This paper shows that a common economic model of limited attention can be understood as a kind of regularized transport problem. The model of “rational inattention” weighs expected utility against an information cost measured by Shannon information. The author casts that objective as a nested version of the Schrödinger Bridge problem from entropic optimal transport and then uses tools from that literature to reproduce and extend earlier results to arbitrary choice sets.

In the usual discrete rational inattention setup an agent chooses a joint distribution over states and recommended actions to maximize expected utility minus the Kullback–Leibler divergence (the information cost) between the joint and the product of its marginals. The paper splits this into two steps. The inner step finds the best joint distribution when the marginal distribution over actions is fixed. That inner problem is exactly a Schrödinger Bridge problem. The outer step then chooses the best action marginal. This nested view makes the structure of the problem clearer.

The entropic optimal transport viewpoint brings concrete mathematical objects into the economic model. The inner problem produces two “potentials” (called Schrödinger potentials) for actions and states. These potentials explain why the conditional choice rule takes a familiar additive multinomial logit form: the log of the partition function for each state appears as a state potential, and the action potential measures the marginal gain of putting more probability on a given action. The paper also connects computational methods used in information economics (the Blahut–Arimoto algorithm) to the Sinkhorn matrix‑scaling algorithm used in entropic transport.