In nonparametric statistics, the following space is often used $$H_{per}^\beta := \left\{f:[0,1]\to\mathbb{R}:\,D^{\beta-1}f\,\text{absolutely continuous and } D^\beta f\in L^2[0,1], \\D^{k}f(0) = D^{k}f(1),\quad \forall k\leq\beta-1 \right\}. $$ I think it is the classical Sobolev space (on a compact domain) when $\beta$ is nonnegative integers. Furthermore, since the space is a subspace of $L^2[0,1]$, we can find Fourier series for each element in $H_{per}^\beta$ and recharaterize the space as follows, $$ \left\{ f:\,f=\sum_{k\in\mathbb{Z}}c_k e^{-ik\pi x},\, \overline{c_k} = c_{-k},\, \sum_{k\in\mathbb{Z}}(k)^{2\beta}c_k^2 <\infty \right\}. $$ The space is extended by allowing $\beta$ to take nonnegative real value.

On the other hand, fractional Sobolev space is defined (in $\mathbb{R}$) by $$W_2^{\beta}(\mathbb{R}):=\left\{ f\in \mathcal{S}': \, \left((1+|\xi|^2)^{\beta/2}\hat{f}\right)^{\vee}\in L^2 \right\}.$$

I would like to know how these two types of spaces relate to each other if we restrict the second space to periodic distributions. If they do hold a relation, can it be extended to higher dimension, $\mathbb{R}^d$?