1. Convex Hull(35.3) Convex Hull, CH(X), is the smallest convex polygon containing all points from X, |X|=n Different methods: –incremental: moving from left to right updates CH(x1..xi). the runtime O(nlogn) –divide-and-conquer divides into two subsets left and right in O(n), then combine into one convex hull in O(n) –prune-and-search O(n logh), where h is the # points in CH uses pruning as for finding median to find upper chain

2. Graham’s Scan (35.3) O(nlogn)

3. Finding the Closest Pair(35.4) Brute-force: O(n 2 ) Divide-and-conquer algorithm with recurrence T(n)=2T(n/2)+O(n) Divide: divide into almost equal parts by a vertical line which divides given x-sorted array X into 2 sorted subarrays Conquer: Recursively find the closest pair in each half of X. Let  = min{  left,  right } Combine: The closest pair is either in distance  or a pair of points from different halves.

4. Combine in D-a-C (35.4) Subarray Y’ (y-sorted) of Y with points in 2  strip  p  Y’ find all in Y’ which are closer than in  –no more than 8 in   2  rectangle –no more than 7 points can be closer than in  If the closest in the strip closer then it is the answer 22  left right 22 

5. Voronoi Graph Voronoi region Vor(p) (p in set S) –the set of points on the plane that are closer to p than to any othe rpoint in S Voronoi Graph VOR(S) –dual to voronoi region graph –two points are adjacent if their voronoi regions have common contiguous boundary (segment)

6. Voronoi Graph Voronoi Graph in the rectilinear plane Rectilinear distance: p = (x, y); p’=(x’,y’) a b c bc ac ab Voronoi region of b

7. NP-Completeness (36.4-5) NPC P NP-hard P: yes and no in pt NP: yes in pt NPH  NPC NP

8. Independent Set Independent set in a graph G: pairwise nonadjacent vertices Max Independent Set is NPC Is there independent set of size k? –Construct a graph G: literal -> vertex two vertices are adjacent iff –they are in the same clause –they are negations of each other –3-CNF with k clauses is satisfiable iff G has independent k-set assign 1’s to literals-vertices of independent set Example: f = (x+z+y’) & (x’+z’+a) & (a’+x+y) x z y’ x’ z’ a a’ x y x, z’, y independent F is satisfiable: f = 1 if x = z’ = y = 1

9. MAX Clique Max Clique (MC): –Find the maximum number of pairwise adjacent vertices MC is in NP –for the answer yes there is certificate of polynomial length = clique which can be checked in polynomial time MC is in NPC –Polynomial time reduction from MIS: For any graph G any independent set in G 1-1 corresponds to clique in the complement graph G’ red independent set red clique G Complement G’ noedge  edge edge  noedge

10. Minimum Vertex Cover Vertex Cover: –the set of vertices which has at least one endpoint in each edge Minimum Vertex Cover (MVC): –the set of vertices which has at least one endpoint in each edge MIN Vertex Cover is NPC –If C is vertex cover, then V - C is an independent set red independent set blue vertex cover

11. Set Cover Given: a set X and a family F of subsets of X, F  2 X, s.t. X covered by F Find : subfamily G of F such that G covers X and |G| is minimize Set Cover is NPC –reduction from Vertex Covert Graph representation: red elements of ground set X blue subsets in family F edge between set A  F and element x  X means x  A A abc d A = {a,b,c}, B = {c,d} B

12. Intermediate Classes NPC NP-hard P  NP  Dense Set Cover is NP but not in P neither in NPC Dense Set Cover: Each element of X belongs to at least half of all sets in F

13. Runtime Complexity Classes Runtime order: –constant –almost constant –logarithmic –sublinear –linear –pseudolinear –quadratic –polynomial –subexponential –exponential –superexponential Example –adding an element in a queue/stack –inverse Ackerman function = O(loglog…log n) n times –extracting minimum from binary heap –n 1/2 –traversing binary search tree, list –O(n log n) sorting n numbers, closest pair, MST, Dijkstra shortest paths –adding two n  n matrices – e n ^ (1/2) – e n,, n! –Ackerman function 2. 2 n times

14. Hamiltonian Cycle and TSP Hamiltonian Cycle: –given an undirected graph G –find a tour which visits each point exactly once Traveling Salesperson Problem –given a positive weighted undirected graph G (with triangle inequality = can make shortcuts) –find a shortest tour which visits all the vertices HC and TSP are NPC NPC problems: SP, ISP, MCP, VCP, SCP, HC, TSP

15. Approximation Algorithms (37.0) When problem is in NPC try to find approximate solution in polynomial-time Performance Bound = Approximation Ratio (APR) (worst-case performance) –Let I be an instance of a minimization problem –Let OPT(I) be cost of the minimum solution for instance I –Let ALG(I) be cost of solution for instance I given by approximate algorithm ALG APR(ALG) = max I {ALG(I) / OPT(I)} APR for maximization problem = max I {ALG(I) / OPT(I)}

16. Vertex Cover Problem (37.1) Find the least number of vertices covering all edges Greedy Algorithm: –while there are edges add the vertex of maximum degree delete all covered edges 2-VC Algorithm: –while there are edges add the both ends of an edge delete all covered edges APR of 2-VC is at most 2 –e 1, e 2,..., e k - edges chosen by 2-VC –the optimal vertex cover has  1 endpoint of e i –2-VC outputs 2k vertices while optimum  k

17. 2-approximation TSP (37.2) Given a graph G with positive weights Find a shortest tour which visits all vertices Triangle inequality w(a,b) + w(b,c)  w(a,c) 2-MST algorithm: –Find the minimum spanning tree MST(G) –Take MST(G) twice: T = 2  MST(G) –The graph T is Eulerian - we can traverse it visiting each edge exactly once –Make shortcuts APR of 2-MST is at most 2 –MST weight  weight of optimum tour any tour is a spanning tree, MST is the minimum

18. 3/2-approximation TSP (Manber) Matching Problem (in P) –given weighted complete (all edges) graph with even # vertecies –find a matching (pairwise disjoint edges) of minimum weight Christofides’s Algorithm (ChA) –find MST(G) –for odd degree vertices find minimum matching M –output shortcutted T = MST(G) + M APR of ChA is at most 3/2 –|MST|  OPT –|M|  OPT/2 –|T|  (3/2) OPT odd

19. 3/2-approximation TSP Christofides’s Algorithm (ChA) –find MST(G) –for odd degree vertices find minimum matching M –output shortcutted T = MST(G) + M The worst case for Christofides heuristic in Euclidean plane: 1 1 1 1234… k k+1k+2… 2k-1k+3 - Minimum Spanning Tree length = 2k - 2 - Minimum Matching of 2 odd degree nodes = k - 1 - Christofides heuristic length = 3k - 3 - Optimal tour length = 2k - 1 - Approximation Ratio of Christofides = 3/2-1/(k-1/2)

20. Non-approximable TSP (37.2) Approximating TSP w/o triangle inequality is NPC –any c-approximation algorithm can solve Hamiltonian Cycle Problem in polynomial time Take an instance of HCP = graph G Assign weight 0 to any edge of G Complete G up to complete graph G’ Assign weight 1 to each new edge c-approximate tour can use only 0-edges - so it gives Hamiltonian cycle of G

21. Steiner Tree Problem Given: A set S of points in the plane = terminals Find: Minimum-cost tree spanning S = minimum Steiner tree 1 1 Cost = 2 Steiner Point Cost =  3 1 Terminals 1 1 Euclidean metric 1 1 1 1 1 1 1 1 1 1 Cost = 6Cost = 4 Rectilinear metric

22. Steiner Tree Problem in Graphs Given a graph G=(V,E,cost) and terminals S in V Find minimum-cost tree spanning all terminals MST algorithm (does not use Steiner points): –find G(S) = complete graph on terminals edge cost = shortest path cost –find T(S) = MST of G(S) –replace each edge of T(S) with the path in G –output T(S)

23. MST -Heuristic Theorem: MST-heuristic is a 2-approximation in graphs Proof: MST < Shortcut Tour  Tour = 2 OPTIMUM

24. Approximation Ratios Euclidean Steiner Tree Problem – approximation ratio = 2/  3 Rectilinear Steiner Tree Problem –approximation ratio = 3/2 Steiner Tree Problem in graphs –approximation ratio = 2 Steiner Point Opt Cost = k 1 2 3 k 5 4 MST Cost = 2k-2 Approximation ratio = 2-2/k  2

25. The Set Cover Problem Sets A i cover a set X if X is a union of A i Weighted Set Cover Problem Given: –A finite set X (the ground set X) –A family of F of subsets of X, with weights w: F   + Find: –sets S  F, such that S covers X, X =  {s | s  S} and S has the minimum total weight  {w(s) | s  S} If w(s) =1 (unweighted), then minimum # of sets

26. Greedy Algorithm for SCP Greedy Algorithm: –While X is not empty find s  F minimizing w(s) / |s  X| X = X - s C = C + s –Return C 6 345 2 1

27. Analysis of Greedy Algorithm Th: APR of the Greedy Algorithm is at most 1+ln k Proof:

28. Approximation Complexity Approximation algorithm = polynomial time approximation algorithm PTAS = a series of approximation algorithms s.t. for any  > 0 there is pt (1+  )-approximation –There is PTAS fro subset sum Remarkable progress in 90’s (assuming P  NP). –No PTAS for Vertex Cover –No clog k-approximation for Set Cover for k < 1 k is the size of the ground set X –No n 1-  approximation for Independent Set n is the number of vertices