#include using namespace std; // #pragma GCC target("arch=skylake-avx512") // #include // using namespace atcoder; // #define NDEBUG #pragma region template // Define using ll = long long; using ull = unsigned long long; using ld = long double; template using pvector = vector>; template using rpriority_queue = priority_queue, greater>; constexpr const ll dx[4] = {1, 0, -1, 0}; constexpr const ll dy[4] = {0, 1, 0, -1}; constexpr const ll MOD = 1e9 + 7; constexpr const ll mod = 998244353; constexpr const ll INF = 1LL << 60; constexpr const ll inf = 1 << 30; constexpr const char rt = '\n'; constexpr const char sp = ' '; #define rt(i, n) (i == (ll) (n) -1 ? rt : sp) #define len(x) ((ll) (x).size()) #define all(x) (x).begin(), (x).end() #define rall(x) (x).rbegin(), (x).rend() #define mp make_pair #define mt make_tuple #define pb push_back #define eb emplace_back #define ifn(x) if (not(x)) #define elif else if #define elifn else ifn #define fi first #define se second #define uniq(x) (sort(all(x)), (x).erase(unique(all(x)), (x).end())) #define bis(x, y) ((ll) (lower_bound(all(x), y) - (x).begin())) using graph = vector>; template using wgraph = vector>>; bool __DIRECTED__ = true; bool __ZERO_INDEXED__ = false; istream &operator>>(istream &is, graph &g) { ll a, b; is >> a >> b; if (__ZERO_INDEXED__ == false) a--, b--; g[a].pb(b); if (__DIRECTED__ == false) g[b].pb(a); return is; } template istream &operator>>(istream &is, wgraph &g) { ll a, b; T c; is >> a >> b >> c; if (__ZERO_INDEXED__ == false) a--, b--; g[a].pb({b, c}); if (__DIRECTED__ == false) g[b].pb({a, c}); return is; } template bool chmax(T &a, const T &b) { if (a < b) { a = b; return 1; } return 0; } template bool chmin(T &a, const T &b) { if (a > b) { a = b; return 1; } return 0; } // Debug #ifdef NDEBUG #define debug(...) #define dumpi(a, h, w) #define vdumpi(a, n) #define dump(a, h, w) #define vdump(a, n) #else #define debug(...) \ { \ cerr << __LINE__ << ": " << #__VA_ARGS__ << " = "; \ for (auto &&__i : {__VA_ARGS__}) cerr << "[" << __i << "] "; \ cerr << rt; \ } #define dumpi(a, h, w) \ { \ cerr << __LINE__ << ": " << #a << " = [" << rt; \ rep(__i, h) { \ if (__i) cerr << ",\n"; \ cerr << "["; \ rep(__j, w) { \ if (__j) cerr << ", "; \ if (abs(a[__i][__j]) >= INF / 2 and a[__i][__j] <= -INF / 2) cerr << '-'; \ if (abs(a[__i][__j]) >= INF / 2) cerr << "∞"; \ else \ cerr << a[__i][__j]; \ } \ cerr << "]"; \ } \ cerr << "\n]" << rt; \ } #define vdumpi(a, n) \ { \ cerr << __LINE__ << ": " << #a << " = ["; \ rep(__i, n) { \ if (__i) cerr << ", "; \ if (abs(a[__i]) >= INF / 2 and a[__i] <= -INF / 2) cerr << '-'; \ if (abs(a[__i]) >= INF / 2) cerr << "∞"; \ else \ cerr << a[__i]; \ } \ cerr << "]" << rt; \ } #define dump(a, h, w) \ { \ cerr << __LINE__ << ": " << #a << " = [" << rt; \ rep(__i, h) { \ if (__i) cerr << ",\n"; \ cerr << "["; \ rep(__j, w) { \ if (__j) cerr << ", "; \ cerr << a[__i][__j]; \ } \ cerr << "]"; \ } \ cerr << "\n]" << rt; \ } #define vdump(a, n) \ { \ cerr << __LINE__ << ": " << #a << " = ["; \ rep(__i, n) { \ if (__i) cerr << ", "; \ cerr << a[__i]; \ } \ cerr << "]" << rt; \ } #endif template istream &operator>>(istream &is, pair &p) { is >> p.first >> p.second; return is; } template ostream &operator<<(ostream &os, const pair &p) { os << p.first << ' ' << p.second; return os; } // Loop #define inc(i, a, n) for (ll i = (a), _##i = (n); i <= _##i; ++i) #define dec(i, a, n) for (ll i = (a), _##i = (n); i >= _##i; --i) #define rep(i, n) for (ll i = 0, _##i = (n); i < _##i; ++i) #define each(i, a) for (auto &&i : a) // Stream #define fout(n) cout << fixed << setprecision(n) struct io { io() { cin.tie(nullptr), ios::sync_with_stdio(false); } } io; // Speed #pragma GCC target("sse,sse2,sse3,ssse3,sse4,popcnt,abm,mmx,avx,avx2,tune=native") #pragma GCC optimize("Ofast,unroll-loops") // Math inline constexpr ll gcd(const ll a, const ll b) { return b ? gcd(b, a % b) : a; } inline constexpr ll lcm(const ll a, const ll b) { return a / gcd(a, b) * b; } inline constexpr ll modulo(const ll n, const ll m = MOD) { ll k = n % m; return k + m * (k < 0); } inline constexpr ll chmod(ll &n, const ll m = MOD) { n %= m; return n += m * (n < 0); } inline constexpr ll mpow(ll a, ll n, const ll m = MOD) { ll r = 1; rep(i, 64) { if (n & (1LL << i)) r *= a; chmod(r, m); a *= a; chmod(a, m); } return r; } inline ll inv(const ll n, const ll m = MOD) { ll a = n, b = m, x = 1, y = 0; while (b) { ll t = a / b; a -= t * b; swap(a, b); x -= t * y; swap(x, y); } return modulo(x, m); } unsigned long long binary_gcd(unsigned long long x, unsigned long long y) { if (!x | !y) return x | y; unsigned long long cx = __builtin_ctzll(x), cy = __builtin_ctzll(y); x >>= cx, y >>= cy; while (x ^ y) { if (x > y) { x = (x - y) >> __builtin_ctzll(x ^ y); } else { y = (y - x) >> __builtin_ctzll(x ^ y); } } return x << min(cx, cy); } inline long long binary_gcd(long long x, long long y) { return binary_gcd((unsigned long long) (abs(x)), (unsigned long long) (abs(y))); } #define codeforces \ ll testcases; \ cin >> testcases; \ rep(testcase, testcases) #define gcj(s) cout << s << testcase + 1 << ": " #pragma endregion signed main() { ll N; cin >> N; ll L[N], res; vector G[100], used(N); rep(i, N) { ll S; cin >> L[i] >> S; G[S - 1].pb(i); } function dfs = [&](ll v, ll x) { if (used[v]) { res += x; } else { used[v] = true; rep(i, G[v].size()) dfs(G[v][i], min(x, L[v])); } }; res = accumulate(L, L + N, 0LL); rep(i, N) if (!used[i]) dfs(i, INF); cout << res / 2; cout << (res & 1 ? ".5" : ".0") << rt; }