1: \begin{proof}[Proof of Lemma \ref{lem:bound_PDE_noisy_a}] 

2: %

3: %\textcolor{red}{Let us first show that $\rho_t$ is sub-Gaussian on the variable $a$.

4: %\paragraph*{Step 1:}

5: %We first recall that $\rho_t$ can be seen as the gradient flow of the following functional over the space of distributions, also known as free energy:

6: %\[

7: %F_{\Gamma, \lambda} (\rho) = \frac{1}{2} \cdot \lb R (\rho ) + \lambda M(\rho ) \rb - \frac{1}{D  \Gamma } \cdot S(\rho),

8: %\]

9: %where we denoted:

10: %\[

11: %M(\rho) \equiv \int_{\R^{D+1}} \norm{\btheta}^2_2 \rho (\btheta) \de \btheta, \qquad S ( \rho ) \equiv - \int_{\R^{D+1}} \rho (\btheta) \log \rho (\btheta) \de \btheta

12: %\]

13: %the second moment and the entropy of the distribution $\rho$ respectively. In particular, $F_{\Gamma, \lambda} (\rho_t) $ is non-increasing along the trajectory, hence:

14: %\[

15: %R ( \rho_t ) \leq R(\rho_0) + \lambda M(\rho_0) + \frac{2}{D  \Gamma } S (\rho_t).

16: %\]

17: %Moreover, we recall the result of lemma 10.1 of PNAS paper: for any distribution $\rho$ with finite second moment,

18: %\[

19: %S ( \rho) \leq 2 + (D+1) \cdot \log (2 \pi M(\rho) )

20: %\]

21: %Let us derive a crude bound on $M(\rho_t)$. For simplicity, we will directly take the derivative of this function. This computation can be made rigorous by considering smooth approximation of a truncated squared function, with bounded second derivative, and recalling that $\rho_t$ admit a continuous density. We get:

22: %\[

23: %\frac{\de}{\de t} M(\rho_t ) = 2D/(D  \Gamma) - 2 \int \lambda \norm{\btheta}_2^2 + \btheta \cdot \lb \na V ( \btheta )  + \int \na  U ( \btheta , \btheta' ) \rho_t ( \de \btheta' )  \rb \rho_t ( \de \btheta ) \leq K + K A(t) 

24: %\]

25: %which implies the result by Gronwall's lemma.}

26: %\end{proof}

27: