Gradient Flow Structure of the Spontaneous Aggregation Model

Abstract: We identify a previously unnoticed gradient-flow structure of the Fokker--Planck equation arising in the spontaneous particle aggregation model. For an exponential response function, the equation is a generalized Wasserstein gradient flow of the attractive McKean--Vlasov free energy with a nonlocal mobility. We show that, within a natural class of local entropies and symmetric interaction energies, this structure essentially singles out the exponential response. We discuss consequences for stationary states, global minimizers, the single-cluster structure in one dimension, and formally outline convergence to equilibrium using a Wasserstein--\mboxŁojasiewicz inequality.

• The paper identifies the exponential-response model as a generalized Wasserstein gradient flow of attractive McKean–Vlasov free energy with density-dependent, nonlocal mobility and proves monotonic energy dissipation.
• The exponential response is essentially selected within a broad entropy–interaction framework, while stationary states, minimizers, and mode-by-mode stability thresholds coincide with those of classical McKean–Vlasov dynamics.
• In one dimension, rearrangement methods establish single-cluster global minimizers, while a formal Łojasiewicz-based argument predicts convergence to equilibrium and highlights open analytical challenges involving positivity and compactness.

The spontaneous aggregation model, introduced by Burger, Haskovec, and Wolfram as a mean-field description of "direct aggregation," posits that individuals diffuse without preferred direction but with a diffusivity that decreases with locally perceived density. Its mean-field limit is the nonlinear, nonlocal Fokker–Planck equation

$$\partial_t\varrho = \tfrac12 \Delta\bigl(G(W*\varrho)^2\varrho\bigr),$$

posed on the $d$-dimensional torus, where $W$ is a normalized, radially symmetric sensing kernel and $G$ is the response function. The paper under review identifies a previously unreported variational structure of this equation: for an exponential response function $G(s)=e^{-\beta s}$, it is a generalized Wasserstein gradient flow of the attractive McKean–Vlasov free energy with a nonlocal mobility. This observation yields a unified treatment of stationary states, global minimizers, single-cluster structure in one dimension, and a formal convergence-to-equilibrium argument.

Gradient flow structure

For $G(s)=\exp(-\beta s)$ with $\beta>0$, define the free energy

$$\mathcal F_\beta(\varrho) = \int_{\mathbb T^d} \varrho(\log\varrho-1)\,dx - \beta\int_{\mathbb T^d}\varrho(W*\varrho)\,dx,$$

whose first term is the Boltzmann entropy and whose second is an attractive nonlocal interaction. The key computation uses the first variation (chemical potential) $$\delta\mathcal F_\beta/\delta\varrho = \log\varrho - 2\beta W*\varrho$$ and the symmetry of $W$: multiplying this chemical potential by the mobility $d$0 reproduces exactly the flux $d$1. Hence the Fokker–Planck equation takes the form $d$2, a generalized gradient flow with Onsager operator $d$3. Along smooth positive solutions,

$d$4

Because the mobility depends nonlocally on $d$5, this differs from the classical 2-Wasserstein gradient flow, whose mobility is proportional to $d$6.

Selection of the exponential response function

A central claim of the paper is that the exponential response is essentially forced by the gradient-flow requirement. Two independent arguments are given. First, allowing a general convex local entropy $d$7 in place of the Boltzmann entropy and comparing fluxes in the ansatz $d$8 with $d$9 yields $W$0. Since the left side depends only on $W$1 and the right only on $W$2, both equal a constant $W$3, forcing $W$4 to be the Boltzmann entropy up to affine terms and $W$5 to be exponential. Second, exploiting the natural factorization $W$6 with $W$7, symmetry of the Fréchet derivative of the candidate chemical potential $W$8 again forces $W$9 to be constant.

The author is careful to state the scope of this result: it is not an impossibility theorem for every conceivable variational formulation. A nonexponential $G$0 might admit a gradient structure based on a different metric or additional state variables; none has been found, apart from the trivial case $G$1 const. Notably, for power-law responses $G$2, all positive stationary states admit an exact variational characterization via the Euler–Lagrange equation of a nonlinear Rayleigh quotient, even though the evolution itself does not appear to be its gradient flow.

Relation to McKean–Vlasov dynamics

The free energy $G$3 coincides with the standard attractive McKean–Vlasov energy, but the two evolutions have different Onsager operators: mobility $G$4 for the granular media equation versus $G$5 here. Consequently, the two models share the same mass-constrained critical points, stationary states, global minimizers, and energetic classification, while their transient dynamics, spectra, and convergence rates may differ substantially. Because the mobility is exponentially suppressed where $G$6 is large, mass rearrangement inside dense clusters proceeds much more slowly than in classical McKean–Vlasov dynamics. The author offers this as a plausible mechanism for the slow coarsening and metastable multi-cluster patterns observed numerically in prior work — an explanation that remains heuristic rather than quantitatively established.

Stationary states and stability thresholds

Positive stationary states are characterized equivalently as constrained critical points of $G$7, solutions of the Euler–Lagrange equation $G$8, or fixed points of the Gibbs-type self-consistency relation

$G$9

This yields quantitative bounds: if $G(s)=e^{-\beta s}$0 then $G(s)=e^{-\beta s}$1, and the density contrast satisfies $G(s)=e^{-\beta s}$2. A relative-entropy identity shows that every nonconstant stationary state must satisfy $G(s)=e^{-\beta s}$3 with $G(s)=e^{-\beta s}$4; via Parseval's identity, if all nonzero Fourier coefficients of $G(s)=e^{-\beta s}$5 are nonpositive, the homogeneous state is the unique stationary state and the energy is strictly convex.

A notable finding is the exact coincidence of two thresholds: the homogeneous state loses local minimality in mode $G(s)=e^{-\beta s}$6 precisely when $G(s)=e^{-\beta s}$7, which is exactly the linear instability condition from the Fourier analysis of the original model. Thus linear instability and loss of energetic local minimality agree mode-by-mode. Additionally, if $G(s)=e^{-\beta s}$8, a Csiszár–Kullback–Pinsker argument proves the homogeneous state is the unique global minimizer.

Single-cluster minimizers in one dimension

Using the periodic Riesz rearrangement inequality on $G(s)=e^{-\beta s}$9, the paper shows that when $G(s)=\exp(-\beta s)$0 (equal to its symmetric decreasing rearrangement), a global minimizer exists that is even and nonincreasing up to translation — i.e., a single symmetric cluster. If $G(s)=\exp(-\beta s)$1 is strictly decreasing on $G(s)=\exp(-\beta s)$2, equality cases in the rearrangement inequality imply that every global minimizer has this form. The restriction to global minimizers is essential: multi-cluster stationary states are not excluded, consistent with the metastable multi-cluster patterns seen in simulations.

Formal long-time convergence

The final section outlines, formally, convergence of bounded trajectories to individual stationary states using the recent Wasserstein–Łojasiewicz inequality for McKean–Vlasov energies. The crucial bridge is the uniform equivalence of the dissipation functionals: since $G(s)=\exp(-\beta s)$3,

$G(s)=\exp(-\beta s)$4

where $G(s)=\exp(-\beta s)$5 is the McKean–Vlasov dissipation. Combining the Łojasiewicz–Simon inequality (which requires quantitative analyticity of $G(s)=\exp(-\beta s)$6) with the energy dissipation identity gives algebraic decay ($G(s)=\exp(-\beta s)$7) or exponential decay ($G(s)=\exp(-\beta s)$8) of the energy gap, and a finite-length estimate in $G(s)=\exp(-\beta s)$9 implies the trajectory is Cauchy and converges to its $\beta>0$0-limit point. The argument is explicitly formal: a rigorous implementation would require globally positive solutions, uniform-in-time parabolic regularity, and precompactness of the trajectory in $\beta>0$1 with $\beta>0$2. These are left open.

Limitations and open questions

The paper is candid about its scope. All derivations assume sufficient smoothness and strict positivity of solutions; no well-posedness, regularity, or asymptotic compactness theory is developed. The long-time convergence result is formal and contingent on compactness assumptions not proved here. The selection theorem for the exponential response applies only within the class of entropies plus symmetric quadratic interaction energies with scalar mobilities; whether a nonexponential $\beta>0$3 admits a gradient structure with a different metric or enlarged state space remains open, as does a rigorous variational treatment of the power-law case beyond stationary states. Finally, the explanation of slow coarsening via exponentially suppressed mobility is qualitative and not backed by rate estimates.

Conclusion

This note establishes that the spontaneous aggregation model with exponential response is a generalized Wasserstein gradient flow of the attractive McKean–Vlasov free energy with nonlocal mobility, and that within a natural class of variational structures this response function is essentially uniquely selected. The shared energy with classical McKean–Vlasov dynamics transfers the entire equilibrium theory — critical points, minimizers, stability thresholds, and single-cluster structure in 1D — while the distinct mobility accounts qualitatively for slow coarsening. The main outstanding tasks are a rigorous well-posedness and compactness framework supporting the formal Łojasiewicz-based convergence argument, and a definitive answer to whether nonexponential responses admit alternative gradient structures.