Prof. Swarat Chaudhuri COMP 482: Design and Analysis of Algorithms Spring 2013 Lecture 2.
482 Lecture - faculty.math.illinois.edu
Transcript of 482 Lecture - faculty.math.illinois.edu
Math 482Lecture 7
Announcements
Awk 2 due Friday
Evening officehours this week thµ8pm_
on class zoom link not OH link
Rearrest supplementary problems if taking
the 4 unit version
ReadingsLect 7 GM 5.5 5,6 5 so
PS 2.5 2.62 I 2.2
handoutBT chap 3 I
Lect 8 DegeneraciesGm 5,7 5,8PS 2,3 2.8chap 3
BJ 24 4
today An unbounded examplematrix notation
Anunbound edexampleMax 2 X 2 Xz
i
as
In standard form
Max 2 X 2 2
Subj X Xz 1 3 I
I Is Elise
Xi 42,43 4 30
Tableau Xi Xz Xs XL solve
I I is
I I 0 O lE not auite solved
Does this work
Ra
EE'd to keep entries in this firstpushescolumn non negative for feasibility notwork
Instead try putting in basis
EERIEI II 0 O I
pivot here
i H 7Bi 17isfENext
try to pivot againMax LP pivot column must be XyBut no restriction on how big X4 grows
Claim this means LP is unbounded
Let Xz o whereDictionary
4 M M isZ X 2 1 4 1 as large
qX XIII the Miz asHE
X M 11
ButI 2 Mtl
MatrixNotationKEIRNconsider the LP
q I o aTyberm
Let Aj be the jth column of A
Given basis 93 let BCD Bc2 Bang
to denote basic indices
Basic variables XBas XBc XBCm
XB X Bcn XBes n XBcm
CB CBcis CBas CBCm
AB mxm sub matrix of A
BCD Bcm
Claim If X is a bfs
AB is invertible by def of bfs
CTX CTzXB since xjjop.frp XB AI't
2 x E'AI b follows fromprevious
2Cx CTXI'BXB
P fpEBAfb
Doing Row operations on A
to turn The columns ABCD ABCm
to the identity matrix Im
is conivalent to multiplying on left
by Ats
EEE KITENE ENAE AB
Row operations
must on left by Ats
Since Apj'AXB Inxs XB
qB
AI Axa'Eb
a
MainpiotingstepSuppose your tableau has form
a fit
pivoting can be thought of as left multiplication
by seauence of elementarymatrices
T Xk XzX T
What is form for X
T dTI dX
Uo U8 Cmas xConti
x if.ECEi aEaI
Suppose XT s solved for basis B
Then UA Im
so U AIWe also know that Iida D
So CBT dTAB
d cigars
xt
ffn.IN bifaaaJTableau solved for basis B
Xss Ats'bFirst column tells us values ofbasic variables
relatist'sIT CT CBTAB'A u reducedcosts
Top
DI Ej.is osEfXj
Cg is cost per unit increase in
variable Xj
CBTABTAj is cost of compensating changein basic variables necessitated
by constraint Ax b
Them If X is a bfs for a
minimization LP then
if c 20 then x is optimal
maximization LP ESO n