You are on page 1of 23

Chapter 8 .

Duality I .
Completneutatyslacknesscondition

Def
-

ups Max EE LD ) Min Jiu

St .
AI EI S t
- .

Atu I

I 75 I 38 .

Convert them into standard form .

Cps Max EE LD ) Min Jiu


- o - -

St .
AE + as = b S t
- .

Atu -
Us = c

- J j s -

K , Ns Zo U , Us 38 .

I I
{ { Tls
(
Kj ) 5=1,2 ( Ui ) is
=
h 1
- -
'
= m
- - -

. .
, . , .

Is = ( Ksi > i =
1.2 .
- -

-
m .
= (
Usj )
j=i .
" '

in .

the slackness condition be started as


complementary can :

For each j
-

- I .
- -

in .

Kj Usj
-

=0 ⇐ E' Tis -

-
o ①

For each i =L ,
- . .

. m . hi -

Ksi =o ⇐ Isth =o .
@

Sine E Jes I Is have


, , , are
nonnegative , we can also .

EuJt2Tu=oCGmp6meutgslacknessonditb#
Rematki
If a constraint is nonbinding , then the associated dual variable is O .
⇐a:÷÷.¥⇐a÷÷.
.
Theme . ( CSC thin )

consider a canonical primal -


dual pair in standard form .

Cps Max EE LD ) Min Jiu

let ( I÷ ) d l thus )
. be feasible sols t
Lps d l D)
resp .

( Igo ) ( ÷)
Csc holds
opt sols
d are ⇐

( hT£sot bison .
= o )

PI .
Since ( Eso ) d l Ens )
. are feasible sols .

A Iot Iso = I UTA E + UT Iso = 5 = Jiro ①

Atu -
Iso = I EA -
Tsi = ET
JE AI .
-
Tin .
-
- EE .

① -
② : UT Iso t UTI Io = Eino -

Esto .

UTRs tusono .
=
o ⇐ E'I .
-
E' Io =o

⇐ Jiro = E' To
⇐ ( EI ) a ( EI ) are
opt sees oops .

(
strong duality ) CDs
& resp .

#
E.G.8.tl
(
p> Max x, -14kt 323 CD > Min 4h , -16ha

St .
22 , -122kt 23 S4 S t -
.
24 ,
t Uz 3 I

N, + 222-1223 E 6 24 , +26234

Ki K2 23 30 U, t 2 Us 33
'

. .
.

hi .
U2 3D .

CA > verify optimality .

Initial tableau :

)
K , Nz 23 Ky Ks ths

& Z Z " " ° 4

25 I 2 2 O l G

I -4 -
3 O O O

tableau
opt :

goin
N , Nz Ks ka 25 ths
-

3
Nz 12 I O l
-

Yz t

a,

2 O O l l 10

opt Sol
to ( (N , Us 24 ( O O O )
p) Na 25 ) = I 2
:

{ opt Sol
.
. , .
.
. , ,

to LD ) : CU . .
Uz uz Ua Us ) =L I ,
I, 2 ,
O ,
O )
- -

( is check feasibility .

di) check Csc :

IITs + ZITI =
0.2+1-0+2 Ot O 1+0-1=0 - -

they are indeed


opt sols to
Cps d CD ) resp .
(B) Construct an
opt
Sol
of Cps from an
apt sod of CD) .

Ci > solve CDs graphically .

a
CD > Mir z '=4U -16ha ,

| S.t.

24
Zu

,
,
-1

+24234
Us 3 1

.. ÷ : :c::
U,
-1242=3
Uit UE 2

ZU , -142=1

hit U2 2 U, I

{ /
= =

Hi -1242 =3 U2 =
I

Also , UG =
Us =D . U3 = 2h , + Uz -

I = 2 .

opt Sol to CD ) is

( Ui .
U2 , Uss , Ua .
45 ) = ( l , I ,
2, O ,
O ) with Z'* =4U ,
+ buz = 10 .

let's use Csc to find the opt Sol of Cp ) .

