TopCoder SRM 590 FoxAndCity | 2017 ICPC Xian E

Problem

Description

As we know, Naomi is poor at math. But Naomi practices math problems every day. The following is one of them.

Naomi has a non-directed connected graph with nn vertices labelled from 11 to nn and mm edges. The length of each edge is 11. Naomi needs to add some edges (lengths of which should be 11 as well) to the graph such that each of them connects two different vertices and minimize the cost of the graph.

We define dist[i] as the length of the shortest path between vertex 11 and vertex ii. Vertex ii has a value A[i]A[i]. The cost of the graph equals to ${\sum }_{i=1}^n(A[i]-dist[i]{)}^2$
Can you help her?

Input

The input contains multiple test cases. （No more than 2020）

In each test case:

The first line contains two numbers nn, mm. ($1\le n\le 40,0\le m\le 1600$)

The following m lines each contains two numbers $x$ , $y$ $(1\le x,y\le n)$ denoting that there is an edge between $x$ and $y$.

And the last line of each test case contains n numbers denoting the array $A$. $0\le A[i]\le 1000$.

Output

For each test case, print the minimum cost of the graph in a single line.

Sample Input

4 3
1 2
2 3
3 4
0 3 3 3

Sample Output

5

题目大意

给定一个无向图，每条边的边权固定为 1，现在你可以自己加一些边权也为 1 的边，最小化
$$\sum _{i=1}^n(A_i-dis{t}_i{)}^2$$
其中 $dis{t}_i$ 表示第 i 个点到 1 的最短路长度。

题解

现在我们要最小化 ${\sum }_{i=1}^m(A_i-dis{t}_i{)}^2$，因此我们要得到一个可行的$dist$的解，由于我们现在要求最短路，因此加边以后对于原来的每条边都要满足松弛条件：$\left|dis{t}_i-dis{t}_j\right|\le 1$，而且 $dis{t}_1=0$，因此实际上我们现在存在一个需要最优化的函数以及一些限定条件，看起来和线性规划很像呢。但其实有问题，因为最优化函数包含平方项，因此：
$$\begin{gathered} z=\sum _{i=1}^n(A_i-x_i{)}^2 \\ s.t.\{\begin{array}{l}\left|x_i-x_j\right|\le 1(adj(i,j))\\ 0\le x_i<n\\ x_1=0\end{array} \end{gathered}$$
是不能直接单纯形的。但是因为这道题的数据范围很小，我们可以修改变量。为了使变量和平方没有关系，我们将 $dis{t}_i$ 拆成 $x_{i,j}=1$ 表示 $dis{t}_i=j$，这样我们的规划式变成：
$$\begin{gathered} z=\sum _{i=1}^n\sum _{j=0}^{n-1}{x}_{i,j}(A_i-j{)}^2 \\ s.t.\{\begin{array}{l}{\sum }_{j=0}^{n-1}{x}_{i,j}=1(i\in [1,n])\\ x_{1,0}=1\\ 若x_{i,p}=1,x_{j,q}=1，那么|p-q|\le 1\end{array} \end{gathered}$$
如果我们换一种形式，令 $x_{i,j}=1$ 表示 $dis{t}_i\le j$，那么式子长得更简单一些
$$\begin{gathered} z=\sum _{i=1}^n\sum _{j=0}^{n-1}(x_{i,j}-x_{i,j-1})(A_i-j{)}^2 \\ s.t.\{\begin{array}{l}{x}_{i,n-1}=1\\ x_{1,0}=1\\ x_{i,p}=0,x_{j,q}=1不合法(p>q)\end{array} \end{gathered}$$
其中对于 $z$，因为$x_{i,j}=1$表示$dis{t}_i\le j$，所以当 $x_{i,j}=1,x_{i,j-1}=0$时表示$dis{t}_i=j$，所以是 $(x_{i,j}-x_{i,j-1})(A_i-j{)}^2$ 。
第二个判定式的意思是说：$dis{t}_i>q$ and $dis{t}_j\le q$的情况不合法。
实际上这种表示方法仍然不方便处理，再转化成最小割的形式：
min ⁡ { ∑ max ⁡ { 0 , x i , k − x i , k − 1 } ⋅ ( A i − k ) 2 + ∑ max ⁡ { 0 , 1 − x i , n − 1 } ⋅ ∞ + ∑ max ⁡ { 0 , x i , k − 1 − x j , k } ⋅ ∞ } \min\left\{\sum \max\{0,x_{i,k}-x_{i,k-1}\}\cdot (A_i-k)^2+ \sum\max\{0,1-x_{i,n-1}\}\cdot \infin+ \sum\max\{0,x_{i,k-1}-x_{j,k}\}\cdot \infin \right\} min{∑max{0,xi,k​−xi,k−1​}⋅(Ai​−k)2+∑max{0,1−xi,n−1​}⋅∞+∑max{0,xi,k−1​−xj,k​}⋅∞}
表示限制 $x_{i,n-1}=1$，且$i<k$时$j>k$不合法。

构出来的图长这样（没错就是原 wiki 的图，我盗走了），节点 $d_i\le j$ 即为我给出的表达式 $x_{i,j}$。其中 i 和 j 在原图中有边相连，因此要限制松弛关系，所以两行的节点直接要连边，边权 inf（图中省略了）。同时，因为点 1 的 dist 必为 0，因此 $c(1,j)=\infty (j>0)$

（这篇博客是为了解释为什么是 $d_j\le k$ 连向 $d_i\le k-1$ 而不是反过来连）

#include <bits/stdc++.h>
using namespace std;
#define FOR(i,j,k) for(int i=j;i<=k;++i)
#define rep(i,j,k) for(int i=j;i<k;++i)
const int N = 2005, M = 5e6 + 10, inf = 0x3f3f3f3f;
int h[N], p[M], v[M], w[M], edge = 1, s, t;
int level[N], cur[N], vis[N], q[N], a[N];

void add(int a, int b, int c) {
    p[++edge] = h[a]; v[edge] = b; w[edge] = c; h[a] = edge;
    p[++edge] = h[b]; v[edge] = a; w[edge] = 0; h[b] = edge;
}

bool bfs() {
    int i, x, f = 0, r = 0;
    FOR(i,0,t) level[i] = -1;
    q[r++] = s; level[s] = 0;
    while (f < r) {
        x = q[f++];
        for (i = h[x]; i; i = p[i])
            if (w[i] && level[v[i]] == -1) {
                level[v[i]] = level[x] + 1;
                q[r++] = v[i];
            }
    }
    return level[t] != -1;
}

int dfs(int x, int low) {
    int i, tmp, res = 0;
    if (x == t) return low;
    for (i = cur[x]; i && res < low; i = p[i])
        if (w[i] && level[v[i]] == level[x] + 1) {
            tmp = dfs(v[i], min(low - res, w[i]));
            w[i] -= tmp; w[i ^ 1] += tmp; res += tmp;
            if (w[i]) cur[x] = i;
        }
    if (!res) level[x] = -1;
    return res;
}

int dinic() {
    int ans = 0, i;
    while (bfs()) {
        FOR(i,0,t) cur[i] = h[i];
        ans += dfs(s, inf);
    }
    return ans;
}

int main() {
    int n, m;
    while (scanf("%d%d", &n, &m) == 2) {
        s = n * n; t = n * n + 1;
        FOR(i,0,t) h[i] = 0;
        edge = 1;

        rep(e,0,m) {
            int i, j;
            scanf("%d%d", &i, &j); --i, --j;
            rep(k,1,n) {
                add(i * n + k, j * n + k - 1, inf);
                add(j * n + k, i * n + k - 1, inf);
            }
        }

        rep(i,0,n) scanf("%d", &a[i]);

        rep(k,1,n) {
            add(k - 1, k, inf);
        }
        add(s, 0, a[0] * a[0]);
        add(n - 1, t, inf);

        rep(i,1,n) {
            rep(k,0,n) {
                add(k == 0 ? s : i * n + k - 1,
                    i * n + k,
                    (a[i] - k) * (a[i] - k));
            }
            add(i * n + n - 1, t, inf);
        }

        printf("%d\n", dinic());
    }
}