Adaptive Dynamic Bitvectors

While operations \emph{rank} and \emph{select} on static bitvectors can be supported in constant time, lower bounds show that supporting updates raises the cost per operation to $\Theta(\log n/ \log\log n)$. This is a shame in scenarios where updates are possible but uncommon. We develop a representation of bitvectors that supports all the operations in $O(\log(n/q))$ amortized time. Our experimental results support the theoretical findings, displaying speedups of orders of magnitude compared to standard dynamic implementations.