Smoluchowski Equation: Coagulation & Diffusion

• The Smoluchowski equation is a family of equations modeling binary coagulation and overdamped diffusion, capturing mass conservation and gelation phenomena.
• It employs weak formulations and moment methods to detail cluster-size evolution, flux balances, and self-similar behaviors in aggregation processes.
• The framework also encompasses the overdamped Fokker–Planck formulation for Brownian motion in potentials, linking aggregation dynamics with geometric and active-matter systems.

The Smoluchowski equation denotes two major, historically related classes of equations. In aggregation theory, the Smoluchowski coagulation equation is a mean-field population-balance model for the time evolution of cluster-size distributions under binary coalescence; in stochastic dynamics, the Smoluchowski equation is also the overdamped Fokker–Planck equation for Brownian motion in a potential. Across these usages, the central structures are gain–loss balance, diffusion or drift, weak formulations, and the competition between conservation laws and singular phenomena such as gelation, flux to infinity, trapping, or phase transition (Kimura et al., 15 May 2025, Diwaker et al., 2015).

1. Classical coagulation equation and weak formulations

In its continuous homogeneous form, the coagulation equation evolves a nonnegative size density $f(t,m)$ or $u(x,t)$ on $(0,\infty)$ by

$$\partial_t f(t,m) = \frac{1}{2}\int_0^m K(m',m-m')\,f(t,m')\,f(t,m-m')\,dm' - f(t,m)\int_0^\infty K(m,m')\,f(t,m')\,dm',$$

with symmetric nonnegative kernel $K$ (Norris, 2014, Kimura et al., 15 May 2025). In the discrete setting, with concentrations $c_m(t)$ or $f_m(x,t)$, the same gain–loss structure appears as

$$\partial_t c_m(t) = \frac{1}{2}\sum_{i+j=m} K(i,j)c_i(t)c_j(t) - c_m(t)\sum_{j\ge 1}K(m,j)c_j(t).$$

The first term counts formation of size $m$ by mergers; the second removes size $m$ by mergers with all partners (Hammond, 2014, Cepeda et al., 2010).

The measure-valued weak formulation is fundamental because the formal identity obtained by testing with $u(x,t)$0 does not automatically justify mass conservation. For a nonnegative Radon measure $u(x,t)$1, the weak equation is

$u(x,t)$2

for suitable bounded measurable test functions $u(x,t)$3 (Cepeda et al., 2010, Fournier, 7 Jan 2025). In the generalized-moment framework of 2025, one instead writes truncated weighted moments

$u(x,t)$4

and requires

$u(x,t)$5

for all continuous weights $u(x,t)$6, all $u(x,t)$7, and all $u(x,t)$8. This definition is proved equivalent to the Escobedo–Laurençot–Mischler–Perthame weak formulation (Kimura et al., 15 May 2025).

Moment functionals organize much of the theory. For $u(x,t)$9,

$(0,\infty)$0

so $(0,\infty)$1 is the total number of clusters and $(0,\infty)$2 is the total mass (Kimura et al., 15 May 2025). In broad classes of non-gelling kernels, $(0,\infty)$3 is conserved; in stronger-growth regimes, $(0,\infty)$4 may decrease because mass is transferred to an infinite cluster. The formal distinction between “mass-conserving” and “gelling” kernels is already visible at the level of homogeneous kernels $(0,\infty)$5, where the degree $(0,\infty)$6 is a first classifier (Kimura et al., 15 May 2025, Fournier, 7 Jan 2025).

2. Spatially inhomogeneous and measure-valued coagulation–diffusion equations

The spatially inhomogeneous Smoluchowski coagulation–diffusion equation augments the homogeneous coagulation operator with mass-dependent diffusion in physical space. In continuous mass variables,

$(0,\infty)$7

and in the discrete version,

$(0,\infty)$8

At fixed spatial point $(0,\infty)$9, coagulation is exactly the homogeneous Smoluchowski mechanism; the novelty is the parabolic transport and the possibility of mass-dependent diffusivities (Hammond, 2014, Norris, 2014).

