rvals,and then are the equal intevals,and at the end are continuous. 3 A Comparative Study of Algorithms in MOD In order to cnsure the quality of the software syslem,it was tested to take in whole of the package MOD based on the preliminaty test of every algori- thm. The problems used in the testing are thirty five classical ones which were selccted so as to include a wide range in various open literatures.These pro- blems are inclusive of the variable number from 2 to 48 and constraints number from 1 to 58 and five classical geometric programming problems.All are the integer,discrete or mixcd variables except for two all continuous variable pro- blems.These problems were made up of two sets:TEST-I and TEST-I. The information obtained during the examining collection data consists of objective and constrained function values and their evaluated numbers,and the solving time for each algorithm on one problem.In addition the relative error in the objective function is defined by following criterion =f()() or er=If(x)] (15) f(x·)川 Where the f(x)is the value of the objective function at a found optimal point on one problem for each algorithm,The f(x)is discrete optimum of reference that is obtained by above two algorithms. The number and time of solved problems of each algorithm are the two main characteristics,but both must be considered at the same time in order to produce a valid indication of the performance.On the basis of this,the cri- terion used in the comparison is the solved number under a reasonable amount of time and relative error accuracy.Here the reasonable amount of time means a series of specified limits on fraction of the average sofution time for all of the algorithms on each problem. This plot at relative error level of 0.05 and relating to the front fiften problems in TEST-I is presented in Fig.3 and Fig.4.This plot,the so-called algorithm preformance cuves,may be thought of as a utility graph for algori- thm quality comparison. Thus a good algorithm has to have a steep rise and attain a high maximum value on the vertical axis,We can quite easily pick out the superior algori- thm from this curve,the MDOD has the best characteristic following it are the MDCP,MDHP and MDGP,and then is MDRP,But worth mentioned that No.14 and No.15 problems include the posynomial function and the superior re- ference optimum is found by means of MDGP,but solving them by making use 344、 ， ， 。 ， 勺 ， 。 了 、 。 ， 。 一 一 ， 。 了 厂 忍 二 一 二 ’ 。 二 ’ ， 。 ， 皿 。 一 。 ， 一 ， 爪 主 主 。 了 ， ， 〔 ， 。 ， 、