Discrete population models track individual entities — cells, organisms, particles — as they move through a structured space. Continuous models describe the same dynamics as smooth densities obeying partial differential equations. The passage from one to the other is upscaling, and it usually works: when the population is large and the structure is fine-grained, the continuum approximation captures the essential dynamics.
Agostinelli, Chambers, Byrne, and Dalwadi identify where it doesn't work. Using matched asymptotic expansions applied to discrete systems — a technique from fluid mechanics adapted for population models — they show that the continuum approximation breaks down in boundary layers: thin regions near the edges of the structured space where the discrete nature of the population cannot be averaged away.
In the bulk domain, the leading-order dynamics reduce to a nonlinear advection equation — the population is transported by its own transition rates, and diffusion is negligible. But near the boundaries, the transition rates change character. The discrete population cannot flux past the edge, creating a buildup that the continuum equation models as an advection-diffusion process in a thin inner layer. The boundary layer thickness scales with the discretization parameter, and the matching between inner and outer solutions determines the effective boundary conditions for the continuum model.
They demonstrate the framework on lipid-structured macrophages in early atherosclerosis — foam cells accumulating lipid droplets, where the discrete “structure” is the number of droplets per cell. The continuum model works in the interior of lipid space but fails at the extremes: cells with zero droplets or maximum capacity require the discrete boundary layer description. The biology imposes the boundaries; the mathematics determines what happens there.