1. LINEAR PROGRAMMING PROBLEM Definition and Examples

2. Linear Program Decision Variables Objective Function Constraints

3. Linear Program (General Form) Objective Function Constraints

4. The Linear Programming Model Standard form

5. Linear Programming Problem (6,11) (6,3) (3,11) (3,0) (0,0) (0,2) Not drawn to scale x 1 +2 x 2 = 0 x 1 +2 x 2 = 35 x 1 +2 x 2 = 28 Solution: x 1 = 6, x 2 = 11 Optimal Objective Value: 28

6. Overview Solving a Linear Program. Visualizing Linear Programs. What does solving a Linear Program mean? Algorithms for Linear Programming. Simplex. Ellipsoidal Methods. Interior Point Methods.

7. VISUALIZING LINEAR PROGRAMS

8. Linear Program (General Form) Objective Function Constraints

9. Feasible Region Feasible Region: Polyhedron (n dimensional)

10. Optimization

11. Will the optimal solution always be at a vertex? Prove it.

12. Solving Linear Programs Outcome #1: Optimal Solution(s) exists. Outcome #2: Objective Function is unbounded. Outcome #3: Feasible Region is empty.

13. Unbounded Problem (Example) y x Feasible Region Feasible Region

14. Infeasible Problem Issue: Constraints contradict each other.

15. Solving Linear Programs 1. Find which of the three cases are applicable. Infeasible? Unbounded? Feasible + Bounded = Optimal? 1. If Optimal, find optimal solution. Note multiple optimal solutions possible.

16. LINEAR PROGRAMMING ALGORITHMS

17. Linear Programming Solving systems of Linear Inequalities. Early work by Fourier (Fourier-Motzkin Elimination Algorithm). In symbolic logic, this is called “Linear Arithmetic”. World War II: Optimal allocation of resources. Advent of electronic/mechanical calculating machines. L.V. Kantorovich in USSR (1940) and G.B. Dantzig et al. in the USA (1947).

18. SIMPLEX Simplex: algorithm for solving LPs. First Published by George B. Dantzig Prof. Dantzig contributed numerous seminal ideas to this field. G.B Dantzig: Maximization of a linear function of variables subject to linear inequalities, 1947. Photo credit: Stanford University

19. Visualizing the Simplex Algorithm (6,11) (6,3) (3,11) (3,0) (0,0) (0,2) Not drawn to scale Solution: x 1 = 6, x 2 = 11pt. Objective Value: 28

20. Linear Programming Theory Duality: John Von Neumann Early work by Lagrange. Connections to game theory. Generalized to Karush-Kuhn-Tucker Conditions. Complexity of Simplex: Exponential time in the worst case (Klee + Minty). Polynomial time in the “average case”. Much remains to be understood.

21. Polynomial Time Algorithms Leonid Khachiyan’s ellipsoidal algorithm [Kachiyan’1980] First polynomial time algorithm. Interior Point Methods Ideas go back to Isaac Newton (Newton-Raphson). First algorithms for Linear Programs by Narendra Karmarkar [Karmarkar’1984] Interior point methods are useful for non-linear programming (Cf. Nocedal + Wright textbook).

22. Applications of Linear Programming Theory Too numerous to list exhaustively… Major application areas: Operations Research. Optimal allocation of resources. Decision making. Computer Science Algorithms, Machine Learning, Automated Reasoning, Robotics. Engineering Control Theory

23. INTEGER LINEAR PROGRAMMING Real vs. Integer Variables

24. Feasible Region Feasible Region: Polyhedron (n dimensional)

25. Linear vs. Integer Linear Programs

26. Integer Linear Programming Feasible Region: Z-Polyhedron (n dimensional)

27. Linear vs. Integer Linear Programs (Complexity) Polynomial Time Linear Programming (Reals) Linear Programming (Integers) Nondeterministic Polynomial Time (NP) Million Dollar Question: Can Integer Linear Programs be solved in polynomial time? ( P = ? = NP)

28. Example #1 (6,11) (6,3) (3,11) (3,0) (0,0) (0,2) Not drawn to scale Solution: x 1 = 6, x 2 = 11 Objective Value: 28

29. Example #2

30. LINEAR PROGRAMMING Formulating a Linear Program

31. Example (H&L) Example 1: Design of radiation therapy for cancer treatment Goal: select best combination of beams and their intensities to generate best possible dose distribution Dose is measured in kilorads 31

32. Example 1: Radiation Therapy Design 32

33. Example 1: Radiation Therapy Design Linear programming model Using data from Table 3.7 33

34. Example 1: Radiation Therapy Design A type of cost-benefit tradeoff problem 34