Norris formulated a general measure-solution theory on a type space $$\partial_t f(t,m) = \frac{1}{2}\int_0^m K(m',m-m')\,f(t,m')\,f(t,m-m')\,dm' - f(t,m)\int_0^\infty K(m,m')\,f(t,m')\,dm',$$0, with measurable mass $$\partial_t f(t,m) = \frac{1}{2}\int_0^m K(m',m-m')\,f(t,m')\,f(t,m-m')\,dm' - f(t,m)\int_0^\infty K(m,m')\,f(t,m')\,dm',$$1, diffusivity $$\partial_t f(t,m) = \frac{1}{2}\int_0^m K(m',m-m')\,f(t,m')\,f(t,m-m')\,dm' - f(t,m)\int_0^\infty K(m,m')\,f(t,m')\,dm',$$2, and symmetric finite coagulation kernel $$\partial_t f(t,m) = \frac{1}{2}\int_0^m K(m',m-m')\,f(t,m')\,f(t,m-m')\,dm' - f(t,m)\int_0^\infty K(m,m')\,f(t,m')\,dm',$$3 satisfying $$\partial_t f(t,m) = \frac{1}{2}\int_0^m K(m',m-m')\,f(t,m')\,f(t,m-m')\,dm' - f(t,m)\int_0^\infty K(m,m')\,f(t,m')\,dm',$$4 $$\partial_t f(t,m) = \frac{1}{2}\int_0^m K(m',m-m')\,f(t,m')\,f(t,m-m')\,dm' - f(t,m)\int_0^\infty K(m,m')\,f(t,m')\,dm',$$5-a.e. The mild equation is

$$\partial_t f(t,m) = \frac{1}{2}\int_0^m K(m',m-m')\,f(t,m')\,f(t,m-m')\,dm' - f(t,m)\int_0^\infty K(m,m')\,f(t,m')\,dm',$$6

where $$\partial_t f(t,m) = \frac{1}{2}\int_0^m K(m',m-m')\,f(t,m')\,f(t,m-m')\,dm' - f(t,m)\int_0^\infty K(m,m')\,f(t,m')\,dm',$$7 is the heat semigroup with type-dependent diffusivity (Norris, 2014). Under the structural bound $$\partial_t f(t,m) = \frac{1}{2}\int_0^m K(m',m-m')\,f(t,m')\,f(t,m-m')\,dm' - f(t,m)\int_0^\infty K(m,m')\,f(t,m')\,dm',$$8, local boundedness assumptions on $$\partial_t f(t,m) = \frac{1}{2}\int_0^m K(m',m-m')\,f(t,m')\,f(t,m-m')\,dm' - f(t,m)\int_0^\infty K(m,m')\,f(t,m')\,dm',$$9, and weighted integrability of the initial data, there exists a maximal strong solution, uniqueness holds within the strong-solution class, and mass is conserved when the initial mass–weight moment is finite (Norris, 2014).

A particularly important example is the Einstein–Smoluchowski model for Brownian coagulation of spherical colloids in $K$0. There,

$K$1

and the Brownian kernel is

$K$2

With $K$3 and a suitable choice of weight $K$4, the Norris framework accommodates the singular behavior as $K$5, yields global existence, and yields mass conservation when

$K$6

(Norris, 2014).

The coagulation–diffusion PDE also arises as a kinetic limit of many Brownian particles undergoing short-range coagulation. In $K$7, with $K$8, Hammond and Rezakhanlou show that the empirical measures converge to weak solutions with macroscopic kernel

$K$9

where $c_m(t)$0 solves

$c_m(t)$1

The factor $c_m(t)$2 is the survival correction accounting for short-range depletion of near-collision pairs (Hammond, 2014).

3. Mass conservation, fluxes, and gelation

A central issue in coagulation theory is that the equation is formally mass-conserving but may lose finite-size mass through gelation. The 2025 generalized-moment analysis introduces a mass flux

$c_m(t)$3

and proves the weak conservation law

$c_m(t)$4

in the sense of distributions (Kimura et al., 15 May 2025). Moreover,

$c_m(t)$5

so that formally

$c_m(t)$6

This identifies gelation as flux of mass to infinity (Kimura et al., 15 May 2025).

The generalized-moment method yields sharp sufficient conditions for conservation and gelation. If a continuous weight $c_m(t)$7 satisfies $c_m(t)$8 and the kernel obeys inequalities of the form

$c_m(t)$9

$f_m(x,t)$0

then any weak solution conserves mass: $f_m(x,t)$1 If instead $f_m(x,t)$2 is nondecreasing, the kernel is nondegenerate at large sizes, and

$f_m(x,t)$3

then gelation occurs; if $f_m(x,t)$4, finite-time gelation is guaranteed whenever

$f_m(x,t)$5

and one may write

$f_m(x,t)$6

These criteria apply to possibly inhomogeneous kernels, not only to homogeneous ones (Kimura et al., 15 May 2025).

For homogeneous kernels

$f_m(x,t)$7

the generalized-moment framework recovers the classical threshold: if $f_m(x,t)$8, then mass is conserved; if $f_m(x,t)$9, the regime is consistent with gelation (Kimura et al., 15 May 2025). The additive kernel $$\partial_t c_m(t) = \frac{1}{2}\sum_{i+j=m} K(i,j)c_i(t)c_j(t) - c_m(t)\sum_{j\ge 1}K(m,j)c_j(t).$$0 sits at the critical line $$\partial_t c_m(t) = \frac{1}{2}\sum_{i+j=m} K(i,j)c_i(t)c_j(t) - c_m(t)\sum_{j\ge 1}K(m,j)c_j(t).$$1 and conserves mass; the multiplicative kernel $$\partial_t c_m(t) = \frac{1}{2}\sum_{i+j=m} K(i,j)c_i(t)c_j(t) - c_m(t)\sum_{j\ge 1}K(m,j)c_j(t).$$2 lies beyond it and is a canonical gelling example (Kimura et al., 15 May 2025).

A separate 2025 analysis revisits the Escobedo–Mischler–Perthame proof and gives a short deterministic criterion for finite-time gelation. If

$$\partial_t c_m(t) = \frac{1}{2}\sum_{i+j=m} K(i,j)c_i(t)c_j(t) - c_m(t)\sum_{j\ge 1}K(m,j)c_j(t).$$3

then any weak solution with $$\partial_t c_m(t) = \frac{1}{2}\sum_{i+j=m} K(i,j)c_i(t)c_j(t) - c_m(t)\sum_{j\ge 1}K(m,j)c_j(t).$$4 and $$\partial_t c_m(t) = \frac{1}{2}\sum_{i+j=m} K(i,j)c_i(t)c_j(t) - c_m(t)\sum_{j\ge 1}K(m,j)c_j(t).$$5 loses mass in finite time (Fournier, 7 Jan 2025). For continuous homogeneous kernels of degree $$\partial_t c_m(t) = \frac{1}{2}\sum_{i+j=m} K(i,j)c_i(t)c_j(t) - c_m(t)\sum_{j\ge 1}K(m,j)c_j(t).$$6 with $$\partial_t c_m(t) = \frac{1}{2}\sum_{i+j=m} K(i,j)c_i(t)c_j(t) - c_m(t)\sum_{j\ge 1}K(m,j)c_j(t).$$7, this condition holds automatically, so any weak solution gels (Fournier, 7 Jan 2025). At the critical homogeneity $$\partial_t c_m(t) = \frac{1}{2}\sum_{i+j=m} K(i,j)c_i(t)c_j(t) - c_m(t)\sum_{j\ge 1}K(m,j)c_j(t).$$8, the power of the logarithmic correction required for gelation depends on the kernel shape: for $$\partial_t c_m(t) = \frac{1}{2}\sum_{i+j=m} K(i,j)c_i(t)c_j(t) - c_m(t)\sum_{j\ge 1}K(m,j)c_j(t).$$9, the threshold is $m$0, whereas for the near-diagonal band kernel

$m$1

the threshold is $m$2 for monodisperse data (Fournier, 7 Jan 2025).

4. Self-similar, stationary, and forced regimes

Self-similarity organizes long-time behavior in non-gelling regimes and singular behavior near gelation. For the multiplicative family

$m$3

the kernel is homogeneous of degree $m$4, so gelation is absent and mass is conserved (Breschi et al., 2022). Global-in-time self-similar solutions are sought as

$m$5

with

$m$6

Equivalently,

$m$7

For $m$8, the profile has a three-region structure consisting of very-fast decay near the origin, a lognormal intermediate region, and a gamma-distribution tail; for $m$9, it has a singular origin, a Pareto-like intermediate region, and the same gamma tail (Breschi et al., 2022).

Near gelation, self-similarity becomes more delicate. For kernels

$m$0

the equation gels in finite time, and the similarity ansatz

$m$1

leads to an anomalous drift exponent

$m$2

The Laplace-transform formulation yields a nonlocal similarity equation for $m$3 and its fractional integral $m$4, and the analysis identifies a stable lower branch continued from the exactly solvable $m$5 product kernel together with an upper branch near $m$6 that violates scaling relations often assumed universal (Eggers et al., 2022). In particular, the linear laws $m$7 and $m$8 are not generally valid (Eggers et al., 2022).

Open and forced versions of the coagulation equation replace exact mass conservation by flux balance. With a constant flux of dust particles entering at zero size, one obtains weak flux solutions satisfying

$m$9

For a broad non-gelling class of kernels with homogeneity $u(x,t)$00 and $u(x,t)$01, there are time-dependent flux solutions with linearly increasing mass

$u(x,t)$02

and size-averaged time-integrated densities are bounded above by the explicit constant-flux profile

$u(x,t)$03

For the constant kernel, the Bernstein transform solves a Riccati ODE and converges to the stationary flux measure $u(x,t)$04 (Ferreira et al., 2024).

In discrete forced coagulation with source $u(x,t)$05 and removal $u(x,t)$06,

$u(x,t)$07

well-posedness holds for a large class of kernels $u(x,t)$08 and removal laws $u(x,t)$09, and under a suitable smallness condition solutions converge exponentially to a unique equilibrium (Kuehn et al., 2018). For aggregation–fragmentation steady states with steeply decaying tails, a fast two-stage numerical algorithm constructs a small “seed system” and extends it through a quadratic recurrence for $u(x,t)$10, yielding orders-of-magnitude speedups over standard schemes (Stadnichuk et al., 2015).

5. Particle systems, stochastic approximations, and numerical methods

A principal finite-particle approximation to coagulation is the Marcus–Lushnikov process. If

$u(x,t)$11

then coalescence of masses $u(x,t)$12 and $u(x,t)$13 replaces $u(x,t)$14 by $u(x,t)$15, and the empirical measure satisfies a martingale formulation converging to the weak Smoluchowski equation (Cepeda et al., 2010). For homogeneous-like kernels with degree up to $u(x,t)$16, Cepeda and Fournier introduce a Wasserstein-type distance $u(x,t)$17 and prove an explicit convergence rate

$u(x,t)$18

uniform on compact time intervals, together with constructive discrete approximations of the initial data achieving the same rate (Cepeda et al., 2010).

For large discrete systems, recent computational work exploits low-rank structure of the kernel. Low-rank Monte Carlo factorizes or upper-bounds the collision kernel as a sum of rank-one terms and samples collisions with segment trees. The per-collision cost is

$u(x,t)$19

where $u(x,t)$20 is the factorization rank and $u(x,t)$21 the number of size classes (Osinsky, 2023). The Brownian benchmark kernel

$u(x,t)$22

is exactly rank $u(x,t)$23, the linear kernel $u(x,t)$24 is rank $u(x,t)$25, and the ballistic kernel admits a low-rank upper bound with acceptance probability at least $u(x,t)$26 (Osinsky, 2023). The method remains exact after acceptance–rejection correction and preserves total mass at the particle level (Osinsky, 2023).

A related low-rank strategy applies to coagulation combined with Ostwald ripening in volume space: $u(x,t)$27 with

$u(x,t)$28

In the dimensionless variables of the 2026 study, the coupled model admits an exact parametric solution of the form

$u(x,t)$29

so the scaled profile is universally exponential,

$u(x,t)$30

independent of the initial volume distribution (Zaks et al., 20 Jan 2026).

The same coagulation formalism has been carried into astrophysical applications. For primordial black-hole clusters, the merger kernel is obtained by velocity-averaging the gravitational-wave capture cross section, yielding

$u(x,t)$31

hence an effective homogeneous degree

$u(x,t)$32

This implies finite-time runaway in the Smoluchowski sense. The simulations use an event-driven full-conditioning Monte Carlo scheme with waiting time

$u(x,t)$33

and show that mass segregation further shortens the runaway time (Zhang et al., 2 Apr 2026).

6. Overdamped Fokker–Planck, geometric, and active-matter variants

In another established usage, the Smoluchowski equation is the overdamped Fokker–Planck equation for the probability density $u(x,t)$34 of a Brownian particle in a potential $u(x,t)$35: $u(x,t)$36 For constant $u(x,t)$37, this is equivalently

$u(x,t)$38

(Diwaker et al., 2015). With localized time-dependent sinks, exact Laplace-transform methods reduce the PDE to an integral equation for the single function $u(x,t)$39; once $u(x,t)$40 is known, the full density follows by convolution with the Green function. This program has been carried out for a flat potential, a piecewise linear potential, and a parabolic potential, with exact solutions for constant, linear, inverse, and exponential sink strengths in the cases treated (Diwaker et al., 2014, Diwaker et al., 2014, Diwaker et al., 2015).

On curved manifolds, the intrinsic Smoluchowski equation reads

$u(x,t)$41

On the sphere $u(x,t)$42,

$u(x,t)$43

with conserved probability and flux

$u(x,t)$44

A geometry-based algorithm evolves particles in the tangent plane and projects them back to the sphere; it converges in the weak sense, captures curvature-induced effects in both transient and stationary regimes, and reproduces the stationary Boltzmann density $u(x,t)$45 (Gómez et al., 2021).

The same overdamped framework also supports nonlinear orientational dynamics. For the modified Smoluchowski equation on the unit sphere with dipolar potential,

$u(x,t)$46

where

$u(x,t)$47

equilibria are von Mises–Fisher states $u(x,t)$48 solving the self-consistency condition

$u(x,t)$49

The theory identifies stable and unstable equilibria, exponential convergence rates, second-order transitions when $u(x,t)$50 is non-increasing, and first-order transitions with hysteresis when the ratio grows in part of the state space (Degond et al., 2012).

For active Brownian swimmers in $u(x,t)$51, the phase-space Smoluchowski equation becomes

$u(x,t)$52

After marginalization over the angle $u(x,t)$53, the long-time position density is Gaussian with effective diffusivity

$u(x,t)$54

and the exact mean-square displacement is

$u(x,t)$55

The kurtosis quantifies non-Gaussianity generated by orientational persistence; for $u(x,t)$56, the deviation from Gaussian scales as $u(x,t)$57 at short times and $u(x,t)$58 asymptotically (Sevilla et al., 2015).

Taken together, these strands show that “Smoluchowski equation” is not a single model but a family of structurally related equations. In coagulation theory it governs mass exchange across size classes, weak solutions, flux to infinity, and gelation; in stochastic transport it governs overdamped diffusion, reactive sinks, curvature, and active matter. The modern literature extends both lines through generalized moments, measure solutions, kinetic limits, low-rank computation, and explicit singular or stationary asymptotics (Kimura et al., 15 May 2025, Hammond, 2014)