Professional Documents
Culture Documents
MITIA DUERINCKX
arXiv:2302.01466v1 [math.AP] 2 Feb 2023
Contents
1. Introduction and main results 1
2. Preliminary on Stokes analysis 11
3. Lipschitz estimate on the fluid velocity 14
4. Dilute expansion of particle velocities 18
5. Mean-field approximation 32
6. Conclusion: proof of main results 40
Acknowledgements 44
References 44
Previous work on particle suspensions have mainly been devoted to the following two
preliminary questions:
— Homogenization for a ‘given’ microstructure:
Given instantaneous particle positions, the Stokes problem defining the fluid velocity
can be approximated on large scales by an effective Stokes equation with some effective
viscosity, say B̄. This is now well-understood in the framework of homogenization
theory [14, 11] and we refer to [15] for a review.
— Semi-dilute expansion of the effective viscosity:
In the dilute regime, the effective viscosity B̄ can be expanded with respect to the par-
ticle volume fraction λ ≪ 1. To first order, this expansion takes form of the celebrated
Einstein formula
B̄ = Id +λB̄ 1 + O(λ2 ), (1.1)
where B̄ 1 only depends on the single-particle distribution of shapes and orientations.
The next-order correction further involves the statistical distribution of pairs of particles
on the microscale. The expansion can be pursued to higher orders in form of a cluster
expansion and is now well understood; see [12, 15] and references therein.
With these results at hand, it remains to couple homogenization with microstructure dy-
namics in the semi-dilute expansion. As Einstein’s approximation (1.1) only involves the
single-particle distribution, we can expect a mean-field description of the dynamics with
accuracy O(λ2 ): it would take form of an effective Stokes equation with viscosity given
by Einstein’s approximation (1.1), coupled to a Vlasov equation for the single-particle dis-
tribution of positions and orientations; see indeed (1.15)–(1.16) below. This accounts for
non-Newtonian effects due to the collective orientation of the particles and corresponds to
the so-called Doi model first derived formally in [31, 24, 3].
The first rigorous results on the dilute dynamics [30, 25, 32] focused on the leading-order
description, thus neglecting Einstein’s O(λ) correction to the effective viscosity (1.1): this
leads to a simpler transport-Stokes system devoid of any non-Newtonian effect. In [29],
Höfer and Schubert went one step further and managed to capture Einstein’s correction
in the effective Stokes equation, but their analysis was restricted to the case of spherical
particles: orientations then play no role and no non-Newtonian effect is obtained. In the
present work, we consider non-spherical particles, we manage to describe the mean-field
distribution of their orientations, and we derive a Doi-type model, rigorously describing
non-Newtonian effects for the first time in a micro-macro limit. We further extend our
analysis to the case of active suspensions, including the effects of particle self-propulsion:
this leads to an additional elastic stress in the effective Stokes equation as was indeed
predicted in [38, 20, 36, 35, 6] and first derived in [18, 2] in the equilibrium setting.
Note that no mean-field description can hold beyond accuracy O(λ2 ) since higher-order
corrections to the effective viscosity (1.1) would involve statistical information on the ar-
rangement of the particles on the microscale, such as the microscopic two-particle distribu-
tion. Such information is beyond the scope of propagation of chaos and mean-field theory,
and its dynamical description is left as an open problem; see Remark 4.2 for more detail.
1.2. Hydrodynamic model for particle suspension. We consider a system of N non-
n
Brownian inertialess rigid particles, denoted by Iε,N for 1 ≤ n ≤ N , of typical size O(ε),
possibly self-propelling, suspended in a Stokes flow. We start by introducing the precise
model that we are going to study: we describe the set of rigid particles, then turn to their
possible self-propulsion, before describing the underlying viscous solvent and the particle
SEMI-DILUTE RHEOLOGY OF PARTICLE SUSPENSIONS 3
dynamics. We assume that the space dimension is d > 2 (the case d = 2 can be treated
similarly, up to obvious modifications due to the logarithmic growth of the Stokeslet in the
whole plane).
• Elongated rigid particles. ◦
´ Let I ⊂ B be an axisymmetric connected closed set, which
we take to be centered at I ◦ x dx = 0 and to be of class C 2 . We then consider N particles
n }
that are disjoint rigid copies {Iε,N ◦
1≤n≤N of the rescaled set εI . More precisely, each
n is characterized by its center X n ∈ Rd and by the direction Rn
particle Iε,N d−1 of
ε,N ε,N ∈ S
its axis, in the sense of
n n n n
Iε,N := Iε (Xε,N , Rε,N ) := Xε,N + εI ◦ (Rε,N
n
),
where we have set I ◦ (r) := Θ(r)I ◦ , where for a direction r ∈ Sd−1 we denote by Θ(r) :
Rd → Rd the rotation that maps the axis of I ◦ to r (the sign is fixed by choosing r 7→ Θ(r)
to be continuous). The set of all rigid particles is denoted by
Iε,N := N n
S
n=1 Iε,N .
n;+ n + εB and we assume that they are disjoint,
We also consider ε-neighborhoods Iε,N := Iε,N
n;+ m;+
Iε,N ∩ Iε,N = ∅, for all 1 ≤ n 6= m ≤ N , (1.2)
which will be shown to be preserved along the dynamics in our regime of interest, cf. Propo-
sition 1.1. The particle volume fraction is then
λ := |Iε,N | = N |εI ◦ | = N εd |I ◦ |,
and we shall consider the macroscopic limit N ↑ ∞, ε ↓ 0, in the dilute regime λ ≪ 1.
• Particle activity. We consider particles that may be active and propel themselves in
the fluid (e.g. by consuming some underlying chemical energy, which is not included in
the model for simplicity). By a balance of forces, self-propulsion must be described by a
couple of forces of same intensity and opposite direction on each rigid particle and on the
surrounding fluid:
— Each particle Iε,Nn is assumed to propel itself in the direction Rn
ε,N of its own axis.
— The force of each particle on the surrounding fluid is typically exerted via a flagellar
bundle, but the detail of the propulsion mechanism is not included in the model for
simplicity: as e.g. in [5, Section 2.1] (see also [2]), we assume that the force exerted by
n
particle Iε,N on the surrounding fluid can be effectively described by a force field
n n n
fε,N := fε (· − Xε,N , Rε,N ) : Rd \ Iε,N
n
→ Rd .
where the function fε takes the form fε (x, r) := ε−d f ( xε , r) for some bounded func-
tion f : Rd ×Sd−1 → Rd . We also assume that f (·, r) is axisymmetric around direction r,
just like I ◦ (r), for all r ∈ Sd−1 .
The balance of propulsion forces then takes form of the following assumption,
ˆ
n n
Rε,N + fε,N = 0, for all 1 ≤ n ≤ N .
n
Rd \Iε,N
or equivalently f (·, r)|I ◦ (r) = r|I ◦ |−1 , so the balance of forces reads
ˆ
f (·, r) = 0, for all r. (1.4)
Rd
We further assume that f (·, r) is compactly supported, say in I ◦ (r) + B, for all r, meaning
that each particle propels itself only by acting on the surrounding fluid at bounded distance.
By the separation assumption (1.2), we then note that the force fields {fε,N n }
1≤n≤N have
pairwise disjoint supports.
In most previous work, e.g. [19, 20, 18], the action of a particle on the surrounding fluid
was represented for simplicity by a point force, typically setting f (x, r) := −δ(x − θr)r for
x∈/ I ◦ (r), for some parameter θ ∈ R with θr ∈ / I ◦ (r). This point-force model is a special
case of ours (regularity issues for f play no important role), and the cases θ > 0 and θ < 0
then correspond to so-called puller and pusher particles, respectively. We refer e.g. to [39,
Sections 2.1–2.2] for a review of other models for self-propulsion, such as squirmer models,
which are different but which we believe could be treated analogously.
• Inertialess particle dynamics in viscous solvent. Particles are suspended in a homoge-
neous viscous fluid, which we assume to be described by the steady Stokes equation with
unit viscosity. More precisely, given the set Iε,N of particles at a given time, the fluid
velocity uε,N and pressure pε,N satisfy the following Stokes equation in the fluid domain,
N
X
n
− △uε,N + ∇pε,N = h + κ fε,N , div(uε,N ) = 0, in Rd \ Iε,N , (1.5)
n=1
where h stands for some internal force in the fluid domain and where κ ≥ 0 is the self-
propulsion intensity. We assume for convenience h ∈ W 1,1 ∩ W 1,∞ (Rd )d . Next, we assume
that the fluid flow satisfies no-slip conditions at particle boundaries, so we may implicitly
extend the fluid velocity uε,N inside the particles to coincide with the particle velocities.
The rigidity of the particles then translates into a boundary condition,
D(uε,N ) = 0, in Iε,N , (1.6)
where D(u) = 12 (∇u + (∇u)′ ) stands for the symmetric gradient. Equivalently, this means
for all 1 ≤ n ≤ N ,
n
uε,N = Vε,N + Ωnε,N (x − xnε,N ), n
in Iε,N , (1.7)
n ∈ Rd and angular velocity tensor Ωn d×d
for some translational velocity Vε,N ε,N ∈ Rskew . As we
neglect the inertia of the particles, Newton’s equations of motion reduce to the balance of
forces and torques, which take form of additional boundary conditions,
ˆ
n
me + κRε,N + σ(uε,N , pε,N )ν = 0,
n
∂Iε,N
ˆ
n
(x − Xε,N ) × σ(uε,N , pε,N )ν = 0, for all 1 ≤ n ≤ N , (1.8)
n
∂Iε,N
where σ(u, p) := 2 D(u) − p Id is the Cauchy stress tensor and where me ∈ Rd stands for
the buoyancy of the particles. (Henceforth, for a, b ∈ Rd , we use the vectorial notation
a × b ∈ Rd×d
skew with (a × b)αβ := aα bβ − aβ bα .) We choose the buoyancy me and the
self-propulsion intensity κ as
n λ κ0 n λ
me := |Iε,N |e = N e, κ := ε |Iε,N | = κ0 εN ,
SEMI-DILUTE RHEOLOGY OF PARTICLE SUSPENSIONS 5
in Rd \ Iε,N
n ,
div(uε,N ) = 0,
D(uε,N ) = 0, n
in Iε,N , (1.9)
λ λ n
´
e + κ 0 R
εN ε,N + n σ(u ε,N , p ε,N )ν = 0, for all 1 ≤ n ≤ N ,
´N ∂Iε,N
n ) × σ(u
∂I n (x − Xε,N ε,N , pε,N )ν = 0, for all 1 ≤ n ≤ N .
ε,N
We recall that the weak formulation of this system takes on the following simple guise:
for any test function v ∈ Ḣ 1 (Rd )d that is incompressible, i.e. div(v) = 0, and that is rigid
inside particles, i.e. D(v) = 0 in Iε,N , we have
ˆ ˆ X
λ n
2 D(v) : D(uε,N ) = v · h1Rd \Iε,N + e1Iε,N + κ0 εN fε,N . (1.10)
Rd \Iε,N Rd n
Once the Stokes system (1.9) is solved for the instantaneous fluid velocity uε,N , particle
positions and orientations can be updated according to
n n n
∂t Xε,N = Vε,N , ∂t Rε,N = Ωnε,N Rε,N
n
, for all 1 ≤ n ≤ N , (1.11)
n , Ωn
where Vε,N ε,N are given by (1.7), or alternatively
n
Vε,N := uε,N ∈ Rd , Ωnε,N := ∇uε,N ∈ Rd×d
skew . (1.12)
n
Iε,N n
Iε,N
In this way, the particles follow the fluid flow and interact with one another via the flow
disturbance that they generate. The resulting dynamics is reputedly complex in view of
the multibody, long-range, and singular nature of hydrodynamic interactions.
1.3. Semi-dilute mean-field description. We aim to investigate the collective macro-
scopic behavior of the fluid velocity uε,N and of the particle empirical measures
N
X
d d−1
νε,N := n ,Rn ) ∈ P(R × S
δ(Xε,N ε,N
), (1.13)
n=1
XN
µε,N := δXε,N
n ∈ P(Rd ).
n=1
1In particular, while a drag force of order O( 1 ε2 ) appeared in the Vlasov equation in [25, 32, 29] due
λ
to the buoyancy, this term would become O(ε2 ) in our scaling and is neglected in this work.
6 M. DUERINCKX
where (uλ,ε , νλ,ε , µλ,ε ) solves the following coupled system on Rd : the macroscopic fluid
velocity uλ,ε satisfies an effective Stokes equation,
◦ ν i D(u ) + ∇p
−div 2 1 + λhΣ λ,ε λ,ε λ,ε
= (1 − λµλ,ε )h + λµλ,ε e + κ0 λβf div h(r ⊗ r − 1d Id) νλ,ε i ,
(1.15)
div(uλ,ε ) = 0,
µλ,ε = hνλ,ε i,
which is coupled to a Vlasov equation for the mean-field distribution νλ,ε of particle posi-
tions and orientations,
∂t νλ,ε + divx νλ,ε (uλ,ε + κ0 εαf r) + divr νλ,ε (Ω◦ ∇uλ,ε )r = 0,
(1.16)
νλ,ε |t=0 = ν ◦ .
´
Here, we use the short-hand notation hgi(x) := Sd−1 g(x, r) dσ(r) for angular averaging,
and the coefficient fields r 7→ Σ◦ (r), Ω◦ (r) and constants αf , βf are defined as follows:
— Passive effective viscosity: The coefficient 1 + λhΣ◦ νλ,ε i in the effective Stokes equa-
tion (1.15) corresponds to Einstein’s formula for the dilute correction of the plain fluid
viscosity due to the presence of rigid particles, see e.g. [12, Section 2] or [23]: for
all r ∈ Sd−1 , the tensor Σ◦ (r) is the symmetric linear map on the set Rd×d sym,0 of trace-
free symmetric matrices, given by
ˆ
E : Σ◦ (r)E := |I ◦ |−1 |D(u◦r,E )|2 , for all E ∈ Rd×d
sym,0 , (1.17)
Rd
which describes the flow disturbance generated by a strain rate E at a single rigid
particle I ◦ (r) oriented in direction r. In other words, Σ◦ (r)E measures the reaction of
a particle oriented in direction r to a strain rate E, and it can equivalently be written
as (half) the associated stresslet,
ˆ
Σ◦ (r)E = 21 |I ◦ |−1 σ(u◦r,E + Ex, p◦r,E )ν ⊗◦s x, (1.19)
∂I ◦ (r)
where u◦r,H sym is the solution of (1.18) with E = H sym the symmetric part of H. Note
that the map r 7→ Ω◦ (r) is easily checked to be smooth.
— Active elastic stress: The particle self-propulsion generates an elastic stress in the Stokes
equation (cf. last right-hand side term in (1.15)): in the dilute regime, it is naturally
given in terms of the stresslet
ˆ ˆ
Σ◦f (r) := σ(u◦r,f , p◦r,f )ν ⊗◦s x − f (·, r) ⊗◦ x, (1.21)
∂I ◦ (r) Rd
where a ⊗◦ b := a ⊗ b − d1 Id(a · b) is the trace-free tensor product and where u◦r,f is the
unique decaying solution of the single-particle problem
◦ ◦ in Rd \ I ◦ (r),
−△ur,f + ∇pr,f = f (·, r),
◦ in Rd \ I ◦ (r),
div(ur,f ) = 0,
D(u´◦r,f ) = 0, in I ◦ (r), (1.22)
r + σ(u ◦ , p ◦ )ν = 0,
∂I ◦ (r) r,f r,f
´
x × σ(u ◦ , p◦ )ν = 0,
∂I (r)
◦ r,f r,f
which describes the flow disturbance generated by the self-propulsion of a single particle
oriented in direction r. Since the inclusion I ◦ (r) and the propulsion force f (·, r) are
both axisymmetric around direction r, this single-particle stresslet (1.21) can be written
by symmetry as
Σ◦f (r) = βf r ⊗◦ r, for some βf ∈ R. (1.23)
The cases βf > 0 and βf < 0 correspond to so-called puller and pusher particles,
respectively. In the macroscopic limit, the active elastic stress is then given by the
angular average hΣ◦f νλ,ε i, which appears as the last right-hand side term in the effective
Stokes equation (1.15). It coincides with the expression predicted e.g. in [38, 20, 36,
35, 6] and first derived in [18, 2] in the equilibrium setting.
— Swimming velocities: The particle self-propulsion also generates a drag force on the
particles, leading to effective swimming velocities: in the dilute regime, we naturally
define Vf◦ (r) as the drag velocity of a single particle oriented in direction r due to its
self-propulsion, ˆ
Vf◦ (r) := u◦r,f ∈ Rd . (1.24)
I ◦ (r)
Since the propulsion force f (·, r) is axisymmetric around r, this swimming velocity can
be written by symmetry as
Vf◦ (r) = αf r, for some αf > 0. (1.25)
The above kinetic model (1.15)–(1.16) is a variant of the so-called Doi model, which
was first introduced in [31, 24, 3] for passive suspensions κ0 = 0 (see also [8, 7, 9]), and
which was adapted by Saintillan and Shelley [38, 36, 37] to active suspensions κ0 6= 0. The
difference of the above with the standard form of the Doi model is twofold:
8 M. DUERINCKX
— Brownian effects are not considered in the present work. The inclusion of Brownian
rotary effects on particle orientations was recently discussed in a simplified setting
in [27], but the general mean-field description of Brownian suspensions with spatial
diffusion remains a delicate open problem and is postponed to future work.
— While we consider particles with a given axisymmetric shape, the Doi model rather
corresponds to the limit of very elongated particles as computed by slender-body theory,
e.g. [3]. This amounts to replacing the effective coefficients Σ◦ , Ω◦ in (1.15)–(1.16) by
Σ◦ (r) ❀ α1 (r ⊗◦ r) ⊗ (r ⊗◦ r),
Ω◦ (r)(∇u)r ❀ Id −r ⊗ r α2 D(u) + 21 ∇ × u r,
1.4. Main results. We start by stating the asymptotically global well-posedness of the
particle dynamics (1.9)–(1.12). It requires to control the evolution of the minimal inter-
particle distance and of the largest distance to the origin,
dmin
ε,N := min n
|Xε,N m
− Xε,N |, ρmax n
ε,N := max |Xε,N |.
1≤n6=m≤N 1≤n≤N
The result is similar to corresponding statements in [25, 32], and a short proof is included
in Section 3 for completeness, cf. Corollary 3.2.
Proposition 1.1 (Well-posedness of particle system). Assume that initial particle positions
satisfy, for some C0 ≥ 1,
− d1
dmin
ε,N (0) ≥
1
C0 N , ρmax
ε,N (0) ≤ C0 .
Given T < ∞, provided that λ log N ≪ 1 is small enough (depending on C0 , T, h), the
particle dynamics (1.9)–(1.12) is well-posed up to time T and satisfies for all 0 ≤ t ≤ T ,
− d1
dmin
ε,N (t) ≥
1
CN , ρmax
ε,N (t) ≤ C,
1 1 1
for some C ≥ 1 (depending on C0 , T, h). In particular, as ε ≃ λ d N − d ≪ N − d , this
ensures that the disjointness condition (1.2) remains satisfied up to time T . ♦
We turn to the justification of the kinetic model (1.15)–(1.16) in the macroscopic limit.
In fact, we only manage to derive it in general with accuracy O((λ + ε)2 + κ0 ε), which
amounts to neglecting O(κ0 ε) swimming velocities: it leads us to the simplified model
−div 2 1 + λhΣ◦ νλ i D(uλ ) + ∇pλ
= (1 − λµλ )h + λµλ e + κ0 λβf div h(r ⊗ r − d1 Id)νλ i ,
div(uλ ) = 0,
(1.26)
µ λ = hνλ i,
◦
∂t νλ + divx (νλ uλ ) + divr (νλ (Ω ∇uλ )r) = 0,
νλ |t=0 = ν ◦ .
SEMI-DILUTE RHEOLOGY OF PARTICLE SUSPENSIONS 9
We start with the well-posedness of these macroscopic models in the smooth class. The
proof is standard and is included in Section 6.1 for completeness.
Proposition 1.2 (Well-posedness of macroscopic models).
(i) Given T < ∞, given γ > 1 non-integer, and given an initial condition ν ◦ ∈ P ∩
W 1,1 ∩ W γ,∞ (Rd × Sd−1 ), there is λ0 > 0 (depending on T, h, γ, ν ◦ ) such that for
all 0 ≤ λ ≤ λ0 there is a unique solution (νλ , uλ ) of (1.26) up to time T with νλ ∈
L∞ ([0, T ]; P ∩ W 1,1 ∩ W γ,∞ (Rd × Sd−1 )) and uλ ∈ L∞ ([0, T ]; W γ+1,∞ (Rd )d ).
(ii) Given T < ∞, given γ > 1 non-integer, and given initial conditions µ◦ ∈ P ∩
W 1,1 ∩ W γ,∞ (Rd ) and r ◦ ∈ W γ,∞ (Rd ; Sd−1 ), there are λ0 , ε0 > 0 (depending on
T, h, γ, µ◦ , r ◦ ) such that for all 0 ≤ λ ≤ λ0 and 0 ≤ ε ≤ ε0 there is a unique solution
(µλ,ε , rλ,ε , uλ,ε ) of (1.27) up to time T with µλ,ε ∈ L∞ ([0, T ]; P ∩ W 1,1 ∩ W γ,∞ (Rd )),
rλ,ε ∈ L∞ ([0, T ]; W γ,∞ (Rd ; Sd−1 )), and uλ,ε ∈ L∞ ([0, T ]; W γ+1,∞ (Rd )d ). ♦
We can now state our main results, that is, the derivation of the Doi-type macroscopic
models (1.26) or (1.27) from the particle dynamics (1.9)–(1.12), thus providing a rigorous
version of (1.14). To our knowledge, this is the first time that non-Newtonian macroscopic
models are rigorously derived from a hydrodynamic description of particle suspensions.
It constitutes both a generalization of [29] to non-spherical and possibly active particles,
and a generalization of [16, 18] to the particle dynamics. We start with the simplified
model (1.26) without swimming velocities, which we derive with accuracy O((λ+ε)2 +κ0 ε)
(up to logarithmic corrections and initial well-preparedness).
Theorem 1.3 (Semi-dilute mean-field approximation). Assume that initial particle posi-
tions satisfy, for some C0 ≥ 1,
− d1
dmin
ε,N (0) ≥
1
C0 N , ρmax
ε,N (0) ≤ C0 ,
and assume that the empirical measure (1.13) is initially close in the ∞-Wasserstein metric
to some density ν ◦ ∈ P ∩ W 1,1 ∩ W γ,∞ (Rd × Sd−1 ) for some γ > 1,
◦
W∞ (νε,N , ν ◦ ), W∞ (µ◦ε,N , µ◦ ) ≤ C0 (λ log N + ε), µ◦ := hν ◦ i.
Given T < ∞, assume that λ log N ≪ 1 is small enough to ensure the well-posedness
of the particle system up to time T as in Proposition 1.1, and let 0 ≤ λ ≤ λ0 be small
enough to ensure that (1.26) admits a unique solution (νλ , µλ , uλ ) up to time T as in
Proposition 1.2(i). Then we have for all t ∈ [0, T ],
W∞ (µtε,N , µtλ ) . (λ log N + ε)2 + κ0 ε + W∞ (µ◦ε,N , µ◦ ),
10 M. DUERINCKX
Notation.
• In order to control the particle dynamics, on top of the minimal interparticle distance
and the largest distance to the origin,
dmin
ε,N := min n
|Xε,N m
− Xε,N |, ρmax n
ε,N := max |Xε,N |. (1.28)
1≤n6=m≤N 1≤n≤N
we also consider the following quantities, as in [25], for any σ ∈ [0, d],
N
X
ασε,N := max 1 n
|Xε,N m σ−d
− Xε,N | . (1.29)
1≤n≤N N
m:m6=n
• We denote by Rd×d
0 , Rd×d
sym,0 ,
and Rd×d
the set of trace-free matrices, of trace-free sym-
skew
metric matrices, and of skew-symmetric matrices, respectively.
SEMI-DILUTE RHEOLOGY OF PARTICLE SUSPENSIONS 11
−△u + ∇p = h, in Rd \ Iε,N ,
then the following relation holds in the weak sense in the whole space Rd ,
N
X
−△u + ∇(p1Rd \Iε,N ) = h1Rd \Iε,N − δ∂I i σ(u, p)ν. ♦
ε,N
n=1
Proof. For a test function v ∈ Ḣ 1 (Rd )d , the incompressibility of u and the rigidity con-
straint in Iε,N yield
ˆ ˆ ˆ
∇v : ∇u = 2 ∇v : D(u) = 2 ∇v : D(u),
Rd Rd Rd \Iε,N
or equivalently, inserting the definition of the Cauchy stress tensor σ(u, p) = 2 D(u) − p Id,
ˆ ˆ ˆ
∇v : ∇u = ∇v : p Id + ∇v : σ(u, p).
Rd Rd \Iε,N Rd \Iε,N
12 M. DUERINCKX
Integrating by parts in the last right-hand side term and noting that the Stokes equation
in (2.1) yields −div(σ(u, p)) = h in Rd \ Iε,N , we deduce
ˆ ˆ ˆ N ˆ
X
∇v : ∇u = ∇v : p Id + v·h− v · σ(u, p)ν,
Rd Rd \Iε,N Rd \Iε,N n
n=1 ∂Iε,N
We state and prove the following basic trace estimate, which is used repeatedly to control
force terms concentrated at particle boundaries.
1 (Rd )d × L2 (Rd ) satisfies
Lemma 2.2 (Trace estimate). Given 1 ≤ n ≤ N , if (u, p) ∈ Hloc loc
n;+
−△u + ∇p = 0, in Iε,N n ,
\ Iε,N
n;+ n ,
div(u) = 0, in Iε,N \ Iε,N
D(u) = 0, n
in Iε,N ,
´ (2.2)
n σ(u, p)ν = 0,
´∂Iε,N
n ) × σ(u, p)ν = 0,
∂I n (x − Xε,N
ε,N
n;+ n + εB.
where we recall Iε,N = Iε,N ♦
´
Proof. The condition ∂I n σ(u, p)ν = 0 allows to rewrite
ε,N
ˆ ˆ
F · σ(u, p)ν = F− F · σ(u, p)ν.
n n n;+
∂Iε,N ∂Iε,N Iε,N
Hence, taking the infimum over Ω ∈ Rd×d skew and appealing to Korn’s inequality,
ˆ
F · σ(u, p)ν . kD(F )k 2 n;+ kσ(u, p)k 2 n;+ n .
L (Iε,N ) L (Iε,N \Iε,N )
n
∂Iε,N
It remains to´estimate the pressure field p in the right-hand side. As the incompressibility
of F implies ∂I n F · ν = 0, any constant can be subtracted to the pressure p in the above,
ε,N
hence in particular
ˆ
F · σ(u, p)ν . kD(F )kL2 (I n;+ ) kD(u)kL2 (I n;+ ) +
p − p
2 n;+ n .
n
∂Iε,N ε,N ε,N n;+
Iε,N n
\Iε,N L (Iε,N \Iε,N )
Now appealing to a local pressure estimate for the steady Stokes equation, which follows
from a standard argument based on the Bogovskii operator, e.g. [16, Lemma 3.3], the
conclusion follows.
Next, we recall the usual definition of the Stokeslet G, that is the Green’s function for
the steady Stokes equation, and we recall its pointwise decay.
Lemma 2.3 (Pointwise Stokeslet estimates). For all 1 ≤ i ≤ d, we can define Gi ∈
1,1
Wloc (Rd )d as the unique decaying distributional solution of
−△Gi + ∇Pi = ei δ0 , div(Gi ) = 0, in Rd ,
and we then set Gi = (Gij )1≤j≤d and G = (Gij )1≤i,j≤d . This Green’s function is explicitly
given by
1
G(x) = 2(d−2)|∂B| |x|2−d Id +(d − 2) |x|x x
⊗ |x| ,
hence it satisfies the following pointwise estimates,
|G(x)| . |x|2−d , |∇G(x)| . |x|1−d , |∇2 G(x)| . |x|−d . ♦
We also define a corresponding notion of Stokeslet for the steady Stokes problem with
a single rigid inclusion and we state that it satisfies a similar pointwise decay. This is a
particular case of the analysis in [12, Appendix A], where we further get a corresponding
result for any finite family of well-separated rigid inclusions.
Lemma 2.4 (Pointwise Stokeslet estimates with rigid inclusions; [12]). For all 1 ≤ n ≤ N ,
1,1
y ∈ Rd \ Iε,N
n , and 1 ≤ i ≤ d, we can define G n d d
ε,N ;i (·, y) ∈ Wloc (R ) as the unique decaying
distributional solution of
−△Gε,N n n in Rd \ Iε,N
n ,
;i (·, y) + ∇Pε,N ;i (·, y) = ei δy ,
n in Rd \ Iε,N
n ,
div(G ε,N ;i (·, y)) = 0,
n
D(Gε,N ;i (·, y)) = 0, n
in Iε,N ,
´ n σ(G n (·, y), P n (·, y))ν = 0,
∂Iε,N ε,N ;i ε,N ;i
´ n (· − X i ) × σ(G n (·, y), P n (·, y))ν = 0,
∂I ε,N
ε,N ε,N ;i ε,N ;i
14 M. DUERINCKX
From the Stokes problem (1.9), using Lemma 2.1, we find that the fluid velocity uε,N
satisfies the following equation in Rd ,
In terms of the Stokeslet G, cf. Lemma 2.3, noting that the assumption dmin
ε,N ≥ 4ε ensures
n } have pairwise disjoint supports, we deduce
that propulsion forces {fε,N n
ˆ N ˆ
X
λ n
uε,N (x) = G(x − ·)h + κ0 εN G(x − ·)fε,N
n;+ n
Rd \I ε,N n=1 Iε,N \Iε,N
N ˆ
X
− G(x − ·) σ(uε,N , pε,N )ν,
n
n=1 ∂Iε,N
hence, using boundary conditions for uε,N , cf. (1.9), and recalling (1.3),
ˆ N ˆ
X N ˆ
X
λ n
uε,N (x) = G(x − ·)h + G(x − ·)e + κ0 εN G(x − ·)fε,N
n n;+
Rd \I ε,N n=1 Iε,N n=1 Iε,N
N ˆ
X
− G(x − ·) − G(x − ·) σ(uε,N , pε,N )ν. (3.3)
n n
n=1 ∂Iε,N Iε,N
Appealing to a trace estimate and to pointwise bounds on G, cf. Lemmas 2.2 and 2.3,
recalling the local balance condition (1.4) for propulsion forces, performing local integrals,
and recalling εd |I ◦ | = Nλ , we find for all x ∈ Rd with dist(x, Iε,N ) ≥ 2ε,
ˆ N
X N
X
|uε,N (x)| . |x − ·|2−d |h| + |e| Nλ n 2−d
|x − Xε,N | + κ0 Nλ n 1−d
|x − Xε,N |
Rd n=1 n=1
N
X 1
n 1−d
|D(uε,N )|2
λ 2
+ N |x − Xε,N | .
n;+
n=1 Iε,N
where we recall that ασε,N is defined in (1.29). Using this bound in (3.4) and recalling the
assumption dmin
ε,N ≥ 4ε, the claim (3.1) follows.
Step 2. Preliminary estimate on the fluid velocity close to rigid particles: for all x ∈
n | ≤ 1 dmin for some 1 ≤ n ≤ N , we have
Rd \ Iε,N with |x − Xε,N 2 ε,N
Let x ∈ Rd \ Iε,N be fixed with |x − Xε,N n | ≤ 1 dmin for some 1 ≤ n ≤ N . Note that
2 ε,N
m } 1 min d n
we then have dist(x, {Xε,N m:m6=n ) ≥ 2 dε,N . Testing equation (3.2) in R \ Iε,N with the
n
Stokeslet Gε,N (·, x) for the Stokes problem with a single rigid inclusion, cf. Lemma 2.4, we
find, instead of (3.3),
ˆ N ˆ
X N ˆ
X
n n λ n m
uε,N (x) = Gε,N (·, x)h + Gε,N (·, x)e + κ0 εN Gε,N (·, x)fε,N
m m;+
Rd \I ε,N m=1 Iε,N m=1 Iε,N
N
X ˆ
n n
− Gε,N (·, x) − Gε,N (·, x) σ(uε,N , pε,N )ν.
m m
m:m6=n ∂Iε,N Iε,N
Now appealing to a trace estimate and to the pointwise bounds on Gε,N n , cf. Lemmas 2.2
and 2.4, and noting that |x − Xε,Nm | & |X n − X m | for all m, the claim (3.6) follows sim-
ε,N ε,N
ilarly as in Step 2. The only difference is that we now need to treat diagonal contributions
separately: for instance,
N ˆ N
X n 2−d
X
|e| Nλ |e| Nλ | · −x|2−d
Gε,N (·, x)e . | · −x| +
m n m
m=1 Iε,N Iε,N m:m6=n Iε,N
N
X
n
. |e| Nλ hx − Xε,N i2−d
ε + |e| Nλ m 2−d
|x − Xε,N |
m:m6=n
2
. |e|(ε + λα2ε,N ),
where we use the short-hand notation hxiε := (ε2 + |x|2 )1/2 , and where the last inequality
m } 1 min
follows from (3.5) with dist(x, {Xε,N m:m6=n ) ≥ 2 dε,N .
This estimate is obtained by post-processing the results of the first two steps in the dilute
regime λ ≪ 1. From (3.6), we deduce in particular, for all 1 ≤ n ≤ N ,
1
|D(uε,N )|2 . Ch + |e|λα1ε,N + κ0 λα0ε,N
2
n;+
Iε,N
N
X 1
n m −d
λ
|D(uε,N )|2
2
+ N |Xε,N − Xε,N | . (3.8)
m;+
m:m6=n Iε,N
XN 1
0 1 n m −d 2 2
+ λαε,N sup N |Xε,N − Xε,N | |D(uε,N )| .
m;+
1≤n≤N m:m6=n Iε,N
Provided that λα0ε,N ≪ 1 is small enough, the last right-hand side term can be absorbed,
to the effect of
XN 1
1 n m −d 2 2
sup N |Xε,N − Xε,N | |D(uε,N )|
m;+
1≤n≤N m:m6=n Iε,N
Inserting this back into (3.8) and using κ0 λα0ε,N ≤ 1, the claim (3.7) follows.
Step 4. Conclusion.
Inserting (3.7) into (3.1) and (3.6), using again (3.5), using Hölder’s inequality in form of
d−1 d−2 d−2
(α1ε,N )2 ≤ α0ε,N α2ε,N , α1ε,N ≤ (α0ε,N ) d , α2ε,N ≤ (α1ε,N ) d−1 ≤ (α0ε,N ) d , (3.9)
and using λα0ε,N ≤ 1, we deduce for all x ∈ Rd \ Iε,N ,
|uε,N (x)| + |∇uε,N (x)| .h 1,
which is the conclusion.
3.2. Proof of Corollary 3.2. As long as particles are separated by a positive distance,
the particle dynamics (1.9)–(1.12) is well-posed, see e.g. [25, Appendix A]. It remains to
control interparticle distances. Given θ > θ0 , consider the maximal time
n o
θ − d1
Tε,N := sup t ≥ 0 : dminε,N (s) ≥ (θN ) , ρmax
ε,N (s) ≤ θ for all 0 ≤ s ≤ t . (3.10)
We split the proof into two steps.
Step 1. Preliminary estimate: for all σ ∈ (0, d],
1
ασε,N . σ −1 (N d dmin −d max σ
ε,N ) (ρε,N ) ,
1
α0ε,N . (N d dmin −d
log 2 + ρmax min −1
ε,N ) ε,N + (dε,N ) . (3.11)
18 M. DUERINCKX
We start from
N
X N
X
1 n m σ−d 1 n σ−d
N |Xε,N − Xε,N | . N | · −Xε,N | .
m ,1
B(Xε,N dmin
m:m6=n m:m6=n 2 ε,N )
By definition, we have {Xε,Nm } ⊂ B(0, ρmax ) and the balls {B(X m , 1 dmin )} are pair-
m ε,N ε,N 2 ε,N m
wise disjoint. For σ ∈ (0, d], we then get
XN ˆ
n m σ−d min −d
1
N |Xε,N − Xε,N | 1
. N (dε,N ) | · |σ−d
m:m6=n B(0,2ρmax
ε,N )
1
. σ −1 (N d dmin −d max σ
ε,N ) (ρε,N ) ,
.h (λα1ε,N + ε) λ(α0ε,N + 1) + ε ;
.h λαε,N + ε λ(α0ε,N + 1) + ε + ( δε )d .
1
♦
Remark 4.2 (Higher orders and failure of mean-field theory). These dilute expansions
can be easily pursued to higher orders. For instance, the next-order expansion of the
n involves in particular the three-body contribution
translational velocity Vε,N
h i n
2 ◦ ◦
ˆ ˆ
λ ∇G ∗ 2hΣ νε,N iD ∇G ∗ (2hΣ νε,N iD(G ∗ h)) (Xε,N ), (4.1)
which physically describes the flow disturbance due to a stress difference at a particle
boundary generated by the flow disturbance due to a stress difference at another parti-
cle boundary. Yet, although the expansion can be pursued to higher orders, a mean-field
approximation cannot be justified to higher accuracy. Indeed, the above three-body contri-
bution (4.1) involves particle interactions via the kernel ∇2 G, which is a Calderón–Zygmund
kernel, with critical decay |∇2 G(x)| ≃ |x|−d . The criticality of this kernel is known to im-
ply the failure of mean-field theory, see e.g. [33]: by scaling, any macroscopic limit should
actually depend on the microscopic arrangement of the particles. This importance of the
microscopic geometry was recently illustrated in [28]. More precisely, the next-order semi-
dilute correction to the mean-field approximation would require to capture the statistical
distribution of pairs of particles on the microscale — as was indeed anticipated in the intro-
duction, in link with corrections to Einstein’s formula (1.1), cf. [12, 17]. This leads us far
beyond the scope of propagation of chaos and mean-field theory: even a formal description
remains unclear and is left open for future work. (We emphasize that it is not about the
two-particle macroscopic density, which commonly appears when describing corrections to
mean field, e.g. [10]: instead, it is here about the distribution of pairs of particles on the
microscale.) ♦
20 M. DUERINCKX
K⊂{1,...,N } I⊂K
NX
−♯I
X
I k N − ♯I
= πε,N (−1)
k
I⊂{1,...,N } k=0
{1,...,N }
= πε,N = πε,N ,
PK
where we used the fact that k=0 (−1)k K
k = 1K=0 .
4.2. Cluster expansion errors. We turn to the accuracy of the cluster expansion (4.3)
upon truncation in the dilute regime λ ≪ 1. For the purpose of this work, we restrict
ourselves to the second-order expansion, but higher orders can be dealt with analogously,
cf. Remark 4.5.
Lemma 4.4. Assume that dmin 0
ε,N ≥ 4ε and that λαε,N ≪ 1 is small enough. Then we have
(0)
kuε,N − uε,N kL∞ (Rd ) .h λα1ε,N + ε, (4.5)
(0) (1)
kuε,N − uε,N − uε,N kL∞ (Rd ) .h λα0ε,N (λα1ε,N + ε), (4.6)
and moreover, for velocity gradients, for all 1 ≤ n ≤ N ,
∇uε,N − ∇π {n} u(0)
∞ 0
ε,N ε,N L (B(X n , 1 dmin )) .h λαε,N . (4.7)
ε,N 2 ε,N
♦
Remark 4.5. The proof below can be immediately generalized to higher orders: we can
show for all 1 ≤ K < N ,
K
(k)
X
uε,N − uε,N
∞ d .h,K (λα0ε,N )K (λα1ε,N + ε),
L (R )
k=0
and moreover, for velocity gradients, for all 1 ≤ n ≤ N ,
K
{n} (k)
X
∇uε,N − ∇πε,N uε,N
∞ .h,K (λα0ε,N )K+1 .
n 1 min
L (B(Xε,N , 2 dε,N ))
k=0
As this will not be used in this work, we omit the detail for shortness. ♦
Proof of Lemma 4.4. We split the proof into three steps, separately proving the different
estimates in the statement.
Step 1. First-order expansion of uε,N : proof of (4.5).
(0)
Subtracting (3.2) from the defining equation for uε,N = vε,N , cf. (4.2), we get the following
equation in Rd ,
N
(0)
X
λ n n σ(uε,N , pε,N )ν . (4.8)
− △(uε,N − uε,N ) + ∇p = − e1Iε,N
n + κ0 εN fε,N 1Iε,N
n + δ∂Iε,N
n=1
22 M. DUERINCKX
In terms of the Stokeslet G, using the boundary conditions for uε,N , we deduce
N ˆ
(0)
X
(uε,N − uε,N )(x) = − G(· − x) − G(· − x) σ(uε,N , pε,N )ν.
n n
n=1 ∂Iε,N Iε,N
Appealing to a trace estimate and to pointwise bounds on G, cf. Lemmas 2.2 and 2.3, and
evalutating local integrals, this leads us to
N 1
(0)
X
n
λ
i1−d |D(uε,N )|2
2
|(uε,N − uε,N )(x)| . N hx − Xε,N ε ,
n;+
n=1 Iε,N
where we recall the short-hand notation hxiε = (ε2 + |x|2 )1/2 . Now note that (3.5) yields
for all x ∈ Rd , after separating the diagonal contribution,
N
X
n
λ
N hx − Xε,N i1−d
ε . λα1ε,N + ε. (4.9)
n=1
Inserting this into the above, and combining with the Lipschitz estimate of Proposition 3.1,
we deduce for all x ∈ Rd ,
(0)
|(uε,N − uε,N )(x)| .h λα1ε,N + ε,
N
(1)
X
uε,N = (unε,N − vε,N ),
n=1
In terms of the Stokeslet G, using boundary conditions, and appealing to a trace estimate
and to pointwise bounds, we deduce as in Step 1,
N 1
uε,N − u(0) − u(1) (x) .
X
n
λ
i1−d |D(uε,N − unε,N )|2
2
ε,N ε,N N hx − Xε,N ε . (4.10)
n;+
n=1 Iε,N
It remains to estimate the last factor. For that purpose, given 1 ≤ n ≤ N , we start by
noting that the difference uε,N − unε,N satisfies the following equation in Rd \ Iε,N
n ,
N
X
unε,N λ m
−△ uε,N − + ∇p = − e1Iε,N
m + κ0
εN f ε,N 1 I m + δ∂I m σ(uε,N , pε,N )ν .
