1: \begin{abstract}

2: A variety of information processing tasks in practice involve recovering

3: $n$ objects from single-shot graph-based measurements, particularly

4: those taken over the edges of some measurement graph $\mathcal{G}$.

5: This paper concerns the situation where each object takes value over

6: a group of $M$ different values, and where one is interested to

7: recover all these values based on observations of certain pairwise

8: relations over $\mathcal{G}$. The imperfection of measurements presents

9: two major challenges for information recovery: 1) \emph{inaccuracy}:

10: a (dominant) portion $1-p$ of measurements are corrupted; 2) \emph{incompleteness}:

11: a significant fraction of pairs are unobservable, i.e. $\mathcal{G}$

12: can be highly sparse. 

13: 

14: Under a natural random outlier model, we characterize the {\em minimax recovery rate}, that is, the critical threshold of non-corruption rate $p$ below which exact information recovery is infeasible. This accommodates a very

15: general class of pairwise relations. For various homogeneous

16: random graph models (e.g. Erd\H{o}s\textendash{}Rényi random graphs,

17: random geometric graphs, small world graphs), the minimax recovery rate

18: depends almost exclusively on the edge sparsity of the measurement graph

19: $\mathcal{G}$ irrespective of other graphical metrics. This fundamental

20: limit decays with the group size $M$ at a square root rate before

21: entering a connectivity-limited regime. Under the Erd\H{o}s\textendash{}Rényi

22: random graph, a tractable combinatorial algorithm is proposed

23: to approach the limit for large $M$ ($M=n^{\Omega(1)}$), while order-optimal recovery is enabled by

24: semidefinite programs in the small $M$

25: regime.  

26: \end{abstract}

27: