1: \begin{abstract}

2: A popular approach for modeling and inference in spatial statistics is to represent Gaussian random fields as solutions to stochastic partial differential equations (SPDEs) of the form $L^{\beta}u = \mathcal{W}$, 

3: where $\mathcal{W}$ is Gaussian white noise, 

4: $L$ is a second-order differential operator, 

5: and $\beta>0$ is a parameter that determines the smoothness of $u$.

6: However, this approach has been limited 

7: to the case $2\beta\in\mathbb{N}$,

8: which excludes several important models 

9: and makes it necessary to keep $\beta$ fixed during inference.

10: 

11: We propose a new method, the rational SPDE approach, which in spatial dimension $d\in\mathbb{N}$ is applicable for any $\beta>d/4$, and

12: thus remedies the mentioned limitation.

13: The presented scheme combines a 

14: finite element discretization 

15: with a rational approximation of the function $x^{-\beta}$ 

16: to approximate~$u$.

17: For the resulting approximation,

18: an explicit rate of convergence 

19: to $u$ in mean-square sense is derived. 

20: Furthermore, we show that our method has

21: the same computational benefits

22: as in the restricted case $2\beta\in\mathbb{N}$.

23: Several numerical experiments and a statistical application 

24: are used to illustrate the accuracy of the method,

25: and to show that it facilitates likelihood-based inference for all model parameters including $\beta$.

26: \end{abstract}

27: