【模板】分治 FFT

题目大意:给定长度为 (n - 1) 的序列 (g),求 (f) 序列,其中 (f)

[f[i]=sum_{j=1}^{i} f[i-j] g[j] ]

学会了分治 (fft)
发现这个式子中也含有卷积,但是这是一个递推式,即:(f) 数组是未知的。
考虑分治策略,即:假设已经算出区间 ([l, mid])(f) 值,现在要计算区间 ([mid + 1, r])(f)
考虑左半部分对右半部分的贡献,对于 $$x in [mid + 1, r], contribution(left ightarrow x) = sumlimits_{i = l}^{mid}f(i)*g(x - i)$$
通过下标转换得:

[egin{aligned} f(x) &=sum_{i=l}^{mid} f(i) * g(x-i) \ &=sum_{i=0}^{mid-l} f(i+l) * g(x-l-i) end{aligned} ]

换元得

[foo(i)=f(i+l), bar(i)=g(i) ]

式子变成了

[f(x)=foo(x-l)=sum_{i=0}^{mid-l} A(i) * B(x-l-i) ]

即:卷积序列的第 (x - l) 项为 (x) 的贡献。

注意:对于 (g) 来说,下标不能移动到 (0) 开始,因为 (f) 数组本身下标就是从 (0) 开始的,这就意味着 (g[0] = 0),即:输入从 (g[1]) 开始。
另外,关于c11的 lambda 表达式,对于 [=] 来说,捕获到的变量均为其常量的副本,这个值仅仅由该变量的初始化决定,不会改变。

// update at 2019.10.11
代码如下

#include <bits/stdc++.h>

using namespace std;

typedef long long LL;

const LL mod = 998244353, g = 3, ig = 332748118;

inline LL fpow(LL a, LL b) {
	LL ret = 1 % mod;
	for (; b; b >>= 1, a = a * a % mod) {
		if (b & 1) {
			ret = ret * a % mod;
		}
	}
	return ret;
}

void ntt(vector<LL> &v, vector<int> &rev, int opt) {
	int tot = v.size();
	for (int i = 0; i < tot; i++) if (i < rev[i]) swap(v[i], v[rev[i]]);
	for (int mid = 1; mid < tot; mid <<= 1) {
		LL wn = fpow(opt == 1 ? g : ig, (mod - 1) / (mid << 1));
		for (int j = 0; j < tot; j += mid << 1) {
			LL w = 1;
			for (int k = 0; k < mid; k++) {
				LL x = v[j + k], y = v[j + mid + k] * w % mod;
				v[j + k] = (x + y) % mod, v[j + mid + k] = (x - y + mod) % mod;
				w = w * wn % mod;
			}
		}
	}
	if (opt == -1) {
		LL itot = fpow(tot, mod - 2);
		for (int i = 0; i < tot; i++) v[i] = v[i] * itot % mod;
	}
}
vector<LL> convolution(vector<LL> &a, int sa, int cnta, vector<LL> &b, int sb, int cntb, const function<LL(LL, LL)> &calc) {
	int bit = 0, tot = 1;
	while (tot <= 2 * max(cnta, cntb)) bit++, tot <<= 1;
	vector<int> rev(tot);
	for (int i = 0; i < tot; i++) rev[i] = rev[i >> 1] >> 1 | (i & 1) << (bit - 1);
	vector<LL> foo(tot), bar(tot);
	for (int i = 0, j = sa; i < cnta; i++, j++) foo[i] = a[j];
	for (int i = 0, j = sb; i < cntb; i++, j++) bar[i] = b[j];
	ntt(foo, rev, 1), ntt(bar, rev, 1);
	for (int i = 0; i < tot; i++) foo[i] = calc(foo[i], bar[i]);
	ntt(foo, rev, -1);
	return foo;
}
void cdq(int l, int r, vector<LL> &f, vector<LL> &w) {
	if (l == r) return;
	int mid = l + r >> 1;
	cdq(l, mid, f, w);
	int sz = r - l + 1;
	vector<LL> h = convolution(f, l, mid - l + 1, w, 0, sz, [&](LL a, LL b) {return a * b % mod;});
	for (int i = mid + 1; i <= r; i++) f[i] = (f[i] + h[i - l]) % mod;
	cdq(mid + 1, r, f, w);
}

int main() {
	ios::sync_with_stdio(false);
	cin.tie(0), cout.tie(0);
	int n;
	cin >> n;
	vector<LL> f(n), w(n);
	f[0] = 1;
	for (int i = 1; i < n; i++) {
		cin >> w[i];
	}
	cdq(0, n - 1, f, w);
	for (auto v : f) {
		cout << v << " ";
	}
	return 0;
}
原文地址:https://www.cnblogs.com/wzj-xhjbk/p/11437578.html