Abstract
We rigorously derive pressureless Euler-type equations with nonlocal dissipative terms in velocity and aggregation equations with nonlocal velocity fields from Newton-type particle descriptions of swarming models with alignment interactions. Crucially, we make use of a discrete version of a modulated kinetic energy together with the bounded Lipschitz distance for measures in order to control terms in its time derivative due to the nonlocal interactions.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
In this work, we analyse the evolution of an indistinguishable N-point particle system given by
subject to the initial data
Here \(x_i = x_i(t) \in {\mathbb {R}}^d\) and \(v_i = v_i(t) \in {\mathbb {R}}^d\) denote the position and velocity of i-particle at time t, respectively. The coefficient \(\gamma \geqq 0\) represents the strength of linear damping in velocity, \(\varepsilon _N>0\) the strength of inertia, \(V:{\mathbb {R}}^d \rightarrow {\mathbb {R}}_+\) and \(W : {\mathbb {R}}^d \rightarrow {\mathbb {R}}\) stand for the confinement and interaction potentials, respectively. \(\psi : {\mathbb {R}}^d \rightarrow {\mathbb {R}}_+\) is a communication weight function. Throughout this paper, we assume that W and \(\psi \) satisfy \(W(x) = W(-x)\) and \(\psi (x) = \psi (-x)\) for \(x \in {\mathbb {R}}^d\). They include basic particle models for collective behaviors, see [12, 20, 25, 34, 36, 46, 47, 63] and the references therein.
Our main goal is to derive the macroscopic collective models rigorously governing the evolution of the particle system (1.1) as the number of particles goes to infinity. On one hand, we will derive hydrodynamic Euler-alignment models given by
in the mean-field limit: when initial particles are close to a monokinetic distribution \(\rho _0(x) \delta _{u_0(x)}(v)\) in certain sense and \(\varepsilon _N=O(1)\) as \(N\rightarrow \infty \). On the other hand, we will show that the particle system can be described by aggregation equations of the form
where
in the combined mean-field/small inertia limit when initial particles are close to a monokinetic distribution \(\rho _0(x) \delta _{u_0(x)}(v)\), \(\gamma >0\) and \(\varepsilon _N \rightarrow 0\) as \(N \rightarrow \infty \). For simplicity of notations when dealing with the mean-field limit, we will take \(\varepsilon _N=1\) in the sequel.
1.1 Mean-field limits: from particles to continuum
As the number of particles N tends to infinity, microscopic descriptions given by the particle system (1.1) become more and more computationally unbearable. Reducing the complexity of the system is of paramount importance in any practical application. The classical multiscale strategy in kinetic modelling is to introduce the number density function \(f = f(x,v,t)\) in phase space \((x,v) \in {\mathbb {R}}^d \times {\mathbb {R}}^d\) at time \(t \in {\mathbb {R}}_+\) and study the time evolution of that density function. Then at the formal level, we can derive the following Vlasov-type equation from the particle system (1.1) as \(N \rightarrow \infty \):
where \(\rho _f = \rho _f(x,t)\) is the local particle density and \(F_a(f) = F_a(f)(x,v,t)\) represents a nonlocal velocity alignment force given by
and
respectively. Let us briefly recall the reader the basic formalism leading to the kinetic equation (1.6) as the limiting system of (1.1). We first define the empirical measure \(\mu ^N\) associated to a solution to the particle system (1.1), that is,
As long as there exists a solution to (1.1), the empirical measure \(\mu ^N\) satisfies (1.6) in the sense of distributions. To be more specific, for any \(\varphi \in {\mathcal {C}}^1_0({\mathbb {R}}^d \times {\mathbb {R}}^d)\), we get
Notice that the particle velocity can also be rewritten in terms of the empirical measure \(\mu ^N\) as
This implies that the right-hand side of (1.7) can also be written in terms of the empirical measure \(\mu ^N\) as
This concludes that \(\mu ^N\) is a solution to (1.6) in the sense of distributions as long as particle paths are well defined. In fact, if the interaction potential W and the communication weight function \(\psi \) in the classical Cucker–Smale alignment model are regular enough, for instance, bounded Lipschitz regularity, then the global-in-time existence of measure-valued solutions can be obtained by establishing a weak-weak stability estimate for the empirical measure, see [46, Section 5] for more details. The mean-field limit has attracted lots of attention in the last years in different settings depending on the regularity of the involved potentials V, W and communication function \(\psi \). Different approaches to the derivation of the Vlasov-like kinetic equations with alignments/interaction terms or the aggregation equations have been taken leading to a very lively interaction between different communities of researchers in analysis and probability. We refer to [3, 4, 10, 20, 30, 31, 35, 44, 47, 50, 54,55,56, 64, 67] for the classical references and non-Lipschitz regularity velocity fields in kinetic cases, to [48, 49] for very related incompressible fluid problems, and to [7, 9, 16, 17, 37, 43, 45, 51, 52, 61, 63, 65, 66] for results with more emphasis on the singular interaction kernels both at the kinetic and the aggregation-diffusion equation cases.
1.2 Local balanced laws, the mono-kinetic ansatz, and the small inertia limit
The classical procedure in kinetic theory of deriving equations for the first 3 moments of the distribution function f leads to the standard problem of how to close the moment system since the equation for the second moment will depend on higher order moments. Suitable closure assumptions are not known so far even in cases where noise/diffusion is added to the system. However, at the formal level, we can take into account the mono-kinetic ansatz for f, as done in [18, 21], leading to
where \(\rho \) and u are the macroscopic density and the mean velocity of particles, that is, the first two moments of f in velocity variable
It is standard to check that the strain tensor and heat flux become zero and the moment system closes becoming the pressureless Euler equations with nonlocal interaction forces (1.3):
and
on the support of \(\rho \). The last equation coming from the closed equation on the evolution of the second moment is redundant but it gives a nice information about the total energy of the system. Although the monokinetic assumption is not fully rigorously justified and it does not have a direct physical motivation, it is observed by particle simulations that the derived hydrodynamic system shares some qualitative behavior with the particle system, see [12, 18, 20,21,22, 33]. Note that (1.3) conserves only the total mass in time in this generality. However, the total free energy is dissipated due to the linear damping and the velocity alignment force as pointed out in [19] for weak solutions of this system. The hydrodynamic system (1.9) has a rich variety of phenomena compared to the plain pressureless Euler system. This fact is due to the competition between attraction/repulsion and alignment leading to sharp thresholds for the global existence of strong solutions versus finite time blow-up and decay to equilibrium, see [13,14,15, 26, 63, 68]. We emphasize that the additional alignment, linear damping and attraction/repulsion terms can promote the existence of global solutions depending on the intial data. We will show that these hydrodynamical solutions can be obtained directly from particle descriptions as long as they exist, so their physical relevance is dictated by the time of existence of these solutions.
It is worth noticing as in [18] that the mono-kinetic ansatz for f is a measure-valued solution of the kinetic equation (1.6). More precisely, one can show that \(\rho (x,t) \delta _{u(x,t)}(v)\) is a solution to the kinetic equation (1.6) in the sense of distributions as long as \((\rho ,u)(x,t)\) is a strong solution to the hydrodynamic equations (1.3). Indeed, for any \(\varphi \in {\mathcal {C}}^1_0({\mathbb {R}}^d \times {\mathbb {R}}^d)\), we obtain
Using the continuity equation in (1.3), \(I_1\) can be easily rewritten as
By multiplying the velocity equation in (1.3) by \(\rho \) and using \((\nabla _v \varphi ) (x, u(x,t))\) as a test function to the resulting equation yields
Then similarly as before, we can rewrite the second and third terms on the right hand side of the equality by using the mono-kinetic ansatz (1.8). This implies
Combining all of the above estimates yields
This shows that \(\rho (x,t) \delta _{u(x,t)}(v)\) satisfies the kinetic equation (1.6) in the sense of distributions.
Finally, we will be also dealing with the small inertia limit for both the kinetic equation (1.6) and the hydrodynamic system (1.3) combined with the mean field limit. In the small inertia asymptotic limit, we want to describe the behavior of the scaled kinetic equation
and the scaled hydrodynamic system
in the limit of small inertia \(\varepsilon \rightarrow 0\). At the formal level, the equations (1.11) will be replaced by (1.4)–(1.5) as \(\varepsilon \rightarrow 0\). The limiting nonlinearly coupled aggregation equations (1.4)–(1.5) have been recently studied in [39, 40]. Several authors have studied particular choices of interactions V, W and comunication functions \(\psi \) for some of the connecting asymptotic limits from the kinetic description (1.10) with/without noise to the hydrodynamic system (1.11) in [8, 11, 42, 57], from the hydrodynamic system (1.11) to the aggregation equation (1.4)–(1.5) in [23, 59, 60], and for the direct limit from the kinetic equation to the aggregation equation (1.4)–(1.5) in [8, 53].
