Download Full Issue in PDF - Academy Publisher
Download Full Issue in PDF - Academy Publisher
Download Full Issue in PDF - Academy Publisher
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
1496 JOURNAL OF COMPUTERS, VOL. 8, NO. 6, JUNE 2013<br />
Improved Feasible SQP Algorithm for Nonl<strong>in</strong>ear<br />
Programs with Equality Constra<strong>in</strong>ed Sub-<br />
Problems<br />
Zhijun Luo 1 , Guohua Chen 3 and Simei Luo 4<br />
Department of Mathematics & Applied Mathematics, Hunan University of Humanities, Science and Technology, Loudi,<br />
Ch<strong>in</strong>a<br />
Email: ldlzj123@163.com<br />
Zhib<strong>in</strong> Zhu 2<br />
School of Mathematics and Comput<strong>in</strong>g Science, Guil<strong>in</strong> University of Electronic Technology, Guil<strong>in</strong>, Ch<strong>in</strong>a<br />
Email: zhu_zhib<strong>in</strong>@163.com<br />
Abstract—This paper proposed an improved feasible<br />
sequential quadratic programm<strong>in</strong>g (FSQP) method for<br />
nonl<strong>in</strong>ear programs. As compared with the exist<strong>in</strong>g SQP<br />
methods which required solv<strong>in</strong>g the QP sub-problem with<br />
<strong>in</strong>equality constra<strong>in</strong>ts <strong>in</strong> s<strong>in</strong>gle iteration, <strong>in</strong> order to obta<strong>in</strong><br />
the feasible direction, the method of this paper is only<br />
necessary to solve an equality constra<strong>in</strong>ed quadratic<br />
programm<strong>in</strong>g sub-problems. Comb<strong>in</strong>ed the generalized<br />
projection technique, a height-order correction direction is<br />
yielded by explicit formulas, which can avoids Maratos<br />
effect. Furthermore, under some mild assumptions, the<br />
algorithm is globally convergent and its rate of convergence<br />
is one-step superl<strong>in</strong>early. Numerical results reported show<br />
that the algorithm <strong>in</strong> this paper is effective.<br />
Index Terms—Nonl<strong>in</strong>ear programs, FSQP method, Equality<br />
constra<strong>in</strong>ed quadratic programm<strong>in</strong>g, Global convergence,<br />
Superl<strong>in</strong>ear convergence rate<br />
I. INTRODUCTION<br />
Consider the follow<strong>in</strong>g nonl<strong>in</strong>ear programs<br />
m<strong>in</strong> f( x)<br />
s.. t g ( x) ≤0, j∈ I = {1,2, , m},<br />
j<br />
Where f ( x), g ( ): n<br />
j<br />
x R → R( j∈I)<br />
are cont<strong>in</strong>uously<br />
differentiable functions. Denote the feasible set for (1) by<br />
n<br />
X = { x∈R | g<br />
j<br />
( x) ≤0, j∈ I}<br />
.<br />
The Lagrangian function associated with (1) is def<strong>in</strong>ed<br />
as follows:<br />
Lx ( , λ) f( x) λ g( x)<br />
m<br />
= +∑<br />
j=<br />
1<br />
A po<strong>in</strong>t x ∈ X is said to be a KKT po<strong>in</strong>t of (1), if it is<br />
satisfies the equalities<br />
This work was supported <strong>in</strong> part by the National Natural Science<br />
Foundation (11061011) of Ch<strong>in</strong>a, and the Educational Reform Research<br />
Fund of Hunan University of Humanities, Science and Technology<br />
(NO.RKJGY1030), correspond<strong>in</strong>g author, E-mail: ldlzj123@163.com .<br />
j<br />
j<br />
(1)<br />
m<br />
∇ f( x) + λ ∇ g ( x) = 0,<br />
j=<br />
1<br />
λ g ( x) = 0, j∈I,<br />
j<br />
j<br />
∑<br />
where λ = ( λ1<br />
, , λ ) T<br />
m<br />
is nonnegative, and λ is said to<br />
be the correspond<strong>in</strong>g KKT multiplier vector.<br />
Method of Sequential Quadratic Programm<strong>in</strong>g (SQP)<br />
is an important method for solv<strong>in</strong>g nonl<strong>in</strong>early<br />
constra<strong>in</strong>ed optimization [1, 2, 18]. It generates<br />
iteratively the ma<strong>in</strong> search direction d 0<br />
by solv<strong>in</strong>g the<br />
follow<strong>in</strong>g quadratic programm<strong>in</strong>g (QP) sub-problem:<br />
j<br />
T 1 T<br />
m<strong>in</strong> ∇ f( x)<br />
d + d Hd<br />
2<br />
T<br />
s.. t g ( x) + ∇g ( x) d ≤0, j∈I,<br />
j<br />
n n<br />
where H ∈ R × is a symmetric positive def<strong>in</strong>ite matrix.<br />
However, such type SQP algorithms have two serious<br />
shortcom<strong>in</strong>gs:<br />
1) SQP algorithms require that the relate QP subproblems<br />
(2) must be consistency;<br />
2) There exists Matatos effect.<br />
Many efforts have been made to overcome the<br />
shortcom<strong>in</strong>gs through modify<strong>in</strong>g the quadratic subproblem<br />
(2) and the direction d [4, 5, 7, 8]. Some<br />
algorithms solve the problem (1) by us<strong>in</strong>g the idea of<br />
filter method or trust-region [13, 16, 17].<br />
For the problem (2), it is also a hot topic to solve the<br />
QP problem like (2) <strong>in</strong> the field of optimization. By us<strong>in</strong>g<br />
the idea of active constra<strong>in</strong>ts set, some algorithms solve<br />
step by step a series of correspond<strong>in</strong>g QP problems with<br />
only equality constra<strong>in</strong>ts to obta<strong>in</strong> the optimum solution<br />
to the QP sub-problem (2). P. Spellucci [6] proposed a<br />
new method, the d 0<br />
is obta<strong>in</strong>ed by solv<strong>in</strong>g QP subproblem<br />
with only equality constra<strong>in</strong>ts:<br />
j<br />
j<br />
(2)<br />
© 2013 ACADEMY PUBLISHER<br />
doi:10.4304/jcp.8.6.1496-1503