This technique is based on the fact that every integer can be represented in binary form.
Through pre-processing, a sparse table dp[i][v] can be calculated which stores the 2^ith parent of vertex v, where 0 <= i <= logN. (Example: 1483. Kth Ancestor of a Tree Node (Hard)) This pre-processing takes O(NlogN) time.
When we want to get the kth parent of v, we can first break k into the binary representation k = 2^p1 + 2^p2 + 2^p3 + ... + 2^pj where p1, ..., pj are the indices where k has bit value 1.
So we can get the 2^p1th, 2^p2th, ... 2^pjth parent in any order to get kth parent of v.
Example: to get the 6th parent of v, since 6 = 2^1 + 2^2, the answer can be P[1][ P[2][v] ] or P[2][ P[1][v] ].
Hence we can reduce the time complexity of getting the Kth parent from O(K) to O(logK).