Transportation problem (TP)

There are m supply stations S1 , ..., Sm for product Q.

There are n destination stations D1 , D2 , ..., Dn where Q is
transported.
cij is the cost of transportation of unit amount of Q from Si
to Dj .
ai is the amount of Q available at Si .
dj is the demand of Q at Dj .
To find xij , i = 1, 2, .., m, j = 1, 2, ..., n, where xij is the
amount of Q to be transported from Si to Dj such that the
demand at each Dj is met and the cost of transportation is
minimum.
The problem
P is given by:
Min i,j c ij xij
subject
Pn to
xij ≤ ai , i = 1, 2, ..., m
Pj=1m
i=1 xij ≥ dj , j = 1, 2, ...., n,
xij ≥ 0 for i = 1, 2, .., m, j = 1, 2, .., n.
It is clear P
that for the
P transportation problem (TP) to be
feasible a
i i ≥ j dj .
A transportation
P P problem (TP) is called balanced if
i ai = j dj .
In that case all the inequalities in the constraints should
hold as equalities.
A balanced
P transportation problem (TP) is given by,
Min i,j c ij x ij
subject
Pn to
xij = ai , i = 1, 2, ..., m
Pj=1m
i=1 xij = dj , j = 1, 2, ...., n,
xij ≥ 0 for i = 1, 2, .., m, j = 1, 2, .., n.
P P
Since i ai = j dj
if x = (xij )mn×1 satisfies any (m + n − 1) equations then it
automatically satisfies all the (m + n) equations.
The constraints are of the form
Ax = b,
where
 z }| { 
111..11 0n 0n ... . 0n
 n 
 z }| { 
 0n 111..11 0n ... . 0n 

 n 

z }| {
0n 0n 111..11 0n . 0n
 
 
 n 

 . . . . ... . 

A(m+n)×mn =  z }| { 
 0n . . ... 0n 111..11 
 n 
 z }| { z }| { z }| { 
 100...0 100...0 . . ... 100...0 
 
 z }| { z }| { z }| { 
 010...0 010...0 . . ... 010...0 
 

 . . . ... ... . 

z }| { z }| { z }| {
000..01 000..01 . . ... 000..01
b = [a1 , a2 , .., am , d1 , d2 , ..., dn ]T (the 0n ’s are row vectors
with n components).
rank(A)= m + n − 1.
Remove the last equation (any equation) from the (m + n)
equations.
In the new Ax = b, the dimension of A is (m + n − 1) × mn
and b = [a1 , a2 , .., am , d1 , d2 , ..., dn−1 ]T .
Any BFS of this problem will have m + n − 1 basic
variables and the order of any basis matrix, B is
(m + n − 1) × (m + n − 1).
Theorem 1 : Let B be a basis matrix then:
1. There exists a row of B with exactly one nonzero entry
(which is a 1).
2. The sub matrix obtained by deleting the corresponding
row and column (containing that nonzero entry ) from B will
again be nonsingular and will have a row with a single
nonzero entry.
If B is a square sub matrix of A with properties 1 and 2 of
theorem 1, then |B| = ±1
D is any nonsingular sub matrix of A if and only if D has
the same structure as B.
If D is any square submatrix of A, then | D | is either 0,1 or
-1.
Such matrices (such as B) are called triangular matrices,
and because of this special structure of B it is easy to
solve system of equations of the form BxB = b (which will
give a basic solution of the transportation problem).
If the i th row of B has a single nonzero entry at the j th
column, then assign xij = bi (where bi is either ai or dj ).
Remove the ith row and the j th column from B which
gives the matrix B1 .
Solve the system B1 x0 = b0 ,where x0 is obtained from x by
removing xij and b0 is obtained from b by removing bi and
changing bj to bj − bi .
Continuing in this way solve the system Bx = b.
In any basic
P solution the basic variables takes values of
the form, i γi bi , where the γi ’s are either 0,1 or -1.
In any BFS x of TP with supplies ai , i = 1, 2, ..., m and
demand dj , j = 1, 2, ..., n the basic variables take values of
the form,
P P
xij = i αi ai + j βj dj where the αi ’s, βj ’s are either 0,1 or
-1.
Transportation Array (TA): The mn variables xij can be
arranged in an m × n array known as the m × n
transportation array.
In a transportation array (TA) each cell corresponds to a
variable, that is the (i, j)th cell corresponds to variable xij .
The m rows correspond to the m supply constraints,
hence the sum of the values of the variables in row i is
given by ai .
Similarly the n columns correspond to the n demand
constraints and the sum of the values of the variables in
column j is given by dj .
Definition 1: A subset of cells of the TA is said to be
linearly independent if the set of column vectors in the
matrix A corresponding to the cells (variables) are linearly
independent( LI). Otherwise they are linearly dependent
(LD).
Any set of ≤ 2 cells is LI.
Any set of ≥ m + n cells is LD.
Definition 2: A subset of (m+n-1) cells of the TA is said to
be a basic set if they are linearly independent. The cells in
a basic set are called basic cells.
Remark 2: A basic set corresponds to a basic solution of
the TP, where the variables corresponding to the basic
cells are basic variables and the rest are nonbasic
variables.
Remark 3: If B be a basic set of cells then the submatrix
of A(m+n−1)×mn with columns corresponding to B, is a
basis matrix B.
There exists a row of B with exactly one nonzero entry.
Since we are now solving BxB = [a1 , .., am , d1 , .., dn−1 ]T
and each row of B corresponds to a constraint (supply or
demand).
There exists a constraint in the TP which has exactly one
basic variable corresponding to B.
Since each row and column of the transportation array
corresponds to a constraint, there exists a row or column
of the transportation array which has exactly one cell from
the basic set B.
If row i contains a single nonzero entry at the (i, j) th
position, then the submatrix obtained from B after deleting
the i th row and the j th column from B again has the
same property.
If B is a basic set of cells and if the row or column having
a single basic cell is struck off from the TA, then in the
reduced (or remaining) array there will again be a row or
column with a single basic cell.
Since every row and column of the array has at least one
basic cell, one can continue till all the rows and columns of
the TA are struck off.
Also any set of m + n − 1 cells which satisfies the above
property is a basic set of cells.
Example 1: Consider the TP with ai and dj as given
below:

j =1 2 3 4 5 6 ai
i =1 7
2 17
3 5
4 24
dj 15 10 9 3 8 8

Let us first start with cell (2, 3) as a basic cell and then try
to construct a BFS of the above problem.
Since the minimum of a2 and d3 is d3 = 9, we take x23 = 9.
Delete the third column and change a2 from 17 to
a20 = 17 − 9 = 8.
In the reduced array choose a basic cell say (2, 4). Take
x24 = 3 since 3 = min{d4 = 3, a20 = 8}. Proceeding in this
way we get the following BFS.

j =1 2 3 4 5 6ai
i =1 [7] 7
2 [9] [3] [5] 17
3 [3] [2] 5
4 [15] [3] [6] 24
dj 15 10 9 3 8 8
North West Corner Rule for Initial BFS

ai
40 40
30
60
dj 50 20 40 20

ai
40 40
10 30
60
dj 50 20 40 20
 North West Corner Rule

Start by choosing a basic cell from the North-West corner

of the array and assign the maximum possible value
(maintaining feasibility) to that variable.

ai
40 40
10 20 30
60
dj 50 20 40 20

Delete the first row of the TA ( the constraint being taken

care of) and repeat the above process with the North-West
corner of the reduced array.
 North West Corner Rule

ai
40 40
10 20 30
0 60
dj 50 20 40 20

Delete the first column of the array and repeat the above
process by starting the North-West corner of the reduced
array. The following tables shows the various steps
following the North-West corner rule till a BFS is obtained.
North West Corner Rule

ai
40 40
10 20 30
0 40 60
dj 50 20 40 20

ai
40 40
10 20 30
0 40 20 60
dj 50 20 40 20
 θ -loops

A collection of cells of the TA is said to form a θ- loop if it

satisfies the following conditions.
1. Nonempty.
2. Every row and column of the TA either has 0 or 2 cells
from this collection.
3. No proper subset of this collection satisfies both
property 1 and property 2.
Consider the following examples.

