Fully-dynamic-to-incremental reductions with known deletion order (e.g. sliding window)

TL;DR

This paper presents a general reduction from fully dynamic algorithms to incremental algorithms under known deletion order, enabling more practical and theoretical solutions with bounded overheads.

Contribution

It introduces a black-box reduction from fully dynamic to incremental algorithms applicable when deletion order is known in advance, with guarantees on worst-case running time.

Findings

Achieves a simple amortized-fully-dynamic to worst-case-incremental reduction with logarithmic overhead.

Provides a theoretical worst-case-fully-dynamic to worst-case-incremental reduction with polylogarithmic overhead.

Applicable to scenarios like sliding window models where deletion order is predetermined.

Abstract

Dynamic algorithms come in three main flavors: $\mathit{incremental}$ (insertions-only), $\mathit{decremental}$ (deletions-only), or $\mathit{fully}$ $\mathit{dynamic}$ (both insertions and deletions). Fully dynamic is the holy grail of dynamic algorithm design; it is obviously more general than the other two, but is it strictly harder? Several works managed to reduce fully dynamic to the incremental or decremental models by taking advantage of either specific structure of the incremental/decremental algorithms (e.g. [HK99, HLT01, BKS12, ADKKP16, BS80, OL81, OvL81]), or specific order of insertions/deletions (e.g. [AW14,HKNS15,KPP16]). Our goal in this work is to get a black-box fully-to-incremental reduction that is as general as possible. We find that the following conditions are necessary: $\bullet$ The incremental algorithm must have a worst-case (rather than amortized) running time guarantee. $\bullet$ The reduction must work in what we call the $\mathit{deletions}$-$\mathit{look}$-$\mathit{ahead}$ $\mathit{model}$, where the order of deletions among current elements is known in advance. A notable practical example is the "sliding window" (FIFO) order of updates. Under those conditions, we design: $\bullet$ A simple, practical, amortized-fully-dynamic to worst-case-incremental reduction with a $\log(T)$-factor overhead on the running time, where $T$ is the total number of updates. $\bullet$ A theoretical worst-case-fully-dynamic to worst-case-incremental reduction with a $\mathsf{polylog}(T)$-factor overhead on the running time.