MIT 公开课 - 高等算法 lec19.pdfVIP

MIT OpenCourseWare 6.854J / 18.415J Advanced Algorithms Fall 2008 �� For information about citing these materials or our Terms of Use, visit: /terms. 18.415/6.854 Advanced Algorithms November 21, 2008 Lecture 19 Lecturer: Michel X. Goemans 1 Introduction In this lecture, we revisit MAXCUT and describe a randomized γ (≈ .87856)-approximation al­ gorithm. We also explore SPARSEST-CUT, an NP-hard problem for which no constant factor approximation is known. We begin to describe an O(log k) approximation using multicommodity fl ows ; here k is the number of commodities. To define the relationship between the optimal values of SPARSEST-CUT and multicommodity flow, we introduce metrics and finite metric spaces. 2 Revisiting MAXCUT Recall the MAXCUT problem: given a graph G = (V , E) and weights w : E → R+ (we could assume ¯ that G is the complete graph and weights are 0 for the original non-edges), maximize w (S : S) (= � wij ) in S ⊂ V . MAXCUT can be formulated as the integer program i∈S ¯ j ∈S max � w (1 − x x )/2 ij i j (i,j )∈E subject to xi ∈ {±1},∀i. The prior lecture described a 1/2-approximation algorithm and an upper bound on the solution to the above optimization, via reduction to a semidefinite program. 2.1 SDP Relaxation of MAXCUT In the SDP relaxation, we replaced the xi with unit vectors in the sphere Sn−1 :=