ε,N ε,N
m:m6=n
SEMI-DILUTE RHEOLOGY OF PARTICLE SUSPENSIONS 23
Inserting this into (4.10), combining with the Lipschitz estimate of Proposition 3.1, and
using (4.9) again, we get for all x ∈ Rd ,
uε,N − u(0) − u(1) (x) .h λα0ε,N (λα1ε,N + ε),
ε,N ε,N
and moreover,
(0) (1) n
n uε,N + uε,N − G ˆ
∗ (1 − λµε,N )h + λµε,N e (Xε,N )
Iε,N
24 M. DUERINCKX
◦
κ0 hΣ◦f νε,N i n n
κ0 εVf◦ (Rε,N
− λ ∇G ˆ
∗ 2hΣ νε,N iD(G ∗ h) + (Xε,N ) − )
where we recall that Σ◦ , Ω◦ , Σ◦f , Vf◦ are defined in (1.17)–(1.25). Moreover, away from the
particles, the fluid velocity satisfies for any δ ∈ [ε, 1],
(0)
kuε,N − G ∗ hkL∞ (Rd \(Iε,N +δB)) .h λ(α2ε,N + κ0 α1ε,N ) + δ(δ + κ0 )( δε )d . (4.15)
and
(0) (1)
uε,N + uε,N − G ∗ (1 − λµε,N )h + λµε,N e (4.16)
◦ ◦
− λ∇G ∗ 2hΣ νε,N iD(G ∗ h) + κ0 hΣf νε,N i
L∞ (Rd \(Iε,N +δB))
For the remaining terms, separating the diagonal contributions, and recalling the nota-
tion (1.29), the pointwise bounds on G directly yield
XN ˆ
G(x − ·)(e − h) dx
n m
m=1 Iε,N Iε,N
X ˆ
d n m 2−d
.h ε |Xε,N − Xε,N | + |x − ·|2−d dx
n
Iε,N n
Iε,N
m:m6=n
. λα2ε,N + ε2 ,
and similarly, further recalling the definition fε,Nm = ε−d f ( 1 (· − X m ), Rm ) and the
ε ε,N ε,N
balance of forces (1.4),
XN ˆ
λ m
G(x − ·)f ε,N dx
εN
n m;+
m=1 Iε,N Iε,N
SEMI-DILUTE RHEOLOGY OF PARTICLE SUSPENSIONS 25
N
X
ˆ
λ m
= εN
G(x − ·) − G(x − ·) fε,N dx
n m;+
m;+
m=1 Iε,N Iε,N Iε,N
XN ˆ
. εd n
|Xε,N m 1−d
− Xε,N | + |x − ·|1−d dx
n n;+
m:m6=n Iε,N Iε,N
. λα1ε,N + ε.
Inserting these different estimates into (4.17), the conclusion (4.12) follows.
(0)
Step 2. Detailed estimates on the linear proxy uε,N = vε,N : for all 1 ≤ n ≤ N , we have
the following version of (3.7),
1
|∇vε,N |2
2
.h 1, (4.18)
n;+
Iε,N
The first bound can be further refined in the following form, where we additionally capture
the leading contribution: denoting by vε,Nn the unique decaying solution of the following
d
Stokes equation in R , 2
(
n + ∇q n = h1 λ P p d
−△vε,N ε,N Rd \Iε,N + e1Iε,N + κ0 εN p:p6=n fε,N , in R ,
n ) = 0, (4.20)
div(vε,N in Rd ,
we have
n n
k∇vε,N − ∇(G ∗ h)(Xε,N )kL∞ (I n;+ ) .h λ(α1ε,N + κ0 α0ε,N ) + ε. (4.21)
ε,N
The bounds (4.18) and (4.19) are easily obtained by similar estimates as in Step 1, recall-
ing λα0ε,N ≤ 1, and we rather focus on the proof of (4.21). For that purpose, we start with
the following identity for the solution of (4.20),
n
∇α vε,N (x) = ∇α (G ∗ h)(x)
XN ˆ N
X ˆ
λ m
+ ∇α G(x − ·)(e − h) + κ0 εN ∇α G(x − ·)fε,N .
m m;+
m=1 Iε,N m:m6=n Iε,N
Using again pointwise bounds on G, we can estimate the last two right-hand side terms
n;+
in Iε,N pointwise by Ch (λα1ε,N + ε) and by κ0 λα0ε,N , respectively, and the claim (4.21)
follows.
Step 3. Second-order cluster contribution: proof of (4.14).
By definition, cf. (4.4), we have
N
(1)
X
m
uε,N = wε,N , (4.22)
m=1
2This equation coincides with (4.2) up to removing the propulsion force of the n-th particle, which
n
would create an additional O(1) self-interaction term at Iε,N .
26 M. DUERINCKX
m := um − v
in terms of wε,N m {m} v
ε,N ε,N , where we recall the short-hand notation uε,N := π ε,N .
m
As in (4.8), we find the following equation for wε,N in R , d
m λ m m m
− △wε,N + ∇p = − e1Iε,N m + κ0
εN f ε,N 1 I m + δ∂I m σ(u
ε,N ε,N ε,N , p ε,N )ν . (4.23)
N
X 1
m n −d
ελ
|D(um 2 2
. N |Xε,N − Xε,N | ε,N )| . (4.25)
m;+
m:m6=n Iε,N
In order to estimate the last factor in the right-hand side, we appeal to an energy estimate:
m with w m itself, and using boundary conditions for
testing the equation (4.23) for wε,N ε,N
m m
wε,N = uε,N − vε,N , we get the energy identity
ˆ ˆ
m 2 m m
|∇wε,N | = − wε,N − wε,N · σ(um m
ε,N , pε,N )ν
Rd m
∂Iε,N m
Iε,N
ˆ
= vε,N − vε,N · σ(um m
ε,N , pε,N )ν,
m
∂Iε,N m
Iε,N
.h ελα0ε,N . (4.27)
n (0)
Recalling (4.22) and writing wε,N = unε,N − uε,N for the diagonal term, we are then led to
N
(0) (1)
X
n n m m 0
n u ε,N + u ε,N − u ε,N − ∇G(X ε,N − X ε,N ) Σ ε,N .h ελαε,N , (4.28)
Iε,N n Iε,N m:m6=n
SEMI-DILUTE RHEOLOGY OF PARTICLE SUSPENSIONS 27
Here, we recall that ⊗◦s stands for the trace-free symmetric tensor product: we have
used both the vanishing torque condition for um ε,N and the incompressibility constraint
div(G) = 0 to restrict Σmε,N to its trace-free symmetric part in (4.28). It remains to evalu-
ate the diagonal term and the stresslets in (4.28): we claim that for all 1 ≤ m ≤ N ,
m m
Σε,N − 2|Iε,N |Σ◦ (Rε,N
m
) D(G ∗ h)(Xε,Nm m
) − κ0 |Iε,N m
|Σ′f (Rε,N )
.h εd λ(α1ε,N + κ0 α0ε,N ) + ε ,
(4.30)
um m
ε,N − G ˆ∗ (1 − λµε,N )h + λµε,N e (Xε,N )
m
Iε,N
m
+ κ0 λ[∇G ˆ
∗ (hνε,N Mf i)](Xε,N ) − κ0 εVf◦ (Rε,N
m
)
.h ε λ(α2ε,N + α1ε,N + α0ε,N ) + ε ,
(4.31)
recalling that (u◦r,f , p◦r,f ) is the unique decaying solution of the single-particle problem (1.22).
The proof of these two estimates (4.30)–(4.31) is split into the following three substeps.
Inserting them into (4.28), noting that we recover Σ′f − Mf = Σ◦f as defined in (1.21), and
further using convolution notation, the conclusion (4.14) follows.
λ PN p
−△um m
ε,N + ∇pε,N = h1Rd \Iε,N + e1Iε,N \Iε,N m + κ0
εN
d m
p=1 fε,N , in R \ Iε,N ,
m in Rd \ Iε,N
m ,
div(uε,N ) = 0,
m
D(uε,N ) = 0, m
in Iε,N ,
λ e + κ λ Rm + ´ m m
0 εN ε,N m σ(uε,N , pε,N )ν = 0,
´N ∂Iε,N
m m m
∂I m (x − Xε,N ) × σ(uε,N , pε,N )ν = 0,
ε,N
um m m m m
ε,N = vε,N + wε,N ;0 + wε,N ;1 + eε,N , (4.34)
m is defined in (4.20), and where:
where we recall that vε,N
28 M. DUERINCKX
m
— wε,N ;0 is the unique decaying solution of the single-particle problem
−△wε,N m m in Rd \ Iε,N
m ,
;0 + ∇qε,N ;0 = 0,
m in Rd \ Iε,N
m ,
div(w ε,N ;0 ) = 0,
m m m
in Iε,N ,
´D(wε,N ;0 ) m + Hε,N = 0,
m
(4.35)
m σ(wε,N ;0 , qε,N ;0 )ν = 0,
∂Iε,N
´ m (x − X m ) × σ(wm , q m )ν = 0,
∂I ε,N
ε,N ε,N ;0 ε,N ;0
m is chosen as
where the strain rate Hε,N
m m
Hε,N := D(G ∗ h)(Xε,N ); (4.36)
m
— wε,N ;1 is the unique decaying solution of the single-particle problem
m m λ m
−△wε,N ;1 + ∇qε,N ;1 = κ0 εN fε,N , in Rd \ Iε,N
m ,
m in Rd \ Iε,N
m ,
div(w ε,N ;1 ) = 0,
m
D(wε,N ;1 ) = 0, m
in Iε,N , (4.37)
λ m m m
´
κ0 R + m σ(wε,N ;1 , qε,N ;1 )ν = 0,
´ εN ε,N m∂Iε,N
∂I m (x − Xε,N ) × σ(wε,N m m
;1 , qε,N ;1 )ν = 0.
ε,N
Using Lemma 2.1, a direct computation then entails that the remainder
em m m m m
ε,N := uε,N − vε,N − wε,N ;0 − wε,N ;1
·σ um m m m m m m
ε,N − wε,N ;0 − Hε,N x − wε,N ;1 , qε,N − qε,N ;0 − qε,N ;1 ν
ˆ
m m m m
= (vε,N − Hε,N x) − (vε,N − Hε,N x)
m
∂Iε,N m
Iε,N
·σ um m m m m m m
ε,N − wε,N ;0 − Hε,N x − wε,N ;1 , qε,N − qε,N ;0 − qε,N ;1 ν,
It remains to evaluate the different terms in the left-hand side. First, we compute
ˆ ˆ
m m ◦ m m
σ(vε,N , qε,N )ν ⊗s (x − Xε,N ) = 2D(vε,N ),
m
∂Iε,N m
Iε,N
.h εd λ(α1ε,N + κ0 α0ε,N ) + ε .
m
Next, as wε,N m
;0 and wε,N ;1 satisfy the single-particle problems (4.35)–(4.37), which can be
compared to (1.18) and (1.22), we simply find by scaling
ˆ
m m ◦ m m ◦ m m m
σ(wε,N ;0 , qε,N ;0 )ν ⊗s (x − Xε,N ) = 2|Iε,N | Σ (Rε,N )Hε,N − Hε,N ,
m
∂Iε,N
ˆ
m m ◦ m m ′ m
σ(wε,N ;1 , qε,N ;1 )ν ⊗s (x − Xε,N ) = κ0 |Iε,N |Σf (Rε,N ),
m
∂Iε,N
where we recall the definition of Σ◦ , Σ′f in (1.19) and (4.33). Inserting these different
m = D(G ∗ h)(X m ), cf. (4.36), the claim (4.30)
computations into (4.39), and recalling Hε,N ε,N
follows.
Substep 3.3. Proof of (4.31).
Using again the decomposition (4.34) for um
ε,N , and noting that, by the Sobolev embedding,
the remainder estimate (4.38) yields
ˆ 2d
d−2
m 1− d2
|em
2d
eε,N . ε ε,N |
d−2
m
Iε,N m
Iε,N
d
ˆ 1
ε1− 2 |∇em 2 2
. ε,N |
m
Iε,N
.h ε λ(α1ε,N + κ0 α0ε,N ) + ε ,
we deduce
um m m m 1 0
ε,N − vε,N − wε,N ;0 − wε,N ;1 .h ε λ(αε,N + κ0 αε,N ) + ε . (4.40)
m
Iε,N m
Iε,N m
Iε,N m
Iε,N
30 M. DUERINCKX
It remains to evaluate the different terms in the left-hand side. First, recalling equa-
m , we can represent, in terms of the Stokeslet G,
tion (4.20) for vε,N
N ˆ N ˆ
p
X X
m λ
vε,N (x) = (G ∗ h)(x) + G(x − ·)(e − h) + κ0 εN G(x − ·)fε,N . (4.41)
p p;+
p=1 Iε,N p:p6=m Iε,N
.h ελ(α2ε,N + α1ε,N ),
N ˆ N
λ X p p p
X
m
λ
− Xε,N )Mf (Rε,N ) . ελα0ε,N ,
εN G(x − ·)fε,N + N ∇G(Xε,N
m p;+
p:p6=m Iε,N Iε,N p:p6=m
in terms of the first moment Mf of the propulsion force, cf. (4.32). In addition, for the
diagonal contribution in the second right-hand side term of (4.41), we can estimate
ˆ ˆ
G(x − ·)(e − h) .h |x − ·|2−d . ε2 .
m
Iε,N m
Iε,N m
Iε,N m
Iε,N
Inserting these different computations into (4.41), and using convolution notation, we get
m
m
m
vε,N − Gˆ
∗ (1 − λµε,N )h + λµε,N e (Xε,N ) + κ0 λ ∇G ∗ (hνε,N Mf i) (Xε,N )
m
Iε,N
m
We turn to the analysis of the last two left-hand side contributions in (4.40). As wε,N ;0
solves the single-particle problem (4.35), we note that it actually satisfies the Dirichlet
m
condition wε,N m m m
;0 (x) = −Hε,N (x − Xε,N ) in Iε,N , hence
m
wε,N ;0 = 0.
m
Iε,N
m
As wε,N ;1 satisfies the single-particle problem (4.37), which can be compared to (1.22), we
find by scaling
m ◦ m
wε,N ;1 = κ0 εVf (Rε,N ),
m
Iε,N
where we recall the definition (1.24) of Vf◦ . Inserting these identities into (4.40), together
m = D(G ∗ h)(X m ), cf. (4.36), and using convolution notation,
with (4.42), recalling Hε,N ε,N
the claim (4.31) follows.
SEMI-DILUTE RHEOLOGY OF PARTICLE SUSPENSIONS 31
It remains to evaluate the different terms in the left-hand side. First, we get from (4.21),
n n
∇vε,N − ∇(G ∗ h)(Xε,N ) .h λ(α1ε,N + κ0 α0ε,N ) + ε.
n
Iε,N
n
Next, as wε,N ;0 satisfies the single-particle problems (4.35), which can be compared to (1.18),
we find by scaling
n
∇wε,N ;0 = ∇u◦Rn n
n ε,N ,Hε,N
Iε,N I ◦ (Rn
ε,N )
= Ω◦ (Rε,N
n n
)∇(G ∗ h)(Xε,N n
) − ∇(G ∗ h)(Xε,N ),
where we recall the definition (1.20) of Ω◦ and the choice (4.36) of Hε,N n as the symmetric
n ). Finally, as w n
part of ∇(G ∗ h)(Xε,N ε,N ;1 satisfies the single-particle problem (4.37) with
n
rigidity constraint D(wε,N ) = 0 in n , and as the inclusion I m and the propulsion
Iε,N
;1 ε,N
m are both axisymmetric in the direction Rm , centered at X m , we find by
force fε,N ε,N ε,N
symmetry
n
∇wε,N ;1 = 0.
n
Iε,N
Inserting these different computations into (4.43), the conclusion (4.13) follows.
Step 5. Fluid velocity away from the particles: proof of (4.15)–(4.16).
Let the boundary-layer thickness δ ∈ [ε, 1] be fixed. The proof of (4.15)–(4.16) is slightly
simpler than that of (4.12)–(4.14) as self-interaction terms can be ignored away from the
(0)
particles. We start with (4.15). Recalling the representation (4.17) for uε,N = vε,N , and
using the pointwise bounds on G, cf. Lemma 2.3, we get for all x ∈ Rd \ (Iε,N + δB),
N
(0)
X
m 2−d m 1−d
kuε,N −G ∗ hkL (Rd \(Iε,N +δB)) .h Nλ
∞ |x − Xε,N | + κ0 |x − Xε,N | .
m=1
To estimate the right-hand side, we use (3.5) in the following modified form away from the
particles: for all x ∈ Rd \ (Iε,N + δB), we can estimate for any σ ∈ [0, d], after separating
the diagonal contribution as in (4.9),
N
X
m σ−d
1
N |x − Xε,N | . ασε,N + 1 σ−d
Nδ ≃ ασε,N + λ1 δσ ( δε )d . (4.44)
m=1
Using this to estimate the above right-hand side, we get the conclusion (4.15).
(0)
We turn to the proof of (4.16) and we start with an improved estimate on uε,N = vε,N .
Replacing G by its Taylor expansion in the representation (4.17) for the latter, we find for
32 M. DUERINCKX
(1)
We turn to the corresponding analysis of uε,N . As in (4.22)–(4.24), we can write
N ˆ
(1)
X
uε,N = − G(x − ·) − G(x − ·) σ(um m
ε,N , pε,N )ν. (4.46)
m m
m=1 ∂Iε,N Iε,N
Now inserting the approximation (4.30) for the stresslets, using (4.44) again, using convo-
lution notation, and noting that ( δε )d δ ≤ ε, we deduce
(1)
uε,N − λ∇G ∗ 2hΣ◦ νε,N iD(G ∗ h) + κ0 hΣ′f νε,N i
∞ d
L (R \(Iε,N +δB))
Combining this with (4.45) and recalling Σ◦f = Σ′f − Mf , we get the conclusion (4.16).
5. Mean-field approximation
Given the dilute expansion of particle velocities in Proposition 4.1, we now appeal to
a mean-field argument to derive a macroscopic equation for the particle density. For
that purpose, as in [25, 32, 29], we use the ∞-Wasserstein method developed by Hauray
and Jabin in [21] (see also [4]), combined with a refinement found in [29]. This method
can generally be applied whenever particle interactions are less singular than Coulomb
forces at short distances, which is indeed the case in the present setting as hydrodynamic
interactions are given to leading order by the Stokeslet. We start with the following mean-
field approximation with accuracy O((λ + ε)2 + κ0 ε), neglecting swimming velocities.
SEMI-DILUTE RHEOLOGY OF PARTICLE SUSPENSIONS 33
Proposition 5.1. Assume that there is a solution ν̃λ ∈ C([0, T ]; P ∩ L∞ (Rd × Sd−1 )) of
the following transport equation up to some time T > 0,
∂t ν̃λ + divx (ν̃λ ũλ ) + divr (ν̃λ Ω̃λ r) = 0,
(5.1)
ν̃λ |t=0 = ν ◦ ,
where the translational and angular velocity fields ũλ and Ω̃λ are given by
ũλ (x) := G ∗ (1 − λµ̃λ )h + λµ̃λ e (x) (5.2)
+λ ∇G ∗ 2hΣ◦ ν̃λ i D(G ∗ h) + κ0 hΣ◦f ν̃λ i (x),
Here, we use the notation sup essx;µ g(x) for the essential supremum of a function g with
respect to a measure µ. We split the proof into three steps.
∞ : proof that
Step 1. First-order control on Zε,N
d+ ∞ ∞
+ λ α0ε,N + kµ̃λ kL1 ∩ L∞ (Rd ) + ε,
dt Zε,N .h Zε,N (5.9)
34 M. DUERINCKX
d+
where dt stands for the right-derivative.
