Scribd
Upload
126 views

Linear Programming 1 PDF

The document discusses linear programming problems and their optimal solutions, noting that the feasible region and objective function must be convex for there to be a global maximum. It provides an example of using linear programming to maximize revenue from two crops given constraints on available land, water, and minimum production, finding the optimal solution that uses all resources while maximizing revenue. Necessary conditions for optimality including stationarity, inequality of Lagrange multipliers, and curvature are examined for the example problem.

Uploaded by

mike
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
126 views

Linear Programming 1 PDF

The document discusses linear programming problems and their optimal solutions, noting that the feasible region and objective function must be convex for there to be a global maximum. It provides an example of using linear programming to maximize revenue from two crops given constraints on available land, water, and minimum production, finding the optimal solution that uses all resources while maximizing revenue. Necessary conditions for optimality including stationarity, inequality of Lagrange multipliers, and curvature are examined for the example problem.

Uploaded by

mike
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

MASSACHUSETTS INSTITUTE OF TECHNOLOGY

Department of Civil and Environmental Engineering

1.731 Water Resource Systems

Lecture 7, Linear Programming Overview, Sept. 28, 2006

Linear Programming Problems (LPP)


Objectives and constraints are all linear functions of decision variables:
Maximize F ( x1 , x 2 ,..., x n ) = c j x j
x1 ,x 2 ,...,x n
such that:
g Ti ( x ) = Aij x j bi i = 1,..., mT Technological constraints (may be
equalities or inequalities)

g Ni ( x ) = xi m 0 i = mT + 1,..., mT + m N Non-negativity constraints

Total constraints = mT +mN = m

Optimal Solutions of Linear Programming Problems:


For LPP:
Linear objective and constraint functions are both convex and concave so:
Feasible region F for LP is convex (i.e. constructed from convex functions gi(x)0)
Objective function for LP is concave
Therefore:
A candidate LPP solution x* that is a local maximum is also a global maximum.

To check whether x* is a local/global maximum use necessary conditions:

Focus on constraints that are active at x*:

G *Aij x *j b *Ai = 0 i C ( x*)

Row i of G *A = Aij if i is a technological constraint


Row i of b*Ai = bi if i is a technological constraint
Row i of G *A = -ij if i is a non-negativity constraint
Row i of b*Ai = 0 if i is a non-negativity constraint

1
1. Feasibility
x* is chosen to be feasible

2. Stationarity
If x* is a local maximum then:
F ( x*) [c j x j ] [G *Aij x j + b *Ai ]
= = c j = i = i G *Aij
x j x j x j

For LPP stationarity condition reduces to a set of linear equations in unknown is


i G *Aij = c j
The stationarity condition is satisfied if this set of linear equations is consistent so:
* = Rank [G *Aij ] = Rank [G *Aij c j ]

There are four ways this can occur:


1). Corner solution: x* lies at intersection of n linearly independent constraints.
* = m*A = n .
2). Trivial interior solution: occurs only if cj =0.
* = m*A = 0
3). Non-corner boundary solution: x* lies along a boundary but not at a corner.
* = m*A < n
4). Degenerate solution: Constraints are linearly dependent (i.e. number of constraints
exceeds rank of G *Aij ).
* < m*A

3. Inequality Lagrange multiplier


If x* is a local maximum then:
i 0 i C ( x*)
In case 1) above there will be only one solution that satisfies this condition.

4. Curvature
In LLP curvature condition applies for any x* since Lagrangian Hessian is always zero.
2 L( x*, ) 2 [c j x j i (G *Aij x j b *Ai )]
Wkl = Z ik Z lj = Z ik Z lj = 0
x j x k x j x k
x = x*

2
Crop Allocation Example
Problem is to maximize revenue from two crops, given constraints on available land and water
and on minimum total crop grown.

Decision variables:
3
x1 = mass of crop 1 grown (tonnes = 10 kg)
3
x2 = mass of crop 2 grown (tonnes = 10 kg)

Maximize 6 x1 + 11x2
x1,x2
such that :
2 x1 + x2 104 Water constraint (103 m3/season)
x1 + 2 x2 76 Land constraint (ha)
- x1 x2 25 Minimum production constraint (tonnes)
- x1 0 x1 non - negativity constraint
- x2 0 x2 non - negativity constraint

Objective, right-hand side, and technological coefficients:


cj -- Crop values ($/tonne)
b1 -- Water available (m3/season)
b2 -- Land available (ha)
3 3 -1
A1j -- Water requirements (10 m /(season tonne) ) = (unit water requirement in 10
m/season)/(yield in tonnes/ha)
-1
A2j -- Land requirement (ha/tonne) = (yield in tonnes/ha)

x2

(0,38)
Increasing F(x)
g4(x) g2(x) land
non-neg

(44,16)
(0,25)
g3(x) F
production
g1(x) water

(25,0) g5(x) non-neg (52,0) (76,0) x1

3
Pairs of constraints active at the 5 corners of the feasible solution are all linearly independent
(i.e. corresponding G *Aij have rank * = m*A = n = 2 ).

So i G *Aij = c j is consistent and stationarity condition is satisfied at each of these corner


points.

There are no interior, non-corner boundary, or degenerate solutions for this example.
So we need only consider the Lagrange multipliers at the 5 corner solutions:
Candidate Active Lagrange Multipliers
Solution Constraints
(25, 0) 3, 5 3 = -6 5 = -5
(52, 0) 5, 1 1 = +3 5 = -8
(44, 16) 1, 2 1 = +1/3 2 = +16/3
(0, 38) 2, 4 2 = +11/2 4 = -1/2
(0, 25) 4, 3 4 = -11 3 = +5

x* = (44, 16) is the local/global maximum since it is the only corner solution with positive
Lagrange multipliers for all active constraints.

In this problem the optimum crop allocation mixes Crop 1 and Crop 2 in a way that uses all
available land and water while giving maximum revenue.

It is possible to generate a non-corner (non-unique) boundary solution to this problem by


changing the objective function to F(x) = 6x1 + 12x2. Then * = m*A < n
( * = 1, m*A = 1, n = 2 ). The objective function contours are parallel to g1(x) and any feasible
solution along g1(x) is local/global maximum.

It is possible to generate a degenerate solution to this problem by changing the water constraint
to g1(x) = 2x1 + x2. 152. Then * < m*A ( * = 2, n = 2, m*A = 3) at the new corner (76, 0).
There is not a unique set of is satisfying stationarity condition at this corner.
So the inequality Lagrange multiplier condition cannot be checked.

You might also like