Tommy has just invented an interesting string encoding algorithm, which is described below.

• For every string , we may define a character-mapping function , which maps every character occurring in to a lowercase letter, as below

where is the -th lowercase Latin letter, and is the number of different characters after the last occurrence of in .

• To encode a string by Tommy's algorithm, replace every character in by simultaneously.

For example, the encoded string of "abc" is "cba", and the encoded string of "cac" is "aba".

You are given a string of length and then encode all the nonempty prefixes. Your task is to find the encoded string that has the greatest lexicographical order among all the encoded strings.

The first line contains an integer  .

The second line contains a string of length , which consists of only the first  lowercase letters, 'a' to 't'.

Output the encoded string that has the greatest lexicographical order among all the encoded strings.

4
aacc

bbaa

In the first sample case, the four nonempty prefixes "a", "aa", "aac" and "aacc" will be encoded to "a", "aa", "bba" and "bbaa" respectively, where "bbaa" has the greatest lexicographical order.

In the second sample case, the three nonempty prefixes "a", "ac" and "aca" will be encoded to "a", "ba" and "aba" respectively, where "ba" has the greatest lexicographical order.

3
aca

ba