Recalling (5.3), we focus on the bound on W∞ (νε,N , ν̃λ ), while the argument is analogous
for W∞ (µε,N , µ̃λ ). In view of the particle dynamics (1.11) and of the macroscopic transport
equation (5.1), we can estimate the time-derivative of the ∞-Wasserstein distance (5.8) by
using characteristics similarly as in [4], to the effect of
d+
dt W (ν
∞ ε,N λ, ν̃ ) ≤ sup ess |ũλ (x) − Vε,N (X̃ε,N (x, r))|
x,r;ν̃λ
+ Ω̃λ (x, r)r − Ωε,N (X̃ε,N (x, r))R̃ε,N (x, r) ,
d+ ∞ ∞
+ (λα1ε,N + ε) λ(α0ε,N + 1) + ε + κ0 ε
dt Wε,N .h Wε,N
1
∞ ∞ 2 ∞ 2
1
) (N d dmin 2−d
+ λ Zε,N log 2 + Z ∞ + (Wε,N ) + (Wε,N ε,N )
ε,N
1
∞
+ Zε,N (N d dmin
ε,N )
1−d
kµ̃λ kL1 ∩ L∞ (Rd ) . (5.10)
Inserting the second-order expansion of particle velocities stated in Proposition 4.1, as well
as the definition (5.2) of limiting velocities, we deduce
d+ ∞
dt Wε,N .h sup ess (G ∗ h)(x) − (G ∗ h)(X̂ε,N (x))
x;µ̃λ
+ λ sup ess G ∗ (µ̃λ (e − h)) (x) − G ˆ∗ (µε,N (e − h)) (X̂ε,N (x))
x;µ̃λ
SEMI-DILUTE RHEOLOGY OF PARTICLE SUSPENSIONS 35
+ λ sup ess ∇G ∗ (hΣ◦f ν̃λ i) (x) − ∇G ˆ∗ (hΣ◦f νε,N i) (X̂ε,N (x))
x;µ̃λ
◦ ◦
+ λ sup ess ∇G ∗ (hΣ ν̃λ i D(G ∗ h)) (x) − ∇G ˆ∗ (hΣ νε,N i D(G ∗ h)) (X̂ε,N (x))
x;µ̃λ
We analyze the different right-hand side terms separately. First, as in Step 1, the first
term is bounded by Ch Wε,N ∞ . Next, the second term can be decomposed as
ˆ
G ∗ (µ̃λ (e − h)) (x) − G ∗ (µε,N (e − h)) (X̂ε,N (x))
ˆ ˆ
≤ G(x − y) µ̃λ (y) dy − G(X̂ε,N (x) − y) µε,N (y) dy
Rd Rd \{X̂ε,N (x)}
ˆ ˆ
+ G(x − y) h(y) µ̃λ (y) dy − G(X̂ε,N (x) − y) h(y) µε,N (y) dy ,
Rd Rd \{X̂ε,N (x)}
G ∗ (µ̃λ (e − h)) (x) − G ˆ∗ (µε,N (e − h)) (X̂ε,N (x))
ˆ
∞
.h Wε,N kµ̃λ kL1 ∩ L∞ (Rd ) + G(x − y) − (16= G)(X̂ε,N (x) − X̂ε,N (y)) µ̃λ (y) dy. (5.12)
Rd
We split the remaining integral into two parts, distinguishing between the contributions of
∞ and |x − y| < 4W ∞ . On the one hand, appealing to pointwise bounds
|x − y| ≥ 4Wε,N ε,N
on the Stokeslet, cf. Lemma 2.3, in form of
|x−x′ |
|G(x) − G(x′ )| . (|x|∧|x′ |)d−1
,
∞ entails
and noting that the condition |x − y| ≥ 4Wε,N
∞ 1
|X̂ε,N (x) − X̂ε,N (y)| ≥ |x − y| − 2Wε,N ≥ 2 |x − y|,
we find
ˆ
G(x − y) − (16= G)(X̂ε,N (x) − X̂ε,N (y)) µ̃λ (y) dy
∞
y:|x−y|≥4Wε,N
ˆ
= G(x − y) − G(X̂ε,N (x) − X̂ε,N (y)) µ̃λ (y) dy
∞
y:|x−y|≥4Wε,N
ˆ
|x−X̂ε,N (x)|+|y−X̂ε,N (y)|
. (|x−y|∧|X̂ε,N (x)−X̂ε,N (y)|)d−1
µ̃λ (y) dy
∞
y:|x−y|≥4Wε,N
36 M. DUERINCKX
ˆ
∞
. Wε,N |x − y|1−d µ̃λ (y) dy
∞
y:|x−y|≥4Wε,N
∞
. Wε,N kµ̃λ kL1 ∩ L∞ (Rd ) . (5.13)
On the other hand, using pointwise bounds on the Stokeslet, cf. Lemma 2.3, we get
ˆ
G(x − y) − (16= G)(X̂ε,N (x) − X̂ε,N (y)) µ̃λ (y) dy
∞
y:|x−y|<4Wε,N
ˆ
. |x − y|2−d µ̃λ (y) dy
∞
y:|x−y|<4Wε,N
ˆ
+ |X̂ε,N (x) − X̂ε,N (y)|2−d 1X̂ε,N (x)6=X̂ε,N (y) µ̃λ (y) dy,
∞
y:|x−y|<4Wε,N
regime. Instead, we use an idea by Höfer and Schubert [29]: as shown in [29, Lemma 3.1],
there holds for any σ ∈ (0, d) and ℓ ≥ Wε,N ∞ ,
X σ 1
n m σ−d
N
1
|X ε,N − Xε,N | 1 |X n −X m |≤ℓ .σ kµ̃λ k ∞
ε,N ε,N
d
L (Rd )
ℓσ (N d dmin
ε,N )
σ−d
. (5.14)
m:m6=n
We turn to the third right-hand side term in (5.11). Using the transport plan (X̃ε,N , R̃ε,N )
given by (5.8), with (X̃ε,N , R̃ε,N )∗ ν̃λ = νε,N , and recalling that Σ◦f is smooth (it is actually
SEMI-DILUTE RHEOLOGY OF PARTICLE SUSPENSIONS 37
On the other hand, using pointwise bounds on the Stokeslet, as well as (5.14), we get
ˆ
∇G(x − y) − (16= ∇G)(X̂ε,N (x) − X̃ε,N (y, r)) ν̃λ (y, r) dydσ(r)
∞
y,r:|x−y|≤4Zε,N
1 1
∞
. Zε,N kµ̃λ kL∞ (Rd ) + kµ̃λ kLd∞ (Rd ) (N d dmin
ε,N )1−d
.
Finally, regarding the fourth right-hand side term in (5.11), a similar argument yields
∇G ∗ (hΣ◦ ν̃λ i D(G ∗ h)) (x) − ∇G ˆ∗ (hΣ◦ νε,N i D(G ∗ h)) (X̂ε,N (x))
1
∞ ∞
1
(N d dmin 1−d
.h Zε,N log 2 + Z ∞ + Zε,N ε,N ) kµ̃λ kL1 ∩ L∞ (Rd ) .
ε,N
Inserting this into (5.11), together with (5.15), and (5.17), the claim (5.10) follows.
Step 3. Conclusion.
We use the notation Ch,µ (t) for any constant that further depends on h, t, and on an upper
38 M. DUERINCKX
bound on kµ̃sλ kL∞ ([0,t];L∞ (Rd )) . Recall the assumption Tε,N θ ≥ T , where T θ
ε,N is the maximal
time in Corollary 3.2,
n o
θ − d1
Tε,N := sup t ≥ 0 : dmin ε,N (s) ≥ (θN ) , ρmax
ε,N (s) ≤ θ for all 0 ≤ s ≤ t .
By the Gronwall inequality, the result (5.9) of Step 1 yields
∞ ∞
(0) + λ α0ε,N + 1 + kµ̃λ kL∞ ([0,t];L1 ∩ L∞ (Rd )) + ε,
Zε,N (t) .h,t Zε,N
and thus, using (3.11), for all t ∈ [0, T ],
∞ ∞
Zε,N (t) ≤ Ch,µ (t) Zε,N (0) + λθ log(θN ) + ε , (5.18)
which already proves (5.4).
We turn to the proof of (5.5). Using (5.18) to control the last O(λ) term in the result (5.10)
of Step 2, and using again (3.11), we get for all t ∈ [0, T ],
d+ ∞ ∞ 2
dt W ε,N . h W ε,N + (λθ + ε) λθ log(θN ) + ε + κ0 ε
∞ 1
+ λCh,µ Zε,N (0) + λθ log(θN ) + ε log 2 + Z ∞ (0)+λ log N +ε
ε,N
+ θ(Zε,N (0)) + θZε,N (0) + θε + λθ 2 log(θN ) .
∞ 2 ∞
where the infimum runs over all measurable maps X : Rd → Rd with X∗ µ̃tλ,ε = µtε,N , and
n ) := Rn
where the map Rε,N is defined by setting Rε,N (Xε,N ε,N for all 1 ≤ n ≤ N . We
t d d
denote by X̃ε,N : R → R an optimal transport map for this problem (5.24).
With this notation, we now turn to the control of Zε,N ∞ . Using characteristics, its time-
where, just as for Rε,N , we have defined Vε,N and Ωε,N by setting Vε,N (Xε,N n ) = Vn
ε,N
n n
and Ωε,N (Xε,N ) = Ωε,N for all 1 ≤ n ≤ N . Inserting the first-order expansions of particle
translational and angular velocities stated in Proposition 4.1, as well as the definition (5.20)
of limiting velocities, we find
d+ ∞
dt Zε,N . h sup ess (G ∗ h)(x) − (G ∗ h)(X̃ε,N (x))
x;µ̃λ,ε
+ Ω◦ (r̃λ,ε (x))∇(G∗h)(x) r̃λ,ε (x)− Ω◦ (Rε,N (X̃ε,N (x)))∇(G∗h)(X̃ε,N (x)) Rε,N (X̃ε,N (x))
To estimate the right-hand side, we note that the Lipschitz continuity of r̃λ,ε yields
∞
sup ess |r̃λ,ε (x) − Rε,N (X̂ε,N (x))| ≤ 1 + k∇r̃λ,ε kL∞ (Rd ) Zε,N .
x;µ̃λ,ε
The first two right-hand side terms in (5.27) are then bounded by
∞
∞
Ch Wε,N + κ0 ε 1 + k∇r̃λ,ε kL∞ (Rd ) Zε,N ,
while the remaining three terms in bracket can be estimated similarly as in Step 2 of the
proof of Proposition 5.1 above. This leads us to
d+ ∞ ∞
∞ 1 0
dt Wε,N .h Wε,N + κ0 ε 1 + k∇r̃λ,ε kL (Rd ) Zε,N + (λαε,N + ε) λ(αε,N + 1) + ε
∞
1
∞ ∞ 2 ∞ 2
1
) (N d dmin 2−d
+ λ Zε,N log 2 + Z ∞ + (Wε,N ) + (Wε,N ε,N )
ε,N
1
∞
+ Zε,N (N d dmin
ε,N )
1−d
kµ̃λ,ε kL1 ∩ L∞ (Rd ) .
Combining this with (5.25) and (5.26), and appealing to the Gronwall inequality, the
conclusion easily follows as in Step 3 of the proof of Proposition 5.1. We skip the detail
for shortness.
6.1. Proof of Proposition 1.2. We focus on the proof of Proposition 1.2(i) for the well-
posedness of (1.26), while the argument is similar for (1.27). Let some non-integer γ > 1
be fixed. We proceed by an iteration argument. For n = 0, let ν0 := ν ◦ and u0 := 0. Next,
for all n ≥ 0, we iteratively define νn+1 as the solution of the linear transport equation
∂t νn+1 + divx (νn+1 un ) + divr (νn+1 (Ω◦ ∇un )r) = 0,
(6.1)
νn+1 |t=0 = ν ◦ ,
and ∇un+1 as the solution of the linear Stokes equation
div(un+1 ) = 0,
which means, in terms of the Stokeslet G,
that initially ν ◦ ∈ P ∩ W 1,1 ∩ W γ,∞(Rd × Sd−1 ), the standard theory of transport equations
in Hölder spaces (e.g. [1, Theorem 3.14]) ensures that (6.1) admits a unique weak solution
νn+1 ∈ L∞ +
loc (R ; P ∩ W
1,1 ∩ W γ,∞ (Rd × Sd−1 )) with
t
kνn+1 kW 1,1 ∩W γ,∞ (Rd ×Sd−1 )
ˆ t
◦
≤ kν kW 1,1 ∩W γ,∞ (Rd ×Sd−1 ) exp Cγ t + Cγ kun kW γ+1,∞ (Rd ) . (6.3)
0
Moreover, if νn ∈ L∞ +
loc (R ; P ∩ W
γ,∞ (Rd × Sd−1 )) and un ∈ L ∞ +
loc (R ; W
γ+1,∞ (Rd )d ), re-
calling that Σ◦ , Σ◦f are smooth and using the standard theory for the Stokes equation in
Hölder spaces, we find that equation (6.2) yields un+1 ∈ L∞ +
loc (R ; W
γ+1,∞ (Rd )d ) with
kun+1 kW γ+1,∞ (Rd ) .h,γ 1 + λkνn kL1 ∩W γ,∞ (Rd ×Sd−1 ) 1 + kun kW γ+1,∞ (Rd ) . (6.4)
By induction, this proves that we can indeed construct unique global weak solutions
for the scheme (6.1)–(6.2) with νn ∈ L∞ +
loc (R ; P ∩ W
1,1 ∩ W γ,∞ (Rd × Sd−1 )) and u ∈
n
∞ + γ+1,∞ d d
Lloc (R ; W (R ) ) for all n ≥ 0. From the above a priori estimates (6.3)–(6.4), ab-
sorbing the nonlinearity for small λ, we conclude the following: given T > 0, provided that
λ ≪T,h,γ,ν ◦ 1 is small enough, we have for all n ≥ 0 and t ∈ [0, T ],
kνnt kW 1,1 ∩W γ,∞ (Rd ×Sd−1 ) ≤ eCh,γ t kν ◦ kW 1,1 ∩W γ,∞ (Rd ×Sd−1 ) , (6.5)
kutn kW γ+1,∞ (Rd ) ≤ Ch,γ .
