FORELASNING 3 (Chapter 4.4 - 4.

7 )

Solving linear programming problems: the Simplex Method

Denition 9 For any LP with m constraints, two feasible basic solutions are said to be adjacent (nrliggande) if they have m 1 basic variables in common. a Notice that The feasible basic solutions The extreme points, or corner points (hrnpunkter) of its o feasible region. Moving from one feasible basic solution to an adjacent feasible basic solution involves switching one variable from basic to nonbasic and vice versa for one other variable. How can we improve the objective function value when a feasible basic solution is moved to an adjacent one? We calculate the reduced cost c associated with point x(k) according to c= where d is the moving direction. If > 0 c If < 0 c = = direction d is an ascent direction. direction d is an descent direction. f (x(k) )T d = cT d

Example 6 (Example in Chapter 4.5 in the book, p.102) max s.t. z = 30x1 + 20x2 2x1 + x2 100 x1 + x2 80 x1 x1 , x2 40 0

4.4

The simplex method

The following are the general description of how the simplex method solves LPs in a maximum problem: Step 1 Find a feasible basic solution the LP in standard form (i.e., a corner point of the feasible region). Step 2 Determine whether the current feasible basic solution is an optimal solution to the LP. If it is , the procedure terminates. Otherwise, go to Step 3. Step 3 Find an adjacent feasible basic solution that has a larger value of the objective function. Use this new feasible basic solution as the current feasible basic solution, and then return to Step 2. The general description of the Simplex Algorithm can be found in Chapter 4.6 in the book(p.107-109).

4.5

The simplex method in tabular form

Example 7 The example in Chapter 4.7 in the book(p.109). Note that the coecients in the objective function row are the reduced cost with a reversed sign. Summary of the Simplex algorithm for tabular form Step 0 Initialization Convert the LP to standard form and nd an initial feasible basic solution x(0) . Set k = 0 Step 1 Optimality test If the reduced costs for all nonbasic variables satisfy

cj 0 j (for minimization problem) (i.e. all nonbasic variables have non-positive ( 0) coecients in the objective function row)

cj 0 j (for maximization problem) (i.e. all nonbasic variables have non-negative ( 0) coecients in the objective function row), then the feasible basic solution x(k) is optimal. Otherwise, go to Step 2.

 The LP has alternative optimal solutions (multiple optimal solutions) if, at least, one of the reduced costs of the nonbasic variable equals to zero. Step 2 Determine the entering basic variable (inkommande basvariabel) Choose the nonbasic variable xp that has cp = min {j | cj < 0} (for min problem) (i.e. the most positive coecient in objective function row) c
j

cp = max {j | cj > 0} (for max problem) (i.e. the most negative coecient in objective function row) c
j

to enter the basis (break tie arbitrarily). Step 3 Determine the leaving basic variable (utg aende basvariabel) Compute the ratio s i b b = min { | aip > 0} i asp aip

where i is the right-hand-side coecient of equation i and aip is the coecient of entering basic b variable (inkommande basvariabel) xp in row i. Then the basic variable in row s is the leaving basic variable (utg aende basvariabel). Break tie arbitrarily. If the coecients aip < 0 for all i, then the LP has unbounded solution. Step 4 Determine the new feasible basic solution x(k+1) by Gaussian Elimination. Update k := k + 1 and go to Step 1

Example 8

Consider the following LP max s.t. z = 3x1 + 2x2 x1 4 2x2 12 3x1 + 2x2 18 x1 , x2 0

The standard form of the problem is: max s.t. z = 3x1 + 2x2 x 1 + s1 = 4 2x2 + s2 = 12 3x1 + 2x2 + s3 = 18 x1 , x2 , s1 , s2 , s3 0

The optimal solution of the problem is given in the following nal simplex tableau: Basvar z x1 s2 x2 z 1 0 0 0 x1 0 1 0 0 x2 0 0 0 1 s1 0 1 3 3 2 s2 0 0 1 0 s3 1 0 1
1 2

b 18 4 6 3

x = (x , x ) = (4, 3), 1 2

z = 18

This LP has alternative optimal solutions (multiple optimal solutions) since the reduced cost of nonbasic variable s1 is zero. What happens if we enter s1 into the basis? The resulting tableau is given in the following. Basvar z x1 s1 x2 z 1 0 0 0 x1 0 1 0 0 x2 0 0 0 1 s1 0 0 1 0 s2 0 1 3
1 3 1 2

s3 1
1 3

b 18 2 2 6

1 3 0

An alternative optimal solution: x = (x , x ) = (2, 6), 1 2

z = 18,

The optimal solutions of the problem can be written in the form x = (4, 3)T + (1 )(2, 6)T , where 0 1. 4