A substantial part of the work was the test of programs. All
tests were performed with a maximum of 20000 function evaluations.
The tables given below contain not the genuine
function values f, but modified f ':
f* is the best function value of a specific testfunction calculated by any program. fm is defined, as by each program the median of all function evaluations of the current test problem is calculatet and then the median of this mediane is determined.
This table
contains the best function
values found after 1000, 2500, 5000, 10000 and 20000 evaluations.
The best function values found (of a test problem) have been underlined.
Not all programs always achieved the necessary number of
function evaluations; In this case the best value found by the program
is given in the table.
The following two links show tables, subject to the
following abort criterion:
fk is the best function value found
after at most k function evaluations.
I also expandet a table showing some results on the testset of Dixon and Szegö originally published by Neumaier and Huyer. Shown are the number of evaluations needet to approximate a global optimum with a relative error of .01% (12000 evaluations maximum).
Here is a short
description of the test functions
,
a
position table of the functions
representing the difficulty to optimize them,
and the following link leads to a descripton of the
tested programs
(including further links).
The page of Arnold Neumaier on global optimization -- Global Optimization (Arnold Neumaier) -- contains still much further information about this topic.