Lecture (04-09)

Linear Programming (Planning

·

Allocating limited resources

among competing activities to optimize output

Example :
Chapter 3 section 3 1
.

(Wyndor

decides to products
Company make2 Product 1 and Product 2
.
·

has
Company 3
manufacturing plants Plant 1 Plant and Plant
·

:
,
2 ,
3

·
Product1 requires Plant 1 and 3

Product 2
requires Plant 2 and 3

Rea hours per

bathch
Hours
Plant Prl Dr 2 available

I I O 4

2 O 2 12

3 3 2 18

Profit per
bathak $3k $54

(Hours/week)

Define variables :

X, # of batches Prl
per week
·
=

Decision variables
·
X= = ot batches Pr2 per week

total objective function

profit
·
z =
week >
per

z =
3x1 + 5X2

Max Z = 3x , + 5x2 subject to X114 X ,, X2 = 0 <

non
negativity
2X2112
functional constraints
3x ,
+
2X2 = 18

Use graphical procedure to solve

X 21

combo of decision variables is called solution

12
any
solution that satisfies all constraints is "feasible"
10
12 -Feasible
6 ·
region
(M)

Y
·
(4 3)
,

2
3x , 2x2118
>
+

-
&
12 6
3 4 XI
Lecture (09-09)

Problem :

Factory making two different products

·
Can make 100 cases of Prl in 6h

·
Can make 100 Cases of Pr2 in 5h

Factory has 60h/week for Pul and Pr2

·

stockroom must keep production (Storage capacity 15000ft) Pul 10ft" and PrC 3

weekly 20ft
·

uses
:
is uses
per case

·
Pul sells for $5/case & customer will not take more than 800)
·
Pr2 sells for $1 5/ case
. (unlimited)
Q1 Find
weekly production to profit
·

: max
?

·
Q2 : Find that profit

Step 1 :

Define decision variables : X = number of cases of Prl/week (in hundreds

Xz =
number of cases of Pr2/week (in hundreds)

Step 2 :
Z = 500X , +
150X2

Step 3 :
functional constraints :
(1) 6X1 +
5x2 = 60 (hours available)
(2) 10x1 +
20x 1150 (storage in 100ft)
(3) X, 18

X , 20 , X2 = 0

Maximize : E =
500X , +
150Xc subject to 6X1 + 5X2 = 60 m

10X1 + 20x2 -150 m

X, S aut

X = 0 ,
X2 = 0

X, N

#
" X = Z -
500 X ,
450 450
10

-z =
<500 4507

5
,

7 .

slope level set for E(X X2)

-10
=
,,

2 Graphically , see that optimal solution is at P ,

that is at the

*
intersection of 6X ,
+ 5x2 =
60
H
2 681012141516 X,
10X X
20X2 150 -Xz
12 637
,
+ : = =
,
,

2
(63 42) 51427
,
=

Integer solution :
(x1 x2) .
= (6 1), <z(6 4) ,
=
4800
Lecture (09-09)

Next topic :
Shadow Prices

Suppose we are
considering increasing hours from 60 to 6

(1) 6X ,
+ 54256
(2) 10x , + 20x27150P =

(65 4) ,
<
z(e) =

5221

The "shadow price" associated with constraint on

production capacity

52213
51426
784
-
=

~60h

Q :
How far can I keep increasing production capacity and keep getting 78/h ? 160 ·
61 : 62. ...
)

P3 = (8 , 3) (cause you
will
moving
this line to the right until it doesn't matter)

60 1 59 158 ..... Pu = (0 , 7 t) 6 .

0 + 5 .

