HDU 6750 Function-莫比乌斯反演进阶好题(一)

题目大意:

求${\sum }_{i=1}^n{\sum }_{t|i}^{}t[gcd(t,\frac{i}{t})=1]$
$n\le 1e12,T\le 10$

题目思路:

这题在推导公式的时候有比较多的技巧需要使用.如下:

①换求和枚举顺序,消去整除符号:

$={\sum }_{t=1}^n{\sum }_{t|i}^{n}t[gcd(t,\frac{i}{t})=1]$

$={\sum }_{t=1}^{n}t{\sum }_{i=1}^{\frac{n}{t}}[gcd(t,i)=1]$

$={\sum }_{t=1}^{n}t{\sum }_{i=1}^{\frac{n}{t}}{\sum }_{d|gcd(t,i)}^{}\mu (d)$

②继续换求和枚举顺序,消去整除符号:

$={\sum }_{d=1}^n\mu (d){\sum }_{t=1}^{n}t{\sum }_{i=1}^{\frac{n}{t}}[d|gcd(t,i)]$

③$\because d|gcd(t,i)\leftrightarrow d|t\bigwedge d|i,同时这里也要注意上下限的变换$

$={\sum }_{d=1}^n\mu (d){\sum }_{t=1}^{\frac{n}{d}}td{\sum }_{i=1}^{\frac{n}{t{d}^2}}1$

$={\sum }_{d=1}^n\mu (d)d{\sum }_{t=1}^{\frac{n}{d}}t{\sum }_{i=1}^{\frac{n}{t{d}^2}}1$

④令$T=t{d}^2$

$={\sum }_{T=1}^n{\sum }_{d=1}^n\mu (d)d{\sum }_{t=1}^{\frac{n}{d}}t{\sum }_{i=1}^{\frac{n}{t{d}^2}}[T=t{d}^2]$

⑤${\sum }_{i=1}^{\frac{n}{t{d}^2}}$和$[T=t{d}^2]$无关,转化成乘法:

$={\sum }_{T=1}^n\lfloor \frac{n}{T}\rfloor {\sum }_{d=1}^n\mu (d)d{\sum }_{t=1}^{\frac{n}{d}}t[T=t{d}^2]$

⑥观察到$[T=t{d}^2]$,考虑枚举$d^2$的倍数

$={\sum }_{T=1}^n\lfloor \frac{n}{T}\rfloor {\sum }_{d^2|T}^n\mu (d)d{\sum }_{t=1}^{\frac{n}{d}}t[T=t{d}^2]$

⑦在已知$T$和$d^2$的情况下,${\sum }_{t=1}^{\frac{n}{d}}t[T=t{d}^2]$部分可转化为:

$={\sum }_{T=1}^n\lfloor \frac{n}{T}\rfloor {\sum }_{d^2|T}^n\mu (d)d{\sum }_{t=1}^{\frac{n}{d}}t[t=\frac{T}{d^2}]$

⑧观察:${\sum }_{t=1}^{\frac{n}{d}}t[t=\frac{T}{d^2}]$

$t\in [1,\frac{n}{d}],$而$\frac{n}{d}\ge \frac{T}{d}\ge \frac{T}{d^2}$

⑨所以$[t=\frac{T}{d^2}]$在$t\in [1,\frac{n}{d}]$中必然成立且恰好只成立一次。

$={\sum }_{T=1}^n\lfloor \frac{n}{T}\rfloor {\sum }_{d^2|T}^n\mu (d)\frac{T}{d}$

⑩变换枚举顺序，消除整除符号:

$={\sum }_{d=1}^{\sqrt{n}}\mu (d){\sum }_{T=1}^n\lfloor \frac{n}{T}\rfloor [d^2|T]\frac{T}{d}$

(十一)从枚举$T$改成枚举$k=\frac{T}{d^2}$

$={\sum }_{d=1}^{\sqrt{n}}\mu (d){\sum }_{k=1}^{\lfloor \frac{n}{d^2}\rfloor }\lfloor \frac{n}{k{d}^2}\rfloor \frac{k{d}^2}{d}$

$={\sum }_{d=1}^{\sqrt{n}}\mu (d)d{\sum }_{k=1}^{\lfloor \frac{n}{d^2}\rfloor }\lfloor \frac{\lfloor \frac{n}{d^2}\rfloor }{k}\rfloor k$

(十二)令$F(n)={\sum }_{i=1}^{n}i\ast \lfloor \frac{n}{i}\rfloor$

$原式={\sum }_{d=1}^{\sqrt{n}}\mu (d)\ast d\ast F(\lfloor \frac{n}{d^2}\rfloor )$

复杂度分析:

外层直接枚举$\sqrt{n}$次，$F(i)$复杂度为:$\sqrt{i}$

所以复杂度可以写成以下和式:
${\sum }_{i=1}^{\sqrt{n}}\sqrt{\frac{n}{i^2}}={\sum }_{i=1}^{\sqrt{n}}\frac{\sqrt{n}}{i}=\sqrt{n}log\sqrt{n}$

---

细节补充:

形如:${\sum }_{d=1}^{n}g(d)\ast f(\lfloor \frac{n}{d^2}\rfloor )$的式子如何整除分块?

直接套用整除分块的定义:$r^2=\lfloor \frac{n}{\lfloor \frac{n}{l^2}\rfloor }\rfloor$

$r=\sqrt{\lfloor \frac{n}{\lfloor \frac{n}{l^2}\rfloor }\rfloor }$

进一步优化

由于这题的常数很大，可能会$T$。尝试从以下几个方面来考虑优化:

一.利用细节补充我们可以将外层的循环转化成一个数论分块。那么问题就变成了数论分块套数论分块。

二.$F$函数输出只和输入有关，可以记忆化一下.

三.考虑预处理$F$函数：

推导:
$F(n)={\sum }_{i=1}^{n}i\ast \lfloor \frac{n}{i}\rfloor ={\sum }_{i=1}^{n}i\ast {\sum }_{i|j}^{n}1={\sum }_{j=1}^n{\sum }_{i|j}^{}i={\sum }_{j=1}^n\sigma (j)$

所以$F(n)$为约数和的$n$项前缀和.可以上筛法

AC代码:

#include<bits/stdc++.h>
using namespace std;
#define ll long long
#define pii pair<int,int>
#define pb push_back
#define mp make_pair
#define vi vector<int>
#define vll vector<ll>
#define fi first
#define se second
const int maxn = 1e6 + 5;
const int mod = 1e9 + 7;
int u[maxn] , bk[maxn] , p[maxn] , cnt;
ll sum[maxn];
ll g[maxn];
void init ()
{
    u[1] = 1;
    for (int i = 2 ; i < maxn ; i++){
        if (!bk[i]){
            p[++cnt] = i;
            u[i] = -1;
        }
        for (int j = 1 ; j <= cnt && i * p[j] < maxn ; j++){
            bk[i * p[j]] = 1;
            if (i % p[j] == 0) {
                u[i * p[j]] = 0;
                break;
            }else {
                u[i * p[j]] = -u[i];
            }
        }
    }
    for (int i = 1 ; i < maxn ; i++){
        for (int j = i ; j < maxn ; j += i){
            g[j] = (g[j] + i) % mod;
        }
    }
    for (int i = 1 ; i < maxn ; i++){
        sum[i] = (sum[i - 1] + 1ll * (u[i] + mod) % mod * i % mod) % mod;
        g[i] = (g[i - 1] + g[i]) % mod;
    }
    return ;
}
ll ksm (ll a , ll b){ ll ans = 1 , base = a;
while (b){if (b & 1) ans = ans * base % mod;b >>= 1;base = base * base % mod;}return ans;}
ll inv2 = ksm(2 , mod - 2);
unordered_map<ll , ll> dp;
ll F (ll n){
    //if (n < maxn) return g[n];
    if ( dp.count(n) ) return dp[n];
    ll ans = 0;
    ll x , y , z;
    for (ll l = 1 , r; l <= n ; l = r + 1){
        r = n / (n / l);
        x = (r - l + 1) % mod;
        y = (l + r) % mod;
        z = (n / l) % mod;
        ll a = x * y % mod;
        a = a * inv2 % mod;
        a = a * z % mod;
        ans = (ans + a) %mod;
    }
    return dp[n] = ans;
}
int main()
{
    ios::sync_with_stdio(false);
    init();
    int t; cin >> t;
    dp.clear();
    while (t--){
        ll n; cin >> n;
        ll ans = 0;
        ll m = sqrt(n);
        ll x;
        for (ll l = 1 , r; l <= m ; l = r + 1){
            x = n / l / l;
            r = min((ll)sqrt(n / x) , m);
            ll res = (sum[r] - sum[l - 1] + mod) %mod;
            res = res * F(x) % mod;
            ans = (ans + res) % mod;
        }
        cout << ans << endl;
    }
    return 0;
}