Monotone Multispecies Flows

Lauren Conger, Franca Hoffmann, Eric Mazumdar, Lillian J. Ratliff · 2025

Read Monotone Multispecies Flows by Lauren Conger, Franca Hoffmann, Eric Mazumdar, Lillian J. Ratliff (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-22947.txt.

Introduction

Lauren Conger, Franca Hoffmann, Eric Mazumdar, and Lillian J Ratliff define a new kind of monotonicity for systems of many interacting species. Their 2025 paper works with flow equations that preserve total mass. Distance is measured by the Wasserstein-2 metric, the cost of moving one pile of mass into another. Monotonicity in that setting turns out to be a powerful organizing property.

When the property holds, the system has exactly one steady state and every solution settles into it. Velocity fields and second moments converge too, and distances shrink at rates set by the strength of the monotonicity. In the special case where each species descends its own energy landscape, the steady state is the Nash equilibrium of those energies, extending known zero-sum results to general games. Examples run from cross-diffusion to mean-field games, and a market model shows the theory working in simulation.

Catalog record

CREATORLauren Conger, Franca Hoffmann, Eric Mazumdar, Lillian J. Ratliff
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE149 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-22947

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Monotone Multispecies Flows Lauren Conger, Franca Hoffmann, Eric Mazumdar, Lillian J. Ratliff Abstract We present a novel notion of λ \lambda -monotonicity for an n n -species system of partial differential equations governed by mass-preserving flow dynamics, extending monotonicity in Banach spaces to the Wasserstein-2 metric space. We show that monotonicity implies the existence of and convergence to a unique steady state, convergence of the velocity fields and second moments, and contraction in the Wasserstein-2 metric, at rates dependent on λ \lambda . In the special setting of Wasserstein-2 gradient descent of different energies for each species, we prove convergence to the unique Nash equilibrium of the associated energies and delineate the relationship between monotonicity and displacement convexity. This extends known zero-sum results in infinite-dimensional game theory to the general-sum setting. We provide a number of examples of monotone coupled gradient flow systems, including cross-diffusion, gradient flows with potentials, nonlocal interaction, linear and nonlinear diffusion, and min-max systems, and draw connections to a class of mean-field games. Numerically, we demonstrate convergence of a four-player economic model for service providers and strategic users competing in a market, and a degenerately monotone game. Keywords. monotonicity, Wasserstein gradient flow, game theory, strategic learning, interacting species AMS subject classification. 35G50, 91A06, 35B40. 1 Introduction In the analysis of nonlinear dynamical systems, understanding the long-time behavior is important for modeling applications. When the dynamical system evolves via an operator on Banach spaces, monotonicity has proven to be a useful criteria under which convergence can be shown. In this work, we investigate mass-preserving flows, which are not in a normed space, and develop a notion of monotonicity in Wasserstein space which gives similar results to those for Banach spaces. Mass-preserving flow equations are commonly used to describe physical systems, such as fluid flow, biological systems [ 43 , Ch 8] , traffic [ 37 ] , and increasingly appear in emerging areas like generative modeling [ 34 ] , opinion dynamics [ 28 ] , and strategic learning environments [ 19 , 20 ] . Importantly, the monotonicity condition we propose has a structure that allows for systems where there is more than one entity evolving, such as in coupled flow equations. Understanding the long-time behavior of the coupled dynamics is a key question in these applications; in particular, monotonicity allows us to determine the existence of and convergence to a unique steady state. Our notion of monotonicity is a natural extension of the well-studied condition for Euclidean dynamical systems in finite dimensions and Banach spaces in infinite dimensions, providing a condition for contraction of nonlinear dynamics. We show how this idea can be applied to velocity fields driving the evolution of probability measures. In the monotone flow setting, we consider an n –species system where each species \rho_{i}\in\mathcal{P}(\mathbb{R}^{d_{i}}) , with \mathcal{P}(\mathbb{R}^{d_{i}}) the space of probability measures on \mathbb{R}^{d_{i}} , evolves according to a velocity field v_{i}[\rho]:\mathbb{R}^{d_{i}}\to\mathbb{R}^{d_{i}} , which can depend on all species, denoted \rho=[\rho_{1},\dots,\rho_{n}] . The mass of each \rho_{i} is conserved due to the divergence structure. We prove that monotonicity is a sufficient condition for dynamics of the form \displaystyle\partial_{t}\rho_{i}+\mathrm{div}\left(\rho_{i}v_{i}[\rho]\right)=0\quad\forall\,i\in[1,\dots,n]\,, (1.1) to contract in the joint Wasserstein-2 metric. We additionally study a particular choice of velocity fields which results in a system of coupled gradient flows. In the Euclidean setting, all norms are equivalent, whereas in infinite dimensions, the choice of metric determines the topology and the gradient flow trajectory. We focus on the Wasserstein-2 metric, which results in a system of partial differential equations (PDEs) where the evolution of each species is written as a gradient flow with respect to its own energy. Let \mathcal{P}_{2}(\mathbb{R}^{d}) be the space of probability measures with bounded second moments on \mathbb{R}^{d} , denote the product space \overline{\mathcal{P}_{2}}=\mathcal{P}_{2}(\mathbb{R}^{d_{1}})\times\cdots\times\mathcal{P}_{2}(\mathbb{R}^{d_{n}}) , and denote by \mathscr{W}_{2} the Wasserstein-2 metric (see Appendix A for details). Consider energy functionals F_{i}:\mathcal{A}\subseteq\overline{\mathcal{P}_{2}}\to\mathbb{R} , for a suitable choice of domain \mathcal{A}\subseteq\overline{\mathcal{P}_{2}} . Selecting v_{i}[\rho]=-\nabla_{x_{i}}\delta_{\rho_{i}}F_{i}[\rho] results in a system in which each species evolves in the direction of steepest descent of F_{i} in the Wasserstein-2 metric, given by \displaystyle\partial_{t}\rho_{i}=-\nabla_{\mathscr{W}_{2},\rho_{i}}F_{i}[\rho]=\mathrm{div}_{x_{i}}\left(\rho_{i}\nabla_{x_{i}}\delta_{\rho_{i}}F_{i}[\rho]\right)\quad\forall\,i\in[1,\dots,n]\,. (1.2) This system also has a game theoretic interpretation; in a game where each player i has a cost functional F_{i} which it seeks to minimize, the player i evolves a distribution \rho_{i}\in\mathcal{P}_{2}(\mathbb{R}^{d_{i}}) to minimize a cost functional F_{i}[\rho_{i},\rho_{-i}]=F_{i}[\rho]:\mathcal{A}\rightarrow\mathbb{R} , where \rho_{-i} denotes the non- i^{th} player measures. We show that monotonicity implies displacement convexity of each F_{i} ; however, displacement convexity of each F_{i} is not enough to ensure monotonicity because monotonicity also depends on the coupling structure among all the species. Additionally, under mild conditions, the steady state of ( 1.2 ) corresponds to the unique Nash equilibrium of the game characterized by (F_{i}) . A primary contribution of this work is the definition of monotonicity for velocity fields and energy functionals in this setting. Denote by \Gamma_{i}(\rho^{(0)}_{i},\rho^{(1)}_{i}) the set of couplings whose marginals are \rho^{(0)}_{i},\rho^{(1)}_{i}\in\mathcal{P}_{2}(\mathbb{R}^{d_{i}}) , and let \Gamma(\rho^{(0)},\rho^{(1)})\coloneqq\Gamma_{1}(\rho^{(0)}_{1},\rho^{(1)}_{1})\times\dots\times\Gamma_{n}(\rho^{(0)}_{n},\rho^{(1)}_{n}) . Let \Gamma^{*} be the set of optimal transport plans, where \Gamma_{i}^{*}=\{\gamma_{i}\,:\,\gamma_{i}\in\argmin_{\hat{\gamma}_{i}\in\Gamma_{i}(\rho^{(0)}_{i},\rho^{(1)}_{i})}\int\left\lVert x_{i}-y_{i}\right\rVert^{2}\mathsf{d}\hat{\gamma}_{i}(x_{i},y_{i}) . Let x=[x_{1},\dots,x_{n}]\in\mathbb{R}^{d} , with d=\sum_{i=1}^{n}d_{i} . Let \overline{\mathscr{W}_{2}} be the joint metric over all species, given by \overline{\mathscr{W}_{2}}(\rho,\mu)^{2}=\sum_{i=1}^{n}\mathscr{W}_{2}(\rho_{i},\mu_{i})^{2} . Definition 1.1 (Monotonicity) . Let \mathcal{A}\subseteq\overline{\mathcal{P}_{2}} be a geodesically convex set. If for any \rho^{(0)},\rho^{(1)}\in\mathcal{A} , \displaystyle-\int\left\langle x-y,\,v[\rho^{(0)}](x)-v[\rho^{(1)}](y)\right\rangle\mathsf{d}\gamma(x,y)\geq\lambda\int\left\lVert x-y\right\rVert^{2}\mathsf{d}\gamma(x,y)=\lambda\overline{\mathscr{W}_{2}}(\rho^{(0)},\rho^{(1)})^{2}\,, (1.3) for all \gamma\in\Gamma^{*}(\rho^{(0)},\rho^{(1)}) , then the velocity fields \{v_{i}:\mathcal{A}\to\mathbb{R}^{d_{i}}\}_{i=1}^{n} are \lambda -monotone in \mathcal{A} . Similarly, we say the energy functionals \{F_{i}:\mathcal{A}\to\mathbb{R}\}_{i=1}^{n} are \lambda -monotone in \mathcal{A} if v_{i}[\rho]=-\nabla_{x_{i}}\delta_{\rho_{i}}F_{i}[\rho] are \lambda -monotone. This definition of monotonicity is a multispecies version of the dissipativity condition given in [ 16 ] , for single-valued v instead of set-valued v . Some texts use the word accretive instead of monotone to describe operators which satisfy positive-definiteness over a Hilbert space [ 5 ] ; however, the space of measures endowed with the Wasserstein-2 metric is not a Hilbert space, and so one can think of our notion of monotonicity as an extension of the notion of accretivity. It was recently shown in [ 15 , Theorem 7.6] that if a velocity field is dissipative, the dissipation inequality holds for all admissible plans, not just optimal couplings. In particular, when the dissipation inequality ( 1.3 ) holds for all pairs of measures on \mathcal{P}_{2} , it holds for all optimal plans between a pair of measures if and only if it holds for all admissible plans between those two measures. We prove a similar type of result for the multispecies setting. Theorem 1.2 (Admissible Couplings) . If for any \rho^{(0)},\rho^{(1)}\in\overline{\mathcal{P}_{2}} and \gamma^{*}\in\Gamma^{*}(\rho^{(0)},\rho^{(1)}) it holds \displaystyle-\int\left\langle x-y,\,v[\rho^{(0)}](x)-v[\rho^{(1)}](y)\right\rangle\mathsf{d}\gamma^{*}(x,y)\geq\lambda\int\left\lVert x-y\right\rVert^{2}\mathsf{d}\gamma^{*}(x,y)\,, then for any \gamma\in\Gamma(\rho^{(0)},\rho^{(1)}) , \displaystyle-\int\left\langle x-y,\,v[\rho^{(0)}](x)-v[\rho^{(1)}](y)\right\rangle\mathsf{d}\gamma(x,y)\geq\lambda\int\left\lVert x-y\right\rVert^{2}\mathsf{d}\gamma(x,y)\,. For a proof of this theorem, and a counterexample of why this result does not hold when v is \lambda -monotone on an arbitrary geodesically convex subset A\subseteq\overline{\mathcal{P}_{2}} , see Appendix C . When the velocity field takes the form of a coupled gradient flow, we use the following notation for the row-wise gradient of the first variation \displaystyle\overline{\nabla\delta_{\rho}}F[\rho](x)\coloneqq\begin{pmatrix}\nabla_{x_{1}}\delta_{\rho_{1}}F_{1}[\rho](x_{1})\\ \vdots\\ \nabla_{x_{n}}\delta_{\rho_{n}}F_{n}[\rho](x_{n})\end{pmatrix}\,. Then \lambda -monotonicity for F=(F_{i}) writes as \displaystyle\int\left\langle x-y,\,\overline{\nabla\delta_{\rho}}F[\rho^{(0)}](x)-\overline{\nabla\delta_{\rho}}F[\rho^{(1)}](y)\right\rangle\mathsf{d}\gamma(x,y)\geq\lambda\overline{\mathscr{W}_{2}}(\rho^{(0)},\rho^{(1)})^{2}\,. (1.4) Theorem 1.3 (Convergence, informal) . Let v be \lambda -monotone in \mathcal{A} and \rho(t)\in\mathcal{A} evolve according to ( 1.1 ) for all t\geq 0 . Then (a) there exists a unique steady state \rho^{\infty} of ( 1.1 ), and \rho(t)

Read the complete text