Proposition 1
8RSCThe expected variance in the Bayesian quadrature \epsilon^{2}_{{\textsc{bq}}{}} is the maximum mean discrepancy between the target distribution p and q_{{\textsc{bq}}{}}(x) = \sum_{n=1}^{N}w^{(n)}_{{\textsc{bq}}{}}\delta_{x_n}(x)
Proof of Proposition 1
The proof involves invoking the representer theorem, using bilinearity of scalar products and the fact that if f is a standard Gaussian process then \forall g\in\mathcal{H}: \left\langle f,g\right\rangle \sim \mathcal{N}(0,\left\|g\right\|_{\mathcal{H}}^2):
\begin{aligned}
&\mathbb{V}_{} \left[ {Z_{f,p}\vert f_{x_1}, \dots, f_{x_N}} \right]=\\
&= \mathbb{E}_{f\sim GP} \left( \int f(x) p(x) dx - \sum_{n=1}^{N}w^{(n)}_{{\textsc{bq}}{}} f(x_n)\right)^2\\
&= \mathbb{E}_{f\sim GP} \left( \int \left\langle f, \phi (x)\right\rangle p(x) dx - \sum_{n=1}^{N}w^{(n)}_{{\textsc{bq}}{}} \left\langle f, \phi (x_n)\right\rangle\right)^2\\
&= \mathbb{E}_{f\sim GP} \left\langle f , \int\phi(x) p(x) dx - \sum_{n=1}^{N}w^{(n)}_{{\textsc{bq}}{}}\phi(x_n)\right\rangle^2\\
&= \left\|\mu_p - \mu_{q_{{\textsc{bq}}{}}}\right\|_{\mathcal{H}}^2\\
&= {\rm MMD}^2(p,q_{{\textsc{bq}}{}})
\end{aligned}