You are on page 1of 12

Chapter 10

Integer programming I .
Itp
Question :
How to solve ILP ?

Assumption : all
parameters are
integers .

ILP bounded
Simple observation :
if feasible region of
is . then there

are
finitely many feasible sols

Example : Min Z = -
2x , -

Na

LO - l . I
S.t. Nit Nz E5

-
K , t Nz E O

GK , + 2Nd E 21

Hi , Nz 70 , int .

Method A :
Enumeration Method .

Simple binary decision variable 1 ja


case : :
Kj = o or n
- -

. .
i. .

Sol T
number
of feasible E
-
exponentially .

Not realistic to solve ILP by enumeration method for .

large n .

By observation , I LP is
closely related with its LPR .

ILP LPR
- -

Max EI Max EI
( P)
St .
AI =3 45 ) St .
AE -
- I

size , it .
I 38 .
FRC ILP > E FR ( LPR )

Method B :
Rounding Method .

( is For the
capital budgeting problem ,
if Kitt is a fraction ,
round

down the number makes no sense


up or .

Even Sol
a-is if it makes sense , it
may not give you the correct
opt .

Example : Min Z = -
2K , -

Us

5.t .
Kit Nz E5

-
K , t Nz E O

Gse , + 2Nd E 21

Hi , Nz 70 , int .

generally speaking .
Solving ILP is harder than solving LP .

However ,
there are a group of Ihop which is
easy to solve .

i.e When A
. is
totaleyun.no#ar .
Let's make some observation

ILP LPR
- -

Max Zia ) -
-
EI Max ICE ) = E' si
( P'
St .
AZ =3 45 )
St .
AE -
- I

size , it .
I 38 .

Then ,

① FRCP ) E FR CF )

② if Zi't is an
opt Sol to ( Dd then
, .

HE E FRCP ) ,
EE EE * C E .
> ,
=) ?
-

E' a- LEE'*J ( since I is int )


-

z*iz , LEE 'T


-

E'Je't provides an bound CPI


upper on

x'*
'

If further .
It is int valued ,
then .
is to CP ) .

12¥ .
for Min
problem ,
LEI 'T E Z
't
CE ) .

③ If I > is infeasible .
then CP ) is
-

Solving LPR can give us some


information on the opt Sol of
ILP ,
like its bound ,
and its
opt Sol .
Question which
type of ILP has its LPR with int opt sols ?

Question : which
type of
LP has ALL its BFS to be tht valued ?

Jeb =
B-
'

I BIB I -
-

By Cramer 's rule : detCB.es


NBK =

det CB )

where Bk is formed by replacing Kth Col of B


by J .

By assumption .
Bk has art elements .

if then .
ht K . Urse is it valued .

IB is int valued .

I 0 .
2 .
I

Definition A square . int matrix B is un


-

. modular ,

if det CB) = -4 .

An int matrix A is
totally unimodalar LTU M ) .

if every
square non
singular sub matrix
of A is unimodal ar .

ipsilateral
Thmpj-gtwttffg.li?nft.matrixA.n
Renata .

Furthermore IF ) BFS
opt Sol to Cp )
,
if has an
opt ,
it's an .
%ffeise.IT?j+scifany,ots--IaeirnlAEeb.a
Think ht int matrix A, the
following statements are
equivalent
:

are int for any int I .

③ every basis matrix has an integral inverse .

PI .
(
skipped .
see lecture notes ) .

Remmark.tt
LP with a TUM coelficieut matrix A yields an int opt Sol for

and int valued 5


any cost
coefficient vector
any .

An lip with a Tum coefficient matrix A can be solved by


solving its LPR
by simplex method .

Question Given a matrix A ,


how do we know it is TOM ?

i: c: :
l÷÷÷÷ ,
l: : : :
O l O O O
.
LemmaAisToM⇐[A1]isToM
"

Lemmata A is TOM .
⇐ A is TUM .

:÷÷÷÷n:
Thm A matrix A is TOM
if
it
aij E lo .
-

i . i Y ti
j

: ÷÷:÷{÷÷÷÷÷÷÷÷÷÷÷÷÷÷ Col J contains


lbs
if a 2 nonzero elements
of opposite signs .

both elements are , n the same subset .

⇐ t Col
j containing exactly 2 nonzero entries ,

i }m aij ,
-
I
itMz
aij = o .

PI .
( By M1 on the order
of submatrices
of A ) .

Rematk
It's just sufficient condition condition
a ,
not a
necessary .

Eamon
ft I :O ! )
'

a -

.
N .

A directed graph (
digraphs G -

-
CV , E) .

T t
vertex set Arc set .

✓ =
f i Y i :
vertex . 2
special vertices : S . t .

E E
Icij > I i EV ,
j EV T .

④ →
⑤ :
permmis.be flow transportation direction .

P
t
cij ) EE
Gj 30
capacity )
{ tcijs
source
.

# E .

Cj -0

A S -
t flow in G : f : Vxv →
IR
"
sink .

satisfying the following two properties :

( is
capacity constraints :

for all i
j EV ,
O E
fcij , E
Cij or OE
fij E
Cij .

I flow conservation :

ti E Vil s ,tY ,

Total flow shipped into i = total flow shipped out


of i .

I
ztji =

cij > EE
fij
.

I
Value
of the flow : V =
fsj = I
fjt
(
Sj > EE Lj 't) EE .

The Max flow problem is to find a s - t flow with Max value

of v .
Max V

St fs , fsz

}
. + -
v = O

+"t + t "
-
f" -
t "
=
°
flow Aberration

f-21 t fzt -
fsz -
fi z = O

V -

fit -
fzt = O

fs , Csi

)
O E E

O E fsz Csa flow constraint


E
capacity .

;
O E fit E Cit

O E fzt E Cut .

Write the flow conservation constraints in matrix form :

( S , t ) C S ,
2 ) ( I .
2 ) (2 ,
l S C 1. t ) ( 2. t)
f-s

:X:X :L
,

it:÷÷÷÷
S l l O O O O fsz V

f- it
t
fat V
-

b
incidence matrix A ( ) i =L IV 1 I El
aij j
-

= i
-
- v
. . .
. . -
, . . . . .

\
Aij = 1 .
if Ej
=
Cio . K )

1
if ej ck.io )
- =

O otherwise .

conduit.cidencematnixofadigraphis.TV#f
general formula for Max flow problem :

Max V

(
V S
if i -

=/ ;
-

ji,¥Ii
s t -

.
-

it
It
"

(p,
.

OE
fij E
Gaj
t ii.
j) EE .

Reinark Coefficient matrix


of Kp ) is TUM .

( i. V
j,
- - - . . -
-
. - r - -

S I I 0

f .io/. . s=f
- .

i t o
'
'
: .
.

: .

i
i 1 : :
:
i
f :
:
'
*
. .
.
. . .

'

.
i O '

i
I :
i
-

1- Et
.

Cij
'

cij SEE :
I
i
! -

i 1 :
l
O
' '

l l '

if Lij are int valued ,


the opt Sol of Max flow problem are int valued .
M.hcostcapacitatedpioblem.tt
digraph G = ( V . E ) .

flow

\
for each arc
cij > EE Nij :

dij :
capacity of Li -

j ,

Cij : unit
transportation cost
of ii.
js

\
: V, : set
of sources or origins
Vz : set
of intermediate vertices

Us : set
of sinks or destinations .

b-

{
i EV , ,
ai :
supply of the
commodity .

t i C- Us ,
bi : demand at vertex i .

Conservation constraint : there is no loss of the flow at intermediate vertex .

Min cost capacited problem : find a flow satisfies the conservation


constraint and the and demand requirement the min
#
meet
-
supply at
I

cost

I
Min
Cijseij
( i
jst 't
Ai if

1=0
E i EV ,

sit

ji
.

i
it '
-

c. .

F bi
E , .
-

O E
Nij E
dij .
ttcij > EE
some
particular cases :

A .

transportation problem .

✓ 2=01 J
to 7

#O
|
.

Kiev , lcj.is/cj.iseEY=o/ O

0*0
,
.

titus ,
Icij >
Icij > EE 's =p .

Source
O
Min
cij > EE
Icij Nij Cais sink .

( bi )

St I
a. ij E ai ti EUI
.

Ci SEE

[ Uji bi
> Vi EVs
cj.is EE
.

OE
Nij E
dij ttcijs EE .

B .

transshipment problem :

transportation problem with


dij-tbV-ci.gs EE .

C .

assignment problem :

In the
transshipment problem , if
its Vi EV A -
1 IV , 1=1 Vz I
{
. .
.
= . as

ti EVs ,
bi -
-

Assignment problem : i :
job j : worker

Min
Icij Nij
cij SEE
St I I
Nij =
fit Vi
.

Lij > EE

[ Uji =L . Hit Us
Cj EE
.

Nij EE
iht ⇐
0EKij El .
.
=
o or 1 .

V-ci.jo .

w e

You might also like