• loj #6686. Stupid GCD 莫比乌斯反演+杜教筛


    题目描述

    传送门

    (sumlimits_{i=1}^ngcd(sqrt[3]{i},i))

    (n leq 10^{30})

    分析

    毒瘤题,读入和计算要用 (int128)

    首先枚举 (sqrt[3]{i}) 的值

    (sumlimits_{d=1}^{sqrt[3]{n}}sumlimits_{i=1}^n[sqrt[3]{i}=d]gcd(i,d))

    方框里的那一个条件就等价于 (d^3 leq i <(d+1)^3)

    式子变为 (sumlimits_{d=1}^{sqrt[3]{n}}sumlimits_{i=d^3}^{min((d+1)^3-1,n)}gcd(i,d))

    (r=sqrt[3]{n}),把 (min) 拆掉

    (sumlimits_{d=1}^{r-1}sumlimits_{i=d^3}^{(d+1)^3-1}gcd(i,d)+sumlimits_{i=r^3}^{n}gcd(r,i))

    先考虑如何求出求 (sumlimits_{i=1}^ngcd(a,i))

    (egin{aligned} sumlimits_{i=1}^ngcd(a,i) &=sumlimits_{d|a}dsumlimits_{i=1}^{n/d}[gcd(i,a/d)=1] \&= sumlimits_{d|a}dsumlimits_{i=1}^{n/d}sumlimits_{k|gcd(i,a/d)}mu(k)\&= sumlimits_{d|a}dsumlimits_{k|(a/d)}mu(k)frac{n}{kd}\&=sumlimits_{T|a}frac{n}{T}sumlimits_{d|T}mu(d)frac{T}{d}\ &=sumlimits_{T|a}frac{n}{T}varphi(T) end{aligned})

    (egin{aligned}& sumlimits_{d=1}^{r-1}sumlimits_{i=d^3}^{(d+1)^3-1}gcd(i,d)+sumlimits_{i=r^3}^{n}gcd(r,i)\ &=sumlimits_{d=1}^{r-1}sumlimits_{T|d}varphi(T)(frac{(d+1)^3-1}{T}-frac{d^3-1}{T})+sumlimits_{T|r}varphi(T)(frac{n}{T}-frac{r^3-1}{T}) end{aligned})

    后面的这一部分就可以枚举约数,对于每一个约数暴力计算 (varphi)

    能过但是复杂度不太对

    正确的做法是把质因子筛出来,然后进行 (dfs),同时计算欧拉函数

    因为暴力能过所以就没有去写

    前面的这一部分可以继续化简

    (egin{aligned}&sumlimits_{d=1}^{r-1}sumlimits_{T|d}varphi(T)(frac{(d+1)^3-1}{T}-frac{d^3-1}{T})\ &=sumlimits_{d=1}^{r-1}sumlimits_{T|d}varphi(T)(frac{(d+1)^3-1}{T}-frac{d^3-1}{T})\&=sumlimits_{T=1}^{r-1}varphi(T)sumlimits_{d=1}^{(r-1)/T}(frac{(dT+1)^3-1}{T}-frac{(dT)^3-1}{T})\&=sumlimits_{T=1}^{r-1}varphi(T)sumlimits_{d=1}^{(r-1)/T}3Td^2+3d+1end{aligned})

    最后一步左边的部分是能整除 (T) 的,右边向下取整后会少一个 (1),前面有一个减号又变成了加 (1)

    这三项可以分别用杜教筛求解

    (sumlimits_{T=1}^{r-1}Tvarphi(T)sumlimits_{d=1}^{(r-1)/T}3d^2+sumlimits_{T=1}^{r-1}varphi(T)sumlimits_{d=1}^{(r-1)/T}(3d+1))

    筛一个 (varphi(n)) 再筛一个 (nvarphi(n)) 即可

    因为洛谷 (IDE) 被禁了,而本机编译不了(int128),所以推荐在这个网站编译

    要注意的是如果用系统自带的 (pow) 函数开三次根号精度可能会出问题,所以要用两个循环判一下

    代码

    #include<cstdio>
    #include<cmath>
    #include<algorithm>
    #define rg register
    const int maxn=8e6+5;
    #define ll __int128
    const int mod=998244353,inv2=499122177,inv6=166374059,Mod=4999999;
    template <class Tp>
    void read(Tp &x) {
    	static char ch;
    	static bool neg;
    	for (ch = neg = 0; ch < '0' || ch > '9'; neg |= (ch == '-'), ch = getchar());
    	for (x = 0; ch >= '0' && ch <= '9'; (x *= 10) += (ch ^ 48), ch = getchar());
    	neg && (x = -x);
    }
    struct has{
    	struct asd{
    		int nxt,val;
    		ll num;
    	}b[maxn];
    	int h[maxn],tot;
    	void insert(rg ll num,rg int val){
    		rg int now=num%Mod;
    		b[tot].nxt=h[now];
    		b[tot].val=val;
    		b[tot].num=num;
    		h[now]=tot++;
    	}
    	int cx(rg ll num){
    		rg int now=num%Mod;
    		for(rg int i=h[now];i!=0;i=b[i].nxt){
    			if(b[i].num==num) return b[i].val;
    		}
    		return -1;
    	}
    }ans1,ans2;
    int getsum1(rg ll now){
    	now%=mod;
    	return (now+1)*now%mod*inv2%mod;
    }
    int getsum2(rg ll now){
    	now%=mod;
    	return (2*now+1)%mod*(now+1)%mod*now%mod*inv6%mod;
    }
    bool not_pri[maxn];
    int phi1[maxn],phi2[maxn],sum1[maxn],sum2[maxn],mmax,pri[maxn],ans;
    void pre(){
    	not_pri[0]=not_pri[1]=1;
    	phi1[1]=1;
    	for(rg int i=2;i<=mmax;i++){
    		if(!not_pri[i]){
    			pri[++pri[0]]=i;
    			phi1[i]=i-1;
    		}
    		for(rg int j=1;j<=pri[0] && 1LL*i*pri[j]<=mmax;j++){
    			not_pri[i*pri[j]]=1;
    			if(i%pri[j]==0){
    				phi1[i*pri[j]]=1LL*phi1[i]*pri[j]%mod;
    				break;
    			} else {
    				phi1[i*pri[j]]=1LL*phi1[i]*phi1[pri[j]]%mod;
    			}
    		}
    	}
    	for(rg int i=1;i<=mmax;i++) phi2[i]=1LL*phi1[i]*i%mod;
    	for(rg int i=1;i<=mmax;i++){
    		sum1[i]=(sum1[i-1]+phi1[i])%mod;
    		sum2[i]=(sum2[i-1]+phi2[i])%mod;
    	}
    }
    int get_ans1(rg ll now){
    	if(now<=mmax) return sum1[now];
    	if(ans1.cx(now)!=-1) return ans1.cx(now);
    	rg int nans=getsum1(now);
    	for(rg ll l=2,r;l<=now;l=r+1){
    		r=(now/(now/l));
    		nans-=1LL*(r-l+1)*get_ans1(now/l)%mod;
    		nans=(nans+mod)%mod;
    	}
    	ans1.insert(now,nans);
    	return nans;
    }
    int get_ans2(rg ll now){
    	if(now<=mmax) return sum2[now];
    	if(ans2.cx(now)!=-1) return ans2.cx(now);
    	rg int nans=getsum2(now);
    	for(rg ll l=2,r;l<=now;l=r+1){
    		r=(now/(now/l));
    		nans-=1LL*(getsum1(r)-getsum1(l-1)+mod)%mod*get_ans2(now/l)%mod;
    		nans=(nans+mod)%mod;
    	}
    	ans2.insert(now,nans);
    	return nans;
    }
    ll n,sqr=1,tot;
    int get_phi(rg ll now){
    	if(now<=mmax) return phi1[now];
    	rg ll cs=now,nans=now;
    	for(rg ll i=2;i*i<=now;i++){
    		if(cs%i==0){
    			nans=nans/i*(i-1);
    			while(cs%i==0) cs/=i;
    			if(cs==1) return nans%mod;
    		}
    	}
    	if(cs>1) nans=nans/cs*(cs-1);
    	return nans%mod;
    }
    int solve(){
    	rg ll now;
    	rg int nans=0;
    	for(rg ll i=1;i*i<=sqr;i++){
    		if(sqr%i==0){
    			now=i;
    			nans=(nans+1LL*(n/now-(tot-1)/now)%mod*get_phi(now)%mod)%mod;
    			if(i*i!=sqr){
    				now=sqr/i;
    				nans=(nans+1LL*(n/now-(tot-1)/now)%mod*get_phi(now)%mod)%mod;
    			}
    		}
    	}
    	return nans;
    }
    int main(){
    	read(n);
    	mmax=8e6;
    	pre();
    	sqr=std::pow(n,1.0/3);
    	while((sqr+1)*(sqr+1)*(sqr+1)<=n) sqr++;
    	while((sqr-1)*(sqr-1)*(sqr-1)>=n) sqr--;
    	sqr--;
    	rg ll cs,tmp;
    	for(rg ll l=1,r;l<=sqr;l=r+1){
    		r=(sqr/(sqr/l));
    		cs=sqr/l;
    		tmp=(get_ans1(r)-get_ans1(l-1)+mod)%mod;
    		ans=(ans+1LL*(get_ans2(r)-get_ans2(l-1)+mod)%mod*3LL%mod*getsum2(cs)%mod)%mod;
    		ans=(ans+3LL*tmp%mod*getsum1(cs)%mod)%mod;
    		ans=(ans+tmp*cs%mod)%mod;
    	}
    	sqr++;
    	tot=sqr*sqr*sqr;
    	ans=(ans+solve())%mod;
    	printf("%d
    ",ans);
    	return 0;
    }
    
  • 相关阅读:
    如何简单使用tensorboard展示(二)
    如何简单使用tensorboard展示(一)
    Cypress 系列之----03 常用API
    Cypress 系列之----02 自定义命令Custom Commands
    Windows下启动Jmeter出现Not able to find Java executable or version问题解决方案
    linux命令行下文件名中包含特殊符号如何的处理方法
    jenkins高级篇 pipeline系列之-—01简介
    Jenkins部署报错问题解决----git低版本引发的问题
    存储过程--使用变量循环调用
    jenkins高级篇 pipeline 系列之-—06 实现自动打增量包
  • 原文地址:https://www.cnblogs.com/liuchanglc/p/14261546.html
Copyright © 2020-2023  润新知