第37卷 第1期 上海师范大学学报(自然科学版) Vol.37,No.1
2 0 0 8 年 2 月 JournalofShanghaiNormalUniversity(NaturalSciences) 2 0 0 8 , F eb .
A lin e search filter tw o p iece u p d ate of red u ced
H essian m ethod for nonlinear equality constrained
optim ization: global convergence
ZHUANG Dong-jiang
1
, Z H U D e -ton g
2
(1. C ollege of M athem atics and Sciences, Shanghai N orm al U niversity, Shanghai200234, C hina;
2. C ollege of B usiness, Shanghai N orm al U niversity, Shanghai200234, C hina)
A bstract:
A line search filter two piece update of reduced H essian m ethod is proposed for nonlinear constrained
p rogram m ing by using F letcher and Leyffer′s filter m ethod, which replaces the traditionalm eritfunction. W e develop
atwo-piece upd ated p rojected H essian algorith m w ith lin e s e a rc h filte r to so lv e n o n lin e a r e q u a lity c o n s tra in e d o p tim i-
zation. U nder m ild assum ptions it is shown thatevery lim itpointofthe sequence of iterates generated by the algorithm
is fe asib le , an d th a t th e re e x ists a t le ast on e lim it p o in t th a t is a statio n ary point for the problem under consideration.
The num erical experim ents indicate that the algorithm is eff.ective.
K ey w ord s:
constrained optim ization; filter m ethod; two-piece update; line search
CLC number:
O 221.2
D ocum ent code:
A
ArticleID:
1000-5137(2008)01-0001-09
R eceived date: 2007-10-22
Foundation item : The authors gratefully acknowledge the partial supports of the N ational Science Foundation Grant
( 10471094) of C hin a; th e P h . D . F ou n d a tion G ra n t ( 0 5 2 7 0 0 3 ) o f C h in e se E d u ca tion M in istry ; the Shanghai Leading A cadem ic
D iscipline Project ( T0401) ; Shanghai E ducational D evelopm ent Foundation and the Science Foundation G rant ( 05D Z11 ) of
Shanghai E ducation Com m ittee.
Biography: ZH UAN G Dong-jiang(1980 - ), m ale, graduate student, College ofM athem atics and Sciences, ShanghaiN or-
m a l U n iv ersity ; Z H U D e -tong(1954 - ), m ale, professor, C ollege of B usiness, Shanghai N orm al U niverstiy.
1 In tro d u c tion
R ecently, F letcher and Leyffer proposed a filte r m e th o d , re p la c in g th e tra d itio n a l m e rit function m ethod,
a s a to o l to g u a ra n te e g lo b a l c o n v e rg e n c e in a lg o rith m s fo r nonlinear constrained program m ing( N LP) . The un-
derlying concept of the filter m ethod is that trial points are accepted if th e y im p ro v e th e o b je c tiv e fu n c tio n o r
im prove the constraint violation instead of a com bination of those tw o m easures defined by a m erit fu nction .
The m otivation given by Fletcher and Leyffer in [2] for the developm ent of th e filte r m e th o d is to a v o id th e n e -
c e s sity o f d e te rm in in g a s u ita b le v a lu e o f th e p e n a lty p a ra m e te r in th e m e rit function. In additio n , th e filte r a p -
proach offers another im portant advantage regarding robustness. If the trial step size becom es too sm all to guar-
antee sufficientprogress tow ard a solution of the problem , the proposed filte r m e th o d re v e rts to a fe a sib ility re s -
toration phase,w hose goal is to deliver a new acceptable iterate by decreasing the constrain t violation , or to
converge to a local m inim izer of infeasibility if this is not possible.