luogu4240 毒瘤之神的考验(毒瘤乌斯反演)
题意:求出\(\sum_{i=1}^n\sum_{j=1}^m\varphi(ij)\),对998244353取模
多组数据,\(T\le 10^4,n,m\le 10^5\)。
前置知识:\(\varphi(ij)=\frac{\varphi(i)\varphi(j)\gcd(i,j)}{\varphi(\gcd(i,j))}\)
证明:我是口胡呢还是好好证呢还是口胡吧
按照欧拉函数的计算式展开,会发现,左边是\(ij\prod_{p|i \mathrm{\color{green}{or}}p|j}\frac{p-1}p\)
右边是\(\frac{i\prod_{p|i}\frac{p-1}pj\prod_{p|j}\frac{p-1}p\gcd(i,j)}{\gcd(i,j)\prod_{p|i\mathrm{\color{green}{and}}p|j}\frac{p-1}p}\)
显然,根据容斥原理,两边是相等的
然后推式子
\(\sum_{i=1}^n\sum_{j=1}^m\varphi(ij)\)
\(=\sum_{i=1}^n\sum_{j=1}^m\frac{\varphi(i)\varphi(j)\gcd(i,j)}{\varphi(\gcd(ij))}\)
\(=\sum_{p=1}^n\frac p{\varphi(p)}\sum_{i=1}^n\sum_{j=1}^m\varphi(i)\varphi(j)[\gcd(i,j)=p]\)
\(=\sum_{p=1}^n\frac p{\varphi(p)}\sum_{i=1}^{n/p}\sum_{j=1}^{m/p}\varphi(ip)\varphi(jp)[\gcd(i,j)=1]\)
\(=\sum_{p=1}^n\frac p{\varphi(p)}\sum_{i=1}^{n/p}\sum_{j=1}^{m/p}\varphi(ip)\varphi(jp)\sum_{d|i,d|j}\mu(d)\)
\(=\sum_{p=1}^n\frac p{\varphi(p)}\sum_{d=1}^n\mu(d)\sum_{i=1}^{n/dp}\sum_{j=1}^{m/dp}\varphi(idp)\varphi(jdp)\)
\(=\sum_{q=1}^n\sum_{p|q}\frac{p\mu(\frac qp)}{\varphi(p)}\sum_{i=1}^{n/q}\sum_{j=1}^{m/q}\varphi(iq)\varphi(jq)\)
\(=\sum_{q=1}^n\left(\sum_{p|q}\frac{p\mu(\frac qp)}{\varphi(p)}\right)\left(\sum_{i=1}^{n/q}\varphi(iq)\right)\left(\sum_{i=1}^{m/q}\varphi(iq)\right)\)
前面这一部分好处理--\(O(n\log n)\)枚举倍数。后面?按照套路?数论分块?怎么分????
观察了你谷题解后,终于懂了
设\(sum(q)=\sum_{p|q}\frac{p\mu(\frac qp)}{\varphi(p)}\),显然可以在\(O(n\log n)\)的时间复杂度内处理出来。
设\(g(x,y)=\sum_{i=1}^x\varphi(iy)\),显然有递推式\(g(x,y)=g(x-1,y)+\varphi(xy)\)。
由于\(xy<=n\),对于每个\(x\),有\(\frac nx\)的数值,我们可以通过动态申请内存,在\(O(n\log n)\)的时间复杂度和空间复杂度内求出\(g\)数组。
设\(T(n,a,b)=\sum_{q=1}^n\left(\sum_{p|q}\frac{p\mu(\frac qp)}{\varphi(p)}\right)\left(\sum_{i=1}^{a}\varphi(iq)\right)\left(\sum_{i=1}^{b}\varphi(iq)\right)=\sum_{q=1}^nsum(q)g(a,q)g(b,q)\)
显然T的递推式为\(T(n,a,b)=T(n-1,a,b)+sum(n)g(a,n)g(b,n)\)
根据数论分块那套理论,对于一个\(n/q\)和\(m/q\)相同的\(q\)的区间,当\(n/q=a,m/q=b\)时,这一区间的\(ans=T(r,a,b)-T(l-1,a,b)\),r和l是这一区间内的最大值和最小值
我们考虑预处理\(n*B*B\)范围的答案,B是我们钦定的一个数字,T数组开的空间复杂度为\(O(nB^2)\)(实际上由于\(a*n,b*n\le 10^5\)的限制,应该开不到\(O(nB^2)\)。
对于每次询问,我们只能在\(n/q\le B\)时候进行数论分块操作通过\(T\)数组计算答案,复杂度根据数论分块那套理论为\(O(\sqrt n)\)。
对于\(n/q>B\)的部分,有\(q<n/B\),暴力枚举\(q\),通过\(g\)数组计算答案,这一部分单次计算的复杂度为\(O(n/B)\)。
总复杂度为\(O(n\log n+nB^2+T(\sqrt n+n/B))\)。实测,B开到50左右跑的快一点,且内存占用超小。
下面是乱七八糟的代码= =
注意讲文明,new来的内存要主动回收垃圾
注意取模(这题如果写的复杂度没错的话不卡常,开#define int long long也是没问题的
#include <cstdio>
#include <functional>
using namespace std;
const int p = 998244353;
const int b = 50;
bool vis[100010];
int prime[100010], tot, fuck = 100000;
int mu[100010], phi[100010], invphi[100010];
int sum[100010];
int *g[100010], *t[100][100]; //注意这里t数组下标是[2][3][1]
int qpow(int x, int y)
{
int res = 1;
for (x %= p; y > 0; x = x * (long long)x % p, y >>= 1) if (y & 1) res = res * (long long)x % p;
return res;
}
int main()
{
//线性筛phi,mu,预处理前面的部分
phi[1] = mu[1] = invphi[1] = 1;
for (int i = 2; i <= fuck; i++)
{
if (vis[i] == false) prime[++tot] = i, phi[i] = i + (mu[i] = -1);
for (int j = 1; j <= tot && i * prime[j] <= fuck; j++)
{
vis[i * prime[j]] = true;
if (i % prime[j] == 0) { phi[i * prime[j]] = phi[i] * prime[j]; break; }
phi[i * prime[j]] = phi[i] * (prime[j] - 1);
mu[i * prime[j]] = -mu[i];
}
invphi[i] = qpow(phi[i], p - 2);
if (phi[i] * (long long)invphi[i] % p != 1) { return -233; printf("cnm\n"); }
}
for (int pp = 1; pp <= fuck; pp++)
for (int q = pp, d = 1; q <= fuck; q += pp, d++)
sum[q] = (sum[q] + pp * (long long)invphi[pp] % p * mu[d]) % p, sum[q] += (sum[q] < 0 ? p : 0);
//处理g数组
for (int i = 1; i <= fuck; i++)
{
g[i] = new int[(fuck / i) + 1], g[i][0] = 0;
for (int j = 1, sb = fuck / i; j <= sb; j++)
g[i][j] = (g[i][j - 1] + phi[i * j]) % p;
}
//处理t数组 注意有第一维<=第二维,因为下面我们强制n<=m了
for (int j = 1; j <= b; j++)
for (int k = j; k <= b; k++)
{
int len = fuck / max(j, k);
t[j][k] = new int[len + 1], t[j][k][0] = 0;
for (int i = 1; i <= len; i++)
t[j][k][i] = (t[j][k][i - 1] + sum[i] * (long long)g[i][j] % p * g[i][k] % p) % p;
}
//处理询问
int tat;
scanf("%d", &tat);
while (tat --> 0)
{
int n, m, res = 0;
scanf("%d%d", &n, &m);
if (n > m) swap(n, m);
//对于n/q>b的部分,暴力,通过g数组和sum数组计算计算
for (int i = 1, sb = m / b; i <= sb; i++)
res = (res + sum[i] * (long long)g[i][n / i] % p * g[i][m / i] % p) % p;
//对于n/q<b的部分,数论分块,通过b数组计算
for (int i = m / b + 1, j; i <= n; i = j + 1)
{
j = min(n / (n / i), m / (m / i));
res = (res + t[n / i][m / i][j] - t[n / i][m / i][i - 1]) % p, res += (res < 0 ? p : 0);
}
printf("%d\n", res);
}
//垃圾回收
for (int i = 1; i <= fuck; i++)
delete []g[i], g[i] = 0;
for (int i = 1; i <= b; i++)
for (int j = i; j <= b; j++)
delete[] t[i][j], t[i][j] = 0;
return 0;
}