Practical Hilbert space approximate Bayesian Gaussian processes for probabilistic programming

No worries. I’m fairly sure it just has to do with the convention of the Fourier transform used. Looking at the \nu=\infty case using the Rasmussen and Williams convention:
S(\omega) = \int_Re^{r^2/2l^2}e^{-2\pi i\omega r}dr=\sqrt{2\pi}le^{-2\pi^2l^2\omega^2} which matches the formula given in the textbook (pg.83)
Using the Solin and Särrkä convention:
S(\omega) = \int_Re^{r^2/2l^2}e^{-i\omega r}dr=\sqrt{2\pi}le^{-1/2l^2\omega^2} which matches the formula in the Riutort et al. paper.

So it seems like for the general case which I wasn’t able to verify the integration for one substitutes \omega = \omega/2\pi in the Rasmussen and Williams equation to get equations (1) - (3).

2 Likes