FOUNDATIONS OF OPTIMIZATION: IE6001

Linear Programs in Standard Form

Napat Rujeerapaiboon
Semester I, AY2019/2020
 LPs in Standard Form

• We want to use computers to solve LP problems

⇒ We need a standardised specification of LP problems

INPUT: OUTPUT:
LP in Optimal Value &
Standard Form Optimal Solution

Definition: An LP is in standard form if:

• The aim is to minimize a linear objective function;
• All constraints are linear equality constraints;
• All constraint right hand sides are non-negative;
• All decision variables are non-negative.
 LPs in Standard Form

An LP in standard form looks as follows:

minimize c1 x1 + c2 x2 + ... + cn xn
subject to a11 x1 + a12 x2 + ... + a1n xn = b1
a21 x1 + a22 x2 + ... + a2n xn = b2
.. .. .. ..
. . . .
am1 x1 + am2 x2 + . . . + amn xn = bm

x1 ≥ 0, x2 ≥ 0, . . . , xn ≥ 0

The input parameters bi , cj , and aij are fixed real constants that
encode the LP problem. We require bi ≥ 0 ∀i = 1, . . . , m. (The
decision variables xi , i = 1, . . . , n, are yet to be found.)
 Compact Notation

Collect the input parameters in vectors and matrices:

 
a11 a12 . . . a1n
 a21 a22 . . . a2n 
A= .
 
.. .. .. 
 .. . . . 
am1 am2 . . . amn
     
b1 x1 c1
 b2   x2   c2 
b=  x =  c=
     
.. .. .. 
 .   .   . 
bm xn cn
c T = [c1 , c2 , . . . , cn ]
 LPs in Standard Form (cont)

• With matrix notation, the LP in standard form reduces to

minimize c T x
subject to Ax = b
x ≥ 0,

where b ≥ 0.

• Inequalities of the type x ≥ 0 are understood to hold

componentwise.
 Standardising General LPs

• General LP problems can

• be maximization (instead of minimization) problems;
• have inequality (instead of equality) constraints;
• have equality constrains with negative (instead of
non-negative) right hand sides;
• have free (instead of non-negative) decision variables.
• These general LPs can be transformed to standard LPs in
a systematic way.
 Maximization → Minimization

 
max f (x) − min −f (x)
=
s.t. x ∈ X s.t. x ∈X
Inverting the objective preserves the optimal solution x ∗

-f(x)

min -f(x)
x* x*
X X
max f(x) - min -f(x)

f(x)
 ≤ Inequalities → Equalities

minimize c1 x1 + c2 x2 + . . . + cn xn

subject to:

a11 x1 + a12 x2 + ... + a1n xn ≤ b1

a21 x1 + a22 x2 + ... + a2n xn ≤ b2
.. .. .. ..
. . . .
am1 x1 + am2 x2 + . . . + amn xn ≤ bm

and

x1 ≥ 0, x2 ≥ 0, . . . , xn ≥ 0
 ≤ Inequalities → Equalities

minimize c1 x1 + c2 x2 + . . . + cn xn

subject to:

a11 x1 + a12 x2 + ... + a1n xn + y1 = b1

a21 x1 + a22 x2 + ... + a2n xn + y2 = b2
.. .. .. ..
. . . .
am1 x1 + am2 x2 + . . . + amn xn + ym = bm

and

x1 ≥ 0, x2 ≥ 0, . . . , xn ≥ 0, y1 ≥ 0, y2 ≥ 0, . . . , ym ≥ 0
 Slack Variables

• To reformulate ≤ inequalities as equalities, we introduced

m slack variables
• Original variables: x1 , x2 , . . . ,xn
• Slack variables: y1 , y2 , . . . ,ym
⇒ After transformation, LP has n + m variables!
• With matrix notation we can write

min c T x  min c T x
 

s.t. Ax ≤ b = s.t. Ax + y = b
x ≥0 x ≥ 0, y ≥ 0,
 

where y = (y1 , . . . , ym )T .
 ≥ Inequalities → Equalities

minimize c1 x1 + c2 x2 + . . . + cn xn

subject to:

a11 x1 + a12 x2 + ... + a1n xn ≥ b1

a21 x1 + a22 x2 + ... + a2n xn ≥ b2
.. .. .. ..
. . . .
am1 x1 + am2 x2 + . . . + amn xn ≥ bm

and

x1 ≥ 0, x2 ≥ 0, . . . , xn ≥ 0
 ≥ Inequalities → Equalities

minimize c1 x1 + c2 x2 + . . . + cn xn

subject to:

a11 x1 + a12 x2 + ... + a1n xn − y1 = b1

a21 x1 + a22 x2 + ... + a2n xn − y2 = b2
.. .. .. ..
. . . .
am1 x1 + am2 x2 + . . . + amn xn − ym = bm

and

x1 ≥ 0, x2 ≥ 0, . . . , xn ≥ 0, y1 ≥ 0, y2 ≥ 0, . . . , ym ≥ 0
 Surplus Variables

• To reformulate ≥ inequalities as equalities, we introduced

m surplus variables
• Original variables: x1 , x2 , . . . ,xn
• Surplus variables: y1 , y2 , . . . ,ym
⇒ After transformation, LP has n + m variables!
• With matrix notation we can write

min c T x  min c T x
 

s.t. Ax ≥ b = s.t. Ax − y = b
x ≥0 x ≥ 0, y ≥ 0,
 

where y = (y1 , . . . , ym )T .
 Negative Right Hand Sides

• If the right hand side of the ith constraint is negative, i.e., if

bi < 0 in
ai1 x1 + ai2 x2 + . . . + ain xn = bi ,
then this constraint should be multiplied by −1.
• This yields

(−ai1 )x1 + (−ai2 )x2 + . . . + (−ain )xn = −bi .

• The new constraint has a non-negative right hand side,

i.e., we have −bi ≥ 0.
 Free Variables (1st Approach)

• Free variables:
• Suppose there is no constraint xj ≥ 0,i.e., xj can be positive
or negative.
• Substitute xj = xj+ − xj− with xj+ , xj− ≥ 0.
• The LP has now (n + 1) variables:

x1 , . . . , xj−1 , xj+ , xj− , xj+1 , . . . , xn

 Free Variables (2nd Approach)

• Free variables:
• Suppose there is no constraint xj ≥ 0,i.e., xj can be positive
or negative.
• Any equality constraint involving xj can be used to eliminate
xj .
• Example: x1 is free

min x1 + 3x2 + 4x3 
  min x2 + 3x3 + 5


s.t. x1 + 2x2 + x3 = 5 (∗)
= s.t. x2 + x3 = 4
2x1 + 3x2 + x3 = 6 
x2 , x3 ≥ 0
 
x2 , x3 ≥ 0


Use (∗) to substitute x1 = 5 − 2x2 − x3 .