您的位置:首页 > 其它

创新实训-6-6

2020-07-13 06:12 267 查看

学校的创新实训,第一天翻译了一本书(近似算法)的第四章的一半,po一下证明工作量

第一次用latex,还不错

 

 

4 Multiway Cut and k-Cut

The theory of cuts occupies a central place in the study of exact algorithms.
切分理论是精确算法的中流砥柱
In this chapter, we will present approximation algorithms for natural gener-alizations of the minimum cut problem. These generalizations are NPNP-hard.
在这章,我们会介绍最小割问题自然生成的近似算法,是个NP难问题
Given a connected, undirected graph G = (V,E) with an assignment of
weights to edges, w : E→R+E→R+, a cut is defined by a partition of V into two
sets, say V′V′ and V−V′V−V′, and consists of all edges that have one endpoint in
each partition. Clearly, the removal of the cut from G disconnects G. Given
terminals s, t ∈ V , consider a partition of V that separates s and t. The cut
defined by such a partition will be called an s–t cut. The problems of finding
a minimum weight cut and a minimum weight s–t cut can be efficiently
solved using a maximum flow algorithm. Let us generalize these two notions:
通俗理解
割是一些边的集合,当去掉了这些边的时候,原本的连通图就变成了不连通的图。最小割就是边的权值最小的割。最小割可以通过求最大流来获得。

(这里最小割的定义可以参考我们大三学的算法里面的最小割的定义)
一个联通无向图G=(V,E),边的权重是w。定义一个割被定义成两个集合,分别是V’和V-V’,这两个集合包含了所有的边,每条边都把定点分配给了这两个集合。(也就是说,每条边的两个定点,一个在V’里,一个在V-V’里?)。端点s和t都是V集合里面的点,从这个割就叫做s-t割,这个问题就是在寻找一个最小重量(w)的割,这个问题可以用最大流问题来解决。归纳下面两个概念:
Problem4.1(Multiwaycut)Problem4.1(Multiwaycut) Given a set of terminals S=s1,s2,...,sk⊆VS=s1,s2,...,sk⊆V , a multiway cut is a set of edges whose removal disconnects the terminals
from each other. The multiway cut problem asks for the minimum weight such set.
给定一组端点S = s1,s2,…,sk⊆V,多向切割是一组边集,将这组边移除会断开端点彼此的连接,也就是变成不连通图。 多向切割要求边权值最小。

Problem4.2(Minimumk−cut)Problem4.2(Minimumk−cut) A set of edges whose removal leaves k
connected components is called a k-cut. The k-cut problem asks for a mini-mum weight k-cut.
移动边之后剩下了k个联通部分,这样的边成为k-割,k-割是最小割。
The problem of finding a minimum weight multiway cut is NPNP-hard for
any fixed k ≥ 3. Observe that the case k = 2 is precisely the minimum s–t cut
problem. The minimum k-cut problem is polynomial time solvable for fixed
k; however, it is NPNP-hard if k is specified as part of the input. In this chapter,
we will obtain factor 2 − 2/k approximation algorithms for both problems.
In Chapter 19 we will improve the guarantee for the multiway cut problem
to 3/2.
任何的固定的k≥3,找到最小重量的多路切割的问题是NP -难问题。观察到k = 2的情况恰好是最小的s–t cut问题。k固定的最小k-割问题是固定时间的多项式时间内可解的; 但是,如果将k指定为输入的一部分,则是NP-难问题。 在这一章当中,对于这两个问题,我们将获得向量2 − 2 / k的近似算法。在第19章中,我们将改进对多路切割问题,提升至3/2。

4.1 The multiway cut problem

Define an isolating cut for sisi to be a set of edges whose removal disconnects
sisi from the rest of the terminals.
从sisi中定义一个割,移除掉了这个割之后,sisi和剩余的端点就不联通了。

Algorithm4.3(Multiwaycut)Algorithm4.3(Multiwaycut)

  1. For each i = 1,…,k, compute a minimum weight isolating cut for sisi,
    say CiCi.
  2. Discard the heaviest of these cuts, and output the union of the rest, say
    CC.

