1. Daniel Kroening and Ofer Strichman Decision Procedures An Algorithmic Point of View Gaussian Elimination and Simplex

2. Gaussian’s elimination Given a linear system Ax = b Manipulate A|b to an upper-triangular form A x = b

3. Gaussian’s elimination Then, solve backwards from the k ’s row according to:

4. Gaussian elimination - example And now… x 3 = -1, x 2 = 3, x 1 = 1 problem solved.

5. Feasibility with Simplex Simplex was originally designed for solving the optimization problem: s.t. We are only interested in the feasibility problem. Is this system optimal ? Is this system feasible ?

6. General simplex We will learn a variant called general simplex. Very suitable for solving the feasibility problem fast. The input:  A is a m £ n coefficient matrix  The problem variables: First step: convert the input to general form

7. General form General form: A combination of:  Linear equalities of the form  Lower and upper bounds on variables.

8. Converting to General Form A: Replace (where ) with and s 1,..., s m are called the additional variables.

9. Example 1 Convert to: It is common to keep the conjunctions implicit

10. Example 2 Convert to:

11. Simplex basics… Linear inequality constraints, geometrically, define a convex polyhedron. © Wikipedia

12. Our example from before, geometrically General Simplex begins in the origin...

13. Matrix form xys1s2s3xys1s2s3 Recall the general form: Due to the additional variables:  now A is an m £ ( n + m ) matrix.

14. The tableau The diagonal part is inherent to the general form We can instead write: xys1s2s3xys1s2s3 xyxy s1s2s3s1s2s3 This is called the tableau

15. The tableau The tableau changes throughout the algorithm, but maintains its m £ n structure Distinguish between basic and nonbasic variables Initially, basic variables = the additional variables. xyxy s1s2s3s1s2s3 Basic variables Nonbasic variables

16. The tableau Denote by  B – Basic variables  N – Nonbasic variables The tableau is simply a rewrite of the system: The basic variables are also called the dependent variables.

17. The general simplex algorithm Simplex maintains:  The tableau,  an assignment ® to all variables  The bounds Initially,  B = additional variables  N = problem variables  ® ( x i ) = 0 for i 2 {1,..., n + m }

18. Invariants Two invariants are maintained throughout: 1. 2. All nonbasic variables satisfy their bounds Can you see why these invariants are maintained initially ? We should check that they are indeed maintained

19. The general simplex algorithm The initial assignment satisfies If the bounds of all basic variables are satisfied by ®, return `Satisfiable’. Otherwise... pivot.

20. Pivoting Find a basic variable x i that violates its bounds.  Suppose that ® ( x i ) < l i Find a nonbasic variable x j such that  a ij > 0 and  ( x j ) < u j, or  a ij l j Why ?

21. Pivoting Find a basic variable x i that violates its bounds.  Suppose that ® ( x i ) < l i Find a nonbasic variable x j such that  a ij > 0 and  ( x j ) < u j, or  a ij l j Such a variable x j is called suitable. If there is no suitable variable – return ‘Unsatisfiable’  Why ?

22. Pivoting x i with x j Solve equation i for x j : From: To: Swap x i and x j, and update the i -th row accordingly. From To: a i1... a ij... a in - a i1 a ij... 1 a ij...- a in a ij

23. Pivoting x i with x j Update all other rows:  Replace x j with its equivalent obtained from row i :

24. Pivoting Update ® as follows: Increase ® ( x j ) by  Now x j is a basic variable: it can violate its bounds. Update ® ( x i ) accordingly  Q: What is now ® ( x i ) ? Update  for all other basic (dependent) variables.

25. Example Recall the tableau and constraints in our example: Initially  assigns 0 to all variables Bounds of s 1 and s 3 are violated

26. Example Recall the tableau and constraints in our example: We will solve s 1 x is a suitable nonbasic variable for pivoting  It has no upper bound So now we pivot s 1 with x

27. Example Recall the tableau and constraints in our example: Solve 1 st row for x : Replace x with s 1 in other rows:

28. Example The new state: Solve 1 st row for x : Replace x with s 1 in other rows:

29. Example The new state: What about the assignment ? We should increase x by  Hence,  ( x ) = 0 + 2 = 2  Now s 1 is equal to its lower bound:  ( s 1 ) = 2  Update all the others

30. Example The new state: Now s 3 violates its lower bound Which nonbasic variable is suitable for pivoting ?  That’s right… y

31. Example The new state: We should increase y by

32. Example The final state: All constraints are now satisfied

33. Observations The additional variables:  Only additional variables have bounds.  These bounds are permanent.  Additional variables exit the base only on extreme points (their lower or upper bounds).  When entering the base, they shift towards the other bound and possibly cross it (violate it).

34. Observations Can it be that we pivot( x i, x j ) and then pivot( x j, x i ) and enter a (local) cycle ?  No.  For example, suppose that a ij > 0.  We increased  ( x j ) so now  ( x i ) = l i.  After pivoting, possibly  ( x j ) > u j  But a ij ’ = 1 / a ij > 0, hence x i is not suitable.

35. Observations Is termination guaranteed ? Not obvious.  Perhaps there are bigger cycles. In order to avoid circles, we use Bland’s rule:  determine a total order on the variables.  Choose the first basic variable that violates its bounds, and first nonbasic suitable variable for pivoting.  It can be proven that this guarantees that no base is repeated, which implies termination.

36. General-Simplex with Bland’s rule