A solution to the Monge transport problem for Brownian martingales (Q2039418)

The classical Monge-Kantorovich optimal transport problem can be expressed as an optimal statistical correlation problem between probability measures \(\mu\) and \(\nu\) in the following way: \[ \inf \{ \mathbb{E} [ c(X,Y)]: X \sim \mu, Y \sim \nu \}, \] where \(c: \mathbb{R}^d \times \mathbb{R}^d\) is a given cost function and \(X \sim \mu\) means that \(\nu\) is the distribution of the random variable \(X\). In this paper, the authors study a constrained version of the above problem called the Brownian martingale transport problem, in which it is required that the target distribution \(\nu\) is reached via the Brownian motion, i.e. \[ \inf_\tau \{ \mathbb{E} [ c(B_0,B_\tau)]: B_0 \sim \mu, B_\tau \sim \nu \}. \] Here, \(B_t\) is a Brownian motion starting with \(\mu\) and ending at a stopping time \(\tau\) such that \(B_\tau\) realizes the target distribution \(\nu\). In particular, this requires a certain condition between \(\mu\) and \(\nu\) in order for at least one competitor to exist, i.e., that \(\mu\) and \(\nu\) are in subharmonic order. The problem consists of studying existence and uniqueness of solutions together with the uniqueness of the optimal stopping time \(\tau^*\). This is achieved in this paper in several steps. First, the authors use the methods of convex analysis and provide a precise statement of the dual problem and show existence of solutions for the dual problem. Uniqueness of solutions requires additional assumptions, and in the unconstrained case a classical assumption is called the twist condition; here, the authors introduce its stochastic version, and under this assumption prove uniqueness of the optimal Brownian martingale and characterise it as the first hitting time to a certain barrier set. Contrary to the unconstrained case, the stochastic twist condition does not cover the case when the cost is quadratic, i.e. \(c(x,y) = |x-y|^2\), but instead it covers the linear case \(c(x,y) = |x-y|\).