Label
灵活变换求和次序的普通莫比乌斯反演
Description
给定 T ( T = 1 0 4 ) T(T=10^4) T(T=104)组 n , m ( 1 ≤ n , m ≤ 1 0 7 ) n,m(1\leq n,m\leq 10^7) n,m(1≤n,m≤107),求:
∑ i = 1 n ∑ j = 1 m [ ( i , j ) ∈ p r i m e ] \sum_{i=1}^{n}\sum_{j=1}^{m}[(i,j)\in prime] i=1∑nj=1∑m[(i,j)∈prime]
Solution
∑ i = 1 n ∑ j = 1 m [ ( i , j ) ∈ p r i m e ] \sum_{i=1}^{n}\sum_{j=1}^{m}[(i,j)\in prime] i=1∑nj=1∑m[(i,j)∈prime]
= ∑ p ∈ p r i m e p ≤ m i n ( n , m ) ∑ i = 1 n ∑ j = 1 m [ ( i , j ) = p ] =\sum_{p\in prime}^{p\le min(n,m)}\sum_{i=1}^{n}\sum_{j=1}^{m}[(i,j)=p] =p∈prime∑p≤min(n,m)i=1∑nj=1∑m[(i,j)=p]
= ∑ p ∈ p r i m e p ≤ m i n ( n , m ) ∑ i = 1 ⌊ n p ⌋ ∑ j = 1 ⌊ m p ⌋ [ ( i , j ) = 1 ] =\sum_{p\in prime}^{p\le min(n,m)}\sum_{i=1}^{\lfloor\frac{n}{p}\rfloor}\sum_{j=1}^{\lfloor\frac{m}{p}\rfloor}[(i,j)=1] =p∈prime∑p≤min(n,m)i=1∑⌊pn⌋j=1∑⌊pm⌋[(i,j)=1]
= ∑ p ∈ p r i m e p ≤ m i n ( n , m ) ∑ i = 1 ⌊ n p ⌋ ∑ j = 1 ⌊ m p ⌋ ∑ d ∣ ( i , j ) μ ( d ) =\sum_{p\in prime}^{p\le min(n,m)}\sum_{i=1}^{\lfloor\frac{n}{p}\rfloor}\sum_{j=1}^{\lfloor\frac{m}{p}\rfloor}\sum_{d|(i,j)}\mu(d) =p∈prime∑p≤min(n,m)i=1∑⌊pn⌋j=1∑⌊pm⌋d∣(i,j)∑μ(d)
= ∑ p ∈ p r i m e p ≤ m i n ( n , m ) ∑ d = 1 m i n ( ⌊ n p ⌋ , ⌊ m p ⌋ ) μ ( d ) ∑ i = 1 ⌊ n p ⌋ ∑ j = 1 ⌊ m p ⌋ [ d ∣ ( i , j ) ] =\sum_{p\in prime}^{p\le min(n,m)}\sum_{d=1}^{min(\lfloor\frac{n}{p}\rfloor,\lfloor\frac{m}{p}\rfloor)}\mu(d)\sum_{i=1}^{\lfloor\frac{n}{p}\rfloor}\sum_{j=1}^{\lfloor\frac{m}{p}\rfloor}[d|(i,j)] =p∈prime∑p≤min(n,m)d=1∑min(⌊pn⌋,⌊pm⌋)μ(d)i=1∑⌊pn⌋j=1∑⌊pm⌋[d∣(i,j)]
= ∑ p ∈ p r i m e p ≤ m i n ( n , m ) ∑ d = 1 m i n ( ⌊ n p ⌋ , ⌊ m p ⌋ ) μ ( d ) ∑ i = 1 ⌊ n p d ⌋ ∑ j = 1 ⌊ m p d ⌋ =\sum_{p\in prime}^{p\le min(n,m)}\sum_{d=1}^{min(\lfloor\frac{n}{p}\rfloor,\lfloor\frac{m}{p}\rfloor)}\mu(d)\sum_{i=1}^{\lfloor\frac{n}{pd}\rfloor}\sum_{j=1}^{\lfloor\frac{m}{pd}\rfloor} =p∈prime∑p≤min(n,m)d=1∑min(⌊pn⌋,⌊pm⌋)μ(d)i=1∑⌊pdn⌋j=1∑⌊pdm⌋
= ∑ p ∈ p r i m e p ≤ m i n ( n , m ) ∑ d = 1 m i n ( ⌊ n p ⌋ , ⌊ m p ⌋ ) μ ( d ) ⌊ n p d ⌋ ⌊ m p d ⌋ ( 1 ) =\sum_{p\in prime}^{p\le min(n,m)}\sum_{d=1}^{min(\lfloor\frac{n}{p}\rfloor,\lfloor\frac{m}{p}\rfloor)}\mu(d)\lfloor\frac{n}{pd}\rfloor\lfloor\frac{m}{pd}\rfloor(1) =p∈prime∑p≤min(n,m)d=1∑min(⌊pn⌋,⌊pm⌋)μ(d)⌊pdn⌋⌊pdm⌋(1)
我们按照常规反演处理方法化简到这里时不难发现:后面的这个式子虽然可以数论分块处理,但此题的数据范围来讲,最前面求和符号代表的枚举素数操作的复杂度无法令人接受。此处,我们必须将素数求和号拿到式子后面去。
事实上,如果把 d d d的求和号提到最前面,后面的 ⌊ n p d ⌋ ⌊ m p d ⌋ \lfloor\frac{n}{pd}\rfloor\lfloor\frac{m}{pd}\rfloor ⌊pdn⌋⌊pdm⌋的计算还是无法避开枚举素数。事实上,最好的处理办法是:将原式改写为第一层求和符号枚举 p d pd pd的式子。
考虑在当前最简式 = ∑ p ∈ p r i m e p ≤ m i n ( n , m ) ∑ d = 1 m i n ( ⌊ n p ⌋ , ⌊ m p ⌋ ) μ ( d ) ⌊ n p d ⌋ ⌊ m p d ⌋ =\sum_{p\in prime}^{p\le min(n,m)}\sum_{d=1}^{min(\lfloor\frac{n}{p}\rfloor,\lfloor\frac{m}{p}\rfloor)}\mu(d)\lfloor\frac{n}{pd}\rfloor\lfloor\frac{m}{pd}\rfloor =p∈prime∑p≤min(n,m)d=1∑min(⌊pn⌋,⌊pm⌋)μ(d)⌊pdn⌋⌊pdm⌋里 p , d p,d p,d的含义:枚举素数的同时,我们又枚举所有 p p p的倍数 p d pd pd。所以,考虑当 p d = i pd=i pd=i时, p , d p,d p,d需满足什么条件才使得其对 ⌊ n i ⌋ ⌊ m i ⌋ \lfloor\frac{n}{i}\rfloor\lfloor\frac{m}{i}\rfloor ⌊in⌋⌊im⌋有贡献,我们可以直接将(1)式变换成如下式子:
∑ i = 1 n ⌊ n i ⌋ ⌊ m i ⌋ ∑ p ∈ p r i m e , p ∣ i p ≤ i μ ( ⌊ i p ⌋ ) ( 2 ) \sum_{i=1}^{n}\lfloor\frac{n}{i}\rfloor\lfloor\frac{m}{i}\rfloor\sum_{p\in prime,p|i}^{p\leq i}\mu(\lfloor\frac{i}{p}\rfloor)(2) i=1∑n⌊in⌋⌊im⌋p∈prime,p∣i∑p≤iμ(⌊pi⌋)(2)
(其实上述式子 p d ( = i ) pd(=i) pd(=i)下界应为2,但 p d = 1 pd=1 pd=1时对应的项值为0,对答案无影响。事实上,(从某篇文章中看到的)我们通过变换求和次序的好处之一就在于:大大省略了对答案无影响的0项的枚举计算。)
对于 ∑ p ∈ p r i m e , p ∣ i p ≤ i μ ( ⌊ i p ⌋ ) \sum_{p\in prime,p|i}^{p\leq i}\mu(\lfloor\frac{i}{p}\rfloor) ∑p∈prime,p∣ip≤iμ(⌊pi⌋),在 O ( n ) O(n) O(n)预处理出 μ ( 1 ) ∼ μ ( n ) \mu(1)\sim \mu(n) μ(1)∼μ(n)的值后,可利用未经优化的埃氏筛预处理。之后便可对(2)式进行数论分块。
算法时间复杂度: O ( n l o g l o g n ) O(nloglogn) O(nloglogn)(瓶颈在于埃氏筛)。
Code
#include<cstdio>
#include<iostream>
#define ri register int
#define ll long long
using namespace std;
const int MAXN=1e7+20;
int T,N,M,prime[MAXN],cnt,mu[MAXN];
ll sum[MAXN],ans;
bool notprime[MAXN];
void Eulasieve()
{
mu[1]=1,notprime[1]=true;
for(ri i=2;i<=MAXN;++i)
{
if(!notprime[i]) prime[++cnt]=i,mu[i]=-1;
for(ri j=1;j<=cnt&&i*prime[j]<=MAXN;++j)
{
notprime[i*prime[j]]=true;
if(i%prime[j]==0) break;
else mu[i*prime[j]]=-mu[i];
}
}
}
void Estsieve()
{
sum[1]=0;
for(ri i=2;i<=MAXN;++i)
if(!notprime[i])
for(ri j=1;i*j<=MAXN;++j) sum[i*j]+=(ll)mu[j];
for(ri i=1;i<=MAXN;++i) sum[i]=sum[i-1]+sum[i];
}
int main()
{
std::ios::sync_with_stdio(false);
Eulasieve(),Estsieve();
cin>>T;
for(ri op=1;op<=T;++op)
{
cin>>N>>M;
if(N>M) swap(N,M);
ans=0LL;
for(ri l=2,r;l<=N;l=r+1)
{
r=min(N/(N/l),M/(M/l));
ans+=(ll)(N/l)*(ll)(M/l)*(sum[r]-sum[l-1]);
}
cout<<ans<<'\n';
}
return 0;
}