1.3 Purpose, mathematical tools and main novelties
Summarizing the main facts of the mean-field limit and the monokinetic ansatz in Sections 1.1 and 1.2, both the empirical measure \(\mu ^N(t)\) associated to the particle system (1.1) and the monokinetic solutions \(\rho (x,t) \delta _{u(x,t)}(v)\), with \((\rho ,u)(x,t)\) satisfying the hydrodynamic equations (1.3) in the strong sense, are distributional solutions of the same kinetic equation (1.6). In order to analyse the convergence of the empirical measure \(\mu ^N\) to \(\rho (x,t) \delta _{u(x,t)}(v)\), the goal is to establish a weak-strong stability estimate where the strong role is played by the distributional solution \(\rho (x,t) \delta _{u(x,t)}(v)\) associated to the strong solution of the hydrodynamic system (1.3). Our main goal is then to quantify the following convergence
in the sense of distributions for both the mean-field and the combined mean-field/small inertia limit for well prepared initial data. Our main mathematical tools are the use of a modulated kinetic energy combined with the bounded Lipschitz distance in order to control terms between the discrete particle system and the hydrodynamic quantities. Let us first introduce the modulated kinetic energy as
where f is a solution of kinetic equation (1.6) and u is the velocity field as part of the solution of the pressureless Euler equations (1.3). Modulated kinetic energies were used in conjunction with relative potential energy terms for quasineutral limits of Vlasov like equations [5, 6, 62] for instance. We would like to emphasize that the quantity (1.12) gives a sharper estimate compared to the classical modulated macroscopic energy. Indeed, the macro energy of the system (1.3) is given by
Thus its modulated energy, also often refereed to as relative energy, can be defined as
A straightforward computation gives
On the other hand, by Hölder inequality, we easily find that
This yields
In fact, we can easily show that
This shows that the convergence of the modulated kinetic energy (1.12) implies the convergence of the modulated macro energy (1.13). We notice that if f is a monokinetic distribution, \( f(x,v,t) = \rho _f(x,t) \delta _{u_f(x,t)}(v), \) then the second term on the right hand side of (1.14) becomes zero, and the two modulated energies (1.12) and (1.13) coincide. For notational simplicity, we denote by \(\mathcal {Z}^N(t) = \{(x_i(t), v_i(t))\}_{i=1}^N\) the set of trajectories associated to the particle system (1.1). Then let us define the first important quantity that will allow us to quantify the distance between particles (1.1) and hydrodynamics (1.3), it is just the discrete version of the modulated kinetic energy (1.12) defined as
The second quantity that will allow us our quantification goal combined with the discrete modulated energy (1.15) is a classical distance between probability measures, the bounded Lipschitz distance, used already by the pioneers in kinetic theory [4, 64, 67] in the early works for the mean-field limit. Notice that the pressureless Euler system (1.3) includes the nonlocal position and velocity interaction and alignment forces. Furthermore, its relative energy/entropy has no strict convexity in terms of density variable due to the lack of pressure term. In order to overcome these difficulties, ideas of combining the modulated macro energy and the first or second order Wasserstein distance have been recently proposed in [8, 11, 32] quantifying the hydrodynamic limit from kinetic equation to the pressureless Euler type system. More recently, in [24], a general theory providing some relation between a modulated macro energy-type function and p-Wasserstein distance is also developed. In particular, in [24, Proposition 3.1], it is discussed that the p-Wasserstein distance with \(p\in [1,2]\) can be controlled by the modulated macro energy functional.
In the present work, we will employ the bounded Lipschitz distance to provide stability estimates between the empirical particle density \(\rho ^N\) defined as
with \(\mu ^N_t\) be the empirical measure associated to the particle system (1.1), and the hydrodynamic particle density \(\rho \) solution to (1.3). More precisely, let \({\mathcal {M}}({\mathbb {R}}^d)\) be the space of signed Radon measures on \({\mathbb {R}}^d\), which can be considered as nonnegative bounded linear functionals on \({\mathcal {C}}_0({\mathbb {R}}^d)\). Let \(\mu , \nu \in {\mathcal {M}}({\mathbb {R}}^d)\) be two Radon measures. Then the bounded Lipschitz distance, which is denoted by \(d_{BL}: {\mathcal {M}}({\mathbb {R}}^d) \times {\mathcal {M}}({\mathbb {R}}^d) \rightarrow {\mathbb {R}}_+\), between \(\mu \) and \(\nu \) is defined by
where the admissible set \(\Omega \) of test functions are given by
We also denote by \(Lip({\mathbb {R}}^d)\) the set of Lipschitz functions on \({\mathbb {R}}^d\). In Proposition 2.2 below, we provide a relation between the bounded Lispchitz distance and the discrete version of the modulated kinetic energy (1.15). This key observation allows us to overcome the difficulties mentioned above.
1.4 Main results and Plan of the paper
We will first assume that the particle system (1.1), the pressureless Euler-type equations (1.3), and the aggregation equations (1.4)–(1.5) have existence of smooth enough solutions up to a fixed time \(T>0\). We postpone further discussion at the end of this subsection, although we make precise now the assumptions needed on these solutions for our main results.
Our first main result shows the rigorous passage from Newton’s equation (1.1) to pressureless Euler equations (1.3) via the mean-field limit as \(N \rightarrow \infty \).
Theorem 1.1
Let \(T > 0\), \(\mathcal {Z}^N(t) = \{(x_i(t), v_i(t))\}_{i=1}^N\) be a solution to the particle system (1.1), and let \((\rho , u)\) be the unique classical solution of the pressureless Euler system with nonlocal interaction forces (1.3) satisfying \(\rho >0\) on \({\mathbb {R}}^d \times [0,T)\), \(\rho \in {\mathcal {C}}([0,T];{\mathcal {P}}({\mathbb {R}}^d))\) and \(u \in L^\infty (0,T;\mathcal {W}^{1,\infty }({\mathbb {R}}^d))\) up to time \(T>0\) with initial data \((\rho _0, u_0)\). Suppose that the interaction potential W and the communication weight function \(\psi \) satisfy \(\nabla _x W \in \mathcal {W}^{1,\infty }({\mathbb {R}}^d)\) and \(\psi \in \mathcal {W}^{1,\infty }({\mathbb {R}}^d)\), respectively. If the initial data for (1.1) and (1.3) are chosen such that
then we have
as \(N \rightarrow \infty \). In fact, we have the following quantitative bound estimate:
where \(C>0\) only depends on \(\Vert u\Vert _{L^\infty (0,T;\mathcal {W}^{1,\infty })}\), \(\Vert \psi \Vert _{\mathcal {W}^{1,\infty }}\), \(\Vert \nabla _x W\Vert _{\mathcal {W}^{1,\infty }}\), and T.
The main novelty of this first result resides in how to control the alignment terms via the modulated energy combined with the bounded Lipschitz distance.
Remark 1.1
(Singular repulsive interaction) The previous result also applies to singular repulsive interaction potentials. In particular, it holds for the Coulomb interaction potential on \({\mathbb {R}}^d\) given by
and for Riesz potentials in a sense to be specified in Section 2.3. Here \(\alpha _d\) denotes the volume of the unit ball in \({\mathbb {R}}^d\). In order to deal with the singularity on the interaction potential, the diagonal term should be eliminated in the modulated energy functional. This has been recently solved in the recent breakthrough result in [66] by introducing a different relative potential energy avoiding the diagonal terms. The details for singular interaction potentials cases are postponed to Section 2.3, see Theorem 2.1.
Section 2 is devoted to the proof of Theorem 1.1 and the generalization to singular repulsive potentials using [66] in its last subsection.
Our second main result is devoted to the asymptotic analysis for the particle system (1.1) under the small inertia regime: \(\varepsilon _N \rightarrow 0\) as \(N \rightarrow \infty \). By Theorem 1.1, we expect that for sufficiently large \(N\gg 1\), the system (1.1) in the mean-field/small inertia limit can be well approximated by
At the formal level, since \(\varepsilon _N \rightarrow 0\) as \(N \rightarrow \infty \), it follows from the momentum equations in the above system that the hydrodynamic system (1.3) should be replaced by (1.4)–(1.5) as \(N \rightarrow \infty \). In order to apply our strategy above, we rewrite the equations (1.4)–(1.5) as
where \({\bar{e}} := \partial _t {\bar{u}} + {\bar{u}} \cdot \nabla _x {\bar{u}}\).
We can now state our second main result related to a weak-strong stability estimate in the combined mean-field/small inertia limit.
Theorem 1.2
Let \(T>0\) and \(d \geqq 1\). Let \(\mathcal {Z}^N(t) = \{(x_i(t), v_i(t))\}_{i=1}^N\) be a solution to the particle system (1.1), and let \(({\bar{\rho }}, {\bar{u}})\) be the unique classical solution of the aggregation-type equation (1.4)–(1.5) satisfying \({\bar{\rho }} \in {\mathcal {C}}([0,T];{\mathcal {P}}({\mathbb {R}}^d))\) and \({\bar{\rho }} > 0\) on \({\mathbb {R}}^d \times [0,T)\), \({\bar{u}} \in L^\infty (0,T; \mathcal {W}^{1,\infty }({\mathbb {R}}^d))\) and \(\partial _t {\bar{u}} \in L^\infty ({\mathbb {R}}^d \times (0,T))\) up to time \(T>0\) with the initial data \({\bar{\rho }}_0\). Suppose that the interaction potential W and the communication weight function \(\psi \) satisfy \(\nabla _x W \in \mathcal {W}^{1,\infty }({\mathbb {R}}^d)\) and \(\psi \in \mathcal {W}^{1,\infty }({\mathbb {R}}^d)\), respectively, and the strength of damping \(\gamma >0\) is large enough. If the initial data for (1.1) and (1.4) are chosen such that
then we have
and
as \(N \rightarrow \infty \) (and thus \(\varepsilon _N \rightarrow 0\)). In fact, we have the following quantitative bound estimate:
and
for all \(t \in [0,T]\), where \(C>0\) is independent of both \(\varepsilon _N\) and N but depending on \(\Vert {\bar{u}}\Vert _{L^\infty (0,T;\mathcal {W}^{1,\infty })}\), \(\Vert \partial _t {\bar{u}}\Vert _{L^\infty }\), \(\Vert \nabla _x W\Vert _{\mathcal {W}^{1,\infty }}\), \(\Vert \psi \Vert _{\mathcal {W}^{1,\infty }}\), and \(\gamma \).
Remark 1.2
Theorem 1.2 implies that if the initial data satisfies
for some \(C_0>0\) which is independent of both \(\varepsilon _N\) and N, then we have
and
for all \(t \in [0,T]\), where \(C>0\) is independent of both \(\varepsilon _N\) and N. This further yields that the convergences (1.17) and (1.18) hold in weakly in \(L^\infty (0,T;{\mathcal {M}}({\mathbb {R}}^d))\) and \(L^\infty (0,T;{\mathcal {M}}({\mathbb {R}}^d \times {\mathbb {R}}^d))\), respectively.
Remark 1.3
If \(V\equiv 0\) and \(\gamma >0\) is sufficiently large, then we can check that \(\Vert {\bar{u}}\Vert _{L^\infty (0,T;\mathcal {W}^{1,\infty })}\) and \(\Vert \partial _t {\bar{u}}\Vert _{L^\infty }\) can be bounded from above by some constant, which depends only on \(\Vert \nabla _x W\Vert _{\mathcal {W}^{1,\infty }}\), \(\Vert \psi \Vert _{\mathcal {W}^{1,\infty }}\), \(\Vert {\bar{\rho }}\Vert _{L^\infty (0,T;L^1)}\), and \(\gamma \). We refer to [24] for details. For general confinement potentials, we can also deal with general strong solutions for compactly supported initial data since their support remains compact for all times. We refer to [1, 15] for particular instances of these results.
Remark 1.4
One may follow a similar argument as in [40, Theorem 2.4] to have the existence and uniqueness of classical solutions \(({\bar{\rho }}, {\bar{u}})\) to the equations (1.4)–(1.5) satisfying the regularity properties and assumptions of Theorem 1.2. For the Coulomb or Riesz interaction, an idea of proof proposed in [28] would be employed to establish the local-in-time existence and uniqueness of classical solutions to the equations (1.4)–(1.5) without the confinement potential.
Section 3 is devoted to the proof of Theorem 1.2 and the generalizations to singular repulsive potentials. Finally, we complement these results by showing the existence of solutions to the particle system (1.1) in Appendix A, and the existence and uniqueness of classical solutions stated in Theorem 1.1 for the hydrodynamic system (1.3) in Section 4.
2 Mean-Field Limit: From Newton to Pressureless Euler
In this section, we provide the details of the proof for Theorem 1.1. As mentioned before, one of our main mathematical tools is the discrete version of the modulated kinetic energy \(\mathcal {E}^N (\mathcal {Z}^N(t) | U(t))\) defined in (1.15).
2.1 Modulated kinetic energy estimate
In this part, our main purpose is to give the quantitative bound estimate of the discrete modulated kinetic energy \(\mathcal {E}^N (\mathcal {Z}^N(t) | U(t))\).
Proposition 2.1
Let \(T > 0\), \(\mathcal {Z}^N(t) = \{(x_i(t), v_i(t))\}_{i=1}^N\) be a solution to the particle system (1.1), and let \((\rho , u)\) be the unique classical solution of the pressureless Euler system with nonlocal interaction forces (1.3) under the assumptions of Theorem 1.1 up to time \(T>0\). Suppose that the interaction potential W and the communication weight function \(\psi \) satisfy \(\nabla _x W \in \mathcal {W}^{1,\infty }({\mathbb {R}}^d)\) and \(\psi \in \mathcal {W}^{1,\infty }({\mathbb {R}}^d)\), respectively. Then we have
where \(C>0\) is independent of N and \(\gamma \).
Proof
By the notion of our classical solution, we obtain from the momentum equation in (1.3) that
Then using this and (1.1), we estimate the discrete modulated kinetic energy functional as
where
Here \(I_1\) can be easily estimated as
By definition, we obtain \( I_2 = -2\gamma \mathcal {E}^N (\mathcal {Z}^N(t) | U(t)). \) We next estimate \(I_3\) as
On the other hand, the fact \(\nabla _x W \in \mathcal {W}^{1,\infty }\) gives
and subsequently this asserts
For the estimate of \(I_4\), we note that
Then we rewrite \(J_2\) as
This yields
Here we can easily estimate \(I_4^1\) as
Note that
that is,
On the other hand, we can estimate
where
Similarly, we also find that
Combining all of the above estimates, we have
This completes the proof. \(\quad \square \)
Remark 2.1
We assumed that the communication weight \(\psi \) is nonnegative, which takes into account the velocity alignment forces, however a similar bound estimate for the discrete kinetic energy \(\mathcal {E}^N\) to that in Proposition 2.1 can be obtained. Indeed, if \(\psi \) can be negative, but bounded, then the third term on the left hand side of (2.1) can be estimated as
This yields
where \(C>0\) is independent of N and \(\gamma \).
In order to close the estimate in Proposition 2.1, we need to estimate the bounded Lipschitz distance between \(\rho ^N\) and \(\rho \). In the proposition below, we provide the relation between the bounded Lipschitz distance and the discrete modulated kinetic energy.
Proposition 2.2
Let \(\rho ^N\) and \(\rho \) be defined as above. Then we have
where \(C > 0\) depends only on \(\Vert u\Vert _{L^\infty (0,T;Lip)}\) and T.
Proof
Consider a forward characteristics \(\eta =\eta (x,t)\) for the system (1.3) satisfying the following ODEs:
subject to the initial data: \( \eta (x,0) = x \in {\mathbb {R}}^d. \) The characteristic \(\eta \) is well-defined because of the Lipschitz continuous regularity of u. Note that along the characteristic, the solution \(\rho \) can be written in the mild form
and thus we get
This together with using the change of variables yields
for \(\phi \in \mathcal {W}^{1,\infty }({\mathbb {R}}^d)\). Moreover, we find from (2.3) that
and applying Grönwall’s lemma to the above gives
where \(C>0\) depends only on \(\Vert u\Vert _{L^\infty (0,T;Lip)}\) and T, that is, \(\eta \) is Lipschitz continuous in \({\mathbb {R}}^d\). We also get
Here the second term on the right hand side of the above inequality can be estimated as
Thus we get
and applying Grönwall’s lemma to the above deduces
where C depends only on \(\Vert u\Vert _{L^\infty (0,T;Lip)}\) and T. In particular, by taking \(x = x_i(0)\), we get
Then for any \(\phi \in \mathcal {W}^{1,\infty }({\mathbb {R}}^d)\) we use (2.4) to estimate
For \(L_1\), we use the Lipschitz continuity together with (2.6) to obtain
For the estimate of \(L_2\), we notice that
Using this identity, the Lipschitz estimate for \(\eta \) in (2.5), and the fact \(\phi \in \mathcal {W}^{1,\infty }({\mathbb {R}}^d)\), we find
Putting (2.8) and (2.9) into (2.7) yields
for \(0 \leqq t \leqq T\), where \(C > 0\) depends only on \(\Vert u\Vert _{L^\infty (0,T;Lip)}\) and T. \(\quad \square \)
2.2 Proof of Theorem 1.1
2.2.1 Quantitative bound estimates
Applying Grönwall’s lemma and Young’s inequality to the differential inequality in Proposition 2.1 yields
where \(C > 0\) is independent of N. We then use Proposition 2.2 to have
We finally apply Grönwall’s to the above to conclude the desired result.
2.2.2 Convergence estimates
For the convergence estimates, it suffices to prove the following lemma:
Lemma 2.1
-
(i)
Convergence of local moment:
$$\begin{aligned} d_{BL}\left( \int _{{\mathbb {R}}^d}v \,\mu ^N(\mathrm{d}v), \, \rho u\right) \leqq \left( \iint _{{\mathbb {R}}^d \times {\mathbb {R}}^d}|v - u(x) |^2\mu ^N(\mathrm{d}x\mathrm{d}v) \right) ^{1/2} + Cd_{BL}(\rho ^N, \rho ). \end{aligned}$$ -
(ii)
Convergence of local energy:
$$\begin{aligned} \begin{aligned}&d_{BL}\left( \int _{{\mathbb {R}}^d}(v \otimes v)\, \mu ^N(\mathrm{d}v), \,\rho u \otimes u\right) \\&\quad \leqq \iint _{{\mathbb {R}}^d \times {\mathbb {R}}^d}|v - u(x) |^2\mu ^N(\mathrm{d}x\mathrm{d}v) + C\left( \iint _{{\mathbb {R}}^d \times {\mathbb {R}}^d}|v - u(x) |^2\mu ^N(\mathrm{d}x\mathrm{d}v) \right) ^{1/2} \\&\quad + Cd_{BL}(\rho ^N, \rho ). \end{aligned} \end{aligned}$$ -
(iii)
Convergence of empirical measure:
$$\begin{aligned} d^2_{BL}(\mu ^N, \rho \delta _{u}) \leqq C\iint _{{\mathbb {R}}^d \times {\mathbb {R}}^d}|v - u(x) |^2\,\mu ^N(\mathrm{d}x\mathrm{d}v) + Cd^2_{BL}(\rho ^N, \rho ). \end{aligned}$$
Here \(C>0\) is independent of N.
Proof
(i) For any \(\phi \in \mathcal {W}^{1,\infty }({\mathbb {R}}^d)\), we get
(ii) Adding and subtracting, we notice that
This yields for \(\phi \in \mathcal {W}^{1,\infty }({\mathbb {R}}^d)\)
(iii) For any \(\varphi \in \mathcal {W}^{1,\infty }({\mathbb {R}}^d \times {\mathbb {R}}^d)\), we find that
\(\square \)
2.3 Singular interaction potential cases: Coulomb and Riesz potentials
In this part, we discuss the singular interaction potentials. Let \(d \geqq 1\) and consider a potential \(\widetilde{W}\) has the form
or
Note that the case \(\alpha =d-2\) with \(d \geqq 3\) or (2.11) with \(d=2\) corresponds to the Coulomb potential, and the other cases are called Riesz potentials. With these types of singular potentials, in a recent work [66], the quantitative mean-field limit from the particle system (1.1) to the pressureless Euler-type system when \(\gamma = 0\), \(V\equiv 0\) and \(\psi \equiv 0\). More precisely, in [66], the following modulated free energy is employed to measure the error between particle and continuum systems:
where \(\Delta \) denotes the diagonal in \({\mathbb {R}}^d \times {\mathbb {R}}^d\).
Theorem 2.1
Let \(T > 0\) and \(\mathcal {Z}^N(t) = \{(x_i(t), v_i(t))\}_{i=1}^N\) be a solution to the particle system (1.1), and let \((\rho , u)\) be the unique classical solution of the pressureless Euler system (1.3) with nonlocal interaction forces \(\widetilde{W}\), which is appeared in (2.10) or (2.11), instead of W up to time \(T>0\) with initial data \((\rho _0, u_0)\). Suppose that the communication weight function \(\psi \) satisfies \(\psi \in \mathcal {W}^{1,\infty }({\mathbb {R}}^d)\). Assume that the classical solution \((\rho ,u)\) satisfies \(\rho \in L^\infty (0,T;({\mathcal {P}} \cap L^\infty )({\mathbb {R}}^d))\) and \(u \in L^\infty (0,T;\mathcal {W}^{1,\infty }({\mathbb {R}}^d))\). In the case \(\alpha \geqq d-1\), we further assume that \(\rho \in L^\infty (0,T;{\mathcal {C}}^\sigma ({\mathbb {R}}^d))\) for some \(\sigma > \alpha -d+1\). Then there exists \(\beta < 2\) such that
where \(C>0\) is independent of N.
Remark 2.2
If the interaction potential W is singular at the origin, then the term related to W in (1.1) should be replaced by \(\frac{1}{N} \sum _{j:j \ne i} \nabla _x W (x_i - x_j)\) since W(0) can not be well defined. This is why the diagonal \(\Delta \) is excluded in the integration in the modulated potential energy.
Remark 2.3
If the right hand side of (2.12) converges to zero as \(N \rightarrow \infty \), then we also have the same convergence estimates in Theorem 1.1.
Remark 2.4
Our quantified mean-field limit estimate from (1.1) to (1.3) also apply with a simple combination of Theorems 1.1 and 2.1 for interaction potentials of the form \({\overline{W}} := W + \widetilde{W}\) with W satisfying \(\nabla W \in \mathcal {W}^{1,\infty }({\mathbb {R}}^d)\) and \(\widetilde{W}\) appeared in (2.10) or (2.11).
Proof of Theorem 2.1
For the proof, we only need to reestimate \(I_3\) term in the proof of Proposition 2.1. Although this proof is almost the same with that of [66], we provide the details here for the completeness of our work. Let us denote by
On the other hand, we find that
Here we used
This implies
We next use (2.13) to get
and
Thus we obtain
This together with the estimates in Proposition 2.1 yields
We then apply [66, Proposition 1.1] to have that the last term on the right hand side of the above inequality can be bounded from above by
for some \(\beta < 2\), where \(C>0\) is independent of N. Applying the Grönwall’s lemma to the resulting inequality concludes the desired quantitative bound estimate. The convergence result can be directly obtained by using Lemma 2.1. This completes the proof. \(\quad \square \)
3 Combined Small Inertia & Mean Field Limits: From Newton to Aggregation
3.1 Proof of Theorem 1.2
We first start with the case of smooth interaction potentials as in previous section and apply a similar strategy to the proof of Proposition 2.1 to the system (1.16). Then we get
where \({\bar{I}}_i, i=1,2,3,4\) are the terms \(I_i,i=1,2,3,4\) in (2.2) with replacing \((\rho ,u)\) by \(({\bar{\rho }}, {\bar{u}})\), and \({\bar{I}}_5\) is given by
where \({\bar{e}} = \partial _t {\bar{u}} + {\bar{u}} \cdot \nabla _x {\bar{u}}\). This can be simply estimated as
where \(C>0\) depends only \(\Vert {\bar{e}}\Vert _{L^\infty }\), independent of N and \(\varepsilon _N\). For the rest, we employ almost the same arguments as before to have
where \(C>0\) is independent of N, \(\varepsilon _N\), and \(\gamma > 0\). This yields
where \(C>0\) is independent of N, \(\varepsilon _N\), and \(\gamma > 0\). On the other hand, by Proposition 2.2, we can bound the first term on the right hand side of the above inequality from above by
where \(C>0\) is independent of N, \(\varepsilon _N\), and \(\gamma > 0\). This together with integrating (3.1) in time implies
We finally apply Grönwall’s lemma to conclude the desired result in Theorem 1.2.
3.2 Singular interaction potential cases
Similarly as before, Theorem 1.2 can be also easily extended to the case with Coulomb or Riesz potentials \(\widetilde{W}\) defined in (2.10) or (2.11). More specifically, we have the following theorem.
Theorem 3.1
Let \(T>0\) and \(\mathcal {Z}^N(t) = \{(x_i(t), v_i(t))\}_{i=1}^N\) be a solution to the particle system (1.1), and let \(({\bar{\rho }}, {\bar{u}})\) be the unique classical solution of the aggregation-type equation (1.4)–(1.5) with \(\widetilde{W}\), which is appeared in (2.10) or (2.11), instead of W, under the assumptions of Theorem 1.2 up to time \(T>0\) with the initial data \({\bar{\rho }}_0\). Suppose that the strength of damping \(\gamma >0\) is large enough and \(({\bar{\rho }},{\bar{u}})\) satisfies \({\bar{\rho }} \in L^\infty ({\mathbb {R}}^d \times (0,T))\). We further assume that \({\bar{\rho }} \in L^\infty (0,T;{\mathcal {C}}^\sigma ({\mathbb {R}}^d))\) for some \(\sigma > \alpha -d+1\) in the case \(s \geqq d-1\). Then there exists \(\beta < 2\) such that
and
for all \(t \in [0,T]\), where \(C>0\) is independent of \(\varepsilon _N\) and N. In particular if
and
for some \(C>0\) which is independent of \(\varepsilon _N\), then we have
and
for all \(t \in [0,T]\), where \(C>0\) is independent of \(\varepsilon _N\) and N.
4 Local Cauchy Problem for Pressureless Euler Equations with Nonlocal Forces
In order to make the analysis for the mean-field limit from the particle system (1.1) to the pressureless Euler-type equations (1.3) fully rigorous, we need to have the existence of solutions for both systems. As mentioned in Introduction, we postpone the existence theory for the particle system (1.1) in Appendix A, and here we provide local-in-time existence and uniqueness of classical solutions for the system (1.3). For the reader’s convenience, let us recall our limiting system:
with the initial data
Here we set the coefficient of linear damping \(\gamma =1\).
For the one dimensional problem, the well-posedness and singularity formation for the system (4.1) without the linear damping, the confinement and interaction potentials, called Euler-alignment system, are discussed in [13]. To be more precise, the sharp critical threshold which distinguishes the global-in-time regularity of classical solutions and finite-time breakdown of smoothness is analyzed. The sharp critical threshold estimate is also obtained in [15] for the pressureless damped Euler–Poisson system with quadratic confinement potential in one dimension, that is the system (4.1) with replacing W by \({\mathcal {N}}\), \(V = |x|^2/2\), and \(\psi \equiv 0\). For the pressureless Euler–Poisson system, the critical threshold is also discussed in [2, 38], see also [69] for the case with pressure. More recently, in [27], the local-in-time existence of classical solutions and finite-time singularity formation are taken into account.
We introduce the exact notion of strong solution to the system (4.1) that we will deal with.
Definition 4.1
Let \(s > d/2+1\). For given \(T\in (0,\infty )\), the pair \((\rho ,u)\) is a strong solution of (4.1) on the time interval [0, T] if and only if the following conditions are satisfied:
-
(i)
\(\rho \in {\mathcal {C}}([0,T];H^s({\mathbb {R}}^d))\), \(u \in {\mathcal {C}}([0,T];Lip({\mathbb {R}}^d)\cap L^2_{loc}({\mathbb {R}}^d))\), and \(\nabla _x^2 u \in {\mathcal {C}}([0,T];H^{s-1}({\mathbb {R}}^d))\),
-
(ii)
\((\rho , u)\) satisfy the system (4.1) in the sense of distributions.
Notice that due to the choice of s in the previous definition, these strong solutions are also classical solutions to (4.1). Our main result of this section is the following local Cauchy problem for the system (4.1).
Theorem 4.1
Let \(s > d/2+1\) and \(R>0\). Suppose that the confinement potential V is given by \(V = |x|^2/2\), the interaction potential \(\nabla _x W \in (\mathcal {W}^{1,1} \cap \mathcal {W}^{1,\infty })({\mathbb {R}}^d)\), and the communication weight function \(\psi \) satisfies
where \(B(0,R) \subset {\mathbb {R}}^d\) denotes a ball of radius R centered at the origin. For any \(N<M\), there is a positive constant \(T^*\) depending only on R, N, and M such that if \(\rho _0 > 0\) on \({\mathbb {R}}^d\) and
then the Cauchy problem (4.1) has a unique strong solution \((\rho ,u)\), in the sense of Definition 4.1, satisfying
Remark 4.1
The assumption on the communication weight function (4.2) implies \( \psi \in \mathcal {W}^{1,p}({\mathbb {R}}^d) \) for any \(p \in [1,\infty ]\).
Remark 4.2
By the standard Sobolev embedding theorem, the solution \((\rho ,u)\) constructed as in Theorem 4.1 is a classical solution, that is \((\rho ,u) \in {\mathcal {C}}^1({\mathbb {R}}^d \times (0,T^*))\).
Remark 4.3
The \(L^2\)-norm of u on the ball is introduced due to the confinement potential V. In fact, if we ignore the confinement potential V in the momentum equation in (4.1), then under the following assumption on the initial data
we have the unique strong solution \((\rho ,u)\) to the system (4.1) satisfying
Remark 4.4
In case of a singular interaction potential beyond the Coulomb case, we refer to [27] for the well-posedness theory for the Euler–Riesz system. More precisely, in [27], the local-in-time existence and uniqueness of classical solutions to the system (4.1) with \({\widetilde{W}}\) defined in (2.10) instead of the regular W, \(\gamma =0\), \(V \equiv 0\), and \(\psi \equiv 0\) are discussed. One may extend the arguments used in [27] to study the well-posedness for the system (4.1) with \({\widetilde{W}}\).
4.1 Linearized system
In this part, we construct approximate solutions \((\rho ^n, u^n)\) for the system (4.1) and provide some uniform bound estimates of it.
Let us first take the initial data as the zeroth approximation:
We next suppose that the nth approximation \((\rho ^n, u^n)\) with \(n \geqq 1\) is given. Then we define the \((n+1)\)th approximation \((\rho ^{n+1}, u^{n+1})\) as a solution to the following linear system.
with the initial data
Let us introduce a solution space \({\mathcal {Y}}_{s,R}(T)\) with \(s > d/2+1\) as
Then by the standard linear solvability theory [58], for any \(T>0\) we have that the approximation \(\{(\rho ^n,u^n)\}_{n=0}^\infty \subset {\mathcal {Y}}_{s,R}(T)\) is well-defined.
For notational simplicity, in the rest of this section, we drop x-dependence of the differential operator \(\nabla _x\).
Proposition 4.1
Suppose that the initial data \((\rho _0, u_0)\) satisfies \(\rho _0 > 0\) on \({\mathbb {R}}^d\) and
and let \((\rho ^n,u^n)\) be a sequence of the approximate solutions of (4.3) with the initial data \((\rho _0, u_0)\). Then for any \(N < M\), there exists \(T^* > 0\) such that
Proof
For the proof, we use the inductive argument. Since we take the initial data for the first iteration step, it is clear to find
We now suppose that
for some \(T_0 > 0\). In the rest of the proof, upon mollifying if necessary we may assume that the communication weight function \(\psi \) is smooth. Since this proof is a rather lengthy, we divide it into four steps:
-
In Step A, we provide the positivity and \(H^s({\mathbb {R}}^d)\)-estimate of \(\rho ^{n+1}\):
$$\begin{aligned} \rho ^{n+1}(x,t) > 0 \quad \forall \, (x,t) \in {\mathbb {R}}^d \times [0,T] \quad \text{ and } \quad \Vert \rho ^{n+1}(\cdot ,t)\Vert _{H^s} \leqq \Vert \rho _0\Vert _{H^s}e^{CMt} \end{aligned}$$for \(t \leqq T_0\), where \(C>0\) is independent of n.
