cplib-cpp
This documentation is automatically generated by online-judge-tools/verification-helper
formal_power_series/test/kitamasa.test.cpp
- View this file on GitHub
- Last update: 2025-08-25 00:44:48+09:00
- Problem: https://yukicoder.me/problems/no/214
Depends on
Code
#define PROBLEM "https://yukicoder.me/problems/no/214"
#include "../../convolution/ntt.hpp"
#include "../linear_recurrence.hpp"
#include "../../modint.hpp"
using mint = ModInt<1000000007>;
#include <iostream>
#include <numeric>
std::vector<mint> gen_dp(std::vector<int> v, int n) {
std::vector<std::vector<mint>> dp(n + 1, std::vector<mint>(v.back() * n + 1));
dp[0][0] = 1;
for (auto x : v) {
for (int i = n - 1; i >= 0; i--) {
for (int j = 0; j < dp[i].size(); j++)
if (dp[i][j]) {
for (int k = 1; i + k <= n; k++) dp[i + k][j + x * k] += dp[i][j];
}
}
}
return dp.back();
}
int main() {
long long N;
int P, C;
std::cin >> N >> P >> C;
std::vector<mint> primes = gen_dp({2, 3, 5, 7, 11, 13}, P),
composites = gen_dp({4, 6, 8, 9, 10, 12}, C);
std::vector<mint> f_reversed = nttconv(primes, composites);
std::vector<mint> dp(f_reversed.size());
dp[0] = 1;
for (int i = 0; i < dp.size(); i++) {
for (int j = 1; i + j < dp.size(); j++) dp[i + j] += dp[i] * f_reversed[j];
}
for (auto &x : f_reversed) x = -x;
f_reversed[0] = 1;
std::vector<mint> g(f_reversed.size() - 1);
g[0] = 1;
if (N > f_reversed.size()) {
long long d = N - f_reversed.size();
N -= d;
g = monomial_mod_polynomial<mint>(d, f_reversed);
}
auto prod_x = [&](std::vector<mint> v) -> std::vector<mint> {
int K = v.size();
std::vector<mint> c(K);
c[0] = -v[K - 1] * f_reversed[K];
for (int i = 1; i < K; i++) { c[i] = v[i - 1] - v[K - 1] * f_reversed[K - i]; }
return c;
};
mint acc = 0;
for (int i = N; i < f_reversed.size(); i++) acc += f_reversed[i];
mint ret = 0;
while (N) {
mint p = std::inner_product(g.begin(), g.end(), dp.begin(), mint(0));
ret -= acc * p;
g = prod_x(g);
N--;
acc += f_reversed[N];
}
std::cout << ret << '\n';
}#line 1 "formal_power_series/test/kitamasa.test.cpp"
#define PROBLEM "https://yukicoder.me/problems/no/214"
#line 2 "modint.hpp"
#include <cassert>
#include <iostream>
#include <set>
#include <vector>
template <int md> struct ModInt {
static_assert(md > 1);
using lint = long long;
constexpr static int mod() { return md; }
static int get_primitive_root() {
static int primitive_root = 0;
if (!primitive_root) {
primitive_root = [&]() {
std::set<int> fac;
int v = md - 1;
for (lint i = 2; i * i <= v; i++)
while (v % i == 0) fac.insert(i), v /= i;
if (v > 1) fac.insert(v);
for (int g = 1; g < md; g++) {
bool ok = true;
for (auto i : fac)
if (ModInt(g).pow((md - 1) / i) == 1) {
ok = false;
break;
}
if (ok) return g;
}
return -1;
}();
}
return primitive_root;
}
int val_;
int val() const noexcept { return val_; }
constexpr ModInt() : val_(0) {}
constexpr ModInt &_setval(lint v) { return val_ = (v >= md ? v - md : v), *this; }
constexpr ModInt(lint v) { _setval(v % md + md); }
constexpr explicit operator bool() const { return val_ != 0; }
constexpr ModInt operator+(const ModInt &x) const {
return ModInt()._setval((lint)val_ + x.val_);
}
constexpr ModInt operator-(const ModInt &x) const {
return ModInt()._setval((lint)val_ - x.val_ + md);
}
constexpr ModInt operator*(const ModInt &x) const {
return ModInt()._setval((lint)val_ * x.val_ % md);
}
constexpr ModInt operator/(const ModInt &x) const {
return ModInt()._setval((lint)val_ * x.inv().val() % md);
}
constexpr ModInt operator-() const { return ModInt()._setval(md - val_); }
constexpr ModInt &operator+=(const ModInt &x) { return *this = *this + x; }
constexpr ModInt &operator-=(const ModInt &x) { return *this = *this - x; }
constexpr ModInt &operator*=(const ModInt &x) { return *this = *this * x; }
constexpr ModInt &operator/=(const ModInt &x) { return *this = *this / x; }
friend constexpr ModInt operator+(lint a, const ModInt &x) { return ModInt(a) + x; }
friend constexpr ModInt operator-(lint a, const ModInt &x) { return ModInt(a) - x; }
friend constexpr ModInt operator*(lint a, const ModInt &x) { return ModInt(a) * x; }
friend constexpr ModInt operator/(lint a, const ModInt &x) { return ModInt(a) / x; }
constexpr bool operator==(const ModInt &x) const { return val_ == x.val_; }
constexpr bool operator!=(const ModInt &x) const { return val_ != x.val_; }
constexpr bool operator<(const ModInt &x) const {
return val_ < x.val_;
} // To use std::map<ModInt, T>
friend std::istream &operator>>(std::istream &is, ModInt &x) {
lint t;
return is >> t, x = ModInt(t), is;
}
constexpr friend std::ostream &operator<<(std::ostream &os, const ModInt &x) {
return os << x.val_;
}
constexpr ModInt pow(lint n) const {
ModInt ans = 1, tmp = *this;
while (n) {
if (n & 1) ans *= tmp;
tmp *= tmp, n >>= 1;
}
return ans;
}
static constexpr int cache_limit = std::min(md, 1 << 21);
static std::vector<ModInt> facs, facinvs, invs;
constexpr static void _precalculation(int N) {
const int l0 = facs.size();
if (N > md) N = md;
if (N <= l0) return;
facs.resize(N), facinvs.resize(N), invs.resize(N);
for (int i = l0; i < N; i++) facs[i] = facs[i - 1] * i;
facinvs[N - 1] = facs.back().pow(md - 2);
for (int i = N - 2; i >= l0; i--) facinvs[i] = facinvs[i + 1] * (i + 1);
for (int i = N - 1; i >= l0; i--) invs[i] = facinvs[i] * facs[i - 1];
}
constexpr ModInt inv() const {
if (this->val_ < cache_limit) {
if (facs.empty()) facs = {1}, facinvs = {1}, invs = {0};
while (this->val_ >= int(facs.size())) _precalculation(facs.size() * 2);
return invs[this->val_];
} else {
return this->pow(md - 2);
}
}
constexpr static ModInt fac(int n) {
assert(n >= 0);
if (n >= md) return ModInt(0);
while (n >= int(facs.size())) _precalculation(facs.size() * 2);
return facs[n];
}
constexpr static ModInt facinv(int n) {
assert(n >= 0);
if (n >= md) return ModInt(0);
while (n >= int(facs.size())) _precalculation(facs.size() * 2);
return facinvs[n];
}
constexpr static ModInt doublefac(int n) {
assert(n >= 0);
if (n >= md) return ModInt(0);
long long k = (n + 1) / 2;
return (n & 1) ? ModInt::fac(k * 2) / (ModInt(2).pow(k) * ModInt::fac(k))
: ModInt::fac(k) * ModInt(2).pow(k);
}
constexpr static ModInt nCr(int n, int r) {
assert(n >= 0);
if (r < 0 or n < r) return ModInt(0);
return ModInt::fac(n) * ModInt::facinv(r) * ModInt::facinv(n - r);
}
constexpr static ModInt nPr(int n, int r) {
assert(n >= 0);
if (r < 0 or n < r) return ModInt(0);
return ModInt::fac(n) * ModInt::facinv(n - r);
}
static ModInt binom(int n, int r) {
static long long bruteforce_times = 0;
if (r < 0 or n < r) return ModInt(0);
if (n <= bruteforce_times or n < (int)facs.size()) return ModInt::nCr(n, r);
r = std::min(r, n - r);
ModInt ret = ModInt::facinv(r);
for (int i = 0; i < r; ++i) ret *= n - i;
bruteforce_times += r;
return ret;
}
// Multinomial coefficient, (k_1 + k_2 + ... + k_m)! / (k_1! k_2! ... k_m!)
