集合冪級数の合成
(set-function/polynomial-composite-set-power-series.hpp)

• View this file on GitHub
• Last update: 2023-08-31 20:44:07+09:00
• Include: #include "set-function/polynomial-composite-set-power-series.hpp"

Depends on

• 多項式/形式的冪級数ライブラリ (fps/formal-power-series.hpp)
• 平行移動 (fps/taylor-shift.hpp)
• modulo/binomial.hpp
• Subset Convolution (set-function/subset-convolution.hpp)

Verified with

• verify/verify-yosupo-math/yosupo-polynomial-composite-set-power-series.test.cpp

Code

#pragma once

#include <cassert>
#include <vector>
using namespace std;

#include "../fps/formal-power-series.hpp"
#include "../fps/taylor-shift.hpp"
#include "subset-convolution.hpp"

template <typename mint, int MAX = 21>
vector<mint> polynomial_composite_set_power_series(int n, FormalPowerSeries<mint> f,
                                            vector<mint> g) {
  assert(0 <= n && n <= MAX);
  static SubsetConvolution<mint, MAX> ss;

  Binomial<mint> binom(f.size());
  if (g[0] != 0) {
    f = TaylorShift(f, g[0], binom);
    g[0] = 0;
  }
  f.resize(n + 1), g.resize(1 << n);

  for (int i = 0; i <= n; i++) f[i] *= binom.fac(i);
  vector h(n + 1, vector(n + 1, vector<mint>{}));
  for (int i = 0; i <= n; i++) h[0][i] = {f[i]};
  for (int k = 1; k <= n; k++) {
    auto A = ss.lift({begin(g) + (1 << (k - 1)), begin(g) + (1 << k)});
    ss.zeta(A);
    for (int j = 0; j <= n - k; j++) {
      h[k][j] = h[k - 1][j];
      auto B = ss.lift(h[k - 1][j + 1]);
      ss.zeta(B);
      ss.prod(B, A);
      ss.mobius(B);
      auto c = ss.unlift(B);
      copy(begin(c), end(c), back_inserter(h[k][j]));
    }
  }
  return h[n][0];
}

/**
 * @brief 集合冪級数の合成
 */

#line 2 "set-function/polynomial-composite-set-power-series.hpp"

#include <cassert>
#include <vector>
using namespace std;

#line 2 "fps/formal-power-series.hpp"

template <typename mint>
struct FormalPowerSeries : vector<mint> {
  using vector<mint>::vector;
  using FPS = FormalPowerSeries;

  FPS &operator+=(const FPS &r) {
    if (r.size() > this->size()) this->resize(r.size());
    for (int i = 0; i < (int)r.size(); i++) (*this)[i] += r[i];
    return *this;
  }

  FPS &operator+=(const mint &r) {
    if (this->empty()) this->resize(1);
    (*this)[0] += r;
    return *this;
  }

  FPS &operator-=(const FPS &r) {
    if (r.size() > this->size()) this->resize(r.size());
    for (int i = 0; i < (int)r.size(); i++) (*this)[i] -= r[i];
    return *this;
  }

  FPS &operator-=(const mint &r) {
    if (this->empty()) this->resize(1);
    (*this)[0] -= r;
    return *this;
  }

  FPS &operator*=(const mint &v) {
    for (int k = 0; k < (int)this->size(); k++) (*this)[k] *= v;
    return *this;
  }

  FPS &operator/=(const FPS &r) {
    if (this->size() < r.size()) {
      this->clear();
      return *this;
    }
    int n = this->size() - r.size() + 1;
    if ((int)r.size() <= 64) {
      FPS f(*this), g(r);
      g.shrink();
      mint coeff = g.back().inverse();
      for (auto &x : g) x *= coeff;
      int deg = (int)f.size() - (int)g.size() + 1;
      int gs = g.size();
      FPS quo(deg);
      for (int i = deg - 1; i >= 0; i--) {
        quo[i] = f[i + gs - 1];
        for (int j = 0; j < gs; j++) f[i + j] -= quo[i] * g[j];
      }
      *this = quo * coeff;
      this->resize(n, mint(0));
      return *this;
    }
    return *this = ((*this).rev().pre(n) * r.rev().inv(n)).pre(n).rev();
  }

  FPS &operator%=(const FPS &r) {
    *this -= *this / r * r;
    shrink();
    return *this;
  }

  FPS operator+(const FPS &r) const { return FPS(*this) += r; }
  FPS operator+(const mint &v) const { return FPS(*this) += v; }
  FPS operator-(const FPS &r) const { return FPS(*this) -= r; }
  FPS operator-(const mint &v) const { return FPS(*this) -= v; }
  FPS operator*(const FPS &r) const { return FPS(*this) *= r; }
  FPS operator*(const mint &v) const { return FPS(*this) *= v; }
  FPS operator/(const FPS &r) const { return FPS(*this) /= r; }
  FPS operator%(const FPS &r) const { return FPS(*this) %= r; }
  FPS operator-() const {
    FPS ret(this->size());
    for (int i = 0; i < (int)this->size(); i++) ret[i] = -(*this)[i];
    return ret;
  }

  void shrink() {
    while (this->size() && this->back() == mint(0)) this->pop_back();
  }

  FPS rev() const {
    FPS ret(*this);
    reverse(begin(ret), end(ret));
    return ret;
  }

  FPS dot(FPS r) const {
    FPS ret(min(this->size(), r.size()));
    for (int i = 0; i < (int)ret.size(); i++) ret[i] = (*this)[i] * r[i];
    return ret;
  }

  // 前 sz 項を取ってくる。sz に足りない項は 0 埋めする
  FPS pre(int sz) const {
    FPS ret(begin(*this), begin(*this) + min((int)this->size(), sz));
    if ((int)ret.size() < sz) ret.resize(sz);
    return ret;
  }

  FPS operator>>(int sz) const {
    if ((int)this->size() <= sz) return {};
    FPS ret(*this);
    ret.erase(ret.begin(), ret.begin() + sz);
    return ret;
  }

  FPS operator<<(int sz) const {
    FPS ret(*this);
    ret.insert(ret.begin(), sz, mint(0));
    return ret;
  }

  FPS diff() const {
    const int n = (int)this->size();
    FPS ret(max(0, n - 1));
    mint one(1), coeff(1);
    for (int i = 1; i < n; i++) {
      ret[i - 1] = (*this)[i] * coeff;
      coeff += one;
    }
    return ret;
  }

  FPS integral() const {
    const int n = (int)this->size();
    FPS ret(n + 1);
    ret[0] = mint(0);
    if (n > 0) ret[1] = mint(1);
    auto mod = mint::get_mod();
    for (int i = 2; i <= n; i++) ret[i] = (-ret[mod % i]) * (mod / i);
    for (int i = 0; i < n; i++) ret[i + 1] *= (*this)[i];
    return ret;
  }

  mint eval(mint x) const {
    mint r = 0, w = 1;
    for (auto &v : *this) r += w * v, w *= x;
    return r;
  }

  FPS log(int deg = -1) const {
    assert(!(*this).empty() && (*this)[0] == mint(1));
    if (deg == -1) deg = (int)this->size();
    return (this->diff() * this->inv(deg)).pre(deg - 1).integral();
  }

  FPS pow(int64_t k, int deg = -1) const {
    const int n = (int)this->size();
    if (deg == -1) deg = n;
    if (k == 0) {
      FPS ret(deg);
      if (deg) ret[0] = 1;
      return ret;
    }
    for (int i = 0; i < n; i++) {
      if ((*this)[i] != mint(0)) {
        mint rev = mint(1) / (*this)[i];
        FPS ret = (((*this * rev) >> i).log(deg) * k).exp(deg);
        ret *= (*this)[i].pow(k);
        ret = (ret << (i * k)).pre(deg);
        if ((int)ret.size() < deg) ret.resize(deg, mint(0));
        return ret;
      }
      if (__int128_t(i + 1) * k >= deg) return FPS(deg, mint(0));
    }
    return FPS(deg, mint(0));
  }

  static void *ntt_ptr;
  static void set_fft();
  FPS &operator*=(const FPS &r);
  void ntt();
  void intt();
  void ntt_doubling();
  static int ntt_pr();
  FPS inv(int deg = -1) const;
  FPS exp(int deg = -1) const;
};
template <typename mint>
void *FormalPowerSeries<mint>::ntt_ptr = nullptr;

/**
 * @brief 多項式/形式的冪級数ライブラリ
 * @docs docs/fps/formal-power-series.md
 */
#line 2 "modulo/binomial.hpp"

#line 4 "modulo/binomial.hpp"
#include <type_traits>
#line 6 "modulo/binomial.hpp"
using namespace std;

// コンストラクタの MAX に 「C(n, r) や fac(n) でクエリを投げる最大の n 」
// を入れると倍速くらいになる
// mod を超えて前計算して 0 割りを踏むバグは対策済み
template <typename T>
struct Binomial {
  vector<T> f, g, h;
  Binomial(int MAX = 0) {
    assert(T::get_mod() != 0 && "Binomial<mint>()");
    f.resize(1, T{1});
    g.resize(1, T{1});
    h.resize(1, T{1});
    if (MAX > 0) extend(MAX + 1);
  }

  void extend(int m = -1) {
    int n = f.size();
    if (m == -1) m = n * 2;
    m = min<int>(m, T::get_mod());
    if (n >= m) return;
    f.resize(m);
    g.resize(m);
    h.resize(m);
    for (int i = n; i < m; i++) f[i] = f[i - 1] * T(i);
    g[m - 1] = f[m - 1].inverse();
    h[m - 1] = g[m - 1] * f[m - 2];
    for (int i = m - 2; i >= n; i--) {
      g[i] = g[i + 1] * T(i + 1);
      h[i] = g[i] * f[i - 1];
    }
  }

  T fac(int i) {
    if (i < 0) return T(0);
    while (i >= (int)f.size()) extend();
    return f[i];
  }

  T finv(int i) {
    if (i < 0) return T(0);
    while (i >= (int)g.size()) extend();
    return g[i];
  }

  T inv(int i) {
    if (i < 0) return -inv(-i);
    while (i >= (int)h.size()) extend();
    return h[i];
  }

  T C(int n, int r) {
    if (n < 0 || n < r || r < 0) return T(0);
    return fac(n) * finv(n - r) * finv(r);
  }

  inline T operator()(int n, int r) { return C(n, r); }

  template <typename I>
  T multinomial(const vector<I>& r) {
    static_assert(is_integral<I>::value == true);
    int n = 0;
    for (auto& x : r) {
      if (x < 0) return T(0);
      n += x;
    }
    T res = fac(n);
    for (auto& x : r) res *= finv(x);
    return res;
  }

  template <typename I>
  T operator()(const vector<I>& r) {
    return multinomial(r);
  }

  T C_naive(int n, int r) {
    if (n < 0 || n < r || r < 0) return T(0);
    T ret = T(1);
    r = min(r, n - r);
    for (int i = 1; i <= r; ++i) ret *= inv(i) * (n--);
    return ret;
  }

  T P(int n, int r) {
    if (n < 0 || n < r || r < 0) return T(0);
    return fac(n) * finv(n - r);
  }

  // [x^r] 1 / (1-x)^n
  T H(int n, int r) {
    if (n < 0 || r < 0) return T(0);
    return r == 0 ? 1 : C(n + r - 1, r);
  }
};
#line 4 "fps/taylor-shift.hpp"

// calculate F(x + a)
template <typename mint>
FormalPowerSeries<mint> TaylorShift(FormalPowerSeries<mint> f, mint a,
                                    Binomial<mint>& C) {
  using fps = FormalPowerSeries<mint>;
  int N = f.size();
  for (int i = 0; i < N; i++) f[i] *= C.fac(i);
  reverse(begin(f), end(f));
  fps g(N, mint(1));
  for (int i = 1; i < N; i++) g[i] = g[i - 1] * a * C.inv(i);
  f = (f * g).pre(N);
  reverse(begin(f), end(f));
  for (int i = 0; i < N; i++) f[i] *= C.finv(i);
  return f;
}

/**
 * @brief 平行移動
 * @docs docs/fps/fps-taylor-shift.md
 */
#line 2 "set-function/subset-convolution.hpp"

#include <array>
#line 5 "set-function/subset-convolution.hpp"
using namespace std;

template <typename mint, int _s>
struct SubsetConvolution {
  using fps = array<mint, _s + 1>;
  static constexpr int s = _s;
  vector<int> pc;

  SubsetConvolution() : pc(1 << s) {
    for (int i = 1; i < (1 << s); i++) pc[i] = pc[i - (i & -i)] + 1;
  }

  void add(fps& l, const fps& r, int d) {
    for (int i = 0; i < d; ++i) l[i] += r[i];
  }

  void sub(fps& l, const fps& r, int d) {
    for (int i = d; i <= s; ++i) l[i] -= r[i];
  }

  void zeta(vector<fps>& a) {
    int n = a.size();
    for (int w = 1; w < n; w *= 2) {
      for (int k = 0; k < n; k += w * 2) {
        for (int i = 0; i < w; ++i) {
          add(a[k + w + i], a[k + i], pc[k + w + i]);
        }
      }
    }
  }

  void mobius(vector<fps>& a) {
    int n = a.size();
    for (int w = n >> 1; w; w >>= 1) {
      for (int k = 0; k < n; k += w * 2) {
        for (int i = 0; i < w; ++i) {
          sub(a[k + w + i], a[k + i], pc[k + w + i]);
        }
      }
    }
  }

  vector<fps> lift(const vector<mint>& a) {
    vector<fps> A(a.size());
    for (int i = 0; i < (int)a.size(); i++) {
      fill(begin(A[i]), end(A[i]), mint());
      A[i][pc[i]] = a[i];
    }
    return A;
  }

  vector<mint> unlift(const vector<fps>& A) {
    vector<mint> a(A.size());
    for (int i = 0; i < (int)A.size(); i++) a[i] = A[i][pc[i]];
    return a;
  }

  void prod(vector<fps>& A, const vector<fps>& B) {
    int n = A.size(), d = __builtin_ctz(n);
    for (int i = 0; i < n; i++) {
      fps c{};
      for (int j = 0; j <= d; j++) {
        for (int k = 0; k <= d - j; k++) {
          c[j + k] += A[i][j] * B[i][k];
        }
      }
      A[i].swap(c);
    }
  }

  vector<mint> multiply(const vector<mint>& a, const vector<mint>& b) {
    vector<fps> A = lift(a), B = lift(b);
    zeta(A), zeta(B);
    prod(A, B);
    mobius(A);
    return unlift(A);
  }
};

/**
 * @brief Subset Convolution
 * @docs docs/set-function/subset-convolution.md
 */
#line 10 "set-function/polynomial-composite-set-power-series.hpp"

template <typename mint, int MAX = 21>
vector<mint> polynomial_composite_set_power_series(int n, FormalPowerSeries<mint> f,
                                            vector<mint> g) {
  assert(0 <= n && n <= MAX);
  static SubsetConvolution<mint, MAX> ss;

  Binomial<mint> binom(f.size());
  if (g[0] != 0) {
    f = TaylorShift(f, g[0], binom);
    g[0] = 0;
  }
  f.resize(n + 1), g.resize(1 << n);

  for (int i = 0; i <= n; i++) f[i] *= binom.fac(i);
  vector h(n + 1, vector(n + 1, vector<mint>{}));
  for (int i = 0; i <= n; i++) h[0][i] = {f[i]};
  for (int k = 1; k <= n; k++) {
    auto A = ss.lift({begin(g) + (1 << (k - 1)), begin(g) + (1 << k)});
    ss.zeta(A);
    for (int j = 0; j <= n - k; j++) {
      h[k][j] = h[k - 1][j];
      auto B = ss.lift(h[k - 1][j + 1]);
      ss.zeta(B);
      ss.prod(B, A);
      ss.mobius(B);
      auto c = ss.unlift(B);
      copy(begin(c), end(c), back_inserter(h[k][j]));
    }
  }
  return h[n][0];
}

/**
 * @brief 集合冪級数の合成
 */

Back to top page