課程名稱︰資訊工程理論基礎
課程性質︰選修
課程教師︰呂育道
開課學院：電資學院
開課系所︰資工系
考試日期（年月日）︰2015.01.13
考試時限（分鐘）：180
試題 :
Theory of Computation
Final Examination on January 13, 2015
Fall Semester, 2014
Problem 1 (20 points)
Does IP contain all languages that have uniformly polynomial circuits?
Ans:
Yes. P equals the class of languages with uniformly polynomial circuits.
Furthermore, any language in P can be decided by an interactive proof system
where the verifier simply decides the language itself and ignores the provers
messages. So P ⊆ IP.
Problem 2 (30 points)
Design a zero-knowledge proof protocol for 3 Colorability.
Ans: See pp.695-696 of the slides.
Problem 3 (30 points)
Suppose that there are n jobs to be assigned to m machines. Let t_i be the
sorted running time for job i ∈ {1, ... , n} in descending order, which means
t_1 > t_2 > ... > t_n, A be an assignment where A[i] = j means that job i is
assigned to machine j ∈ {1, ... , m}, and T[j] = (Σ_(A[i] = j) t_i) be the
total running time for machine j. The makespan of A is the maximum time that
any machine is busy, or
makespan(A) = max T[j].
j
The Sorted Load Balance problem is to find the assignment which has minimal
makespan over all assignments A, denoted by OPT. It is known to be NP-hard.
Consider the following algorithm for Sorted Load Balance:
1: T[j] ← 0 for j = 1, 2, ... , m
2: for i ← 1 to n do
3: Let min be the j such that T[j] is the smallest
(with ties broken arbitrarily)
4: A[i] ← min
5: T[min] ← T[min] + t_i
6: end for
7: return A
Show that this algorithm for Sorted Load Balance is an approximation algorithm
which returns a solution that is at most (3/2) × OPT. (You may use the fact
that T[j] - t_i ≦ OPT for machine j and job i.)
Ans:
Suppose n ≦ m. Then the algorithm is trivially optimal and the claim holds.
Now suppose n > m. Assume that the optimal algorithm assigns the first m jobs
to distinct machines. Then job (m + 1) must be paired with one of the first
m jobs. Recall that each of the first m jobs has a running time at least t_m.
So t_m + t_(m+1) ≦ OPT, and therefore 2×t_(m+1) < OPT, which implies
t_(m+1) < OPT/2.
On the other hand, assume that two (say i and j, where i < j) of the first
m jobs are assigned by the optimal algorithm to the same machine. Then
t_i + t_j ≦ OPT. It implies that 2×t_j < OPT, which further implies
t_(m+1) ≦ OPT/2 because t_(m+1) < t_j.
Let machine j* be the busiest machine after running our greedy algorithm. This
means T[j*] equals the makespan, i.e., the largest among all T[j]s. Let i* be
the last job assigned to machine j*. Suppose i* ≦ m. Then machine j* has only
one job i* because each job of the first m jobs is assigned to distinct
machines. Since t_1 is the largest running time among the first m jobs, this
implies that i* = 1 and T[j*] = t_1. Recall that t_i ≦ OPT for all i. So
T[j*] = t_1 = OPT and the claim holds. Now suppose i* > m. Then
t_(i*) < t_(m+1) ≦ OPT/2. By the hint, T[j*] ≦ OPT + t_i ≦ (3/2) × OPT.
Hence, the claim is proved.
Problem 4 (20 points)
Argue that if all monotone languages in P have polynomial monotone circuits,
then P ≠ NP.
Ans: See p.804 of the slides.