Professional Documents
Culture Documents
LECTURE NOTES
Acknowledgements:
I would like to acknowledge Prof. W.L. Winston's "Operations Research: Applications and Algorithms“
(slides submitted by Brooks/Cole, a division of Thomson Learning, Inc.) and Prof. J.E. Beasley's
lecture notes which greatly influence these notes...
I retain responsibility for all errors and would love to hear from readers...
www.isl.itu.edu.tr/ya
Interior Point Methods (IPM) still follow the improving search paradigm for LP, but they
employ moves quite different from those in simplex method.
Much more effort turns out to be required per move (iteration) with IPM but the number
of moves decreases dramatically.
The simplex algorithm is an exponential time algorithm for solving LPs.
If an LP of size n is solved by the simplex, then there exists a positive number a
such that for any n, the simplex algorithm will find the optimal solution in a time of
at most a2n.
IPM, on the other hand, is a polynomial time algorithm.
This implies that if an LP of size n is solved by an IPM, then there exist positive
numbers b and c such that for any n, LP can be solved in a time of at most bnc
Instead of staying on the boundary of the feasible region and passing from extreme point
to extreme point, IPM proceed directly across the interior.
IPM begin at and move through a sequence of interior feasible solutions, converging to
the boundary of the feasible region only at an optimal solution.
Popular methods
• Karmarkar’s Projective Transformation
• Affine Scaling
• Log Barrier
such that x = D~ x
~ −1
Rescaled variable: x = D x [Centering scheme (1, 1, 1)]
For new coordinates
~
A = AD
~c = Dc
~ ~~ ~
Projection matrix:P = I − A T (AA T ) −1 A
~
Projected gradient: c = ± P c
P