24
0

Fast Machine-Precision Spectral Likelihoods for Stationary Time Series

Abstract

We provide in this work an algorithm for approximating a very broad class of symmetric Toeplitz matrices to machine precision in O(nlogn)\mathcal{O}(n \log n) time with applications to fitting time series models. In particular, for a symmetric Toeplitz matrix Σ\mathbf{\Sigma} with values Σj,k=hjk=1/21/2e2πijkωS(ω)dω\mathbf{\Sigma}_{j,k} = h_{|j-k|} = \int_{-1/2}^{1/2} e^{2 \pi i |j-k| \omega} S(\omega) \mathrm{d} \omega where S(ω)S(\omega) is piecewise smooth, we give an approximation FΣFHD+UVH\mathbf{\mathcal{F}} \mathbf{\Sigma} \mathbf{\mathcal{F}}^H \approx \mathbf{D} + \mathbf{U} \mathbf{V}^H, where F\mathbf{\mathcal{F}} is the DFT matrix, D\mathbf{D} is diagonal, and the matrices U\mathbf{U} and V\mathbf{V} are in Cn×r\mathbb{C}^{n \times r} with rnr \ll n. Studying these matrices in the context of time series, we offer a theoretical explanation of this structure and connect it to existing spectral-domain approximation frameworks. We then give a complete discussion of the numerical method for assembling the approximation and demonstrate its efficiency for improving Whittle-type likelihood approximations, including dramatic examples where a correction of rank r=2r = 2 to the standard Whittle approximation increases the accuracy from 33 to 1414 digits for a matrix ΣR105×105\mathbf{\Sigma} \in \mathbb{R}^{10^5 \times 10^5}. The method and analysis of this work applies well beyond time series analysis, providing an algorithm for extremely accurate direct solves with a wide variety of symmetric Toeplitz matrices. The analysis employed here largely depends on asymptotic expansions of oscillatory integrals, and also provides a new perspective on when existing spectral-domain approximation methods for Gaussian log-likelihoods can be particularly problematic.

View on arXiv
Comments on this paper