library
This documentation is automatically generated by online-judge-tools/verification-helper
graph/tree_all_distances.hpp
- View this file on GitHub
- Last update: 2024-04-19 02:20:22+09:00
- Include:
#include "graph/tree_all_distances.hpp"
Depends on
- graph/base.hpp
- graph/centroid_decomposition.hpp
- graph/shortest_path/bfs01.hpp
- mod/crt3.hpp
- mod/mod_inv.hpp
- mod/modint.hpp
- mod/modint_common.hpp
- poly/convolution.hpp
- poly/convolution_karatsuba.hpp
- poly/convolution_naive.hpp
- poly/fft.hpp
- poly/ntt.hpp
Verified with
Code
#include "graph/centroid_decomposition.hpp"
#include "poly/convolution.hpp"
// frequency table of distance of all directed pairs.
// sum of result array = N^2
template <typename GT>
vi tree_all_distances(GT& G) {
assert(G.is_prepared());
int N = G.N;
vi ANS(N);
auto f = [&](vc<int>& par, vc<int>& V, int n1, int n2) -> void {
int N = len(par);
vc<int> dist(N);
FOR(i, 1, N) { dist[i] = 1 + dist[par[i]]; }
int mx = MAX(dist);
vi f(1 + mx), g(1 + mx);
FOR(i, 1, 1 + n1) f[dist[i]]++;
FOR(i, 1 + n1, 1 + n1 + n2) g[dist[i]]++;
while (len(f) && f.back() == 0) POP(f);
while (len(g) && g.back() == 0) POP(g);
f = convolution(f, g);
FOR(i, len(f)) ANS[i] += f[i] * 2;
};
centroid_decomposition<1>(G, f);
ANS[0] = N, ANS[1] = 2 * (N - 1);
return ANS;
}#line 2 "graph/base.hpp"
template <typename T>
struct Edge {
int frm, to;
T cost;
int id;
};
template <typename T = int, bool directed = false>
struct Graph {
static constexpr bool is_directed = directed;
int N, M;
using cost_type = T;
using edge_type = Edge<T>;
vector<edge_type> edges;
vector<int> indptr;
vector<edge_type> csr_edges;
vc<int> vc_deg, vc_indeg, vc_outdeg;
bool prepared;
class OutgoingEdges {
public:
OutgoingEdges(const Graph* G, int l, int r) : G(G), l(l), r(r) {}
const edge_type* begin() const {
if (l == r) { return 0; }
return &G->csr_edges[l];
}
const edge_type* end() const {
if (l == r) { return 0; }
return &G->csr_edges[r];
}
private:
const Graph* G;
int l, r;
};
bool is_prepared() { return prepared; }
Graph() : N(0), M(0), prepared(0) {}
Graph(int N) : N(N), M(0), prepared(0) {}
void build(int n) {
N = n, M = 0;
prepared = 0;
edges.clear();
indptr.clear();
csr_edges.clear();
vc_deg.clear();
vc_indeg.clear();
vc_outdeg.clear();
}
void add(int frm, int to, T cost = 1, int i = -1) {
assert(!prepared);
assert(0 <= frm && 0 <= to && to < N);
if (i == -1) i = M;
auto e = edge_type({frm, to, cost, i});
edges.eb(e);
++M;
}
#ifdef FASTIO
// wt, off
void read_tree(bool wt = false, int off = 1) { read_graph(N - 1, wt, off); }
void read_graph(int M, bool wt = false, int off = 1) {
for (int m = 0; m < M; ++m) {
INT(a, b);
a -= off, b -= off;
if (!wt) {
add(a, b);
} else {
T c;
read(c);
add(a, b, c);
}
}
build();
}
#endif
void build() {
assert(!prepared);
prepared = true;
indptr.assign(N + 1, 0);
for (auto&& e: edges) {
indptr[e.frm + 1]++;
if (!directed) indptr[e.to + 1]++;
}
for (int v = 0; v < N; ++v) { indptr[v + 1] += indptr[v]; }
auto counter = indptr;
csr_edges.resize(indptr.back() + 1);
for (auto&& e: edges) {
csr_edges[counter[e.frm]++] = e;
if (!directed)
csr_edges[counter[e.to]++] = edge_type({e.to, e.frm, e.cost, e.id});
}
}
OutgoingEdges operator[](int v) const {
assert(prepared);
return {this, indptr[v], indptr[v + 1]};
}
vc<int> deg_array() {
if (vc_deg.empty()) calc_deg();
return vc_deg;
}
pair<vc<int>, vc<int>> deg_array_inout() {
if (vc_indeg.empty()) calc_deg_inout();
return {vc_indeg, vc_outdeg};
}
int deg(int v) {
if (vc_deg.empty()) calc_deg();
return vc_deg[v];
}
int in_deg(int v) {
if (vc_indeg.empty()) calc_deg_inout();
return vc_indeg[v];
}
int out_deg(int v) {
if (vc_outdeg.empty()) calc_deg_inout();
return vc_outdeg[v];
}
#ifdef FASTIO
void debug() {
print("Graph");
if (!prepared) {
print("frm to cost id");
for (auto&& e: edges) print(e.frm, e.to, e.cost, e.id);
} else {
print("indptr", indptr);
print("frm to cost id");
FOR(v, N) for (auto&& e: (*this)[v]) print(e.frm, e.to, e.cost, e.id);
}
}
#endif
vc<int> new_idx;
vc<bool> used_e;
// G における頂点 V[i] が、新しいグラフで i になるようにする
// {G, es}
Graph<T, directed> rearrange(vc<int> V, bool keep_eid = 0) {
if (len(new_idx) != N) new_idx.assign(N, -1);
int n = len(V);
FOR(i, n) new_idx[V[i]] = i;
Graph<T, directed> G(n);
vc<int> history;
FOR(i, n) {
for (auto&& e: (*this)[V[i]]) {
if (len(used_e) <= e.id) used_e.resize(e.id + 1);
if (used_e[e.id]) continue;
int a = e.frm, b = e.to;
if (new_idx[a] != -1 && new_idx[b] != -1) {
history.eb(e.id);
used_e[e.id] = 1;
int eid = (keep_eid ? e.id : -1);
G.add(new_idx[a], new_idx[b], e.cost, eid);
}
}
}
FOR(i, n) new_idx[V[i]] = -1;
for (auto&& eid: history) used_e[eid] = 0;
G.build();
return G;
}
private:
void calc_deg() {
assert(vc_deg.empty());
vc_deg.resize(N);
for (auto&& e: edges) vc_deg[e.frm]++, vc_deg[e.to]++;
}
void calc_deg_inout() {
assert(vc_indeg.empty());
vc_indeg.resize(N);
vc_outdeg.resize(N);
for (auto&& e: edges) { vc_indeg[e.to]++, vc_outdeg[e.frm]++; }
}
};
#line 3 "graph/shortest_path/bfs01.hpp"
template <typename T, typename GT>
pair<vc<T>, vc<int>> bfs01(GT& G, int v) {
assert(G.is_prepared());
int N = G.N;
vc<T> dist(N, infty<T>);
vc<int> par(N, -1);
deque<int> que;
dist[v] = 0;
que.push_front(v);
while (!que.empty()) {
auto v = que.front();
que.pop_front();
for (auto&& e: G[v]) {
if (dist[e.to] == infty<T> || dist[e.to] > dist[e.frm] + e.cost) {
dist[e.to] = dist[e.frm] + e.cost;
par[e.to] = e.frm;
if (e.cost == 0)
que.push_front(e.to);
else
que.push_back(e.to);
}
}
}
return {dist, par};
}
// 多点スタート。[dist, par, root]
template <typename T, typename GT>
tuple<vc<T>, vc<int>, vc<int>> bfs01(GT& G, vc<int> vs) {
assert(G.is_prepared());
int N = G.N;
vc<T> dist(N, infty<T>);
vc<int> par(N, -1);
vc<int> root(N, -1);
deque<int> que;
for (auto&& v: vs) {
dist[v] = 0;
root[v] = v;
que.push_front(v);
}
while (!que.empty()) {
auto v = que.front();
que.pop_front();
for (auto&& e: G[v]) {
if (dist[e.to] == infty<T> || dist[e.to] > dist[e.frm] + e.cost) {
dist[e.to] = dist[e.frm] + e.cost;
root[e.to] = root[e.frm];
par[e.to] = e.frm;
if (e.cost == 0)
que.push_front(e.to);
else
que.push_back(e.to);
}
}
}
return {dist, par, root};
}
#line 3 "graph/centroid_decomposition.hpp"
// 頂点ベースの重心分解
// f(par, V, indptr)
template <typename F>
void centroid_decomposition_0_dfs(vc<int>& par, vc<int>& vs, F f) {
const int N = len(par);
assert(N >= 1);
int c = -1;
vc<int> sz(N, 1);
FOR_R(i, N) {
if (sz[i] >= ceil<int>(N, 2)) {
c = i;
break;
}
sz[par[i]] += sz[i];
}
vc<int> color(N);
vc<int> V = {c};
int nc = 1;
FOR(v, 1, N) {
if (par[v] == c) { V.eb(v), color[v] = nc++; }
}
if (c > 0) {
for (int a = par[c]; a != -1; a = par[a]) { color[a] = nc, V.eb(a); }
++nc;
}
FOR(i, N) {
if (i != c && color[i] == 0) color[i] = color[par[i]], V.eb(i);
}
vc<int> indptr(nc + 1);
FOR(i, N) indptr[1 + color[i]]++;
FOR(i, nc) indptr[i + 1] += indptr[i];
vc<int> counter = indptr;
vc<int> ord(N);
for (auto& v: V) { ord[counter[color[v]]++] = v; }
vc<int> new_idx(N);
FOR(i, N) new_idx[ord[i]] = i;
vc<int> name(N);
FOR(i, N) name[new_idx[i]] = vs[i];
{
vc<int> tmp(N, -1);
FOR(i, 1, N) {
int a = new_idx[i], b = new_idx[par[i]];
if (a > b) swap(a, b);
tmp[b] = a;
}
swap(par, tmp);
}
f(par, name, indptr);
FOR(k, 1, nc) {
int L = indptr[k], R = indptr[k + 1];
vc<int> par1(R - L, -1);
vc<int> name1(R - L, -1);
name1[0] = name[0];
FOR(i, L, R) name1[i - L] = name[i];
FOR(i, L, R) { par1[i - L] = max(par[i] - L, -1); }
centroid_decomposition_0_dfs(par1, name1, f);
}
}
/*
https://maspypy.com/%e9%87%8d%e5%bf%83%e5%88%86%e8%a7%a3%e3%83%bb1-3%e9%87%8d%e5%bf%83%e5%88%86%e8%a7%a3%e3%81%ae%e3%81%8a%e7%b5%b5%e6%8f%8f%e3%81%8d
centroid_decomposition_1：長さ 2 以上のパス全体
f(par, V, n1, n2)
[1,1+n1]: color 1
[1+n1,1+n1+n2]: color 2
*/
template <typename F>
void centroid_decomposition_1_dfs(vc<int>& par, vc<int> vs, F f) {
const int N = len(par);
assert(N > 1);
if (N == 2) { return; }
int c = -1;
vc<int> sz(N, 1);
FOR_R(i, N) {
if (sz[i] >= ceil<int>(N, 2)) {
c = i;
break;
}
sz[par[i]] += sz[i];
}
vc<int> color(N, -1);
int take = 0;
vc<int> ord(N, -1);
ord[c] = 0;
int p = 1;
FOR(v, 1, N) {
if (par[v] == c && take + sz[v] <= floor<int>(N - 1, 2)) {
color[v] = 0, ord[v] = p++, take += sz[v];
}
}
FOR(i, 1, N) {
if (color[par[i]] == 0) color[i] = 0, ord[i] = p++;
}
int n0 = p - 1;
for (int a = par[c]; a != -1; a = par[a]) { color[a] = 1, ord[a] = p++; }
FOR(i, N) {
if (i != c && color[i] == -1) color[i] = 1, ord[i] = p++;
}
assert(p == N);
int n1 = N - 1 - n0;
vc<int> par0(n0 + 1, -1), par1(n1 + 1, -1), par2(N, -1);
vc<int> V0(n0 + 1), V1(n1 + 1), V2(N);
FOR(v, N) {
int i = ord[v];
V2[i] = vs[v];
if (color[v] != 1) { V0[i] = vs[v]; }
if (color[v] != 0) { V1[max(i - n0, 0)] = vs[v]; }
}
FOR(v, 1, N) {
int a = ord[v], b = ord[par[v]];
if (a > b) swap(a, b);
par2[b] = a;
if (color[v] != 1 && color[par[v]] != 1) par0[b] = a;
if (color[v] != 0 && color[par[v]] != 0)
par1[max(b - n0, 0)] = max(a - n0, 0);
}
f(par2, V2, n0, n1);
centroid_decomposition_1_dfs(par0, V0, f);
centroid_decomposition_1_dfs(par1, V1, f);
}
/*
https://maspypy.com/%e9%87%8d%e5%bf%83%e5%88%86%e8%a7%a3%e3%83%bb1-3%e9%87%8d%e5%bf%83%e5%88%86%e8%a7%a3%e3%81%ae%e3%81%8a%e7%b5%b5%e6%8f%8f%e3%81%8d
f(par2, V2, color)
color in [-1,0,1], -1 is virtual.
*/
template <typename F>
void centroid_decomposition_2_dfs(vc<int>& par, vc<int>& vs, vc<int>& real,
F f) {
const int N = len(par);
assert(N > 1);
if (N == 2) {
if (real[0] && real[1]) {
vc<int> color = {0, 1};
f(par, vs, color);
}
return;
}
int c = -1;
vc<int> sz(N, 1);
FOR_R(i, N) {
if (sz[i] >= ceil<int>(N, 2)) {
c = i;
break;
}
sz[par[i]] += sz[i];
}
vc<int> color(N, -1);
int take = 0;
vc<int> ord(N, -1);
ord[c] = 0;
int p = 1;
FOR(v, 1, N) {
if (par[v] == c && take + sz[v] <= floor<int>(N - 1, 2)) {
color[v] = 0, ord[v] = p++, take += sz[v];
}
}
FOR(i, 1, N) {
if (color[par[i]] == 0) color[i] = 0, ord[i] = p++;
}
int n0 = p - 1;
for (int a = par[c]; a != -1; a = par[a]) { color[a] = 1, ord[a] = p++; }
FOR(i, N) {
if (i != c && color[i] == -1) color[i] = 1, ord[i] = p++;
}
assert(p == N);
int n1 = N - 1 - n0;
vc<int> par0(n0 + 1, -1), par1(n1 + 1, -1), par2(N, -1);
vc<int> V0(n0 + 1), V1(n1 + 1), V2(N);
vc<int> rea0(n0 + 1), rea1(n1 + 1), rea2(N);
FOR(v, N) {
int i = ord[v];
V2[i] = vs[v], rea2[i] = real[v];
if (color[v] != 1) { V0[i] = vs[v], rea0[i] = real[v]; }
if (color[v] != 0) {
V1[max(i - n0, 0)] = vs[v], rea1[max(i - n0, 0)] = real[v];
}
}
FOR(v, 1, N) {
int a = ord[v], b = ord[par[v]];
if (a > b) swap(a, b);
par2[b] = a;
if (color[v] != 1 && color[par[v]] != 1) par0[b] = a;
if (color[v] != 0 && color[par[v]] != 0)
par1[max(b - n0, 0)] = max(a - n0, 0);
}
if (real[c]) {
color.assign(N, -1);
color[0] = 0;
FOR(i, 1, N) color[i] = rea2[i] ? 1 : -1;
f(par2, V2, color);
rea0[0] = rea1[0] = rea2[0] = 0;
}
color.assign(N, -1);
FOR(i, 1, N) if (rea2[i]) color[i] = (i <= n0 ? 0 : 1);
f(par2, V2, color);
centroid_decomposition_2_dfs(par0, V0, rea0, f);
centroid_decomposition_2_dfs(par1, V1, rea1, f);
}
// f(par, V, color)
// V: label in original tree, dfs order
// color in [-1,0,1], color=-1: virtual
template <int MODE, typename GT, typename F>
void centroid_decomposition(GT& G, F f) {
static_assert(!GT::is_directed);
const int N = G.N;
if (N == 1) return;
vc<int> V(N), par(N, -1);
int l = 0, r = 0;
V[r++] = 0;
while (l < r) {
int v = V[l++];
for (auto& e: G[v]) {
if (e.to != par[v]) V[r++] = e.to, par[e.to] = v;
}
}
assert(r == N);
vc<int> new_idx(N);
FOR(i, N) new_idx[V[i]] = i;
vc<int> tmp(N, -1);
FOR(i, 1, N) {
int j = par[i];
tmp[new_idx[i]] = new_idx[j];
}
swap(par, tmp);
static_assert(MODE == 0 || MODE == 1 || MODE == 2);
if constexpr (MODE == 0) { centroid_decomposition_0_dfs(par, V, f); }
elif constexpr(MODE == 1) { centroid_decomposition_1_dfs(par, V, f); }
else {
vc<int> real(N, 1);
centroid_decomposition_2_dfs(par, V, real, f);
}
}
#line 2 "mod/modint_common.hpp"
struct has_mod_impl {
template <class T>
static auto check(T &&x) -> decltype(x.get_mod(), std::true_type{});
template <class T>
static auto check(...) -> std::false_type;
};
template <class T>
class has_mod : public decltype(has_mod_impl::check<T>(std::declval<T>())) {};
template <typename mint>
mint inv(int n) {
static const int mod = mint::get_mod();
static vector<mint> dat = {0, 1};
assert(0 <= n);
if (n >= mod) n %= mod;
while (len(dat) <= n) {
int k = len(dat);
int q = (mod + k - 1) / k;
dat.eb(dat[k * q - mod] * mint::raw(q));
}
return dat[n];
}
template <typename mint>
mint fact(int n) {
static const int mod = mint::get_mod();
assert(0 <= n && n < mod);
static vector<mint> dat = {1, 1};
while (len(dat) <= n) dat.eb(dat[len(dat) - 1] * mint::raw(len(dat)));
return dat[n];
}
template <typename mint>
mint fact_inv(int n) {
static vector<mint> dat = {1, 1};
if (n < 0) return mint(0);
while (len(dat) <= n) dat.eb(dat[len(dat) - 1] * inv<mint>(len(dat)));
return dat[n];
}
template <class mint, class... Ts>
mint fact_invs(Ts... xs) {
return (mint(1) * ... * fact_inv<mint>(xs));
}
template <typename mint, class Head, class... Tail>
mint multinomial(Head &&head, Tail &&... tail) {
return fact<mint>(head) * fact_invs<mint>(std::forward<Tail>(tail)...);
}
template <typename mint>
mint C_dense(int n, int k) {
static vvc<mint> C;
static int H = 0, W = 0;
auto calc = [&](int i, int j) -> mint {
if (i == 0) return (j == 0 ? mint(1) : mint(0));
return C[i - 1][j] + (j ? C[i - 1][j - 1] : 0);
};
if (W <= k) {
FOR(i, H) {
C[i].resize(k + 1);
FOR(j, W, k + 1) { C[i][j] = calc(i, j); }
}
W = k + 1;
}
if (H <= n) {
C.resize(n + 1);
FOR(i, H, n + 1) {
C[i].resize(W);
FOR(j, W) { C[i][j] = calc(i, j); }
}
H = n + 1;
}
return C[n][k];
}
template <typename mint, bool large = false, bool dense = false>
mint C(ll n, ll k) {
assert(n >= 0);
if (k < 0 || n < k) return 0;
if constexpr (dense) return C_dense<mint>(n, k);
if constexpr (!large) return multinomial<mint>(n, k, n - k);
k = min(k, n - k);
mint x(1);
FOR(i, k) x *= mint(n - i);
return x * fact_inv<mint>(k);
}
template <typename mint, bool large = false>
mint C_inv(ll n, ll k) {
assert(n >= 0);
assert(0 <= k && k <= n);
if (!large) return fact_inv<mint>(n) * fact<mint>(k) * fact<mint>(n - k);
return mint(1) / C<mint, 1>(n, k);
}
// [x^d](1-x)^{-n}
template <typename mint, bool large = false, bool dense = false>
mint C_negative(ll n, ll d) {
assert(n >= 0);
if (d < 0) return mint(0);
if (n == 0) { return (d == 0 ? mint(1) : mint(0)); }
return C<mint, large, dense>(n + d - 1, d);
}
#line 3 "mod/modint.hpp"
template <int mod>
struct modint {
static constexpr u32 umod = u32(mod);
static_assert(umod < u32(1) << 31);
u32 val;
static modint raw(u32 v) {
modint x;
x.val = v;
return x;
}
constexpr modint() : val(0) {}
constexpr modint(u32 x) : val(x % umod) {}
constexpr modint(u64 x) : val(x % umod) {}
constexpr modint(u128 x) : val(x % umod) {}
constexpr modint(int x) : val((x %= mod) < 0 ? x + mod : x){};
constexpr modint(ll x) : val((x %= mod) < 0 ? x + mod : x){};
constexpr modint(i128 x) : val((x %= mod) < 0 ? x + mod : x){};
bool operator<(const modint &other) const { return val < other.val; }
modint &operator+=(const modint &p) {
if ((val += p.val) >= umod) val -= umod;
return *this;
}
modint &operator-=(const modint &p) {
if ((val += umod - p.val) >= umod) val -= umod;
return *this;
}
modint &operator*=(const modint &p) {
val = u64(val) * p.val % umod;
return *this;
}
modint &operator/=(const modint &p) {
*this *= p.inverse();
return *this;
}
modint operator-() const { return modint::raw(val ? mod - val : u32(0)); }
modint operator+(const modint &p) const { return modint(*this) += p; }
modint operator-(const modint &p) const { return modint(*this) -= p; }
modint operator*(const modint &p) const { return modint(*this) *= p; }
modint operator/(const modint &p) const { return modint(*this) /= p; }
bool operator==(const modint &p) const { return val == p.val; }
bool operator!=(const modint &p) const { return val != p.val; }
modint inverse() const {
int a = val, b = mod, u = 1, v = 0, t;
while (b > 0) {
t = a / b;
swap(a -= t * b, b), swap(u -= t * v, v);
}
return modint(u);
}
modint pow(ll n) const {
assert(n >= 0);
modint ret(1), mul(val);
while (n > 0) {
if (n & 1) ret *= mul;
mul *= mul;
n >>= 1;
}
return ret;
}
static constexpr int get_mod() { return mod; }
// (n, r), r は 1 の 2^n 乗根
static constexpr pair<int, int> ntt_info() {
if (mod == 120586241) return {20, 74066978};
if (mod == 167772161) return {25, 17};
if (mod == 469762049) return {26, 30};
if (mod == 754974721) return {24, 362};
if (mod == 880803841) return {23, 211};
if (mod == 943718401) return {22, 663003469};
if (mod == 998244353) return {23, 31};
if (mod == 1045430273) return {20, 363};
if (mod == 1051721729) return {20, 330};
if (mod == 1053818881) return {20, 2789};
return {-1, -1};
}
static constexpr bool can_ntt() { return ntt_info().fi != -1; }
};
#ifdef FASTIO
template <int mod>
void rd(modint<mod> &x) {
fastio::rd(x.val);
x.val %= mod;
// assert(0 <= x.val && x.val < mod);
}
template <int mod>
void wt(modint<mod> x) {
fastio::wt(x.val);
}
#endif
using modint107 = modint<1000000007>;
using modint998 = modint<998244353>;
#line 2 "mod/mod_inv.hpp"
// long でも大丈夫
// (val * x - 1) が mod の倍数になるようにする
// 特に mod=0 なら x=0 が満たす
ll mod_inv(ll val, ll mod) {
if (mod == 0) return 0;
mod = abs(mod);
val %= mod;
if (val < 0) val += mod;
ll a = val, b = mod, u = 1, v = 0, t;
while (b > 0) {
t = a / b;
swap(a -= t * b, b), swap(u -= t * v, v);
}
if (u < 0) u += mod;
return u;
}
#line 1 "mod/crt3.hpp"
constexpr u32 mod_pow_constexpr(u64 a, u64 n, u32 mod) {
a %= mod;
u64 res = 1;
FOR(32) {
if (n & 1) res = res * a % mod;
a = a * a % mod, n /= 2;
}
return res;
}
template <typename T, u32 p0, u32 p1, u32 p2>
T CRT3(u64 a0, u64 a1, u64 a2) {
static_assert(p0 < p1 && p1 < p2);
static constexpr u64 x0_1 = mod_pow_constexpr(p0, p1 - 2, p1);
static constexpr u64 x01_2 = mod_pow_constexpr(u64(p0) * p1 % p2, p2 - 2, p2);
u64 c = (a1 - a0 + p1) * x0_1 % p1;
u64 a = a0 + c * p0;
c = (a2 - a % p2 + p2) * x01_2 % p2;
return T(a) + T(c) * T(p0) * T(p1);
}
#line 2 "poly/convolution_naive.hpp"
template <class T, typename enable_if<!has_mod<T>::value>::type* = nullptr>
vc<T> convolution_naive(const vc<T>& a, const vc<T>& b) {
int n = int(a.size()), m = int(b.size());
if (n > m) return convolution_naive<T>(b, a);
if (n == 0) return {};
vector<T> ans(n + m - 1);
FOR(i, n) FOR(j, m) ans[i + j] += a[i] * b[j];
return ans;
}
template <class T, typename enable_if<has_mod<T>::value>::type* = nullptr>
vc<T> convolution_naive(const vc<T>& a, const vc<T>& b) {
int n = int(a.size()), m = int(b.size());
if (n > m) return convolution_naive<T>(b, a);
if (n == 0) return {};
vc<T> ans(n + m - 1);
if (n <= 16 && (T::get_mod() < (1 << 30))) {
for (int k = 0; k < n + m - 1; ++k) {
int s = max(0, k - m + 1);
int t = min(n, k + 1);
u64 sm = 0;
