---
title: Multinomial Diffusion Equation (MDE)
url: https://www.emergentmind.com/topics/multinomial-diffusion-equation-mde
type: topic
---

# Multinomial Diffusion Equation (MDE)

The Multinomial Diffusion Equation (MDE) is a finite-difference, discrete-time model of diffusion that captures stochastic fluctuations resulting from particle-level discreteness. Unlike classical deterministic or continuum stochastic formulations, the MDE provides a particle-conserving, synchronously updated evolution on an Eulerian grid, accurately reproducing ensemble statistics—including higher cumulants characteristic of multinomial fluctuations—across a broad range of regimes. The MDE has recently emerged as a foundation both for physically accurate simulation of discrete particle diffusion [1010.0719] and as a machine learning generative model for categorical data within the denoising diffusion probabilistic modeling paradigm [2102.05379].

## 1. Microscopic Dynamics of the Multinomial Diffusion Equation

The MDE models a 1D periodic spatial domain of length $L$ partitioned into $M$ voxels of size $\Delta x = L/M$. Let $N_i^t$ denote the integer number of particles in voxel $i$ at time $t$, with total $N_0 = \sum_i N_i^t$ fixed. At each discrete timestep $\Delta t$, each particle in voxel $i$ independently hops to its left neighbor with probability $k = D\,\Delta t/\Delta x^2$, to the right with probability $k$, or remains in place with probability $1-2k$ (with the constraint $k \ll 1$, $2k \leq 1$).

Let $L_i^t$ and $R_i^t$ denote the numbers of particles moving from voxel $i$ to $i-1$ and $i+1$, respectively, during $[t, t+\Delta t]$. Conditional on $N_i^t$, the pair $(L_i^t, R_i^t)$ follows a trinomial law:
\[
P[L_i^t = l, R_i^t = r] = \frac{N_i^t!}{l!~r!~(N_i^t - l - r)!}k^{l}k^{r}(1-2k)^{N_i^t-l-r}
\]
The update rule, expressing conservation and nearest-neighbor coupling, reads:
\[
N_i^{t+\Delta t} = N_i^t + L_{i+1}^t - R_{i+1}^t - L_i^t + R_i^t
\]
This preserves total mass and strictly enforces discrete particle counts, unlike continuum or stochastic partial differential equation approaches [1010.0719].

## 2. Continuum and Stochastic Limits

In the thermodynamic limit ($N_i \to \infty$, $k \ll 1$), bin exchange statistics can be approximated using the Central Limit Theorem. The mean and variance of outgoing hops satisfy:
\[
E[L_i^t]=E[R_i^t]=kN_i^t, \quad Var(L_i^t)=Var(R_i^t)=kN_i^t+o(k)
\]
Neglecting higher-order covariances and using $N_i^t = \rho_i^t \Delta x$, the macroscopic update, after combining Gaussian noise contributions, yields:
\[
\rho_i^{t+\Delta t} = \rho_i^t + \frac{D}{\Delta x^2}\left(\rho_{i+1}^t - 2\rho_i^t + \rho_{i-1}^t\right) +
\sqrt{\frac{D}{\Delta x^2} \left[\sqrt{\rho_{i+1}^t+\rho_i^t}\,\xi_{i,1}^t-\sqrt{\rho_i^t+\rho_{i-1}^t}\,\xi_{i,2}^t\right]}
\]
In the continuum ($\Delta x \to 0$, $\Delta t \to 0$, $D$ fixed), this converges to the stochastic diffusion equation (SDE):
\[
\partial_t \rho = D \partial_x^2\rho + \partial_x [\sqrt{2D\rho}\;\xi(x,t)]
\]
where $\xi(x,t)$ is space-time white noise. The MDE thus bridges particle-based and macroscopic stochastic diffusion models [1010.0719].

## 3. Equilibrium and Fluctuations: Ensemble Statistics

At equilibrium, the MDE yields a multinomial distribution over $\{N_i\}$ with equal bin probabilities $1/M$. The resulting statistics are:
- $E[N_i] = N_0/M$
- $Var(N_i) = N_0(1/M)(1-1/M)$
- $Cov(N_i, N_j) = -N_0/M^2$ for $i \neq j$

This translates in densities to
- $Var(\rho_i) = E[\rho_i]/\Delta x - E[\rho_i]^2\Delta x$

The SDE steady-state is Gaussian. The MDE, however, exactly reproduces all factorial moments (i.e., all cumulants) of the multinomial, including non-vanishing skewness and kurtosis at low $N_0$, where the SDE instead predicts vanishing higher cumulants. For $N_0 \gg 1$ particles per bin, SDE and MDE converge, but for $N_0 \sim 1$ the SDE systematically overestimates spatial variance [1010.0719].

## 4. Multinomial Diffusion in Generative Modeling

The MDE formalism has been adapted for machine learning in the modeling of one-hot categorical data $x \in \{0,1\}^K$ through a discrete-time forward–reverse process [2102.05379].

- **Forward (noising) process:** At each timestep $t$, the data is mixed with uniform noise (probability $\beta_t$), giving the categorical transition
  \[
  q(x_t\mid x_{t-1}) = \Cat\Big(x_t ~|~ (1-\beta_t)x_{t-1} + \frac{\beta_t}{K}\mathbf{1}\Big)
  \]
  This generates a Markov chain over the simplex, with cumulative signal retention $\bar\alpha_t = \prod_{\tau=1}^t (1-\beta_\tau)$.

- **Marginalization:** The closed-form marginal after $t$ steps is
  \[
  q(x_t|x_0) = \Cat\Big(x_t ~|~ \bar\alpha_t x_0 + (1-\bar\alpha_t) \tfrac{1}{K}\mathbf{1}\Big)
  \]

- **Reverse (generative) process:** To sample from the target data distribution, a parameterized network $\mu_\theta(x_t, t)$ (producing $\hat{x}_0$ on the simplex) learns to approximate the true posterior
  \[
  p_\theta(x_{t-1}|x_t) = \Cat\big(x_{t-1}|\theta^{\rm post}(x_t, \hat x_0)\big)
  \]
  with loss function 
  \[
  \mathcal{L} = \mathbb{E}_{t, x_0, x_t} \Big[ \mathrm{KL}\big(q(x_{t-1}|x_t, x_0)\| p_\theta(x_{t-1}|x_t)\big) \Big] + \text{(special term at $t=1$)}
  \]
  No score matching or stochastic approximation is required; all quantities admit closed-form [2102.05379].

This discrete-time MDE generalizes the Gaussian denoising diffusion models—the dynamical equation on the simplex plays the role of a finite-difference Fokker–Planck equation in the categorical case.

## 5. Validation and Regimes of Applicability

Numerical comparison of the MDE, SDE, and direct particle-tracking (overdamped Langevin) show:
- All recover correct mean as $\Delta t \to 0$
- For large $n_0$, variances match across methods
- For small $n_0$ ($O(1)$ particle/bin), MDE matches true variance, but SDE systematically overestimates variance
- SDE breakdown becomes pronounced near $n_0 \approx 1$ particle/bin or less

MDE thus provides a valid stochastic description in regimes inaccessible to the SDE—a critical feature for low-density, finite-population, and reaction–diffusion settings [1010.0719].

## 6. Computational and Modeling Implications

The MDE is a $\Delta t$-synchronous, grid-based, particle-conserving model suitable for:
- Capturing all cumulants of the stochastic diffusion process at the Eulerian level
- Multiscale modeling frameworks—naturally interfacing with finite difference solvers
- Efficient simulation of reaction–diffusion systems without stochastic time-step handling
- Operating between the exact but inefficient Multivariate Master Equation (asynchronous, event-based) and the SDE (efficient, Gaussian-approximate for large $N$)

A key operational constraint is the Courant–Friedrichs–Lewy (CFL) condition: $k = D\Delta t/\Delta x^2 \leq 1/2$ to avoid negative populations. The limit $N_i \gg 1$ is required for the SDE approximation, while the MDE remains exact for all $N_i \geq 0$.

In reactive systems, especially with bimolecular reactions, the MDE offers computational efficiency ($O(M)$ scaling) and accuracy at moderate-to-low densities compared to particle-tracking ($O(M^2)$) [1010.0719].

## 7. Broader Impact and Analytical Paradigms

The MDE framework, as originally formulated, underpins the exact treatment of stochastic fluctuations at discrete particle scales, bridging the gap between particle-based and macroscopic continuum models. Its adaptation for categorical diffusion in machine learning enables direct, tractable training and generation of categorical data, extending denoising diffusion approaches beyond continuous domains [2102.05379]. This intersection points toward unified stochastic models for both physical and data-generative systems, providing consistent ways to incorporate discreteness, mass conservation, and closed-form optimization for efficient inference and synthesis.

Source: https://www.emergentmind.com/topics/multinomial-diffusion-equation-mde