7 =

37th

if get below 37th only making Pr2

Lecture (Il-0a)

Sensitivity Analysis
Max Z =
500X ,
+
130X2 s t
. .
6X ,
+ 5x2 = 60 hours production aup

storage
U
10X 2012 = 150
,
+
space
Wo
X ,
= 8 ,
X , 20
,
X = 0

Xzn
X =
production rate of Pr
16
X =

production rate of Pr2 14

i
Xz = z -

500X
,
450 450

246810121416 Y

Suppose we not sure about selling Prl for $5/case

Z =
CX
, ,
+ 150X2

From above
picture it's clear that optimal
,
solution (P1) remains of P , as
long as obj function lies between slope of prod const (-6/5)
and storage constraint (-1) .
So for P , to be optimal , must have-6/s =
objfunction slope = -12
-
Ci
1130

5402C = 223

As
long as Prl sells between C .
25/case and 5 .
4/case
production levels atM will
give optimal solution

Wyndor Problem : Max Z =

3X , + 5X2 s .
t
.

2X2
X=4

= 12
i
6 (2 , 6)

3X + 2x2118
Y
(4 3)
.

X = 0 X2 2 O

..
,
2

2 46
S

Any combination of decision variables is called a "solution" Some solutions. are allowed (satisfied all constraints) called "feasible solutions"
"Best" feasible solution is called optimal solution . Possible for feasible region to be a

CPF >
-

corner point feasible

Any (Problem with bounded feasible region will posess CPF solutions at least one CPF solution will be optimal
If CP Problem only has one optimal solution it must only be a CPF solution . If LP problem has more than I optimal solution ,
at least 2 will be

CPF solutions

(Non-linear) Lagrange mult method extreme points happen when level curves of objfunction (surface) tangent to b-day
Look for point with gradient parallel
Lecture (Il-09)

z = (3 3)
,

Ig =
< 2X ,, 2x2]

#z = X Tg +
(3 3) ,
=
x(2X 2x2)
,,

() 3 =
X2X ,
-> X, =
33x
(2) 5 = x2x2 < Xz =
3
2x (2) (c)"
+ =
1

(3) X 2 ,
+
Xz = 1
Lecture (16-0a)

Assumptions / Properties that

go along with QP problems
Linear function

f(ax by) + =
a f(x) +
bf(y)

1.
Proportionality :
Contribution of each activity to value of objective function (and constraint) is
proportional to level of that activity.
Only have power I terms
2 .

Additivity Every function :

is the sum of individual constributions of each respective activity. f(x ,, 0) +
f(0
,
xx) =

f(x ,, X2) ·

additivity
holds for "separable" functions
.
3
Divisibility Any fractional :
solution is valid as
long as it is inside feasible region (i . e not
dealing with integer programming
Certainty : Each parameter in obj funct and in functional constraint given (fixed) is a

If it doesn't hold in small number of vars perform sensitivity analysis

Solid waste reclamation example :

A center collects 4
types of waste material ; and generates 3
grades of salable product
Grades ABC ;
Types 1 ,
2, 3 7
,

Il I
Grade Specs Amalgamation cost/1b Sale price/lb mat lb/week avai treat cost/1b
mat I = 30% total
A matc = 40 % total 3 .5
8 I 3000 3

matz = 50 % 2 2000 6

maty =
20 % 3 4000 Y

matl = 50% U 1000 5

B mat 2 7
= 10 % 2 5 .

maty =
10%

C mat 1 = 70 % 2 55 &

·
At least half of all incoming material must be used up per week (30k per week available to treat all
material)

↓ P formulation define : :

yi
=
pounds product grate (i
number =
A B , , c) to produce :

Ya , Yes Ya ,

Zij =

Proportion of matj in grade i (j .

= 1 ,
2
,
3 ,
4)

Consider : = 3000 lbs mat

o
in
grade A grade B

Don't want cross terms because not LP problem

Define Xij : = number of pounds of mat ; allocated to product i

Xa1 +
Xaz Xaz + Xau =
number of product A produced week
+
per
↓ ecture (16-09)

X B1 +
XB2 +
XB3 +
Xin =

product B
per week

Xai +
XBi + Xc1 =
number lbs of mat I used

XAz +
XBz +
Xc2 =
mat 2
per week

XAI =
proportion of mat I in prod A

Xa1 +
Xaz + Xaz + Xan Y lbs prod A

XAz >
0 4 .

Xa + XAz +
XAz + Xau

XAz = 0 .
4 (XA1 +
Xaz + XA3 +
Xan) <
0 6 XA2-0
. .
4 Xal- 0 4 XA3-0 UXau
. .
= O

Xij O

XA2 Xaz Xan) (7 XB2 XB3 Xiu) (5 5

2)/X Xe Xcu)
z 18
5.3) 5) X Xcz
+ + + + +
2 + + + -
=
.
+ + -
.

amount of grade A grade B

grade C

Treatment cost 31 Xi
Xc) 6Xan) 4(XA3 Xcs) X
5 Xi
+Xi
: + + + + + + +

-
mat I mat 2 mat H

12 Decision variables
16 or 17 functional constraints

Simplex method >

(SM) uses
following property of LP problems .
If a corner point feasible solution such that none of it's "neighbors" CPF are "better" then such solution is

z =
3X , + 5X2 optimal
X, H

2X2[12
Ashley young

3X ,
+ 2X2 = 18

X = 0
,
X = 0
Lecture (23-0a)

Use
Wynder problem

Xer .
(2 , 6) max z =
3X ,
+ 5x2 subj X . 14

Li
2 x 2 1/2
5

H
3X ,
+ 2x2 = 18

X, 20 , Xz = 0
3

↓ Any CPF lies on intersection of "n" constraint b dries .

12356
>
XI
Def for LP problem with "n" decision vars two CPF "adjacent" to each other if they share n -1

constraint boundaries

Two adjacent CPF solutions are connected by straight line

segment called "edge" of feasible region

SM
only examines CPF solutions and "trends" of objective function along edges

Optimality test :
If objective function value at all adjacent points is not "better" than current CPF such
,
CPF is
optimal

Geometric solution concepts :

1. Focus
solenly on CPF solutions .

Finding "best" CPF solution provides an optimal solution.

2
. Simplex Method (SM) is iterative algorithm Repeat . same
sequence of steps
: Initialization (start at
origin
CPF)
V

Optimality test :
yes
:
stop
to another "adjacent"
~
no : iterate to move CPF

3
. When possible initialization ,
choose
origin . When origin is not CPF , we will transform system to make it CPF

. Given
9 a CPF solution ,
it is computational efficient to assest adjacent CPF solutions . Moves along edges towards optimal solution
. At
5 a "current" CPF solution SM examines
every edge of feasible region for rae of improvement when
moving along each edge
.

"best"
rate of improvement is the selects to to next
The edge that
gives edge SM move CPF
6
.
Optimality test Check wheter edge emanating
:

any from CPF

gives positive rate of improvement of z

Algebra of SM

Need to
change (convert) all inequality constraint to equalities . Does not include non-negativity
X, 4 max z =
3X1 + 5X2

2X2 = 12
XM

X,
3 + 2x218

Xi ,
220

4
" cononical form
X ,
+ Xz =
>
4 XI

2xz + Xu = 12

3X, + 2x2 + Xi = 18
only do "this" to functional constraints
non-negativity
Xi ,
Xz = 0
; X3 , 4 , 3
= O

new vars like X3 .

4 , s are called "Slack variables". All slack vars must have non-negativity constraint to ensure we stay inside feasible
region .

