I'm reading some papers on spin-ice models and in a few of them, they state the correlation function of the polarization in momentum space (in 3D) as being
$$\langle P_i(\mathbf{k}) P_j(\mathbf{-k})\rangle = \frac{1}{\kappa}\left(\delta_{ij} - \frac{k_i k_j}{|\mathbf{k}|^2} \right)$$
where $\kappa$ is some constant. It is then stated that Fourier transforming this back to real space gives
$$\langle P_i(0)P_j(\mathbf{r})\rangle = \frac{4\pi}{\kappa}\left(\delta^3(\mathbf{r}) + \frac{1}{r^3}\left(\delta_{ij} - \frac{3 x_i x_j}{|\mathbf{x}|^2} \right) \right).$$
I have two questions -
1) Why is the Fourier transform of $<P_i(k)P_j(-k)> = <P_i(0) P_j(r)>$? I get that the system is translationally invariant and so all correlations will only depend on the difference $x-y$ and that we can set $y = 0$, but I can't explicitly relate these using a Fourier transform.
2) Similarly, where did the $\delta$ function come from after the Fourier transform to real space? I can't reproduce this result.