Equilibrium selection and sharp convergence rates for a nonlinear Fokker--Planck equation with simplex-volume interactions

arXiv:2609.16978 · math.AP · Submitted 2026-09-15 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Mathematics. The day in mathematics: new theorems, new proofs, and what they open.

Ada: We're looking at the paper: "Equilibrium selection and sharp convergence rates for a nonlinear Fokker--Planck equation with simplex-volume interactions".

Emmy: Here's the gist of the paper. The paper studies a diffusive nonlocal interaction equation generated by squared volumes of random n-simplices to prove global well-posedness and classify its stationary states based on dimension Model Formulation and Dynamics The model involves an extended generalized variance functional, Vn

ρ: , which is the expected squared volume of an n-simplex whose vertices are independently sampled from ρ The dynamics are governed by the equation ∂tρt + ∇ · (ρtvt) = γ∆ρt, where vt(x):= v

ρt: (x), vt is defined as −∇δVnδρ, and γ > 0 is the diffusion coefficient The associated free energy is defined by E

ρ: := γ Ent(ρ) + Vn

ρ: , ρ ∈ P2(R d) Global Existence and Representation Theorem 1.1 establishes the global well-posedness of the equation, stating that it admits a unique global measure-valued weak solution The full solution is recovered from the initial measure by an affine pushforward followed by convolution with a nondegenerate Gaussian measure This implies that ρt has a smooth strictly positive density for every t > 0, even when ρ0 is singular Equilibrium Classification and Symmetry Breaking The stationary weak solutions coincide with the global minimizers of E, which are classified based on dimension 1.

Ada: If 1 ≤ n < d, every equilibrium is of the form N (m, λ∗Id) with m ∈ R d, where λ∗ = γn!2(n + 1)d−1/n-1 2.

Emmy: If n = d, every equilibrium is of the form N (m, Σ) with m ∈ R d and Σ positive definite satisfying det Σ = γd!2(d + 1) Long-Time Convergence Rates The long-time behavior of the covariance matrix is studied, showing qualitative differences between the full-dimensional case n = d and the subcritical case 1 ≤ n 0 such that max λi(t) − λj(t) ≤ Ce−βt for every t ≥ 0, where β is the synchronization rate Convergence to Equilibrium Theorem 5.1 proves that W2(ρt, ρ∞) → 0 as t → ∞, where ρ∞ is a global minimizer of the free energy E and hence an equilibrium The convergence rate is given by W2(ρt, ρ∞) ≤ Ce−β∗t + e−αcovt for every t ≥ 0, where αcov is defined based on the initial covariance Σ0 Geometric Interpretation of Rates The two decay rates α∥ and α⊥ reflect the linearized geometry at the isotropic equilibrium, where a symmetric covariance perturbation splits into a component parallel to the isotropic line span and a traceless component orthogonal to that line For n > 1, this rate is strictly larger than α⊥, which explains the faster convergence in the isotropic case Final Conclusion on Convergence The paper concludes that for non-isotropic initial covariance, the covariance rate is no larger than the rate associated with transport memory exactly when d ≤ 2n − 1 In those cases, W2 decay is exactly e−αcovt regardless of the relation between αcov and β∗ For Gaussian initial data, the exact decay exponent is αcov regardless of the relation between αcov and β∗ The solution is stationary if ρ0 = ρ∞, and the Wasserstein distance is identically zero References 1.

Ada: Luigi Ambrosio, Nicola Gigli, and Giuseppe Savaré, Gradient flows in metric spaces and in the space of probability measures, Lectures in Mathematics ETH Zürich, Birkhäuser Verlag, Basel, 2005 2.

Paper summary: Ada: So we're looking at this paper titled "Equilibrium selection and sharp convergence rates for a nonlinear Fokker--Planck equation with simplex-volume interactions." It deals with a diffusive nonlocal interaction equation that comes from the squared volumes of random n-simplices. The main idea is to figure out when this system settles down and how fast it gets there based on the dimension of those simplices.

