# https://github.com/tatyam-prime/SortedSet/blob/main/SortedSet.py
import math
from bisect import bisect_left, bisect_right
from typing import Generic, Iterable, Iterator, TypeVar, Union, List
T = TypeVar('T')

class SortedSet(Generic[T]):
    BUCKET_RATIO = 50
    REBUILD_RATIO = 170

    def _build(self, a=None) -> None:
        "Evenly divide `a` into buckets."
        if a is None: a = list(self)
        size = self.size = len(a)
        bucket_size = int(math.ceil(math.sqrt(size / self.BUCKET_RATIO)))
        self.a = [a[size * i // bucket_size : size * (i + 1) // bucket_size] for i in range(bucket_size)]
    
    def __init__(self, a: Iterable[T] = []) -> None:
        "Make a new SortedSet from iterable. / O(N) if sorted and unique / O(N log N)"
        a = list(a)
        if not all(a[i] < a[i + 1] for i in range(len(a) - 1)):
            a = sorted(set(a))
        self._build(a)

    def __iter__(self) -> Iterator[T]:
        for i in self.a:
            for j in i: yield j

    def __reversed__(self) -> Iterator[T]:
        for i in reversed(self.a):
            for j in reversed(i): yield j
    
    def __len__(self) -> int:
        return self.size
    
    def __repr__(self) -> str:
        return "SortedSet" + str(self.a)
    
    def __str__(self) -> str:
        s = str(list(self))
        return "{" + s[1 : len(s) - 1] + "}"

    def _find_bucket(self, x: T) -> List[T]:
        "Find the bucket which should contain x. self must not be empty."
        for a in self.a:
            if x <= a[-1]: return a
        return a

    def __contains__(self, x: T) -> bool:
        if self.size == 0: return False
        a = self._find_bucket(x)
        i = bisect_left(a, x)
        return i != len(a) and a[i] == x

    def add(self, x: T) -> bool:
        "Add an element and return True if added. / O(√N)"
        if self.size == 0:
            self.a = [[x]]
            self.size = 1
            return True
        a = self._find_bucket(x)
        i = bisect_left(a, x)
        if i != len(a) and a[i] == x: return False
        a.insert(i, x)
        self.size += 1
        if len(a) > len(self.a) * self.REBUILD_RATIO:
            self._build()
        return True

    def discard(self, x: T) -> bool:
        "Remove an element and return True if removed. / O(√N)"
        if self.size == 0: return False
        a = self._find_bucket(x)
        i = bisect_left(a, x)
        if i == len(a) or a[i] != x: return False
        a.pop(i)
        self.size -= 1
        if len(a) == 0: self._build()
        return True
    
    def lt(self, x: T) -> Union[T, None]:
        "Find the largest element < x, or None if it doesn't exist."
        for a in reversed(self.a):
            if a[0] < x:
                return a[bisect_left(a, x) - 1]

    def le(self, x: T) -> Union[T, None]:
        "Find the largest element <= x, or None if it doesn't exist."
        for a in reversed(self.a):
            if a[0] <= x:
                return a[bisect_right(a, x) - 1]

    def gt(self, x: T) -> Union[T, None]:
        "Find the smallest element > x, or None if it doesn't exist."
        for a in self.a:
            if a[-1] > x:
                return a[bisect_right(a, x)]

    def ge(self, x: T) -> Union[T, None]:
        "Find the smallest element >= x, or None if it doesn't exist."
        for a in self.a:
            if a[-1] >= x:
                return a[bisect_left(a, x)]
    
    def __getitem__(self, x: int) -> T:
        "Return the x-th element, or IndexError if it doesn't exist."
        if x < 0: x += self.size
        if x < 0: raise IndexError
        for a in self.a:
            if x < len(a): return a[x]
            x -= len(a)
        raise IndexError
    
    def index(self, x: T) -> int:
        "Count the number of elements < x."
        ans = 0
        for a in self.a:
            if a[-1] >= x:
                return ans + bisect_left(a, x)
            ans += len(a)
        return ans

    def index_right(self, x: T) -> int:
        "Count the number of elements <= x."
        ans = 0
        for a in self.a:
            if a[-1] > x:
                return ans + bisect_right(a, x)
            ans += len(a)
        return ans

import sys
input = sys.stdin.buffer.readline
sys.setrecursionlimit(10 ** 7)

class SegTree(object):
    def __init__(self, N, op_data, u_data):
        self._n = N
        self.log = (N-1).bit_length()
        self.size = 1 << self.log

        self.op = op_data
        self.e = u_data

        self.data = [u_data] * (2 * self.size)
        # self.len = [1] * (2 * self.size)

    def _update(self, i):
        self.data[i] = self.op(self.data[i << 1], self.data[i << 1 | 1])

    def initialize(self, arr=None):
        """ segtreeをarrで初期化する。len(arr) == Nにすること """
        if arr:
            for i, a in enumerate(arr, self.size):
                self.data[i] = a
        for i in reversed(range(1, self.size)):
            self._update(i)
            # self.len[i] = self.len[i << 1] + self.len[i << 1 | 1]

    def update(self, p, x):
        """ data[p] = x とする (0-indexed)"""
        p += self.size
        self.data[p] = x
        for i in range(1, self.log + 1):
            self._update(p >> i)

    def get(self, p):
        """ data[p]を返す """
        return self.data[p + self.size]

    def prod(self, l, r):
        """
        op_data(data[l], data[l+1], ..., data[r-1])を返す (0-indexed)
        """
        sml = self.e
        smr = self.e
        l += self.size
        r += self.size

        while l < r:
            if l & 1:
                sml = self.op(sml, self.data[l])
                l += 1
            if r & 1:
                r -= 1
                smr = self.op(self.data[r], smr)
            l >>= 1
            r >>= 1
        return self.op(sml, smr)

    def all_prod(self):
        """ op(data[0], data[1], ... data[N-1])を返す """
        return self.data[1]

    def max_right(self, l, func):
        """
        func(l, l+1, ..., r-1) = True,
        func(l, l+1, ..., r-1, r) = Falseとなる r を返す
        """
        if l == self._n:
            return self._n
        l += self.size
        sm = self.e
        while True:
            while l % 2 == 0:
                l >>= 1
            if not func(self.op(sm, self.data[l])):
                while l < self.size:
                    l <<= 1
                    if func(self.op(sm, self.data[l])):
                        sm = self.op(sm, self.data[l])
                        l += 1
                return l - self.size
            sm = self.op(sm, self.data[l])
            l += 1
            if (l & -l) == l:
                break
        return self._n

    def min_left(self, r, func):
        """
        func(     l, l+1, ..., r-1) = True,
        func(l-1, l, l+1, ..., r-1) = Falseとなる l を返す
        """
        if r == 0:
            return 0
        r += self.size
        sm = self.e
        while True:
            r -= 1
            while r > 1 and r & 1:
                r >>= 1
            if not func(self.op(self.data[r], sm)):
                while r < self.size:
                    r = r << 1 | 1
                    if func(self.op(self.data[r], sm)):
                        sm = self.op(self.data[r], sm)
                        r -= 1
                return r + 1 - self.size
            sm = self.op(self.data[r], sm)
            if (r & -r) == r:
                break
        return 0


class LazySegTree(SegTree):
    def __init__(self, N, op_data, u_data, op_lazy, u_lazy, op_merge):
        super().__init__(N, op_data, u_data)
        self.composition = op_lazy
        self.mapping = op_merge
        self.id = u_lazy

        self.lazy = [u_lazy] * self.size

    def _all_apply(self, i, F):
        # self.data[i] = self.mapping(F, self.data[i], self.len[i])
        self.data[i] = self.mapping(F, self.data[i])
        if i < self.size:
            self.lazy[i] = self.composition(F, self.lazy[i])

    def _push(self, i):
        self._all_apply(i << 1, self.lazy[i])
        self._all_apply(i << 1 | 1, self.lazy[i])
        self.lazy[i] = self.id

    def update(self, p, x):
        """ data[p] = x とする (0-indexed)"""
        p += self.size
        for i in reversed(range(1, self.log + 1)):
            self._push(p >> i)
        self.data[p] = x
        for i in range(1, self.log + 1):
            self._update(p >> i)

    def apply(self, p, F):
        """ data[p]にFを作用させる(data[p] = op_merge(F, data[p])とする, 0-indexed) """
        p += self.size
        for i in reversed(range(1, self.log + 1)):
            self._push(p >> i)
        # self.data[p] = self.mapping(F, self.data[p], self.len[p])
        self.data[p] = self.mapping(F, self.data[p])
        for i in range(1, self.log + 1):
            self._update(p >> i)

    def range_apply(self, l, r, F):
        """ i = l, l+1, ..., r-1 について、Fを作用させる(op_merge(F, data[i]), 0-indexed) """
        if l == r:
            return

        l += self.size
        r += self.size
        for i in reversed(range(1, self.log + 1)):  # too->down
            if ((l >> i) << i) != l:
                self._push(l >> i)
            if ((r >> i) << i) != r:
                self._push((r - 1) >> i)

        l2, r2 = l, r
        while l < r:
            if l & 1:
                self._all_apply(l, F)
                l += 1
            if r & 1:
                r -= 1
                self._all_apply(r, F)
            l >>= 1
            r >>= 1
        l, r = l2, r2

        for i in range(1, self.log + 1):
            if ((l >> i) << i) != l:
                self._update(l >> i)
            if ((r >> i) << i) != r:
                self._update((r - 1) >> i)

    def get(self, p):
        """ data[p]を返す """
        p += self.size
        for i in reversed(range(1, self.log + 1)):
            self._push(p >> i)
        return self.data[p]

    def prod(self, l, r):
        """
        op_data(data[l], data[l+1], ..., data[r-1])を返す (0-indexed)
        l == rの時は単位元u_dataを返す
        """
        if l == r:
            return self.e

        l += self.size
        r += self.size
        for i in reversed(range(1, self.log + 1)):
            if ((l >> i) << i) != l:
                self._push(l >> i)
            if ((r >> i) << i) != r:
                self._push(r >> i)

        sml = self.e
        smr = self.e
        while l < r:
            if l & 1:
                sml = self.op(sml, self.data[l])
                l += 1
            if r & 1:
                r -= 1
                smr = self.op(self.data[r], smr)
            l >>= 1
            r >>= 1
        return self.op(sml, smr)

    def max_right(self, l, func):
        """
        func(l, l+1, ..., r-1) = True,
        func(l, l+1, ..., r-1, r) = Falseとなる r を返す
        """
        if l == self._n:
            return self._n
        l += self.size
        for i in reversed(range(1, self.log + 1)):
            self._push(l >> i)

        sm = self.e
        while True:
            while l % 2 == 0:
                l >>= 1
            if not func(self.op(sm, self.data[[l]])):
                while l < self.size:
                    self._push(l)
                    l <<= 1
                    if func(self.op(sm, self.data[l])):
                        sm = self.op(sm, self.data[l])
                        l += 1
                return l - self.size
            sm = self.op(sm, self.data[l])
            l += 1
            if (l & -l) == l:
                break
        return self._n

    def min_left(self, r, func):
        """
        func(     l, l+1, ..., r-1) = True,
        func(l-1, l, l+1, ..., r-1) = Falseとなる l を返す
        """
        if r == 0:
            return 0
        r += self.size
        for i in reversed(range(1, self.log + 1)):
            self._push((r - 1) >> i)
        sm = self.e
        while True:
            r -= 1
            while r > 1 and r & 1:
                r >>= 1
            if not func(self.op(self.data[r], sm)):
                while r < self.size:
                    self._push(r)
                    r = r << 1 | 1
                    if func(self.op(self.data[r], sm)):
                        sm = self.op(self.data[r], sm)
                        r -= 1
                return r + 1 - self.size
            sm = self.op(self.data[r], sm)
            if (r & -r) == r:
                break
        return 0
    """
    遅延セグ木(ac-library移植)
    op_data(d_L, d_R) : d_Lとd_Rの二項演算, dataを返す
    op_lazy(lz_new, lz_orig) : lz_origにlz_newを作用させる, lazyを返す
    op_merge(lz, d) : dにlzを作用させる, dataを返す
    """

import heapq

n = int(input())
cd = [[int(i) for i in input().split()] for j in range(n)]
st = SortedSet()
st.add(0)
st.add(n)

wdt = n
dp = []
konp = []
ans = 0

for i in range(n):
    dp.append(cd[i][0])
    if i != 0:
        konp.append([-cd[i][1],i])
    ans += cd[i][0]+cd[i][1]

ans -= cd[0][1]

heapq.heapify(konp)

bv = 10**10

u_data = 0
op_data = max

segmax = SegTree(wdt, op_data, u_data)
segmax.initialize(dp)

u_data = bv
op_data = min

segmin = SegTree(wdt, op_data, u_data)
segmin.initialize(dp)

tmp = segmin.prod(0,n)
idx = segmin.max_right(0,lambda x: x > tmp)
ans += tmp
segmin.update(idx,bv)
segmax.update(idx,bv)

for i in range(n-1):
    kp,i = heapq.heappop(konp)
    kp *= -1
    bigidx = st.gt(i)
    smlidx = st.lt(i)
    if bv != segmax.prod(i,bigidx):
        tmp = segmin.prod(i,bigidx)
        segidx = segmin.max_right(i,lambda x: x > tmp)
    else:
        tmp = segmin.prod(smlidx,i)
        segidx = segmin.max_right(smlidx,lambda x: x > tmp)
    if tmp < kp:
        st.add(i)
        segmax.update(segidx,bv)
        segmin.update(segidx,bv)
        ans -= (kp-tmp)
    else:
        break

print(ans)