UVa 1395 Slim Span

Given an undirected weighted graph G , you should find one of spanning trees specified as follows.

The graph G is an ordered pair (V, E) , where V is a set of vertices {v₁, v₂,..., v_n} and E is a set of undirected edges {e₁, e₂,..., e_m} . Each edge e  E has its weight w(e) .

A spanning tree T is a tree (a connected subgraph without cycles) which connects all the n vertices with n - 1 edges. The slimness of a spanning tree T is defined as the difference between the largest weight and the smallest weight among the n - 1 edges of T .

For example, a graph G in Figure 5(a) has four vertices {v₁, v₂, v₃, v₄} and five undirected edges {e₁, e₂, e₃, e₄, e₅} . The weights of the edges are w(e₁) = 3 , w(e₂) = 5 , w(e₃) = 6 , w(e₄) = 6 , w(e₅) = 7 as shown in Figure 5(b).

=6in 

There are several spanning trees for G . Four of them are depicted in Figure 6(a)∼(d). The spanning tree T_a in Figure 6(a) has three edges whose weights are 3, 6 and 7. The largest weight is 7 and the smallest weight is 3 so that the slimness of the tree T_a is 4. The slimnesses of spanning trees T_b , T_c and T_d shown in Figure 6(b), (c) and (d) are 3, 2 and 1, respectively. You can easily see the slimness of any other spanning tree is greater than or equal to 1, thus the spanning tree T_d in Figure 6(d) is one of the slimmest spanning trees whose slimness is 1.

Your job is to write a program that computes the smallest slimness.

Input 

The input consists of multiple datasets, followed by a line containing two zeros separated by a space. Each dataset has the following format.

n m 
a₁ b₁ w₁ 
 
a_m b_m w_m 

Every input item in a dataset is a non-negative integer. Items in a line are separated by a space.

n is the number of the vertices and m the number of the edges. You can assume 2n100 and 0mn(n - 1)/2 . a_k and b_k (k = 1,...,m) are positive integers less than or equal to n , which represent the two vertices v_ak and v_bk connected by the k -th edge e_k . w_k is a positive integer less than or equal to 10000, which indicates the weight of e_k . You can assume that the graph G = (V, E) is simple, that is, there are no self-loops (that connect the same vertex) nor parallel edges (that are two or more edges whose both ends are the same two vertices).

Output 

For each dataset, if the graph has spanning trees, the smallest slimness among them should be printed. Otherwise, `-1' should be printed. An output should not contain extra characters.

Sample Input 

4 5 
1 2 3
1 3 5
1 4 6
2 4 6
3 4 7
4 6 
1 2 10 
1 3 100 
1 4 90 
2 3 20 
2 4 80 
3 4 40 
2 1 
1 2 1
3 0 
3 1 
1 2 1
3 3 
1 2 2
2 3 5 
1 3 6 
5 10 
1 2 110 
1 3 120 
1 4 130 
1 5 120 
2 3 110 
2 4 120 
2 5 130 
3 4 120 
3 5 110 
4 5 120 
5 10 
1 2 9384 
1 3 887 
1 4 2778 
1 5 6916 
2 3 7794 
2 4 8336 
2 5 5387 
3 4 493 
3 5 6650 
4 5 1422 
5 8 
1 2 1 
2 3 100 
3 4 100 
4 5 100 
1 5 50 
2 5 50 
3 5 50 
4 1 150 
0 0

Sample Output 

1 
20 
0 
-1 
-1 
1 
0 
1686 
50

#include <cstdio>
#include <algorithm>
#include <vector>
using namespace std;

// 边的结构体
typedef struct e_node
{
	int a, b;	// 边的端点
	int w;		// 边的权重
	bool operator < (const struct e_node& x) const
	{
		return w < x.w;
	}
}e_node;

vector<e_node> e_array;

// p[i]代表结点i所在连通分量的父节点
// p[i]==i代表结点i为所在连通分量的树根
int p[120];

int n, m;

int inf = (1<<30);


int find(int x);


int main()
{
	while(scanf("%d %d", &n, &m) == 2 && !(n == 0 && m == 0))
	{
		// 读入所有边
		e_array = vector<e_node>();
		for(int i = 1; i <= m; i++)
		{
			e_node x;
			scanf("%d %d %d", &x.a, &x.b, &x.w);
			e_array.push_back(x);
		}

		// 将所有边按权重从小到大排序
		sort(e_array.begin(), e_array.end());

/*
		for(int i = 0; i < m; i++)
			printf("a:%d b:%d w:%d\n", e_array[i].a, e_array[i].b, e_array[i].w);
*/

		// 枚举所有可能情况的最小生成树
		int result = inf;
		for(int i = 0; i < m; i++)
		{
			// 将所有点作为单独连通分量
			for(int j = 1; j <= n; j++)
				p[j] = j;
			int c_count = n;
			// 建立最小生成树
			for(int j = i; j < m; j++)	
			{
				// 查看该边的两个端点是否在一个连通分量中
				// 如果不是，将该边加入生成树
				// 即将该两点所在连通分量合并
				int f_a = find(e_array[j].a);
				int f_b = find(e_array[j].b);

				if(f_a != f_b)
				{
					c_count--;
					p[f_a] = f_b;
					if(c_count == 1)
					{
						result = min(result, e_array[j].w-e_array[i].w);
		//				printf("here: result %d\n", result);
						break;
					}
				}	
			}			
		}				

		if(result == inf)
			printf("-1\n");
		else
			printf("%d\n", result);	
	}	
	return 0;	
}

// 找到该节点的连通分量代表元素
int find(int x)
{
	if(p[x] == x)
		return x;
	else
	{
		int f_x = find(p[x]);
		p[x] = f_x;
		return f_x;	
	}
}

这是道好题，想着用Kruskal求最小生成树框架思路，但是这里求的是最大最小权重差值最小的生成树。

想到枚举最小权重的边，来看最大权重的情况，但是最小生成树只是保证权重之和最小，如果保证最大权重最小。

后来发现一条性质：最小生成树不仅权重之和最小，而且最大权重最小。

即：最小生成树为瓶颈生成树。证明可以在网上找。

这样，就可以枚举最小权重，来生成最小生成树，这样得到的就为该最小权重下的最大最小权重差值最小的生成树。