Emmy: Right, so we're looking at this paper on the "Equilibrium selection and sharp convergence rates for a nonlinear Fokker--Planck equation with simplex-volume interactions." It claims they can prove global well-posedness for any initial data in P2(R d), which means there's always a unique solution that exists forever. They also classify the stationary states based on dimension, which is interesting because it links the geometry of the interaction to what the final stable shapes look like.

Kolya: The paper claims they can classify those stationary states and show they match the global minimizers of an extended generalized variance functional called Vnrho. That's a big statement about finding where the system will end up. What this means for us is that we have a way to predict the long-term state of this kind of random interaction model.

Ada: Exactly, and it goes further by showing how fast things converge to those states. They prove that every solution converges to one of these equilibria in the two Wasserstein distance, and they give us explicit exponential convergence rates. That's a lot of detail about the dynamics once you get close to equilibrium.

Emmy: The paper explains that the convergence rate depends on whether we are looking at a full-dimensional case where n equals d, or a subcritical case where n is less than d. This dimension dependence is key because it dictates the structure of those final states.

Kolya: And when they look at the long-time behavior of the covariance matrix itself, they find that there are different ways things decay depending on whether you're in the full-dimensional case or not, with a synchronization rate β governing that decay.

Ada: They also connect these convergence rates to some geometric interpretation of those rates, where they show how a symmetric perturbation in the covariance splits into parts parallel and orthogonal to an isotropic line span. This helps explain why things converge faster in the isotropic case when n is greater than one.

Emmy: The paper also states that for non-isotropic initial covariance, the convergence rate is no larger than the rate associated with transport memory exactly when d equals two n minus one, and in those specific dimensions, the Wasserstein distance decay is exactly e to negative alpha covt regardless of how alpha cov and beta star relate.

Kolya: And for Gaussian initial data specifically, they say that the exact decay exponent is just alpha cov regardless of the relation between alpha cov and beta star. This is a strong result about how quickly these systems thermalize when starting from simple random distributions.

Ada: So to summarize, this paper on "Equilibrium selection and sharp convergence rates for a nonlinear Fokker--Planck equation with simplex-volume interactions" gives us global well-posedness, classifies the equilibria based on dimension n versus d, and provides explicit exponential convergence rates in the Wasserstein distance.

Emmy: It's about how this specific type of nonlocal interaction, driven by random n-simplices, settles down into different stable configurations depending on whether you are in a lower or higher dimensional setting relative to the space you're living in.

Kolya: And for anyone who only listens to the show, it means that if you have a system governed by these kinds of interactions, you can predict its final configuration and how quickly it arrives at that state based on the initial setup.

Ada: So we'll stick with this idea as we go into the conclusion about what all this stuff actually means for us in a broader sense.

Conclusion: Ada: So, this paper is about taking a nonlinear equation that describes how things interact nonlocally—like particles pushing each other based on the shape of random triangles—and figuring out exactly where those interactions end up settling down and how fast they get there.

Emmy: Yeah, the title itself tells you it’s two things at once: finding the stable states, and measuring just how quickly you hit them. The authors are digging into this because these types of models pop up a lot in physics when you look at granular media or complex fluids.

Kolya: What I find interesting is that they don't just say it settles down; they give you sharp convergence rates, which means we can actually calculate the time scale for thermalization. That’s a big deal for proving how systems reach equilibrium in practice.

Ada: Exactly, and the real punch here is how they classify those final states based on the dimension of the simplex—the number of sides it has versus the space we're living in. It shows that whether you have more dimensions or fewer, there are fundamentally different stable shapes possible.

Emmy: And they link these final shapes to something called a free energy functional, which is basically a measure of how much "energy" the system has left after all the dynamics have settled. So finding the minimum of that functional tells you what the universe is trying to settle into.

Kolya: If this holds up under scrutiny, it means that for many random systems, we can predict not just *if* they will reach a certain state, but *how* they get there and with what speed. That moves us from qualitative behavior to quantitative prediction.

Ada: It’s about taking something incredibly complex—nonlocal interactions based on high-dimensional geometry—and giving us a concrete roadmap for its long-term destiny. But the next thing we need to look at is how those specific convergence rates actually behave when you start with non-uniform, bumpy initial conditions instead of just smooth starting points.

DOHYUN KIM, HANSOL PARK, WOOJOO SHIM

National Research Foundation of Korea · Korea Institute for Advanced Study · National Science and Technology Council, Taiwan

math.AP

Submitted: 2026-09-15

Updated: 2026-11-16

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 83/100

The gist: The paper studies a diffusive nonlocal interaction equation generated by squared volumes of random n-simplices to prove global well-posedness and classify its stationary states based on dimension…

Key concepts

Extended generalized variance functional (Vn[ρ])
This functional measures the expected squared volume of an n-simplex whose vertices are drawn independently from a probability distribution $\rho$. It quantifies how the interaction energy depends on the shape and spread of the measure $\rho$ in space.
Stationary Weak Solutions (Equilibria)
These are stable states where the system stops changing over time. The paper shows that these equilibria correspond to global minimizers of a specific free energy functional $E[\rho]$. The structure of these minimizers changes significantly based on the relationship between the dimension of the simplex ($n$) and the dimension of the space ($d).
Wasserstein Distance Convergence Rate
This describes how quickly a system's state $\rho(t)$ approaches its final equilibrium state $\rho\infty$. The convergence is governed by two rates, one related to synchronization ($\beta$) and another related to the initial spread of the system ($\alpha$). This analysis reveals geometric insights into how perturbations decay.
Dimension Dependence of Equilibria
The paper classifies the possible stable states based on $n$ relative to $d$. When $n < d$, equilibria are simple forms. When $n=d$, the equilibrium involves a positive definite matrix $\Sigma$. These distinct classifications reveal fundamental differences in how the system settles down in different dimensional regimes.

Terminology

Summary

The paper studies a diffusive nonlocal interaction equation generated by squared volumes of random n-simplices to prove global well-posedness and classify its stationary states based on dimension

Model Formulation and Dynamics

The model involves an extended generalized variance functional, Vn[ρ], which is the expected squared volume of an n-simplex whose vertices are independently sampled from ρ The dynamics are governed by the equation ∂tρt + ∇ · (ρtvt) = γ∆ρt, where vt(x):= vρt, vt is defined as −∇δVnδρ, and γ > 0 is the diffusion coefficient The associated free energy is defined by E[ρ]:= γ Ent(ρ) + Vn[ρ], ρ ∈ P2(R d)

Global Existence and Representation

Theorem 1.1 establishes the global well-posedness of the equation, stating that it admits a unique global measure-valued weak solution The full solution is recovered from the initial measure by an affine pushforward followed by convolution with a nondegenerate Gaussian measure This implies that ρt has a smooth strictly positive density for every t > 0, even when ρ0 is singular

Equilibrium Classification and Symmetry Breaking

The stationary weak solutions coincide with the global minimizers of E, which are classified based on dimension

  1. If 1 ≤ n < d, every equilibrium is of the form N (m, λ∗Id) with m ∈ R d, where λ∗ = γn! p. 2

    Where the paper says it

    2) coincide with the global minimizers of E, which can be classified as follows. (1) If 1 ≤ n < d, every equilibrium is of the form N (m, λ∗Id) with m ∈ R d , where λ∗ = γn!

    2(n + 1)d−1/n-1

  2. If n = d, every equilibrium is of the form N (m, Σ) with m ∈ R d and Σ positive definite satisfying det Σ = γd! p. 2

    Where the paper says it

    2(n + 1)d−1 n−1 ! (2) If n = d, every equilibrium is of the form N (m, Σ) with m ∈ R d and Σ positive definite satisfying det Σ = γd!

    2(d + 1)

Long-Time Convergence Rates

The long-time behavior of the covariance matrix is studied, showing qualitative differences between the full-dimensional case n = d and the subcritical case 1 ≤ n 0 such that max λi(t) − λj(t) ≤ Ce−βt for every t ≥ 0, where β is the synchronization rate

Convergence to Equilibrium

Theorem 5.1 proves that W2(ρt, ρ∞) → 0 as t → ∞, where ρ∞ is a global minimizer of the free energy E and hence an equilibrium The convergence rate is given by W2(ρt, ρ∞) ≤ Ce−β∗t + e−αcovt for every t ≥ 0, where αcov is defined based on the initial covariance Σ0

Geometric Interpretation of Rates

The two decay rates α∥ and α⊥ reflect the linearized geometry at the isotropic equilibrium, where a symmetric covariance perturbation splits into a component parallel to the isotropic line span and a traceless component orthogonal to that line For n > 1, this rate is strictly larger than α⊥, which explains the faster convergence in the isotropic case

Final Conclusion on Convergence

The paper concludes that for non-isotropic initial covariance, the covariance rate is no larger than the rate associated with transport memory exactly when d ≤ 2n − 1 In those cases, W2 decay is exactly e−αcovt regardless of the relation between αcov and β∗ For Gaussian initial data, the exact decay exponent is αcov regardless of the relation between αcov and β∗ The solution is stationary if ρ0 = ρ∞, and the Wasserstein distance is identically zero

References

  1. Luigi Ambrosio, Nicola Gigli, and Giuseppe Savaré, Gradient flows in metric spaces and in the space of probability measures, Lectures in Mathematics ETH Zürich, Birkhäuser Verlag, Basel, 2005

  2. Daniel Balagué, José Antonio Carrillo, Thomas Laurent, and Gael Raoul, Dimensionality of local minimizers of the interaction energy, Arch. p. 0

    Where the paper says it

    We use this reduction to prove global well-posedness for arbitrary initial data in P2(R d ). We then classify the stationary states and show that they coincide with the minimizers of the associated free energy.

    Ration. Mech. Anal. 209 (2013), no. 3, 1055–1088

  3. François Bolley, Ivan Gentil, and Arnaud Guillin, Uniform convergence to equilibrium for granular media, Arch. p. 0

    Where the paper says it

    Within this framework, an important class of nonlinear aggregation and Fokker–Planck equations combines the entropy with a pairwise interaction term of the form 1 2 RR W(x−y) dρ(x) dρ(y). Such models have been studied with emphasis on measure-valued well-posedness, the geometry of minimizers, and convergence to equilibrium [2, 3, 4, 6, 7].

    Ration. Mech. Anal. 208 (2013), no. 2, 429–445

  4. José A. Carrillo, Marco Di Francesco, Alessio Figalli, Thomas Laurent, and Dejan Slepčev, Global-intime weak measure solutions and finite-time aggregation for nonlocal interaction equations, Duke Math. p. 0

    Where the paper says it

    Within this framework, an important class of nonlinear aggregation and Fokker–Planck equations combines the entropy with a pairwise interaction term of the form 1 2 RR W(x−y) dρ(x) dρ(y). Such models have been studied with emphasis on measure-valued well-posedness, the geometry of minimizers, and convergence to equilibrium [2, 3, 4, 6, 7].

    J. 156 (2011), no. 2, 229–271

  5. José A. Carrillo, Rishabh S. Gvalani, Grigorios A. Pavliotis, and André Schlichting, Long-time behaviour and phase transitions for the McKean–Vlasov equation on the torus, Arch. p. 0

    Where the paper says it

    Such models have been studied with emphasis on measure-valued well-posedness, the geometry of minimizers, and convergence to equilibrium [2, 3, 4, 6, 7]. The same competition between diffusion and interaction can also produce multiple equilibria and phase transitions, as shown for McKean–Vlasov-type equations in [5].

    Ration. Mech. Anal. 235 (2020), no. p. 0
    Where the paper says it

    For 1 ≤ n ≤ d and Date: September 16, 2026. 2020 Mathematics Subject Classification.

    1, 635–690

  6. José A. Carrillo, Robert J. McCann, and Cédric Villani, Kinetic equilibration rates for granular media and related equations: entropy dissipation and mass transportation estimates, Rev. Mat. Iberoam. 19 (2003), no. 3, 971–1018

  7. José A. Carrillo, Robert J. McCann, and Cédric Villani, Contractions in the 2-Wasserstein length space and thermalization of granular media, Arch. p. 0

    Where the paper says it

    Finally, we prove convergence of every solution to one of these equilibria in quadratic Wasserstein distance and obtain explicit exponential convergence rates. Introduction Nonlinear Fokker–Planck equations with nonlocal interactions often admit a variational interpretation through free energies on Wasserstein space.

    Ration. Mech. Anal. 179 (2006), no. 2, 217–263

  8. Matthias Gelbrich, On a formula for the L2 Wasserstein metric between measures on Euclidean and Hilbert spaces, Math Nachr. p. 0

    Where the paper says it

    This viewpoint provides a useful connection between diffusion, transport, and energy dissipation. Its development was initiated by the variational construction of the Fokker–Planck equation through the Jordan– Kinderlehrer–Otto scheme [9] and was further developed through Otto’s Riemannian calculus on Wasserstein space [11] and the metric theory in [1].

    147 (1990), no. 1, 185–203

  9. Richard Jordan, David Kinderlehrer, and Felix Otto, The variational formulation of the Fokker–Planck equation, SIAM J. p. 0

    Where the paper says it

    This viewpoint provides a useful connection between diffusion, transport, and energy dissipation. Its development was initiated by the variational construction of the Fokker–Planck equation through the Jordan– Kinderlehrer–Otto scheme [9] and was further developed through Otto’s Riemannian calculus on Wasserstein space [11] and the metric theory in [1].

    Math. Anal. 29 (1998), no. 1–17

  10. Dohyun Kim, Hansol Park, and Woojoo Shim, Finite-dimensional reduction of a Wasserstein gradient flow and sharp decay rates, 2026, arXiv preprint arXiv:2606. p. 0

    Where the paper says it

    DIMENSION-DEPENDENT SYMMETRY BREAKING IN A DIFFUSIVE NONLOCAL INTERACTION MODEL DOHYUN KIM, HANSOL PARK, AND WOOJOO SHIM Abstract. We study a diffusive nonlocal interaction equation generated by the squared volumes of random n-simplices.

    21918

  11. Felix Otto, The geometry of dissipative evolution equations: the porous medium equation, Comm. p. 0

    Where the paper says it

    Within this framework, an important class of nonlinear aggregation and Fokker–Planck equations combines the entropy with a pairwise interaction term of the form 1 2 RR W(x−y) dρ(x) dρ(y). Such models have been studied with emphasis on measure-valued well-posedness, the geometry of minimizers, and convergence to equilibrium [2, 3, 4, 6, 7].

    Partial Differential Equations 26 (2001), no. p. 0
    Where the paper says it

    Finally, we prove convergence of every solution to one of these equilibria in quadratic Wasserstein distance and obtain explicit exponential convergence rates. Introduction Nonlinear Fokker–Planck equations with nonlocal interactions often admit a variational interpretation through free energies on Wasserstein space.

    1–2, 101–174

  12. Luc Pronzato, Henry P. Wynn, and Anatoly A. Zhigljavsky, Extended generalised variances, with applications, Bernoulli 23 (2017), no. p. 1

    Where the paper says it

    q det(B⊤B). For ρ ∈ P2(R d ), the extended generalized variance is defined by Vn[ρ] := Z (Rd)n+1 Voln(x0, .

    4A, 2617–2642

13., Simplicial variances, potentials and Mahalanobis distances, J. Multivariate Anal. 168 (2018), 276–289

14., Bregman divergences based on optimal design criteria and simplicial measures of dispersion, Statistical Papers 60 (2019), no. p. 1

Where the paper says it

1) Thus Vn[ρ] is the expected squared volume of an n-simplex whose vertices are independently sampled from ρ. This functional belongs to the family of dispersion measures introduced and studied in [12], with related developments in [13, 14, 18].

