2020 CCPC 威海站 L Clock Master （数论+分组背包）

L Clock Master

Description

With the rapid development of society, the demand for high-precision clocks is constantly rising. Recently, the China Clock Production Company is developing a new type of clock, which can represent a wide range of times.

The novel clock displays the current time in an unusual fashion. The clock consists of several pointers, each controlled by a gear. All gears rotate synchronously – one tooth per period. However, the numbers of teeth of the gears may differ. If a gear has t teeth, then the corresponding pointer can point to t different directions, denoted $0,1,2,\cdots ,t-1$, respectively, where 0 is the initial direction. Furthermore, if a clock is equipped with n pointers, the i-th of which is controlled by a t​i​​-tooth gear, then the i-th pointer will point to kmodt​i​​ after k periods of time.

The price for a t-tooth gear is t yuan. Given a total budget of $b$ yuan, you need to design a combination of gears, such that the number of valid combinations of directions of pointers is maximized, and the total cost on gears does not exceed the budget. A combination of directions ($d1,d2,\cdots ,dn$) is valid, if it can be written for some nonnegative integer $k$, where t​i​​ is the number of teeth of the $ i$-th gear. Since the answer may be too large, output the answer in natural logarithm (logarithm with base $e=2.718281828\cdots$).
Input Specification:

The first line of input is a single integer $T(1\le T\le 30000)$, indicating the number of test cases. Each test case is a single line of an integer $b(1\le b\le 30000)$, denoting the total budget.
Output Specification:

For each test case, print the natural logarithm, within an absolute or relative error of no more than $10^{-6}$​​, of the maximum number of valid combinations, in a single line.

Sample Input 1:

3
2
7
10

Sample Output:

0.693147181
2.484906650
3.401197382

Notes:

For the second sample data, a $3$-tooth gear along with a $4$-tooth gear may yield $12$ different combinations of directions, with total cost exactly being $7$. So you should print the value of ln12, which is approximately $2.484906650$.

分析

题目要求可以提炼为：给定一个数$n(1\le n\le 30000)$，令$n=a_1+a_2+a_3+....$，求$lcm(a_1,a_2,a_3,.....)$的最大值。

进一步思考可以想到，要使$lcm$最大，就得保证$a_1,a_2,a_3,.....$均为不同的素数或素数的整数次幂。求30000内的素数可以用欧拉筛，筛出素数后再…挺惭愧的，赛场上想到这里就卡住了，到最后都每没出来怎么继续做，又打铁了。赛后一看题解恍然大悟，这就是一个典型的“分组背包”问题啊，因为条件“$a_1,a_2,a_3,.....$均为不同的素数或素数的整数次幂”，所以每个素数和它的整数次幂就相当于一组背包。把$30000$以内的所有素数都各自为一组，然后一组组的DP就行。

求结果的时候需要利用对数运算法则将$ln(a_1\ast a_2\ast a_3\ast ....)$转化为$ln(a_1)+ln(a_2)+ln(a_3)....$

简单算一下$x$以内的素数约为$x/ln(x)$个，30000以内约有3000个素数，极限情况下总执行代码数约$1.5\ast 10^8$。

可能是我算法太慢，这计算量还挺极限的，得提前求出来每个数的log值，否则DP的时候现求就TLE了。最后运行时间300+ms，勉强能接受吧。下面是照着大佬优化后的版本，耗时又再减了一半。

AC代码：

#include <bits/stdc++.h>
using namespace std;

const int maxn=30000+5;
bool number[maxn+5];
int prime[maxn+5];
void euler_sieve(){
	int i,j,cnt=0;
	memset(number,true,sizeof(number));
	memset(prime,0,sizeof(prime));
	
	for(int i=2;i<=maxn;i++){
		if(number[i]) prime[cnt++]=i;
		for(j=0;j<cnt&&prime[j]*i<=maxn;j++){
			number[prime[j]*i]=false;
			if(i%prime[j]==0) break; 
 		}
	}
}

double dp[maxn];
double t_log[maxn];

void dp_run(){
	for(int i=0;prime[i]!=0;i++){
		for(int j=maxn-1;j>=prime[i];j--){
			for(int k=prime[i];k<=j;k*=prime[i]){
				dp[j]=max(dp[j],dp[j-k]+t_log[k]);
			}
		}
	}
}

void init(){
	for(int i=0;i<maxn;i++){
		dp[i]=0.0;
		t_log[i]=log(i);
	}
	euler_sieve();
}

int main(){
	init();
	dp_run();
	int T;
	scanf("%d",&T);
	while(T--){
		int x;
		scanf("%d",&x);
		printf("%.9lf\n",dp[x]);
	}
	return 0;
}

参考资料：

【动态规划】分组背包_yu121380的博客-CSDN博客
2020CCPC(威海) - Clock Master(数论+分组背包)_Falcon的博客-CSDN博客