好的序列 - 莫比乌斯反演 - 杜教筛 - 拉格朗日插值

题目大意：
令$f_k(n)$表示长度为$k$的序列，每个元素在$[1,n]$，并且$\gcd$为$1$的数量。求：${\sum }_{i=1}^n{f}_k(i),n\le 10^9,k\le 10^5$
题解：
$$\begin{gathered} Ans=\sum _{i=1}^n{f}_k(i)=\sum _{i=1}^n\sum _{j=1}^i{\lfloor \frac{i}{j}\rfloor }^k\mu (j)=\sum _{j=1}^n\mu (j)\sum _{i=1}^n{\lfloor \frac{i}{j}\rfloor }^k \\ =\sum _{j=1}^n\left[\left(\sum _{i=1}^{\lfloor \frac{n}{j}\rfloor -1}{i}^{k}j\right)+{\lfloor \frac{n}{j}\rfloor }^k\left(n-\lfloor \frac{n}{j}\rfloor j+1\right)\right] \\ =\sum _{j=1}^n\mu (j)\left(j{S}_k\left(\lfloor \frac{n}{j}\rfloor -1\right)+{\lfloor \frac{n}{j}\rfloor }^k(n+1)-{\lfloor \frac{n}{j}\rfloor }^{k+1}j\right) \\ =\sum _{j=1}^n\mu (j)j{S}_k\left(\lfloor \frac{n}{j}\rfloor -1\right)+\mu (j){\lfloor \frac{n}{j}\rfloor }^k(n+1)-\mu (j)j{\lfloor \frac{n}{j}\rfloor }^{k+1} \end{gathered}$$
其中：
$$S_k(n)=\sum _{i=1}^n{i}^k$$
然后对$\mu (j)$和$\mu (j)j$求杜教筛，对$S_k(n)$做插值即可。
取$\mathrm{b}\mathrm{l}\mathrm{o}\mathrm{c}\mathrm{k}\_\mathrm{s}\mathrm{i}\mathrm{z}\mathrm{e}=\sqrt{nk}$，可以做到$O\left(n^{\frac{2}{3}}+\sqrt{nk}\right)$
$\mathrm{s}\mathrm{t}\mathrm{d}$本意是要写一个多点插值把后半部分做到$O\left(\sqrt{n}lgk\right)$来着，但是数据出小了，就暴过去了……

#include<bits/stdc++.h>
#define gc getchar()
#define rep(i,a,b) for(int i=a;i<=b;i++)
#define Rep(i,v) rep(i,0,(int)v.size()-1)
#define lint long long
#define mod 998244353
#define db double
#define pb push_back
#define mp make_pair
#define fir first
#define sec second
#define N 10000010
#define K 100020
#define debug(x) cerr<<#x<<"="<<x
#define sp <<" "
#define ln <<endl
using namespace std;
typedef pair<int,int> pii;
typedef set<int>::iterator sit;
typedef unordered_map<int,int> mpii;
typedef unordered_map<int,lint> mpil;
mpil savg;lint sg[N];int fac[K],facinv[K];
mpii savh;int sh[N],k,mu[N],ik[N],sk[N];
bool np[N];int p[N],pre[K],suf[K],y[K];
inline int inn()
{
	int x,ch;while((ch=gc)<'0'||ch>'9');
	x=ch^'0';while((ch=gc)>='0'&&ch<='9')
		x=(x<<1)+(x<<3)+(ch^'0');return x;
}
inline int fast_pow(int x,int k,int ans=1) { for(;k;k>>=1,x=(lint)x*x%mod) (k&1)?ans=(lint)ans*x%mod:0;return ans; }
inline int prelude(int n)
{
	mu[1]=1,sg[1]=mu[1]*1,sh[1]=mu[1],ik[1]=1,sk[1]=1;
	for(int i=2,c=0;i<=n;i++)
	{
		if(!np[i]) p[++c]=i,mu[i]=-1,ik[i]=fast_pow(i,k);
		sg[i]=sg[i-1]+mu[i]*i,sh[i]=sh[i-1]+mu[i],
		sk[i]=sk[i-1]+ik[i],(sk[i]>=mod?sk[i]-=mod:0);
		for(int j=1,u=n/i;j<=c&&p[j]<=u;j++)
		{
			int x=p[j]*i;np[x]=1,ik[x]=(lint)ik[i]*ik[p[j]]%mod;
			if(i%p[j]==0) { mu[x]=0;break; } else mu[x]=-mu[i];
		}
	}
	return 0;
}
inline int prelude2(int n)
{
	rep(i,1,n) y[i]=y[i-1]+fast_pow(i,k),(y[i]>=mod?y[i]-=mod:0);
	rep(i,fac[0]=1,n) fac[i]=(lint)fac[i-1]*i%mod;
	facinv[n]=fast_pow(fac[n],mod-2);
	for(int i=n-1;i>=0;i--) facinv[i]=(i+1ll)*facinv[i+1]%mod;
	return 0;
}
inline lint g(int n)//mu(i)*i
{
	if(n<N) return sg[n];lint ans=0;
	if(savg.count(n)) return savg[n];
	for(int s=2,t;s<=n;s=t+1) t=n/(n/s),ans+=(s+t)*(t-s+1ll)/2*g(n/s);
	return savg[n]=1-ans;
}
inline int h(int n)//mu(i)
{
	if(n<N) return sh[n];int ans=0;
	if(savh.count(n)) return savh[n];
	for(int s=2,t;s<=n;s=t+1) t=n/(n/s),ans+=h(n/s)*(t-s+1);
	return savh[n]=1-ans;
}
inline int g(int l,int r) { return ((g(r)-g(l-1))%mod+mod)%mod; }
inline int h(int l,int r) { return (h(r)-h(l-1)+mod)%mod; }
inline int S(int x)
{
	if(x<N) return sk[x];int n=k+3;static lint xs[3],ans;
	pre[0]=suf[n+1]=1,xs[0]=1,xs[1]=-1,ans=0;
	for(int i=1;i<=n;i++) pre[i]=(lint)pre[i-1]*(x-i)%mod;
	for(int i=n;i>=1;i--) suf[i]=(lint)suf[i+1]*(x-i)%mod;
	for(int i=1;i<=n;i++)
		ans+=xs[(n-i)&1]*y[i]*pre[i-1]%mod*suf[i+1]%mod*facinv[i-1]%mod*facinv[n-i]%mod;
	return int((ans%mod+mod)%mod);
}
inline int IK(int i) { if(i<N) return ik[i];return fast_pow(i,k); }
inline int qs(int n,int kk=k) { int ans=0;rep(i,1,n) (ans+=fast_pow(i,kk))%=mod;return ans; }
int main()
{
	int n=inn();k=inn();lint ans=0;prelude(N-1),prelude2(K-1);
	for(int l=1,r,t;l<=n;l=r+1) r=n/(n/l),t=n/l,
		ans+=(g(l,r)*(S(t-1)-(lint)IK(t)*t%mod)+h(l,r)*(n+1ll)%mod*IK(t)%mod)%mod;
	return !printf("%lld\n",(ans%mod+mod)%mod);
}