( Max
p) 2- = x, t 4kt 323

S.t. 22 , -122kt 23 E 4

N, +222 t 22cg E 6

Ki .
K2 . 23 30 .

Cg
u, = , Ns , =
24=0 2K , -122Gt 343=4 22kt 23=4

42=1 262=25 =O N, -1 222 +2kg = 6 .


I 22h -1223=6

43=2 K, = O Nz = I 2cg = 2

opt Sol to Ip) ( Ni Ks ) ( O )


: Nz Us 24 = O, I, 2 O .
, . .
,
,

with Z* = K, -14221-323 = 4+6=10 .


pkx
l .

In the c
primal ) simplex method ,
we start with a BFS ,
maintain As

feasibility and work towards


optimality .

Simplex tableau ut't basis B - CLP is feasible canonical form ) .

→ -

K Ks rhs

:÷.a ÷÷t::
feasibility B- to 38
'

primal : .

£513 E 38
"

{
primal optimality
-
' -

IT
N
Ezio -
i
>E

CB-5 Es
{ Atu 3,8g
T
'
dual feasibility .
I let -
-
d .

for Basic Sol I satisfies


some ,

if it

primal feasibility +
primal optimality I is opt

dual
feasibility .

In
primal simplex method ,
we maintain primal feasibility ,
want to

get primal optimality ,


i.e .
dual feasibility .

Question Is it that start computation with basic Sol


:
possible we a

that is dual feasible but not pinnae feasible .


and work towards

a basic sod that is both


primal and dual feasible ?

Dual simplex method .


In the dual method the conducted
simplex ,
computation is still on

the primal simplex tableau with


different ways to
pick entering
and departing variables .

E g. 8.2
-

.
I

Min 3K , t 422-1523

Sit . N, t 222-1323 75

2K , t 2ns t 23 36

Ki ,
K2 ,
Ks 3D

Method Solve it by Big M / Two


phase method
-
-
. .

Method Dual simplex method .

Convert it into canonical form .

Max -
3N , -
422 -

5K3 Max -
3N , -
422 -

5K3

S.t. -

K
,
-

zu
-

323 E -
5 S.t. -

K
,
-

zu
-

323+24=-5
-

2K , -
2K -

Ks E -

b -

2x , -
2K -

Nz toes -6

Ni . R2 , Nz 30 .
Ni . R2 . N3 ,
Ka . . 25 30

Step .
get an initial Basic Sol .

z-z.io#
Hi Nz Ks Uk Us rhs
F

,
o
primal infeasible .

Ni -
I -2 -

3 I 0 -5

a,

3 4 5 O O 0
A

Dual feasible

If B-
'
I 38 opt otherwise go to 2
step
.
: .
.
choose variable for
Step departing NB .r with - O .

a.z.io#
Hi Nz Ks 24 Hs rhs

in g
¥i÷¥
-

3 -5 -
I -2 -

I 0
.

*
j boo

3 4 5 O O O

St .
Choose
entering variable Ks to maintain dual
feasibility .

Nos .r
Assume is the
departing variable .

The pivoting row is


aport I
yrjzej =
por . (
pro )
N B. V
.

Ci , if all
yrj >o .

NBR E l
yrj ) Uj t
for Epr < 0
= -

.
.

N B. V
.

Cps is
infeasible
Question How about (D > ?

Cii )
if I some N B. V
-
.
Ks with yrs so .

NBR I
t yrs Ks +
No B. U yrj Kj
=
for .

j #s

as -
-

Fgs -

TTT -

Iggy Tyr
'
, Kj ①
Sub ① into the fun
obj :

Z t
§¥, rj Nj t rs Ks yoo
=

2- +

÷¥ , rjsg
-
t rs ( Ifrs T.IT- -

Eg
yrs
) =
yoo

Zt

y;¥, rtyj) Nj -

Msr =
Yoo -

FIS Br .

T T
'

tbr
r
'

to maintain dual feasibility ,

'r
%
grab
=
so yrs so
-

Frs
'

rj rj yrj 70
= -
-

Thit
yirj
|
so .
No .

effect !

if
y7÷
yr
Tris
so
; .

treys =
max
f T÷j I yr
;
co )
ie .

¥s , -
-
min
17¥ , I yrjao ) .

( Nih ratio test )

""i÷::i÷÷÷:
b

z*,z,,,
TI N , Nz Ks Ma Hs rhs
-

NG -
I -2 -

3 I 0 -5

← a,

3 4 5 O O O

ratio :
3/2 42 5h

Step .
do
pivot operation at
yrs .
go to
step 1 .

11420¥
T2 N Us rhs
-
, Nz Nz 24
-

N4 Itt -42
O -
-

5h I -2

Ki

0 I 72 0 % -9

ratio I
75 3

,o#,
-13
-
N , Nz Nz Sea Hs rhs
-

Kz O l 5/2 -
I Yz 2

z ,

O O l l l -
11

Jett 05
opt Sol = ( I 2 O O with Z*= C 113=11
-
-
, , , , .

Also , dual opt Sol : VT 't =L I , I ,


O .
O ,
I 5
Question when should we use dual
simplex method ?

① like the above


example ,
in which the dual feasibility has already
been
satisfied .

② in sensitivity
-
analysis .
or
post optimality analysis
-

Suppose we have got an


opt
Sol to a LP and
. we want to solve the LP

again for a new set


of this .

dual feasibility £5 B- A E'


35
{ 85
\
: -

No
change .

i
B
-

primal feasibility Io changes


'
:
B- .

Instead
of solving the new LP from scratch ,
we can
apply the

dual simplex method ,


starting from the
opt basis
of the original
problem .
Seusiivityanalysis

study the
impact of changing the
parameter of a LP on

the
opt sod of it .

Max 2- = I a BFS wit B .

Ate =3

St .
se rhs

:÷a*¥¥is
"is .

criteria for determining the opt Sol :

primal feasibility B- I 38
'

{
:

"

F' Is E ST
'

dual feasibility B 'A >


-

: = -

ET E
'

( primal optimality ) = -

If the modification affects :

the primal feasibility use dual simplex method


|
.

the dual feasibility use


primal simplex method .

se rhs
a

cl)§→b_
"
.mg?.on::agg..E.e7IIa.elEis
I'
E .

But no
impact on dual feasibility .

It -

- E- E' = Tis B- A E '


-

if Is B-
'
I feasible dual
is not apply the
simplex
=
,

method to derive a new opt sols .

EG .
8. 3.1

.
A Max LP has the following initial tableau .

ths

,¢,,o#
N , K2 23 NG 25

24 I 3 4 I 0 30

a,

-2 -7 3 O O 0

,¢,,#
Opt tableau :
KiNzK3K¢N5
24 O -
I 5 I
-

l ZO

se ,

O l l O 2 20
b'
'
= ( Yo )

( 1) ( Yo ) ( II )
' -

Irs I infeasible
-_

new Basic Sol :


-
-
B- '
=
o

new z = EBT EB =
Cca .
C, > EB =
co ,
2 ) ( II ) -
40

update tableau :

,q,o#
Ki Na 23 N4 25 this
~

*
Ki O -
I 5 I -
I TO

a,

O l l O 2 40

ratio I 2

,o,g¢z*|#
TI Ki22Ks2C42st#
Nz O l -5 -
l l lo

se ,

O O 6 I I 30

ratio 43

-Y3o-i%-%1|20#
-3L Ny22K3K4N5r#
Zz 43 I 4/3 43 O
'
%

25

43 0 373 73 O 7%

New opt Sol :


Jit -

- ( O,
'
% ,
o.O ,
2%5 with z*=7% .

x rhs

czschanges.ntcck-s.ci#
IEI7a-E#7I
Tis B- A E
'

have impact primal optimality


'

{
on : -

value Eri B- I '

obj
:

tis
if of
aN.B.v.kk.FI
Ck is the cost
coefficient

- Ea -
EI = EIB ai -
' -
at
=( IF B- ai '
-
Ck ) + C Ck -
Ek )

= 8kt CCK -
Ee ) ( let Ock = Ee -
Ga )

=
Vk -
OCK

For other
Tj , Fj =IsT3- a'j '
-

Cj =

zj
-

Cj =

rj .

II. initial tableau :


opt tableau :

,¢,,o# ,¢,,#
K , K2 23 NG 25 ths k, Nz Nz 24 25 this

-4 224 O 4
5 20
I 3 I 0 30 -
I I
-

a, y ,

-2 -7 3 O O O O l l O 2 20

Ez = 10 .

I = Zz - E = ( Zz -
Cs > + Ccs -
Ez )

= V2 -
D Cz

= I -
C lo - 7 > = I -
3 = -2 CO .

So , we lose the dual feasibility .


,¢*yo÷
TI 2,26232425in
Kc, O -

I 5 I -
I 20

a,

O -2 I O Z 20

-2L K2 23 Ks Vhs

441-44044/5/242
N, Ny
-

No, 74 O
'
% ,
I -3k, 4%

Kz
'

O 12 O 5/2 25

New opt Sol


£*=C 0 ,
5/2 ,
O, 4% ,
05 with # = 25 .

Lii ,
if Ck is the cost coefficient of a B. v .
Kia CEB → Ers )
-

have impact on

Isi B- A E
'

primal optimality
'
=
EBTB -
'
A E
{
: - -

Eri B- I Est 8 cistp


'
value
obj B-
: '
-

= =
.

if primal optimality is violated .


apply primal simplex method to

get a new opt Sol .

II. initial tableau :


opt tableau :

,¢,o# ,¢,,#
N , K2 23 24 25 ths k, Nz Nz 24 Us this

-4 224 O 4
5I20 3 I 0 30 -
I I
-

a, y ,

-2 -7 3 O O O O l l O 2 20

I = ( 3 ,
6 ,
-3,0 ,
03T
I = E -
EE Eri B- A
'
-
E

(? I T)
'

Cs 6
-

= co . 3 >
o
-
3. o.o )
,
. .

=
C 3 .
12 ,
-3 ,
O ,
33 -
C3 ,
6 ,
-3 ,
O .
O )

= C O , 6 ,
O, O , 33 38?

opt see remains unchanged ,


but opt value changes !
Est pi ( %)
'
z'* = = C o, 3 ) =
30 .

opt Sol
Question : How about its dual ?

dual opt Sol changes :

ii -
- cos
-

ist Es =
(
'

o ? 519 ) = (9)
z* = £5 B- I '
= 30 .

I
'

Remark :
,
= r'* = 0 .

rj of B. v . are still 0 .
I Ex ) -

Just need to
update in associated N B V
rj row O with . -
.

.
J
x rhs

③ changes in A
Caij Ej )
fEj7a.E#

ti )
if aij is associated with a B. v .

Nj .
then .

B B- I
{ GTB
'

changes
"

change .

IA -
E

Need to solve the


problem from scratch .

associated
Hi ,
if aij is with a N B. v
>
.

Nj ,
then .

B-' I remains unchanged


\
.

Ii B- E Ij
'
'
A ( )
-

changes changes .

y'j Ej
'
'
= B-

Fj -

Ej -

g- =E5B- '
a'
j
-

g-
=
Iii's .
-

g-

if Fj still opt
\ opt Sol
au >o ,
the current is

otherwise ,
apply primal simplex method to restore
optimality .

II. initial tableau :


opt tableau :

,¢,o# ,¢,,o#
N , K2 23 24 25 ths k, Nz Nz 24 Us rhs
-

24 I 3 4 I 0 30 24 O -
I 5 I
-

I 20

a, y ,

-2 -7 3 O O O O l l O 2 20

ii. =/ 's )
(
'

5. =
'
B- a''s = o T ) ( b) =
(3)
To =
Giga -
Cz

= Co , z > I}) -

= 6-7 = -
I C O .

current opt Sol is no


longer opt .

,z*,,÷
I H,KzN32l4K5r#
Na O -2 5 I -

I IO

y,

O -
l l O 2 20

431-43043/1%43
It NiN2N3K4k5r#
Hq 43 O 13/3 I -43 843
Nz

O 43 O 73 70/3

opt
Sol # =L 0 ,
'
% ,
O, 80/3 , 05
'
with # = 70/3 .
143
I
Addition
of a new variable . ( Kk ,
AI ,
Ck ) .

impact of Uk on the current


opt Sol :

if effect at all
opt Sol is still
nonbasic No
Uk stays opt
{
.

Sol
if Uk becomes basic current
opt is no
longer opt .

check rk :

re -
-
CFB - '
a'e -
Ck

Sol
if 0 current is still opt
{
he
opt
> .
.

should the basis with Tia B tape


-

otherwise Kia enter


-

-
. .

C method )
primal simplex

II. initial tableau :


opt tableau :

,¢,o# ,¢,,#
K , K2 23 Not 25 ths k, Nz Nz 2.4 25 this

-4 224 O 4
5 20I 3 I 0 30 -
I I
-

a, y ,

-2 -7 3 O O O O l l O 2 20

No is added with aJ= (L) co -


-
lo .

No = GTB tab -

Cb =
Jiao -

Cb =
co . 25
'
(L ) -

co = -
6 co .

current opt Sol is no longer opt .

E- ai ( i) K) II )
' -

B-' -

- -

o
,¢,,o,z*÷
I N, NL Hs 2K,
25 No rhs
#

24 O I 5 I l I 20
-
-
-

a,

O l l O 2 -
b 20

yzz%oyz,|#
It HiNzKzk4H526t#
Me
,
Yz I % I -

Yz O
25

×,

3 13 -2 O 5 O 50

:

L53Ad itonofanewconstraihtorigshalon.g.nu
region
l .

itaaYT-n.ro:5?oesaeisties+a
then ,
it is still
opt
!

P C check primal feasibility )

New constraint .

New
14¥ 8
constraint
.io:7?e.itffraf-wopotnssmeamd.oesntsansty

.
(
then ,

incorporate
tableau
it is

and
the
no
longer opt
new

apply
constraint

dual simplex
(
infeasible)
into the

method ) .

II. initial tableau :


opt tableau :

,¢,o# ,¢,,o#
N , K2 23 NG 25 ths k, Nz Nz 24 Us this

-4 224 O 4
5 20
I 3 I 0 30 -
I I
-

a, y ,

-
2 -7 3 O O O O l l O 2 20

Add Kit 22k E 8 .

Garrett opt Sol : n # = 10 ,


Kat -_ 0 .
Kitt 2nF = to > 8 .

the opt Sol is no


longer opt !

Add Kit 2kt 26=8 into the tableau .


/
Hi Ms Hs Ha 25 26 Hrs
-

Na O -
I 5 I -
I 0 20

se
, i ¢ -

, o , o co

26 ① 2 O O O l f
-

O l l O Z O 20

/
TI Ki Ns Hs Ha 25 No rhs
-

Na O -
I 5 I -
I 0 20

se
, i ¢ -

, o , o co

26 O -2 # l O -
l l -2
-

O l l O Z O 20

ratio 42 2

)
2,262324252Gt
No, O O % I -

Yz -

Yz 21

a ,
I o , o -
, z 6

Yz Yz 42
-
-

U2 O l O I
-

O O 3/2 0 3/2 42 19

opt Sol
!

You might also like