Stochastic PNP Equation & Homogenization

• Stochastic PNP equation is a family of electro-diffusion models that introduces randomness via stochastic coefficients and probabilistic linearizations.
• It captures microscale ionic transport and electrostatic interactions in porous media through a rigorously developed triple-scale homogenization framework.
• The model effectively addresses heterogeneous material properties and interfacial charge dynamics, ensuring mass conservation and non-local boundary effects.

Searching arXiv for relevant papers on stochastic Poisson–Nernst–Planck formulations and homogenization. The stochastic Poisson–Nernst–Planck equation denotes a class of electro-diffusion models in which the classical Poisson–Nernst–Planck (PNP) system is coupled either to random media through stochastic coefficients and stochastic homogenization, or to an equivalent stochastic-process representation after linearization. In the porous-media setting, the stochasticity enters through material coefficients depending on a random dynamical system together with a periodic microstructure, yielding a “stochastic-periodic” model with triple-scale structure in the macroscopic variable $x$, the random variable $\omega$, and the fast periodic variable $y=x/\varepsilon^{2}$ (Tchinda et al., 2024). In a distinct one-dimensional linearized setting with closed ends, the same PNP dynamics can be recast as a damped heat equation with non-local boundary conditions and interpreted as the forward Kolmogorov equation of a reflected Brownian motion with Poisson-distributed resets to the endpoints (Wolansky, 2021). Taken together, these formulations show that “stochastic PNP” is not a single equation but a family of rigorously related models in which randomness may describe either heterogeneous media or an equivalent probabilistic dynamics.

1. Microscopic electro-diffusion model in random porous media

In porous media, the microscopic unknowns are the ionic concentrations for cations and anions, denoted $\varpi_{\pm,\varepsilon}(t,x,\omega)$, and the electric potential $\varUpsilon_{\varepsilon}(t,x,\omega)$, with $\varepsilon>0$ the small parameter associated with pore-scale oscillations (Tchinda et al., 2024). The fluid phase is $\Omega^f_\varepsilon$ and the solid phase is $\Omega^s_\varepsilon$.

The ionic transport is governed by the Nernst–Planck equations with isotropic diffusion and no convection: $$\frac{\partial \varpi_{\pm,\varepsilon}}{\partial t} - D_{\pm}\,\mathrm{div}\!\left(\nabla \varpi_{\pm,\varepsilon} + \frac{F}{R\Theta}\,z_{\pm}\varpi_{\pm,\varepsilon}\,\nabla \varUpsilon_{\varepsilon}\right) = 0 \quad \text{in } Q_{\varepsilon}^{f}\times\Lambda.$$ Here $D_{\pm}$ are constant molecular diffusion coefficients, $\omega$0 is Faraday’s constant, $\omega$1 is the gas constant, and $\omega$2 is temperature. No terms of the form $\omega$3 appear; fluid flow is neglected (Tchinda et al., 2024).

The electrostatic coupling is described by Poisson equations posed separately in the fluid and solid phases: $\omega$4

$\omega$5

The dielectric constants $\omega$6 and $\omega$7 depend on the random variable through the dynamical system $\omega$8 and on the periodic microscale $\omega$9, which is the defining feature of the stochastic-periodic dependence (Tchinda et al., 2024).

A global formulation is obtained by introducing

$y=x/\varepsilon^{2}$0

so that, in distributional form,

$y=x/\varepsilon^{2}$1

The measure term on $y=x/\varepsilon^{2}$2 encodes the jump of electric flux across the solid–fluid interface (Tchinda et al., 2024).

This formulation places the stochastic PNP system in a rigorous heterogeneous-medium setting: randomness affects constitutive coefficients, while geometry remains periodic. A plausible implication is that the model is designed to capture porous solids whose morphology is regular at the cell level but whose dielectric response varies randomly from one realization to another.

2. Geometry, interfaces, and stochastic-periodic structure

The macroscopic domain is a bounded connected open set $y=x/\varepsilon^{2}$3, with $y=x/\varepsilon^{2}$4 or $y=x/\varepsilon^{2}$5, and Lipschitz boundary $y=x/\varepsilon^{2}$6 (Tchinda et al., 2024). The periodicity cell is $y=x/\varepsilon^{2}$7, decomposed into fluid and solid parts,

$y=x/\varepsilon^{2}$8

with $y=x/\varepsilon^{2}$9 the internal interface.

The perforated domains are defined by

$\varpi_{\pm,\varepsilon}(t,x,\omega)$0

and the periodic surface is

$\varpi_{\pm,\varepsilon}(t,x,\omega)$1

The interface $\varpi_{\pm,\varepsilon}(t,x,\omega)$2 is assumed smooth enough, and all cell sets are extended by $\varpi_{\pm,\varepsilon}(t,x,\omega)$3-periodicity to $\varpi_{\pm,\varepsilon}(t,x,\omega)$4 (Tchinda et al., 2024).

The stochastic framework is built on a probability space $\varpi_{\pm,\varepsilon}(t,x,\omega)$5, with stochastic stationarity modeled by an $\varpi_{\pm,\varepsilon}(t,x,\omega)$6-dimensional measure-preserving dynamical system

$\varpi_{\pm,\varepsilon}(t,x,\omega)$7

The associated stochastic derivatives are denoted $\varpi_{\pm,\varepsilon}(t,x,\omega)$8, collected into $\varpi_{\pm,\varepsilon}(t,x,\omega)$9, and the stochastic divergence operator $\varUpsilon_{\varepsilon}(t,x,\omega)$0 is dual to $\varUpsilon_{\varepsilon}(t,x,\omega)$1 (Tchinda et al., 2024).

The coefficient field satisfies

$\varUpsilon_{\varepsilon}(t,x,\omega)$2

while the interfacial permittivity satisfies $\varUpsilon_{\varepsilon}(t,x,\omega)$3 and $\varUpsilon_{\varepsilon}(t,x,\omega)$4 (Tchinda et al., 2024). The terminology “stochastic-periodic” refers precisely to evaluation at

$\varUpsilon_{\varepsilon}(t,x,\omega)$5

combining stochastic stationarity in $\varUpsilon_{\varepsilon}(t,x,\omega)$6 with periodicity in $\varUpsilon_{\varepsilon}(t,x,\omega)$7.

This geometric setup is central because the surface $\varUpsilon_{\varepsilon}(t,x,\omega)$8 contributes to the limiting equations through surface two-scale convergence rather than through an ordinary volumetric limit. That distinction is essential for interfacial charge effects.

3. Boundary and interface conditions, including the Grahame relation

The Nernst–Planck part is equipped with no-flux conditions both on the internal interface and on the external fluid boundary: $\varUpsilon_{\varepsilon}(t,x,\omega)$9 where $\varepsilon>0$0 is the outward unit normal to $\varepsilon>0$1 (Tchinda et al., 2024).

For the potential, homogeneous Neumann conditions are imposed on external boundaries: $\varepsilon>0$2

At the solid–fluid interface, the electric double layer is encoded by the jump condition

$\varepsilon>0$3

where the scaled surface charge density is given by the Grahame relation

$\varepsilon>0$4

The function $\varepsilon>0$5 is monotone Lipschitz with $\varepsilon>0$6 and $\varepsilon>0$7 (Tchinda et al., 2024).

The initial conditions are

$\varepsilon>0$8

No nondimensionalization based on Debye length, Peclet number, or related dimensionless groups is introduced in the porous-media study; the model is written in physical units. The essential scaling is instead the small parameter $\varepsilon>0$9, together with the second fast variable $\Omega^f_\varepsilon$0 and the interfacial term scaled by $\Omega^f_\varepsilon$1 (Tchinda et al., 2024).

A common misconception is to identify stochastic PNP exclusively with noise-driven SPDEs. The present framework shows a different usage: stochasticity may enter through stationary random coefficients and interfacial constitutive laws, without any explicit stochastic forcing term in time (Tchinda et al., 2024). The stochastic component lies in the medium, not in additive or multiplicative temporal noise.

4. Stochastic two-scale convergence and homogenization mechanism

The central analytical tool is stochastic two-scale convergence, extended to periodic surfaces (Tchinda et al., 2024). In the bulk, a bounded sequence $\Omega^f_\varepsilon$2 in $\Omega^f_\varepsilon$3 is said to converge stochastically weakly two-scale to $\Omega^f_\varepsilon$4 if, for admissible test functions,

$\Omega^f_\varepsilon$5

The corresponding strong stochastic two-scale convergence is defined by weak two-scale convergence plus norm convergence (Tchinda et al., 2024).

A compactness result yields gradient decomposition: if $\Omega^f_\varepsilon$6 is bounded in $\Omega^f_\varepsilon$7, then along a subsequence

$\Omega^f_\varepsilon$8

for suitable limit fields $\Omega^f_\varepsilon$9, $\Omega^s_\varepsilon$0, and $\Omega^s_\varepsilon$1 (Tchinda et al., 2024). This is the origin of the triple-scale structure appearing in the homogenized PNP system.

For periodic surfaces, if

$\Omega^s_\varepsilon$2

then there exists a subsequence and a surface limit $\Omega^s_\varepsilon$3 such that

$\Omega^s_\varepsilon$4

for all admissible test functions (Tchinda et al., 2024).

This extension to periodic surfaces is decisive because the interfacial Grahame term contributes in the macroscopic limit through surface measures rather than by being absorbed into an effective bulk coefficient. That feature distinguishes the stochastic-periodic PNP homogenization from more standard bulk-only diffusion homogenization.

5. Homogenized variational system and effective description

The homogenization theorem states that, for the unique microscopic weak solution $\Omega^s_\varepsilon$5,

$\Omega^s_\varepsilon$6

$\Omega^s_\varepsilon$7

$\Omega^s_\varepsilon$8

$\Omega^s_\varepsilon$9

as $$\frac{\partial \varpi_{\pm,\varepsilon}}{\partial t} - D_{\pm}\,\mathrm{div}\!\left(\nabla \varpi_{\pm,\varepsilon} + \frac{F}{R\Theta}\,z_{\pm}\varpi_{\pm,\varepsilon}\,\nabla \varUpsilon_{\varepsilon}\right) = 0 \quad \text{in } Q_{\varepsilon}^{f}\times\Lambda.$$0 (Tchinda et al., 2024).

The limiting fields belong to the spaces

$$\frac{\partial \varpi_{\pm,\varepsilon}}{\partial t} - D_{\pm}\,\mathrm{div}\!\left(\nabla \varpi_{\pm,\varepsilon} + \frac{F}{R\Theta}\,z_{\pm}\varpi_{\pm,\varepsilon}\,\nabla \varUpsilon_{\varepsilon}\right) = 0 \quad \text{in } Q_{\varepsilon}^{f}\times\Lambda.$$1

$$\frac{\partial \varpi_{\pm,\varepsilon}}{\partial t} - D_{\pm}\,\mathrm{div}\!\left(\nabla \varpi_{\pm,\varepsilon} + \frac{F}{R\Theta}\,z_{\pm}\varpi_{\pm,\varepsilon}\,\nabla \varUpsilon_{\varepsilon}\right) = 0 \quad \text{in } Q_{\varepsilon}^{f}\times\Lambda.$$2

$$\frac{\partial \varpi_{\pm,\varepsilon}}{\partial t} - D_{\pm}\,\mathrm{div}\!\left(\nabla \varpi_{\pm,\varepsilon} + \frac{F}{R\Theta}\,z_{\pm}\varpi_{\pm,\varepsilon}\,\nabla \varUpsilon_{\varepsilon}\right) = 0 \quad \text{in } Q_{\varepsilon}^{f}\times\Lambda.$$3

The homogenized Nernst–Planck equation is stated in a global variational form involving $$\frac{\partial \varpi_{\pm,\varepsilon}}{\partial t} - D_{\pm}\,\mathrm{div}\!\left(\nabla \varpi_{\pm,\varepsilon} + \frac{F}{R\Theta}\,z_{\pm}\varpi_{\pm,\varepsilon}\,\nabla \varUpsilon_{\varepsilon}\right) = 0 \quad \text{in } Q_{\varepsilon}^{f}\times\Lambda.$$4, the correctors $$\frac{\partial \varpi_{\pm,\varepsilon}}{\partial t} - D_{\pm}\,\mathrm{div}\!\left(\nabla \varpi_{\pm,\varepsilon} + \frac{F}{R\Theta}\,z_{\pm}\varpi_{\pm,\varepsilon}\,\nabla \varUpsilon_{\varepsilon}\right) = 0 \quad \text{in } Q_{\varepsilon}^{f}\times\Lambda.$$5, and the limiting electric field through $$\frac{\partial \varpi_{\pm,\varepsilon}}{\partial t} - D_{\pm}\,\mathrm{div}\!\left(\nabla \varpi_{\pm,\varepsilon} + \frac{F}{R\Theta}\,z_{\pm}\varpi_{\pm,\varepsilon}\,\nabla \varUpsilon_{\varepsilon}\right) = 0 \quad \text{in } Q_{\varepsilon}^{f}\times\Lambda.$$6: $$\frac{\partial \varpi_{\pm,\varepsilon}}{\partial t} - D_{\pm}\,\mathrm{div}\!\left(\nabla \varpi_{\pm,\varepsilon} + \frac{F}{R\Theta}\,z_{\pm}\varpi_{\pm,\varepsilon}\,\nabla \varUpsilon_{\varepsilon}\right) = 0 \quad \text{in } Q_{\varepsilon}^{f}\times\Lambda.$$7

