1. Global Optimization The Problem minimize f(x) subject to g i (x)>b i i=1,…,m h j (x)=c j j=1,…,n When x is discrete we call this combinatorial optimization i.e. an optimization problem with a finite number of feasible solutions. Note that when the objective function and/or constraints cannot be expressed analytically, solution techniques used in combinatorial problems can be used to solve this problem

2. P vs. NP P: Optimization problems for which there exists an algorithm to solve it with polynomial time complexity. This means that the time it takes to solve the problem can be expressed as a polynomial that is a function of the dimension of the problem. NP: Stands for ‘non-deterministic polynomial’; not P. NP hard: If a problem Q is such that every problem in NP is polynomially transformable to Q, we say that Q is NP hard. NP problems and their existence give justification for heuristic methods for solving optimization problems.

3. Assignment Problem A set of n people is available to carry out n tasks. If person i does task j, it costs c ij units. Find an assignment {x 1,…x n } that minimizes  n i=1 c ix i The solution is represented by the permutation {x 1,…x n } of the numbers {1,…,n}. Example tasksolution x1 does task 2 1 2 3 x2 does task 3 x1| 3 2 5 x3 does task 1 x2| 9 7 1 cost x3| 4 5 8 person

4. Knapsack Problem A set of n items is available to be packed into a knapsack with capacity C units. Item i has value v i and uses up c i units of capacity. Determine the subset I of items which should be packed in order to minimize  I v i subject to  I c i < C Here the solution is represented by the subset I of the set {1,…,n}. Example valuecapacitysolution 12.7C/2I={1,3} 23.2C/4 31.1C/2

5. Traveling Salesman Problem (TSP) A salesperson must visit n cities once before returning home. The distance from city i to city j is d ij. What ordering of the cities minimizes the distance the salesperson must travel before returning home? SETUP minimize  n i,j=1 d ij x ij subject to  n i=1 x ij =1,  n j=1 x ij =1 where x ij = 1 if go from city i to city j = 0 otherwise Note that this is an integer programming problem and there are (n-1)! possible solutions to this problem.

6. Integer Programming Integer problems involve large numbers of variables and constraint and quickly become very large problems. Finding a Solution If the function is piecewise linear, the problem can be solved exactly with a mixed integer program method that uses branch and bound (later). Otherwise, Heuristic methods (‘finding’ methods) can be used to find approximate solutions. What are heuristic methods? Definition: A heuristic is a technique which seeks good (i.e. near optimal) solutions at a reasonable computational cost without being able to guarantee either feasibility or optimality, or even in many cases to state how close to optimality a particular feasible solution is.

7. Branch and Bound (In general) Branch and bound is a general search method used to find the minimum of a function, f(x), where x is restricted to some feasible region. f 0L =2 f 0U =9 f 1L =2 f 1U =7 f 2L =4 f 2U =9 Step 0Step 1Step 2 f 1L =2 f 1U =7 f 4L =5 f 4U =9 f 3L =4 f 3U =4 In step 2 the lower bound equals the upper bound in region 3, so 4 is a optimal solution for region 3. Region 4 can be removed from consideration since it has a lower bound of 5, which is greater than 4. Continue branching until the location of the global optimal is found. The difficulty with this method is in determining the lower and upper bounds on each of the regions. L = lower bound U = upper bound

8. Branch and Bound applied to Mixed Integer Programming A Mixed Integer Program The solution to the optimization problem includes elements that are integers. So minimize f(x) where x=(x 1,x 2,…,x n ) and x i = integer, for at least one i. Branch and Bound: Suppose x 1 and x 2 must be an integers. x1x1 x2x2 * x1x1 x2x2 Find a global minimum for a relaxed problem. x1x1 x2x2 Find minima for the subproblems I and II. * Find minima for the subproblems III, IV and V. * 5 6 5.1 3.5 4343 5 6 II I III IV V

9. Clustering Methods Clustering methods are an improvement to multistart methods. Multistart methods: These are methods of optimization that determine the global optimal by comparing the local optimal attained from a large number of different starting points. These methods are inefficient because many starting points may lead to the same local minimum. Clustering methods: A form of a multistart method, with one major difference. Neighborhoods of starting points that lead to the same local minimum are estimated. This decreases redundancies in local minimum values that are found. x1x1 x2x2 x1x1 x2x2 x1x1 x2x2 * * * * * * * * Step 0: Sample pointsStep 1: Create groupsStep 2: Continue sampling * * * * * * * * * * * * * * * * * * ** ***The challenge is how to identify the groups.

10. Simulated Annealing A method that is useful in solving combinatorial optimization problems. At the the heart of this method is the annealing process studied in thermodynamics. High TemperatureLow Temperature Thermal mobility is lost as the temperature decreases. Thermodynamic StructureCombinatorial Optimization System statesFeasible solutions EnergyCost Change of stateNeighboring solution TemperatureControl parameter Frozen stateHeuristic solution

11. Simulated Annealing The Boltzmann probability distribution: Prob(E)~exp(-E/kT). k is the Boltzmann’s constant which relates temperature to energy. A system in thermal equilibrium at temperature T has its energy probabilistically distributed among all different energy states E. Even if the temperature is low, there is a chance that the energy state will be high. (This is a small chance, but it is a chance nonetheless.) There is a chance for the system to get out of local energy minimums in search of the global minimum energy state. The general scheme of always taking downhill steps while sometimes taking an uphill step has come to be known as the Metropolis algorithm, named after Metropolis who first incorporated simulated annealing ideas in an optimization problem in 1953.

12. Simulated Annealing Algorithm to minimize the cost function, f(s). 1) Select an initial solution s 0, an initial temperature t 0 and set iter=0. 2) Select a temperature reduction function, , and a maximum number of iterations, nrep. 3) Randomly select a new point s in the neighborhood of s 0 ; set iter=iter+1. 4) If f(s) nrep. 5) Generate random x between 0 and 1. 6) If x nrep. 7) Let s 0 remain the same and go to step 3 until iter > nrep. 8) Set t=  (t) until stopping criteria is met. 9) The approximation to the optimal solution is s 0. All of the following parameters affect the speed and quality of the solution. t 0 :a high value for free exchange. N(s 0 ): by swap, mutation, random, etc.  :cooling should be gradual t decay. stopping criteria:minimum temp; total nrep: related to the dimension;number of iterations exceeded; can vary with t (higher with low t).proportion of acceptable moves.

13. Hybrid Methods MINLP: Mixed Integer Nonlinear Programming Branch and Bound (Talked about this earlier.) 1) Relax the integer constraints forming a nonlinear problem. 2) Fix the integer values found to be closest to the solution found in step 1. 3) Solve the new nonlinear programming problems for the fixed integer values until all of the integer parameters are determined. **Requires a large number of NLP problems to be solved.

14. Hybrid Methods Tree Annealing: Simulated annealing applied to continuous functions Algorithm 1) Randomly choose an initial point, x, over a search interval, S 0. 2) Randomly travel down the tree to an arbitrary terminal node i, and generate a candidate point, y, over the subspace defined by S i. 3) If f(y) < f(x), then replace x with y and go to step 5. 4) Compute P=exp(-(f(y)-f(x))/T). If P > R, where R is a random number uniformly distributed between 0 and 1, then replace x with y. 5) If y replaced x, then decrease T slightly and update the tree. 6) Set i=i+1, and go to 2 until T < Tmin.

15. Hybrid Methods Differences between tree and simulated annealing: 1) The points x and y are sampled from a continuous space. 2) A minimum value is determined by an increasing density of nodes in a given region. The subspace over which candidate points are chosen decreases in size as a minimum value is approached. 3) The probability of accepting y is now governed by a modified criterion: P=g(x)p(y)/g(y)p(x) where p(y)=exp(-f(x)/T) and g(x)=(1/V x )q i g is dependent upon the volume of the node associated with the subspace defined by x, V x, as well as the path from the root node to the current node, q i. Tree annealing is not guaranteed to converge and often convergence is very slow. Use tree annealing as a first step, then use a gradient method to attain the minimum.

16. Statistical Global Optimization A statistical model of the objective function is used to bias the selection of new sample points. This is a Bayesian argument, where we use information about the objective function gathered to make decisions about where to sample new points. Problems 1) The statistical model used may not truly represent the objective function. If the statistical model is determined prematurely, the optimization algorithm may be biased and lead to unreliable solutions. 2) Determining a statistical model can be mathematically intense.

17. Tabu Search The tabu search is designed to cross boundaries of feasibility or local optimality normally treated as barriers, and systematically to impose and release constraints to permit exploration of otherwise forbidden regions. Example: 21435 2543121435 1)Assume an initial solution: 2)Define a neighborhood by some type of operation applied to this solution, such as a swap exchange: The 10 adjacent solutions attained by such a swap for a neighborhood. 3) For each swap define a move value or a change in fitness value. 4) Classify a subset of the moves in a neighborhood as forbidden or tabu, such as pairs that have been swapped cannot be swapped for 3 iterations. Call this the tabu classification. 5) Define an aspiration criteria that allows us to override the tabu classification, such as if the move results in a new global minimum.

18. Tabu Search Example 21435 Iteration 0 Value = 4 Tabu structure 0 0 000 000 0 0 1 2 3 4 2345 Swap value 5,4 -6 3,4 -2 1,2 0 2,3 2 0: free to swap >0: tabu * Iteration 1 21534 Tabu structure 0 0 000 000 0 3 1 2 3 4 2345 Swap value 3,1 -2 2,3 1,5 1 3,4 2 * Value = 10

19. Tabu Search Example (cont.) 23514 Iteration 2 Value = 4 Tabu structure 0 0 003 000 0 2 1 2 3 4 2345 Swap value 1,3 2 3,4 2 4,5 3 2,3 4 T Iteration 3 21543 Tabu structure 0 3 002 000 0 1 1 2 3 4 2345 Swap value 4,5 -3 2,4 1,3 1 3,5 2 Value = 2 T * Choose a move that does not improve the solution. Although 4,5 is tabu, the resulting value obtained from this search is less than the lowest value attained thus far (4-3=1 < 2). T* T

20. Tabu Search Some of the factors that affect the efficiency and the quality of the tabu search include: The number of swaps per iteration. The number of moves that become tabu. The tenure of the tabu. Tabu restrictions (can restrict any swap that includes one member of a tabu pair). Can take into account the frequency of swaps and penalize move values between those pairs that have high swap frequencies.

21. Nested Partitions Method ( Lyeuan Shi, Operations Research, May 2000 ) This method systematically partitions the feasible region into subregions, and then concentrates the computational effort in the most promising region. 1) Assume that we have a most promising region, . 2) Partition this region into M subregions and aggregate the entire surrounding region into one region. 3) Sample each of these M+1 regions and determine the promising index for each. 4) Set  equal to the most promising region. Go to step 1. If the surrounding region is found to be the best, the algorithm backtracks to a larger region that contains the old most promising region.

22. Nested Partitions Example Set M = 2 Set  =  3  = {1,2,3,4,5,6,7,8} partition: partition:  5 = {1} --> prom index = 2   1 = {1,2,3,4} --> prom index = 5  6 = {2}--> prom index = 3  2 = {5,6,7,8}--> prom index = 4  0 \(  5 U  6 ) = {3,4,5,6,7,8}--> prom index = 4 Set  =  1 partition: Backtrack to  =  1  3 = {1,2} --> prom index = 5 new partition:   4 = {3,4}--> prom index = 3  7 = {1,2,3} --> prom index = 5  2 = {5,6,7,8} --> prom index = 4  8 = {4} --> prom index = 2  2 = {5,6,7,8} --> prom index = 4 continue until a minimum is found

23. The Tunneling Method f(q) = original objective function n = the pole strength f*= the local minimum determined in the minimization phase q* = the pumping rate for the local minimum determined in the minimization phase The Tunneling Function Cycle through the minimization and tunneling phases until no roots can be determined for the tunneling function. Minimization Phase: A local minimum value is determined. Tunneling Phase: The objective function is transformed into a tunneling function, whose roots become the starting points for the next minimization phase.

24. Tunneling Method Illustrated.67 Starting point: x=0 1st root found