Description
给定(n*n)的矩阵(B)和(1*n)的矩阵(C)。求一个(1*N)的01矩阵(A)使得((AB-C)A^T)最大。(A^T)是(A)的转置。(nleq500)，所有输入为不超过(1000)的非负整数。
Solution
[egin{aligned}
(AB)_i&=sum_{j=1}^n A_jB_{j,i}\
(AB-C)_i&=left(sum_{j=1}^n A_jB_{j,i}
ight) - C_i\
(AB-C)A^T&=sum_{i=1}^n A_ileft(sum_{j=1}^nA_jB_{j,i}
ight)-A_iC_i\
&=sum_{i,j=1dots n} A_iA_jB_{i,j} -sum_{i=1}^nA_iC_i
end{aligned}
]
观察最后一个式子，可以看成：
(n)个物品，第(i,j)两个物品同时选会获得(B_{i,j})的收益；选第(i)个物品会付出(C_i)的代价；求最大净收益。
最小割即可。
Code
#include <algorithm>
#include <cstdio>
#include <cstring>
const int N = 500;
const int NN = 500000;
const int M = 3000050;
int pre[NN], nxt[M], to[M], ret[M], cnt;
int dis[NN], que[NN];
bool BFS(int S, int T) {
int hd = 0, tl = 0;
memset(dis, -1, sizeof dis);
for (dis[que[tl++] = S] = 0; hd < tl; ++hd)
for (int i = pre[que[hd]]; ~i; i = nxt[i])
if (ret[i] && !~dis[to[i]]) dis[que[tl++] = to[i]] = dis[que[hd]] + 1;
return dis[T] != -1;
}
int DFS(int x, int T, int maxf) {
if (!maxf) return 0;
if (x == T) return maxf;
int ans = 0;
for (int i = pre[x]; ~i; i = nxt[i])
if (ret[i] && dis[to[i]] == dis[x] + 1) {
int t = DFS(to[i], T, std::min(maxf - ans, ret[i]));
ret[i] -= t; ret[i ^ 1] += t;
ans += t;
}
if (ans < maxf) dis[x] = -1;
return ans;
}
int solve(int S, int T) {
int ans = 0;
while (BFS(S, T)) ans += DFS(S, T, 1000000000);
return ans;
}
inline void addEdge(int x, int y, int c) {
nxt[cnt] = pre[x];
ret[cnt] = c;
to[pre[x] = cnt++] = y;
nxt[cnt] = pre[y];
ret[cnt] = 0;
to[pre[y] = cnt++] = x;
}
int main() {
int n;
scanf("%d", &n);
memset(pre, -1, sizeof pre);
int S = n * (n + 1), T = S + 1;
int ans = 0;
for (int i = 0; i < n; ++i)
for (int j = 0, v; j < n; ++j) {
scanf("%d", &v);
ans += v;
int t = (i + 1) * n + j;
addEdge(t, T, v);
addEdge(i, t, 1000000000);
addEdge(j, t, 1000000000);
}
for (int i = 0, v; i < n; ++i) {
scanf("%d", &v);
addEdge(S, i, v);
}
printf("%d
", ans - solve(S, T));
return 0;
}