MEAN FIELD LIMIT FOR COULOMB-TYPE FLOWS

SYLVIA SERFATY, APPENDIX WITH MITIA DUERINCKX

Abstract. We establish the mean-field convergence for systems of points evolving along
the gradient flow of their interaction energy when the interaction is the Coulomb potential
or a super-coulombic Riesz potential, for the first time in arbitrary dimension. The proof
arXiv:1803.08345v3 [math.AP] 1 Apr 2019

is based on a modulated energy method using a Coulomb or Riesz distance, assumes that
the solutions of the limiting equation are regular enough and exploits a weak-strong stability
property for them. The method can handle the addition of a regular interaction kernel, and
applies also to conservative and mixed flows. In the appendix, it is also adapted to prove the
mean-field convergence of the solutions to Newton’s law with Coulomb or Riesz interaction
in the monokinetic case to solutions of an Euler-Poisson type system.

1. Introduction
1.1. Problem and background. We study here the large N limit of gradient flow evolutions
  

 1 X

ẋi = − ∇xi HN (x1 , . . . , xN ) + F(xi − xj ) , i = 1, . . . , N
(1.1) N
 j6 = i
xi (0) = x0i



or conservative evolutions of the form

 ẋ = − 1 J∇ H (x , . . . , x ) i = 1, . . . , N

i xi N 1 N
(1.2) N
xi (0) = x0i


where J is an antisymmetric matrix. The points xi evolve in the whole space Rd and their
energy HN is given by
X
(1.3) HN (x1 , . . . , xN ) = g(xi − xj )
i6=j

and g is a repulsive interaction kernel which assumed to be logarithmic, Coulomb, or Riesz;

more precisely of the form
(1.4) g(x) = |x|−s max(d − 2, 0) ≤ s < d for any d ≥ 1
or
(1.5) g(x) = − log |x| for d = 1 or 2.
In the case (1.4) with s = d − 2 and d ≥ 3, or (1.5) and d = 2, g is exactly (a multiple of) the
Coulomb kernel. In the other cases of (1.4) it is called a Riesz kernel. The map F : Rd → Rd
is an additional interaction force that we can add in the dissipative case to illustrate the
robustness of the method, and which enjoys some Sobolev and Hölder regularity.
1
2 SYLVIA SERFATY, APPENDIX WITH MITIA DUERINCKX

Mixed flows of the form

 
1 X
(1.6) ẋi = − (αI + βJ) ∇xi HN (x1 , . . . , xN ) + F(xi − xj ) α>0
N j6=i

can be treated with exactly the same proof, so can the same dynamics (1.1), (1.2) or (1.6)
with an additional forcing N1 N
P
i=1 V(xi ) with V Lipschitz. These generalizations are left to
the reader, see in particular Remark 2.2.
The limiting evolutions as N → ∞ of the systems (1.1) or (1.2), or their mean-field limit, is
a well-known and natural question : from a physical point of view they can model interacting
particles, from a numerical point of view they correspond to particle approximations of the
limiting PDEs or can serve to approximate their equilibrium states. More general interaction
kernels including attractive terms are also of interest as aggregation and swarming models,
see for instance [CCH]. In dimension 2, choosing (1.5) and J the rotation by π/2 in (1.2)
corresponds to the so-called point vortex system which is well-known in fluid mechanics (cf.
for instance [MP]), and its mean-field convergence to the Euler equation in vorticity form was
already established [Scho2].
Consider the empirical measure
N
1 X
(1.7) µtN := δ t
N i=1 xi
associated to a solution XNt := (xt , . . . , xt ) of the flow (1.1) or (1.2). If the points x0 ,
1 N i
which themselves depend on N , are such that µ0N converges to some regular measure µ0 , then
a formal derivation leads to expecting that for t > 0, µtN converges to the solution of the
Cauchy problem with initial data µ0 for the limiting evolution equation
(1.8) ∂t µ = div ((∇g + F) ∗ µ)µ)
in the dissipative case (1.1) or
(1.9) ∂t µ = div (J∇(g ∗ µ)µ)
in the conservative case (1.2).
These equations should be understood in a weak sense. Equation (1.8) (with F = 0) is
sometimes called the fractional porous medium equation. The two-dimensional Coulomb ver-
sion also arises as a model for the dynamics of vortices in superconductors. The construction
of solutions, their regularity and basic properties, are addressed in [LZ, DZ, AS, SV] for the
Coulomb case of (1.8), in [CSV,CV,XZ] for the case d− 2 < s < d of (1.8), and [De,Yu,Scho1]
for the two-dimensional Coulomb case of (1.2).
Establishing the convergence of the empirical measures to solutions of the limiting equa-
tions is nontrivial because of the nonlinear terms in the equation and the singularity of the
interaction g. In fact, because of the strength of the singularity, treating the case of Coulomb
interactions in dimension d ≥ 3 (and even more so that of super-coulombic interactions) had
remained an open question for a long time, see for instance the introduction of [JW2] and the
review [Jab]. It was not even completely clear that the result was true without expressing it
in some statistical sense (with respect to the initial data).
In [JW1, JW2], Jabin and Wang introduced a new approach for the related problem of
the mean-field convergence of the solutions of Newton’s second order system of ODEs to the
Vlasov equation, which allowed them to treat all interactions kernels with bounded gradients,
 MEAN FIELD LIMIT FOR COULOMB FLOWS 3

but still not Coulomb interactions. The same problem has been addressed in [BP,LP,La] with
results that still require a cutoff of the Coulomb interactions. Our method already allows to
unlock the case of Coulomb interaction for monokinetic data, this is the topic of Appendix A.
The non-monokinetic case is even much more challenging and remains open. Finally, study-
ing the same evolutions with added noise is also very interesting and has motivations from
Random Matrix Theory, it is done in particular in [JW1, JW2, BO, FHM, BCF, LLY] (see also
references therein), but we do not address this question here.

The previously known results on the problems we are addressing were the following:
• Schochet [Scho2] proved the convergence of the conservative flow (1.2) to (1.9) in the
two-dimensional logarithmic case, and his proof can be readapted to treat (1.1) as
well in that case. Results of similar nature were also obtained in [GHL].
• Hauray [Hau] (see also [CCH]) treated the case of all sub-Coulombic interactions
(s < d − 2) for (a possibly higher-dimensional generalization of) (1.2), where particles
can have positive and negative charges and thus can attract as well as repel. His
proof, which relies on the stability in ∞-Wasserstein distance of the limiting solution,
cannot be adapted to s ≥ d − 2.
• In dimension 1 and in the dissipative case (1.1), Carrillo-Ferreira-Precioso and Berman-
Onnheim [CFP, BO] proved the unconditional convergence for all 0 < s < 1 using the
framework of Wasserstein gradient flows but their method, based on the convexity of
the interaction in dimension 1, does not extend to higher dimensions.
• Duerinckx [Du], inspired by the modulated energy method of [Sy] for Ginzburg-Landau
equations (where vortices also interact like Coulomb particles in dimension 2), was
able to prove the result in the dissipative case for d = 1 and d = 2 with s < 1,
conditional to the regularity of the limiting solution, as we have here.

In this paper, we extend Duerinckx’s result to all cases of (1.4) or (1.5), which involves
overcoming serious difficulties, as described further below, and we add the possible additional
interaction force F in the dissipative case. We are limited to s < d and this is natural since
for s ≥ d the interaction kernel g is no longer integrable and the limiting equation is expected
to be a different one.
As in [Du], our proof is a modulated energy argument inspired from [Sy], which is a way of
exploiting a weak-strong uniqueness principle for the limiting equation. As mentioned above,
looking for a stability principle in some Wasserstein distance fails at the Coulomb singularity.
Instead we use a distance which is built as a Coulomb (or Riesz) metric, associated to the
norm
¨
(1.10) kµk2 = g(x − y)dµ(x)dµ(y).

We are able to show by a Gronwall argument on this metric that the equations (1.8) and (1.9)
satisfy a weak-strong uniqueness principle, and this can be translated into a proof of stability
and convergence to 0 of the norm of µtN − µt (if it is initially small, it remains small for all
further times).
The proof is self-contained and quantitative. It does not require understanding any quali-
tative property of the trajectories of the particles, such as for instance their minimal distances
along the flow.
4 SYLVIA SERFATY, APPENDIX WITH MITIA DUERINCKX

1.2. Main result. Let XN denote (x1 , . . . , xN ) and let us define for any probability measure
µ,
¨ N
X  N
X 
(1.11) FN (XN , µ) = g(x − y)d δxi − N µ (x)d δxi − N µ (y)
Rd ×Rd \△ i=1 i=1

where △ denotes the diagonal in Rd × Rd . We choose for “modulated energy”

t
FN (XN , µt )

where XN t = (xt , . . . , xt ) are the solutions to (1.1) or (1.2), µt solves the expected limiting
1 N
PDE. It turns out that FN is a good notion of distance from µtN to µt and metrizes at least
weak convergence, as described in Proposition 3.5.
Our main result is a Gronwall inequality on the time-derivative of FN (XN t , µt ), which
t t
implies a quantitative rate of convergence of µN to µ in that metric.
Throughout the paper, s should be understood as equal to 0 in the formulae in the cases
(1.5), and (·)+ denotes the positive part of a number. The space Ḣ m (Rd ) is the homogeneous
Sobolev space of functions u whose Fourier transform û satisfies |ξ|m û(ξ) ∈ L2 (Rd ).
d−s
Theorem 1. Assume that g is of the form (1.4) or (1.5). Assume that F ∈ Ḣ 2 (Rd ) ∩
C 0,α (Rd ) for some α > 0 and ∇F ∈ Lq (Rd ) for some 1 ≤ q ≤ ∞. Assume (1.8), resp. (1.9),
admits a solution µt such that, for some T > 0,
(1.12)
(
µt ∈ L∞ ([0, T ], L∞ (Rd )), and ∇2 g ∗ µt ∈ L∞ ([0, T ], L∞ (Rd )) if s < d − 1
µt ∈ L∞ ([0, T ], C σ (Rd )) with σ > s − d + 1, and ∇2 g ∗ µt ∈ L∞ ([0, T ], L∞ (Rd )) if s ≥ d − 1.

Let XNt solve (1.1), respectively (1.2). Then there exist positive constants C , C depending
1 2
only on the norms of µt controlled by (1.12) and those of F, and an exponent β < 2 depending
only on d, s, α, σ, such that for every t ∈ [0, T ] we have
 
t
(1.13) FN (XN , µt ) ≤ FN (XN
0
, µ0 ) + C1 N β eC2 t .

In particular, using the notation (1.7), if µ0N ⇀ µ0 and is such that

1 0
lim FN (XN , µ0 ) = 0,
N →∞ N2
then the same is true for every t ∈ [0, T ] and

(1.14) µtN ⇀ µt

in the weak sense.

Establishing the convergence of the empirical measures is essentially equivalent to proving

propagation of molecular chaos (see [Go,HM,Jab] and references therein) which means showing
that if fN0 (x , . . . , x ) is the initial probability density of seeing particles at (x , . . . , x ) and
1 N 1 N
0 converges to some factorized state µ0 ⊗· · ·⊗µ0 , then the k-point marginals f t
if fN N,k converge
for all time to (µt )⊗k . With Remark 3.6, our result implies a convergence of this type as well.
 MEAN FIELD LIMIT FOR COULOMB FLOWS 5

1.3. Comments on the assumptions. Let us now comment on the regularity assumption
made in (1.12). First of all, one can check (see Lemma 3.1) that the assumption (1.12) is
implied by
(1.15) µ ∈ L∞ ([0, T ], C θ (Rd )) for some θ > s − d + 2,
which coincides with the assumption made in [Du] and is a bit stronger. This weakening
of the assumption allows to include for instance the case of measures which are (a regular
function times) the characteristic function of some regular set, such as in the situation of
vortex patches for the Euler equation in vorticity form, corresponding to (1.2) in the two-
dimensional logarithmic case. These vortex patches were first studied in [Ch2,BC,Si] where it
was shown that if the patch initially has a C 1,α boundary this remains the case over time, and
our second assumption that the velocity ∇g ∗ µt be Lipschitz holds as well (see also [BK]). It
is not too difficult to check that in all dimensions this second condition holds any time µ is C σ
with σ > 0 away from a finite number of C 1,α hypersurfaces. More generally, such situations
with patches can be expected to naturally arise in all the Coulomb cases. For instance, in
the dissipative Coulomb case (1.1) with F = 0, in any dimension a self-similar solution in the
form of (a constant multiple of) the characteristic function of an expanding ball was exhibited
in [SV] and shown to be an attractor of the dynamics. For the non-Coulomb dissipative cases,
the corresponding self-similar solutions, called Barenblatt solutions, are of the form
d 2 s−d+2
t− 2+s (a − bx2 t− 2+s )+ 2
as shown in [BIK, CV] (and this formula retrieves the solution of [SV] for s = d − 2).
If the initial µ0 is sufficiently regular, the stronger assumption (1.15) is known to hold with
T = ∞ for the Coulomb case (see [LZ] where the proof works as well in higher dimensions),
and it is known up to some T > 0 in the case (d − 2)+ < s ≤ d − 1 [XZ]. As for (1.9), to
our knowledge the desired regularity is only known in dimension 2 for the Euler equation in
vorticity form (see [Wo,Ch2]), although the arguments of [XZ] written for the dissipative case
seem to also apply to the conservative one. Our convergence result thus holds in these cases,
under the assumption that the limit µ0 of µ0N is sufficiently regular and that FN (XN 0 , µ0 ) =

o(N 2 ). Note that, as shown in [Du], the latter is implied by the convergence of the initial
energy
1 X
¨
0 0
lim g(xi − xj ) = g(x − y)dµ0 (x)dµ0 (y)
N →∞ N 2 d
R ×R d
i6=j

which can be viewed as a well-preparedness condition.

For d − 1 < s < d, even the local in time propagation of regularity of solutions of (1.8)
remains an open problem. Note that the uniqueness of regular enough solutions is always
implied by the weak-strong stability argument we use, reproduced in Section 1.4 below.
Requiring some regularity of the solutions to the limiting equation for establishing con-
vergence with relative entropy / modulated entropy / modulated energy methods is fairly
common: the same situation appears for instance in [JW1, JW2] or in the derivation of the
Euler equations from the Boltzmann equation via the modulated entropy method, see [SR]
and references therein.

1.4. The method. As mentioned, our method exploits a weak-strong uniqueness principle
for the solutions of (1.8), resp. (1.9), which is exactly the same as [Du, Lemma 2.1, Lemma
6 SYLVIA SERFATY, APPENDIX WITH MITIA DUERINCKX

2.2] (and can be easily readapted to the conservative case) and states that if µt1 and µt2 are
two L∞ solutions to (1.8) such that ∇2 (g ∗ µ2 ) ∈ L1 ([0, T ], L∞ ), we have

¨
(1.16) g(x − y)d(µt1 − µt2 )(x)d(µt1 − µt2 )(y)
Rd ×Rd
´t ¨
2 s
≤ eC 0 k∇ (g∗µ2 )kds g(x − y)d(µ01 − µ02 )(x)d(µ01 − µ02 )(y).
Rd ×Rd

But the Coulomb or Riesz energy (1.10) is nothing else than the fractional Sobolev H −α norm
of µ with α = d−s 2 , hence this is a good metric of convergence and implies the weak-strong
uniqueness property.
A crucial ingredient is the use of the stress-energy (or energy-momentum) tensor which
naturally appears when taking the inner variations of the energy (1.10), i.e. computing
d 2
dt |t=0 kµ ◦ (I + tψ)k (this is standard in the calculus of variations, see for instance [He, Sec.
1.3.2]). To explain further, let us restrict for now to the Coulomb case, and set hµ = g ∗ µ.
In that case, we have

(1.17) − ∆hµ = cd µ

for some constant cd depending only on d. The first key is to reexpress the Coulomb energy
(1.10) as a single integral in hµ , more precisely we easily find via an integration by parts that

1 1
¨ ˆ ˆ ˆ
g(x − y)dµ(x)dµ(y) = hµ dµ = − hµ ∆hµ = |∇hµ |2 .
Rd ×Rd Rd cd Rd cd Rd

The stress-energy tensor is then defined as the d × d tensor with coefficients

(1.18) [hµ , hµ ]ij = 2∂i hµ ∂j hµ − |∇hµ |2 δij ,

with δij the Kronecker symbol. We may compute that

(1.19) div [hµ , hµ ] = 2∆hµ ∇hµ = −2cd µ∇hµ .

(Here the divergence is a vector with entries equal to the divergence of each row of [hµ , hµ ].)
We thus see how this stress-energy tensor allows to give a weak meaning to the product
µ∇hµ = µ∇g ∗ µ, with [hµ , hµ ] well-defined in energy space and pointwise controlled by
|∇hµ |2 , which can by the way serve to give a notion of weak solutions of the equation in the
energy space (as in [De, LZ]). Note that in dimension 2, it is known since [De] that even
though [hµ , hµ ] is nonlinear, it is stable under weak limits in energy space provided µ has a
sign, but this fact does not extend to higher dimension.
Let us now present the short proof of (1.16) as it will be a model for the main proof.
We focus on the dissipative case (the conservative one is an easy adaptation) and still the
Coulomb case for simplicity with no additional interaction F (when present, the additional
terms can be absorbed thanks to the dissipation). Let µ1 and µ2 be two solutions to (1.8)
 MEAN FIELD LIMIT FOR COULOMB FLOWS 7

and hi = g ∗ µi the associated potentials, which solve (1.17). Let us compute

ˆ ˆ
2
(1.20) ∂t |∇(h1 − h2 )| = 2cd (h1 − h2 )∂t (µ1 − µ2 )
Rd Rd
ˆ
= 2cd (h1 − h2 )div (µ1 ∇h1 − µ2 ∇h2 )
Rd
ˆ
= −2cd (∇h1 − ∇h2 ) · (µ1 ∇h1 − µ2 ∇h2 )
Rd
ˆ ˆ
2
= −2cd |∇(h1 − h2 )| µ1 − 2cd ∇h2 · ∇(h1 − h2 )(µ1 − µ2 )
Rd Rd
In the right-hand side, we recognize from (1.19) the divergence of the stress-energy tensor
[h1 − h2 , h1 − h2 ], hence
ˆ ˆ
2
∂t |∇(h1 − h2 )| ≤ ∇h2 · div [h1 − h2 , h1 − h2 ]
Rd Rd
so if ∇2 h2 is bounded, we may integrate by parts the right-hand side and bound it pointwise
by ˆ ˆ
2 2
k∇ h2 kL∞ |[h1 − h2 , h1 − h2 ]| ≤ 2k∇ h2 kL∞ |∇(h1 − h2 )|2 ,
Rd Rd
and the claimed result follows by Gronwall’s lemma.
In the Riesz case, the Riesz potential hµ = g∗µ is no longer the solution to a local equation,
and to find a replacement to (1.17)–(1.19), we use an extension procedure as popularized
by [CaffSi] in order to obtain a local integral in hµ in the extended space Rd+1 .
In the discrete case of the original ODE system, all the above integrals are singular and
this constitutes the main difficulty overcome in this paper. In place of the second term in the
right-hand side of (1.20), we then have to control a term which by symmetry can be written
in the form
¨
(1.21) (∇hµ1 (x) − ∇hµ2 (y)) · ∇g(x − y)d(µ1 − µ2 )(x)d(µ1 − µ2 )(y)
Rd ×Rd \△

where △ denotes the diagonal, µ1 is the limiting measure µt and µ2 is the discrete empirical
measure µtN . Such terms are well known (see for instance [Scho2]), and create the main
difficulty due to the singularity of g. When removing the diagonal, the positivity of the
interaction is in effect lost.
The key is to interpret (1.21) in a suitable “renormalized” way. We do this via a truncation
of hµ , at a lengthscale ri depending on the point xi and equal to the minimal distance from
xi to its nearest neighbors. An observation made for the first time in this paper is that when
using these particular truncation parameters, the truncated energy can be controlled by the
full energy and conversely (see Corollary 3.4). This would not be true when using truncation
parameters that are independent of i, as we do not know any control on the minimal distances
between points. The next crucial result is that, even though positivity is lost, when using
the ri ’s as truncation parameters we can still control for each time slice the expression in
(1.21) by the modulated energy FN (XN t , µt ) itself, up to a small error (Proposition 2.3), and

provided the limiting solution is regular enough. This, which is the most difficult part of the
proof, uses two ingredients: the first is to reexpress (1.21) as a single integral in terms of a
stress-energy tensor, and the second is to show that the expression in (1.21) is in fact close
to the same expression with truncated fields.
8 SYLVIA SERFATY, APPENDIX WITH MITIA DUERINCKX

This was the main obstacle to treating the case of general dimension. The idea of express-
ing the interaction energy as a local integral in hµ and its renormalization procedure, were
previously used in the study of Coulomb and Riesz energies in [RS, PS, LS1, LS2], but it was
not clear how to adapt these ideas to control (1.21): in fact in [Du] this was dealt with by
a ball-construction procedure inspired from the analysis of Ginzburg-Landau vortices, which
led to nonoptimal estimates and to the restriction to the dissipative case only and to s < 1
and d ≤ 2.
We also note that the idea of using truncations for proving mean-field limits is common,
however what is usually done is to truncate the interaction g itself (at lengthscales possibly
depending on N ), and try to take limits in the resulting flow. What we do here is very
different: we do not modify the interaction but desingularize the charges themselves as an
intermediate step to control the singular terms.
Acknowledgments: I thank Mitia Duerinckx for his careful reading and helpful sugges-
tions. This research was supported by the NSF grant DMS-1700278.

2. Main proof
In all the paper, we will use the notation 1(1.5) to indicate a term which is only present
in the logarithmic cases (1.5) and 1s<d−1 for a term which is present only if s ´< d − 1.
Below, the principal value (which may be omitted for s < d − 1) is defined by P.V. Rd \{x} :=
´
limr→0 Rd \B(x,r) .
Differentiating from formula (1.11), we have
t is a solution of (1.1), then
Lemma 2.1. If XN
ˆ  ˆ 2
 
t
(2.1) ∂t FN (XN , µt ) = −2N 2
P.V. ∇g(x − y)d(µN − µ )(y) dµtN (x)
t t
 
R d  d
R \{x} 
ˆ
2 t t
+ 2N F ∗ (µt − µtN ) · ∇(hµN − hµ )dµtN
Rd
¨  
µt t
−N 2 (∇h + F ∗ µt )(x) − (∇hµ + F ∗ µt )(y) ·∇g(x−y)d(µtN −µt )(x)d(µtN −µt )(y).
Rd ×Rd \△
t is a solution of (1.2), then
If XN

(2.2) ¨  t t

t
∂t FN (XN , µt ) = −N 2
J ∇hµ (x) − ∇hµ (y) ·∇g(x−y)(µtN −µt )(x)(µtN −µt )(y).
Rd ×Rd \△

Remark 2.2. In the case of an added term N1 N

P
i=1 V(xi ) in the evolutions, one obtains an
additional term
¨
2
−2N (V(x) − V(y)) · ∇g(x − y)d(µtN − µt )(x)d(µtN − µt )(y)
Rd ×Rd \△

which can be handled like the others using Proposition 2.3.

Proof. We note that if s ≥ d − 1, ∇g is not integrable near 0, so ∇g ∗ µ should be understood
in a distributional sense and µ∇(g ∗ µ) = µg ∗ ∇µ as well, assuming that µ is regular enough.
 MEAN FIELD LIMIT FOR COULOMB FLOWS 9

We may also check that this distributional definition is equivalent to defining

ˆ
µ
∇h (x) = P.V. ∇g(x − y)dµ(y).
Rd \{x}

In the case (1.1), we have

¨
t
, µt ) 2
g(x − y)dµt (x)dµt (y) + ∂t g(xti − xtj )
X
∂t FN (XN = N ∂t
Rd ×Rd i6=j
N ˆ
g(xti − y)dµt (y)
X
−2N ∂t
Rd
ˆi=1 ˆ
2 µt 2 t 2 t
= −2N |∇h | (x)dµ (x) − 2N ∇hµ (x) · F ∗ µt (x)dµt (x)
Rd Rd
 2  
N X  N
∇g(xti − xtj ) − 2 ∇g(xti − xtj ) · F(xti − xtj )
X  X X X
−2 


i=1  j6=i  i=1 j6=i j6=i
t
 
∇hµ (xti ) · ∇g(xti − xtj ) + F(xti − xtj )
X
+2N
i6=j
N ˆ
t
(∇hµ + F ∗ µt )(x) · ∇g(x − xti )dµt (x).
X
+2N P.V.
i=1 Rd \{xti }

We then recombine the terms to obtain

ˆ  ˆ 2
 
t
∂t FN (XN , µt ) = −2N 2
P.V. ∇g(x − y)d(µN − µ )(y) dµtN (x)
t t
 
R d  d
R \{x} 
ˆ ˆ
2 µt 2 t 2 t
−2N |∇h | dµ − 2N ∇hµ (x) · F ∗ µt (x)dµt (x)
Rd Rd
ˆ ˆ ! ˆ
µtN t
−2N ∇h (x) · F(x − y)dµtN (y) dµtN (x) + 2N 2
|∇hµ |2 dµtN
Rd Rd \{x} Rd
ˆ ˆ
t
+2N 2 ∇hµ (x) · (∇g + F)(x − y)dµtN (y)dµtN (x)
Rd Rd \{x}
ˆ ˆ
t
+2N 2 P.V. (∇hµ + F ∗ µt )(x) · ∇g(x − y)dµt (x)dµtN (y).
Rd Rd \{y}

We next recognize that all the terms except the first can be recombined and symmetrized
into
¨  
t t
−N2 (∇hµ + F ∗ µt )(x) − (∇hµ + F ∗ µt )(y) · ∇g(x − y)d(µtN − µt )(x)d(µtN − µt )(y)
△c
ˆ
2 t t
+ 2N F ∗ (µt − µtN ) · ∇(hµN − hµ )dµtN
Rd

which gives the desired formula.

10 SYLVIA SERFATY, APPENDIX WITH MITIA DUERINCKX

In the case (1.2) we have

¨ N ˆ
t
, µt ) = N 2 ∂t g(x−y)dµt (x)dµt (y)+∂t g(xti −xtj )−2N ∂t g(xti −y)dµt (y)
X X
∂t FN (XN
i6=j i=1 Rd
N ˆ
t t
∇hµ (xti ) · J∇g(xti − xtj ) + 2N J∇hµ (x) · ∇g(x − xti )dµt (x)
X X
= 2N P.V.
i6=j i=1 Rd \{xti }

We then rewrite this as

ˆ ˆ
t µt
∂t FN (XN , µt ) = 2N 2
∇h (x) · J∇g(x − y)dµtN (y)dµtN (x)
Rd Rd \{x}
ˆ ˆ
2 t
+2N P.V. J∇hµ (x) · ∇g(x − y)dµt (x)dµtN (y).
Rd Rd \{y}

By antisymmetry of J, we recognize that the right-hand side can be symmetrized into

¨  
2 t t
−N J ∇hµ (x) − ∇hµ (y) · ∇g(x − y)d(µtN − µt )(x)d(µtN − µt )(y).
△c

The main point is thus to control the last term in the right-hand side of (2.1) or (2.2) :
we may reinterpret it via an appropriate stress-energy tensor, which we may show is itself
pointwise controlled by the modulated energy density, once we have given a renormalized
sense to all the quantities.
The crucial result we obtain this way is the following proposition.

Proposition 2.3. Assume that µ is a probability density, with µ ∈ C σ (Rd ) with σ > s − d + 1
if s ≥ d− 1; respectively µ ∈ L∞ (Rd ) or µ ∈ C σ (Rd ) with σ > 0 if s < d− 1. For any Lipschitz
map ψ : Rd → Rd , we have

N N
¨ 
 X X 
(2.3) (ψ(x) − ψ(y)) · ∇g(x − y)d( δxi − N µ)(x)d( δxi − N µ)(y)
 

 △c 
i=1 i=1
N
   
s s+1
≤ Ck∇ψkL∞ FN (XN , µ) + (1 + kµkL∞ )N 1+ d + N 2− 2 + log N 1(1.5)
d
s+1 s s+1−σ
 
+ C min kψkL∞ kµkL∞ N 1+ d + k∇ψkL∞ kµkL∞ N 1+ d , kψkW 1,∞ kµkC σ N 1+ d

( 1
k∇ψkL∞ (1 + kµkC σ )N 2− d if s ≥ d − 1
+C 1
k∇ψkL∞ (1 + kµkL∞ )N 2− d if s < d − 1,

where C depends only on s, d.

The proof of this proposition will occupy Section 4. Here, all the additive terms should
be considered as moderate perturbations of FN , in particular what matters is that they are
o(N 2 ).
For the dissipative case with the added force, we will also need
 MEAN FIELD LIMIT FOR COULOMB FLOWS 11

d−s
Lemma 2.4. Assume F ∈ Ḣ 2 (Rd ) ∩ C 0,α (Rd ) for some α > 0. Then there exists λ > 0 and
C > 0 depending only on α, s, d such that for every t,
ˆ ˆ  ˆ 2
t t
 
2 t
N F∗(µ −µtN )·∇(hµN −hµ )dµtN ≤N 2
P.V. ∇g(x − y)d(µN − µ )(y) dµtN (x)
t t
 
Rd R d  d
R \{x} 
N
   
s 2λ
+CkFk2 d−s
t
FN (XN , µt ) + (1 + kµt kL∞ )N 1+ d + log N 1(1.5) +CkFk2C 0,α (Rd ) N 2− d .
Ḣ 2 (Rd ) d
With these two results at hand, and noting that by assumption µt ∈ ∩∞ p d
p=1 L (R ) and
taking p to be the conjuguate exponent to q
k∇F ∗ µt kL∞ ≤ k∇FkLq kµt kLp ,
we immediately deduce from Lemma 2.1 that
 
t 2 µt
∂t FN (XN , µt ) ≤ C k∇ h kL∞ (Rd ) + k∇FkLq (Rd ) + kFk 2
d−s
Ḣ 2 (Rd )
h N
  
s 3 2λ
t
× FN (XN , µt ) + (1 + kµt kL∞ )N 1+ d + N 2 + log N 1(1.5) + CkFk2C 0,α (Rd ) N 2− d
d
1 s+1−σ
   i
(1 + kµt kC σ )N 2− d + k∇hµt kL∞ + k∇2 hµt kL∞ kµkC σ N 1+ d if s ≥ d − 1
+C 1 s+1
i
2− t
(1 + kµt k ∞ )N d + k∇hµ k ∞ kµt k ∞ N 1+ t
d + k∇2 hµ kL∞ kµt kL∞ )N
1+ ds
L L L if s < d − 1.

Since s < d and σ > s − d + 1, this implies by Gronwall’s lemma and in view of (1.12) that
for every t ≤ T ,
 
t
FN (XN , µt ) ≤ FN (XN
0
, µ0 ) + C1 N β eC2 t for some β < 2.

In view Proposition 3.5 below, this proves the main theorem.

3. Formulation via the electric potential

3.1. The extension representation for the fractional Laplacian. In general, the kernel
g is not the convolution kernel of a local operator, but rather of a fractional Laplacian. Here
we use the extension representation popularized by [CaffSi]: by adding one space variable
y ∈ R to the space Rd , the nonlocal operator can be transformed into a local operator of the
form −div (|z|γ ∇·).
In what follows, k will denote the dimension extension. We will take k = 0 in the Coulomb
cases for which g itself is the kernel of a local operator. In all other cases, we will take k = 1.
For now, points in the space Rd will be denoted by x, and points in the extended space Rd+k
by X, with X = (x, z), x ∈ Rd , z ∈ Rk . We will often identify Rd × {0} and Rd and thus
(xi , 0) with xi .
If γ is chosen such that
(3.1) d − 2 + k + γ = s,
then, given a probability measure µ on Rd , the g-potential generated by µ, defined in Rd by
ˆ
µ
h (x) := g(x − x̃) dµ(x̃)
Rd
12 SYLVIA SERFATY, APPENDIX WITH MITIA DUERINCKX

can be extended to a function hµ on Rd+k defined by

ˆ
µ
h (X) := g(X − (x̃, 0)) dµ(x̃),
Rd
and this function satisfies
(3.2) − div (|z|γ ∇hµ ) = cd,s µδRd
where by δRd we mean the uniform measure on Rd × {0}. The corresponding values of the
constants cd,s are given in [PS, Section 1.2]. In particular, the potential g seen as a function
of Rd+k satisfies
(3.3) − div (|z|γ ∇g) = cd,s δ0 .
To summarize, we will take
• k = 0, γ = 0 in the Coulomb cases. The reader only interested in the Coulomb cases
may thus just ignore the k and the weight |z|γ in all the integrals.
• k = 1, γ = s − d + 2 − k in the Riesz cases and in the one-dimensional logarithmic case
(then we mean s = 0). Note that our assumption (d − 2)+ ≤ s < d implies that γ is
always in (−1, 1). We refer to [PS, Section 1.2] for more details.
We now make a remark on the regularity of hµ :
Lemma 3.1. Assume µ is a probability density in C θ (Rd ) for some θ > s − d + 2, then we
have
 
(3.4) k∇hµ kL∞ (Rd ) ≤ C kµkC θ−1 (Rd ) + kµkL1 (Rd ) ,
and
 
(3.5) k∇2 hµ kL∞ (Rd ) ≤ C kµkC θ (Rd ) + kµkL1 (Rd ) .
d−s s−d
Proof. As is well known, g is (up to a constant) the kernel of ∆ 2 , hence hµ = cd,s ∆ 2 µ and
the relations follow (cf. also [Du, Lemma 2.5]). 

3.2. Electring rewriting of the energy. We briefly recall the procedure used in [RS, PS]
for truncating the interaction or, equivalently, spreading out the point charges. It will also
be crucial to use the variant introduced in [LSZ, LS2] where we let the truncation distance
depend on the point.
For any η ∈ (0, 1), we define
(3.6) gη := min(g, g(η)), fη := g − gη
and
(η) 1
(3.7) δ0 := − div (|z|γ ∇gη ),
cd,s
which is a positive measure supported on ∂B(0, η).
Remark 3.2. This nonsmooth truncation of gη can be replaced with no change by a smooth
one such that
1
gη (x) = g(x) for |x| ≥ η, gη (x) = cst for |x| ≤ η − ε ε< η
2
 MEAN FIELD LIMIT FOR COULOMB FLOWS 13

(η)
and this way δ0 gets replaced by a probability measure with a regular density supported in
B(0, η)\B(0, η − ε). We make this modification whenever the integrals against the singular
measures may not be well-defined.
We will also let
(3.8) fα,η := fα − fη = gη − gα ,
and we observe that fα,η has the sign of α − η, vanishes outside B(0, max(α, η)), and satisfies
(3.9) g ∗ (δx(η) − δx(α) ) = fα,η (· − x)
and
(η) (α)
(3.10) − div (|z|γ ∇fα,η ) = cd,s (δ0 − δ0 ).
For any configuration XN = (x1 , . . . , xN ), we define for any i the minimal distance
1
 
1
(3.11) ri = min min |xi − xj |, N − d .
4 j6=i
For any ~η = (η1 , . . . , ηN ) ∈ RN and measure µ, we define the electric potential
N
ˆ !
µ
X
(3.12) HN [XN ] = g(x − y)d δxi − N µδRd (y)
Rd+k i=1
and the truncated potential
N
ˆ !
µ
δx(ηi i )
X
(3.13) HN,~
η [XN ] = g(x − y)d − N µδRd (y),
Rd+k i=1
where we will quickly drop the dependence in XN . We note that
N
µ µ
X
(3.14) HN,~
η [XN ] = HN [XN ] − fη (x − xi ).
i=1

These functions are viewed in the extended space Rd+k as described in the previous subsection,
and solve
N
!
γ µ
X
(3.15) − div (|z| ∇HN ) = cd,s δxi − N µδRd in Rd+k ,
i=1
and
N
!
γ µ
δx(ηi i )
X
(3.16) − div (|z| ∇HN,~
η) = cd,s − N µδRd in Rd+k .
i=1
The following proposition shows how to express FN in terms of the truncated electric fields
µ
∇HN,~
η . In addition, we show that the quantities
ˆ N
µ 2
|z|γ |∇HN,~
X
η | − cd,s g(ηi )
Rd+k i=1

converge almost monotonically (i.e. up to a small error) to FN , while the discrepancy between
the two can serve to control the energy of close pairs of points.
14 SYLVIA SERFATY, APPENDIX WITH MITIA DUERINCKX

Proposition 3.3. Let µ be a bounded probability density on Rd and XN be in (Rd )N . We

may re-write FN (XN , µ) as
N
!
1
ˆ
γ µ 2
X
(3.17) FN (XN , µ) := lim |z| |∇HN,~η | − cd,s g(ηi ) ,
cd,s η→0 Rd+k i=1

and we have the bound

X
(3.18) (g(xi − xj ) − g(ηi ))+
i6=j
N N
!
1
ˆ
γ µ 2
X X
≤ FN (XN , µ) − |z| |∇HN,~
η| − g(ηi ) + CN kµkL∞ ηid−s ,
cd,s Rd+k i=1 i=1

for some C depending only on d and s.

The proof, which is an adaptation and improvement of [PS,LS2], is postponed to Section 5.

What
´ makes ourµ main proof work is the ability to find someP choice of truncation ~η such
N
that Rd+k |z|γ |∇HN,~ η |2 (without the renormalizing term −c
d,s i=1 g(ηi )) is controlled by
FN (XN , µ) and the balls B(xi , ηi ) are disjoint. In view of (3.18) the former could easily be
achieved by taking the ηi ’s large enough, say ηi = N −1/d , but the balls would not necessarily
be disjoint. Instead the choice of ηi = ri where ri are the minimal distances as in (3.11) allows
to fulfill both requirements, as seen in the following
Corollary 3.4. We have
N
N
   
s
g(ri ) ≤ C FN (XN , µ) + (1 + kµkL∞ )N 1+ d +
X
(3.19) log N 1(1.5)
i=1
d
and
N
ˆ    
µ 2 s
(3.20) |z|γ |∇HN,~
r | ≤ C FN (XN , µ) + (1 + kµkL )N
∞
1+ d
+ log N 1(1.5)
Rd+k d
for some C depending only on s, d.
Proof. Let us choose ηi = N −1/d for all i in (3.18) and observe that for each i, by definition
(3.11) there exists j 6= i such that (g(|xi − xj |) − g(N −1/d ))+ = g(4ri ) − g(N −1/d ). We may
thus write that
(3.21)
N
1
ˆ
− 1d 1 s
|z|γ |∇HN,~η |2 + N g(N − d ) + O(N kµkL∞ )N d .
X
(g(4ri ) − g(N )) ≤ FN (XN , µ) −
i=1
cd,s Rd+k

from which (3.19) follows after rearranging and using the definition of g.
Let us next choose ηi = ri in (3.18). Using that ri ≤ N −1/d , this yields
N
1
ˆ
s
|z|γ |∇HN,~r |2 +
X
0 ≤ FN (XN , µ) − g(ri ) + O(N kµkL∞ )N d .
cd,s Rd+k i=1

Combining with (3.19), (3.20) follows. 

 MEAN FIELD LIMIT FOR COULOMB FLOWS 15

3.3. Coerciveness of the modulated energy. Here we prove that the modulated energy
does metrize the convergence of µtN to µt .
Proposition 3.5. For any α > 0, there exists λ > 0 depending only on α, d, s, such that for
d−s
any XN ∈ (Rd )N and probability density µ, if ξ ∈ C 0,α (Rd ) ∩ Ḣ 2 (Rd ), we have
ˆ 
λ
ξd (δxi − N µ) ≤ CkξkC 0,α (Rd ) N 1− d
 
(3.22) 

Rd
1
N
  
2
1+ ds
+ Ckξk d−s FN (XN , µ) + (1 + kµkL∞ )N + log N 1(1.5) .
Ḣ 2 (Rd ) d
1
In particular, if N 2 FN (XN , µ) → 0 as N → ∞, we have that
N
1 X
(3.23) δx ⇀ µ in the weak sense.
N i=1 i

Proof. Let ξ be a smooth test function on Rd . Let ξ̄ denote an extension of ξ to Rd+k satisfying
−div (|z|γ ∇ξ̄) = 0 in {z 6= 0}.
By [FKS], |z|γ being a Muckenhoupt A2 weight, the function ξ̄ is in C 0,λ (Rd+k ) for some
λ > 0 depending on the other parameters, with kξ̄kC 0,λ ≤ CkξkC 0,α . This can also be seen
from the Poisson kernel representation given in [CaffSi]. In addition, we also have (this can
be seen in Fourier, see [CaffSi, Section 3.2]
ˆ
(3.24) |z|γ |∇ξ̄|2 = Ckξk2 d−s .
Rd+k Ḣ 2 (Rd )
Using (3.16) let us write for any probability density µ
1
ˆ ˆ   ˆ
µ
(3.25) ξ d (δxi − N µ) = ξ δxi − δx(rii ) − ξ¯ div (|z|γ ∇HN,~r [XN ]).
R d R d c d,s R d+k

For the fist term in the right-hand side we use the Hölder continuity of ξ¯ and the fact that
(r )
δxii is supported in B(xi , ri ) and write
ˆ  N
α

(ri ) 
riλ ≤ CkξkC 0,α (Rd ) N 1− d .
 X
(3.26) 
 ξ δxi − δxi  ≤ CkξkC 0,α (Rd )
Rd i=1
For the second term, we integrate by parts and use the Cauchy-Schwarz inequality to write
ˆ  ˆ 

¯ γ µ   γ µ 
(3.27)  ξ div (|z| ∇HN,~r [XN ]) = 
  |z| ∇ξ̄ · ∇HN,~r 
Rd+k Rd+k
ˆ  1 ˆ 1
2 2
γ 2 γ µ 2
≤ |z| |∇ξ̄| |z| |∇HN,~
r|
Rd+k Rd+k
In view of (3.20) and (3.24) we thus find
ˆ 
µ
ξ¯ div (|z|γ ∇HN,~
 
(3.28)  r [X ])
N 

Rd+k
1
N
  
2
1+ ds
≤ Ckξk d−s FN (XN , µ) + (1 + kµkL∞ )N + log N 1(1.5) .
Ḣ 2 (Rd ) d
Inserting (3.26) and (3.28) into (3.25) we conclude the result. 
16 SYLVIA SERFATY, APPENDIX WITH MITIA DUERINCKX

Remark 3.6. In a density formulation aiming at proving propagation of chaos, arguing

exactly as in [RS, Lemma 8.4] for instance, we may deduce from this result and the main
theorem the convergence of the k-marginal densities in the dual of some Sobolev space, with
rate k/N times the right-hand side of (3.22).
3.4. Proof of Lemma 2.4. Using the Cauchy-Schwarz inequality we have the bound
¨
F ∗ (µt − µtN ) · ∇g(x − y)d(µtN − µt )(y)dµtN (x)
△c
 2 1
ˆ 1 ˆ  ˆ 2
2

≤ |F ∗ (µt − µtN )|2 dµtN P.V. ∇g(x − y)d(µtN t t
− µ )(y) dµN (x)
   
Rd  Rd \{x} 

thus to prove the lemma, it suffices to show that

λ
(3.29) N kF ∗ (µt − µtN )kL∞ ≤ CkFkC 0,α (Rd ) N 1− d
1
N
  
2
1+ ds
+ CkFk d−s FN (XN , µ) + (1 + kµkL∞ )N + log N 1(1.5) ,
Ḣ 2 (Rd ) d
which is a direct consequence of (3.22).

4. Proof of Proposition 2.3

4.1. Stress-energy tensor.
Definition 4.1. For any functions h, f in Rd+k such that Rd+k |z|γ |∇h|2 and Rd+k |z|γ |∇f |2
´ ´
are finite, we define the stress tensor [h, f ] as the (d + k) × (d + k) tensor
(4.1) [h, f ] = |z|γ (∂i h∂j f + ∂i h∂j f ) − |z|γ ∇h · ∇f δij
where δij = 1 if i = j and 0 otherwise.
We note that
Lemma 4.2. If h and f are regular enough, we have
(4.2) div [h, f ] = div (|z|γ ∇h)∇f + div (|z|γ ∇f )∇h − ∇|z|γ ∇h · ∇f
where div T here denotes the vector with components
P
i ∂i Tij , with j ranging from 1 to d + k.
Proof. This is a direct computation. Below, all sums range from 1 to d + k.
X
∂i [h, f ]ij
i
!
γ γ γ γ γ
X X
= [∂i (|z| ∂i h)∂j f + ∂i (|z| ∂i f )∂j h + |z| ∂ij h∂i f + |z| ∂ij f ∂i h] − ∂j |z| ∂i h∂i f
i i
= div (|z|γ ∇h) + div (|z|γ ∇f ) − ∇h · ∇f ∂j |z|γ .


In view of (4.2), we have

 MEAN FIELD LIMIT FOR COULOMB FLOWS 17

Lemma 4.3. Let ψ : Rd → Rd be Lipschitz, and if k = 1 let ψ̂ be an extension of it to a map

from Rd+k to Rd+k , whose last component identically vanishes, which tends to 0 as |z| → ∞
and has the same pointwise and Lipschitz bounds as ψ. 1 For any measures ´ µ, νγ on R
d+k , if
γ µ γ ν µ 2
´−div (|z|γ ∇hν )2 = cd,s µ and −div (|z| ∇h ) = cd,s ν, and assuming that Rd+k |z| |∇h | and
Rd+k |z| |∇h | are finite and the left-hand side in (4.3) is well-defined, we have
1
¨ ˆ
(4.3) (ψ̂(x) − ψ̂(y)) · ∇g(x − y)dµ(x)dν(y) = ∇ψ̂(x) : [hµ , hν ].
R d+k ×R d+k cd,s R d+k

Proof. If µ is smooth enough then we may use (4.2) to write

¨ ˆ
(ψ̂(x) − ψ̂(y)) · ∇g(x − y)dµ(x)dν(y) = ψ̂ · (∇hµ dν + ∇hν dµ)
Rd+k ×Rd+k Rd+k
1
ˆ
= − ψ̂ · div [hµ , hν ]
cd,s Rd+k
since the last component of ψ̂ vanishes identically. Integrating by parts, we obtain
1
¨ ˆ
(ψ̂(x) − ψ̂(y)) · ∇g(x − y)dµ(x)dν(y) = ∇ψ̂ : [hµ , hν ].
R d+k ×R d+k cd,s R d+k

By density, we may extend this relation to all measures µ, ν such that both sides of (4.3)
make sense. 
4.2. Proof of Proposition 2.3. We now proceed to the proof. Given the Lipschitz map
ψ : Rd → Rd , we choose an extension ψ̂ to Rd+k which satisfies the same conditions as in
Lemma 4.3.
Step 1: renormalizing the quantity and expressing it with the stress-energy tensor. Clearly,
¨ N
X XN
(4.4) (ψ(x) − ψ(y)) · ∇g(x − y)d( δxi − N µ)(x)d( δxi − N µ)(y)
△c i=1 i=1
h¨   N N
δx(η) δx(η)
X X
= lim ψ̂(x) − ψ̂(y) · ∇g(x − y)d( i
− N µδRd )(x)d( i
− N µδRd )(y)
η→0 Rd+k ×Rd+k i=1 i=1
N ¨   i
ψ̂(x) − ψ̂(y) · ∇g(x − y)dδx(η) (x)dδx(η)
X
− i i
(y) .
i=1 Rd+k ×Rd+k

Applying Lemma 4.3, in view of (3.13), (3.16) we find that

¨   N N
δx(η) δx(η)
X X
(4.5) ψ̂(x) − ψ̂(y) · ∇g(x − y)d( i
− N µδRd )(x)d( i
− N µδRd )(y)
Rd+k ×Rd+k i=1 i=1
1
ˆ
µ µ
= ∇ψ̂ : [HN,~
η , HN,~
η ].
cd,s Rd+k

Step 2: analysis of the diagonal terms. The main point is to understand how they vary
with ηi . Let α
~ be such that αi ≥ ηi for every i.
1Such an extension exists, for instance by solving the ∞-Laplacian in a strip, which provides an “absolutely
minimal Lipschitz extension”
18 SYLVIA SERFATY, APPENDIX WITH MITIA DUERINCKX

We may write that

¨  
(4.6) ψ̂(x) − ψ̂(y) · ∇g(x − y)dδx(ηi i ) (x)dδx(ηi i ) (y)
Rd+k ×Rd+k
¨  
− ψ̂(x) − ψ̂(y) · ∇g(x − y)dδx(αi i ) (x)dδx(αi i ) (y)
Rd+k ×Rd+k
¨
= (ψ̂(x) − ψ̂(y)) · ∇g(x − y)d(δx(ηi i ) − δx(αi i ) )(x)d(δx(ηi i ) − δx(αi i ) )(y)
R d+k ×R d+k
¨
+2 (ψ̂(x) − ψ̂(y)) · ∇g(x − y)dδx(αi i ) (x)d(δx(ηi i ) − δx(αi i ) )(y).
Rd+k ×Rd+k

We claim that
¨
(4.7) (ψ̂(x) − ψ̂(y)) · ∇g(x − y)dδx(αi i ) (x)d(δx(ηi i ) − δx(αi i ) )(y) = 0.
Rd+k ×Rd+k

Assuming this, inserting it to (4.6) and using (3.9) and Lemma 4.3, we conclude that
¨  
(4.8) ψ̂(x) − ψ̂(y) · ∇g(x − y)dδx(η) i
(x)dδx(η)
i
(y)
Rd+k ×Rd+k
¨  
− ψ̂(x) − ψ̂(y) · ∇g(x − y)dδx(αi i ) (x)dδx(αi i ) (y)
Rd+k ×Rd+k
1
ˆ
= ∇ψ̂ : [fαi ,ηi (· − xi ), fαi ,ηi (· − xi )].
cd,s Rd+k

Step 3: proof of (4.7). Let us write the quantity in (4.7) as

¨
(4.9) 2 (ψ̂(x) − ψ̂(y)) · ∇g(x − y)dδx(2α
i
i)
(x)d(δx(ηi i ) − δx(αi i ) )(y)
R d+k ×R d+k
¨    
+2 (ψ̂(x) − ψ̂(y)) · ∇g(x − y)d δx(αi i ) − δx(2α
i
i)
(x)d δx
(ηi )
i
− δx
(αi )
i
(y).
Rd+k ×Rd+k
(2α ) (η ) (α )
In view of (3.7), ∇g ∗ δxi i = ∇g2αi (· − xi ) and in view of (3.9), ∇g ∗ (δxi i − δxi i ) =
∇fαi ,ηi (x − xi ). We may thus rewrite the first term in (4.9) as
ˆ ˆ  
(2αi )
2P.V. ψ̂ · ∇fαi ,ηi (· − xi )dδxi + 2P.V. ψ̂ · ∇g2αi (· − xi )d δx(ηi i ) − δx(αi i ) .
Rd+k Rd+k
(2α )
But ∇fαi ,ηi (· − xi ) is supported in B(xi , αi ) while δxi i is supported on ∂B(xi , 2αi ), and
(η ) (α )
in the same way ∇gαi (· − xi ) vanishes in B(xi , 2αi ) where δxi i − δxi i is supported, so we
conclude that the first term in (4.9) is zero. The second term in (4.9) is equal by (3.9) and
Lemma 4.3 to
1
ˆ
∇ψ̂ : [f2αi ,αi (· − xi ), fαi ,ηi (· − xi )]
cd,s Rd+k
and it is zero, since f2αi ,αi and fαi ,ηi have disjoint supports. This finishes the proof of (4.7).
Step 4: combining (4.5) and (4.8). The following lemma allows to recombine the terms
obtained at different values of ηi while making only a small error.
 MEAN FIELD LIMIT FOR COULOMB FLOWS 19

Lemma 4.4. Assume that µ ∈ C σ (Rd ) with σ > s − d + 1 if s ≥ d − 1. Assume µ ∈ L∞ (Rd )

or µ ∈ C σ (Rd ) with σ > 0 if s < d − 1. If for each i we have ηi < αi ≤ ri , then
ˆ ˆ N ˆ
µ µ µ µ
X
∇ψ̂ : [HN,~
η , HN,~
η] = ∇ψ̂ : [HN,~
α , HN,~
α ]+ ∇ψ̂ : [fαi ,ηi (·−xi ), fαi ,ηi (·−xi )]+E
Rd+k Rd+k i=1 Rd+k

with

N
   
2− s+1 s
(4.10) |E| ≤ Ck∇ψkL∞ N 2 + FN (XN , µ) + log N 1(1.5) + (1 + kµkL∞ )N 1+ d
d
N N N
!
X X X
+CN min kψkL∞ kµkL∞ αid−s−1 + k∇ψkL∞ kµkL∞ αd−s
i , kψkW 1,∞ kµkC σ αd−s+σ−1
i
i=1 i=1 i=1
( PN
k∇ψkL∞ (1 + kµkC σ ) i=1 αi if s ≥ d − 1
+ CN
k∇ψkL∞ (1 + kµkL∞ ) N
P
i=1 αi if s < d − 1

where C depends only on s, d.

Assuming this, and combining (4.4), (4.5) and (4.8) we find that for any αi ≤ ri ,

N N
1
¨ ˆ
µ µ
X X
(ψ(x) − ψ(y))·∇g(x−y)d( δxi −N µ)(x)d( δxi −N µ)(y) = ∇ψ̂ : [HN,~
α , HN,~
α]
△c i=1 i=1
cd,s Rd+k
N ¨  
ψ̂(x) − ψ̂(y) · ∇g(x − y)dδx(αi i ) (x)dδx(αi i ) (y) + O(E)
X
−
i=1 Rd+k ×Rd+k

where E is as in (4.10). Using the Lipschitz character of ψ and the expression of g, we find
that the second term on the right-hand side can be bounded by 2

N ¨
g(x − y)dδx(αi i ) (x)dδx(αi i ) (y)
X
Ck∇ψkL∞
i=1 Rd+k ×Rd+k
N ˆ N
gαi (· − xi )dδx(αi i ) = Ck∇ψkL∞
X X
= Ck∇ψkL∞ g(αi )
i=1 Rd+k i=1

µ µ
where we have used (3.7). Choosing finally αi = ri ≤ N −1/d , bounding pointwise [HN,~ r , HN,~r ]
µ 2 PN
by 2|z|γ |∇HN,~
r | and using (3.20), while using (3.19) to bound i=1 g(ri ), we conclude the
proof of Proposition 2.3.

µ µ µ µ
Proof of Lemma 4.4. First, we observe from (3.14) that [HN,~ α , HN,~
α ] and [HN,~
η , HN,~η ] only
differ in the balls B(xi , αi ) which are disjoint since αi ≤ ri , and that in each B(xi , αi ) we have
µ µ
HN,~
η = HN,~
α + fαi ,ηi (· − xi ).

2In the case (1.5) we bound instead |x − y||∇g(x − y)| by 1, which yields an even better control.
20 SYLVIA SERFATY, APPENDIX WITH MITIA DUERINCKX

We thus deduce that

ˆ  
µ µ µ µ
(4.11) ∇ψ̂ : [HN,~ η , HN,~η ] − [HN,~ α , HN,~ α]
B(xi ,αi )
ˆ  
µ
= ∇ψ̂ : [fαi ,ηi , fαi ,ηi ](· − xi ) + 2[fαi ,ηi (· − xi ), HN,~α]
B(xi ,αi )
ˆ  
µ
= ∇ψ̂ : [fαi ,ηi , fαi ,ηi ](· − xi ) + 2[fαi ,ηi (· − xi ), HN,~
α .
]
Rd+k
There only remains to control the second part of the right-hand side. By Lemma 4.3, we have
ˆ
µ
∇ψ̂ : [fαi ,ηi (· − xi ), HN,~
α]
Rd+k
 
¨ N
(α )
 
δxj j − N µδRd  (x)d δx(ηi i ) − δx(αi i ) (y).
X
= cd,s (ψ̂(x) − ψ̂(y)) · ∇g(x − y)d 
Rd+k ×Rd+k j=1

In view of (4.7), we just need to bound the sum over i of

(4.12)
 
¨
(α )
 
δxj j − N µδRd  (x)d δx(ηi i ) − δx(αi i ) (y)
X
cd,s (ψ̂(x) − ψ̂(y)) · ∇g(x − y)d 
Rd+k ×Rd+k j:j6=i
 
ˆ X (α )
= cd,s ψ̂ · ∇fαi ,ηi (· − xi )d  δxj j − N µδRd 
Rd+k j:j6=i
 
ˆ  
∇gαj (x − xj ) − N ∇hµ  d δx(ηi i ) − δx(αi i ) ,
X
+ cd,s ψ̂ · 
Rd+k j:j6=i

where we used (3.9).

(α )
Step 1: first term in (4.12). Since fαi ,ηi (· − xi ) is supported in B(xi , αi ), δxj j in B(xj , αj )
and the balls are disjoint, one type of terms vanishes and there remains
ˆ
−N cd,s ψ · ∇fαi ,ηi (· − xi )dµ.
Rd
Thanks to the explicit form of fα,η we have
(
g(x − xi ) − g(αi ) for η < |x − xi | ≤ αi
fαi ,ηi (· − xi ) =
g(ηi ) − g(αi ) for |x − xi | ≤ ηi
and
∇fαi ,ηi (· − xi ) = ∇g(x − xi )1ηi ≤|x−xi |≤αi .
It follows that
ˆ ˆ ˆ
(4.13) |fα | ≤ Cα d−s
, |fαi ,ηi | ≤ Cαd−s
i , |∇fαi ,ηi | ≤ Cαid−s−1 .
Rd Rd Rd
Indeed, it suffices to observe that
ˆ α
C r ′
ˆ ˆ
(4.14) fα = C (g(r) − g(α))r d−1 dr = − g (r)r d dr,
B(0,η) 0 d 0
 MEAN FIELD LIMIT FOR COULOMB FLOWS 21

with an integration by parts.

We may always write
αi
rd
ˆ  ˆ
 
(4.15)  ψ · ∇fαi ,ηi (· − xi )(µ − µ(xi )) ≤ CkψkL∞ kµkC 0,1 dr
Rd ηi r s+1
≤ CkψkL∞ kµkC 0,1 αd−s
i ,
and, integrating by parts and using (4.13),
ˆ  ˆ
|fαi ,ηi | ≤ kµkL∞ k∇ψkL∞ αd−s
 
(4.16) 
 ψ · ∇fαi ,ηi (· − xi )µ(xi ) ≤ kµkL∞ k∇ψkL∞

i .
Rd Rd

Alternatively, we may use the simpler bound derived from (4.13),

ˆ 
ψ · ∇fαi ,ηi (· − xi )dµ ≤ CkψkL∞ kµkL∞ αid−s−1 .
 
(4.17) 

Rd

A standard interpolation argument yields that kgk(C σ )∗ ≤ kgkσ(C 1 )∗ kgk1−σ

(C 0 )∗ so interpolating
between (4.15)–(4.16) and (4.17), we obtain
ˆ 
ψ · ∇fαi ,ηi (· − xi )dµ ≤ Ckψk1−σ σ d−s+σ−1
 
L∞ kψkW 1,∞ kµkC σ αi .


Rd

We conclude that the sum over i of the first terms in (4.12) is bounded by both
1−σ σ
X X
(4.18) CkψkL∞ kψkW 1,∞ kµkC σ αid−s+σ−1 and CkψkL∞ kµkL∞ αid−s−1 .
i i

Step 2: second term in (4.12). We may rewrite the integral as

ˆ  
(4.19) − N ψ̂(xi ) · ∇Rd hµ d δx(ηi i ) − δx(αi i )
Rd+k
ˆ  
ψ̂(xj ) · ∇gαj (x − xj )d δx(ηi i ) − δx(αi i )
X
+
d+k
j:j6=i R
ˆ  
(ψ̂ − ψ̂(xj )) · ∇gαj (x − xj )d δx(ηi i ) − δx(αi i )
X
+
d+k
j:j6=i R
 ˆ  
µ
+ O N k∇ψk L∞ |x − xi | |∇Rd h | d δx(ηi i ) + δx(αi i ) ,
Rd+k

where we used that the last component of ψ̂ vanishes, so that only the derivatives along the
Rd directions appear.
(η ) (α ) 1
Substep 2.1: first term of (4.19). We may write that δxi i −δxi i = − cd,s div (|z|γ ∇fαi ,ηi (·−
xi )) and integrate by parts twice to get
1
ˆ   ˆ
γ µ
|z| ∇ ψ̂(xi ) · ∇Rd h · ∇fαi ,ηi (· − xi ) = (ψ(xi ) · ∇µ)fαi ,ηi .
cd,s Rd+k Rd

Here, we used that −div (|z|γ ∇hµ ) = cd,s µδRd and took the ψ̂(xi ) · ∇Rd of this relation. In
view of (4.13), this is then bounded by
CkψkL∞ k∇µkL∞ αd−s
i .
22 SYLVIA SERFATY, APPENDIX WITH MITIA DUERINCKX

Alternatively, we may integrate by parts in Rd to bound it by

 ˆ 
CkµkL∞ kψkL∞ |∇fαi ,ηi | + k∇ψkL∞ αd−s
i .
Rd
Interpolating as above, we conclude with (4.13) that the sum over i of these terms is bounded
by both
!
kψkσL∞ k∇ψk1−σ
X
d−s
(4.20) CkµkC σ L∞ αi + kψkL∞ αd−s+σ−1
i
i
!
X X
and kµkL∞ kψkL∞ αid−s−1 + k∇ψkL∞ αd−s
i .
i i

Substep 2.2: second term of (4.19). Arguing in the same way as for the first term, using
(α )
that −div (|z|γ ∇gαj (· − xj )) = cd,s δxj j and the disjointness of the balls, we find that this
term vanishes.
Substep 2.3: third term in (4.19). We separate the sum into two pieces and bound this
term by
ˆ  
(ψ̂ − ψ̂(xj )) · ∇gαj (· − xj )d δx(ηi i ) − δx(αi i )
X
(4.21)
−1
Rd+k
j6=i,|xi −xj |≥N d
ˆ  
|x − xj ||∇gαj (x − xj )|d δx(ηi i ) + δx(αi i ) (x)
X
+ k∇ψkL∞
−1
Rd+k
j6=i,|xi −xj |≤N d

1
For the first term of (4.21), we may use that |xi − xj | ≥ N − d to write
s+1
 
k∇ ψ̂ − ψ̂(xj ) · ∇gαj (x − xj )kL∞ (B(xi ,ri )) ≤ Ck∇ψkL∞ N − d

and we may thus bound it by

s+1
CN k∇ψkL∞ N − d .
Since |∇gα | ≤ |∇g|, we may bound the second term in (4.21) by
|xi − xj |−s .
X
(4.22) k∇ψkL∞
−1
j6=i,|xi−xj |≤N d

1
To bound this, let us choose ηi = 2N − d in (3.18) to obtain that
X 1 1 s

g(xi − xj ) − g(2N − d ) ≤ FN (XN , µ) + N g(2N − d ) + CkµkL∞ N 1+ d .
+
i6=j

In the cases (1.4), it follows that

s
g(xi − xj ) ≤ FN (XN , µ) + C(1 + kµkL∞ )N 1+ d .
X

−1
i6=j,|xi−xj |≤N d

In the cases (1.5), it follows that

N s
log N + CkµkL∞ N 1+ d ,
X
log 2 ≤ FN (XN , µ) +
d
−1
i6=j,|xi −xj |≤N d
 MEAN FIELD LIMIT FOR COULOMB FLOWS 23

and this suffices to bound (4.22) as well. We conclude in all cases that the sum over i of the
third terms in (4.19) is bounded by
N
   
s+1 s
2− 1+
(4.23) Ck∇ψkL∞ N d + FN (XN , µ) + (1 + kµkL∞ )N d + log N 1(1.5) .
d
 
Substep 2.4: fourth term in (4.19). We may bound it by O k∇ψkL∞ αi k∇Rd hµ kL∞ (Rd+k ) .
But since hµ = g ∗ µ, it is straightforward to check that k∇Rd hµ kL∞ (Rd+k ) ≤ k∇hµ kL∞ (Rd ) .
Using (3.4), we conclude the sum of these terms is bounded by
(
C i αi k∇ψkL∞ (1 + kµkL∞ ) if s < d − 1
P
(4.24) P
C i αi k∇ψkL∞ (1 + kµkC σ ) if s ≥ d − 1 and σ > s − d + 1.
Combining the bounds (4.18), (4.20), (4.23), (4.24) concludes the proof of the lemma. 

5. Proof of Proposition 3.3

We drop the superscripts µ. First, Rd+k |z|γ |∇HN,~η |2 is a convergent integral and
´

(5.1)
N N
ˆ ¨ ! !
γ 2 (ηi ) (ηi )
X X
|z| |∇HN,~η | = cd,s g(x−y)d δxi − N µδRd (x)d δxi − N µδRd (y).
Rd+k Rd+k ×Rd+k i=1 i=1
Indeed, we may choose R large enough so that all the points of XN are contained in the ball
BR = B(0, R) in Rd+k . By Green’s formula and (3.16), we have
N
!
∂HN
ˆ ˆ ˆ
γ 2 γ (ηi )
X
|z| |∇HN,~η | = |z| HN,~η + cd,s HN,~η d δxi − N µδRd .
BR ∂BR ∂n BR i=1

Since d( i δxi − N µ) = 0, the function HN,~η decreases like 1/|x|s+1 and ∇HN,~η like 1/|x|s+2
´ P
as |x| → ∞, hence the boundary integral tends to 0 as R → ∞, and we may write
N
ˆ ˆ !
γ 2 (ηi )
X
|z| |∇HN,~η | = cd,s HN,~η d δxi − N µδRd
Rd+k Rd+k i=1
and thus by (3.16), (5.1) holds. We may next write that
N N N
! !
h¨ i
δx(ηi i ) δx(ηi i )
X X X
lim g(x − y)d − N µδRd (x)d − N µδRd (y) − g(ηi )
η→0 Rd+k ×Rd+k i=1 i=1 i=1
N N
¨ ! !
X X
=− g(x − y)d δxi − N µδRd (x)d δxi − N µδRd (y)
△c i=1 i=1
and we deduce in view of (5.1) that (3.17) holds.
We next turn to the proof of (3.18), adapted from [PS]. From (3.16) applied with ~η = α ~
PN
and in view of (3.13), we have ∇HN,~η = ∇HN,~α + i=1 ∇fαi ,ηi (· − xi ) thus
ˆ ˆ ˆ
γ 2 γ 2
|z|γ ∇fαi ,ηi (x − xi ) · ∇fαi ,ηi (x − xj )
X
|z| |∇HN,~η | = |z| |∇HN,~α | +
Rd+k Rd+k i,j Rd+k
N ˆ
|z|γ ∇fαi ,ηi (x − xi ) · ∇HN,~α .
X
+2
i=1 Rd+k
24 SYLVIA SERFATY, APPENDIX WITH MITIA DUERINCKX

Using (3.10), we first write

ˆ
|z|γ ∇fαi ,ηi (x − xi ) · ∇fαi ,ηi (x − xj )
X

i,j Rd+k
ˆ ˆ
γ
fαi ,ηi (x − xi )d(δx(ηji ) − δx(αj i ) ).
X X
=− fαi ,ηi (x − xi )div (|z| ∇fαi ,ηi (x − xj )) = cd,s
i,j Rd+k i,j Rd+k

Next, using (3.16), we write

N ˆ N ˆ
|z|γ ∇fαi ,ηi (x − xi ) · ∇HN,~α = −2 fαi ,ηi (x − xi )div (|z|γ ∇HN,~α )
X X
2
i=1 Rd+k i=1 Rd+k
N ˆ N
X 
δx(αj i ) − µδRd .
X
= 2cd,s fαi ,ηi (x − xi ) d
i=1 Rd+k j=1
These last two equations add up to give a right-hand side equal to
ˆ N ˆ
X
(αi ) (ηj ) X
(5.2) cd,s fαi ,ηi (x − xi )d(δxj + δxj ) − 2cd,s fαi ,ηi (x − xi )dµδRd
i6=j Rd+k i=1
ˆ
(αi ) (η )
+ N cd,s fαi ,ηi d(δ0 + δ0 i ) .
Rd+k
´ (α ) (η ) ´ (α )
We then note that fαi ,ηi d(δ0 i + δ0 i ) = − fηi dδ0 i = −(g(αi ) − g(ηi )) by definition of fη
(α)
and the fact that
´ δ0 is a measure supported on ∂B(0, α) and of mass 1. Secondly, by (4.13)
d−s
we may bound Rd fαi ,ηi (x − xi )µδRd by CkµkL∞ αi .
Thirdly, we observe that since fαi ,ηi ≤ 0, the first term in (5.2) is nonpositive and we may
bound it above by
ˆ ˆ
X (αj ) X (α )
cd,s fαi ,ηi dδxj ≤ cd,s (gηi (x − xi ) − gαi (x − xi )) dδxj j
i6=j Rd+k i6=j Rd+k
X
ˆ
≤ cd,s (g(ηi ) − gαi (|xi − xj | + αj ))−
i6=j Rd+k

where we used the fact that gα is radial decreasing. Combining the previous relations yields
N
X X
− CN kµkL∞ ηid−s + cd,s (gαi (|xi − xj | + αj ) − g(ηi ))+
i=1 i6=j
N N
ˆ ! ˆ !
γ 2 γ 2
X X
≤ |z| |∇HN,~α | − cd,s g(αi ) − |z| |∇HN,~η | − cd,s g(ηi )
Rd+k i=1 Rd+k i=1
and letting all αi → 0 finishes the proof in view of (3.17).
 MEAN FIELD LIMIT FOR COULOMB FLOWS 25

Appendix A. Mean-field limit for monokinetic Vlasov systems,

with Mitia Duerinckx
In this appendix, we turn to examining the mean-field limit of solutions of Newton’s second-
order system of ODEs, that is,

ẋi = vi ,


(A.1) v̇ = − 1 ∇ H (x , . . . , xN ),
i N xi N 1
i = 1, . . . , N

x (0) = x0 , v (0) = v 0 ,

i i i i

where HN is the interaction energy defined in (1.3). We then consider the phase-space em-
pirical measure
N
t 1 X
fN := δ t t ∈ P(Rd × Rd ),
N i=1 (xi ,vi )
where P denotes probability measures, associated to a solution ZN t := ((xt , v t ), . . . , (xt , v t ))
1 1 N N
0 0
of the flow (A.1). If the points (xi , vi ), which themselves depend on N , are such that fN 0

converges to some regular measure f 0 , then formal computations indicate that for t > 0, fN t
0
should converge to the solution of the Cauchy problem with initial data f for the following
Vlasov equation,
ˆ
(A.2) ∂t f + v · ∇x f + (∇g ∗ µ) · ∇v f = 0, µt (x) := f t (x, v) dv.
Rd
In the case of a smooth interaction kernel g, such a mean-field result was first established
in the 1970s by a weak compactness argument [NW, BH], while Dobrushin [Do] gave the
first quantitative proof in 1-Wasserstein distance. In recent years much attention has been
given to the physically more relevant case of singular Coulomb-like kernels, but only very
partial results have been obtained. In [HJ2, HJ1], exploiting a Grönwall argument on the
∞-Wasserstein distance between fN and f , Hauray and Jabin treated all interaction kernels
g satisfying |∇g(x)| ≤ C|x|−s and |∇2 g(x)| ≤ C|x|−s−1 for some s < 1. In [JW1], Jabin
and Wang introduced a new approach, allowing them to treat all interaction kernels with
bounded gradient. The same problem has been addressed in [BP, La, LP], leading to results
that require a small N -dependent cutoff of the interaction kernel.
In this appendix, we show that the method presented in the article allows to unlock the
mean-field problem with Coulomb-like interaction in the simpler case of monokinetic data,
that is, if there exists a regular velocity field u0 : Rd → Rd such that vi0 ≈ u0 (x0i ) for
i = 1, . . . , N , which implies that the solutions remain “monokinetic”, at least for short times.
Justifying a complete mean-field result for non-monokinetic solutions in the Coulomb case
in dimensions d ≥ 2 remains one of the main open problems in the field. It is expected to
be rendered difficult by the possibility of concentration and filamentation in both space and
velocity variables.
In the monokinetic setting, we focus on the (spatial) empirical measure
N
1 X
µtN := δ t ∈ P(Rd ),
N i=1 xi
t of the flow (A.1). If the points x0 are such that µ0 converges to
associated to a solution ZN i N
0
some regular measure µ , then formal computations (see for instance [LZ]) lead to expecting
26 SYLVIA SERFATY, APPENDIX WITH MITIA DUERINCKX

that for t > 0, µtN converges to the solution µt of the Cauchy problem with initial data (µ0 , u0 )
of the following monokinetic version of (A.2)3,
(A.3) ∂t µ + div (µu) = 0, ∂t u + u · ∇u = −∇g ∗ µ.
In the Coulomb case, these equations are known as the (pressureless) Euler-Poisson system.
Since for the second-order system (A.1) the total energy splits into a potential and a kinetic
part, we naturally introduce a modulated total energy taking both parts into account: for
ZN := ((x1 , v1 ), . . . , (xN , vN )) and for a couple (µ, u) ∈ P(Rd ) × C(Rd ), we set
N
|u(xi ) − vi |2
X
HN (ZN , (µ, u)) := N
i=1
¨ N
X  N
X 
+ g(x − y) d δxi − N µ (x) d δxi − N µ (y),
Rd ×Rd \△ i=1 i=1

and we view t , (µt , ut ))

HN (ZN as a good notion of distance between µtN and µt that is adapted
to the monokinetic setting. The condition HN (ZN t , (µt , ut )) = o(N 2 ) indeed entails the weak

convergence µtN ⇀ µt as in Proposition 3.5, while due to the kinetic part it also implies that
the flow ZNt remains monokinetic, that is, v t ≈ ut (xt ). In parallel with Theorem 1, our main
i i
result is a Gronwall inequality on the time-derivative of HN (ZN t , (µt , ut )), which implies a
t t
quantitative rate of convergence of µN to µ in that metric.
Theorem 2. Assume that g is of the form (1.4) or (1.5). Assume that (A.3) admits a
solution (µ, u) in [0, T ] × Rd for some T > 0, with µ ∈ L∞ ([0, T ]; P ∩ L∞ (Rd )) and u ∈
L∞ ([0, T ]; W 1,∞ (Rd )d ). In the case s ≥ d − 1, also assume µ ∈ L∞ ([0, T ]; C σ (Rd )) for some
σ > s − d + 1. Let ZN t solve (A.1). Then there exist constants C , C depending only on
1 2
controlled norms of (µ, u) and on d, s, σ, and there exists an exponent β < 2 depending only
on d, s, σ, such that for every t ∈ [0, T ] we have
 
t
HN (ZN , (µt , ut )) ≤ HN (ZN
0
, (µ0 , u0 )) + C1 N β eC2 t .

In particular, if µ0N ⇀ µ0 and is such that

1 0
lim HN (ZN , µ0 ) = 0,
N →∞ N 2

then the same is true for every t ∈ [0, T ], hence µtN ⇀ µt in the weak sense.
Remark A.1. We briefly comment on the regularity assumptions on the solution (µ, u) of
the mean-field system (A.3) in the Coulomb case. While global existence of a weak solution is
known in some settings [CW, Gu, IP], existence of a smooth solution can in general only hold
locally in time due to the possible wave breakdown [Pe, En]. The local-in-time existence of a
smooth solution has then been established in [Ma, MU, Ga], ensuring that the above regularity
assumptions on (µ, u) indeed hold locally in time for smooth initial data. In addition, a
critical threshold phenomenon was discovered by Engelberg, Liu, and Tadmor [ELT,LT]: global
existence of a smooth solution actually holds whenever the initial data satisfies some critical
condition, while finite-time breakdown occurs otherwise. For some initial data, the required
regularity assumptions on (µ, u) may thus even hold globally in time.
3Formally, the pair (µ, u) indeed satisfies the system (A.3) if and only if the monokinetic ansatz f t (x, v) :=
µt (x)δv=ut (x) satisfies the Vlasov equation (A.2).
 MEAN FIELD LIMIT FOR COULOMB FLOWS 27

Before turning to the proof of Theorem 2, we start with a weak-strong stability principle
for the mean-field system (A.3), analogous to (1.16) for the mean-field equation (1.8) in the
first-order case.

Lemma A.2. Let (µt1 , ut1 ) and (µt2 , ut2 ) denote two smooth solutions of (A.3) in [0, T ] × Rd .
Then there is a constant C depending only on d such that, in terms of
ˆ ˆ ˆ
H((µt1 , ut1 ), (µt2 , ut2 )) := |ut1 − ut2 |2 µt1 + g(x − y) d(µt1 − µt2 )(x) d(µt1 − µt2 )(y),
Rd Rd Rd

we have
´t
k∇uu
H((µt1 , ut1 ), (µt2 , ut2 )) ≤ eC 0 2 kL∞ du H((µ01 , u01 ), (µ02 , u02 )).

Proof. Let hi := g ∗ µi . We compute

ˆ ˆ
2
∂t H((µ1 , u1 ), (µ2 , u2 )) = |u1 − u2 | ∂t µ1 + 2 (u1 − u2 ) · (∂t u1 − ∂t u2 ) µ1
Rd Rd
ˆ
+2 (h1 − h2 )(∂t µ1 − ∂t µ2 )
Rd
ˆ ˆ
2
= ∇|u1 − u2 | · µ1 u1 − 2 (u1 − u2 ) · (u1 · ∇u1 − u2 · ∇u2 ) µ1
Rd Rd
ˆ ˆ
−2 ∇(h1 − h2 ) · (u1 − u2 ) µ1 + 2 ∇(h1 − h2 ) · (µ1 u1 − µ2 u2 ).
Rd Rd

Decomposing 2(u1 − u2 ) · (u1 · ∇u1 − u2 · ∇u2 ) = u1 · ∇|u1 − u2 |2 + 2∇u2 : (u1 − u2 ) ⊗ (u1 − u2 ),

we find after straightforward simplifications,

∂t H((µ1 , u1 ), (µ2 , u2 ))
ˆ ˆ
= −2 µ1 ∇u2 : (u1 − u2 ) ⊗ (u1 − u2 ) + 2 u2 · ∇(h1 − h2 )(µ1 − µ2 ).
Rd Rd

In the Coulomb case, we recognize from (1.19) in the second right-hand side term the diver-
gence of the stress-energy tensor [h1 − h2 , h1 − h2 ], hence

∂t H((µ1 , u1 ), (µ2 , u2 ))
1
ˆ ˆ
= −2 µ1 ∇u2 : (u1 − u2 ) ⊗ (u1 − u2 ) − u2 · div [h1 − h2 , h1 − h2 ]
Rd cd Rd
C
ˆ ˆ
≤ Ck∇u2 kL∞ |u1 − u2 |2 µ1 + k∇u2 kL∞ |∇(h1 − h2 )|2
R d cd R d

= Ck∇u2 kL∞ H((µ1 , u1 ), (µ2 , u2 )),

and the result follows by Gronwall’s lemma. In the Riesz case, replacing (1.19) by (3.2) and
by (4.1)–(4.2), the same conclusion follows. 

Taking inspiration from the above calculations for the weak-strong stability principle, we
now turn to the proof of Theorem 2.
28 SYLVIA SERFATY, APPENDIX WITH MITIA DUERINCKX

Proof of Theorem 2. Let h := g ∗ µ. We decompose

N
X

∂t HN (ZN , (µ, u)) = 2N (u(xi ) − vi ) · (∂t u)(xi ) + ẋi · ∇u(xi ) − v̇i
i=1
ˆ N ˆ N ˆ
ẋi ·∇g(xi −xj )+2N 2
X X X
+2 h∂t µ−2N ẋi ·∇g(xi −y) dµ(y)−2N g(xi −·) ∂t µ.
i6=j Rd i=1 Rd i=1 Rd

Inserting equations (A.1) and (A.3), we obtain

∂t HN (ZN , (µ, u))
N
X  1 X 
= 2N (u(xi ) − vi ) · − (u · ∇u)(xi ) − ∇h(xi ) + vi · ∇u(xi ) + ∇g(xi − xj )
i=1
N j:j6=i
ˆ
2
X
+2 vi · ∇g(xi − xj ) + 2N ∇h · uµ
i6=j Rd
N
X N
X
ˆ
− 2N vi · ∇h(xi ) − 2N ∇g(· − xi ) · u dµ,
i=1 i=1 Rd

and hence, after straightforward simplifications,

N
X
∂t HN (ZN , (µ, u)) = −2N ∇u(xi ) : (u(xi ) − vi ) ⊗ (u(xi ) − vi )
i=1
¨
2
+N (u(x) − u(y)) · ∇g(x − y) d(µN − µ)(x) d(µN − µ)(y).
Rd ×Rd \△

Applying Proposition 2.3 to estimate the second right-hand side term, we are led to

∂t HN (ZN , (µ, u))

 s s+1
N  
≤ Ck∇ukL∞ HN (ZN , (µ, u)) + (1 + kµkL∞ )N 1+ d + N 2− 2 + log N 1(1.5)
d
2− 1d

1+ s+1−σ
(1 + kµkC )k∇ukL N
σ ∞ + kukW 1,∞ kµkC N
σ d , if s ≥ d − 1,


+C 2− 1d 1+ s+1
(1 + kµk L∞ )k∇uk L∞ N + kuk kµk N
L∞ L∞ d
 s
+k∇ukL∞ kµkL∞ N 1+ d , if s < d − 1,



and the conclusion follows by Gronwall’s lemma. 

References
[AS] L. Ambrosio, S. Serfaty, A gradient flow approach for an evolution problem arising in superconductivity,
Comm. Pure Appl. Math 61 (2008), no. 11, 1495–1539.
[BK] H. Bae, J. Kelliher, The vortex patches of Serfati, arXiv:1409.5169.
[BO] R. Berman, M. Onnheim, Propagation of chaos for a class of first order models with singular mean field
interactions, SIAM J. Math. Anal. 51, No 1, 159–196.
[BC] A. Bertozzi, P. Constantin, Global regularity for vortex patches, Comm. Math. Phys., No. 1 152
(1993),19–28.
[BIK] P. Biler, C. Imbert, G. Karch, Barenblatt profiles for a nonlocal porous medium equation, C. R. Acad.
Sci. Paris, Ser. I 349, (2011), 641–645.
[BP] N. Boers, P. Pickl, On mean field limits for dynamical systems, J. Stat. Phys. 164, No. 1 (2016), 1–16.
 MEAN FIELD LIMIT FOR COULOMB FLOWS 29

[BCF] F. Bolley, D. Chafai, J. Fontbona, Dynamics of a planar Coulomb gas, Ann. Appl. Probab. 28, No. 5
(2018), 3152–3183.
[BH] W. Braun, K. Hepp, The Vlasov dynamics and its fluctuations in the 1/N limit of interacting classical
particles, Comm. Math. Phys. 56, No. 2 (1977), 101–113.
[CaffSi] L. Caffarelli, L. Silvestre, An extension problem related to the fractional Laplacian. Comm. PDE 32,
No. 8, (2007), 1245–1260.
[CSV] L. Caffarelli, F. Soria, J.L. Vázquez, Regularity of solutions of the fractional porous medium flow. J.
Eur. Math. Soc. 15, (2013), no. 5, 1701-1746.
[CV] L. Caffarelli, J.L. Vázquez. Nonlinear porous medium flow with fractional potential pressure. Arch. Rat.
Mech. Anal. 202, (2011), no. 2, 537–565.
[CCS] J. A. Carrillo, Y. P. Choi, S. Salem, Propagation of chaos for the VPFP equation with a polynomial
cut-off, arXiv:1802.01929.
[CCH] J.A. Carrillo, Y.P. Choi, M. Hauray, The derivation of swarming models: Mean-field limit and Wasser-
stein distances. In Collective Dynamics from Bacteria to Crowds, Springer, (2014), p. 1–46.
[CFP] J. A. Carrillo, L. C. F. Ferreira, J. C. Precioso, A mass-transportation approach to a one dimensional
fluid mechanics model with nonlocal velocity, Adv. in Math. 231 (2012) 306–327.
[Ch1] J. Y Chemin, Persistance de structures géométriques dans les fluides incompressibles bidimensionnels,
Ann. Sci. Ecole Norm. Sup. 26 (1993), no. 4, 517–542.
[Ch2] J.-Y. Chemin, Perfect incompressible fluids. Oxford Lecture Series in Mathematics and its Applications,
14. T Oxford University Press, 1998.
[CW] G.- Q. Chen, D. Wang, Convergence of shock capturing schemes for the compressible Euler-Poisson
equations, Comm. Math. Phys. 179, No. 2, (1996), 333–364.
[De] J.-M. Delort, Existence de nappes de tourbillon en dimension deux, J. Amer. Math. Soc 4 (1991), 553–586.
[Do] R. L. Dobrushin, Funktsional. Anal. i Prilozhen. Fjournal Akademiya Nauk SSSR. Funktsional′nyui
Analiz i ego Prilozheniya 13 (1979), No. 2, 48–58, 96.
[DZ] Q. Du, P. Zhang, Existence of weak solutions to some vortex density models, SIAM J. Math. Anal. 34
(2003), no. 6, 1279–1299.
[Du] M. Duerinckx, Mean-field limits for some Riesz interaction gradient flows, SIAM J. Math. Anal. 48
(2016), no. 3, 2269–2300.
[En] S. Engelberg, Formation of singularities in the Euler and Euler-Poisson equations, Physica D 98, 67–74.
[ELT] S. Engelberg, H. Liu, E. Tadmor, Critical thresholds in Euler-Poisson equations, Indiana Univ. Math.
J. 50 (2001), 109–157.
[FKS] E. Fabes, C. Kenig, R. Serapioni, The local regularity of solutions of degenerate elliptic equations,
Comm. PDE, 7 No. 1 (1982), 77–116.
[FHM] N. Fournier, M. Hauray, S. Mischler, Propagation of chaos for the 2D viscous vortex model, J. Eur.
Math. Soc. 16 (2014), no. 7, 1423–1466.
[Ga] P. Gamblin, Solution régulière à temps petit pour l’équation d’Euler-Poisson, Comm. PDE 18, No. 5-6,
(1993), 731–745.
[Go] F. Golse, 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, Lect. Notes Appl. Math. Mech. 3
(2016), 1-144.
[GHL] J. Goodman, T. Hou, J. Lowengrub, Convergence of the point vortex method for the 2-D Euler equa-
tions. Comm. Pure Appl. Math. 43 (1990), no. 3, 415–430.
[Gu] Y. Guo, Smooth irrotational flows in the large to the Euler-Poisson system in R3 + 1, Comm. Math.
Phys. 195, No. 2 (1998), 249–265.
[Hau] M. Hauray, Wasserstein distances for vortices approximation of Euler-type equations, Math. Models
Methods Appl. Sci. 19(8) (2009), 1357–1384.
[HM] M. Hauray, S. Mischler, On Kac’s chaos and related problems, JFA 266, No. 10 (2014), 6055–6157.
[HJ1] M. Hauray, P-E. Jabin, Particle approximation of Vlasov equations with singular forces: propagation of
chaos, Ann. Sci. Ec. Norm. Super. (4) 48 (2015), no. 4, 891–940.
[HJ2] M. Hauray, P-E. Jabin, N -particles approximation of the Vlasov equations with singular potential, Arch.
Ration. Mech. Anal. 183, No. 3 (2007), 489–524.
[He] F. Hélein, Harmonic Maps, Conservation Laws and Moving Frames, Cambridge University Press, 2002.
[IP] A. D. Ionescu, B. Pausader, The Euler-Poisson system in 2D: global stability of the constant equilibrium
solution, Inter. Math. Research Not. 4 (2013), 761–826.
30 SYLVIA SERFATY, APPENDIX WITH MITIA DUERINCKX

[Jab] P. E. Jabin, A review of the mean field limits for Vlasov equations. Kinet. Rel. Mod. 7 (2014), 661–711.
[JW1] P. E. Jabin, Z. Wang, Mean Field Limit and Propagation of Chaos for Vlasov Systems with Bounded
Forces, J. Funct. Anal. 271 (2016), no. 12, 3588–3627.
[JW2] P. E. Jabin, Z. Wang, Quantitative estimates of propagation of chaos for stochastic systems with W −1,∞
kernels, Invent. Math. 214, No 1, 523–591.
[La] D. Lazarovici, The Vlasov-Poisson dynamics as the mean field limit of extended charges, Comm. Math.
Phys. 347, No. 1, (2016), 271–289.
[LP] D. Lazarovici, P. Pickl, A mean field limit for the Vlasov-Poisson system. Arch. Ration. Mech. Anal. 225
(2017), no. 3, 1201–1231.
[LS1] T. Leblé, S. Serfaty, Large Deviation Principle for Empirical Fields of Log and Riesz gases, Invent. Math.
210, No. 3, 645–757.
[LS2] T. Leblé, S. Serfaty, Fluctuations of Two-Dimensional Coulomb Gases, GAFA 28, No. 2, (2018), 443–508.
[LSZ] T. Leblé, S. Serfaty, O. Zeitouni, Large deviations for the two-dimensional two-component plasma.
Comm. Math. Phys. 350 (2017), no. 1, 301–360.
[LLY] L. Lo, J-G. Liu, P. Yu, On mean field limit for Brownian particles with Coulomb interaction in 3D,
arXiv:1811.07797.
[LWJB] X. Li, J. G. Wöhlbier, S. Jin, J. H. Booske, Eulerian method for computing multivalued solutions
of the Euler-Poisson equations and applications to wave breaking in klystrons, Phys. Rev. E 70 (2004),
016502.
[LZ] F. H. Lin, P. Zhang, On the hydrodynamic limit of Ginzburg-Landau vortices, Disc. Cont. Dyn. Systems
6 (2000), 121–142.
[LT] H. Liu, E. Tadmor, Critical thresholds in 2D restricted Euler-Poisson equations, SIAM J. Appl. Math.
63, No. 6, (2003), 1889–1910.
[Ma] T. Makino, On a local existence theorem for the evolution equation of gaseous stars, Patterns and waves,
Stud. Math. Appl. 18 (1986), 459–479.
[MU] T. Makino, S. Ukai, Sur l’existence des solutions locales de l’équation d’Euler-Poisson pour l’évolution
d’étoiles gazeuses, J. Math. Kyoto Univ. 27 (1987), No. 3, 387–399.
[MP] C. Marchioro, M. Pulvirenti, Mathematical theory of incompressible nonviscous fluids. Applied Mathe-
matical Sciences, 96. Springer-Verlag, New York, 1994.
[NW] H. Neunzert, J. Wick, Die Approximation der Lösung von Integro-Differentialgleichungen durch
endliche Punktmengen, Numerische Behandlung nichtlinearer Integrodifferential- und Differentialgleichun-
gen, Springer, 1974, 275–290.
[Pe] B. Perthame, Nonexistence of global solutions to Euler-Poisson equations for repulsive forces, Japan J.
Appl. Math. 7, No. 2 (1990), 363–367.
[PS] M. Petrache, S. Serfaty, Next Order Asymptotics and Renormalized Energy for Riesz Interactions, J.
Inst. Math. Jussieu 16 (2017) No. 3, 501–569.
[RS] N. Rougerie, S. Serfaty, Higher Dimensional Coulomb Gases and Renormalized Energy Functionals,
Comm. Pure Appl. Math 69 (2016), 519–605.
[SR] L. Saint Raymond, Hydrodynamic limits of the Boltzmann equation. Lecture Notes in Mathematics 1971,
Springer, (2009).
[Scho1] S. Schochet, The weak vorticity formulation of the 2-D Euler equations and concentration-cancellation,
Comm. PDE 20 (1995), No. 5-6, 1077–1104.
[Scho2] S. Schochet, The point-vortex method for periodic weak solutions of the 2-D Euler equations. Comm.
Pure Appl. Math. 49 (9) (1996), 911–965.
[Si] P. Serfati, Une preuve directe d’existence globale des vortex patches 2D, C. R. Acad. Sci. Paris Ser. I
Math. 318, (1994), no. 6, 515–518.
[Sy] S. Serfaty, Mean Field Limits for the Gross-Pitaevskii and Parabolic Ginzburg-Landau Equations, J.
Amer. Math. Soc. 30 (2017), no. 3, 713–768.
[SV] S. Serfaty, J. L. Vázquez, A Mean Field Equation as Limit of Nonlinear Diffusions with Fractional
Laplacian Operators, Calc Var. PDE 49 (2014), no. 3-4, 1091–1120.
[Wo] W. Wolibner, Un théorème sur l’existence du mouvement plan d’un fluide parfait, homogène, incom-
pressible, pendant un temps infiniment long. Math. Z. 37 (1933), no. 1, 698–726.
[XZ] W. Xiao, X. Zhou, Well-posedness of a porous medium flow with fractional pressure in Sobolev spaces,
Electron. J. Differential Equations 2017, Paper No. 238, 7 pp.
[Yu] V.I Yudovich, Non-stationary flow of an ideal incompressible liquid. Zh. Vych. Mat. 3, (1963) 1032–1066.