对于每一个sisi,i从1到k,都计算一个最小独立割,称为CiCi
丢弃掉这些割里最重的割,然后把剩余的割组成一个割集,称为CC

Each computation in Step 1 can be accomplished by identifying the terminals in S − {si} into a single node, and finding a minimum cut separating
this node from si; this takes one max-flow computation. Clearly, removing C
from the graph disconnects every pair of terminals, and so is a multiway cut.
步骤1中的每个计算都可以通过将S-{sisi}中的终点识别为单个节点并找到最小割把这个节点从sisi重分割出来; 这需要一次最大流计算。 显然,将C从图中移除会使得端点的每个部分都不连通,多路切割也是如此。

Theorem4.4Theorem4.4 Algorithm 4.3 achieves an approximation guarantee of 2−2/k.
定理4.3实现了2-2/k的近似保证
ProofProof: Let AA be an optimal multiway cut in GG. We can view AA as the
union of k cuts as follows: The removal of AA from GG will create k connected
components, each having one terminal (since A is a minimum weight multiway
cut, no more than k components will be created). Let AiAi be the cut separating
the component containing sisi from the rest of the graph. Then A = ⋃ki=1Ai⋃i=1kAi.
令A为G中的最优多路切割,我们可以把A看作是K-割的并集如下:把A从G中移除会导致生成k个联通部分,每个部分都含有一个终点(因为A是一个最小多路分割,最多会生成K个联通部分),定义一个AiAi,AiAi是把剩余的图中包含sisi的联通部分分离出来的割。然后就有 A = ⋃ki=1Ai⋃i=1kAi.

Since each edge of A is incident at two of these components, each edge
will be in two of the cuts AiAi.
这里的is incident at不是很会翻译,感觉意思指的就是A的每条边,边有两个顶点,这两个顶点都不在同一个联通部分上。每条边都属于两个不同的AiAi,也就因此得到了下面那个公式,AiAi的权重之后等于二倍的AA的权值.

Hence,
Σki=1Σi=1kww(AiAi)=2ωω(AA).–4.4.14.4.1
Clearly, AiAi is an isolating cut for sisi. Since CiCi is a minimum weight isolating
cut for SiSi, w(CiCi) ≤ w(AiAi). Notice that this already gives a factor 2 algorithm,
by taking the union of all k cuts CiCi. Finally, since CC is obtained by discarding
the heaviest of the cuts CiCi,

显然,AiAi是一个从sisi中分离出来的割,,因为CiCi是从SiSi中分离出来的最小割,所以有w(CiCi) ≤ w(AiAi)。注意这里已经根据k-割的并集CiCi给了一个向量两种算法。最后,因为C是由抛弃最重的割所获得的,因此有下面这个公式:

ω(c)≤(1−1k)∑ki=1ω(ci)≤(1−1k)∑ki=1ω(Ai)=2(1−1k)ω(Ai)ω(c)≤(1−1k)∑i=1kω(ci)≤(1−1k)∑i=1kω(Ai)=2(1−1k)ω(Ai)

c的权值小于等于cici中的权值之和乘(1−1k)(1−1k)(这一部分是表示抛弃最重割).上一段的翻译中提到,AiAi是一个从sisi中分离出来的割,,因为CiCi是从SiSi中分离出来的最小割,所以有w(CiCi) ≤ w(AiAi),因此得到了(1−1k)∑ki=1ω(ci)≤(1−1k)∑ki=1ω(Ai)(1−1k)∑i=1kω(ci)≤(1−1k)∑i=1kω(Ai)。最后的等于号的成立在于4.4.1的公式

Once again, Algorithm 4.3 is not based on a lower bounding scheme. Exercise 19.2 gives an algorithm with the same guarantee using an LP-relaxation
as the lower bound. The use of LP-relaxations is fruitful for this problem
as well. Section 19.1 gives an algorithm with an improved guarantee, using
another LP-relaxation.
再一次,算法4.3不是基于较低的边界方案。 练习19.2给出了一种使用LP松弛与下限具有相同保证的算法。 LP松弛的使用对于解决此问题也很有用。 第19.1节通过使用另一个LP松弛,给了一个改进的保证。

Example4.5Example4.5 A tight example for this algorithm is given by a graph on
2k vertices consisting of a k-cycle and a distinct terminal attached to each
vertex of the cycle. The edges of the cycle have weight 1 and edges attaching terminals to the cycle have weight 2 − ε for a small fraction ε > 0.
该算法的一个紧密示例由以下图表给出:2k个顶点,包括一个k圈和圈中一个分别连接到每个顶点的相异终点。 圈的边的权重为1,边连接到圈的终点的权重为2-ε,其中小部分ε> 0。
For example, the graph corresponding to k = 4 is
例如,对应于k = 4的图为

For each terminal sisi, the minimum weight isolating cuts for sisi is given by
the edge incident to sisi. So, the cut C returned by the algorithm has weight
(k − 1)(2 − ε). On the other hand, the optimal multiway cut is given by the
cycle edges, and has weight k.
对于每个终点$ s_i ,, s_i 的最轻隔离割是通过边incidentto的最轻隔离割是通过边incidentto s_i $。 因此,算法返回的割点C具有权重(k − 1)(2 −ε)。 另一方面,最优的多路切割是由圈的边给出,权重为k。

内容来自用户分享和网络整理,不保证内容的准确性,如有侵权内容,可联系管理员处理 点击这里给我发消息
标签: