1: \begin{proof}

2: The proof will use the following properties of $\nabla\varphi^*$:

3: \begin{enumerate}[(P1)]

4: \item $\nabla\varphi^*$ is odd, i.e., $\nabla\varphi^*(z)=-\nabla\varphi^*(-z)$. This is because $\varphi^*(z)=f(|z|)$ for some function $f$.

5: \item $\nabla\varphi^*$ is injective, i.e., if $\nabla\varphi^*(z_1)=\nabla\varphi^*(z_2)$ then $z_1=z_2$. This is because $\varphi^*$ is a convex function that implies that $\nabla\varphi^*$ is monotone.

6: \end{enumerate}

7: We recall that $u$ is called an equilibrium solution of Eq.~\eqref{eq: gRHE eqn} if it annihilates the total flux. Now, suppose that $u_{eq}$ is an equilibrium solution, i.e.,

8: \begin{equation*}

9: \nabla\varphi^*(\nabla V)+\nabla \varphi^*(\nabla \log u_{eq})=0.

10: \end{equation*}

11: Hence by (P1)

12: \begin{equation*}

13: \nabla \varphi^*(\nabla \log u_{eq})=-\nabla\varphi^*(\nabla V)=\nabla\varphi^*(-\nabla V),

14: \end{equation*}

15: and then by $(P2)$, 

16: \begin{equation*}

17: \nabla \log u_{eq}=-\nabla V,

18: \end{equation*}

19: which implies that $u_{eq}=C\exp(-V)$ for some constant $C$. If $\int\exp(-V(x))\,dx<\infty$, there exists only one value of $C$, namely $C=\left(\int\exp(-V(x))\,dx\right)^{-1}$, such that $u_{eq}$ is a probability measure on $\R^d$. This is exactly the Gibbs measure. 

20: 

21: However, we note that Eq. \eqref{eq: gRHE eqn} is still meaningful without this extra condition. This is also true for the classical drift-diffusion equation. In addition, its Wasserstein gradient flow structure is rigorously well-defined without the extra assumption on the growth of $V$, see \cite{JKO98}.

22: \end{proof}

23: