Codeforces Round #225 (Div. 1) B 模拟：

题目：http://codeforces.com/contest/383/problem/B

参考：http://www.xuebuyuan.com/2068138.html

384D - Вулканы

Our first observation is that if there is a path from (1, 1) to (N, N), then the length of path is 2 * N – 2. Since all paths have length 2 * N – 2, it follows that if there is at least one path, the answer is 2 * N – 2 and if there isn’t, the answer is -1. How to prove it? Every path from (1, 1) to (N, N) has exactly N – 1 down directions and exactly N – 1 right directions. So, total length for each path is N – 1 + N – 1 = 2 * N – 2.

So we reduced our problem to determine if there is at least one path from (1, 1) to (N, N). This is the challenging part of this task, considering that N <= 10 ^ 9. How would you do it for a decently small N, let’s say N <= 10^3 . One possible approach would be, for each row, keep a set of reachable columns. We could easily solve this one by doing this: if (i, j) denotes element from ith row and jth column, then (i, j) is (is not) reachable if:

if (i, j) contains a volcano, then (i, j) is not reachable. Otherwise, if at least one of (i – 1, j) and (i, j – 1) is reachable, then (i, j) is reachable. Otherwise, (i, j) is not reachable.

What’s the main problem of this approach? It needs to keep track of 10^9 lines and in worst case, each of those lines can have 10^9 reachable elements. So, worst case we need 10^9 * 10^9 = 10^18 operations and memory.

Can we optimize it? We can note for beginning that we don’t need to keep track of 10^9 lines, only m lines are really necessarily. We need only lines containing at least one obstacle (in worst case when each line contains only one obstacle, we need m lines). How to solve it this way? Suppose line number x contains some obstacles and lines x + 1, x + 2, x + 3 do not contain any obstacle. Suppose we calculated set S = {y | cell (x, y) is reachable}. How would look S1, S2, S3 corresponding to lines x + 1, x + 2, x + 3? For S1, we can reach cell (x + 1, ymin), where ymin is minimal value from set S. Then, we can also reach {ymin + 1, ymin + 2, …, N}, by moving right from (x + 1, ymin). So S1 = {ymin, ymin + 1, …, N}. How do S2 and S3 look? It’s easy to see that they’ll be as well {ymin, ymin + 1, …, N}. So we get following optimization: suppose set of lines containing at least one obstacle is {L1, L2, …, Lk}. We need to run algorithm only for lines L1, L1 + 1, L2, L2 + 1, L3, L3 + 1, …, Lk, Lk + 1.

It looks like we didn’t make anything with this optimization. Even if we calculate for m lines, each line can still have 10^9 reachable positions. So worst case we perform 10^14 operations. We need something better for managing information from a line. You can note that for a given line y, there are a lot of positions having consecutive values. There are a lot of positions (x, y) and (x, y + 1) both reachable. This should give us following idea: what if instead of keeping reachable positions, we keep reachable ranges? That is, for each line x we keep a set of ranges S = {(a, b) | all cells (x, k) with a <= k <= b are reachable}.

How many ranges can it be for a line? If the line contains m obstacles, there are m + 1 ranges. Suppose for line x all cells are reachable, but for line x + 1 cells (x + 1, 3) (x + 1, 5) (x + 1, N – 1) are blocked. Then, the ranges of reachable cells are [1, 2] [4, 4], [6, N – 2] and [N, N]. By now, we get worst case m lines and worst case each line having m elements, so in worst case we’d have to handle m * m = 10 ^ 10 events. This may still look too much, but happily this bound is over estimated. If a line has o obstacles, there can be at most o + 1 ranges. If lines L1, L2, …, Lk have {o1, o2, …, ok} obstacles, there’ll be at most o1 + o2 + … + ok + k ranges. But o1 + o2 + … + ok = m and also k is at most m (proved above why we’re interested in at most m lines), so in worst case we get m + m = 2 * m ranges. Yaay, finally a decent number of states for this problem :)

So, we iterate each line we’re interested in. Let’s find set of ranges for this line, thinking that all cells from line above are reachable. This is easy to do. After we get our ranges like all cells from above can be visited, let’s think how having obstacles above can influence current ranges. After adding ranges from above, current ranges can’t increase (obviously), they can only decrease, remain the same or some of them can become empty. So, let’s take each range [a, b] from current line and see how it will transform after adding ranges from previous line.

Given range [a, b], it can transform only in [a’ , b] with a’ >= a. If a’ > b, then obviously range is empty. Why second number of range keeps constant? Let a’ smallest reachable column from current line which is in range [a, b]. It’s enough to check a’ >= a, as if a’ > b, range will be empty. It’s obviously why we need to keep a’ smallest value possible >= a: we’re interested to keep range as big as possible and as less as we cut from left, as big it is. Once we’ve found a’ in range [a, b] (or a’ > b if range is empty) all cells {a’ + 1, a’ + 2, …, b} are reachable as well by going right from a’, so if interval is not empty, then second number defining it remains b.

Next question is how to find a’ fast enough. In order a point a’ to be reachable on current range, it also needs to exist a range on previous line containing it. If the range from previous line is [pa, pb] then a’ needs to follow 3 conditions:

a’ minimal such as

pa <= a’ <= pb

a’ >= a

What if instead of finding a’ we find [pa, pb]? Then a’ is max(pa, a). In order a’ to be as small as possible, since a is constant, pa needs to be as small as possible. So we reduced it to:

pa minimal pb >= a’ >= a <=> pb >= a

Intervals from previous line are disjoint, no 2 intervals cross each other. It means that if pb is minimal, than pa is minimal too (if we increase pb, then pa will increase too, so it won’t be minimal). Hence, you need to find an interval [pa, pb] such as pb is minimal and pb >= a. Then, a’ is max(a, pa). This is easy to do if we sort all intervals from previous line increasing by second value (pb), then we binary search for value a.

Finally, after running algorithm for all lines, last range from last line has second number N (assuming ranges are sorted increasing by second value), then there exist a path, otherwise there does not exist. This algorithm should run O(m * logm) worst case, good enough to pass.

Code: http://pastie.org/8651817

#include<cstdio>
#include<cstdlib>
#include<cstring>
#include<algorithm>
#include<iostream>
#include<fstream>
#include<map>
#include<ctime>
#include<set>
#include<queue>
#include<cmath>
#include<vector>
#include<bitset>
#include<functional>
#define x first
#define y second
#define mp make_pair
#define pb push_back
#define REP(i,l,r) for((i)=(l);(i)<=(r);++(i))
#define REP2(i,l,r) for((i)=(l);(i)!=(r);++(i))
using namespace std;

typedef long long LL;
typedef double ld;

const int MAX=100000+10;

int n,m;
pair<int,int> d[MAX];

int main()
{
#ifndef ONLINE_JUDGE
    freopen("input.txt","r",stdin);freopen("output.txt","w",stdout);
#endif
    int i;
    scanf("%d%d",&n,&m);
    REP(i,1,m)
        scanf("%d%d",&d[i].x,&d[i].y);
    sort(d+1,d+m+1);
    vector< pair<int,int> > last;
    last.pb( mp(1,1) );
    int lx=0;
    for(i=1;i<=m;)
    {
        int x=d[i].x;
        if(x!=lx+1)
        {
            int u=last.begin()->x;
            last.clear();
            last.pb(mp(u,n));
        }
        lx=x;
        vector<int> now;
        now.pb(0);
        now.pb(n+1);
        for(;i<=m && d[i].x==x;++i)
            now.pb(d[i].y);
        sort(now.begin(),now.end());
        int ll=0;
        int j;
        vector< pair<int,int> > ne;
        for(j=0;j+1<(int)now.size();++j)
        {
            int l=now[j]+1;
            int r=now[j+1]-1;
            if(l>r)
                continue;
            for(;ll<(int)last.size() && last[ll].y<l;++ll)
                ;
            if(ll==(int)last.size() || last[ll].x>r)
                continue;
            ne.pb( mp(max( last[ll].x, l) , r) );
        }
        last=ne;
        if(!ne.size() || (x==n && (--ne.end())->y!=n))
        {
            printf("-1\n");
            return 0;
        }
    }
    printf("%d\n",n*2-2);
    return 0;
}