1: \begin{abstract}

2: In this paper, we exploit the theory of dense graph limits to provide a

3: new framework to study the stability of graph partitioning methods, which

4: we call {\em structural consistency}. Both stability under perturbation

5: as well as asymptotic consistency (i.e., convergence with probability $1$

6: as the sample size goes to infinity under a fixed probability model)

7: follow from our notion of structural consistency. By formulating

8: structural consistency as a continuity result on the graphon space, we

9: obtain robust results that are completely independent of the data

10: generating mechanism.

11: In particular, our results apply in settings where observations are not

12: independent, thereby significantly generalizing the common probabilistic

13: approach where data are assumed to be i.i.d.

14: 

15: In order to make precise the notion of structural consistency of graph

16: partitioning, we begin by extending the theory of graph limits to include

17: vertex colored graphons. We then define \textit{continuous node-level

18: statistics} and prove that graph partitioning based on such statistics is

19: consistent. Finally, we derive the structural consistency of commonly

20: used clustering algorithms in a general model-free setting. These include

21: clustering based on local graph statistics such as homomorphism

22: densities, as well as the popular spectral clustering using the

23: normalized Laplacian.

24: 

25: We posit that proving the continuity of clustering algorithms in the

26: graph limit topology can stand on its own as a more robust form of

27: model-free consistency. We also believe that the mathematical framework

28: developed in this paper goes beyond the study of clustering algorithms,

29: and will guide the development of similar model-free frameworks

30: to analyze other procedures in the broader mathematical sciences.

31: \end{abstract}

32: