HDU6607Easy Math Problem(min_25筛+杜教筛+拉格朗日插值)
题目大意
求
∑i=1n∑j=1ngcd(i,j)klcm(i,j)[gcd(i,j)∈prime]
\sum_{i=1}^n\sum_{j=1}^ngcd(i,j)^klcm(i,j)[gcd(i,j)\in prime]
i=1∑nj=1∑ngcd(i,j)klcm(i,j)[gcd(i,j)∈prime]
其中1≤n≤1010,1≤k≤1001\le n\le 10^{10},1\le k\le 1001≤n≤1010,1≤k≤100
解题思路
∑i=1n∑j=1ngcd(i,j)klcm(i,j)[gcd(i,j)∈prime]=∑p∑i=1⌊np⌋∑j=1⌊np⌋pk+1⋅(i⋅j)[gcd(i,j)=1]=∑ppk+1∑i=1⌊np⌋∑j=1⌊np⌋⋅(i⋅j)[gcd(i,j)=1] \begin{aligned} \sum_{i=1}^n\sum_{j=1}^ngcd(i,j)^klcm(i,j)[gcd(i,j)\in prime]&=\sum_p\sum_{i=1}^{\left\lfloor\frac{n}{p}\right\rfloor}\sum_{j=1}^{\left\lfloor\frac{n}{p}\right\rfloor}p^{k+1}\cdot(i\cdot j)[gcd(i,j)=1]\\ &=\sum_pp^{k+1}\sum_{i=1}^{\left\lfloor\frac{n}{p}\right\rfloor}\sum_{j=1}^{\left\lfloor\frac{n}{p}\right\rfloor}\cdot(i\cdot j)[gcd(i,j)=1] \end{aligned} i=1∑nj=1∑ngcd(i,j)klcm(i,j)[gcd(i,j)∈prime]=p∑i=1∑⌊pn⌋j=1∑⌊pn⌋pk+1⋅(i⋅j)[gcd(i,j)=1]=p∑pk+1i=1∑⌊pn⌋j=1∑⌊pn⌋⋅(i⋅j)[gcd(i,j)=1]
对于∑i=1⌊np⌋∑j=1⌊np⌋⋅(i⋅j)[gcd(i,j)=1]\sum_{i=1}^{\left\lfloor\frac{n}{p}\right\rfloor}\sum_{j=1}^{\left\lfloor\frac{n}{p}\right\rfloor}\cdot(i\cdot j)[gcd(i,j)=1]∑i=1⌊pn⌋∑j=1⌊pn⌋⋅(i⋅j)[gcd(i,j)=1]单独分析
∑i=1n∑j=1nj⋅i[gcd(i,j)=1]
\sum_{i=1}^n\sum_{j=1}^nj\cdot i[gcd(i,j)=1]
i=1∑nj=1∑nj⋅i[gcd(i,j)=1]
可以对原本的区间进行拆分
[1≤i≤n][1≤j≤n]=[1≤i≤n][1≤j<i]+[1≤i<j][1≤j≤n]+[i=j][1≤i≤n]
[1\le i\le n][1\le j\le n]=[1\le i\le n][1\le j< i]+[1\le i<j][1\le j\le n]+[i=j][1\le i\le n]
[1≤i≤n][1≤j≤n]=[1≤i≤n][1≤j<i]+[1≤i<j][1≤j≤n]+[i=j][1≤i≤n]
对第一部分求和
∑i=1ni∑j=1i−1j[gcd(i,j)]
\sum_{i=1}^ni\sum_{j=1}^{i-1}j [gcd(i,j)]
i=1∑nij=1∑i−1j[gcd(i,j)]
根据公式KaTeX parse error: \tag works only in display equations就有
∑i=1ni[i=1]+i2φ(i)2
\sum_{i=1}^n\frac{i[i=1]+i^2\varphi(i)}{2}
i=1∑n2i[i=1]+i2φ(i)
而区间的第二部分和第一部分实质上是一样的,那么前面两部分就是∑i=1ni[i=1]+i2φ(i)\sum_{i=1}^ni[i=1]+i^2\varphi(i)∑i=1ni[i=1]+i2φ(i)而第三部分除了i=j=1i=j=1i=j=1的情况以外,其总为0,而当其为1时答案恰好为1,故最终的答案就是
∑i=1n∑j=1nj⋅i[gcd(i,j)=1]=∑i=1ni2φ(i)
\sum_{i=1}^n\sum_{j=1}^nj\cdot i[gcd(i,j)=1]=\sum_{i=1}^ni^2\varphi(i)
i=1∑nj=1∑nj⋅i[gcd(i,j)=1]=i=1∑ni2φ(i)
而对于id2⋅φid^2\cdot \varphiid2⋅φ这个函数可以通过与id2id^2id2卷积
(id2⋅φ)∗(id2)=∑d∣nd2φ(d)n2d2=n2∑d∣nφ(d)=n3
(id^2\cdot\varphi)*(id^2)=\sum_{d|n}d^2\varphi(d)\frac{n^2}{d^2}=n^2\sum_{d|n}\varphi(d)=n^3
(id2⋅φ)∗(id2)=d∣n∑d2φ(d)d2n2=n2d∣n∑φ(d)=n3
其中n3n^3n3和n2n^2n2都有公式可以O(1)O(1)O(1)求出,因此可以很方便地应用杜教筛求解
而对于∑ppk+1\sum_pp^{k+1}∑ppk+1可以通过min25筛中的子方案求出所有的整除分块中需要用到的前缀和
这个问题的主要复杂度就是通过拉格朗日插值处理出前n\sqrt nn的∑i=1nik+1\sum_{i=1}^ni^{k+1}∑i=1nik+1的前缀和,这部分的复杂度是O(kn)O(k\sqrt n)O(kn)
加上min25筛加上杜教筛的复杂度,总的复杂度就是O(kn+n34logn+n23)O(k\sqrt n+\frac{n^{\frac{3}{4}}}{logn}+n^{2\over 3})O(kn+lognn43+n32)
AC代码
#include<bits/stdc++.h>
using namespace std;
#define int long long
const int size=2e5+5;
const int mod=1e9+7;
typedef long long LL;
bool prime[size]; int p[size];
int tot=0;
int inv6,inv2;
int phi[size];
int sumiiphi[size];
int quick_pow(int a,int b)
{
register int ans=1;
while(b)
{
if(b&1) ans=1LL*ans*a%mod;
a=1LL*a*a%mod;
b>>=1;
}
return ans;
}
int inv[205];
void init()
{
tot=0;
phi[1]=1;
for(int i=1;i<size;i++) prime[i]=true;
for(int i=2;i<size;i++)
{
if(prime[i])
p[++tot]=i,phi[i]=i-1;
for(int j=1;j<=tot&&p[j]*i<size;j++)
{
prime[i*p[j]]=false;
if(i%p[j]==0)
{
phi[i*p[j]]=phi[i]*p[j];
break;
}
else
phi[i*p[j]]=phi[i]*phi[p[j]];
}
}
sumiiphi[0]=0;
for(int i=1;i<size;i++) sumiiphi[i]=(sumiiphi[i-1]+1LL*i*i%mod*phi[i]%mod)%mod;
inv6=quick_pow(6,mod-2);
inv2=quick_pow(2,mod-2);
inv[0]=inv[1]=1;
for(int i=2;i<=200;i++)
inv[i] = (mod - (mod / i)) * inv[mod % i] % mod;
for (int i = 1; i <= 200; ++i)inv[i] = inv[i] * inv[i - 1] % mod;
}
int tol;
int hk[size<<1];LL num[size<<1];
int pre[size];
int coeff[205];
void pre_lage(int k)
{
for(int i=1;i<=k+2;i++) coeff[i]=quick_pow(i,k);
coeff[0]=0;
for(int i=1;i<=k+2;i++) coeff[i]=(coeff[i-1]+coeff[i])%mod;
}
LL suf[205],bef[205];
int lage(int n,int k)
{
if(n<=k+2) return coeff[n];
bef[0] = suf[0] = 1;
for (int i = 1; i <= k + 2; ++i) {
bef[i] = bef[i - 1] * ((n - i) % mod) % mod;
suf[i] = suf[i - 1] * ((n + i - k - 3) % mod) % mod;
}
LL res = 0;
for (int i = 1; i <= k + 2; ++i) {
LL s = coeff[i] * bef[i - 1] % mod * suf[k - i + 2] % mod * inv[i - 1] % mod * inv[k + 2 - i] % mod;
if ((k + 2 - i) & 1)s = -s;
res += s;
res = (res % mod + mod) % mod;
}
return res;
}
int s;LL n;
int id1[size],id2[size];
inline int ID(LL x){return x<=s?x:tol-n/x+1;}
void get_h(int k)
{
s=sqrt(n);
while(1LL*s*s<=n) s++;
while(1LL*s*s>n) s--;
pre[0]=0;
for(register int i=1;p[i]<=s;i++)
{
pre[i]=(1LL*pre[i-1]+quick_pow(p[i],k+1))%mod;
}
tol=0;
for(int i=1;i<=s;i++) num[++tol]=i;
for(int i=s;i>=1;i--) if(n/i>s) num[++tol]=n/i;
pre_lage(k+1);
for(register int i=1;i<=tol;i++) hk[i]=(lage(num[i]%mod,k+1)-1+mod)%mod;
hk[0]=0;
int x=1;
for(int j=1;j<=tot&&p[j]<=s;j++)
{
while(num[x]<p[j]*p[j]) x++;
for(int i=tol;i>=x;i--)
{
hk[i]=((hk[i]-1LL*(pre[j]-pre[j-1]+mod)%mod*(hk[ID(num[i]/p[j])]-pre[j-1]))%mod+mod)%mod;
}
}
}
unordered_map<LL,int> mp;
inline int sum2(int n)
{
return 1LL*n*(n+1)%mod*(2*n+1)%mod*inv6%mod;
}
inline int sum3(int n)
{
int ans=1LL*n*(n+1)%mod*inv2%mod;
return 1LL*ans*ans%mod;
}
inline int S(LL n)
{
if(n<size) return sumiiphi[n];
if(mp.count(n)) return mp[n];
LL ans=sum3(n%mod);//n取模
for(LL l=2,r;l<=n;l=r+1)
{
r=n/(n/l);
ans=((ans-1LL*S(n/r)*(sum2(r%mod)-sum2((l-1)%mod)))%mod+mod)%mod;
}
return mp[n]=ans;
}
int32_t main()
{
int t;
scanf("%lld",&t);
int k;
init();
while(t--)
{
scanf("%lld%lld",&n,&k);
mp.clear();
get_h(k);
int ans=0;
for(int i=1;i<=tol;i++)
{
ans=(ans+1LL*(hk[i]-hk[i-1])%mod*S(n/num[i])%mod+mod)%mod;
}
printf("%lld\n",ans);
}
}