The homogenized Poisson equation is likewise global and variational: $$\frac{\partial \varpi_{\pm,\varepsilon}}{\partial t} - D_{\pm}\,\mathrm{div}\!\left(\nabla \varpi_{\pm,\varepsilon} + \frac{F}{R\Theta}\,z_{\pm}\varpi_{\pm,\varepsilon}\,\nabla \varUpsilon_{\varepsilon}\right) = 0 \quad \text{in } Q_{\varepsilon}^{f}\times\Lambda.$$8 The paper does not provide closed-form effective tensors such as $$\frac{\partial \varpi_{\pm,\varepsilon}}{\partial t} - D_{\pm}\,\mathrm{div}\!\left(\nabla \varpi_{\pm,\varepsilon} + \frac{F}{R\Theta}\,z_{\pm}\varpi_{\pm,\varepsilon}\,\nabla \varUpsilon_{\varepsilon}\right) = 0 \quad \text{in } Q_{\varepsilon}^{f}\times\Lambda.$$9, $D_{\pm}$0, or $D_{\pm}$1; the macroscopic description remains in global variational form with microscopic correctors and surface integrals (Tchinda et al., 2024).

This feature is methodologically significant. In standard periodic homogenization, one often isolates cell problems that directly define effective tensors. Here, the correctors are identified implicitly through the variational structure. A plausible implication is that the framework is optimized for existence and convergence rather than for direct numerical upscaling formulas.

6. Well-posedness, invariants, and the ergodic reduction

Under suitable regularity and positivity assumptions, including $D_{\pm}$2 and nonnegative initial data,

$D_{\pm}$3

the microscopic variational problem admits a unique solution for $D_{\pm}$4-almost every $D_{\pm}$5: $D_{\pm}$6 (Tchinda et al., 2024).

A priori estimates are obtained after extension to the full domain: $D_{\pm}$7 and

$D_{\pm}$8

These bounds give the compactness needed for passage to the stochastic two-scale limit (Tchinda et al., 2024).

The homogenized system satisfies nonnegativity and conserved mass of the concentrations: $D_{\pm}$9 It also satisfies a global electric equilibrium: $\omega$00 Consequently,

$\omega$01

is constant in time (Tchinda et al., 2024).

In the ergodic case, for example $\omega$02 with $\omega$03, effective coefficients and source terms can be averaged over $\omega$04, and the stochastic-periodic model reduces to a periodic reiterated homogenization problem (Tchinda et al., 2024). This clarifies that the stochastic description does not necessarily survive at the macroscopic level: under ergodicity, the effective equations may become deterministic.

7. Linearized stochastic interpretation on an interval

A different meaning of “stochastic Poisson–Nernst–Planck” appears in the linearized one-dimensional system with closed ends (Wolansky, 2021). On the interval $\omega$05, the full PNP model involves concentrations $\omega$06, potential $\omega$07, and electric field $\omega$08, with drift-diffusion equations

$\omega$09

and zero flux at both ends: $\omega$10 (Wolansky, 2021).

After nondimensionalization to $\omega$11 and linearization about the neutral state $\omega$12, $\omega$13, the charge density

$\omega$14

satisfies

$\omega$15

with $\omega$16 the square of the inverse Debye length in scaled units (Wolansky, 2021). The closed-end conditions become non-local: $\omega$17

This deterministic PDE is the forward Kolmogorov equation of a stochastic process on $\omega$18: reflected Brownian motion between jumps, with jumps occurring at Poisson times of rate $\omega$19, and reset to $\omega$20 or $\omega$21 with probabilities $\omega$22 and $\omega$23, respectively, when the pre-jump position is $\omega$24 (Wolansky, 2021). The backward generator is

$\omega$25

The linearized system admits an explicit spectral structure. The transformed heat operator has a ground eigenvalue $\omega$26 with eigenfunction

$\omega$27

an even cosine family

$\omega$28

and shifted sine modes determined by the transcendental equation

$\omega$29

(Wolansky, 2021). The heat kernel is recovered by inverse Laplace transform of the resolvent and has an explicit modal expansion.

The dynamics preserve total charge,

$\omega$30

preserve positivity, and satisfy the bound

$\omega$31

(Wolansky, 2021). The long-time behavior is governed by a steady component associated with the zero eigenvalue of the shifted generator and exponential decay of all remaining modes.

This one-dimensional theory should not be conflated with the stochastic-periodic homogenization framework. In the former, stochasticity is a probabilistic representation of an already linearized deterministic PDE. In the latter, stochasticity resides in the heterogeneous coefficients and the homogenization procedure itself (Wolansky, 2021, Tchinda et al., 2024). The two usages are mathematically distinct, but both illuminate how PNP systems can generate nonlocal or effective behavior once microscopic structure is encoded either statistically or probabilistically.