On physical diffusion and stochastic diffusion

Although the same mathematical expression is used to describe physical diffusion and stochastic diffusion, there are intrinsic similarities and differences in their nature. A comparative study shows that characteristic terms of physical and stochastic diffusion cannot be placed exactly in one-to-one correspondence. Therefore, judgment needs to be exercised in transferring ideas between physical and stochastic diffusion.

💡 Research Summary

The paper “On physical diffusion and stochastic diffusion” presents a systematic comparison of two phenomena that are mathematically described by the same diffusion equation, ∂u/∂t = D∇²u, yet differ fundamentally in their physical interpretation, underlying assumptions, and the meaning of the parameters involved. The authors begin by outlining the classic derivation of the diffusion equation in the context of continuum mechanics. In physical diffusion—such as heat conduction, mass transport, or viscous momentum transfer—the equation emerges from conservation of energy or mass combined with constitutive laws (Fourier’s law, Fick’s law, etc.). The diffusion coefficient D is a material property that can be measured experimentally and depends on temperature, pressure, composition, and microscopic collision dynamics.

In contrast, stochastic diffusion arises from the statistical description of random walks, Brownian motion, or Markov processes. Here u(x,t) represents a probability density function, and D is defined as the rate at which the mean‑square displacement grows with time (the variance per unit time). Although the partial differential form is identical, the physical content of D is entirely different: it is a statistical parameter derived from ensemble averages rather than a directly measurable transport property.

The paper then examines the role of initial and boundary conditions. For physical diffusion, initial conditions correspond to actual temperature or concentration fields, while boundary conditions are imposed by real physical constraints such as insulated walls (Neumann), fixed temperature plates (Dirichlet), or convective exchange (Robin). These conditions guarantee the conservation of energy or mass across the domain. In stochastic diffusion, the initial condition is a probability distribution of particle positions, and the boundaries are mathematical constructs that enforce probability conservation—e.g., reflecting boundaries that prevent probability loss, or absorbing boundaries that model particle removal. The authors stress that the same mathematical symbols for flux or gradient have distinct interpretations in the two contexts.

A major part of the analysis focuses on the dimensional and conceptual differences of the diffusion coefficient. Although both have units of length² / time, the physical D is tied to microscopic collision frequencies and mean free paths, whereas the stochastic D is tied to the statistical step size and time increment of the underlying random walk. Consequently, substituting one for the other without proper justification can lead to severe quantitative errors.

The authors explore possible bridges between the two frameworks. They discuss how a Markovian random walk can be coarse‑grained to recover the macroscopic diffusion equation, and how stochastic variational methods (e.g., the stochastic Lagrangian) can embed random fluctuations into a deterministic transport model. However, they point out that such mappings require additional assumptions—such as isotropy, scale separation, or a specific flux‑flux relationship—that are not universally valid. Conversely, applying stochastic diffusion models to physical problems (e.g., diffusion of molecules in crowded cellular environments or price dynamics in financial markets) demands the inclusion of interaction forces, external potentials, and possibly non‑Gaussian step statistics, which go beyond the simple diffusion equation.

The paper concludes by emphasizing that, despite the shared mathematical form, the characteristic terms of physical and stochastic diffusion—diffusion coefficient, initial/boundary conditions, conserved quantities—cannot be placed in a strict one‑to‑one correspondence. Transfer of ideas between the two domains must therefore be performed with caution, explicitly accounting for the underlying physical or statistical assumptions. The authors recommend that researchers developing multi‑scale or hybrid models explicitly state which version of diffusion they are employing, validate the chosen parameters against appropriate experimental or ensemble data, and incorporate correction terms when necessary. This disciplined approach, they argue, will prevent misinterpretations and improve the reliability of models that straddle the boundary between deterministic transport and random processes.