1: \begin{abstract}

2:   The fastest deterministic algorithms for connected components take logarithmic

3:   time and perform superlinear work on a Parallel Random Access Machine

4:   (PRAM). These algorithms maintain a spanning forest by merging and compressing

5:   trees, which requires pointer-chasing operations that increase memory access

6:   latency and are limited to shared-memory systems. Many of these PRAM

7:   algorithms are also very complicated to implement. Another popular method is

8:   ``leader-contraction'' where the challenge is to select a constant fraction of

9:   leaders that are adjacent to a constant fraction of non-leaders with high

10:   probability. Instead we investigate label propagation because it is

11:   deterministic and does not rely on pointer-chasing. Label propagation

12:   exchanges representative labels within a component using simple graph

13:   traversal, but it is inherently difficult to complete in a sublinear number of

14:   steps. We are able to solve the problems with label propagation for graph

15:   connectivity.

16: 

17:   We introduce a simple framework for deterministic graph connectivity using

18:   label propagation that is easily adaptable to many computational models. It

19:   propagates directed edges and alternates edge direction to achieve linear edge

20:   count each step and sublinear convergence. We present new algorithms in PRAM,

21:   Stream, and MapReduce for a simple, undirected graph $G=(V,E)$ with $n=|V|$

22:   vertices, $m=|E|$ edges. Our approach takes $O(m)$ work each step, but we can

23:   only prove logarithmic convergence on a path graph. It was conjectured by Liu

24:   and Tarjan (2019) to take $O(\log n)$ steps or possibly $O(\log^2 n)$

25:   steps. We leave the proof of convergence as an open problem.

26: 

27:   \textit{Keywords:} graph connectivity, connected components, parallel

28:   algorithm, pram, stream, mapreduce

29: \end{abstract}