-
In Step B, we show \({{\dot{\mathcal {W}}}}^{1,\infty }({\mathbb {R}}^d)\)-estimate and \(L^2(B(0,R))\)-estimate of \(u^{n+1}\):
$$\begin{aligned} \begin{aligned}&\Vert \nabla u^{n+1}(\cdot ,t)\Vert _{L^\infty } + \Vert u^{n+1}(\cdot ,t)\Vert _{L^2(B(0,R))} \\&\quad \leqq \Vert \nabla u_0\Vert _{L^\infty }e^{(CM - 1)t} + \Vert u_0\Vert _{L^2(B(0,R))} + E(t) \end{aligned} \end{aligned}$$for \(t \leqq T_0\), where \(C>0\) is independent of n, and \(E: [0,T_0] \rightarrow [0,\infty )\) is continuous on \([0,T_0]\) satisfying \(E(t) \rightarrow 0\) as \(t \rightarrow 0^+\).
-
In Step C, we estimate the higher order derivative of \(u^{n+1}\):
$$\begin{aligned} \Vert \nabla ^2 u^{n+1}\Vert _{H^{s-1}} \leqq \Vert \nabla ^2 u_0\Vert _{H^{s-1}} e^{CMt} + E(t) \end{aligned}$$for \(t \leqq T_0\), where \(C>0\) is independent of n, and E satisfies the same property as in Step B.
-
In Step D, we finally combine all of the estimates in Steps A, B, & C to conclude our desired result.
Step A.- We first show the positivity of \(\rho ^{n+1}\). Consider the following characteristic flow \(\eta ^{n+1}\) associated to the fluid velocity \(u^n\) by
with the initial data \( \eta ^{n+1}(x,0) =x \in {\mathbb {R}}^d. \) Since \(u^n\) is globally Lipschitz, the characteristic equations (4.4) are well-defined. Then by using the method of characteristics, we obtain
and applying Grönwall’s lemma yields
We next estimate \(H^s\)-norm of \(\rho ^{n+1}\). We first easily find
and
for \(2 \leqq k \leqq s\). Here we use Moser-type inequality to estimate \(I_i,i=1,\cdots ,4\) as
Combining all of the above estimates entails
for \(t \leqq T_0\), where \(C > 0\) is independent of n.
Step B.- Due to the positivity of \(\rho ^{n+1}\), it follows from the momentum equation in (4.3) that \(u^{n+1}\) satisfies
Taking the differential operator \(\nabla \) to (4.6) gives
where we used \(\nabla V = x\) and \({\mathbb {I}}_d\) denotes the \(n \times n\) identity matrix. Note that
and
We also estimate the last terms on the right hand side of (4.7) as
and
These estimates together with integrating (4.7) along the characteristic flow \(\eta ^{n+1}\) implies
By using Grönwall’s lemma, we obtain
This together with (4.5) asserts
where \(E_1: [0,T_0] \rightarrow [0,\infty )\) is continuous on \([0,T_0]\) satisfying \(E_1(t) \rightarrow 0\) as \(t \rightarrow 0^+\).
For the \(L^2\)-estimate of \(u^{n+1}\) on B(0, R), we multiply (4.6) by \(u^{n+1}\) and integrate it over B(0, R) to yield
Here we used
Thus we obtain
where \(C>0\) depends only on R and \(\Vert \psi \Vert _{L^2}\). Integrating this over [0, t] with \(t \leqq T_0\) and using the estimates (4.5) and (4.8) imply
where \(E_2: [0,T_0] \rightarrow [0,\infty )\) is continuous on \([0,T_0]\) satisfying \(E_2(t) \rightarrow 0\) as \(t \rightarrow 0^+\).
Step C.- For \(2 \leqq k \leqq s+1\), we find
where \(J_1\) and \(J_2\) can be estimated as
and
For the estimate of \(J_4\), we use the fact that W is the Coulombian potential to deduce
We next divide \(J_5\) into two terms:
Note that
Thus for \(\ell = k-1\) we get
and for \(0 \leqq \ell \leqq k-2\) we obtain
This asserts
Similarly, by integration by parts, we notice that
On the other hand, we find that
and
Moreover, for \(0 \leqq \ell \leqq k-3\) we obtain
Thus we have
and subsequently we get
We finally combine all of the above estimate to have
and applying Grönwall’s lemma gives
where we used the estimates in Steps B & C and \(E_3: [0,T_0] \rightarrow [0,\infty )\) is continuous on \([0,T_0]\) satisfying \(E_3(t) \rightarrow 0\) as \(t \rightarrow 0^+\).
Step D.- We now combine (4.5), (4.8), (4.9), and (4.10) to have
for \(t \leqq T_0\), where \(C>0\) is independent of n, and \(E: [0,T_0] \rightarrow [0,\infty )\) is continuous on \([0,T_0]\) satisfying \(E(t) \rightarrow 0\) as \(t \rightarrow 0^+\). On the other hand, the right hand side of (4.11) converges to \(\Vert \rho _0\Vert _{H^s} + \Vert u_0\Vert _{L^2(B(0,R))} + \Vert \nabla u_0\Vert _{L^\infty } + \Vert \nabla ^2 u_0\Vert _{H^{s-1}}\) as \(t \rightarrow 0^+\) and that is strictly less than N. This asserts that there exists \(T_* \leqq T_0\) such that
This completes the proof. \(\quad \square \)
4.2 Proof of Theorem 4.1
We first show the existence of a solution \((\rho ,u) \in {\mathcal {Y}}_{s,R}(T_*)\). Note that \(\rho ^{n+1} - \rho ^n\) and \(u^{n+1} - u^n\) satisfy
and
respectively. Then multiplying (4.12) by \(\rho ^{n+1} - \rho ^n\) and integrating it over \({\mathbb {R}}^d\) gives
where \(C > 0\) is independent of n. On the other hand, for \(k=0,1\), we find that
where we easily estimate
Here \(C>0\) is independent of n. We next use the following estimates
and
to have \( K_4 \leqq C\Vert u^{n+1} - u^n\Vert _{H^1}^2 + C\Vert \rho ^{n+1} - \rho ^n\Vert _{L^2}^2. \) For the rest, if \(k=0\), then
On the other hand, if \(k=1\), we obtain
We now combine all of the above estimates to have
and subsequently this yields
where \(C > 0\) is independent of n. This together with (4.13) asserts that \((\rho ^n,u^n)\) is a Cauchy sequence in \({\mathcal {C}}([0,T];L^2({\mathbb {R}}^d)) \times {\mathcal {C}}([0,T];H^1({\mathbb {R}}^d))\). Interpolating this strong convergences with the above uniform-in-n bound estimates gives
due to \(s > d/2+1\). We then use a standard functional analytic arguments, see for instances [29, Section 2.1], to have that the limiting functions \(\rho \) and u satisfy the regularity in Theorem 4.1. We easily show that the limiting functions \(\rho \) and u are solutions to (4.1) with regularity properties and assumptions of Theorem 1.2.
We finally provide the uniqueness of strong solutions. Let \((\rho ,u)\) and \(({\tilde{\rho }}, {{\tilde{u}}})\) be the strong solutions obtained above with the same initial data \((\rho _0, u_0)\). Set \(\Delta (t)\) a difference between two strong solutions:
Then by using almost the same argument as above, we have
This concludes that \(\Delta (t) \equiv 0\) on \([0,T_*]\) and completes the proof.
References
Balagué, D., Carrillo, J.A., Laurent, T., Raoul, G.: Nonlocal interactions by repulsive-attractive potentials: radial ins/stability. Physica D 260, 5–25, 2013
Bhatnagar, M., Liu, H.: Critical thresholds in one-dimensional damped Euler–Poisson systems. Math. Models Methods Appl. Sci. 30, 891–916, 2020
Bolley, F., Cañizo, J.A., Carrillo, J.A.: Stochastic mean-field limit: non-Lipschitz forces and swarming. Math. Models Methods Appl. Sci. 21, 2179–2210, 2011
Braun, W., Hepp, K.: The Vlasov dynamics and its fluctuations in the \(1/N\) limit of interacting classical particles. Comm. Math. Phys. 56, 101–113, 1977
Brenier, Y.: Convergence of the Vlasov–Poisson system to the incompressible Euler equations. Commun. Partial Differ. Equ. 25, 737–754, 2000
Brenier, Y., Mauser, N., Norbert, Puel, M.: Incompressible Euler and e-MHD as scaling limits of the Vlasov–Maxwell system. Commun. Math. Sci. 1, 437–447, 2003
Bresch, D., Jabin, P.-E., Wang, Z.: On mean-field limits and quantitative estimates with a large class of singular kernels: application to the Patlak–Keller–Segel model. C. R. Math. Acad. Sci. Paris. 357, 708–720, 2019
Carrillo, J.A., Choi, Y.-P.: Quantitative error estimates for the large friction limit of Vlasov equation with nonlocal forces. Ann. Inst. H. Poincaré Anal. Non Linéaire 3737, 925–954, 2020
Carrillo, J.A., Choi, Y.-P., Hauray, M.: The derivation of swarming models: Mean-field limit and Wasserstein distances. In: Collective Dynamics from Bacteria to Crowds, CISM Courses and Lect., vol. 553. Springer, pp. 1–46, 2014
Carrillo, J.A., Choi, Y.-P., Hauray, M., Salem, S.: Mean-field limit for collective behavior models with sharp sensitivity regions. J. Eur. Math. Soc. 21, 121–161, 2019
Carrillo, J.A., Choi, Y.-P., Jung, J.: Quantifying the hydrodynamic limit of Vlasov-type equations with alignment and nonlocal forces. Math. Models Methods Appl. Sci. 31, 327–408, 2021
Carrillo, J.A., Choi, Y.-P., Pérez, S.: A review on attractive-repulsive hydrodynamics for consensus in collective behavior, Active particles. Vol. 1. Advances in theory, models, and applications. Model. Simul. Sci. Eng. Technol., Birkhäuser/Springer, Cham, pp. 259–298, 2017
Carrillo, J.A., Choi, Y.-P., Tadmor, E., Tan, C.: Critical thresholds in 1D Euler equations with nonlocal forces. Math. Models Methods Appl. Sci. 26, 85–206, 2016
Carrillo, J.A., Choi, Y.-P., Tse, O.: Convergence to equilibrium in Wasserstein distance for damped Euler equations with interaction forces. Commun. Math. Phys. 365, 329–361, 2019
Carrillo, J.A., Choi, Y.-P., Zatorska, E.: On the pressureless damped Euler–Poisson equations with quadratic confinement: critical thresholds and large-time behavior. Math. Models Methods Appl. Sci. 26, 2311–2340, 2016
Carrillo, J.A., Delgadino, M.G., Pavliotis, G.A.: A proof of the mean-field limit for \(\lambda \)-convex potentials by \(\Gamma \)-convergence. J. Funct. Anal. 279, 108734, 2020
Carrillo, J.A., DiFrancesco, M., Figalli, A., Laurent, T., Slepčev, D.: Global-in-time weak measure solutions and finite-time aggregation for nonlocal interaction equations. Duke Math. J. 156, 229–271, 2011
Carrillo, J.A., D’Orsogna, M.R., Panferov, V.: Double milling in self-propelled swarms from kinetic theory. Kinet. Relat. Models 2, 363–378, 2009
Carrillo, J.A., Feireisl, E., Gwiazda, P., Świerczewska-Gwiazda, A.: Weak solutions for Euler systems with non-local interactions. J. Lond. Math. Soc. 95, 705–724, 2017
Carrillo, J.A., Fornasier, M., Toscani, G., Vecil, F.: Particle, kinetic, and hydrodynamic models of swarming. Mathematical Modeling of Collective Behavior in Socio-Economic and Life Sciences, Series: Modelling and Simulation in Science and Technology, Birkhäuser, pp. 297–336, 2010
Carrillo, J.A., Klar, A., Martin, S., Tiwari, S.: Self-propelled interacting particle systems with roosting force. Math. Models Methods Appl. Sci. 20, 1533–1552, 2010
Carrillo, J.A., Klar, A., Roth, A.: Single to double mill small noise transition via semi-Lagrangian finite volume methods. Commun. Math. Sci. 14, 1111–1136, 2016
Carrillo, J.A., Peng, Y., Wróblewska-Kamińska, A.: Relative entropy method for the relaxation limit of hydrodynamic models. Netw. Heterog. Media 15, 369–387, 2020
Choi, Y.-P.: Large friction limit of pressureless Euler equations with nonlocal forces, preprint
Choi, Y.-P., Ha, S.-Y., Li, Z.: Emergent dynamics of the Cucker–Smale flocking model and its variants, Active particles. Vol. 1. Advances in Theory, Models, and Applications. Model. Simul. Sci. Eng. Technol., Birkhäuser/Springer, Cham, pp. 299–331, 2017
Choi, Y.-P., Haskovec, J.: Hydrodynamic Cucker–Smale model with normalized communication weights and time delay. SIAM J. Math. Anal. 51, 2660–2685, 2019
Choi, Y.-P., Jeong, I.-J.: On well-posedness and singularity formation for the Euler–Riesz system, preprint
Choi, Y.-P., Jeong, I.-J.: Classical solutions to the fractional porous medium flow. Nonlinear Anal. 210, 112393, 2021
Choi, Y.-P., Kwon, B.: The Cauchy problem for the pressureless Euler/isentropic Navier–Stokes equations. J. Differ. Equ. 261, 654–711, 2016
Choi, Y.-P., Salem, S.: Propagation of chaos for aggregation equations with no-flux boundary conditions and sharp sensing zones. Math. Models Methods Appl. Sci. 28, 223–258, 2018
Choi, Y.-P., Salem, S.: Collective behavior models with vision geometrical constraints: truncated noises and propagation of chaos. J. Differential Equations. 266, 6109–6148, 2019
Choi, Y.-P., Yun, S.-B.: Existence and hydrodynamic limit for a Paveri–Fontana type kinetic traffic model. SIAM J. Math. Anal. 53, 2631–2659, 2021
Chuang, Y.-L., D’Orsogna, M.R., Marthaler, D., Bertozzi, A.L., Chayes, L.: State transitions and the continuum limit for a 2D interacting, self-propelled particle system. Physica D 232, 33–47, 2007
Cucker, F., Smale, S.: Emergent behavior in flocks. IEEE Trans. Autom. Control. 52, 852–862, 2007
Dobrushin, R.: Vlasov equations. Funct. Anal. Appl. 13, 115–123, 1979
D’Orsogna, M.R., Chuang, Y.-L., Bertozzi, A.L., Chayes, L.: Self-propelled particles with soft-core interactions: patterns, stability, and collapse. Phys. Rev. Lett. 9696, 104302-1/4, 2006
Duerinckx, M.: Mean-field limits for some Riesz interaction gradient flows. SIAM J. Math. Anal. 48, 2269–2300, 2016
Engelberg, S., Liu, H., Tadmor, E.: Critical thresholds in Euler–Poisson equations. Indiana Univ. Math. J. 50, 109–157, 2001
Fetecau, R., Sun, W.: First-order aggregation models and zero inertia limits. J. Differ. Equ. 259, 6774–6802, 2015
Fetecau, R., Sun, W., Tan, C.: First-order aggregation models with alignment. Physica D 325, 146–163, 2016
Filippov, A.F.: Differential Equations with Discontinuous Righthand Sides. Mathematics Applications Soviet Series, vol. 8. Kluwer, Dordrecht, 1988
Figalli, A., Kang, M.-J.: A rigorous derivation from the kinetic Cucker–Smale model to the pressureless Euler system with nonlocal alignment. Anal. PDE 12, 843–866, 2019
Fournier, N., Hauray, M., Mischler, S.: Propagation of chaos for the 2d viscous vortex model. J. Eur. Math. Soc. 16, 1423–1466, 2014
Golse, F.: The mean-field limit for the dynamics of large particle systems. Journées équations aux dérivées partielles 9, 1–47, 2003
Golse, F.: On the Dynamics of Large Particle Systems in the Mean Field Limit, in Macroscopic and Large Scale Phenomena: Coarse Graining, Mean Field Limits and Ergodicity. Lecturer Notes Applied Mathematics and Mechanics, vol. 3. Springer, Cham, pp. 1–144, 2016
Ha, S.-Y., Liu, J.-G.: A simple proof of the Cucker–Smale flocking dynamics and mean-field limit. Commun. Math. Sci. 7, 297–325, 2009
Ha, S.-Y., Tadmor, E.: From particle to kinetic and hydrodynamic descriptions of flocking. Kinet. Relat. Models 1, 415–435, 2008
Han-Kwan, D., Iacobelli, M.: From Newton’s second law to Euler’s equations of perfect fluids. Proc. Am. Math. Soc., to appear
Hauray, M.: Wasserstein distances for vortices approximation of Euler-type equations. Math. Models Methods Appl. Sci. 19, 1357–1384, 2009
Hauray, M.: Mean field limit for the one dimensional Vlasov–Poisson equation. In: Sém. Laurent Schwartz 2012–2013, exp. 21, 16 pp. ,2014
Hauray, M., Jabin, P.-E.: \(N\)-particles approximation of the Vlasov equations with singular potential. Arch. Ration. Mech. Anal. 183, 489–524, 2007
Hauray, M., Jabin, P.-E.: Particle approximations of Vlasov equations with singular forces: propagation of chaos. Ann. Sci. École Norm. Sup. 48, 891–940, 2015
Jabin, P.-E.: Macroscopic limit of Vlasov type equations with friction. Ann. Inst. H. Poincaré Anal. Non Linéaire. 17, 651–672, 2000
Jabin, P.-E., Wang, Z.: Mean field limit and propagation of chaos for Vlasov systems with bounded forces. J. Funct. Anal. 271, 3588–3627, 2016
Jabin, P.-E., Wang, Z.: Mean Field Limit for Stochastic Particle Systems, Active Particles. Vol. 1. Advances in Theory, Models, and Applications. Model. Simul. Sci. Eng. Technol., Birkhäuser/Springer, Cham, pp. 379–402, 2017
Jabin, P.-E., Wang, Z.: Quantitative estimates of propagation of chaos for stochastic systems with \(W^{-1,\infty }\) kernels. Invent. Math. 214, 523–591, 2018
Karper, T.K., Mellet, A., Trivisa, K.: Hydrodynamic limit of the kinetic Cucker–Smale flocking model. Math. Models Methods Appl. Sci. 25, 131–163, 2015
Kato, T.: Linear evolution equations of “hyperbolic” type II. J. Math. Soc. Jpn. 25, 648–666, 1973
Lattanzio, C., Tzavaras, A.E.: Relative entropy in diffusive relaxation. SIAM J. Math. Anal. 45, 1563–1584, 2013
Lattanzio, C., Tzavaras, A.E.: From gas dynamics with large friction to gradient flows describing diffusion theories. Commun. Partial Differ. Equ. 42, 261–290, 2017
Lazarovici, D., Pickl, P.: A mean field limit for the Vlasov–Poisson system. Arch. Ration. Mech. Anal. 225, 1201–1231, 2017
Masmoudi, N.: From Vlasov–Poisson system to the incompressible Euler system. Commun. Partial Differ. Equ. 26, 1913–1928, 2001
Minakowski, P., Mucha, P. B., Peszek, J., Zatorska, E.: Singular Cucker–Smale Dynamics, Active Particles. Vol. 2. Advances in Theory, Models, and Applications. Model. Simul. Sci. Eng. Technol., Birkhäuser/Springer, Cham, pp. 201–243, 2019
Neunzert, H.: An introduction to the nonlinear Boltzmann–Vlasov equation, In Kinetic theories and the Boltzmann equation (Montecatini Terme, 1981), Lecture Notes in Mathematics, vol. 1048. Springer, Berlin, 1984
Petrache, M., Serfaty, S.: Next order asymptotics and renormalized energy for Riesz interactions. J. Inst. Math. Jussieu. 16, 501–569, 2017
Serfaty, S.: Mean field limit for coulomb-type flows. Duke Math. J. 169, 2887–2935, 2020
Spohn, H.: Large Scale Dynamics of Interacting Particles. Texts and Monographs in Physics. Springer, Berlin 1991
Tadmor, E., Tan, C.: Critical thresholds in flocking hydrodynamics with nonlocal alignment. Philos. Trans. A Math. Phys. Eng. Sci. 372, 20130401, 2014
Tadmor, E., Wei, D.: On the global regularity of subcritical Euler–Poisson equations with pressure. J. Eur. Math. Soc. 10, 757–769, 2008
Acknowledgements
JAC was partially supported by EPSRC Grant Numbers EP/P031587/1 and EP/V051121/1, and the Advanced Grant Nonlocal-CPD (Nonlocal PDEs for Complex Particle Dynamics: Phase Transitions, Patterns and Synchronization) of the European Research Council Executive Agency (ERC) under the European Union’s Horizon 2020 research and innovation programme (Grant Agreement No. 883363). YPC was supported by NRF Grant (No. 2017R1C1B2012918), POSCO Science Fellowship of POSCO TJ Park Foundation, and Yonsei University Research Fund of 2019-22-0212.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by J. Bedrossian.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix A. Well-Posedness of the Particle System
Appendix A. Well-Posedness of the Particle System
In this appendix, we study the global existence and uniqueness of classical solutions to the particle system (1.1)–(1.2).
Let us first consider the case with singular interaction potentials with \(d\geqq 2\). In this case, we can use the repulsive effect from the interaction forces, and this also enables us to have the uniqueness of solutions.
Theorem A.1
Let \(d \geqq 2\). Suppose that \(\widetilde{W}\) is of the form (2.10) or (2.11) and the confinement potential V satisfies either \(V \rightarrow +\infty \) as \(|x| \rightarrow \infty \) or \(\nabla _x V\) has linear growth as \(|x| \rightarrow \infty \). If the initial data \(x_0\) satisfy
Then there exists a unique global smooth solution to the system (1.1)–(1.2) with \(\widetilde{W}\) instead of W satisfying
for \(t \geqq 0\), where \(C>0\) is independent of t.
Proof
For the proof, we first introduce the maximal life-span \(T_0 = T(x_0)\) of the initial data data \(x_0\) as
Then by the assumption and continuity of solutions, we get \(T_0>0\). We now claim that \(T_0 = \infty \) and for this it suffices to show that there is no collision between particles for all \(t \geqq 0\) and that particles cannot escape to infinity in finite time.
A straightforward computation yields
for \(t \in [0,T_0)\). Note that
and
where we used \(\nabla \widetilde{W}(-x) = -\nabla \widetilde{W}(x)\). Similarly, we also find
Combining all of the above estimates, we obtain
for \(t \in [0,T_0)\), where \({\mathcal {F}}^N(x,v)\) denotes the discrete free energy given by
If \(d=2\), then we have either
where \(\alpha \in (0,2)\). On the other hand, if \(d \geqq 3\), we obtain
for all \(t \in [0,T_0)\), where \(\alpha \in (d-2,d)\). Since the right hand side of the above inequality is uniformly bounded in t, we conclude \(T_0 = \infty \) for the case \(d \geqq 2\). An upper bound estimate of the distance between particles is a simple consequence of the uniform-in-time bound estimate of the free energy \({\mathcal {F}}^N\) due to the confinement potential whenever is present. If \(V=0\), one can obtain that particles cannot escape to infinity in finite time as soon as \(\nabla _x V\) has linear growth as \(|x|\rightarrow \infty \). \(\quad \square \)
Remark A.1
If the interaction and confinement potentials W and V are regular enough, that is \(\nabla _x W \in \mathcal {W}^{1,\infty }({\mathbb {R}}^d)\) and \(\nabla _x V \in \mathcal {W}^{1,\infty }({\mathbb {R}}^d)\), we have global-in-time existence and uniqueness of solutions by the standard Cauchy-Lipschitz theory. Moreover, an uniform-in-time bound of the distance between particles can also obtained due to the confinement potential if \(V \rightarrow +\infty \) as \(|x| \rightarrow \infty \).
Let us finally comment on the one dimensional case. If \(d=1\) and the interaction potential \(\widetilde{W}\) is given by (2.11), then we apply Theorem A.1 to get the global unique classical solution and uniform-in-time bound estimate. If W is given by the Coulomb potential, that is
Thus the interaction force \(- W'\) is discontinuous, but bounded. In this sense, it is not so singular compared to the other cases. Since the velocity alignment force is regular, we can use a similar argument as in [50, Proposition 1.2], see also [10, 41], to have the following proposition.
Proposition A.1
Let \(d=1\). For any initial configuration \(\mathcal {Z}^N(0)\), there exists at least one global-in-time solution to the system of (1.1) with (A.1) in the sense that \((x_i(t),v_i(t))\) satisfies the integral system:
Even though Proposition A.1 does not provide the uniqueness of solutions, it is not necessary for the analysis of mean-field limit or mean-field/small inertia limit from the particle system (1.1) to the pressureless Euler system (1.3) or the aggregation equation (1.4).
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Carrillo, J.A., Choi, YP. Mean-Field Limits: From Particle Descriptions to Macroscopic Equations. Arch Rational Mech Anal 241, 1529–1573 (2021). https://doi.org/10.1007/s00205-021-01676-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00205-021-01676-x