Submission #387750

#TimeUsernameProblemLanguageResultExecution timeMemory
387750rama_pangEscape Route (JOI21_escape_route)C++17
100 / 100
8074 ms197792 KiB
#include "escape_route.h" #include <bits/stdc++.h> using namespace std; using lint = long long; const lint inf = 1e18; vector<lint> calculate_necessary_time( int N, int M, lint S, int Q, vector<int> A, vector<int> B, vector<lint> L, vector<lint> C, vector<int> U, vector<int> V, vector<lint> T) { // Solution: // // First, we compute the minimum time between any 2 cities // x and y, if we start from city x at time 0. We can // compute this in O(N^3) with Dijkstra. // // Next, process each starting city independently. // Generate the shortest path, assuming there is no restriction // on roads, and day is indefinitely long. Then, there is an // edge which is the "bottleneck" - the maximum time we can start // at source to generate this particular shortest path tree. // The bottleneck is the minimum C[i] - L[i] - dist[A[i]] if we // use the i-th edge in shortest path tree to go from A[i] to B[i]. // // After identifying the bottleneck, if we start at any time before // the bottleneck, we can go to the other cities in time dist[v]. // After arriving at city v, we will have to wait until the next day, // then start the journey again. However, we already computed the minimum // time to go from any 2 cities if we start at time 0. Thus, we can // process a single query by bruteforcing the city where we wait until // the next day. // // Since after deleting a bottleneck, we recompute the shortest path tree, // we need O(N^2) recomputations in O(N^2) time each. There are O(N) sources, // and each query is done in O(N). This yields O(N^5 + Q N). // // To speed it up, for every vertex u, and edge adjacent to u, we calculate // the minimum time to reach all vertices v in the same day. Then, instead // of banning an edge, we add an edge instead by sweeping backwards. // Observet the condition when banning an edge: the minimum C[i] - L[i] - dist[i]. // For all edges e adjacent to u, we sort them by C[e] - L[e]. Then, for every vertex, // we only need to consider a single edge (the largest C[e] - L[e]), and add them one // by one. This can be done via a priority queue or segment tree. Then, after getting // the maximum edge e, we reupdate all distances in O(N). // // Time complexity: O(N^4 log N + Q N). M = M + M; for (int i = 0; i < M / 2; i++) { A.push_back(B[i]); B.push_back(A[i]); L.push_back(L[i]); C.push_back(C[i]); } vector<vector<int>> adj(N); for (int i = 0; i < M; i++) { adj[A[i]].push_back(i); } const auto Dijkstra = [&](int s, vector<lint> &dist, lint init) { dist.assign(N, inf); vector<int> done(N); dist[s] = init; for (int rep = 0; rep < N; rep++) { int u = -1; for (int v = 0; v < N; v++) if (!done[v]) { if (u == -1 || dist[u] > dist[v]) { u = v; } } if (dist[u] == inf) { break; } done[u] = 1; for (auto e : adj[u]) { int v = B[e]; if ((dist[u] % S) <= C[e] - L[e]) { dist[v] = min(dist[v], dist[u] + L[e]); } else { dist[v] = min(dist[v], dist[u] + (S - dist[u] % S) % S + L[e]); } } } }; // distance between cities (i, v) if we start at time 0. vector<vector<lint>> dist0(N, vector<lint>(N, inf)); for (int s = 0; s < N; s++) { Dijkstra(s, dist0[s], 0); } // distance between cities (A[i], v) if we start at time C[e] - L[e] // must arrive at same day vector<vector<lint>> distE(M, vector<lint>(N)); for (int e = 0; e < M; e++) { Dijkstra(A[e], distE[e], C[e] - L[e]); for (int i = 0; i < N; i++) { if (distE[e][i] < S) { distE[e][i] -= C[e] - L[e]; } else { distE[e][i] = inf; } } } for (int u = 0; u < N; u++) { sort(begin(adj[u]), end(adj[u]), [&](int i, int j) { return C[i] - L[i] > C[j] - L[j]; }); } vector<vector<int>> queries(N); for (int q = 0; q < Q; q++) { queries[U[q]].emplace_back(q); } vector<lint> ans(Q, inf); for (int s = 0; s < N; s++) { auto &query = queries[s]; sort(begin(query), end(query), [&](int i, int j) { return T[i] < T[j]; }); vector<int> ptr(N); vector<lint> dist(N, inf); dist[s] = 0; // Segment Tree vector<pair<lint, int>> tree(2 * N, {-1, -1}); const auto Update = [&](int p, lint x) { tree[p + N] = {x, p}; p += N; for (p /= 2; p > 0; p /= 2) { tree[p] = max(tree[p * 2], tree[p * 2 + 1]); } }; const auto Query = [&](int l, int r) { pair<lint, int> res = {-1, -1}; for (l += N, r += N; l < r; l /= 2, r /= 2) { if (l & 1) res = max(res, tree[l++]); if (r & 1) res = max(res, tree[--r]); } return res; }; Update(s, C[adj[s][ptr[s]]] - L[adj[s][ptr[s]]]); while (true) { auto top = Query(0, N); lint cur_time = top.first; while (!query.empty() && T[query.back()] > cur_time) { int q = query.back(); query.pop_back(); for (int m = 0; m < N; m++) { // answer query, brute force when first day ends ans[q] = min(ans[q], dist[m] + (m != V[q]) * ((S - (dist[m] + T[q]) % S) % S) + dist0[m][V[q]]); } } if (cur_time < 0) break; int u = top.second; int e = adj[u][ptr[u]]; ptr[u]++; if (ptr[u] < adj[u].size()) { Update(u, C[adj[u][ptr[u]]] - L[adj[u][ptr[u]]] - dist[u]); } else { Update(u, -1); } assert(u == A[e]); for (int v = 0; v < N; v++) if (distE[e][v] != inf) { dist[v] = min(dist[v], dist[u] + distE[e][v]); if (ptr[v] < adj[v].size()) { Update(v, min(cur_time, C[adj[v][ptr[v]]] - L[adj[v][ptr[v]]] - dist[v])); } } } } return ans; }

Compilation message (stderr)

escape_route.cpp: In function 'std::vector<long long int> calculate_necessary_time(int, int, lint, int, std::vector<int>, std::vector<int>, std::vector<long long int>, std::vector<long long int>, std::vector<int>, std::vector<int>, std::vector<long long int>)':
escape_route.cpp:166:18: warning: comparison of integer expressions of different signedness: '__gnu_cxx::__alloc_traits<std::allocator<int>, int>::value_type' {aka 'int'} and 'std::vector<int>::size_type' {aka 'long unsigned int'} [-Wsign-compare]
  166 |       if (ptr[u] < adj[u].size()) {
escape_route.cpp:175:20: warning: comparison of integer expressions of different signedness: '__gnu_cxx::__alloc_traits<std::allocator<int>, int>::value_type' {aka 'int'} and 'std::vector<int>::size_type' {aka 'long unsigned int'} [-Wsign-compare]
  175 |         if (ptr[v] < adj[v].size()) {
#Verdict Execution timeMemoryGrader output
Fetching results...
#Verdict Execution timeMemoryGrader output
Fetching results...
#Verdict Execution timeMemoryGrader output
Fetching results...
#Verdict Execution timeMemoryGrader output
Fetching results...
#Verdict Execution timeMemoryGrader output
Fetching results...