The Simplex Method. Standard form (max) z c T x = 0 such that Ax = b.

Transcription

1 The Simplex Method Standard form (max) z c T x = 0 such that Ax = b.

2 The Simplex Method Standard form (max) z c T x = 0 such that Ax = b. Build initial tableau. z c T 0 0 A b

3 The Simplex Method Standard form (max) z c T x = 0 such that Ax = b. Build initial tableau. z c T 0 0 A b Find an initial BFS.

4 The Simplex Method Standard form (max) z c T x = 0 such that Ax = b. Build initial tableau. Find an initial BFS. z c T 0 0 A b Look at Row 0 for neg coeffs:

5 The Simplex Method Standard form (max) z c T x = 0 such that Ax = b. Build initial tableau. Find an initial BFS. z c T 0 0 A b Look at Row 0 for neg coeffs: 1. Choose the column most negative coef.

6 The Simplex Method Standard form (max) z c T x = 0 such that Ax = b. Build initial tableau. Find an initial BFS. z c T 0 0 A b Look at Row 0 for neg coeffs: 1. Choose the column most negative coef. 2. Perform a ratio test by taking RHS/Lead Coeff.

7 The Simplex Method Standard form (max) z c T x = 0 such that Ax = b. Build initial tableau. Find an initial BFS. z c T 0 0 A b Look at Row 0 for neg coeffs: 1. Choose the column most negative coef. 2. Perform a ratio test by taking RHS/Lead Coeff. Exceptions: Ignore zeros and neg coeffs.

8 The Simplex Method Standard form (max) z c T x = 0 such that Ax = b. Build initial tableau. Find an initial BFS. z c T 0 0 A b Look at Row 0 for neg coeffs: 1. Choose the column most negative coef. 2. Perform a ratio test by taking RHS/Lead Coeff. Exceptions: Ignore zeros and neg coeffs. Choose the row with the smallest ratio.

9 The Simplex Method Standard form (max) z c T x = 0 such that Ax = b. Build initial tableau. Find an initial BFS. z c T 0 0 A b Look at Row 0 for neg coeffs: 1. Choose the column most negative coef. 2. Perform a ratio test by taking RHS/Lead Coeff. Exceptions: Ignore zeros and neg coeffs. Choose the row with the smallest ratio. 3. Pivot using the column/row we found.

10 The Simplex Method Standard form (max) z c T x = 0 such that Ax = b. Build initial tableau. Find an initial BFS. z c T 0 0 A b Look at Row 0 for neg coeffs: 1. Choose the column most negative coef. 2. Perform a ratio test by taking RHS/Lead Coeff. Exceptions: Ignore zeros and neg coeffs. Choose the row with the smallest ratio. 3. Pivot using the column/row we found. 4. Repeat.

11 Stopping Criteria 1 Stop if all Row 0 coeffs are non-negative. The current solution is optimal and unique.

12 Stopping Criteria 1 Stop if all Row 0 coeffs are non-negative. The current solution is optimal and unique. Example Final Tableau: z x 1 x 2 s 1 s 2 rhs

13 Stopping Criteria 1 Stop if all Row 0 coeffs are non-negative. The current solution is optimal and unique. Example Final Tableau: z x 1 x 2 s 1 s 2 rhs Maximizer: x 1 = 2, x 2 = 3 and s 1 = s 2 = 0. Notice also the top row: z = 27 3s 1 s 2

14 From Wed: Unbounded LP Here was the final tableau for an unbounded region and an unbounded LP: z x 1 x 2 x 3 s 1 s 2 s 3 rhs s x x

15 From Wed: Unbounded LP Here was the final tableau for an unbounded region and an unbounded LP: z x 1 x 2 x 3 s 1 s 2 s 3 rhs s x x Translation: max z = 50 x 1 3s 2 + 2s 3

16 From Wed: Unbounded LP Here was the final tableau for an unbounded region and an unbounded LP: z x 1 x 2 x 3 s 1 s 2 s 3 rhs s x x Translation: max z = 50 x 1 3s 2 + 2s 3 with x = s

17 Example 3, A Little Foreshadowing Build the initial tableau. max z = x 1 +x 2 st 2x 1 +x 2 4 x 1 +2x 2 = 6 x 1, x 2 0

18 Example 3, A Little Foreshadowing Build the initial tableau. max z = x 1 +x 2 st 2x 1 +x 2 4 x 1 +2x 2 = 6 x 1, x 2 0 z x 1 x 2 e 1 rhs Not standard form... No initial BFS.

19 Choosing x 1, x 2 as the basic variables, we can perform the row reduction using an inverse matrix: [ ] 2 1 B = 1 2

20 Choosing x 1, x 2 as the basic variables, we can perform the row reduction using an inverse matrix: [ ] 2 1 B = 1 2 B 1 [A b] = 1 3 [ ] [ ] =

21 Choosing x 1, x 2 as the basic variables, we can perform the row reduction using an inverse matrix: [ ] 2 1 B = 1 2 B 1 [A b] = 1 [ ] [ ] = [ 1 0 2/3 ] 2/ /3 8/3

22 Choosing x 1, x 2 as the basic variables, we can perform the row reduction using an inverse matrix: [ ] 2 1 B = 1 2 B 1 [A b] = 1 [ ] [ ] = [ 1 0 2/3 ] 2/ /3 8/3 The new tableau: z x 1 x 2 e 1 rhs /3 10/ /3 2/ /3 8/3

23 Bring e 1 into the set of BV (take x 2 out): z x 1 x 2 e 1 rhs Optimal?

24 Bring e 1 into the set of BV (take x 2 out): z x 1 x 2 e 1 rhs Optimal? Yes.

25 Back to the Example z x 1 x 2 e 1 rhs What if we had chosen x 2 and e 1 as our basic variables?

26 Back to the Example z x 1 x 2 e 1 rhs What if we had chosen x 2 and e 1 as our basic variables? z x 1 x 2 e 1 rhs 1 1/ / / Conclusion?

27 Back to the Example z x 1 x 2 e 1 rhs What if we had chosen x 2 and e 1 as our basic variables? z x 1 x 2 e 1 rhs 1 1/ / / Conclusion? Not every choice of basic variables will be feasible!

28 Example 4 max x 6x 1 + 5x 2 s.t. x 1 + x 2 5 3x 1 + 2x 2 12 x 1, x Solution is (2, 3)

29 Slight Change: max x 6x 1 + 4x 2 s.t. x 1 + x 2 5 3x 1 + 2x 2 12 x 1, x /3 1 1/ /3 0 1/3 4 New solution is (4, 0)

30 We can bring in x 2 with no change to z: /3 1 1/ /3 0 1/ New solution is (2, 3). Any point on that line segment is also a maximizer.

31 Alternative Optimal Solutions If a NBV in Row 0 is 0, and we can pivot in this column (and maintain the same value of z), then we may have alternative optimal solutions.

32 Alternative Optimal Solutions If a NBV in Row 0 is 0, and we can pivot in this column (and maintain the same value of z), then we may have alternative optimal solutions. If two BFS are optimal, the line segment joining them is also optimal (by convexity).

33 Alternative Optimal Solutions If a NBV in Row 0 is 0, and we can pivot in this column (and maintain the same value of z), then we may have alternative optimal solutions. If two BFS are optimal, the line segment joining them is also optimal (by convexity). There may be stalling behavior.

34 Alternative Optimal Solutions If a NBV in Row 0 is 0, and we can pivot in this column (and maintain the same value of z), then we may have alternative optimal solutions. If two BFS are optimal, the line segment joining them is also optimal (by convexity). There may be stalling behavior. That is, it may be that once we pivot into the new BV, there may be a new way to proceed (for a better value of z).

35 Example Consider the following final tableau: Interpretation? z x 1 x 2 x 3 x 4 rhs

36 Example Consider the following final tableau: z x 1 x 2 x 3 x 4 rhs Interpretation? Row 0 may have a 0 for x 3, but we cannot pivot in that column. How many solutions to the optim. problem do we have?

37 Example Consider the following LP: Start as usual: max 4x 1 + 3x 2 st x 1 6x 2 5 3x 1 11 x 1, x What s going on? /3 44/ /3 4/ /3 11/3

38 Write out the system of equations: max z = 44/3 +3x 2 (4/3)s 2 st x 1 = 1/3 (1/3)s 2 x 2 = x 2 s 1 = 4/3 +6x 2 +(1/3)s 2 s 2 = s 2 How should we maximize this?

39 Write out the system of equations: max z = 44/3 +3x 2 (4/3)s 2 st x 1 = 1/3 (1/3)s 2 x 2 = x 2 s 1 = 4/3 +6x 2 +(1/3)s 2 s 2 = s 2 How should we maximize this? Set s 2 = 0, and 1 0 x = x This increases the value of z without bound. Notice that c T d 0.

40 Proceed as usual: min z = x 1 + 2x 2 st x 1 x 2 1 x 1 2x 2 2 x 1, x 2 0 Interpretation?

41 There is an optimal solution: (1, 0) The feasible set is unbounded.

42 Two Types of Unboundedness The objective function is unbounded (as is the feasible region). The feasible region is unbounded, but the objective function is not. The LP is unbounded if there is a negative coefficient in Row 0, and all the remaining elements in the column are negative or zero