Р. Г. Стронгина. Ниж- ний Новгород: Изд-во Нижегородского университета, 2002, 217 с


PARALLEL GLOBAL OPTIMIZATION ALGORITHMS



Pdf көрінісі
бет134/151
Дата26.01.2022
өлшемі1,64 Mb.
#24342
түріСеминар
1   ...   130   131   132   133   134   135   136   137   ...   151
Байланысты:
Seminar 1

PARALLEL GLOBAL OPTIMIZATION ALGORITHMS
*
 
Yaroslav D. Sergeyev 
 University of Calabria, ITALY  
and University of Nizhni Novgorod, RUSSIA 
Abstract 
In this paper two types of global optimization problems with the objec-
tive function determined over a hyperinterval are considered. The first one 
is the class of Lipschitz functions and the second one is its subclass: func-
tions having Lipschitz first derivatives. For solving these problems we use 
sequential and parallel characteristical algorithms with and without deriva-
tives. Convergence conditions and conditions, which guarantee significant 
speed up in comparison with the sequential versions of the parallel methods, 
are investigated from a general viewpoint. Numerical experiments executed 
on parallel computers illustrate performance of the parallel methods. 
 
                                                           
*
 This work was supported by the grant № 01-01-578 of Russian Fund of Basic 
Research. 


 
189 
This paper deals with the following well known global optimization 
problem of finding the global minimum of a function 
Φ(z) of variables 
over a hyperinterval D: 
 
 
Φ(z*) = min{Φ(z) : zD}, (1) 
 
 
D = {z 
R
N
 : a
j
 
≤ z
j
 
≤ b
j
1< j <N], 
(2)  
 
where 
Φ(z) is multiextremal. If Φ(z) is continuous then, for the function 
 
 
φ(x) = Φ(z(x)),  x∈[a,b], (3) 
 
where  z(x)  is the continuous Peano-type mapping of closed interval [a, b] 
onto the hyperinterval D, we have 
 
 min{
φ(x) : x∈[a,b]} = min{Φ(z) : zD }. 
 
Therefore, solving the multidimensional problem (1), (2) is reduced to 
solving the one-dimensional problem 
 
 
φ(x*) = min{φ(x) : x∈[a,b]}, (4) 
 
In addition, if 
Φ(z) is Lipschitzean with a constant K > 0 then φ(x) satis-
fies the Hölder condition 
 
 
|
φ(x′) – φ(x′′)| ≤ H |x′ – x′′|
1/N
,  x′, x′′
∈[a,b], (5) 
 
with a constant H 
≥ 0 (Hölder constant). For N = 1 we will also consider the 
problem (4) where the first derivative 
φ(x′) satisfies the Lipschitz condition 
 
 
|
φ′(x′) – φ′(x′′)| ≤ L |x′ – x′′|,  x′, x′′∈[a,b], (6) 
 
For solving the problems (4)–(5), (4)–(6) it is proposed to use parallel 
characteristical algorithms generalizing sequential characteristical algo-
rithms. A global optimization algorithm is called a parallel characteristical 
algorithm if trial points (i.e. that ones where we evaluate 
φ(x) and for the 
problem (4)–(6) also 
φ′(x)) are chosen according to the following rules. 
Trials of the first 
≥ 1 iterations are performed in arbitrary 
 
 
K = k(n) = p(l) + p(2) + . . . + p(n
 
points of the interval [ab]where p(i), i 
≥ 1, denotes the number of trials of 
the  i-th iteration. Trial points corresponding to any next Q-th iteration, 


Достарыңызбен бөлісу:
1   ...   130   131   132   133   134   135   136   137   ...   151




©emirsaba.org 2024
әкімшілігінің қараңыз

    Басты бет