Scribd
Upload
58 views6 pages

Sequential Quadratic Programming

Sequential quadratic programming (SQP) is an algorithm for solving constrained optimization problems. It works by iteratively solving quadratic programming (QP) subproblems that approximate the original problem. Specifically, it uses Newton's method on the Karush-Kuhn-Tucker (KKT) conditions to find search directions, then solves a QP to ensure feasibility. Under certain assumptions like linear independence and positive definiteness, each SQP iteration has a unique solution converging quadratically to a KKT point. The algorithm can be applied to problems with equality and inequality constraints by appropriately defining the QP subproblems. Global convergence can be improved by including techniques like line search, trust regions, and filters.

Uploaded by

Alneo Lesag
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
58 views6 pages

Sequential Quadratic Programming

Sequential quadratic programming (SQP) is an algorithm for solving constrained optimization problems. It works by iteratively solving quadratic programming (QP) subproblems that approximate the original problem. Specifically, it uses Newton's method on the Karush-Kuhn-Tucker (KKT) conditions to find search directions, then solves a QP to ensure feasibility. Under certain assumptions like linear independence and positive definiteness, each SQP iteration has a unique solution converging quadratically to a KKT point. The algorithm can be applied to problems with equality and inequality constraints by appropriately defining the QP subproblems. Global convergence can be improved by including techniques like line search, trust regions, and filters.

Uploaded by

Alneo Lesag
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

Sequential quadratic programming

Michal Kocvara
Institute of Information Theory and Automation
Academy of Sciences of the Czech Republic
and
Czech Technical University
[email protected]
http://www.utia.cas.cz/kocvara

Sequential quadratic programming p.1/6

Sequential quadratic programming


Consider equality constrained mathematical program:
min f (x)

(OPE )

s.t. gi (x) = 0,

iE

with Lagrangian
L(x, ) := f (x) +

T
i gi (x)

iE

Denote AT (x) = (g1 (x) . . . gm (x))


KKT: F (x, ) :=


f (x) A (x)
=0
g(x)

Nonlinear equation - solve it by Newtons method

Sequential quadratic programming p.2/6

Sequential quadratic programming


KKT: F (x, ) :=


f (x) A (x)
=0
g(x)

Nonlinear equation - solve it by Newtons method:




  
T
T
Wk Ak
f (x) + A (x)
dk
=
g(x)
Ak
0
d
Wk = 2xx L(x, )
xk+1 = xk + dk ,

k+1 = k + d

Well-defined when the KKT matrix nonsingular


Lemma:

Let Ak has full row rank ( LICQ) and


dT Wk d > 0 d 6= 0, Ak d = 0

Then the KKT matrix is nonsingular.

Sequential quadratic programming p.3/6

Sequential quadratic programming


Define a quadratic program
min

dT Wk d + fkT d

2
s.t. Ak d + gk = 0

(QPE )

Assumptions of L: ! solution of (QPE ) and it solves the KKT system


Algorithm (local SQP for (OPE ))
Choose (x0 , 0 )
for k = 0, 1, 2, . . .
solve (QPE ) to get dk , k
xk+1 = xk + dk , k+1 = k
test convergence

Sequential quadratic programming p.4/6

Sequential quadratic programming


For a general optimization problem
min f (x)

(CO)

s.t.
gi (x) = 0,

iE

gi (x) 0,

iI

define a quadratic program


min

1
2

dT Wk d + fkT d

(QP)

s.t.
giT (xk )d + gi (xk ) = 0,

iE

giT (xk )d + gi (xk ) 0,

iI

and use the same SQP algorithm


Sequential quadratic programming p.5/6

Sequential quadratic programming


Theory: Under assumptions of Lemma: local quadratic convergence
(Newtons method)
Global convergence?
Include
line-search (merit function)
trust region
filter

Sequential quadratic programming p.6/6

You might also like