首页 > 代码库 > Codeforces Round #257 (Div. 2)B 矩阵快速幂
Codeforces Round #257 (Div. 2)B 矩阵快速幂
B. Jzzhu and Sequences
time limit per test
1 secondmemory limit per test
256 megabytesinput
standard inputoutput
standard outputJzzhu has invented a kind of sequences, they meet the following property:
You are given x and y, please calculate fn modulo 1000000007 (109?+?7).
Input
The first line contains two integers x and y (|x|,?|y|?≤?109). The second line contains a single integer n (1?≤?n?≤?2·109).
Output
Output a single integer representing fn modulo 1000000007 (109?+?7).
Sample test(s)
input
2 3 3
output
1
input
0 -1 2
output
1000000006
Note
In the first sample, f2?=?f1?+?f3, 3?=?2?+?f3, f3?=?1.
In the second sample, f2?=??-?1; ?-?1 modulo (109?+?7) equals (109?+?6).
题解
很简单的一个递推法的题目,我直接用矩阵快速幂去过了。
构造方法也很简单。
f1 + f3 = f2
f3 = f2 - f1
f3 = 1 * f2 + -1 * f1
f2 = 1 * f2 + 0 * f1
所以矩阵就是
1, -1
1, 0
然后就是取模运算和负数处理了。负数取模会返回负数。因为我是快速幂的写法,最后返回的数一定是在(-Mod, Mod)之间的,所以直接加上一个Mod,然后再取一次模就好了。
代码示例
/****************************************************************************** * COPYRIGHT NOTICE * Copyright (c) 2014 All rights reserved * ----Stay Hungry Stay Foolish---- * * @author :Shen * @name :B * @file :G:\My Source Code\【ACM】比赛\0719 - CF\B.cpp * @date :2014/07/19 20:57 * @algorithm :Matrix Fast Power Method ******************************************************************************/ #include <cstdio> #include <string> #include <cstring> #include <iostream> #include <algorithm> using namespace std; typedef long long int64; const int MAXN = 2; const int MAXM = 2; const int Mod = 1000000007; struct Matrax{ int n, m; int64 mat[MAXN][MAXM]; Matrax(): n(-1), m(-1){} Matrax(int _n, int _m): n(_n), m(_m){ memset(mat, 0, sizeof(mat)); } void Unit(int _s){ n = _s; m = _s; for (int i = 0; i < n; i++){ for (int j = 0; j < n; j++){ mat[i][j] = (i == j)? 1: 0; } } } void print(){ printf("n = %d, m = %d\n", n, m); for (int i = 0; i < n; i++){ for (int j = 0; j < m; j++) printf("%8d", mat[i][j]); printf("\n"); } } }; Matrax add_mod(const Matrax& a,const Matrax& b,const int64 mod){ Matrax ans(a.n, a.m); for (int i = 0; i < a.n; i++){ for (int j = 0; j < a.m; j++){ ans.mat[i][j] = (a.mat[i][j] + b.mat[i][j]) % mod; } } return ans; } Matrax mul(const Matrax& a,const Matrax& b){ Matrax ans(a.n, b.m); for (int i = 0; i < a.n; i++){ for (int j = 0; j < b.m; j++){ int64 tmp = 0; for (int k = 0; k < a.m; k++){ tmp += a.mat[i][k] * b.mat[k][j]; } ans.mat[i][j] = tmp; } } return ans; } Matrax mul_mod(const Matrax& a, const Matrax& b, const int mod){ Matrax ans(a.n, b.m); for (int i = 0; i < a.n; i++){ for (int j = 0; j < b.m; j++){ int64 tmp = 0; for (int k = 0; k < a.m; k++){ tmp += (a.mat[i][k] * b.mat[k][j]) % mod; } ans.mat[i][j] = tmp % mod; } } return ans; } Matrax pow_mod(const Matrax& a, int64 k, const int mod){ Matrax p(a.n, a.m), ans(a.n, a.m); p = a; ans.Unit(a.n); if (k==0) return ans; else if (k==1) return a; else { while (k){ if (k & 1){ ans = mul_mod(ans, p, mod); k--; } else { k /= 2; p = mul_mod(p, p, mod); } } return ans; } } int64 x, y, n, res; void solve(){ cin >> x >> y >> n; if (n == 1) res = x; else if (n == 2) res = y; else { Matrax ans(2, 1); //tmp = cef ^ (n - 2); //ans = tmp * beg; //res = ans.mat[0][0]; Matrax cef(2, 2); cef.mat[0][0] = 1; cef.mat[0][1] = -1; cef.mat[1][0] = 1; cef.mat[1][1] = 0; //cef.print(); Matrax beg(2, 1); beg.mat[0][0] = y; beg.mat[1][0] = x; Matrax tmp(2, 2); tmp = pow_mod(cef, n - 2, Mod); //tmp.print(); ans = mul_mod(tmp, beg, Mod); //ans.print(); res = ans.mat[0][0]; } if (res < 0) res += Mod; cout << res << endl; } int main() { solve(); return 0; }
Codeforces Round #257 (Div. 2)B 矩阵快速幂
声明:以上内容来自用户投稿及互联网公开渠道收集整理发布,本网站不拥有所有权,未作人工编辑处理,也不承担相关法律责任,若内容有误或涉及侵权可进行投诉: 投诉/举报 工作人员会在5个工作日内联系你,一经查实,本站将立刻删除涉嫌侵权内容。