Solving Linear Programming Problems:
The Simplex Method
BENG 2213B
The Essence of the Simplex Method
Algebraic procedure
– Underlying concepts are geometric
Revisit Wyndor example
– Figure 4.1 shows constraint boundary lines
• Points of intersection are corner-point solutions
• Five points on corners of feasible region are CPF solutions
Adjacent CPF solutions
– Share a constraint boundary
Optimality test
– If a CPF solution has no adjacent CPF solution that is better (as measured
by Z):
• It must be an optimal solution
Solving the example with the simplex method
– Choose an initial CPF solution (0,0) and decide if it is optimal
– Move to a better adjacent CPF solution
– Iterate until an optimal solution is found
Setting Up the Simplex Method
First step: convert functional inequality constraints into equality constraints
– Done by introducing slack variables
– Resulting form known as augmented form
– Example: constraint 𝑥1 ≤ 4 is equivalent to
𝑥1 + 𝑥3 = 4 𝑎𝑛𝑑 𝑥3 ≥ 0
Augmented solution
– Solution for the original decision variables augmented by the slack
variables
Basic solution
– Augmented corner-point solution
Basic feasible (BF) solution
– Augmented CPF solution
Properties of a basic solution
– Each variable designated basic or nonbasic
– Number of basic variables equals number of functional constraints
– The nonbasic variables are set equal to zero
– Values of basic variables obtained as simultaneous solution of system of
– equations
– If basic variables satisfy nonnegativity constraints, basic solution is a BF
solution
The Simplex Method in Tabular Form
Tabular form
– Records only the essential information:
• Coefficients of the variables
• Constants on the right-hand sides of the equations
• The basic variable appearing in each equation
– Example shown in Table 4.3 on next slide
Summary of the simplex method
– Initialization
• Introduce slack variables
– Optimality test
• Optimal if and only if every coefficient in row 0 is nonnegative
– Iterate (if necessary) to obtain the next BF solution
• Determine entering and leaving basic variables
• Minimum ratio test
Tie Breaking in the Simplex Method
Tie for the entering basic variable
– Decision may be made arbitrarily
Tie for the leaving basic variable
– Matters theoretically but rarely in practice
– Choose arbitrarily
Condition of no leaving basic variable
– Z is unbounded
– Indicates a mistake has been made
Multiple optimal solutions
– Simplex method stops after one optimal BF solution is found
– Often other optimal solutions exist and would be meaningful choices
– Method exists to detect and find other optimal BF solutions
Adapting to Other Model Forms
Simplex method adjustments
– Needed when problem is not in standard form
– Made during initialization step
Artificial-variable technique
– Dummy variable introduced into each constraint that needs one
– Becomes initial basic variable for that equation
Adapting to Other Model Forms
Types of nonstandard forms
– Equality constraints
– Negative right-hand sides
– Functional constraints in greater-than-or-equal-to form
– Minimizing Z
– No feasible solutions
– Constructing an artificial feasible solution may lead to a false optimal
solution
– Artificial-variable technique provides a way to indicate whether this is the
case
– Variables are allowed to be negative
– Example: negative value indicates a decrease in production rate
– Negative values may have a bound or no bound
Postoptimality Analysis
Simplex method role
Postoptimality Analysis
Reoptimization
– Alternative to solving the problem again with small changes
– Involves deducing how changes in the model get carried along to the final
simplex tableau
– Optimal solution for the revised model:
• Will be much closer to the prior optimal solution than to an initial
BF solution constructed the usual way
Shadow price
– Measures the marginal value of resource i
– The rate at which Z would increase if more of the resource could be ma
de available
– Given by the coefficient of the ith slack variable in row 0 of the final si
mplex tableau
Sensitivity analysis
- Purpose: to identify the sensitive parameters
These must be estimated with special care
- Can be done graphically if there are just two variables
Postoptimality Analysis
Parametric linear programming
– Study of how the optimal solution changes as many of the parameters c
hange simultaneously over some range
– Used for investigation of trade-offs in parameter values
