cf I Photo Processing （二分答案+dp）

I. Photo Processing

time limit per test

3 seconds

memory limit per test

256 megabytes

input

standard input

output

standard output

Evlampiy has found one more cool application to process photos. However the application has certain limitations.

Each photo i has a contrast v_i. In order for the processing to be truly of high quality, the application must receive at least k photos with contrasts which differ as little as possible.

Evlampiy already knows the contrast v_i for each of his n photos. Now he wants to split the photos into groups, so that each group contains at least k photos. As a result, each photo must belong to exactly one group.

He considers a processing time of the j-th group to be the difference between the maximum and minimum values of v_i in the group. Because of multithreading the processing time of a division into groups is the maximum processing time among all groups.

Split n photos into groups in a such way that the processing time of the division is the minimum possible, i.e. that the the maximum processing time over all groups as least as possible.

Input

The first line contains two integers n and k (1 ≤ k ≤ n ≤ 3·10⁵) — number of photos and minimum size of a group.

The second line contains n integers v₁, v₂, ..., v_n (1 ≤ v_i ≤ 10⁹), where v_i is the contrast of the i-th photo.

Output

Print the minimal processing time of the division into groups.

Examples

Input

Copy

5 2
50 110 130 40 120

Output

20

Input

Copy

4 1
2 3 4 1

Output

0

Note

In the first example the photos should be split into 2 groups: [40, 50] and [110, 120, 130]. The processing time of the first group is 10, and the processing time of the second group is 20. Maximum among 10 and 20 is 20. It is impossible to split the photos into groups in a such way that the processing time of division is less than 20.

In the second example the photos should be split into four groups, each containing one photo. So the minimal possible processing time of a division is 0.

题意:

给你一个n个数字的数列，让你分组，每组最少m个，使得最大价值最小(每组的价值就是组里面最大数-最小数)

解析:

logn优化:二分答案

最后配合O(n)的dp

这里dp网上有好多版本,最牛是无常数的O(n)，然后就是可能带点常数的O(n)

#include <cstdio>
#include <cstring>
#include <algorithm>
using namespace std;
const int MAXN = 3e5+10;
#define INF 0x3f3f3f3f3f3f3f3f

typedef long long int lli;

lli aa[MAXN];
int n,m;

int dp[MAXN];



/*bool check(lli x)
{
    int last=0;  //
    for(int i=m;i<=n;i++)
    {
        int tmp=dp[i-m];  //[1,d[i-m]]是已经切好了,选择i-m的原因是只有这里一定能保证切的区间没有交集
        if(aa[i]-aa[tmp+1]<=x) last=i;
        dp[i]=last;  //将d[i]赋值成包括i的那一块的开头
    }
    return dp[n]==n;
}*/


bool check(lli x)
{
    memset(dp,0,sizeof(dp));  //dp[i]表示[1,i]能否被完整分成满足条件的几段
    dp[0]=1;
    int s=1;
    for(int i=1;i<=n;i++)
    {
        while(aa[i]-aa[s]>x) s++;
        while(i-s+1>=m)
        {
            if(dp[s-1])
            {
                dp[i]=1;
                break;
            }
            s++;
        }
    }
    return dp[n];
}

int main()
{
    scanf("%d%d",&n,&m);
    for(int i=1;i<=n;i++)
    {
        scanf("%lld",&aa[i]);
    }
    aa[0]=INF;
    sort(aa+1,aa+1+n);
    lli l=0;
    lli r=aa[n]-aa[1];
    if(m==1)
    {
        printf("0\n");
        return 0;
    }
    while(l<r)
    {
        lli mid=(l+r)>>1;
        if(check(mid)) r=mid;
        else l=mid+1;
    }
    printf("%lld\n",r);
    return 0;
}