Вы находитесь на странице: 1из 13

MS-C2105 Introduction to Optimization

Lecture 2: The LP model - an example and its graphical solution

H. Ehtamo / M. Mattila
Systems Analysis Laboratory
Aalto University

December 2017
Example: Paint factory

A paint factory produces interior and exterior paint from raw


materials M1 and M2.
The maximum demand of interior paint is 2 tons/day.
The maximum demand of interior paint is also at most 1 ton/day
more than that of exterior paint.
The paint factory wants to maximize its total daily profit.

MS-C2105 Introduction to Optimization 2/13


H. Ehtamo / M. Mattila
Systems Analysis Laboratory
Aalto University
Variables and objective function

material (ton)/paint (ton)


exterior paint interior paint daily availability
material M1 6 4 24
material M2 1 2 6
profit (1000 e/ton) 5 4

Decision variables x1 , x2 : exterior/interior paint produced


(tons/day)
Objective function z = 5x1 + 4x2 (maximize profit)

MS-C2105 Introduction to Optimization 3/13


H. Ehtamo / M. Mattila
Systems Analysis Laboratory
Aalto University
Constraints

material (ton)/paint (ton)


exterior paint interior paint daily availability
material M1 6 4 24
material M2 1 2 6
profit (1000 e/ton) 5 4

Material M1: 6x1 + 4x2 ≤ 24


Material M2: x1 + 2x2 ≤ 6
Demand: x2 ≤ x1 + 1
Demand: x2 ≤ 2
Non-negativity: x1 , x2 ≥ 0

MS-C2105 Introduction to Optimization 4/13


H. Ehtamo / M. Mattila
Systems Analysis Laboratory
Aalto University
The model

max z = 5x1 + 4x2


st. 6x1 + 4x2 ≤ 24 (1)
x1 + 2x2 ≤ 6 (2)
−x1 + x2 ≤ 1 (3)
x2 ≤ 2 (4)
x1 , x2 ≥ 0

The objective function and constraints are linear, so the example is


a linear optimization or linear programming (LP) model.

MS-C2105 Introduction to Optimization 5/13


H. Ehtamo / M. Mattila
Systems Analysis Laboratory
Aalto University
Graphical solution

The feasible region of the


LP model (yellow) consists
of the points that satisfy all
of the constraints (1)-(4)
and the non-negativity
constraints.

MS-C2105 Introduction to Optimization 6/13


H. Ehtamo / M. Mattila
Systems Analysis Laboratory
Aalto University
Graphical solution

The gradient of the objective


function:
∇z = [ ∂∂xz1 , ∂∂xz2 ]T = [5, 4]T .
The gradient is orthogonal to
any line
z = 5x1 + 4x2 = constant.
The gradient indicates the
direction in which the
objective function z increases.
The optimum is the corner
point C.

MS-C2105 Introduction to Optimization 7/13


H. Ehtamo / M. Mattila
Systems Analysis Laboratory
Aalto University
Graphical solution

If the objective function is


minimized, we follow the
negative gradient −∇z to find
the minimum.
Note that minimizing z is
equivalent to maximizing −z.
In this picture, min x1 + 4x2 is
equivalent to max −x1 − 4x2 .

MS-C2105 Introduction to Optimization 8/13


H. Ehtamo / M. Mattila
Systems Analysis Laboratory
Aalto University
Graphical sensitivity analysis

The corner point C remains


optimal as long as the slope
of z lies between those of
constraints (1) and (2).
If z = c1 x1 + c2 x2 , this
condition is 46 ≤ cc12 ≤ 21 , or
1 c1
equivalently, 2 ≤ c2 ≤ 64 .

MS-C2105 Introduction to Optimization 9/13


H. Ehtamo / M. Mattila
Systems Analysis Laboratory
Aalto University
Value of resource M1 at optimal solution

Let the availability of raw


material M1 vary between
20-36 tons.
The optimum at corner point
C ’slides’ along line DG.
If the availability of M1 is not
on this range, the intersection
of (1) and (2) is no longer in
the feasible region and the
solution is changed.

MS-C2105 Introduction to Optimization 10/13


H. Ehtamo / M. Mattila
Systems Analysis Laboratory
Aalto University
Value of resource M1 at optimal solution

Denote M1 = 6x1 + 4x2 ;


D = (2, 2); G = (6, 0).
Then define
z (D )−z (G)
y1 = ∆∆M1
z
= M1(D )−M1(G)
.
5·2+4·2−(5·6+4·0)
= 20−36
−12000 e
= −16 ton = 750 e/ton.
A change of 1 ton in M1, in
the interval 20 ≤ M1 ≤ 36,
makes a change of 750 e to
the optimal z-value.

MS-C2105 Introduction to Optimization 11/13


H. Ehtamo / M. Mattila
Systems Analysis Laboratory
Aalto University
Value of resource M2 in optimal solution

Similarly, now the availability


of M2 = x1 + 2x2 varies.
The optimum at corner point
C moves along line HB.
We may calculate
y2 = ∆∆M2
z
= M2z ((H
H )−z (B )
)−M2(B )
=500 e/ton.
This holds in the region
4 ≤ M2 ≤ 20/3.

MS-C2105 Introduction to Optimization 12/13


H. Ehtamo / M. Mattila
Systems Analysis Laboratory
Aalto University
Dual variables

We have obtained y1 = 750 e/ton and y2 = 500 e/ton.


For all other constraints (including non-negativity constraints), we
get y3 , y4 , y5 , y6 = 0. Thus they are not active constraints at the
optimal solution.
Coefficients yi are called sensitivity coefficients, or dual variables
(more in Lecture 4).
In Lecture 9 we show that yi is the Lagrange variable associated
with constraint i.

MS-C2105 Introduction to Optimization 13/13


H. Ehtamo / M. Mattila
Systems Analysis Laboratory
Aalto University

Вам также может понравиться