The paper tackles a fundamental question about how certain mathematical models of physical systems relate to one another. In many real-world situations, particles or agents interact with neighbors within some finite range, leading to "nonlocal" equations where the behavior at one point depends on conditions nearby. A natural question is whether these nonlocal models become consistent with simpler "local" models when the interaction range shrinks to zero. This paper proves rigorously that yes, under the right conditions, the nonlocal model converges to the local one in a setting that includes randomness, meaning the equations are driven by noise that varies across space.
The technical setting involves conservation laws, which are equations that track quantities like mass, momentum, or traffic density over time. Adding randomness makes these equations stochastic, and the particular noise considered here depends on the gradient of the solution, which creates extra mathematical difficulties. The authors handle these challenges by introducing modified versions of the equations with carefully chosen boundaries and by deriving new estimates that keep the solutions stable throughout the limiting process. These tools allow them to track what happens to solutions as the interaction range is gradually reduced to nothing.
A particularly striking finding is that the convergence does not always work. The authors construct a concrete counterexample showing that if a small diffusion term (called a viscous term) is absent from the equations, the limiting solution can fail to satisfy a physically meaningful condition known as the entropy condition. This condition rules out unphysical solutions that mathematics might allow but nature would not. The result therefore clarifies not just when nonlocal stochastic models can be trusted to approximate local ones, but also why certain mathematical ingredients are genuinely necessary and not just convenient assumptions.