Further appealing to the Aubin lemma, we may then extract a subsequence of (νn , un )n that
converges strongly to some limit (ν, u) in C([0, T ]; L1 (Rd × Sd−1 )) × C([0, T ]; W 1,∞ (Rd )d ).
Passing to the limit in the iterative scheme (6.1)–(6.2), we find that the limit (ν, u) precisely
satisfies the Vlasov–Stokes system (1.26), and we may also pass to the limit in the a priori
estimates (6.5).
It remains to establish the uniqueness of the solution of (1.26). Let (ν, u), (ν∗ , u∗ ) be
two solutions in L∞loc ([0, T ]; P ∩ W
1,1 ∩ W γ,∞ (Rd × Sd−1 )) × L∞ ([0, T ]; W γ+1,∞ (Rd )d ) with
loc
the same initial data ν|t=0 = ν∗ |t=0 = ν ◦ . The equation for the difference ν − ν∗ can be
42 M. DUERINCKX
written as
6.2. Proof of Theorems 1.3 and 1.4. We focus on the proof of Theorem 1.3, which is
a simple post-processing of Proposition 5.1. Theorem 1.4 is similarly obtained by post-
processing of Proposition 5.2 and we skip the corresponding detail for shortness.
First note that a simpler version of the proof of Proposition 1.2(i) also yields the following
well-posedness result for (5.1)–(5.2): given T > 0, non-integer γ > 1, and given an initial
condition ν ◦ ∈ P ∩ W 1,1 ∩ W γ,∞ (Rd × Sd−1 ), there is λ0 > 0 (depending on T, h, γ, ν ◦ )
such that for all 0 ≤ λ ≤ λ0 there is a unique solution (ν̃λ , ũλ ) of (5.1)–(5.2) up to time T
with ν̃λ ∈ C([0, T ]; P ∩ W 1,1 ∩ W γ,∞ (Rd × Sd−1 )) and ũλ ∈ C([0, T ]; W γ+1,∞ (Rd )d ).
In order to deduce Theorem 1.3 from Proposition 5.1, it remains to compare (ν̃λ , ũλ ) to
the solution (νλ , uλ ) of (1.26): we shall prove for all t ∈ [0, T ],
kutλ − ũtλ kW γ,∞ (Rd ) .t,h,γ λ2 , (6.7)
W∞ (µtλ , µ̃tλ ) .t,h λ2 , (6.8)
W∞ (νλt , ν̃λt ) .t,h λ. (6.9)
First, taking the difference between the equations for νλ and ν̃λ , we find
∂t (νλ − ν̃λ ) + divx (νλ − ν̃λ )uλ + divr (νλ − ν̃λ )(Ω◦ ∇uλ )r
from which we can deduce the following estimate, similarly as in (6.6), further using a
priori bounds on (νλ , uλ ), (ν̃λ , ũλ ), for all t ∈ [0, T ],
ˆ t
t t
kνλ − ν̃λ kL1 ∩W γ−1,∞ (Rd ×Sd−1 ) .t,h,γ,ν ◦ kuλ − ũλ kW γ−1,∞ (Rd ) + kuλ − G ∗ hkW γ,∞ (Rd ) .
0
SEMI-DILUTE RHEOLOGY OF PARTICLE SUSPENSIONS 43
Decomposing uλ − G ∗ h = (uλ − ũλ ) + (ũλ − G ∗ h) and using (5.2) to estimate the second
piece, this yields for all t ∈ [0, T ],
ˆ t
t t
kνλ − ν̃λ kL1 ∩W γ−1,∞ (Rd ×Sd−1 ) .t,h,γ,ν ◦ λ + kuλ − ũλ kW γ,∞ (Rd ) . (6.10)
0
Next, comparing the Stokes equation for uλ in (1.26) with the definition (5.2) of ũλ ,
which implies, using again a priori bounds on (νλ , uλ ), (ν̃λ , ũλ ), for all t ∈ [0, T ],
kutλ − ũtλ kW γ,∞ (Rd ) .t,h,γ,ν ◦ λkνλt − ν̃λt kL1 ∩W γ−1,∞ (Rd ×Sd−1 ) + λkutλ − ũtλ kW γ,∞ (Rd ) ,
and thus, for λ ≪T,h,γ,ν ◦ 1 small enough,
kutλ − ũtλ kW γ,∞ (Rd ) .t,h,γ,ν ◦ λkνλt − ν̃λt kL1 ∩W γ−1,∞ (Rd ×Sd−1 ) .
Combining this with (6.10) and appealing to the Gronwall inequality, we deduce
kνλt − ν̃λt kW γ−1,∞ (Rd ×Sd−1 ) .t,h,γ,ν ◦ λ,
kutλ − ũtλ kW γ,∞ (Rd ) .t,h,γ,ν ◦ λ2 . (6.12)
The last estimate is already (6.7). We turn to proof of (6.8) and (6.9), that is, corresponding
estimates for the ∞-Wasserstein distances between νλ , µλ = hνλ i and ν̃λ , µ̃λ = hν̃λ i. For
all t ∈ [0, T ], denote by X̂λt : Rd → Rd an optimal transport map such that
(X̂λt )∗ µtλ = µ̃tλ , W∞ (µtλ , µ̃tλ ) = sup ess |x − X̂λt (x)|,
x;µtλ
and denote by (X̃λt , R̃λt ) : Rd × Sd−1 → Rd × Sd−1 an optimal transport map such that
(X̃λt , R̃λt )∗ νλt = ν̃λt , W∞ (νλt , ν̃λt ) = sup ess (x, r) − (X̃λt (x, r), R̃λt (x, r)).
x,r;νλt
Comparing the transport equations for µλ and µ̃λ , which are obtained by integrating out
equations (1.26) and (5.1) with respect to r,
∂t µλ + div(uλ µλ ) = 0, ∂t µ̃λ + div(ũλ µ̃λ ) = 0,
and using characteristics, we can estimate
d+
dt W∞ (µλ , µ̃λ ) ≤ sup ess |uλ (x) − ũλ (X̂λ (x))|,
x;µλ
Inserting the definition (5.2) of Ω̃λ , and using again the a priori Lipschitz bound on ũλ
and the smoothness of Ω◦ , we are led to
d+
dt W∞ (νλ , ν̃λ ) .t,h W∞ (νλ , ν̃λ ) + kuλ − ũλ kL∞ (Rd ) + k∇(uλ − G ∗ h)kL∞ (Rd ) .
Decomposing again uλ − G ∗ h = (uλ − ũλ ) + (ũλ − G ∗ h), using (5.2) to estimate the second
piece, and then combining the result with (6.12) and appealing to the Gronwall inequality,
the claim (6.9) follows. This ends the proof of (6.7)–(6.9). The conclusion of Theorem 1.3
can then be deduced from Proposition 5.1, replacing (ν̃λ , ũλ ) by (νλ , uλ ).
Acknowledgements
The author thanks Alexandre Girodroux-Lavigne, Antoine Gloria, Richard Höfer, and
Amina Mecherbet for related discussions, as well as the F.R.S.-FNRS for financial support.
References
[1] H. Bahouri, J.-Y. Chemin, and R. Danchin. Fourier analysis and nonlinear partial differential equa-
tions, volume 343 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of
Mathematical Sciences]. Springer, Heidelberg, 2011.
[2] A. Bernou, M. Duerinckx, and A. Gloria. Homogenization of active suspensions and reduction of
effective viscosity. Preprint, arXiv:2301.00166.
[3] H. Brenner. Rheology of a dilute suspension of axisymmetric Brownian particles. Int. J. Multiphase
Flow, 1:195–341, 1974.
[4] J. A. Carrillo, Y.-P. Choi, and M. Hauray. The derivation of swarming models: Mean-field limit and
Wasserstein distances. In Collective Dynamics from Bacteria to Crowds, CISM International Centre
for Mechanical Sciences, pages 1–46. Springer, 2014.
[5] A. Decoene, S. Martin, and B. Maury. Microscopic modelling of active bacterial suspensions. Math.
Model. Nat. Phenom., 6:98–129, 2011.
[6] S. Degond, P. ad Merino-Aceituno, F. Vergnet, and H. Yu. Coupled Self-Organized Hydrodynamics
and Stokes Models for Suspensions of Active Particles. J. Math. Fluid Mech., 21(6):1–36, 2019.
[7] M. Doi. Molecular-dynamics and rheological properties of concentrated-solutions of rodlike polymers
in isotropic and liquid-crystalline phases. J. Polym. Sci. Polym. Phys. Ed., 19:229–243, 1981.
[8] M. Doi and S. F. Edwards. Dynamics of rod-like macromolecules in concentrated solution. Part 1. J.
Chem. Soc., Faraday Trans. 2, 74:560–570, 1978.
[9] M. Doi and S. F. Edwards. The theory of polymer dynamics. Oxford University Press, 1988.
[10] M. Duerinckx. On the size of chaos via Glauber calculus in the classical mean-field dynamics. Commun.
Math. Phys., 382:613–653, 2021.
[11] M. Duerinckx and A. Gloria. Continuum percolation in stochastic homogenization and the effective
viscosity problem. Preprint, arXiv:2108.09654.
[12] M. Duerinckx and A. Gloria. On Einstein’s effective viscosity formula. Preprint, arXiv:2008.03837.
[13] M. Duerinckx and A. Gloria. Analyticity of homogenized coefficients under Bernoulli perturbations
and the Clausius-Mossotti formulas. Arch. Ration. Mech. Anal., 220(1):297–361, 2016.
[14] M. Duerinckx and A. Gloria. Corrector equations in fluid mechanics: Effective viscosity of colloidal
suspensions. Arch. Ration. Mech. Anal., 239:1025–1060, 2021.
[15] M. Duerinckx and A. Gloria. Effective viscosity of semi-dilute suspensions. Séminaire Laurent
Schwartz, EDP et applications, 2021-2022. Exposé n◦ III.
[16] M. Duerinckx and A. Gloria. Quantitative homogenization theory for random suspensions in steady
stokes flow. J. Éc. Polytech. - Math., 9:1183–1244, 2022.
[17] D. Gérard-Varet. Derivation of the Batchelor-Green formula for random suspensions. J. Math. Pures
Appl. (9), 152:211–250, 2021.
[18] A. Girodroux-Lavigne. Derivation of an effective rheology for dilute suspensions of micro-swimmers.
Preprint, arXiv:2204.04967.
[19] B. M. Haines, I. S. Aranson, L. Berlyand, and D. A. Karpeev. Effective viscosity of dilute bacterial
suspensions: a two-dimensional model. Phys. Biol., 5:046003, 2008.
SEMI-DILUTE RHEOLOGY OF PARTICLE SUSPENSIONS 45