for (int i = s; i < t; ++i) { sm += u64(a[i].val) * (b[k - i].val); }
ans[k] = sm;
}
} else {
for (int k = 0; k < n + m - 1; ++k) {
int s = max(0, k - m + 1);
int t = min(n, k + 1);
u128 sm = 0;
for (int i = s; i < t; ++i) { sm += u64(a[i].val) * (b[k - i].val); }
ans[k] = T::raw(sm % T::get_mod());
}
}
return ans;
}
#line 2 "poly/convolution_karatsuba.hpp"
// 任意の環でできる
template <typename T>
vc<T> convolution_karatsuba(const vc<T>& f, const vc<T>& g) {
const int thresh = 30;
if (min(len(f), len(g)) <= thresh) return convolution_naive(f, g);
int n = max(len(f), len(g));
int m = ceil(n, 2);
vc<T> f1, f2, g1, g2;
if (len(f) < m) f1 = f;
if (len(f) >= m) f1 = {f.begin(), f.begin() + m};
if (len(f) >= m) f2 = {f.begin() + m, f.end()};
if (len(g) < m) g1 = g;
if (len(g) >= m) g1 = {g.begin(), g.begin() + m};
if (len(g) >= m) g2 = {g.begin() + m, g.end()};
vc<T> a = convolution_karatsuba(f1, g1);
vc<T> b = convolution_karatsuba(f2, g2);
FOR(i, len(f2)) f1[i] += f2[i];
FOR(i, len(g2)) g1[i] += g2[i];
vc<T> c = convolution_karatsuba(f1, g1);
vc<T> F(len(f) + len(g) - 1);
assert(2 * m + len(b) <= len(F));
FOR(i, len(a)) F[i] += a[i], c[i] -= a[i];
FOR(i, len(b)) F[2 * m + i] += b[i], c[i] -= b[i];
if (c.back() == T(0)) c.pop_back();
FOR(i, len(c)) if (c[i] != T(0)) F[m + i] += c[i];
return F;
}
#line 2 "poly/ntt.hpp"
template <class mint>
void ntt(vector<mint>& a, bool inverse) {
assert(mint::can_ntt());
const int rank2 = mint::ntt_info().fi;
const int mod = mint::get_mod();
static array<mint, 30> root, iroot;
static array<mint, 30> rate2, irate2;
static array<mint, 30> rate3, irate3;
assert(rank2 != -1 && len(a) <= (1 << max(0, rank2)));
static bool prepared = 0;
if (!prepared) {
prepared = 1;
root[rank2] = mint::ntt_info().se;
iroot[rank2] = mint(1) / root[rank2];
FOR_R(i, rank2) {
root[i] = root[i + 1] * root[i + 1];
iroot[i] = iroot[i + 1] * iroot[i + 1];
}
mint prod = 1, iprod = 1;
for (int i = 0; i <= rank2 - 2; i++) {
rate2[i] = root[i + 2] * prod;
irate2[i] = iroot[i + 2] * iprod;
prod *= iroot[i + 2];
iprod *= root[i + 2];
}
prod = 1, iprod = 1;
for (int i = 0; i <= rank2 - 3; i++) {
rate3[i] = root[i + 3] * prod;
irate3[i] = iroot[i + 3] * iprod;
prod *= iroot[i + 3];
iprod *= root[i + 3];
}
}
int n = int(a.size());
int h = topbit(n);
assert(n == 1 << h);
if (!inverse) {
int len = 0;
while (len < h) {
if (h - len == 1) {
int p = 1 << (h - len - 1);
mint rot = 1;
FOR(s, 1 << len) {
int offset = s << (h - len);
FOR(i, p) {
auto l = a[i + offset];
auto r = a[i + offset + p] * rot;
a[i + offset] = l + r;
a[i + offset + p] = l - r;
}
rot *= rate2[topbit(~s & -~s)];
}
len++;
} else {
int p = 1 << (h - len - 2);
mint rot = 1, imag = root[2];
for (int s = 0; s < (1 << len); s++) {
mint rot2 = rot * rot;
mint rot3 = rot2 * rot;
int offset = s << (h - len);
for (int i = 0; i < p; i++) {
u64 mod2 = u64(mod) * mod;
u64 a0 = a[i + offset].val;
u64 a1 = u64(a[i + offset + p].val) * rot.val;
u64 a2 = u64(a[i + offset + 2 * p].val) * rot2.val;
u64 a3 = u64(a[i + offset + 3 * p].val) * rot3.val;
u64 a1na3imag = (a1 + mod2 - a3) % mod * imag.val;
u64 na2 = mod2 - a2;
a[i + offset] = a0 + a2 + a1 + a3;
a[i + offset + 1 * p] = a0 + a2 + (2 * mod2 - (a1 + a3));
a[i + offset + 2 * p] = a0 + na2 + a1na3imag;
a[i + offset + 3 * p] = a0 + na2 + (mod2 - a1na3imag);
}
rot *= rate3[topbit(~s & -~s)];
}
len += 2;
}
}
} else {
mint coef = mint(1) / mint(len(a));
FOR(i, len(a)) a[i] *= coef;
int len = h;
while (len) {
if (len == 1) {
int p = 1 << (h - len);
mint irot = 1;
FOR(s, 1 << (len - 1)) {
int offset = s << (h - len + 1);
FOR(i, p) {
u64 l = a[i + offset].val;
u64 r = a[i + offset + p].val;
a[i + offset] = l + r;
a[i + offset + p] = (mod + l - r) * irot.val;
}
irot *= irate2[topbit(~s & -~s)];
}
len--;
} else {
int p = 1 << (h - len);
mint irot = 1, iimag = iroot[2];
FOR(s, (1 << (len - 2))) {
mint irot2 = irot * irot;
mint irot3 = irot2 * irot;
int offset = s << (h - len + 2);
for (int i = 0; i < p; i++) {
u64 a0 = a[i + offset + 0 * p].val;
u64 a1 = a[i + offset + 1 * p].val;
u64 a2 = a[i + offset + 2 * p].val;
u64 a3 = a[i + offset + 3 * p].val;
u64 x = (mod + a2 - a3) * iimag.val % mod;
a[i + offset] = a0 + a1 + a2 + a3;
a[i + offset + 1 * p] = (a0 + mod - a1 + x) * irot.val;
a[i + offset + 2 * p] = (a0 + a1 + 2 * mod - a2 - a3) * irot2.val;
a[i + offset + 3 * p] = (a0 + 2 * mod - a1 - x) * irot3.val;
}
irot *= irate3[topbit(~s & -~s)];
}
len -= 2;
}
}
}
}
#line 1 "poly/fft.hpp"
namespace CFFT {
using real = double;
struct C {
real x, y;
C() : x(0), y(0) {}
C(real x, real y) : x(x), y(y) {}
inline C operator+(const C& c) const { return C(x + c.x, y + c.y); }
inline C operator-(const C& c) const { return C(x - c.x, y - c.y); }
inline C operator*(const C& c) const {
return C(x * c.x - y * c.y, x * c.y + y * c.x);
}
inline C conj() const { return C(x, -y); }
};
const real PI = acosl(-1);
int base = 1;
vector<C> rts = {{0, 0}, {1, 0}};
vector<int> rev = {0, 1};
void ensure_base(int nbase) {
if (nbase <= base) return;
rev.resize(1 << nbase);
rts.resize(1 << nbase);
for (int i = 0; i < (1 << nbase); i++) {
rev[i] = (rev[i >> 1] >> 1) + ((i & 1) << (nbase - 1));
}
while (base < nbase) {
real angle = PI * 2.0 / (1 << (base + 1));
for (int i = 1 << (base - 1); i < (1 << base); i++) {
rts[i << 1] = rts[i];
real angle_i = angle * (2 * i + 1 - (1 << base));
rts[(i << 1) + 1] = C(cos(angle_i), sin(angle_i));
}
++base;
}
}
void fft(vector<C>& a, int n) {
assert((n & (n - 1)) == 0);
int zeros = __builtin_ctz(n);
ensure_base(zeros);
int shift = base - zeros;
for (int i = 0; i < n; i++) {
if (i < (rev[i] >> shift)) { swap(a[i], a[rev[i] >> shift]); }
}
for (int k = 1; k < n; k <<= 1) {
for (int i = 0; i < n; i += 2 * k) {
for (int j = 0; j < k; j++) {
C z = a[i + j + k] * rts[j + k];
a[i + j + k] = a[i + j] - z;
a[i + j] = a[i + j] + z;
}
}
}
}
} // namespace CFFT
#line 9 "poly/convolution.hpp"
template <class mint>
vector<mint> convolution_ntt(vector<mint> a, vector<mint> b) {
if (a.empty() || b.empty()) return {};
int n = int(a.size()), m = int(b.size());
int sz = 1;
while (sz < n + m - 1) sz *= 2;
// sz = 2^k のときの高速化。分割統治的なやつで損しまくるので。
if ((n + m - 3) <= sz / 2) {
auto a_last = a.back(), b_last = b.back();
a.pop_back(), b.pop_back();
auto c = convolution(a, b);
c.resize(n + m - 1);
c[n + m - 2] = a_last * b_last;
FOR(i, len(a)) c[i + len(b)] += a[i] * b_last;
FOR(i, len(b)) c[i + len(a)] += b[i] * a_last;
return c;
}
a.resize(sz), b.resize(sz);
bool same = a == b;
ntt(a, 0);
if (same) {
b = a;
} else {
ntt(b, 0);
}
FOR(i, sz) a[i] *= b[i];
ntt(a, 1);
a.resize(n + m - 1);
return a;
}
template <typename mint>
vector<mint> convolution_garner(const vector<mint>& a, const vector<mint>& b) {
int n = len(a), m = len(b);
if (!n || !m) return {};
static constexpr int p0 = 167772161;
static constexpr int p1 = 469762049;
static constexpr int p2 = 754974721;
using mint0 = modint<p0>;
using mint1 = modint<p1>;
using mint2 = modint<p2>;
vc<mint0> a0(n), b0(m);
vc<mint1> a1(n), b1(m);
vc<mint2> a2(n), b2(m);
FOR(i, n) a0[i] = a[i].val, a1[i] = a[i].val, a2[i] = a[i].val;
FOR(i, m) b0[i] = b[i].val, b1[i] = b[i].val, b2[i] = b[i].val;
auto c0 = convolution_ntt<mint0>(a0, b0);
auto c1 = convolution_ntt<mint1>(a1, b1);
auto c2 = convolution_ntt<mint2>(a2, b2);
vc<mint> c(len(c0));
FOR(i, n + m - 1) {
c[i] = CRT3<mint, p0, p1, p2>(c0[i].val, c1[i].val, c2[i].val);
}
return c;
}
template <typename R>
vc<double> convolution_fft(const vc<R>& a, const vc<R>& b) {
using C = CFFT::C;
int need = (int)a.size() + (int)b.size() - 1;
int nbase = 1;
while ((1 << nbase) < need) nbase++;
CFFT::ensure_base(nbase);
int sz = 1 << nbase;
vector<C> fa(sz);
for (int i = 0; i < sz; i++) {
double x = (i < (int)a.size() ? a[i] : 0);
double y = (i < (int)b.size() ? b[i] : 0);
fa[i] = C(x, y);
}
CFFT::fft(fa, sz);
C r(0, -0.25 / (sz >> 1)), s(0, 1), t(0.5, 0);
for (int i = 0; i <= (sz >> 1); i++) {
int j = (sz - i) & (sz - 1);
C z = (fa[j] * fa[j] - (fa[i] * fa[i]).conj()) * r;
fa[j] = (fa[i] * fa[i] - (fa[j] * fa[j]).conj()) * r;
fa[i] = z;
}
for (int i = 0; i < (sz >> 1); i++) {
C A0 = (fa[i] + fa[i + (sz >> 1)]) * t;
C A1 = (fa[i] - fa[i + (sz >> 1)]) * t * CFFT::rts[(sz >> 1) + i];
fa[i] = A0 + A1 * s;
}
CFFT::fft(fa, sz >> 1);
vector<double> ret(need);
for (int i = 0; i < need; i++) {
ret[i] = (i & 1 ? fa[i >> 1].y : fa[i >> 1].x);
}
return ret;
}
vector<ll> convolution(const vector<ll>& a, const vector<ll>& b) {
int n = len(a), m = len(b);
if (!n || !m) return {};
if (min(n, m) <= 2500) return convolution_naive(a, b);
ll abs_sum_a = 0, abs_sum_b = 0;
ll LIM = 1e15;
FOR(i, n) abs_sum_a = min(LIM, abs_sum_a + abs(a[i]));
FOR(i, m) abs_sum_b = min(LIM, abs_sum_b + abs(b[i]));
if (i128(abs_sum_a) * abs_sum_b < 1e15) {
vc<double> c = convolution_fft<ll>(a, b);
vc<ll> res(len(c));
FOR(i, len(c)) res[i] = ll(floor(c[i] + .5));
return res;
}
static constexpr unsigned long long MOD1 = 754974721; // 2^24
static constexpr unsigned long long MOD2 = 167772161; // 2^25
static constexpr unsigned long long MOD3 = 469762049; // 2^26
static constexpr unsigned long long M2M3 = MOD2 * MOD3;
static constexpr unsigned long long M1M3 = MOD1 * MOD3;
static constexpr unsigned long long M1M2 = MOD1 * MOD2;
static constexpr unsigned long long M1M2M3 = MOD1 * MOD2 * MOD3;
static const unsigned long long i1 = mod_inv(MOD2 * MOD3, MOD1);
static const unsigned long long i2 = mod_inv(MOD1 * MOD3, MOD2);
static const unsigned long long i3 = mod_inv(MOD1 * MOD2, MOD3);
using mint1 = modint<MOD1>;
using mint2 = modint<MOD2>;
using mint3 = modint<MOD3>;
vc<mint1> a1(n), b1(m);
vc<mint2> a2(n), b2(m);
vc<mint3> a3(n), b3(m);
FOR(i, n) a1[i] = a[i], a2[i] = a[i], a3[i] = a[i];
FOR(i, m) b1[i] = b[i], b2[i] = b[i], b3[i] = b[i];
auto c1 = convolution_ntt<mint1>(a1, b1);
auto c2 = convolution_ntt<mint2>(a2, b2);
auto c3 = convolution_ntt<mint3>(a3, b3);
vc<ll> c(n + m - 1);
FOR(i, n + m - 1) {
u64 x = 0;
x += (c1[i].val * i1) % MOD1 * M2M3;
x += (c2[i].val * i2) % MOD2 * M1M3;
x += (c3[i].val * i3) % MOD3 * M1M2;
ll diff = c1[i].val - ((long long)(x) % (long long)(MOD1));
if (diff < 0) diff += MOD1;
static constexpr unsigned long long offset[5]
= {0, 0, M1M2M3, 2 * M1M2M3, 3 * M1M2M3};
x -= offset[diff % 5];
c[i] = x;
}
return c;
}
template <typename mint>
vc<mint> convolution(const vc<mint>& a, const vc<mint>& b) {
int n = len(a), m = len(b);
if (!n || !m) return {};
if (mint::can_ntt()) {
if (min(n, m) <= 50) return convolution_karatsuba<mint>(a, b);
return convolution_ntt(a, b);
}
if (min(n, m) <= 200) return convolution_karatsuba<mint>(a, b);
return convolution_garner(a, b);
}
#line 3 "graph/tree_all_distances.hpp"
// frequency table of distance of all directed pairs.
// sum of result array = N^2
template <typename GT>
vi tree_all_distances(GT& G) {
assert(G.is_prepared());
int N = G.N;
vi ANS(N);
auto f = [&](vc<int>& par, vc<int>& V, int n1, int n2) -> void {
int N = len(par);
vc<int> dist(N);
FOR(i, 1, N) { dist[i] = 1 + dist[par[i]]; }
int mx = MAX(dist);
vi f(1 + mx), g(1 + mx);
FOR(i, 1, 1 + n1) f[dist[i]]++;
FOR(i, 1 + n1, 1 + n1 + n2) g[dist[i]]++;
while (len(f) && f.back() == 0) POP(f);
while (len(g) && g.back() == 0) POP(g);
f = convolution(f, g);
FOR(i, len(f)) ANS[i] += f[i] * 2;
};
centroid_decomposition<1>(G, f);
ANS[0] = N, ANS[1] = 2 * (N - 1);
return ANS;
}