DS424 – Multi-objective Programming- Prof. Dr. Tarek H.M.

Abou-El-Enien

Notes on Lesson (6)

The Constraint Method (Ɛ- Constraint Method)

- The multi-objective programming problem can be represented as:

𝑀𝑎𝑥𝑖𝑚𝑖𝑧𝑒 [𝑓1 (𝑋), 𝑓2 (𝑋), … , 𝑓𝑘 (𝑋)]

𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 } (1)
𝑋∈𝕄

where 𝑋 is an 𝑛-dimensional decision variable vector. The problem consists of 𝑛 decision variables, 𝑚
constraints and 𝑘 objectives. Any or all of the functions may be nonlinear. In literature this problem is
often referred to as a vector Maximum problem (VMP), or vector optimization problem (VOP).

The algorithm:
Step (1): Construct a payoff table.
(a) Solve p individual maximization problems to find the optimal solution for each of the p
objectives. Call the solutions that maximizes objective k, 𝒙𝑘 = (𝑥1𝑘 , … . . , 𝑥𝑛𝑘 ). If there are
alternative optima for any of these problems, then choose those solutions from among the
alternative optima that are noninferior.
(b) Compute the value of each objective at each of the p optimal solutions : 𝑓1 (𝒙𝒌 ), 𝑓2 (𝒙𝒌 ), ….,
𝑓𝑝 (𝒙𝒌 ), k=1,2,….,p. This gives us p values for each of the p objectives.
(c) Array the p values of each of the p objectives in a table in which the rows correspond to
𝒙1 , 𝒙2 , … … , 𝒙𝑝 and the columns are labeled by the objectives, (see the following table).
𝑓1 (𝒙𝒌 ) 𝑓2 (𝒙𝒌 ) ……. 𝑓𝑝 (𝒙𝒌 )
𝒙1 𝑓1 (𝒙𝟏 ) 𝑓2 (𝒙𝟏 ) ……. 𝑓𝑝 (𝒙𝟏 )
𝒙2 𝑓1 (𝒙𝟐 ) 𝑓2 (𝒙𝟐 ) …… 𝑓𝑝 (𝒙𝟐 )
……… …….. ……. ……. ……..
𝒙𝑝 𝑓1 (𝒙𝒑 ) 𝑓2 (𝒙𝒑 ) …….. 𝑓𝑝 (𝒙𝒑 )

Page 1 of 5
 DS424 – Multi-objective Programming- Prof. Dr. Tarek H.M. Abou-El-Enien

(d) Find the largest number in the kth column, denote it by Mk. Find the smallest number in the
kth column, denote it by nk. Do this for k=1,2,….,p.
Step (2):
Convert the following multi-objective programming problem:

𝑀𝑎𝑥𝑖𝑚𝑖𝑧𝑒 [𝑓1 (𝑋), 𝑓2 (𝑋), … , 𝑓𝑘 (𝑋)]

𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 } (1)
𝑋∈𝕄

to the following corresponding constrained problem:

𝑀𝑎𝑥𝑖𝑚𝑖𝑧𝑒𝑓ℎ (𝒙)
𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜
} (2)
𝑋∈𝕄
𝑓𝑘 (𝒙) ≥ 𝜀𝑘 , 𝑘 = 1,2, … . , ℎ − 1, ℎ + 1, … . , 𝑝
Step (3):
The nk and Mk from step 1 represent a range for objective kin the noninferior set:
nk ≤𝑓𝑘 ≤Mk . This range applies as well to 𝜀𝑘 , the right-hand side of the constraint on
objective k. Choose the number of different values of 𝜀𝑘 that will be used in the generation of
noninferior solutions. Call this r.
Step (4):
Solve the constrained problem set up in step (2) for every combination of values for the
𝜀𝑘 , 𝑘 = 1,2, … . , ℎ − 1, ℎ + 1, … . , 𝑝, where

𝜀𝑘 = nk + [t / (r-1)] (Mk - nk), t=0,1,2,….,(r-1)

Example:
Maximize Z(x1,x2) =[Z1(x1,x2) , Z2(x1,x2)]
S.T.
-x1+x2 ≤3,
x1+x2 ≤6,
x1 ≤3,

Page 2 of 5
 DS424 – Multi-objective Programming- Prof. Dr. Tarek H.M. Abou-El-Enien

x2 ≤4,
x1, x2 ≥0,
where
Z1(x1,x2) = 5x1 - 2x2
Z2(x1,x2) = -x1 + 4x2 .
Solution:
Step(1)
Maximize Z1(x1,x2) = 5x1 - 2x2
S.T.
-x1+x2 ≤3,
x1+x2 ≤6,
x1 ≤3,
x2 ≤4,
x1, x2 ≥0,

𝑋1∗ = (𝑥11∗ , 𝑥21∗ ) = (6 , 0), 𝑍1∗ (𝑋1∗ ) = 30, 𝑍2 (𝑋1∗ ) = −6

Maximize Z2(x1,x2) = -x1 + 4x2

S.T.
-x1+x2 ≤3,
x1+x2 ≤6,
x1 ≤3,
x2 ≤4,
x1, x2 ≥0,

𝑋 2∗ = (𝑥12∗ , 𝑥22∗ ) = (1 , 4), 𝑍1 (𝑋 2∗ ) = 15, 𝑍2∗ (𝑋 2∗ ) = −3

Page 3 of 5
 DS424 – Multi-objective Programming- Prof. Dr. Tarek H.M. Abou-El-Enien

Payoff Table:
k 𝑋𝑘 𝑍1 (𝑋 𝑘 ) = 5𝑥1𝑘 − 2𝑥2𝑘 𝑍2 (𝑋 𝑘 ) = −𝑥1𝑘 + 4𝑥2𝑘
1 𝑋1 = (6 , 0) 𝑀1 = 30 𝑛2 = −6
2 𝑋 2 = (1 , 4) 𝑛1 = −3 𝑀2 = 15

Step(2):
Maximize Z1(x1,x2) = 5x1 - 2x2
S.T.
-x1+x2 ≤3,
x1+x2 ≤6,
x1 ≤3,
x2 ≤4,
x1, x2 ≥0,
Z2(x1,x2) = -x1 + 4x2 ≥ 𝜀2 .
Step(3):
𝑛2 ≤ 𝜀2 ≤ 𝑀2 ,
−6 ≤ 𝜀2 ≤ 15 ,
r = 4 (The number of different values for 𝜀2 to be used in generating noninferior solutions)
(Also, the number of values of functions in payoff table)
Step(4):
1
𝜀2 = −6 + 3 𝑡[15 − (−6)], 𝑡 = 0,1,2,3

= −6 + 7𝑡, 𝑡 = 0,1,2,3
𝜀2 = −6 ,1 ,8 ,15

Page 4 of 5
 DS424 – Multi-objective Programming- Prof. Dr. Tarek H.M. Abou-El-Enien

Maximize Z1(x1,x2) = 5x1 - 2x2

S.T.
-x1+x2 ≤3,
x1+x2 ≤6,
x1 ≤3, -6

x2 ≤4,
x1, x2 ≥0,
1
Z2(x1,x2) = -x1 + 4x2 ≥ 𝜀2 =

15

Solve the above problem for 𝜀2 = −6 ,1 ,8 ,15, thus, we get

𝑋∗ 𝑍1 𝑍2
(6 , 0) -3 15
(4.8 , 3.2) 17.6 8
(6 , 1.75) 26.5 1
(6 , 0) 30 -6

Page 5 of 5