1: \begin{abstract}

2:   We show non-asymptotic exponential convergence of Sinkhorn iterates to the Schrödinger potentials, solutions of the quadratic Entropic Optimal Transport problem on $\rset^d$. 

3: Our results hold under mild assumptions on the marginal inputs: in particular, we only assume that they admit an asymptotically positive log-concavity profile, covering as special cases log-concave distributions and bounded smooth perturbations of quadratic potentials. 

4: More precisely, we provide exponential $\rml^1$ and pointwise

5: convergence of the iterates (resp. their gradient and Hessian) to

6: Schrödinger potentials (resp. their gradient and Hessian) for large enough values of the regularization parameter. As a

7: corollary, we establish exponential convergence of Sinkhorn plans with respect to a symmetric relative entropy and in $\bfW_1$.  Up to the authors'

8: knowledge, these are the first results which establish exponential

9: convergence of Sinkhorn's algorithm in a general setting without assuming bounded cost

10: functions or compactly supported marginals. %Our results are proven following a probabilistic approach that rests on integrated semiconvexity estimates for Sinkhorn iterates that are of independent interest. %Likewise, convergence of gradients and Hessians of iterates appears  in the unbounded setting,

11: %opening new perspectives in applications.

12: % Our proof strategy is built upon the connection between the gradients of Sinkhorn iterates and one-sided

13: % Lipschitz estimates of solutions to the Hamilton-Jacobi-Bellman equation,

14: % which we establish using coupling techniques.

15: \end{abstract}

16: