Range:	pp. 791-797
Publication Status:	Published
URI:	http://wrap.warwick.ac.uk/id/eprint/15324

Data sourced from Thomson Reuters' Web of Knowledge

Request changes to a record

Actions (login required)

View Item