Description

You are given an undirected unweighted tree consisting of \(n\) vertices.

An undirected tree is a connected undirected graph with \(n−1\) edges.

Your task is to choose two pairs of vertices of this tree (all the chosen vertices should be distinct) \((x_1,y_1)\) and \((x_2,y_2)\) in such a way that neither \(x_1\) nor \(y_1\) belong to the simple path from \(x_2\) to \(y_2\) and vice versa (neither \(x_2\) nor \(y_2\) should not belong to the simple path from \(x_1\) to \(y_1\)).

It is guaranteed that it is possible to choose such pairs for the given tree.

Among all possible ways to choose such pairs you have to choose one with the maximum number of common vertices between paths from \(x_1\) to \(y_1\) and from \(x_2\) to \(y_2\). And among all such pairs you have to choose one with the maximum total length of these two paths.

It is guaranteed that the answer with at least two common vertices exists for the given tree.

The length of the path is the number of edges in it.

The simple path is the path that visits each vertex at most once.

Input

The first line contains an integer \(n\) — the number of vertices in the tree \((6 \le n \le 2 \cdot 10^5)\).

Each of the next \(n−1\) lines describes the edges of the tree.

Edge \(i\) is denoted by two integers \(u_i\) and \(v_i\), the labels of vertices it connects \((1\le u_i,v_i\le n, u_i \neq v_i)\).

It is guaranteed that the given edges form a tree.

It is guaranteed that the answer with at least two common vertices exists for the given tree.

Output

Print any two pairs of vertices satisfying the conditions described in the problem statement.

It is guaranteed that it is possible to choose such pairs for the given tree.

Examples

Input

7
1 4
1 5
1 6
2 3
2 4
4 7

Output

3 6
7 5

Input

9
9 3
3 5
1 2
4 3
4 7
1 7
4 6
3 8

Output

2 9
6 8

Input

10
6 8
10 3
3 7
5 8
1 7
7 2
2 9
2 8
1 4

Output

10 6
4 5

Input

11
1 2
2 3
3 4
1 5
1 6
6 7
5 8
5 9
4 10
4 11

Output

9 11
8 10

Note

The picture corresponding to the first example:

The intersection of two paths is \(2\) (vertices \(1\) and \(4\)) and the total length is \(4+3=7\).

The picture corresponding to the second example:

The intersection of two paths is \(2\) (vertices \(3\) and \(4\)) and the total length is \(5+3=8\).

The picture corresponding to the third example:

The intersection of two paths is \(3\) (vertices \(2\), \(7\) and \(8\)) and the total length is \(5+5=10\).

The picture corresponding to the fourth example:

The intersection of two paths is \(5\)(vertices \(1\), \(2\), \(3\), \(4\) and \(5\)) and the total length is \(6+6=12\).

Solution

题意:给定一棵树,找两组点\((x_1, y_1)\)和\((x_2, y_2)\),使得\(x_1,y_1\)不在\(x_2\)和\(y_2\)之间的路径上,\(x_2,y_2\)不在\(x_1\)和\(y_1\)之间的路径上,要求:

  • \(x_1,y_1\)之间的路径与\(x_2,y_2\)之间的路径的重合边数最多
  • 满足第一个条件的前提下,两条路径的长度之和最大

我们考虑两条路径的公共路径,不妨记作\((x, y)\),\(x\)和\(y\)的LCA记作\(a\),则\(a\)或者是\(x\)和\(y\)中的一个,或者是\(x\)与\(y\)路径上的其他节点,所以我们先求出每个点的度大于2的后代的最大深度,以及每个点往父亲方向能够到达的最远距离,然后再一次DFS,对于任何一个点\(u\):

  • 如果\(u\)有两个孩子节点具有度大于2的后代,则尝试更新答案
  • 否则,若\(u\)只有一个孩子节点具有度大于2的后代,且\(u\)自身的度大于2,则尝试更新答案
#include <bits/stdc++.h>
using namespace std;
const int maxn = 200011;
struct triple {
  triple(int _u = 0, int _v1 = 0, int _v2 = 0) : u(_u), v1(_v1), v2(_v2) {}
  int u, v1, v2;
  bool operator<(const triple &b) const {return u < b.u;}
};
vector<int> w[maxn];
int deg[maxn], dep[maxn];
int x1, y1, x2, y2;
pair<pair<int, int>, triple> val[maxn]; // <<deg=3的后代(u)的最大深度, u到两个最远后代(v1, v2)的距离之和>, <u, v1, v2>>
pair<int, int> ans;
pair<int, int> mxdep[maxn], updis[maxn]; // <最远距离, u>
vector<pair<pair<int, int>, int>> downdis[maxn]; // <<后代(u)的最大深度, u>, 到该后代的路径上的第一个点>
void dfs1(int u, int d, int pre) {
  dep[u] = d;
  mxdep[u] = make_pair(d, u);
  for (int v : w[u]) {
    if (v == pre) continue;
    dfs1(v, d + 1, u);
    mxdep[u] = max(mxdep[u], mxdep[v]);
    downdis[u].push_back(make_pair(mxdep[v], v));
  }
  sort(downdis[u].begin(), downdis[u].end(), greater<pair<pair<int, int>, int>>());
}
void dfs2(int u, int pre) {
  if (~pre) {
    updis[u] = make_pair(1 + updis[pre].first, updis[pre].second);
    auto tp = downdis[pre][0].second == u ? downdis[pre][1].first : downdis[pre][0].first;
    if (downdis[pre].size() > 1) {
      updis[u] = max(updis[u], make_pair(tp.first + 1, tp.second));
    }
  } else {
    updis[u] = make_pair(0, u);
  }
  for (int v : w[u]) {
    if (v == pre) continue;
    dfs2(v, u);
  }
}
void dfs3(int u, int pre) {
  vector<pair<pair<pair<int, int>, triple>, int>> vec;
  for (int v : w[u]) {
    if (v == pre) continue;
    dfs3(v, u);
    if (val[v].first.first) {
      vec.push_back(make_pair(val[v], v));
    }
  }
  if (vec.size() >= 2) {
    sort(vec.begin(), vec.end(), greater<pair<pair<pair<int, int>, triple>, int>>());
    auto &x = vec[0].first, &y = vec[1].first;
    val[u] = x;
    int a = x.first.first + y.first.first - 2 * dep[u];
    int b = x.first.second + y.first.second;
    auto c = make_pair(a, b);
    if (c > ans) {
      ans = c;
      x1 = x.second.v1, y1 = y.second.v1;
      x2 = x.second.v2, y2 = y.second.v2;
    }
  } else {
    if (vec.size() == 1) {
      val[u] = vec[0].first;
    } else if (deg[u] >= 3) {
      assert(downdis[u].size() >= 2);
      auto &x = downdis[u][0].first, &y = downdis[u][1].first;
      int tp = x.first + y.first - 2 * dep[u];
      val[u] = make_pair(make_pair(dep[u], tp), triple(u, x.second, y.second));
    } else {
      val[u] = make_pair(make_pair(0, 0), triple());
    }
    if (vec.size() == 1 && deg[u] >= 3) {
      vector<pair<int, int>> cand;
      cand.push_back(updis[u]);
      int up = min(3, (int)downdis[u].size());
      for (int i = 0; i < up; ++i) {
        if (downdis[u][i].second == vec[0].second)  continue;
        cand.push_back(downdis[u][i].first);
      }
      assert(cand.size() >= 2);
      sort(cand.begin(), cand.end(), greater<pair<int, int>>());
      auto &x = vec[0].first;
      int a = x.first.first - dep[u];
      int b = x.first.second + cand[0].first + cand[1].first;
      auto c = make_pair(a, b);
      if (c > ans) {
        ans = c;
        x1 = x.second.v1, y1 = cand[0].second;
        x2 = x.second.v2, y2 = cand[1].second;
      }
    }
  }
}
int main() {
  int n;
  scanf("%d", &n);
  for (int i = 1; i < n; ++i) {
    int u, v;
    scanf("%d%d", &u, &v);
    w[u].push_back(v);
    w[v].push_back(u);
    ++deg[u]; ++deg[v];
  }
  ans = make_pair(0, 0);
  dfs1(1, 0, -1);
  dfs2(1, -1);
  dfs3(1, -1);
  printf("%d %d\n%d %d\n", x1, y1, x2, y2);
  return 0;
}

CodeForces 1073F Choosing Two Paths的更多相关文章

  1. [codeforces 293]B. Distinct Paths

    [codeforces 293]B. Distinct Paths 试题描述 You have a rectangular n × m-cell board. Some cells are alrea ...

  2. Codeforces 219D. Choosing Capital for Treeland (树dp)

    题目链接:http://codeforces.com/contest/219/problem/D 树dp //#pragma comment(linker, "/STACK:10240000 ...

  3. Codeforces 219D Choosing Capital for Treeland

    http://codeforces.com/problemset/problem/219/D 题目大意: 给出一棵树,但是它的边是有向边,选择一个城市,问最少调整多少条边的方向能使一个选中城市可以到达 ...

  4. (纪念第一道完全自己想的树DP)CodeForces 219D Choosing Capital for Treeland

    Choosing Capital for Treeland time limit per test 3 seconds memory limit per test 256 megabytes inpu ...

  5. CodeForces 219D Choosing Capit

    题目链接:http://codeforces.com/contest/219/problem/D 题目大意: 给定一个n个节点的数和连接n个节点的n - 1条有向边,现在要选定一个节点作为起始节点,从 ...

  6. Codeforces 219D - Choosing Capital for Treeland(树形dp)

    http://codeforces.com/problemset/problem/219/D 题意 给一颗树但边是单向边,求至少旋转多少条单向边的方向,可以使得树上有一点可以到达树上任意一点,若有多个 ...

  7. 【CodeForces】870 F. Paths

    [题目]F. Paths [题意]给定数字n,图上有编号为1~n的点,两点当且仅当gcd(u,v)≠1时有连边,定义d(u,v)为两点间最短距离(若不连通则为0),求Σd(u,v),1<=u&l ...

  8. Codeforces 219D Choosing Capital for Treeland:Tree dp

    题目链接:http://codeforces.com/problemset/problem/219/D 题意: 给你一棵树,n个节点. 树上的边都是有向边,并且不一定是从父亲指向儿子的. 你可以任意翻 ...

  9. Codeforces 643G - Choosing Ads(线段树)

    Codeforces 题目传送门 & 洛谷题目传送门 首先考虑 \(p>50\) 的时候怎么处理,也就是求一个区间的绝对众数.我们知道众数这个东西是不能用线段树直接维护的,因为对于区间 ...

随机推荐

  1. PostMan请求不到接口问题

    在些接口的时候经常需要调试,调试的有很多选择,比如swagger.postman,我就是使用过两个都用:为了避免被swagger坑到就再去用postman试试确认看行不行,结果太小白了还是遇到了一些问 ...

  2. Linux load average负载量分析与解决思路

    一.load average top命令中load average显示的是最近1分钟.5分钟和15分钟的系统平均负载.系统平均负载表示 系统平均负载被定义为在特定时间间隔内运行队列中(在CPU上运行或 ...

  3. C# 引用的程序集没有强名称

    首先查一下什么是强名称程序集,见百度百科帖子:http://baike.baidu.com/view/1145682.htm简单来说,就是为了解决Windows Dll Hell问题的,即不同公司开发 ...

  4. python基础学习21----进程

    python中的多线程其实并不是真正的多线程,如果想要充分地使用多核CPU的资源,在python中大部分情况需要使用多进程. 进程与线程的使用有很多相似之处,有关线程方面的知识请参考https://w ...

  5. wing ide 6.0 注册

    1.wing ide介绍 wing ide ,用过python的都知道是干嘛用的了吧,官网已经更新到6.0.0-1版本. 链接如下: Wing IDE Professional - Version 6 ...

  6. [Spark RDD_add_1] groupByKey & reduceBykey 的区别

    [groupByKey & reduceBykey 的区别] 在都能实现相同功能的情况下优先使用 reduceBykey Combine 是为了减少网络负载 1. groupByKey 是没有 ...

  7. Linux nmap命令详解

    nmap,也就是Network Mapper,是Linux下的网络扫描和嗅探工具包. nmap是在网络安全渗透测试中经常会用到的强大的扫描器.功能之强大,不言而喻.下面介绍一下它的几种扫描命令.具体的 ...

  8. 利用windows的计划任务和eKing.CmdReadFileAndSendEmailOper(控制台小程序)实现远程登录服务器的邮件告警提醒

    一.场景摘要: 1.windows计划任务中,有一个用户登录时候触发的事件 2.cmd命令:netstat -ano   | find "3389" 可以看到当前远程登录的IP 3 ...

  9. selenium - pycharm三种案例运行模式

    1.unittest 运行单个用例 (1)将鼠标放到对应的用例,右键运行即可 2.unittest运行整个脚本案例 将鼠标放到if __name__ == "__main__": ...

  10. fedora安装视频播放器

    添加RPMFusion仓库后才能安装VLC.Mplayer,其他库中没有 直接 sudo dnf install vlc sudo dnf install mplayer