no. 17 · Wednesday 18 November 2026
From Schrodinger Bridge to Optimal Transport over Sub-Riemannian Manifolds
Abstract
We study the least-energy way to reshape a probability distribution when motion is constrained to a horizontal subbundle, i.e., optimal transport and distribution steering in sub-Riemannian geometry. It is well-known that the deterministic sub-Riemannian optimal transport problem can be complicated by the non-smooth geometry. To obtain a smooth and numerically tractable formulation, we introduce small noise that aligns with the geometry and study the associated Schrödinger bridge problem. We discuss the case where the given marginals can have compact support or have sub-Gaussian tails. Under bracket-generating hypotheses, we obtain smooth, strictly positive transition densities and a forward–backward characterization of the optimal bridge, which acts as a convex selection principle among potentially many deterministic optima. This leads to a practical Sinkhorn-type algorithm with geometric constraints and we discuss it convergence in both regimes. We show that as the noise level vanishes, the unique Schrödinger bridge minimizers converge to a minimizer of the deterministic sub-Riemannian optimal transport problem. In the case of full support regime, our result illustrates that, the existence and characterization of the unique Schrödinger bridge, Sinkhorn convergence and the convex selection principle is possible only after quantifying the smoothness and positivity of the transition density. We demonstrate with a benchmark Heisenberg group.
- fields
- Optimization and Control