ARC081E. Don't Be a Subsequence
This observation leads us to a solution. Let \(\dp[i]\) be the maximum number of the aforementioned segments that the suffix of \(S\) that starts at index \(i\) can be divided into. The DP can be done in \(O(|S|)\) time. The shortest string that is not a subsequence of \(S\) has a length of \(M = \dp[0] + 1\) (\(S\) is 0-indexed).
Let \(\next[i][j]\) be the position of the first occurrence of letter \(j\) to the right of position \(i\) (including position \(i\)). We can compute the \(\next\) array in \(O(26|S|)\) time.
Using the \(\next\) and \(\dp\) arrays, we can construct the answer as follows:
Start with an empty string \(T\). Iterate the \(\dp[0] + 1\) positions of the answer string from left to right. For each position \(i\), iterate over the letters from 'a' to 'z'. For each letter \(j\), check whether it is possible to get an answer if we append \(j\) to \(T\). Let \(k\) be position of the last letter of the first occurrence of \(Tj\) in \(S\) as a subsequence, it is ok to append letter \(j\) to \(T\) if the suffix \(S[k + 1, |S|)\) does not contain all subsequences of length \(M - |T| - 1\) i.e. \(\dp[k + 1] < M - |T| - 1\). This check can be done efficiently, see the following code for detail.
code
int main() { string s; scan(s); int n = SZ(s); vb vis(26); int cnt = 0; vi dp(n + 1); int length = 0; down (i, n - 1, 0) { if (!vis[s[i] - 'a']) { vis[s[i] - 'a'] = true; ++cnt; if (cnt == 26) { ++length; fill(all(vis), false); cnt = 0; } } dp[i] = length; }
vv
next(n, vi(26));
fill(all(next.back()), n);
next.back()[s.back() - 'a'] = n - 1;
down (i, n - 2, 0) {
rng(j, 0, 26) {
next[i][j] = s[i] - 'a' == j ? i : next[i + 1][j];
}
}++length;
int pos = 0;
while (length > 0) {
rng (j, 0, 26) {
int t = next[pos][j];
if (t < n && dp[t + 1] == length - 1) continue;
if (t < n) {
pos = t + 1;
}
cout << char('a' + j);
break;
}
--length;
}
cout << '\n';
return 0;
}