// Complexity: O(sum(ks))
template <class Vec> static ModInt multinomial(const Vec &ks) {
ModInt ret{1};
int sum = 0;
for (int k : ks) {
assert(k >= 0);
ret *= ModInt::facinv(k), sum += k;
}
return ret * ModInt::fac(sum);
}
template <class... Args> static ModInt multinomial(Args... args) {
int sum = (0 + ... + args);
ModInt result = (1 * ... * ModInt::facinv(args));
return ModInt::fac(sum) * result;
}
// Catalan number, C_n = binom(2n, n) / (n + 1) = # of Dyck words of length 2n
// C_0 = 1, C_1 = 1, C_2 = 2, C_3 = 5, C_4 = 14, ...
// https://oeis.org/A000108
// Complexity: O(n)
static ModInt catalan(int n) {
if (n < 0) return ModInt(0);
return ModInt::fac(n * 2) * ModInt::facinv(n + 1) * ModInt::facinv(n);
}
ModInt sqrt() const {
if (val_ == 0) return 0;
if (md == 2) return val_;
if (pow((md - 1) / 2) != 1) return 0;
ModInt b = 1;
while (b.pow((md - 1) / 2) == 1) b += 1;
int e = 0, m = md - 1;
while (m % 2 == 0) m >>= 1, e++;
ModInt x = pow((m - 1) / 2), y = (*this) * x * x;
x *= (*this);
ModInt z = b.pow(m);
while (y != 1) {
int j = 0;
ModInt t = y;
while (t != 1) j++, t *= t;
z = z.pow(1LL << (e - j - 1));
x *= z, z *= z, y *= z;
e = j;
}
return ModInt(std::min(x.val_, md - x.val_));
}
};
template <int md> std::vector<ModInt<md>> ModInt<md>::facs = {1};
template <int md> std::vector<ModInt<md>> ModInt<md>::facinvs = {1};
template <int md> std::vector<ModInt<md>> ModInt<md>::invs = {0};
using ModInt998244353 = ModInt<998244353>;
// using mint = ModInt<998244353>;
// using mint = ModInt<1000000007>;
#line 3 "convolution/ntt.hpp"
#include <algorithm>
#include <array>
#line 7 "convolution/ntt.hpp"
#include <tuple>
#line 9 "convolution/ntt.hpp"
// CUT begin
// Integer convolution for arbitrary mod
// with NTT (and Garner's algorithm) for ModInt / ModIntRuntime class.
// We skip Garner's algorithm if `skip_garner` is true or mod is in `nttprimes`.
// input: a (size: n), b (size: m)
// return: vector (size: n + m - 1)
template <typename MODINT>
std::vector<MODINT> nttconv(std::vector<MODINT> a, std::vector<MODINT> b, bool skip_garner);
constexpr int nttprimes[3] = {998244353, 167772161, 469762049};
// Integer FFT (Fast Fourier Transform) for ModInt class
// (Also known as Number Theoretic Transform, NTT)
// is_inverse: inverse transform
// ** Input size must be 2^n **
template <typename MODINT> void ntt(std::vector<MODINT> &a, bool is_inverse = false) {
int n = a.size();
if (n == 1) return;
static const int mod = MODINT::mod();
static const MODINT root = MODINT::get_primitive_root();
assert(__builtin_popcount(n) == 1 and (mod - 1) % n == 0);
static std::vector<MODINT> w{1}, iw{1};
for (int m = w.size(); m < n / 2; m *= 2) {
MODINT dw = root.pow((mod - 1) / (4 * m)), dwinv = 1 / dw;
w.resize(m * 2), iw.resize(m * 2);
for (int i = 0; i < m; i++) w[m + i] = w[i] * dw, iw[m + i] = iw[i] * dwinv;
}
if (!is_inverse) {
for (int m = n; m >>= 1;) {
for (int s = 0, k = 0; s < n; s += 2 * m, k++) {
for (int i = s; i < s + m; i++) {
MODINT x = a[i], y = a[i + m] * w[k];
a[i] = x + y, a[i + m] = x - y;
}
}
}
} else {
for (int m = 1; m < n; m *= 2) {
for (int s = 0, k = 0; s < n; s += 2 * m, k++) {
for (int i = s; i < s + m; i++) {
MODINT x = a[i], y = a[i + m];
a[i] = x + y, a[i + m] = (x - y) * iw[k];
}
}
}
int n_inv = MODINT(n).inv().val();
for (auto &v : a) v *= n_inv;
}
}
template <int MOD>
std::vector<ModInt<MOD>> nttconv_(const std::vector<int> &a, const std::vector<int> &b) {
int sz = a.size();
assert(a.size() == b.size() and __builtin_popcount(sz) == 1);
std::vector<ModInt<MOD>> ap(sz), bp(sz);
for (int i = 0; i < sz; i++) ap[i] = a[i], bp[i] = b[i];
ntt(ap, false);
if (a == b)
bp = ap;
else
ntt(bp, false);
for (int i = 0; i < sz; i++) ap[i] *= bp[i];
ntt(ap, true);
return ap;
}
long long garner_ntt_(int r0, int r1, int r2, int mod) {
using mint2 = ModInt<nttprimes[2]>;
static const long long m01 = 1LL * nttprimes[0] * nttprimes[1];
static const long long m0_inv_m1 = ModInt<nttprimes[1]>(nttprimes[0]).inv().val();
static const long long m01_inv_m2 = mint2(m01).inv().val();
int v1 = (m0_inv_m1 * (r1 + nttprimes[1] - r0)) % nttprimes[1];
auto v2 = (mint2(r2) - r0 - mint2(nttprimes[0]) * v1) * m01_inv_m2;
return (r0 + 1LL * nttprimes[0] * v1 + m01 % mod * v2.val()) % mod;
}
template <typename MODINT>
std::vector<MODINT> nttconv(std::vector<MODINT> a, std::vector<MODINT> b, bool skip_garner) {
if (a.empty() or b.empty()) return {};
int sz = 1, n = a.size(), m = b.size();
while (sz < n + m) sz <<= 1;
if (sz <= 16) {
std::vector<MODINT> ret(n + m - 1);
for (int i = 0; i < n; i++) {
for (int j = 0; j < m; j++) ret[i + j] += a[i] * b[j];
}
return ret;
}
int mod = MODINT::mod();
if (skip_garner or
std::find(std::begin(nttprimes), std::end(nttprimes), mod) != std::end(nttprimes)) {
a.resize(sz), b.resize(sz);
if (a == b) {
ntt(a, false);
b = a;
} else {
ntt(a, false), ntt(b, false);
}
for (int i = 0; i < sz; i++) a[i] *= b[i];
ntt(a, true);
a.resize(n + m - 1);
} else {
std::vector<int> ai(sz), bi(sz);
for (int i = 0; i < n; i++) ai[i] = a[i].val();
for (int i = 0; i < m; i++) bi[i] = b[i].val();
auto ntt0 = nttconv_<nttprimes[0]>(ai, bi);
auto ntt1 = nttconv_<nttprimes[1]>(ai, bi);
auto ntt2 = nttconv_<nttprimes[2]>(ai, bi);
a.resize(n + m - 1);
for (int i = 0; i < n + m - 1; i++)
a[i] = garner_ntt_(ntt0[i].val(), ntt1[i].val(), ntt2[i].val(), mod);
}
return a;
}
template <typename MODINT>
std::vector<MODINT> nttconv(const std::vector<MODINT> &a, const std::vector<MODINT> &b) {
return nttconv<MODINT>(a, b, false);
}
#line 4 "formal_power_series/linear_recurrence.hpp"
#include <utility>
#line 6 "formal_power_series/linear_recurrence.hpp"
// CUT begin
// Berlekamp–Massey algorithm
// https://en.wikipedia.org/wiki/Berlekamp%E2%80%93Massey_algorithm
// Complexity: O(N^2)
// input: S = sequence from field K
// return: L = degree of minimal polynomial,
// C_reversed = monic min. polynomial (size = L + 1, reversed order, C_reversed[0] = 1))
// Formula: convolve(S, C_reversed)[i] = 0 for i >= L
// Example:
// - [1, 2, 4, 8, 16] -> (1, [1, -2])
// - [1, 1, 2, 3, 5, 8] -> (2, [1, -1, -1])
// - [0, 0, 0, 0, 1] -> (5, [1, 0, 0, 0, 0, 998244352]) (mod 998244353)
// - [] -> (0, [1])
// - [0, 0, 0] -> (0, [1])
// - [-2] -> (1, [1, 2])
template <typename Tfield>
std::pair<int, std::vector<Tfield>> find_linear_recurrence(const std::vector<Tfield> &S) {
int N = S.size();
using poly = std::vector<Tfield>;
poly C_reversed{1}, B{1};
int L = 0, m = 1;
Tfield b = 1;
// adjust: C(x) <- C(x) - (d / b) x^m B(x)
auto adjust = [](poly C, const poly &B, Tfield d, Tfield b, int m) -> poly {
C.resize(std::max(C.size(), B.size() + m));
Tfield a = d / b;
for (unsigned i = 0; i < B.size(); i++) C[i + m] -= a * B[i];
return C;
};
for (int n = 0; n < N; n++) {
Tfield d = S[n];
for (int i = 1; i <= L; i++) d += C_reversed[i] * S[n - i];
if (d == 0)
m++;
else if (2 * L <= n) {
poly T = C_reversed;
C_reversed = adjust(C_reversed, B, d, b, m);
L = n + 1 - L;
B = T;
b = d;
m = 1;
} else
C_reversed = adjust(C_reversed, B, d, b, m++);
}
return std::make_pair(L, C_reversed);
}
// Calculate $x^N \bmod f(x)$
// Known as `Kitamasa method`
// Input: f_reversed: monic, reversed (f_reversed[0] = 1)
// Complexity: $O(K^2 \log N)$ ($K$: deg. of $f$)
// Example: (4, [1, -1, -1]) -> [2, 3]
// ( x^4 = (x^2 + x + 2)(x^2 - x - 1) + 3x + 2 )
// Reference: http://misawa.github.io/others/fast_kitamasa_method.html
// http://sugarknri.hatenablog.com/entry/2017/11/18/233936
template <typename Tfield>
std::vector<Tfield> monomial_mod_polynomial(long long N, const std::vector<Tfield> &f_reversed) {
assert(!f_reversed.empty() and f_reversed[0] == 1);
int K = f_reversed.size() - 1;
if (!K) return {};
int D = 64 - __builtin_clzll(N);
std::vector<Tfield> ret(K, 0);
ret[0] = 1;
auto self_conv = [](std::vector<Tfield> x) -> std::vector<Tfield> {
int d = x.size();
std::vector<Tfield> ret(d * 2 - 1);
for (int i = 0; i < d; i++) {
ret[i * 2] += x[i] * x[i];
for (int j = 0; j < i; j++) ret[i + j] += x[i] * x[j] * 2;
}
return ret;
};
for (int d = D; d--;) {
ret = self_conv(ret);
for (int i = 2 * K - 2; i >= K; i--) {
for (int j = 1; j <= K; j++) ret[i - j] -= ret[i] * f_reversed[j];
}
ret.resize(K);
if ((N >> d) & 1) {
std::vector<Tfield> c(K);
c[0] = -ret[K - 1] * f_reversed[K];
for (int i = 1; i < K; i++) { c[i] = ret[i - 1] - ret[K - 1] * f_reversed[K - i]; }
ret = c;
}
}
return ret;
}
// Guess k-th element of the sequence, assuming linear recurrence
// initial_elements: 0-ORIGIN
// Verify: abc198f https://atcoder.jp/contests/abc198/submissions/21837815
template <typename Tfield>
Tfield guess_kth_term(const std::vector<Tfield> &initial_elements, long long k) {
assert(k >= 0);
if (k < static_cast<long long>(initial_elements.size())) return initial_elements[k];
const auto f = find_linear_recurrence<Tfield>(initial_elements).second;
const auto g = monomial_mod_polynomial<Tfield>(k, f);
Tfield ret = 0;
for (unsigned i = 0; i < g.size(); i++) ret += g[i] * initial_elements[i];
return ret;
}
#line 5 "formal_power_series/test/kitamasa.test.cpp"
using mint = ModInt<1000000007>;
#line 8 "formal_power_series/test/kitamasa.test.cpp"
#include <numeric>
std::vector<mint> gen_dp(std::vector<int> v, int n) {
std::vector<std::vector<mint>> dp(n + 1, std::vector<mint>(v.back() * n + 1));
dp[0][0] = 1;
for (auto x : v) {
for (int i = n - 1; i >= 0; i--) {
for (int j = 0; j < dp[i].size(); j++)
if (dp[i][j]) {
for (int k = 1; i + k <= n; k++) dp[i + k][j + x * k] += dp[i][j];
}
}
}
return dp.back();
}
int main() {
long long N;
int P, C;
std::cin >> N >> P >> C;
std::vector<mint> primes = gen_dp({2, 3, 5, 7, 11, 13}, P),
composites = gen_dp({4, 6, 8, 9, 10, 12}, C);
std::vector<mint> f_reversed = nttconv(primes, composites);
std::vector<mint> dp(f_reversed.size());
dp[0] = 1;
for (int i = 0; i < dp.size(); i++) {
for (int j = 1; i + j < dp.size(); j++) dp[i + j] += dp[i] * f_reversed[j];
}
for (auto &x : f_reversed) x = -x;
f_reversed[0] = 1;
std::vector<mint> g(f_reversed.size() - 1);
g[0] = 1;
if (N > f_reversed.size()) {
long long d = N - f_reversed.size();
N -= d;
g = monomial_mod_polynomial<mint>(d, f_reversed);
}
auto prod_x = [&](std::vector<mint> v) -> std::vector<mint> {
int K = v.size();
std::vector<mint> c(K);
c[0] = -v[K - 1] * f_reversed[K];
for (int i = 1; i < K; i++) { c[i] = v[i - 1] - v[K - 1] * f_reversed[K - i]; }
return c;
};
mint acc = 0;
for (int i = N; i < f_reversed.size(); i++) acc += f_reversed[i];
mint ret = 0;
while (N) {
mint p = std::inner_product(g.begin(), g.end(), dp.begin(), mint(0));
ret -= acc * p;
g = prod_x(g);
N--;
acc += f_reversed[N];
}
std::cout << ret << '\n';
}