1: \begin{abstract}

2: Singularities of a statistical model are the elements of the model's parameter space 

3: which make the corresponding Fisher information matrix degenerate.

4: These are the points for which estimation techniques

5: such as the maximum likelihood estimator and standard Bayesian procedures

6: do not admit the root-$n$ parametric rate of convergence. 

7: We propose a general framework for the identification of singularity structures 

8: of the parameter space of finite mixtures, and study the impacts of the 

9: singularity structures on minimax lower bounds and rates of convergence for the 

10: maximum likelihood estimator over a compact parameter space. 

11: Our study makes explicit the deep links between model singularities, parameter

12: estimation convergence rates and minimax lower bounds, and the algebraic geometry 

13: of the parameter space for mixtures of continuous distributions. 

14: The theory is applied to establish concrete convergence rates of parameter 

15: estimation for finite mixture of skew-normal distributions. This

16: rich and increasingly popular mixture model is shown to exhibit a 

17: remarkably complex range of asymptotic behaviors which have not

18: been hitherto reported in the literature.

19: %

20: \footnote{This research is supported in part by grants

21: NSF CAREER DMS-1351362 and NSF CNS-1409303 to XN.

22: The authors would like to acknowledge Larry Wasserman and Ya'acov Ritov for valuable 

23: discussions related to this work.

24: 

25: AMS 2000 subject classification: Primary 62F15, 62G05; secondary 62G20.

26: 

27: Keywords and phrases: Fisher singularities, system of polynomial 

28: equations, semialgebraic set, mixture model, 

29: non-linear partial differential equation, minimax lower bound, 

30: maximum likelihood estimation, convergence rates, Wasserstein distances.}

31: \end{abstract}

32: