#include<bits/stdc++.h>
#pragma GCC optimize("-Ofast")
//#pragma GCC optimize("trapv")
#pragma GCC target("sse,sse2,sse3,ssse3,sse4,sse4.2,popcnt,abm,mmx,avx2,tune=native")
#pragma GCC optimize("-ffast-math")
#pragma GCC optimize("-funroll-loops")
#define I inline void
#define S struct
#define vi vector<int>
#define vii vector<pair<int,int>>
#define pii pair<int,int>
#define pll pair<ll,ll>
using namespace std ;
using ll = long long ;
using ld = long double ;
const int N = 2e5 + 7 , mod = 1e9 + 7 ;
const ll inf = 2e18 ;
// How interesting!
int n ;
vector<double> d[2], pd[2];
double ans ;
void solve2(){
double sum1 = 0.0 ;
for(int i = 0 ;i < n; ++ i){
sum1 += d[0][i] ;
int k = lower_bound(pd[1].begin() , pd[1].end() , sum1 + i + 1 ) - pd[1].begin() ;
if(k != pd[1].size()){
ans = max(ans , sum1 - k - 1) ;
}
}
}
int main(){
ios_base::sync_with_stdio(0) ;
cin.tie(0) ;
//freopen("in.in" , "r" , stdin) ;
cin >> n;
for(int i = 0 ;i < n; ++ i){
double a , b;
cin >> a >> b ;
a-=1.0 ; b-=1.0 ;
d[0].push_back(a) ;
d[1].push_back(b) ;
}
for(int i = 0 ;i < 2 ; ++ i){
sort(d[i].begin() , d[i].end(), greater<double>()) ;
pd[i].resize(n) ;
}
pd[0][0] = d[0][0] +1.0;
pd[1][0] = d[1][0] +1.0;
for(int i = 0 ; i < 2 ; ++ i){
for(int j = 1 ;j < n ; ++ j)
pd[i][j] = pd[i][j-1] + d[i][j] +1.0;
}
solve2() ;
swap(d[0] , d[1]) ;
swap(pd[0] , pd[1]) ;
solve2() ;
cout<< fixed <<setprecision(4) << ans ;
return 0 ;
}
/*
- bounds sir (segtree = 4N, eulerTour = 2N, ...)
- a variable defined twice?
- will overflow?
- is it a good complexity?
- don't mess up indices (0-indexed vs 1-indexed)
- reset everything between testcases.
*/