1 2 3 4
1 ◦ ◦
2 ◦ ◦
3
4

1 2 3 4
1 ◦ ◦
2 ◦ ◦
3 ◦ ◦
4 ◦ ◦
 1 2 3 4
1 ◦ ◦
2 ◦ ◦
3 ◦
4

1 2 3 4
1 ◦ ◦
2 ◦ ◦
3 ◦ ◦
4 ◦ ◦

In the third and fourth example, the marked cells do not

form a θ- loop of the 4 × 4 transportation array, since it
violates properties 2 and 3, respectively.
The first and the second however are θ- loops.
Theorem 4: The cells in a θ- loop are LD.
Theorem 5 : If 4 is a nonempty collection of cells of the
TA which contains no θ- loop then it satisfies,
1. There exists a row or column of the array with exactly
one cell from 4.
2. Every nonempty subset of 4 should satisfy property 1.
Theorem 6 : If 4 =
6 φ is a collection of cells of the TA
which contains no θ- loop, then 4 is LI.
Corollary 6: A subset of cells 4 of the TA is LI if and only
if it contains no θ- loop.
Theorem 7: If B is a collection of m + n − 1 basic cells of
the TA and (p, q) ∈
/ B, then B ∪ {(p, q)} contains one and
only one θ-loop and this loop includes the cell (p, q).
How to get the optimal solution from a given basic
feasible solution?
Let x = (xij ) be the initial BFS.
The Pdual of the TPPnis given by
m
Max i=1 ai ui + j=1 bj vj
subject to,
ui + vj ≤ cij for all i = 1, .., m, j = 1, .., n.
Step 1: For the basic cells corresponding to x, cij = ui + vj .
Solve this set of m + n − 1 equations for ui and vj .
Since there are m + n − 1 equations and m + n, ui , vj ’s, fix
the value of any one variable and solve for the others.
Since any one of the (m + n) equations of the TP can be
removed, one can take the corresponding variable of the
dual say vn = 0 and can consider that variable as absent
from the equations cij = ui + vj .
This set of equations is obtained from yT B = cTB , where
yT = [u1 , ..., um , v1 , ..., vn−1 ].
We have m + n − 1 equations and m + n − 1 unknowns,
which can be easily solved.
Step 2: Check if this y is feasible for the dual, that is if
ui + vj ≤ cij for all the non basic cells. If yes, then stop.
The corresponding BFS is then optimal for the primal.
If not, then go to Step 3.
Step 3: Find the θ- loop in B ∪ {(p, q)}, where the cell
(p, q) is such that
cp,q − up − vq = min{cij − ui − vj : cij − ui − vj < 0}.
Step 4: Assign value +θ to cell (p, q) and alternately
assign +θ and −θ to all the cells in the θ- loop, so that the
sum of the allocations (+θ and −θ allocations) in each row
and column add up to zero.
Take
+θ = min{xij ∈ θ-loop : cell (i, j) is assigned value − θ}.
Find the new BFS say x0 where xij0 is either equal to xij ,
xij + θ or xij − θ.
Now (p, q) is a basic cell.
If xrs = min{xij ∈ θ-loop : (i, j) is assigned value − θ}, then
the variable xrs becomes a non basic variable in x0 .
If there is a tie for this minimum value, choose any one
as the leaving variable arbitrarily such that there are
(m + n − 1) basic cells in the next iteration.
Step 5: Go to Step 1.
If xpq is a non basic variable in a BFS then the
corresponding column in the simplex table
B −1 ãp,q = upq , is such that uk,pq = −1, 1, or 0
depending on whether the k th basic variable gets the
allocation θ, −θ or is not included in the θ-loop in
B ∪ {(p, q)}, respectively.
If xpq (or cell (p, q)) is the entering variable of the new
basis then according to the minimum ratio rule of simplex,
the leaving variable is (r , s) if
xrs = min{xij ∈ θ-loop : cell (i, j) is assigned value − θ}.
Example: Consider the following transportation problem
(P) with cij ’s, ai ’s (40,30,30) and dj ’s (30,50,20) as given
below:
2 5 1 40
1 4 5 30
1 5 3 30
30 50 20

Check whether the initial basic feasible solution x0 with

basic cells
B = {(1, 1), (1, 2), (2, 2), (2, 3), (3, 2)}, is optimal for (P) (by
taking v2 = 0, where v2 is the dual variable corresponding
to the second demand constraint).
Also find the optimal solution.
The BFS with B = {(1, 1), (1, 2), (2, 2), (2, 3), (3, 2)} as the
basic cells is given by
x11 = 30, x12 = 10, x22 = 10, x23 = 20, x32 = 30 as the
values of the basic variables.
The other ui , vj values are obtained by solving the
equations given by cij − ui − vj = 0 for the basic cells, that
is by solving the 5 equations given below:
c11 − u1 − v1 = 0, where c11 = 2
c12 − u1 − v2 = 0, where c12 = 5
c22 − u2 − v2 = 0, where c22 = 4
c23 − u2 − v3 = 0, where c23 = 5
c32 − u3 − v2 = 0, where c32 = 5.
On solving we get, u1 = 5, v1 = −3, u2 = 4, v3 = 1, u3 = 5.
The following table shows the cij − ui − vj values against
each cell, where we have taken v2 = 0 for easier
calculations.
Check that
c13 −u1 −v3 = 1−5−1 = −5, c21 −u2 −v1 = 1−4−(−3) = 0,
c31 − u3 − v1 = 1 − 5 − (−3) = −1,
c33 − u3 − v3 = 3 − 5 − 1 = −3, which is indicated in the
following table:

0 0 −5 40
0 0 0 30
−1 0 −3 30
30 50 20

Since all the cij − ui − vj values are not non negative, the
above table is not optimal.
The most negative value of cij − ui − vj is in cell (1, 3), so
this will be the entering variable of the new BFS.
The unique θ- loop in B ∪ {(1, 3)} is given by
{(1, 2), (2, 2), (2, 3), (1, 3)}.
Since (1, 3) is the entering variable, so if we give +θ
allocation to cell (1, 3) ( or value of x13 = +θ ) then
x12 = 10 − θ, x22 = 10 + θ, x23 = 20 − θ.
x13 = 10 is in the basis of the new BFS and x12 leaves the
basis.
New B = {(1, 1), (1, 3), (2, 2), (2, 3), (3, 2)} and the values
of the basic variables are given by:
x11 = 30, x13 = 10, x22 = 20, x23 = 10, x32 = 30.
If we take u1 = 0, then solving for cij − ui − vj = 0 for the
basic cells, that is by solving the 5 equations given below,
c11 − u1 − v1 = 0, where c11 =2
c13 − u1 − v3 = 0, where c13 =1
c23 − u2 − v3 = 0, where c23 =5
c22 − u2 − v2 = 0, where c22 =4
c32 − u3 − v2 = 0, where c32 = 5.
we get
v1 = 2, v2 = 0, v3 = 1, u2 = 4, u3 = 5.
c21 − u2 − v1 = 1 − 4 − 2 = −5, c12 − u1 − v2 = 5 − 0 − 0 = 5,
c31 − u3 − v1 = 1 − 5 − 2 = −6,
c33 − u3 − v3 = 3 − 5 − 1 = −3.
The following table gives the cij − ui − vj values for the
above BFS with
B = {(1, 1), (1, 3), (2, 3), (2, 2), (3, 2)}.

0 5 0 40
−5 0 0 30
−6 0 −3 30
30 50 20

The entering variable for the new BFS is x31 .

The unique θ- loop in B ∪ {(3, 1)} is given by
{(3, 1), (3, 2), (2, 2), (2, 3), (1, 3), (1, 1)}.
(3, 1) is the entering variable, so if we give +θ allocation to
cell (3, 1) ( or value of x31 = +θ ) then x11 = 30 − θ,
x13 = 10 + θ, x23 = 10 − θ, x22 = 20 + θ, x32 = 30 − θ.
So θ = 10.
The entering variable for the new BFS is x31 = 10 and x23
is the leaving variable.
The values of the basic variables in the new BFS is given
by x11 = 20, x13 = 20, x22 = 30, x31 = 10, x32 = 20.
The basic set of cells is given by
B = {(1, 1), (1, 3), (2, 2), (3, 1), (3, 2)}.
We take u1 = 0, then by solving the 5 equations given
below:
c11 − u1 − v1 = 0, where c11 = 2
c13 − u1 − v3 = 0, where c13 = 1
c22 − u2 − v2 = 0, where c22 = 4
c31 − u3 − v1 = 0, where c31 = 1
c32 − u3 − v2 = 0, where c32 = 5.
Check that v1 = 2, v2 = 6, v3 = 1, u2 = −2, u3 = −1.
Check that c23 − u2 − v3 = 5 − (−2) − 1 = 6, c21 − u2 − v1 =
1 − (−2) − 2 = 1, c12 − u1 − v2 = 5 − 0 − 6 = −1,
c33 − u3 − v3 = 3 − (−1) − 1 = 3.
The following table gives the cij − ui − vj values for the
above BFS with
B = {(1, 1), (1, 3), (2, 2), (3, 1), (3, 2)}.

0 −1 0 40
1 0 6 30
0 0 3 30
30 50 20

The entering variable is x12 .

The θ-loop is {(3, 1), (3, 2), (1, 2), (1, 1)}.
If x12 = +θ ) then x11 = 20 − θ, x31 = 10 + θ, x32 = 20 − θ.
Take θ = 20.
Any one of x11 or x32 can be the leaving variable.
Let x32 leave the basis.
If we take u1 = 0, then by solving the 5 equations given
below:
c11 − u1 − v1 = 0, where c11 = 2
c13 − u1 − v3 = 0, where c13 = 1
c22 − u2 − v2 = 0, where c22 = 4
c31 − u3 − v1 = 0, where c31 = 1
c12 − u1 − v2 = 0, where c12 = 5.
we get
v1 = 2, v2 = 5, v3 = 1, u2 = −1, u3 = −1.
c23 − u2 − v3 = 5 − (−1) − 1 = 5, c21 − u2 − v1 =
1 − (−1) − 2 = 0, c32 − u3 − v2 = 5 − (−1) − 5 = 1,
c33 − u3 − v3 = 3 − (−1) − 1 = 3.
Since cij − ui − vj ≥ 0 for all i, j,
the above BFS is optimal and the optimal value is given by:
c11 x11 + c12 x12 + c13 x13 + c22 x22 + c31 x31 =
2 × 0 + 5 × 20 + 1 × 20 + 4 × 30 + 1 × 30 = 270.