1: \begin{abstract}

2: 

3: 

4: Fourier series of smooth, non-periodic functions on $[-1,1]$ are known to

5: exhibit the Gibbs phenomenon, and exhibit overall slow convergence.

6: One way of overcoming these problems is by using a Fourier series on a larger

7: domain, say $[-T,T]$ with $T>1$, a technique called Fourier extension or Fourier

8: continuation. When constructed as the discrete least squares minimizer in

9: equidistant points, the Fourier extension has been shown shown to converge

10: geometrically in the truncation parameter $N$. A fast ${\mathcal O}(N \log^2 N)$

11: algorithm has been described to compute Fourier extensions for the case where

12: $T=2$, compared to ${\mathcal O}(N^3)$ for solving the dense discrete least

13: squares problem. We present two ${\mathcal O}(N\log^2 N )$ algorithms for the

14: computation of these approximations for the case of general $T$, made possible

15: by exploiting the connection between Fourier extensions and Prolate Spheroidal

16: Wave theory. The first algorithm is based on the explicit computation of

17: so-called periodic discrete prolate spheroidal sequences, while the second

18: algorithm is purely algebraic and only implicitly based on the theory. 

19: % An interesting consequence is that the second algorithm can also be used to reduce the computational complexity of the computation of Fourier extensions on higher-dimensional domains of any shape, even if no analogue of the prolate sequences is known for the domain at hand.

20: \end{abstract}

21: