欢迎您访问程序员文章站本站旨在为大家提供分享程序员计算机编程知识!
您现在的位置是: 首页  >  IT编程

洛谷P4007 小 Y 和恐怖的奴隶主(期望dp 矩阵乘法)

程序员文章站 2024-01-30 11:42:40
题意 "题目链接" Sol 首先不难想到一种暴力dp,设$f[i][a][b][c]$表示还有$i$轮没打,场上有$a$个1血,$b$个2血,$c$个三血 发现状态数只有$s = 166$个,复杂度为$O(ns)$ 矩乘优化一下复杂度为$O(s^3 logn T)$,还是过不去。 因为每次询问都是独 ......

题意

题目链接

sol

首先不难想到一种暴力dp,设\(f[i][a][b][c]\)表示还有\(i\)轮没打,场上有\(a\)个1血,\(b\)个2血,\(c\)个三血

发现状态数只有\(s = 166\)个,复杂度为\(o(ns)\)

矩乘优化一下复杂度为\(o(s^3 logn t)\),还是过不去。

因为每次询问都是独立的,那么可以预处理出\(2^i\)的转移矩阵,回答询问只需要拿一个行向量去乘log个矩阵

构造矩阵的时候可以加一个列向量表示期望

#include<bits/stdc++.h>
#define ll long long 
using namespace std;
const int b = 60, mod = 998244353;
template <typename a, typename b> inline bool chmin(a &a, b b){if(a > b) {a = b; return 1;} return 0;}
template <typename a, typename b> inline bool chmax(a &a, b b){if(a < b) {a = b; return 1;} return 0;}
template <typename a, typename b> inline ll add(a x, b y) {if(x + y < 0) return x + y + mod; return x + y >= mod ? x + y - mod : x + y;}
template <typename a, typename b> inline void add2(a &x, b y) {if(x + y < 0) x = x + y + mod; else x = (x + y >= mod ? x + y - mod : x + y);}
ll mul(int x, int y) {return 1ll * x * y % mod;}

inline ll read() {
    char c = getchar(); ll x = 0, f = 1;
    while(c < '0' || c > '9') {if(c == '-') f = -1; c = getchar();}
    while(c >= '0' && c <= '9') x = x * 10 + c - '0', c = getchar();
    return x * f;
}
int fp(int a, int p) {
    int base = 1;
    while(p) {
        if(p & 1) base = mul(base, a);
        a = mul(a, a); p >>= 1;
    }
    return base;
}
int t, m, k;
 
namespace s3 {
    int id[11][11][11], cnt, lim;
    int ans[168];
    ll inv[11];
    
    struct ma {
        int m[168][168];
        ma() {
            memset(m, 0, sizeof(m));    
        }
        void init() {
            for(int i = 0; i <= lim; i++) m[i][i] = 1;
        }
        void print() {
            for(int i = 1; i <= lim; i++, puts(""))
                for(int j = 1; j <= lim; j++)
                    printf("%d ", m[i][j]);
        }
        ma operator * (const ma &rhs) const {
            ma gg = {};
            for(int i = 1; i <= lim; i++)
                for(int j = 1; j <= lim; j++) {
                    __int128 tmp = 0;
                    for(int k = 1; k <= lim; k++) 
                        tmp += mul(m[i][k], rhs.m[k][j]);
                    tmp %= mod;
                    gg.m[i][j] = tmp;
                }
                        
            return gg;
        }
    }f[b + 1];
    void pre() {
        for(int i = 1; i <= k + 1; i++) inv[i] = fp(i, mod - 2);
        for(int a = 0; a <= k; a++) 
            for(int b = 0; a + b <= k; b++)
                for(int c = 0; a + b + c <= k; c++)
                    id[a][b][c] = ++cnt;
        for(int a = 0; a <= k; a++) 
            for(int b = 0; a + b <= k; b++)
                for(int c = 0; a + b + c <= k; c++) {
                    int down = inv[a + b + c + 1], tag = (a + b + c < k), now = id[a][b][c];
                    if(a) f[0].m[now][id[a - 1][b][c]] = mul(a, down);
                    if(b) f[0].m[now][id[a + 1][b - 1][c + tag]] = mul(b, down);
                    if(c) f[0].m[now][id[a][b + 1][c - 1 + tag]] = mul(c, down);
                    f[0].m[now][now] = down;
                    f[0].m[now][cnt + 1] = down;
                }
        f[0].m[cnt + 1][cnt + 1] = 1;
        lim = cnt + 1;
        for(int i = 1; i <= b; i++) f[i] = f[i - 1] * f[i - 1];
    }
    int tmp[168];
    void mul(ma a) {
        memset(tmp, 0, sizeof(tmp));
        for(int j = 1; j <= lim; j++)
            for(int i = 1; i <= lim; i++)
                add2(tmp[j], 1ll * ans[i] * a.m[i][j] % mod);
        memcpy(ans, tmp, sizeof(tmp));
    }
    void matrixpow(ll p) {
        for(int i = 0; p; p >>= 1, i++)
            if(p & 1) 
                mul(f[i]);
    }   
    void work() {
        pre();
        while(t--) {
            ll n = read();
            memset(ans, 0, sizeof(ans)); ans[id[0][0][1]] = 1;
            matrixpow(n);
            cout << ans[cnt + 1] << '\n';
        }   
    }
}


namespace s2 {
    int id[11][11], cnt, lim;
    int ans[168];
    ll inv[11];
    
    struct ma {
        int m[168][168];
        ma() {
            memset(m, 0, sizeof(m));    
        }
        void init() {
            for(int i = 0; i <= lim; i++) m[i][i] = 1;
        }
        void print() {
            for(int i = 1; i <= lim; i++, puts(""))
                for(int j = 1; j <= lim; j++)
                    printf("%d ", m[i][j]);
        }
        ma operator * (const ma &rhs) const {
            ma gg = {};
            for(int i = 1; i <= lim; i++)
                for(int j = 1; j <= lim; j++) {
                    __int128 tmp = 0;
                    for(int k = 1; k <= lim; k++) 
                        tmp += mul(m[i][k], rhs.m[k][j]);
                    tmp %= mod;
                    gg.m[i][j] = tmp;
                }
                        
            return gg;
        }
    }f[b + 1];
    void pre() {
        for(int i = 1; i <= k + 1; i++) inv[i] = fp(i, mod - 2);
        for(int a = 0; a <= k; a++) 
            for(int b = 0; a + b <= k; b++)
                id[a][b] = ++cnt;
        for(int a = 0; a <= k; a++) 
            for(int b = 0; a + b <= k; b++) {
                int down = inv[a + b + 1], tag = (a + b < k), now = id[a][b];
                if(a) f[0].m[now][id[a - 1][b]] = mul(a, down);
                if(b) f[0].m[now][id[a + 1][b - 1 + tag]] = mul(b, down);
                f[0].m[now][now] = down;
                f[0].m[now][cnt + 1] = down;
            }
        f[0].m[cnt + 1][cnt + 1] = 1;
        lim = cnt + 1;
        for(int i = 1; i <= b; i++) f[i] = f[i - 1] * f[i - 1];
    }
    int tmp[168];
    void mul(ma a) {
        memset(tmp, 0, sizeof(tmp));
        for(int j = 1; j <= lim; j++)
            for(int i = 1; i <= lim; i++)
                add2(tmp[j], 1ll * ans[i] * a.m[i][j] % mod);
        memcpy(ans, tmp, sizeof(tmp));
    }
    void matrixpow(ll p) {
        for(int i = 0; p; p >>= 1, i++)
            if(p & 1) 
                mul(f[i]);
    }   
    void work() {
        pre();
        while(t--) {
            ll n = read();
            memset(ans, 0, sizeof(ans)); ans[id[0][1]] = 1;
            matrixpow(n);
            cout << ans[cnt + 1] << '\n';
        }   
    }
}

namespace s1 {
    int n,  f[12][9][9][9]; 
    int inv(int a) {
        return fp(a, mod - 2);
    }
    void work() {
        n = 11;
        for(int i = 1; i <= n; i++) {
            for(int a = 0; a <= k; a++) {
                for(int b = 0; a + b <= k; b++) {
                    for(int c = 0; a + b + c <= k; c++) {
                        int down = a + b + c + 1;
                        if(a) add2(f[i][a][b][c], mul(mul(a, inv(down)), f[i - 1][a - 1][b][c]));
                        if(b) {
                            if(down <= k) add2(f[i][a][b][c], mul(mul(b, inv(down)), f[i - 1][a + 1][b - 1 + (m == 2)][c + (m == 3)]));
                            else add2(f[i][a][b][c], mul(mul(b, inv(down)), f[i - 1][a + 1][b - 1][c]));
                        }
                        if(c) {
                            if(down <= k) add2(f[i][a][b][c], mul(mul(c, inv(down)), f[i - 1][a][b + 1 + (m == 2)][c - 1 + (m == 3)]));
                            else add2(f[i][a][b][c], mul(mul(c, inv(down)), f[i - 1][a][b + 1][c - 1]));
                        }
                        add2(f[i][a][b][c], mul(inv(down), f[i - 1][a][b][c] + 1));
                    }
                }
            }
        }
        while(t--) {
            int n = read();
            printf("%d\n", f[n][m == 1][m == 2][m == 3]);
        }
    }
}

int main() {
    t = read(); m = read(); k = read();
    if(m == 1) s1::work();
    else if(m == 2) s2::work();
    else s3::work();

    return 0;
}