2, 545–564

  1. Filippo Santambrogio, Euclidean, metric, and Wasserstein gradient flows: an overview, Bull. p. 0

    Where the paper says it

    Its development was initiated by the variational construction of the Fokker–Planck equation through the Jordan– Kinderlehrer–Otto scheme [9] and was further developed through Otto’s Riemannian calculus on Wasserstein space [11] and the metric theory in [1]. We refer to [15] for an overview.

    Math. Sci. 7 (2017), no. 1, 87–154

  2. Cédric Villani, Optimal transport: old and new, Grundlehren der mathematischen Wissenschaften, vol. p. 0

    Where the paper says it

    Introduction Nonlinear Fokker–Planck equations with nonlocal interactions often admit a variational interpretation through free energies on Wasserstein space. This viewpoint provides a useful connection between diffusion, transport, and energy dissipation.

    338, Springer, Berlin, Heidelberg, 2009

  3. Samuel S. Wilks, Certain generalizations in the analysis of variance, Biometrika 24 (1932), no. p. 1

    Where the paper says it

    q det(B⊤B). For ρ ∈ P2(R d ), the extended generalized variance is defined by Vn[ρ] := Z (Rd)n+1 Voln(x0, .

    3–4, 471–494

  4. Henry P. Wynn and Anatoly A. Zhigljavsky, An exterior algebra approach to generalised variances and cross-covariances, Soft Computing 29 (2025), no. p. 3

    Where the paper says it

    We use the convention that an elementary symmetric polynomial is zero when its degree is negative or exceeds the number of available variables. The simplex-volume interaction admits the covariance representation established in [12] and revisited from an exterior-algebra viewpoint in [18]: Vn[ρ] = cnen(Σ(ρ)), cn := n + 1 n!

    8, 4247–4257

  5. This paper is titled Equilibrium selection and sharp convergence rates for a nonlinear Fokker--Planck equation with simplex-volume interactions

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improved AI can perform exact mathematical derivations for moment equations by utilizing the closed finite-dimensional covariance system, which reduces complex measure-valued problems to solving a closed autonomous finite-dimensional system.

  2. The improved system can solve for the explicit representation of solutions by performing an affine pushforward followed by Gaussian convolution, even starting from singular initial data without requiring absolute continuity assumptions.

  3. The AI can classify stationary states and determine the exact form of equilibrium measures based on the dimension n, distinguishing between isotropic Gaussian equilibria when 1 <= n < d and anisotropic Gaussian equilibria when n = d.

  4. The system can predict the long-time behavior of a solution by analyzing the dynamics of the covariance matrix eigenvalues, specifically identifying whether convergence is exponential or algebraic decay based on whether 1 <= n < d or n = d.

  5. The improved AI can calculate sharp exponential convergence rates for solutions in quadratic Wasserstein distance by determining covariance decay exponents alpha cov, alpha parallel, and alpha perpendicular based on the initial covariance structure.

Abstract

We study a nonlinear Fokker--Planck equation with diffusion and nonlocal interactions generated by squared simplex volumes. The associated free energy is not isplacement semiconvex when the simplex dimension n is at least two. For 1 at most n<d, the stationary states are isotropic Gaussian measures, and every solution converges to the equilibrium determined by its conserved mean. At n=d, the free energy fixes only the covariance determinant, allowing a noncompact continuum of anisotropic Gaussian equilibria when d at least2. The limit is then selected by the initial covariance, whose eigenvectors and eigenvalue gaps are preserved by the flow. We establish global well-posedness for arbitrary initial probability measures with finite second moment, identify stationary states with the global minimizers of the free energy, and prove convergence in quadratic Wasserstein distance with explicit exponential rates. For compactly supported initial data with nonstationary covariance, the sharp rate is the smaller of the covariance relaxation rate and three times the slowest transport rate, provided that the third moments do not vanish along the directions of slowest transport. The skewness of the initial datum, rather than its covariance, may therefore dictate the long-time behavior.

Sources

Related papers