English

Quicksort, Largest Bucket, and Min-Wise Hashing with Limited Independence

Data Structures and Algorithms 2015-02-23 v1

Abstract

Randomized algorithms and data structures are often analyzed under the assumption of access to a perfect source of randomness. The most fundamental metric used to measure how "random" a hash function or a random number generator is, is its independence: a sequence of random variables is said to be kk-independent if every variable is uniform and every size kk subset is independent. In this paper we consider three classic algorithms under limited independence. We provide new bounds for randomized quicksort, min-wise hashing and largest bucket size under limited independence. Our results can be summarized as follows. -Randomized quicksort. When pivot elements are computed using a 55-independent hash function, Karloff and Raghavan, J.ACM'93 showed O(nlogn)O ( n \log n) expected worst-case running time for a special version of quicksort. We improve upon this, showing that the same running time is achieved with only 44-independence. -Min-wise hashing. For a set AA, consider the probability of a particular element being mapped to the smallest hash value. It is known that 55-independence implies the optimal probability O(1/n)O (1 /n). Broder et al., STOC'98 showed that 22-independence implies it is O(1/A)O(1 / \sqrt{|A|}). We show a matching lower bound as well as new tight bounds for 33- and 44-independent hash functions. -Largest bucket. We consider the case where nn balls are distributed to nn buckets using a kk-independent hash function and analyze the largest bucket size. Alon et. al, STOC'97 showed that there exists a 22-independent hash function implying a bucket of size Ω(n1/2)\Omega ( n^{1/2}). We generalize the bound, providing a kk-independent family of functions that imply size Ω(n1/k)\Omega ( n^{1/k}).

Keywords

Cite

@article{arxiv.1502.05729,
  title  = {Quicksort, Largest Bucket, and Min-Wise Hashing with Limited Independence},
  author = {Mathias Bæk Tejs Knudsen and Morten Stöckel},
  journal= {arXiv preprint arXiv:1502.05729},
  year   = {2015}
}

Comments

Submitted to ICALP 2015