导数卷积
题目描述
题解
参考了一下标程的推导过程,因为这个推导对我这种数学弱渣真的有点难鸭.
[1] f ( x ) f(x) f(x)的 i i i次导函数:
f ( i ) ( x ) = a i ∗ i ! 0 ! + a i + 1 ∗ ( i + 1 ) ! 1 ! ∗ x 1 + . . . + a n − 1 ∗ ( n − 1 ) ! ( n − 1 − i ) ! ∗ x n − 1 − i f^{(i)}(x) = a_{i}*\frac{i!}{0!} + a_{i+1}*\frac{(i+1)!}{1!}*x^{1} +...+a_{n-1}*\frac{(n-1)!}{(n-1-i)!}*x^{n-1-i} f(i)(x)=ai∗0!i!+ai+1∗1!(i+1)!∗x1+...+an−1∗(n−1−i)!(n−1)!∗xn−1−i
[2] f ( x ) f(x) f(x)的 n − i − 1 n-i-1 n−i−1次导函数:
f ( n − i − 1 ) ( x ) = a n − i − 1 ∗ ( n − 1 − i ) ! 0 ! + a n − i ∗ ( n − i ) ! 1 ! ∗ x 1 + . . . + a n − 1 ∗ ( n − 1 ) ! i ! ∗ x i f^{(n-i-1)}(x) = a_{n-i-1}*\frac{(n-1-i)!}{0!} + a_{n-i}*\frac{(n-i)!}{1!}*x^{1} +...+a_{n-1}*\frac{(n-1)!}{i!}*x^{i} f(n−i−1)(x)=an−i−1∗0!(n−1−i)!+an−i∗1!(n−i)!∗x1+...+an−1∗i!(n−1)!∗xi
对于积式 f ( i ) ( x ) f ( n − 1 − i ) ( x ) f^{(i)}(x)f^{(n-1-i)}(x) f(i)(x)f(n−1−i)(x)结果中,次数为 d d d的项 x d x^d xd前的系数应该是:
∑ t = 0 d a i + t ∗ ( i + t ) ! t ! ∗ a n − i − 1 + ( d − t ) ∗ ( n − i − 1 + ( d − t ) ) ! ( d − t ) ! \sum_{t = 0}^{d}a_{i+t}*\frac{(i+t)!}{t!}*a_{n-i-1+(d-t)}*\frac{(n-i-1+(d-t))!}{(d-t)!} ∑t=0dai+t∗t!(i+t)!∗an−i−1+(d−t)∗(d−t)!(n−i−1+(d−t))!
令 F i = a i ∗ i ! F_i = a_i * i! Fi=ai∗i!
换种写法:
∑ t = 0 d F i + t ∗ 1 t ! ∗ F n − 1 + d − ( i + t ) ∗ 1 ( d − t ) ! \sum_{t = 0}^{d}F_{i+t}*\frac{1}{t!}*F_{n-1+d-(i+t)}*\frac{1}{(d-t)!} ∑t=0dFi+t∗t!1∗Fn−1+d−(i+t)∗(d−t)!1
对所有的积式求和得到:
∑ i = 0 n − 1 ∑ t = 0 d F i + t ∗ 1 t ! ∗ F n − 1 + d − ( i + t ) ∗ 1 ( d − t ) ! \sum_{i= 0}^{n-1}\sum_{t = 0}^{d}F_{i+t}*\frac{1}{t!}*F_{n-1+d-(i+t)}*\frac{1}{(d-t)!} ∑i=0n−1∑t=0dFi+t∗t!1∗Fn−1+d−(i+t)∗(d−t)!1
= ∑ i = 0 n − 1 + d F i ∗ F n − 1 + d − i ∗ ∑ j = 0 d 1 j ! ( d − j ) ! =\sum^{n-1+d}_{i=0}F_{i}*F_{n-1+d-i}*\sum^{d}_{j=0}\frac{1}{j!(d-j)!} =∑i=0n−1+dFi∗Fn−1+d−i∗∑j=0dj!(d−j)!1
在上个式子中必须要满足 n − 1 + d − i ≤ n − 1 n-1+d-i \le n-1 n−1+d−i≤n−1, 即, d ≤ i d \le i d≤i ,不然 F n − 1 + d − i F_{n-1+d-i} Fn−1+d−i将变成 0 0 0,不过对答案没有影响.
j j j的取值范围是 j : [ 0 , m i n ( d , i ) ] j:[0,min(d,i)] j:[0,min(d,i)],也就是 j : [ 0 , d ] j:[0,d] j:[0,d].
因此上个式子就是,
= ∑ i = 0 n − 1 + d F i ∗ F n − 1 + d − i ∗ 2 d d ! = 2 d d ! ∗ ∑ i = 0 n − 1 + d F i ∗ F n − 1 + d − i =\sum^{n-1+d}_{i=0}F_{i}*F_{n-1+d-i}*\frac{2^d}{d!} = \frac{2^d}{d!} *\sum^{n-1+d}_{i=0}F_{i}*F_{n-1+d-i} =∑i=0n−1+dFi∗Fn−1+d−i∗d!2d=d!2d∗∑i=0n−1+dFi∗Fn−1+d−i
对后面的部分 ∑ i = 0 n − 1 + d F i ∗ F n − 1 + d − i \sum^{n-1+d}_{i=0}F_{i}*F_{n-1+d-i} ∑i=0n−1+dFi∗Fn−1+d−i直接套用 N T T NTT NTT来求就可以了.
代码
#include <iostream>
#include <cstring>
#include <cstdio>
using namespace std;
#define rep(i,a,b) for(int i = a;i <= b;++i)
typedef long long LL;
const int N = 1 << 20;
const int P = 998244353;
const int G = 3;
const int NUM = 20;
LL wn[NUM];
LL a[N], b[N];
LL quick_mod(LL a, LL b, LL m)
{
LL ans = 1;
a %= m;
while(b)
{
if(b & 1)
{
ans = ans * a % m;
b--;
}
b >>= 1;
a = a * a % m;
}
return ans;
}
void GetWn()
{
for(int i = 0; i < NUM; i++)
{
int t = 1 << i;
wn[i] = quick_mod(G, (P - 1) / t, P);
}
}
void Rader(LL a[], int len)
{
int j = len >> 1;
for(int i = 1; i < len - 1; i++)
{
if(i < j) swap(a[i], a[j]);
int k = len >> 1;
while(j >= k)
{
j -= k;
k >>= 1;
}
if(j < k) j += k;
}
}
void NTT(LL a[], int len, int on)
{
Rader(a, len);
int id = 0;
for(int h = 2; h <= len; h <<= 1)
{
id++;
for(int j = 0; j < len; j += h)
{
LL w = 1;
for(int k = j; k < j + h / 2; k++)
{
LL u = a[k] % P;
LL t = w * a[k + h / 2] % P;
a[k] = (u + t) % P;
a[k + h / 2] = (u - t + P) % P;
w = w * wn[id] % P;
}
}
}
if(on == -1)
{
for(int i = 1; i < len / 2; i++)
swap(a[i], a[len - i]);
LL inv = quick_mod(len, P - 2, P);
for(int i = 0; i < len; i++)
a[i] = a[i] * inv % P;
}
}
void Conv(LL a[], LL b[], int n)
{
NTT(a, n, 1);
NTT(b, n, 1);
for(int i = 0; i < n; i++)
a[i] = a[i] * b[i] % P;
NTT(a, n, -1);
}
int n;
LL Fac[N],iFac[N];
int main()
{
GetWn();
Fac[0] = 1;
rep(i,1,N-1)
Fac[i] = Fac[i-1] * i % P;
rep(i,0,N-1)
iFac[i] = quick_mod(Fac[i],P-2,P);
ios::sync_with_stdio(false);
cin >> n;
rep(i,0,n-1) {
std::cin >> a[i];
a[i] = b[i] = a[i] * Fac[i] % P;
}
int len = 1;
while(len < n) len <<= 1;
len <<= 1;
Conv(a,b,len);
rep(i,0,n-1) {
std::cout << a[n-1+i] * quick_mod(2,i,P) % P * iFac[i] % P << " ";
}
return 0;
}