1: \begin{proof}

2: The proof is based on a standard energy estimates, together with applications of Young convolution inequality, so we omit some details, to show that solutions are sufficiently smooth so that the following computation is rigorous. In particular, we focus on uniqueness (which is the part that we need for our study of convergence). Let $\rho$, $\tilde{\rho} \in C([0,T]; \cP(\Td)) \cap L^2([0,T]; W^{1,2}(\Td))$ be two solutions, and consider their difference $\rho- \tilde\rho$. Then,

3: \[ t \mapsto \frac 1 2\norm{\rho_t -\tilde \rho_t}_2^2 = \frac 1 2 \int (\rho_t(x)- \tilde \rho_t(x))^2 (x) dx\]

4: is absolutely continuous, with weak derivative

5: \begin{equation*}\begin{split} \partial_t\frac 1 2 \norm{\rho_t -\tilde\rho_t}_2^2 

6: & =  \int \nabla \bra{\rho_t- \tilde\rho_t} \cdot \sqa{ \rho_t b_t \conv [g(\rho_t)] - \tilde\rho b_t \conv [g(\tilde\rho_t)] } - \int  \abs{\nabla \bra{ \rho_t- \tilde\rho_t} }^2\\

7: & \le \frac 1 4 \int \abs{ \rho_t b_t \conv [g(\rho_t)] - \tilde{\rho} b_t \conv [g(\tilde{\rho}_t)] }^2 \\

8: & \le \frac { \norm{b_t \conv g(\rho_t) }_\infty}{4}  \norm{ \rho_t - \tilde{\rho}_t}^2_2 + \norm{\tilde{\rho}_t}_2^2 \norm{ b_t \conv \sqa{g(\rho_t)-g(\tilde{\rho}_t)} }^2_\infty \\

9: &\le  c^2 \norm{b_t}_\infty^2   \norm{ \rho_t - \tilde{\rho}_t}^2_2  + \bra{ \operatorname{Lip}g}^2  \norm{\tilde{\rho}_t}_2^2 \norm{ b_t}_\infty^2  \norm{ \rho_t - \tilde{\rho}_t }^2_1\\

10:  &\le  \sqa{ c^2   +  \bra{ \operatorname{Lip}g}^2  \norm{\tilde{\rho}_t}_2^2 } \norm{b_t}_\infty^2\norm{\rho_t - \rho_t^2 }_2^2,

11: \end{split}\end{equation*}

12: where $c \ge 0$ is some constant such that $g(z) \le c(1+ z)$, for $z \in [0,\infty)$. By Gronwall lemma

13: \[ \sup_{t \in [0,T]} \norm{\rho_t - \tilde \rho_t}_2^2 \le \exp\cur{ \bra{T c^2   + \bra{\operatorname{Lip}g }^2  \norm{\tilde{\rho}}_{L^2_t(L^2_x)}^2 } \norm{b}_{L^\infty_t(L^\infty_x)}^2 }\norm{\rho_0 - \tilde \rho_0}_2^2,\]

14: from which uniqueness follows.

15: \end{proof}

16: