As a data-oblivious algorithm, our randomized Shellsort
method is a Monte Carlo algorithm (e.g., see [32,33]), in that
it always runs in the same amount of time but can sometimes
fail to sort. It can easily be converted into a data-dependent
Las Vegas algorithm, however, which always succeeds but
has a randomized running time, by testing if its output is
sorted and repeating the algorithm if it is not. Such a datadependent
version of randomized Shellsort would run in
O(n log n) time with very high probability; hence, it would
provide the first version of Shellsort that provably runs in
O(n log n) time with very high probability.