Chapter 3

Basic Concepts for Simplex Method

03/18/25 Berhanu G (Dr) 1

• Simplex method is a general algebraic procedure to solve LP.
• In this chapter, we consider its basic concepts and steps.

Recall the Corner Point Theorem for LP:

Optimal solution of an LP occurs at a vertex (corner point) of its
feasible polyhedral.
• Thus, in principle, if the set of all corner points are identified, one can
- evaluate the objective function at each of the corner points
- Comparing these values, the point that yields the best value is an
optimal solution.
• But plotting a feasible set (polyhedral) is possible only in two
dimension (two decision variables)
• In this chapter, we will establish algebraic representation of corner
points of a feasible polyhedral irrespective of the number of variables

03/18/25 Berhanu G (Dr) 2

 3.1 Linear Programs in Standard Form
A linear program is said to be in standard form if it
is given as
max c1 x1  c2 x2  ...  cn xn
s.t. a11 x1  a12 x2  ...  a1n xn b1 Max problem
&
a21 x1  a22 x2  ...  a2 n xn b2 Constraints:
 system of
equations
am1 x1  am 2 x2  ...  amn xn bm
x j 0, j 1...., n
Or : max cx cRn, - row vector
A - m×n matrix
s.t Ax b (Usually m  n )
bRm - column vector
x 0 xRn - vector of decision
variables
03/18/25 Berhanu G (Dr) 3
 In fact, a general LP problem can be
 either a maximization or minimization problem.
 Some of the constraints can be inequalities
connected by either “ ” or “ ” .

However, any LP can be changed to the standard form

as shown below.

1. Minimization problem min cx can be changed to

its equivalent maximization problem max –cx
2. Inequalities can be changed to equations using
slack and/or surplus variables defined as follows.

03/18/25 Berhanu G (Dr) 4

 A) Slack Variables:
The “ ” constraint, say
ai1 x1  ai 2 x2 +  +ain xn bi ,
can be transformed into equality by introducing a
variable s, called slack variable, where
s bi  (ai1 x1  ai 2 x2 +  +ain xn )
Observe that, s  0.
Hence, the given inequality is equivalent to
ai1 x1  ai 2 x2 +  +ain xn  s bi , s 0.
For instance, x1  2 x2 +3x3 10
is equivalent to x1  2 x2 +3x3  x4 10,
x4 0.
( x4 is slack variable)
03/18/25 Berhanu G (Dr) 5
 B) Surplus Variables:
The “ ” constraint, say
ai1 x1  ai 2 x2 +  +ain xn bi ,
can be transformed into equality by introducing
a variable s, called surplus variable, where
s ai1 x1  ai 2 x2 +  +ain xn  bi
Observe that, s  0.
Hence, the given inequality is equivalent to
ai1 x1  ai 2 x2 +  +ain xn  s bi , s 0.
For instance, x1  2 x2 +3x3 10
is equivalent to x1  2 x2 +3x3  x4 10,
x4 0.
( x4 is surplus variable)
03/18/25 Berhanu G (Dr) 6
 Thus, any LP problem (min or max) with any form of
inequality constraint can be converted into equivalent
standard LP form (max problem with equality const.)

Example
Change the following LP problem into standard form

min  3x1  x2  x3
s.t x1  3 x2 +5x3 15
2x1  x2 +5x3 10
5x1  2 x2 +x3 270
x1 , x2 , x3 0

03/18/25 Berhanu G (Dr) 7

 3.2 Basic Feasible Solutions
 If B is mm non-singular matrix and bm, then the system Bx= b
has a unique solution given by x = B1b
 Consider a system of m linear equations with n variables,
Ax b.
where A is an mxn matrix bRm, and rank(A) = rank(A, b) = m.
( i.e., A has invertible mm submatrix )

 Any mm invertible submatrix , denoted by B, is called basic and

its columns are called basic columns of A. The remaining
submatix (with n–m columns) , denoted by N, is called non-basic
and its columns are called non-basic columns.
 The m variables attached to the basic columns are called basic
variables (BV), and the remaining n–m variables are called non-basic
variables (NBV).
03/18/25 Berhanu G (Dr) 8
Thus, rearranging if needed, A can be partitioned as A = (B, N )

Let also x be partitioned as x = (xB , xN ), where

xB is the m-vector of BVs attached to the basic columns in B,

xN is the nm vector of NBVs attached to the columns in N,

Then, Ax = b  BxB + NxN = b

 BxB = b  NxN
 xB = B1b  B1NxN
(called basis representation of x wrt B )

03/18/25 Berhanu G (Dr) 9

 Hence, for any choice of values of xN , say xN= unm ,
the values of xB is uniquely determined by xB = B1b  B1Nu
so that x = (xB, u) is a solution of Ax = b.
Thus, if we choose xN = 0 = (0, 0, …, 0)T, then xB = B–1 b;
and 1
 xB   B b 
x    
 xN   0 
is the solution of Ax = b, called basic solution wrt B.
Definition: A basic solution x = (xB , 0)T is called basic
feasible solution (BFS) with respect to B if
xB = B–1 b  0. (nonnegative BS)
Note: 1. A BFS is a solution of Ax = b, x  0.
2. There are only finite number of BFSs.
03/18/25 Berhanu G (Dr) 10
 Example 1: Find all basic solutions (BS) of
x1  x2  2 x3  3 x4 3
x1  2 x2  2 x3 4
(and also identify basic feasible solutions (BFS) )
Solution: The coefficient matrix is A = (A1, A2, A3, A4), where
 1 1    2  3 3
A1   , A2   , A3  , A4   and b   ,
 1 2
    2   0  4

A has 6 sub-matrices of order 2:

B1  A1 , A2 , B2  A1 , A3 , B3  A1 , A4 ,
B4  A2 , A3 , B5  A2 , A4 , B6  A3 , A4 

and det( B1 ) 1, det( B2 ) 0; det( B3 )  3,

det( B4 ) 2, det( B5 )  6; det( B6 ) 6.

 5 basic sub-matrices: B1, B3, B4, B5, B6

03/18/25 Berhanu G (Dr) 12
 So, there can be 5 basic solutions w.r.t each of the basic
sub-matrices. In Particular,
2
• W.r.t B1  A1 , A2 , X B1 B1 1b  1 
 
 ( x1 , x 2 , x3 , x 4 ) ( 2,1,0,0is
) a BS, which is also BFS

• W.r.t B3 A1 , A4  4 
X B 3 B3 1b  
  1 / 3 

 ( x1 , x 2 , x3 , x 4 ) ( 4,0, 0, 1 / 3) is a BS, but NOT a BFS

• Similarly, w.r.t B4, B5 & B6 we obtain, respectively:

( x1 , x2 , x3 , x4 ) (0,1,  1, 0) a BS, but NOT BFS

( x1 , x2 , x3 , x4 ) (0,4, 0, 1/3) a BS, and also BFS

( x1 , x 2 , x3 , x 4 ) (0,0, 2,  2/3) a BS, but NOT BFS

03/18/25 13
Berhanu G (Dr)
 3.3 Algebraic Analysis of the Constraint Set
From Chapter 2: the maximum or minimum of an LP (if it exists)
occurs at a corner (vertex) point of the feasible polyhedral set.
Next, we will show that a corner point is BFS .
To see this, consider the following constraints:

 x1  2 x2 4
x1  x2 5  x1  2 x2 4
x1 , x2 0
P3

Corner points (vertices) of S:

P2
S
x1  x2 5
P1 , P2 , P3, & P4
P4
P1

03/18/25 Berhanu G (Dr) 14

 Now, introducing two slack variables x3 , x4  0, the above two
inequalities are equivalent to
 x1  2 x2  x3 4 BFS is obtained by setting
x x
1 2  x 5 0 to two variables
4
Note that one of the variables is 0 on an edge of the polyhedral S.

A corner (vertex) corresponds to  x1  2 x2 4

point where two variables are 0 ,
i.e.,non basic, are: P3
P1 : x1= 0, x2= 0, x 3= 0 x4 = 0
P2 x1  x2 5
P2 : x1= 0, x3= 0, S
P3 : x3= 0, x4= 0, x1= 0
P4: x2= 0, x4= 0, P1 x2= 0 P4

Thus, the corner points are basic feasible solutions

03/18/25 Berhanu G (Dr) 15

 This suggests that an optimal solution of an LP
occurs at its BFS.
In fact, this is true for any LP.
Hence, we have the following important Theorem.

Theorem 1 (Fundamental Theorem of LP):

Consider an LP max/ min cx

s.t. Ax b
x 0
where A is an m×n matrix.
An optimal solution of the LP, if it exists, will
occur at a BFS of its constraint equations.

03/18/25 Berhanu G (Dr) 16

 Based on the Fundamental of LP, one may solve a bounded LP
problem, max {cx | Ax=b, x≥0 }, using the following procedure:
• Step 1 Find all BFS of the constraint system Ax=b
• Step 2. Evaluate the objective function cx at each BFS
• Step 3. Comparing the values in Step 2; a BFS at which cx
attained the highest value is an optimal solution.

Example: Consider max x1  4 x2  5 x3  3 x4

s.t x1  x2  2 x3  3 x4 3
x1  2 x2  2 x3 4
x i 0
• We have seen that (sec. 3.2), the BFS of the constraints are
X1= (2, 1, 0, 0) and X2 = (0, 2, 0, 1/3)
• The objective function at X1 is 6, and at X2 it is 9.
Hence (assuming it is bounded) optimal solution is
(x1,x2,x3,x4) = (0, 2, 0, 1/3) and optimal value (max value) = 9
03/18/25 Berhanu G (Dr) 17
03/18/25 Berhanu G (Dr) 18