LC 882. 细分图中的可到达节点

题目描述

这是 LeetCode 上的 882. 细分图中的可到达节点 ,难度为 困难

给你一个无向图(原始图),图中有 n 个节点,编号从 0n - 1 。你决定将图中的每条边 细分 为一条节点链,每条边之间的新节点数各不相同。

图用由边组成的二维数组 edges 表示,其中 $edges[i] = [u{i}, v{i}, cnt{i}]$ 表示原始图中节点 $u{i}$ 和 $v{i}$ 之间存在一条边,$cnt{i}$ 是将边 细分 后的新节点总数。注意,$cnt_{i} = 0$ 表示边不可细分。

要 细分边 $[u{i}, v{i}]$ ,需要将其替换为 $(cnt{i} + 1)$ 条新边,和 $cnt{i}$ 个新节点。新节点为 $x1, x_2, …, x{cnt{i}}$ ,新边为 $[u{i}, x{1}], [x{1}, x{2}], [x{2}, x{3}], …, [x{cnt{i}}+1, x{cnt{i}}], [x{cnt{i}}, v{i}]$ 。

现在得到一个 新的细分图 ,请你计算从节点 0 出发,可以到达多少个节点?如果节点间距离是 maxMoves 或更少,则视为 可以到达 。

给你原始图和 maxMoves ,返回 新的细分图中从节点 0 出发 可到达的节点数 。

示例 1:

1
2
3
4
5
6
输入:edges = [[0,1,10],[0,2,1],[1,2,2]], maxMoves = 6, n = 3

输出:13

解释:边的细分情况如上图所示。
可以到达的节点已经用黄色标注出来。

示例 2:
1
2
3
输入:edges = [[0,1,4],[1,2,6],[0,2,8],[1,3,1]], maxMoves = 10, n = 4

输出:23

示例 3:
1
2
3
4
5
输入:edges = [[1,2,4],[1,4,5],[1,3,1],[2,3,4],[3,4,5]], maxMoves = 17, n = 5

输出:1

解释:节点 0 与图的其余部分没有连通,所以只有节点 0 可以到达。

提示:

  • $0 <= edges.length <= \min(n * (n - 1) / 2, 10^4)$
  • $edges[i].length = 3$
  • $0 <= u{i} < v{i} < n$
  • 图中 不存在平行边
  • $0 <= cnt_{i} <= 10^4$
  • $0 <= maxMoves <= 10^9$
  • $1 <= n <= 3000$

朴素 Dijkstra

为了方便,我们将原始图边的数量记为 m,因此对于原始图而言,点的数量 $3000$,边的数量为 $10000$。

题目要我们求新图上,从 0 点出发可到达的点的数量,我们将原图上存在的点称为「原点」,细分边上增加的点称为「细分点」,两类点中可达点的数量即是答案。

在分别考虑如何统计两类点之前,我们需要重新定义一下边的权重:若原点 u 和原点 v 的边上存在 c 个细分点,我们将原点 u 和原点 v 之间的边看作是一条权重为 c + 1 的无向边(结合题意,c 个点存在 c + 1 个分段/距离)

重新定义边的权重后,因为该图是「稠密图」,我们可以使用「朴素 Dijkstra」来求解最短路,得到 $dist$ 数组,其中 $dist[x] = t$ 含义为从原点 0 点出发,到达原点 x 的最短距离为 t

不了解最短路的同学可以看前置 🧀 : 涵盖所有的「存图方式」与「最短路算法(详尽注释)」

随后考虑如何统计答案(可达点的数量),根据统计点的类型分情况讨论:

  1. 对于原点:若有 $dist[x] \leq max$ 的话,说明原点 x 可达,累加到答案中;

  2. 对于细分点:由于所有的细分点都在原图边上,因此我们可以统计所有原图边上有多少细分点可达。
    对于任意一条边 $e(u, v)$ 而言,该边上可达点数量包含「经过原点 u 可达」以及「经过原点 v 可达」的并集,其中原点 0 到达原点 u 以及原点 v 的距离,我们是已知的。因此经过原点 u 可达的数量为 $\max(0, max - dist[u])$,经过原点 v 可达的数量为 $\max(0, max - dist[v])$,两者之和与该边上细分点的总数取 min 即是这条边可达点的数量。

代码:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
class Solution {
static int N = 3010, INF = 0x3f3f3f3f;
static int[][] g = new int[N][N];
static int[] dist = new int[N];
static boolean[] vis = new boolean[N];
public int reachableNodes(int[][] edges, int max, int n) {
// 建图
for (int i = 0; i < n; i++) Arrays.fill(g[i], INF);
for (int[] info : edges) {
int a = info[0], b = info[1], c = info[2] + 1;
g[a][b] = g[b][a] = c;
}
// 朴素 Dijkstra
Arrays.fill(dist, INF);
Arrays.fill(vis, false);
dist[0] = 0;
for (int i = 0; i < n; i++) {
int t = -1;
for (int j = 0; j < n; j++) {
if (!vis[j] && (t == -1 || dist[j] < dist[t])) t = j;
}
vis[t] = true;
for (int j = 0; j < n; j++) dist[j] = Math.min(dist[j], dist[t] + g[t][j]);
}
// 统计答案
int ans = 0;
for (int i = 0; i < n; i++) {
if (dist[i] <= max) ans++;
}
for (int[] info : edges) {
int a = info[0], b = info[1], c = info[2];
int c1 = Math.max(0, max - dist[a]), c2 = Math.max(0, max - dist[b]);
ans += Math.min(c, c1 + c2);
}
return ans;
}
}

  • 时间复杂度:建图复杂度为 $O(m)$;使用朴素 Dijkstra 求最短路复杂度为 $O(n^2)$;统计答案复杂度为 $O(n + m)$。整体复杂度为 $O(m + n^2)$
  • 空间复杂度:$O(n^2)$

SPFA

从数据范围来看,无论是朴素 Dijkstra 还是堆优化版的 Dijkstra 都可以过,复杂度分别为 $O(n^2)$ 和 $O(m\log{n})$。

那 Bellman Ford 类的单源最短路就无法通过了吗?

理论上,无论是 Bellman Ford 还是 SPFA 复杂度均为 $O(n \times m)$,均无法通过本题。但实际上 SPFA 由于使用队列对松弛顺序进行了调整,因此在应对「非菊花图」时均表现良好,复杂度可视为 $O(k \times m)$,近似 $O(m)$。

代码:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
class Solution {
static int N = 3010, M = 20010, INF = 0x3f3f3f3f, idx = 0;
static int[] he = new int[N], e = new int[M], ne = new int[M], w = new int[M];
static int[] dist = new int[N];
static boolean[] vis = new boolean[N];
void add(int a, int b, int c) {
e[idx] = b;
ne[idx] = he[a];
w[idx] = c;
he[a] = idx++;
}
public int reachableNodes(int[][] edges, int max, int n) {
// 建图
Arrays.fill(he, -1);
idx = 0;
for (int[] info : edges) {
int a = info[0], b = info[1], c = info[2] + 1;
add(a, b, c); add(b, a, c);
}
// SPFA
Arrays.fill(dist, INF);
Arrays.fill(vis, false);
Deque<Integer> d = new ArrayDeque<>();
d.addLast(0);
dist[0] = 0;
vis[0] = true;
while (!d.isEmpty()) {
int t = d.pollFirst();
vis[t] = false;
for (int i = he[t]; i != -1; i = ne[i]) {
int j = e[i];
if (dist[j] > dist[t] + w[i]) {
dist[j] = dist[t] + w[i];
if (vis[j]) continue;
d.addLast(j);
vis[j] = true;
}
}
}
// 统计答案
int ans = 0;
for (int i = 0; i < n; i++) {
if (dist[i] <= max) ans++;
}
for (int[] info : edges) {
int a = info[0], b = info[1], c = info[2];
int c1 = Math.max(0, max - dist[a]), c2 = Math.max(0, max - dist[b]);
ans += Math.min(c, c1 + c2);
}
return ans;
}
}

  • 时间复杂度:建图复杂度为 $O(m)$;使用 SPFA 求最短路复杂度为 $O(n \times m)$;统计答案复杂度为 $O(n + m)$。整体复杂度为 $O(n \times m)$
  • 空间复杂度:$O(n + m)$

最后

这是我们「刷穿 LeetCode」系列文章的第 No.882 篇,系列开始于 2021/01/01,截止于起始日 LeetCode 上共有 1916 道题目,部分是有锁题,我们将先把所有不带锁的题目刷完。

在这个系列文章里面,除了讲解解题思路以外,还会尽可能给出最为简洁的代码。如果涉及通解还会相应的代码模板。

为了方便各位同学能够电脑上进行调试和提交代码,我建立了相关的仓库:https://github.com/SharingSource/LogicStack-LeetCode

在仓库地址里,你可以看到系列文章的题解链接、系列文章的相应代码、LeetCode 原题链接和其他优选题解。