Recast our LP problem to "Augmented Form"

Lecture (23-09)

Original Form :
max Z =
3X , + 3X2 subj X , 24 Q

2X2 = 12 ②

3 x, +
2X2 = 18 ⑬
X, 2 - 0

Augmented form :
max z = 3X , + 5x2 subj X, +
Xz = ⑪

2xz +
Xy =
12 Q

3 +
2xz + Xs =
18

X1 , 2, 3 , 4, 3 - 0

Solution corresp to origin :

(X .
X , Xa ,
x, Xs) = 10 ,
0, 4 , 12 ,
18)
Lecture (25-09)
X2 a

Homework

3
:
max Z =
CX + X2 subj X ,
+ X2 = 6 7

·
X, +
2X2 10

X ,,
= O

to be at
for opt solution P must have 13-c- *
1 = c =

I
if < c < 1 then optimal solution at P
12345673X
if c =
1
-opt solution is
segment marking
"side" between (2 4),

and (3 4) ,

if ( > 1 <
(6 0 ,

if <
<
2 <
(0 5),

X2n
subj I

3
max Z =
CX ,
+ CaXz 2X ,
+ X2 -
6

-
X, +
2X2 k2
5

X, 2,
O
Y

of obj fun

Slope
Casel : C2 =
O
, ,
Z =
C XI ,

subcase C, 0 > at (52 @) ,

subcase C , <O' at line segment between (0 0) and (0 1) , ,

Case 2 : C =
0

z = C2Xz

subcase C30 = (4 3) ,

subcase G 1O <
between 10 0) and (52 0 , ,

Try all combinations C ,

+ 0 ,
C + 0

Slope-C compare
to slopes an -

X2n
max Z =
X, + CX2 sobj X2-X , = 2 3 ·
(2 3)
,

Y
·

kX ,
+
X2

X =
3

(k + 3 <
X2 * (2-Xi) + 32
-
X, 2 = O ↑
slope = -

k =
slope of obj fun =
-
*0
I 2 3X

want-k=- =
1
Lecture (25-09)

SM X21

Wyndor : max Z =
3X , + 5X2
6 ·
(2 6),

X ,
14
4

2x2 = 12 3 ·
(4 , 3)

3X1 +
2x2 = 18 2

X, , 2
= O
I

>
XI
I 2 345 6

Original form :
max Z =
3X , + 5X2 subj X14 (1)
2X2 = 12 (2)

3X, +
CX2 = 18(3)

Augmented form : max Z =

3X , + 3X2 subj X + X3 = 4
subj X1 , 2, 3, 4 , O

2x2 + Xu = 12

3X ,
+ 2x2 + X3 = 18

"Augmented solution" is a solution that includes both decision vars and slack vars

Let (X ,,
X2) =
(3 2) ,
<
can solve for (Xs ,
Xy , Xs)
from (1) =
X3 =
-Xi =
X3 =

from (2) > Xy =

12-2X2 <
Xu =
8

from (3): Xs =
18-3X -2x2cXs
,
= 8

augmented solution = (X ,, Xn ,
X3 ,
Xu ,
Xs) = (3 ,
2, 1 . 8 ,
3) <
feasible

CPF =
(4 3)
,

(X , X2 , X3 Xy , ,
Xs) = (4 , 3,0 , 6 ,
0) :
Basic Feasible Solution (BF)
Corner Point Solution in
augmented form are called Basic Solutions

CPI =
(4 6) .
< Basic Infeasible > BI =
(4 , 6 , 0, 0 . -6)

Solution (X, X2) (10 10) ,

=
,

augmented solution (X ,,
X2 , X3 Xy , ,
Xs) = (10 , 10 ,
-6 ,
-8 ,
-32) Violates all constraints

Each variable gets designated as "basic" or "non-basic"

At initialization a basic variable will be slack variable
(X ,,
Xz) = 10 0),
=
10 ,
0, 4 ,
12 , 18) <
Xi =
non-basic
X2 =
non-basic

X3 , 4, s
= basic

number of non-basic variables number of decision variables non-basic variables are always O
is :

number of basic variables is the number of functional constraints (ignore non-negativity) the basic variables .
are non-zero (except in

cases of degeneracies
Lecture (25-0a)

pI
CPF BF basic vars non basic vaus

XI ,
X2

X ,, Xu

(2 6) ,
(2 , 6 ,
2, 0 . 0) Xy , s

Xc Xz
,

(4 3)
,

Two adjacent CPF solutions share all but one basic var and all but one non-basic var