课件43(2010下学期).pptVIP

Agenda Classification LAS VEGAS ALGORITHMS ONE-SIDED-ERROR MONTE CARLO ALGORITHMS BOUNDED-ERROR MONTE CARLO ALGORITHMS UNBOUNDED-ERROR MONTE CARLO ALGORITHMS Definition 2.4.41. A randomized algorithm A is called a Las Vegas algorithm computing a function F if, for any input x (any argument x of F), Prob(A(x) = F(x)) = 1. Example 2.4.44. We consider again the model of communication protocols. We have two computers RI and RII. The input of RI consists of ten strings, x1, x2, . . ., x10 ∈ {0, 1}n, and RII also has ten strings, y1, y2, . . ., y10 ∈ {0, 1}n. The task is to estimate whether there is a j ∈ {1, . . ., 10} such that xj = yj . If such a j exists, then the protocol has to accept the input ((x1, . . ., x10), (y1, . . ., y10)), and if not, the input has to be rejected. As usual, the complexity is measured in the number of exchanged bits. One can prove that every deterministic protocol solving this task must allow a communication of 10n bits. Now, we design a Las Vegas protocol that solves the task with communication complexity of n + O(log n). Protocol LV10 Initial situation: RI has ten strings, x1, x2, . . ., x10, xi ∈ {0, 1}n for all i =1, . . . , 10, and RII has ten strings, y1, y2, . . ., y10, yi ∈ {0, 1}n for all i = 1, . . . , 10. Phase 1: RI uniformly chooses 10 primes, p1, p2, . . ., p10, from PRIM (n2) at random. Phase 2: RI computes si = Number(xi) mod pi for i = 1, 2, . . ., 10 and sends p1, p2, . . ., p10, s1, s2, . . ., s10 to RII. First, we analyze the worst-case complexity of LV10. The maximal possible communication consists of sending p1, p2, . . ., p10, s1, s2, . . ., s10 from RI to RII, and of sending yj and j from RII to RI . All the numbers p1, p2, . . ., p10, s1, s2, . . ., s10 are smaller than n2, and so every of them can be represented by 2log2 n bits. The integer j can be represented by 4 bits, and sending the whole yj costs n bits. Altogether, the communicat